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>
43 * dcache->d_inode->i_lock protects:
44 * - i_dentry, d_alias, d_inode of aliases
45 * dcache_hash_bucket lock protects:
46 * - the dcache hash table
47 * s_anon bl list spinlock protects:
48 * - the s_anon list (see __d_drop)
49 * dcache_lru_lock protects:
50 * - the dcache lru lists and counters
57 * - d_parent and d_subdirs
58 * - childrens' d_child and d_parent
62 * dentry->d_inode->i_lock
65 * dcache_hash_bucket lock
68 * If there is an ancestor relationship:
69 * dentry->d_parent->...->d_parent->d_lock
71 * dentry->d_parent->d_lock
74 * If no ancestor relationship:
75 * if (dentry1 < dentry2)
79 int sysctl_vfs_cache_pressure __read_mostly
= 100;
80 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
82 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
83 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
85 EXPORT_SYMBOL(rename_lock
);
87 static struct kmem_cache
*dentry_cache __read_mostly
;
90 * This is the single most critical data structure when it comes
91 * to the dcache: the hashtable for lookups. Somebody should try
92 * to make this good - I've just made it work.
94 * This hash-function tries to avoid losing too many bits of hash
95 * information, yet avoid using a prime hash-size or similar.
97 #define D_HASHBITS d_hash_shift
98 #define D_HASHMASK d_hash_mask
100 static unsigned int d_hash_mask __read_mostly
;
101 static unsigned int d_hash_shift __read_mostly
;
103 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
105 static inline struct hlist_bl_head
*d_hash(struct dentry
*parent
,
108 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
109 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
110 return dentry_hashtable
+ (hash
& D_HASHMASK
);
113 /* Statistics gathering. */
114 struct dentry_stat_t dentry_stat
= {
118 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
121 static int get_nr_dentry(void)
125 for_each_possible_cpu(i
)
126 sum
+= per_cpu(nr_dentry
, i
);
127 return sum
< 0 ? 0 : sum
;
130 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
131 size_t *lenp
, loff_t
*ppos
)
133 dentry_stat
.nr_dentry
= get_nr_dentry();
134 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
138 static void __d_free(struct rcu_head
*head
)
140 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
142 WARN_ON(!list_empty(&dentry
->d_alias
));
143 if (dname_external(dentry
))
144 kfree(dentry
->d_name
.name
);
145 kmem_cache_free(dentry_cache
, dentry
);
151 static void d_free(struct dentry
*dentry
)
153 BUG_ON(dentry
->d_count
);
154 this_cpu_dec(nr_dentry
);
155 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
156 dentry
->d_op
->d_release(dentry
);
158 /* if dentry was never visible to RCU, immediate free is OK */
159 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
160 __d_free(&dentry
->d_u
.d_rcu
);
162 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
166 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
167 * @dentry: the target dentry
168 * After this call, in-progress rcu-walk path lookup will fail. This
169 * should be called after unhashing, and after changing d_inode (if
170 * the dentry has not already been unhashed).
172 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
174 assert_spin_locked(&dentry
->d_lock
);
175 /* Go through a barrier */
176 write_seqcount_barrier(&dentry
->d_seq
);
180 * Release the dentry's inode, using the filesystem
181 * d_iput() operation if defined. Dentry has no refcount
184 static void dentry_iput(struct dentry
* dentry
)
185 __releases(dentry
->d_lock
)
186 __releases(dentry
->d_inode
->i_lock
)
188 struct inode
*inode
= dentry
->d_inode
;
190 dentry
->d_inode
= NULL
;
191 list_del_init(&dentry
->d_alias
);
192 spin_unlock(&dentry
->d_lock
);
193 spin_unlock(&inode
->i_lock
);
195 fsnotify_inoderemove(inode
);
196 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
197 dentry
->d_op
->d_iput(dentry
, inode
);
201 spin_unlock(&dentry
->d_lock
);
206 * Release the dentry's inode, using the filesystem
207 * d_iput() operation if defined. dentry remains in-use.
209 static void dentry_unlink_inode(struct dentry
* dentry
)
210 __releases(dentry
->d_lock
)
211 __releases(dentry
->d_inode
->i_lock
)
213 struct inode
*inode
= dentry
->d_inode
;
214 dentry
->d_inode
= NULL
;
215 list_del_init(&dentry
->d_alias
);
216 dentry_rcuwalk_barrier(dentry
);
217 spin_unlock(&dentry
->d_lock
);
218 spin_unlock(&inode
->i_lock
);
220 fsnotify_inoderemove(inode
);
221 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
222 dentry
->d_op
->d_iput(dentry
, inode
);
228 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
230 static void dentry_lru_add(struct dentry
*dentry
)
232 if (list_empty(&dentry
->d_lru
)) {
233 spin_lock(&dcache_lru_lock
);
234 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
235 dentry
->d_sb
->s_nr_dentry_unused
++;
236 dentry_stat
.nr_unused
++;
237 spin_unlock(&dcache_lru_lock
);
241 static void __dentry_lru_del(struct dentry
*dentry
)
243 list_del_init(&dentry
->d_lru
);
244 dentry
->d_sb
->s_nr_dentry_unused
--;
245 dentry_stat
.nr_unused
--;
248 static void dentry_lru_del(struct dentry
*dentry
)
250 if (!list_empty(&dentry
->d_lru
)) {
251 spin_lock(&dcache_lru_lock
);
252 __dentry_lru_del(dentry
);
253 spin_unlock(&dcache_lru_lock
);
257 static void dentry_lru_move_tail(struct dentry
*dentry
)
259 spin_lock(&dcache_lru_lock
);
260 if (list_empty(&dentry
->d_lru
)) {
261 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
262 dentry
->d_sb
->s_nr_dentry_unused
++;
263 dentry_stat
.nr_unused
++;
265 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
267 spin_unlock(&dcache_lru_lock
);
271 * d_kill - kill dentry and return parent
272 * @dentry: dentry to kill
273 * @parent: parent dentry
275 * The dentry must already be unhashed and removed from the LRU.
277 * If this is the root of the dentry tree, return NULL.
279 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
282 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
283 __releases(dentry
->d_lock
)
284 __releases(parent
->d_lock
)
285 __releases(dentry
->d_inode
->i_lock
)
287 list_del(&dentry
->d_u
.d_child
);
289 * Inform try_to_ascend() that we are no longer attached to the
292 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
294 spin_unlock(&parent
->d_lock
);
297 * dentry_iput drops the locks, at which point nobody (except
298 * transient RCU lookups) can reach this dentry.
305 * Unhash a dentry without inserting an RCU walk barrier or checking that
306 * dentry->d_lock is locked. The caller must take care of that, if
309 static void __d_shrink(struct dentry
*dentry
)
311 if (!d_unhashed(dentry
)) {
312 struct hlist_bl_head
*b
;
313 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
314 b
= &dentry
->d_sb
->s_anon
;
316 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
319 __hlist_bl_del(&dentry
->d_hash
);
320 dentry
->d_hash
.pprev
= NULL
;
326 * d_drop - drop a dentry
327 * @dentry: dentry to drop
329 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
330 * be found through a VFS lookup any more. Note that this is different from
331 * deleting the dentry - d_delete will try to mark the dentry negative if
332 * possible, giving a successful _negative_ lookup, while d_drop will
333 * just make the cache lookup fail.
335 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
336 * reason (NFS timeouts or autofs deletes).
338 * __d_drop requires dentry->d_lock.
340 void __d_drop(struct dentry
*dentry
)
342 if (!d_unhashed(dentry
)) {
344 dentry_rcuwalk_barrier(dentry
);
347 EXPORT_SYMBOL(__d_drop
);
349 void d_drop(struct dentry
*dentry
)
351 spin_lock(&dentry
->d_lock
);
353 spin_unlock(&dentry
->d_lock
);
355 EXPORT_SYMBOL(d_drop
);
358 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
359 * @dentry: dentry to drop
361 * This is called when we do a lookup on a placeholder dentry that needed to be
362 * looked up. The dentry should have been hashed in order for it to be found by
363 * the lookup code, but now needs to be unhashed while we do the actual lookup
364 * and clear the DCACHE_NEED_LOOKUP flag.
366 void d_clear_need_lookup(struct dentry
*dentry
)
368 spin_lock(&dentry
->d_lock
);
370 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
371 spin_unlock(&dentry
->d_lock
);
373 EXPORT_SYMBOL(d_clear_need_lookup
);
376 * Finish off a dentry we've decided to kill.
377 * dentry->d_lock must be held, returns with it unlocked.
378 * If ref is non-zero, then decrement the refcount too.
379 * Returns dentry requiring refcount drop, or NULL if we're done.
381 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
382 __releases(dentry
->d_lock
)
385 struct dentry
*parent
;
387 inode
= dentry
->d_inode
;
388 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
390 spin_unlock(&dentry
->d_lock
);
392 return dentry
; /* try again with same dentry */
397 parent
= dentry
->d_parent
;
398 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
400 spin_unlock(&inode
->i_lock
);
406 /* if dentry was on the d_lru list delete it from there */
407 dentry_lru_del(dentry
);
408 /* if it was on the hash then remove it */
410 return d_kill(dentry
, parent
);
416 * This is complicated by the fact that we do not want to put
417 * dentries that are no longer on any hash chain on the unused
418 * list: we'd much rather just get rid of them immediately.
420 * However, that implies that we have to traverse the dentry
421 * tree upwards to the parents which might _also_ now be
422 * scheduled for deletion (it may have been only waiting for
423 * its last child to go away).
425 * This tail recursion is done by hand as we don't want to depend
426 * on the compiler to always get this right (gcc generally doesn't).
427 * Real recursion would eat up our stack space.
431 * dput - release a dentry
432 * @dentry: dentry to release
434 * Release a dentry. This will drop the usage count and if appropriate
435 * call the dentry unlink method as well as removing it from the queues and
436 * releasing its resources. If the parent dentries were scheduled for release
437 * they too may now get deleted.
439 void dput(struct dentry
*dentry
)
445 if (dentry
->d_count
== 1)
447 spin_lock(&dentry
->d_lock
);
448 BUG_ON(!dentry
->d_count
);
449 if (dentry
->d_count
> 1) {
451 spin_unlock(&dentry
->d_lock
);
455 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
456 if (dentry
->d_op
->d_delete(dentry
))
460 /* Unreachable? Get rid of it */
461 if (d_unhashed(dentry
))
465 * If this dentry needs lookup, don't set the referenced flag so that it
466 * is more likely to be cleaned up by the dcache shrinker in case of
469 if (!d_need_lookup(dentry
))
470 dentry
->d_flags
|= DCACHE_REFERENCED
;
471 dentry_lru_add(dentry
);
474 spin_unlock(&dentry
->d_lock
);
478 dentry
= dentry_kill(dentry
, 1);
485 * d_invalidate - invalidate a dentry
486 * @dentry: dentry to invalidate
488 * Try to invalidate the dentry if it turns out to be
489 * possible. If there are other dentries that can be
490 * reached through this one we can't delete it and we
491 * return -EBUSY. On success we return 0.
496 int d_invalidate(struct dentry
* dentry
)
499 * If it's already been dropped, return OK.
501 spin_lock(&dentry
->d_lock
);
502 if (d_unhashed(dentry
)) {
503 spin_unlock(&dentry
->d_lock
);
507 * Check whether to do a partial shrink_dcache
508 * to get rid of unused child entries.
510 if (!list_empty(&dentry
->d_subdirs
)) {
511 spin_unlock(&dentry
->d_lock
);
512 shrink_dcache_parent(dentry
);
513 spin_lock(&dentry
->d_lock
);
517 * Somebody else still using it?
519 * If it's a directory, we can't drop it
520 * for fear of somebody re-populating it
521 * with children (even though dropping it
522 * would make it unreachable from the root,
523 * we might still populate it if it was a
524 * working directory or similar).
526 if (dentry
->d_count
> 1) {
527 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
528 spin_unlock(&dentry
->d_lock
);
534 spin_unlock(&dentry
->d_lock
);
537 EXPORT_SYMBOL(d_invalidate
);
539 /* This must be called with d_lock held */
540 static inline void __dget_dlock(struct dentry
*dentry
)
545 static inline void __dget(struct dentry
*dentry
)
547 spin_lock(&dentry
->d_lock
);
548 __dget_dlock(dentry
);
549 spin_unlock(&dentry
->d_lock
);
552 struct dentry
*dget_parent(struct dentry
*dentry
)
558 * Don't need rcu_dereference because we re-check it was correct under
562 ret
= dentry
->d_parent
;
563 spin_lock(&ret
->d_lock
);
564 if (unlikely(ret
!= dentry
->d_parent
)) {
565 spin_unlock(&ret
->d_lock
);
570 BUG_ON(!ret
->d_count
);
572 spin_unlock(&ret
->d_lock
);
575 EXPORT_SYMBOL(dget_parent
);
578 * d_find_alias - grab a hashed alias of inode
579 * @inode: inode in question
580 * @want_discon: flag, used by d_splice_alias, to request
581 * that only a DISCONNECTED alias be returned.
583 * If inode has a hashed alias, or is a directory and has any alias,
584 * acquire the reference to alias and return it. Otherwise return NULL.
585 * Notice that if inode is a directory there can be only one alias and
586 * it can be unhashed only if it has no children, or if it is the root
589 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
590 * any other hashed alias over that one unless @want_discon is set,
591 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
593 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
595 struct dentry
*alias
, *discon_alias
;
599 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
600 spin_lock(&alias
->d_lock
);
601 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
602 if (IS_ROOT(alias
) &&
603 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
604 discon_alias
= alias
;
605 } else if (!want_discon
) {
607 spin_unlock(&alias
->d_lock
);
611 spin_unlock(&alias
->d_lock
);
614 alias
= discon_alias
;
615 spin_lock(&alias
->d_lock
);
616 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
617 if (IS_ROOT(alias
) &&
618 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
620 spin_unlock(&alias
->d_lock
);
624 spin_unlock(&alias
->d_lock
);
630 struct dentry
*d_find_alias(struct inode
*inode
)
632 struct dentry
*de
= NULL
;
634 if (!list_empty(&inode
->i_dentry
)) {
635 spin_lock(&inode
->i_lock
);
636 de
= __d_find_alias(inode
, 0);
637 spin_unlock(&inode
->i_lock
);
641 EXPORT_SYMBOL(d_find_alias
);
644 * Try to kill dentries associated with this inode.
645 * WARNING: you must own a reference to inode.
647 void d_prune_aliases(struct inode
*inode
)
649 struct dentry
*dentry
;
651 spin_lock(&inode
->i_lock
);
652 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
653 spin_lock(&dentry
->d_lock
);
654 if (!dentry
->d_count
) {
655 __dget_dlock(dentry
);
657 spin_unlock(&dentry
->d_lock
);
658 spin_unlock(&inode
->i_lock
);
662 spin_unlock(&dentry
->d_lock
);
664 spin_unlock(&inode
->i_lock
);
666 EXPORT_SYMBOL(d_prune_aliases
);
669 * Try to throw away a dentry - free the inode, dput the parent.
670 * Requires dentry->d_lock is held, and dentry->d_count == 0.
671 * Releases dentry->d_lock.
673 * This may fail if locks cannot be acquired no problem, just try again.
675 static void try_prune_one_dentry(struct dentry
*dentry
)
676 __releases(dentry
->d_lock
)
678 struct dentry
*parent
;
680 parent
= dentry_kill(dentry
, 0);
682 * If dentry_kill returns NULL, we have nothing more to do.
683 * if it returns the same dentry, trylocks failed. In either
684 * case, just loop again.
686 * Otherwise, we need to prune ancestors too. This is necessary
687 * to prevent quadratic behavior of shrink_dcache_parent(), but
688 * is also expected to be beneficial in reducing dentry cache
693 if (parent
== dentry
)
696 /* Prune ancestors. */
699 spin_lock(&dentry
->d_lock
);
700 if (dentry
->d_count
> 1) {
702 spin_unlock(&dentry
->d_lock
);
705 dentry
= dentry_kill(dentry
, 1);
709 static void shrink_dentry_list(struct list_head
*list
)
711 struct dentry
*dentry
;
715 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
716 if (&dentry
->d_lru
== list
)
718 spin_lock(&dentry
->d_lock
);
719 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
720 spin_unlock(&dentry
->d_lock
);
725 * We found an inuse dentry which was not removed from
726 * the LRU because of laziness during lookup. Do not free
727 * it - just keep it off the LRU list.
729 if (dentry
->d_count
) {
730 dentry_lru_del(dentry
);
731 spin_unlock(&dentry
->d_lock
);
737 try_prune_one_dentry(dentry
);
745 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
746 * @sb: superblock to shrink dentry LRU.
747 * @count: number of entries to prune
748 * @flags: flags to control the dentry processing
750 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
752 static void __shrink_dcache_sb(struct super_block
*sb
, int count
, int flags
)
754 struct dentry
*dentry
;
755 LIST_HEAD(referenced
);
759 spin_lock(&dcache_lru_lock
);
760 while (!list_empty(&sb
->s_dentry_lru
)) {
761 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
762 struct dentry
, d_lru
);
763 BUG_ON(dentry
->d_sb
!= sb
);
765 if (!spin_trylock(&dentry
->d_lock
)) {
766 spin_unlock(&dcache_lru_lock
);
772 * If we are honouring the DCACHE_REFERENCED flag and the
773 * dentry has this flag set, don't free it. Clear the flag
774 * and put it back on the LRU.
776 if (flags
& DCACHE_REFERENCED
&&
777 dentry
->d_flags
& DCACHE_REFERENCED
) {
778 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
779 list_move(&dentry
->d_lru
, &referenced
);
780 spin_unlock(&dentry
->d_lock
);
782 list_move_tail(&dentry
->d_lru
, &tmp
);
783 spin_unlock(&dentry
->d_lock
);
787 cond_resched_lock(&dcache_lru_lock
);
789 if (!list_empty(&referenced
))
790 list_splice(&referenced
, &sb
->s_dentry_lru
);
791 spin_unlock(&dcache_lru_lock
);
793 shrink_dentry_list(&tmp
);
797 * prune_dcache_sb - shrink the dcache
798 * @nr_to_scan: number of entries to try to free
800 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
801 * done when we need more memory an called from the superblock shrinker
804 * This function may fail to free any resources if all the dentries are in
807 void prune_dcache_sb(struct super_block
*sb
, int nr_to_scan
)
809 __shrink_dcache_sb(sb
, nr_to_scan
, DCACHE_REFERENCED
);
813 * shrink_dcache_sb - shrink dcache for a superblock
816 * Shrink the dcache for the specified super block. This is used to free
817 * the dcache before unmounting a file system.
819 void shrink_dcache_sb(struct super_block
*sb
)
823 spin_lock(&dcache_lru_lock
);
824 while (!list_empty(&sb
->s_dentry_lru
)) {
825 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
826 spin_unlock(&dcache_lru_lock
);
827 shrink_dentry_list(&tmp
);
828 spin_lock(&dcache_lru_lock
);
830 spin_unlock(&dcache_lru_lock
);
832 EXPORT_SYMBOL(shrink_dcache_sb
);
835 * destroy a single subtree of dentries for unmount
836 * - see the comments on shrink_dcache_for_umount() for a description of the
839 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
841 struct dentry
*parent
;
843 BUG_ON(!IS_ROOT(dentry
));
846 /* descend to the first leaf in the current subtree */
847 while (!list_empty(&dentry
->d_subdirs
))
848 dentry
= list_entry(dentry
->d_subdirs
.next
,
849 struct dentry
, d_u
.d_child
);
851 /* consume the dentries from this leaf up through its parents
852 * until we find one with children or run out altogether */
856 /* detach from the system */
857 dentry_lru_del(dentry
);
860 if (dentry
->d_count
!= 0) {
862 "BUG: Dentry %p{i=%lx,n=%s}"
864 " [unmount of %s %s]\n",
867 dentry
->d_inode
->i_ino
: 0UL,
870 dentry
->d_sb
->s_type
->name
,
875 if (IS_ROOT(dentry
)) {
877 list_del(&dentry
->d_u
.d_child
);
879 parent
= dentry
->d_parent
;
881 list_del(&dentry
->d_u
.d_child
);
884 inode
= dentry
->d_inode
;
886 dentry
->d_inode
= NULL
;
887 list_del_init(&dentry
->d_alias
);
888 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
889 dentry
->d_op
->d_iput(dentry
, inode
);
896 /* finished when we fall off the top of the tree,
897 * otherwise we ascend to the parent and move to the
898 * next sibling if there is one */
902 } while (list_empty(&dentry
->d_subdirs
));
904 dentry
= list_entry(dentry
->d_subdirs
.next
,
905 struct dentry
, d_u
.d_child
);
910 * destroy the dentries attached to a superblock on unmounting
911 * - we don't need to use dentry->d_lock because:
912 * - the superblock is detached from all mountings and open files, so the
913 * dentry trees will not be rearranged by the VFS
914 * - s_umount is write-locked, so the memory pressure shrinker will ignore
915 * any dentries belonging to this superblock that it comes across
916 * - the filesystem itself is no longer permitted to rearrange the dentries
919 void shrink_dcache_for_umount(struct super_block
*sb
)
921 struct dentry
*dentry
;
923 if (down_read_trylock(&sb
->s_umount
))
929 shrink_dcache_for_umount_subtree(dentry
);
931 while (!hlist_bl_empty(&sb
->s_anon
)) {
932 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
933 shrink_dcache_for_umount_subtree(dentry
);
938 * This tries to ascend one level of parenthood, but
939 * we can race with renaming, so we need to re-check
940 * the parenthood after dropping the lock and check
941 * that the sequence number still matches.
943 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
945 struct dentry
*new = old
->d_parent
;
948 spin_unlock(&old
->d_lock
);
949 spin_lock(&new->d_lock
);
952 * might go back up the wrong parent if we have had a rename
955 if (new != old
->d_parent
||
956 (old
->d_flags
& DCACHE_DISCONNECTED
) ||
957 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
958 spin_unlock(&new->d_lock
);
967 * Search for at least 1 mount point in the dentry's subdirs.
968 * We descend to the next level whenever the d_subdirs
969 * list is non-empty and continue searching.
973 * have_submounts - check for mounts over a dentry
974 * @parent: dentry to check.
976 * Return true if the parent or its subdirectories contain
979 int have_submounts(struct dentry
*parent
)
981 struct dentry
*this_parent
;
982 struct list_head
*next
;
986 seq
= read_seqbegin(&rename_lock
);
988 this_parent
= parent
;
990 if (d_mountpoint(parent
))
992 spin_lock(&this_parent
->d_lock
);
994 next
= this_parent
->d_subdirs
.next
;
996 while (next
!= &this_parent
->d_subdirs
) {
997 struct list_head
*tmp
= next
;
998 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1001 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1002 /* Have we found a mount point ? */
1003 if (d_mountpoint(dentry
)) {
1004 spin_unlock(&dentry
->d_lock
);
1005 spin_unlock(&this_parent
->d_lock
);
1008 if (!list_empty(&dentry
->d_subdirs
)) {
1009 spin_unlock(&this_parent
->d_lock
);
1010 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1011 this_parent
= dentry
;
1012 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1015 spin_unlock(&dentry
->d_lock
);
1018 * All done at this level ... ascend and resume the search.
1020 if (this_parent
!= parent
) {
1021 struct dentry
*child
= this_parent
;
1022 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1025 next
= child
->d_u
.d_child
.next
;
1028 spin_unlock(&this_parent
->d_lock
);
1029 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1032 write_sequnlock(&rename_lock
);
1033 return 0; /* No mount points found in tree */
1035 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1038 write_sequnlock(&rename_lock
);
1043 write_seqlock(&rename_lock
);
1046 EXPORT_SYMBOL(have_submounts
);
1049 * Search the dentry child list for the specified parent,
1050 * and move any unused dentries to the end of the unused
1051 * list for prune_dcache(). We descend to the next level
1052 * whenever the d_subdirs list is non-empty and continue
1055 * It returns zero iff there are no unused children,
1056 * otherwise it returns the number of children moved to
1057 * the end of the unused list. This may not be the total
1058 * number of unused children, because select_parent can
1059 * drop the lock and return early due to latency
1062 static int select_parent(struct dentry
* parent
)
1064 struct dentry
*this_parent
;
1065 struct list_head
*next
;
1070 seq
= read_seqbegin(&rename_lock
);
1072 this_parent
= parent
;
1073 spin_lock(&this_parent
->d_lock
);
1075 next
= this_parent
->d_subdirs
.next
;
1077 while (next
!= &this_parent
->d_subdirs
) {
1078 struct list_head
*tmp
= next
;
1079 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1082 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1085 * move only zero ref count dentries to the end
1086 * of the unused list for prune_dcache
1088 if (!dentry
->d_count
) {
1089 dentry_lru_move_tail(dentry
);
1092 dentry_lru_del(dentry
);
1096 * We can return to the caller if we have found some (this
1097 * ensures forward progress). We'll be coming back to find
1100 if (found
&& need_resched()) {
1101 spin_unlock(&dentry
->d_lock
);
1106 * Descend a level if the d_subdirs list is non-empty.
1108 if (!list_empty(&dentry
->d_subdirs
)) {
1109 spin_unlock(&this_parent
->d_lock
);
1110 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1111 this_parent
= dentry
;
1112 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1116 spin_unlock(&dentry
->d_lock
);
1119 * All done at this level ... ascend and resume the search.
1121 if (this_parent
!= parent
) {
1122 struct dentry
*child
= this_parent
;
1123 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1126 next
= child
->d_u
.d_child
.next
;
1130 spin_unlock(&this_parent
->d_lock
);
1131 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1134 write_sequnlock(&rename_lock
);
1141 write_seqlock(&rename_lock
);
1146 * shrink_dcache_parent - prune dcache
1147 * @parent: parent of entries to prune
1149 * Prune the dcache to remove unused children of the parent dentry.
1152 void shrink_dcache_parent(struct dentry
* parent
)
1154 struct super_block
*sb
= parent
->d_sb
;
1157 while ((found
= select_parent(parent
)) != 0)
1158 __shrink_dcache_sb(sb
, found
, 0);
1160 EXPORT_SYMBOL(shrink_dcache_parent
);
1163 * __d_alloc - allocate a dcache entry
1164 * @sb: filesystem it will belong to
1165 * @name: qstr of the name
1167 * Allocates a dentry. It returns %NULL if there is insufficient memory
1168 * available. On a success the dentry is returned. The name passed in is
1169 * copied and the copy passed in may be reused after this call.
1172 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1174 struct dentry
*dentry
;
1177 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1181 if (name
->len
> DNAME_INLINE_LEN
-1) {
1182 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1184 kmem_cache_free(dentry_cache
, dentry
);
1188 dname
= dentry
->d_iname
;
1190 dentry
->d_name
.name
= dname
;
1192 dentry
->d_name
.len
= name
->len
;
1193 dentry
->d_name
.hash
= name
->hash
;
1194 memcpy(dname
, name
->name
, name
->len
);
1195 dname
[name
->len
] = 0;
1197 dentry
->d_count
= 1;
1198 dentry
->d_flags
= 0;
1199 spin_lock_init(&dentry
->d_lock
);
1200 seqcount_init(&dentry
->d_seq
);
1201 dentry
->d_inode
= NULL
;
1202 dentry
->d_parent
= dentry
;
1204 dentry
->d_op
= NULL
;
1205 dentry
->d_fsdata
= NULL
;
1206 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1207 INIT_LIST_HEAD(&dentry
->d_lru
);
1208 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1209 INIT_LIST_HEAD(&dentry
->d_alias
);
1210 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1211 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1213 this_cpu_inc(nr_dentry
);
1219 * d_alloc - allocate a dcache entry
1220 * @parent: parent of entry to allocate
1221 * @name: qstr of the name
1223 * Allocates a dentry. It returns %NULL if there is insufficient memory
1224 * available. On a success the dentry is returned. The name passed in is
1225 * copied and the copy passed in may be reused after this call.
1227 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1229 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1233 spin_lock(&parent
->d_lock
);
1235 * don't need child lock because it is not subject
1236 * to concurrency here
1238 __dget_dlock(parent
);
1239 dentry
->d_parent
= parent
;
1240 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1241 spin_unlock(&parent
->d_lock
);
1245 EXPORT_SYMBOL(d_alloc
);
1247 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1249 struct dentry
*dentry
= __d_alloc(sb
, name
);
1251 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1254 EXPORT_SYMBOL(d_alloc_pseudo
);
1256 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1261 q
.len
= strlen(name
);
1262 q
.hash
= full_name_hash(q
.name
, q
.len
);
1263 return d_alloc(parent
, &q
);
1265 EXPORT_SYMBOL(d_alloc_name
);
1267 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1269 WARN_ON_ONCE(dentry
->d_op
);
1270 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1272 DCACHE_OP_REVALIDATE
|
1273 DCACHE_OP_DELETE
));
1278 dentry
->d_flags
|= DCACHE_OP_HASH
;
1280 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1281 if (op
->d_revalidate
)
1282 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1284 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1287 EXPORT_SYMBOL(d_set_d_op
);
1289 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1291 spin_lock(&dentry
->d_lock
);
1293 if (unlikely(IS_AUTOMOUNT(inode
)))
1294 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1295 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1297 dentry
->d_inode
= inode
;
1298 dentry_rcuwalk_barrier(dentry
);
1299 spin_unlock(&dentry
->d_lock
);
1300 fsnotify_d_instantiate(dentry
, inode
);
1304 * d_instantiate - fill in inode information for a dentry
1305 * @entry: dentry to complete
1306 * @inode: inode to attach to this dentry
1308 * Fill in inode information in the entry.
1310 * This turns negative dentries into productive full members
1313 * NOTE! This assumes that the inode count has been incremented
1314 * (or otherwise set) by the caller to indicate that it is now
1315 * in use by the dcache.
1318 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1320 BUG_ON(!list_empty(&entry
->d_alias
));
1322 spin_lock(&inode
->i_lock
);
1323 __d_instantiate(entry
, inode
);
1325 spin_unlock(&inode
->i_lock
);
1326 security_d_instantiate(entry
, inode
);
1328 EXPORT_SYMBOL(d_instantiate
);
1331 * d_instantiate_unique - instantiate a non-aliased dentry
1332 * @entry: dentry to instantiate
1333 * @inode: inode to attach to this dentry
1335 * Fill in inode information in the entry. On success, it returns NULL.
1336 * If an unhashed alias of "entry" already exists, then we return the
1337 * aliased dentry instead and drop one reference to inode.
1339 * Note that in order to avoid conflicts with rename() etc, the caller
1340 * had better be holding the parent directory semaphore.
1342 * This also assumes that the inode count has been incremented
1343 * (or otherwise set) by the caller to indicate that it is now
1344 * in use by the dcache.
1346 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1347 struct inode
*inode
)
1349 struct dentry
*alias
;
1350 int len
= entry
->d_name
.len
;
1351 const char *name
= entry
->d_name
.name
;
1352 unsigned int hash
= entry
->d_name
.hash
;
1355 __d_instantiate(entry
, NULL
);
1359 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1360 struct qstr
*qstr
= &alias
->d_name
;
1363 * Don't need alias->d_lock here, because aliases with
1364 * d_parent == entry->d_parent are not subject to name or
1365 * parent changes, because the parent inode i_mutex is held.
1367 if (qstr
->hash
!= hash
)
1369 if (alias
->d_parent
!= entry
->d_parent
)
1371 if (dentry_cmp(qstr
->name
, qstr
->len
, name
, len
))
1377 __d_instantiate(entry
, inode
);
1381 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1383 struct dentry
*result
;
1385 BUG_ON(!list_empty(&entry
->d_alias
));
1388 spin_lock(&inode
->i_lock
);
1389 result
= __d_instantiate_unique(entry
, inode
);
1391 spin_unlock(&inode
->i_lock
);
1394 security_d_instantiate(entry
, inode
);
1398 BUG_ON(!d_unhashed(result
));
1403 EXPORT_SYMBOL(d_instantiate_unique
);
1406 * d_alloc_root - allocate root dentry
1407 * @root_inode: inode to allocate the root for
1409 * Allocate a root ("/") dentry for the inode given. The inode is
1410 * instantiated and returned. %NULL is returned if there is insufficient
1411 * memory or the inode passed is %NULL.
1414 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1416 struct dentry
*res
= NULL
;
1419 static const struct qstr name
= { .name
= "/", .len
= 1 };
1421 res
= __d_alloc(root_inode
->i_sb
, &name
);
1423 d_instantiate(res
, root_inode
);
1427 EXPORT_SYMBOL(d_alloc_root
);
1429 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1431 struct dentry
*alias
;
1433 if (list_empty(&inode
->i_dentry
))
1435 alias
= list_first_entry(&inode
->i_dentry
, struct dentry
, d_alias
);
1440 static struct dentry
* d_find_any_alias(struct inode
*inode
)
1444 spin_lock(&inode
->i_lock
);
1445 de
= __d_find_any_alias(inode
);
1446 spin_unlock(&inode
->i_lock
);
1452 * d_obtain_alias - find or allocate a dentry for a given inode
1453 * @inode: inode to allocate the dentry for
1455 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1456 * similar open by handle operations. The returned dentry may be anonymous,
1457 * or may have a full name (if the inode was already in the cache).
1459 * When called on a directory inode, we must ensure that the inode only ever
1460 * has one dentry. If a dentry is found, that is returned instead of
1461 * allocating a new one.
1463 * On successful return, the reference to the inode has been transferred
1464 * to the dentry. In case of an error the reference on the inode is released.
1465 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1466 * be passed in and will be the error will be propagate to the return value,
1467 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1469 struct dentry
*d_obtain_alias(struct inode
*inode
)
1471 static const struct qstr anonstring
= { .name
= "" };
1476 return ERR_PTR(-ESTALE
);
1478 return ERR_CAST(inode
);
1480 res
= d_find_any_alias(inode
);
1484 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1486 res
= ERR_PTR(-ENOMEM
);
1490 spin_lock(&inode
->i_lock
);
1491 res
= __d_find_any_alias(inode
);
1493 spin_unlock(&inode
->i_lock
);
1498 /* attach a disconnected dentry */
1499 spin_lock(&tmp
->d_lock
);
1500 tmp
->d_inode
= inode
;
1501 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1502 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1503 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1504 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1505 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1506 spin_unlock(&tmp
->d_lock
);
1507 spin_unlock(&inode
->i_lock
);
1508 security_d_instantiate(tmp
, inode
);
1513 if (res
&& !IS_ERR(res
))
1514 security_d_instantiate(res
, inode
);
1518 EXPORT_SYMBOL(d_obtain_alias
);
1521 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1522 * @inode: the inode which may have a disconnected dentry
1523 * @dentry: a negative dentry which we want to point to the inode.
1525 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1526 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1527 * and return it, else simply d_add the inode to the dentry and return NULL.
1529 * This is needed in the lookup routine of any filesystem that is exportable
1530 * (via knfsd) so that we can build dcache paths to directories effectively.
1532 * If a dentry was found and moved, then it is returned. Otherwise NULL
1533 * is returned. This matches the expected return value of ->lookup.
1536 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1538 struct dentry
*new = NULL
;
1541 return ERR_CAST(inode
);
1543 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1544 spin_lock(&inode
->i_lock
);
1545 new = __d_find_alias(inode
, 1);
1547 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1548 spin_unlock(&inode
->i_lock
);
1549 security_d_instantiate(new, inode
);
1550 d_move(new, dentry
);
1553 /* already taking inode->i_lock, so d_add() by hand */
1554 __d_instantiate(dentry
, inode
);
1555 spin_unlock(&inode
->i_lock
);
1556 security_d_instantiate(dentry
, inode
);
1560 d_add(dentry
, inode
);
1563 EXPORT_SYMBOL(d_splice_alias
);
1566 * d_add_ci - lookup or allocate new dentry with case-exact name
1567 * @inode: the inode case-insensitive lookup has found
1568 * @dentry: the negative dentry that was passed to the parent's lookup func
1569 * @name: the case-exact name to be associated with the returned dentry
1571 * This is to avoid filling the dcache with case-insensitive names to the
1572 * same inode, only the actual correct case is stored in the dcache for
1573 * case-insensitive filesystems.
1575 * For a case-insensitive lookup match and if the the case-exact dentry
1576 * already exists in in the dcache, use it and return it.
1578 * If no entry exists with the exact case name, allocate new dentry with
1579 * the exact case, and return the spliced entry.
1581 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1585 struct dentry
*found
;
1589 * First check if a dentry matching the name already exists,
1590 * if not go ahead and create it now.
1592 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1594 new = d_alloc(dentry
->d_parent
, name
);
1600 found
= d_splice_alias(inode
, new);
1609 * If a matching dentry exists, and it's not negative use it.
1611 * Decrement the reference count to balance the iget() done
1614 if (found
->d_inode
) {
1615 if (unlikely(found
->d_inode
!= inode
)) {
1616 /* This can't happen because bad inodes are unhashed. */
1617 BUG_ON(!is_bad_inode(inode
));
1618 BUG_ON(!is_bad_inode(found
->d_inode
));
1625 * We are going to instantiate this dentry, unhash it and clear the
1626 * lookup flag so we can do that.
1628 if (unlikely(d_need_lookup(found
)))
1629 d_clear_need_lookup(found
);
1632 * Negative dentry: instantiate it unless the inode is a directory and
1633 * already has a dentry.
1635 new = d_splice_alias(inode
, found
);
1644 return ERR_PTR(error
);
1646 EXPORT_SYMBOL(d_add_ci
);
1649 * __d_lookup_rcu - search for a dentry (racy, store-free)
1650 * @parent: parent dentry
1651 * @name: qstr of name we wish to find
1652 * @seq: returns d_seq value at the point where the dentry was found
1653 * @inode: returns dentry->d_inode when the inode was found valid.
1654 * Returns: dentry, or NULL
1656 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1657 * resolution (store-free path walking) design described in
1658 * Documentation/filesystems/path-lookup.txt.
1660 * This is not to be used outside core vfs.
1662 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1663 * held, and rcu_read_lock held. The returned dentry must not be stored into
1664 * without taking d_lock and checking d_seq sequence count against @seq
1667 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1670 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1671 * the returned dentry, so long as its parent's seqlock is checked after the
1672 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1673 * is formed, giving integrity down the path walk.
1675 struct dentry
*__d_lookup_rcu(struct dentry
*parent
, struct qstr
*name
,
1676 unsigned *seq
, struct inode
**inode
)
1678 unsigned int len
= name
->len
;
1679 unsigned int hash
= name
->hash
;
1680 const unsigned char *str
= name
->name
;
1681 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1682 struct hlist_bl_node
*node
;
1683 struct dentry
*dentry
;
1686 * Note: There is significant duplication with __d_lookup_rcu which is
1687 * required to prevent single threaded performance regressions
1688 * especially on architectures where smp_rmb (in seqcounts) are costly.
1689 * Keep the two functions in sync.
1693 * The hash list is protected using RCU.
1695 * Carefully use d_seq when comparing a candidate dentry, to avoid
1696 * races with d_move().
1698 * It is possible that concurrent renames can mess up our list
1699 * walk here and result in missing our dentry, resulting in the
1700 * false-negative result. d_lookup() protects against concurrent
1701 * renames using rename_lock seqlock.
1703 * See Documentation/filesystems/path-lookup.txt for more details.
1705 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1710 if (dentry
->d_name
.hash
!= hash
)
1714 *seq
= read_seqcount_begin(&dentry
->d_seq
);
1715 if (dentry
->d_parent
!= parent
)
1717 if (d_unhashed(dentry
))
1719 tlen
= dentry
->d_name
.len
;
1720 tname
= dentry
->d_name
.name
;
1721 i
= dentry
->d_inode
;
1724 * This seqcount check is required to ensure name and
1725 * len are loaded atomically, so as not to walk off the
1726 * edge of memory when walking. If we could load this
1727 * atomically some other way, we could drop this check.
1729 if (read_seqcount_retry(&dentry
->d_seq
, *seq
))
1731 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1732 if (parent
->d_op
->d_compare(parent
, *inode
,
1737 if (dentry_cmp(tname
, tlen
, str
, len
))
1741 * No extra seqcount check is required after the name
1742 * compare. The caller must perform a seqcount check in
1743 * order to do anything useful with the returned dentry
1753 * d_lookup - search for a dentry
1754 * @parent: parent dentry
1755 * @name: qstr of name we wish to find
1756 * Returns: dentry, or NULL
1758 * d_lookup searches the children of the parent dentry for the name in
1759 * question. If the dentry is found its reference count is incremented and the
1760 * dentry is returned. The caller must use dput to free the entry when it has
1761 * finished using it. %NULL is returned if the dentry does not exist.
1763 struct dentry
*d_lookup(struct dentry
*parent
, struct qstr
*name
)
1765 struct dentry
*dentry
;
1769 seq
= read_seqbegin(&rename_lock
);
1770 dentry
= __d_lookup(parent
, name
);
1773 } while (read_seqretry(&rename_lock
, seq
));
1776 EXPORT_SYMBOL(d_lookup
);
1779 * __d_lookup - search for a dentry (racy)
1780 * @parent: parent dentry
1781 * @name: qstr of name we wish to find
1782 * Returns: dentry, or NULL
1784 * __d_lookup is like d_lookup, however it may (rarely) return a
1785 * false-negative result due to unrelated rename activity.
1787 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1788 * however it must be used carefully, eg. with a following d_lookup in
1789 * the case of failure.
1791 * __d_lookup callers must be commented.
1793 struct dentry
*__d_lookup(struct dentry
*parent
, struct qstr
*name
)
1795 unsigned int len
= name
->len
;
1796 unsigned int hash
= name
->hash
;
1797 const unsigned char *str
= name
->name
;
1798 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1799 struct hlist_bl_node
*node
;
1800 struct dentry
*found
= NULL
;
1801 struct dentry
*dentry
;
1804 * Note: There is significant duplication with __d_lookup_rcu which is
1805 * required to prevent single threaded performance regressions
1806 * especially on architectures where smp_rmb (in seqcounts) are costly.
1807 * Keep the two functions in sync.
1811 * The hash list is protected using RCU.
1813 * Take d_lock when comparing a candidate dentry, to avoid races
1816 * It is possible that concurrent renames can mess up our list
1817 * walk here and result in missing our dentry, resulting in the
1818 * false-negative result. d_lookup() protects against concurrent
1819 * renames using rename_lock seqlock.
1821 * See Documentation/filesystems/path-lookup.txt for more details.
1825 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1829 if (dentry
->d_name
.hash
!= hash
)
1832 spin_lock(&dentry
->d_lock
);
1833 if (dentry
->d_parent
!= parent
)
1835 if (d_unhashed(dentry
))
1839 * It is safe to compare names since d_move() cannot
1840 * change the qstr (protected by d_lock).
1842 tlen
= dentry
->d_name
.len
;
1843 tname
= dentry
->d_name
.name
;
1844 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1845 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1846 dentry
, dentry
->d_inode
,
1850 if (dentry_cmp(tname
, tlen
, str
, len
))
1856 spin_unlock(&dentry
->d_lock
);
1859 spin_unlock(&dentry
->d_lock
);
1867 * d_hash_and_lookup - hash the qstr then search for a dentry
1868 * @dir: Directory to search in
1869 * @name: qstr of name we wish to find
1871 * On hash failure or on lookup failure NULL is returned.
1873 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1875 struct dentry
*dentry
= NULL
;
1878 * Check for a fs-specific hash function. Note that we must
1879 * calculate the standard hash first, as the d_op->d_hash()
1880 * routine may choose to leave the hash value unchanged.
1882 name
->hash
= full_name_hash(name
->name
, name
->len
);
1883 if (dir
->d_flags
& DCACHE_OP_HASH
) {
1884 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1887 dentry
= d_lookup(dir
, name
);
1893 * d_validate - verify dentry provided from insecure source (deprecated)
1894 * @dentry: The dentry alleged to be valid child of @dparent
1895 * @dparent: The parent dentry (known to be valid)
1897 * An insecure source has sent us a dentry, here we verify it and dget() it.
1898 * This is used by ncpfs in its readdir implementation.
1899 * Zero is returned in the dentry is invalid.
1901 * This function is slow for big directories, and deprecated, do not use it.
1903 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1905 struct dentry
*child
;
1907 spin_lock(&dparent
->d_lock
);
1908 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1909 if (dentry
== child
) {
1910 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1911 __dget_dlock(dentry
);
1912 spin_unlock(&dentry
->d_lock
);
1913 spin_unlock(&dparent
->d_lock
);
1917 spin_unlock(&dparent
->d_lock
);
1921 EXPORT_SYMBOL(d_validate
);
1924 * When a file is deleted, we have two options:
1925 * - turn this dentry into a negative dentry
1926 * - unhash this dentry and free it.
1928 * Usually, we want to just turn this into
1929 * a negative dentry, but if anybody else is
1930 * currently using the dentry or the inode
1931 * we can't do that and we fall back on removing
1932 * it from the hash queues and waiting for
1933 * it to be deleted later when it has no users
1937 * d_delete - delete a dentry
1938 * @dentry: The dentry to delete
1940 * Turn the dentry into a negative dentry if possible, otherwise
1941 * remove it from the hash queues so it can be deleted later
1944 void d_delete(struct dentry
* dentry
)
1946 struct inode
*inode
;
1949 * Are we the only user?
1952 spin_lock(&dentry
->d_lock
);
1953 inode
= dentry
->d_inode
;
1954 isdir
= S_ISDIR(inode
->i_mode
);
1955 if (dentry
->d_count
== 1) {
1956 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
1957 spin_unlock(&dentry
->d_lock
);
1961 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1962 dentry_unlink_inode(dentry
);
1963 fsnotify_nameremove(dentry
, isdir
);
1967 if (!d_unhashed(dentry
))
1970 spin_unlock(&dentry
->d_lock
);
1972 fsnotify_nameremove(dentry
, isdir
);
1974 EXPORT_SYMBOL(d_delete
);
1976 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
1978 BUG_ON(!d_unhashed(entry
));
1980 entry
->d_flags
|= DCACHE_RCUACCESS
;
1981 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
1985 static void _d_rehash(struct dentry
* entry
)
1987 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1991 * d_rehash - add an entry back to the hash
1992 * @entry: dentry to add to the hash
1994 * Adds a dentry to the hash according to its name.
1997 void d_rehash(struct dentry
* entry
)
1999 spin_lock(&entry
->d_lock
);
2001 spin_unlock(&entry
->d_lock
);
2003 EXPORT_SYMBOL(d_rehash
);
2006 * dentry_update_name_case - update case insensitive dentry with a new name
2007 * @dentry: dentry to be updated
2010 * Update a case insensitive dentry with new case of name.
2012 * dentry must have been returned by d_lookup with name @name. Old and new
2013 * name lengths must match (ie. no d_compare which allows mismatched name
2016 * Parent inode i_mutex must be held over d_lookup and into this call (to
2017 * keep renames and concurrent inserts, and readdir(2) away).
2019 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2021 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2022 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2024 spin_lock(&dentry
->d_lock
);
2025 write_seqcount_begin(&dentry
->d_seq
);
2026 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2027 write_seqcount_end(&dentry
->d_seq
);
2028 spin_unlock(&dentry
->d_lock
);
2030 EXPORT_SYMBOL(dentry_update_name_case
);
2032 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2034 if (dname_external(target
)) {
2035 if (dname_external(dentry
)) {
2037 * Both external: swap the pointers
2039 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2042 * dentry:internal, target:external. Steal target's
2043 * storage and make target internal.
2045 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2046 dentry
->d_name
.len
+ 1);
2047 dentry
->d_name
.name
= target
->d_name
.name
;
2048 target
->d_name
.name
= target
->d_iname
;
2051 if (dname_external(dentry
)) {
2053 * dentry:external, target:internal. Give dentry's
2054 * storage to target and make dentry internal
2056 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2057 target
->d_name
.len
+ 1);
2058 target
->d_name
.name
= dentry
->d_name
.name
;
2059 dentry
->d_name
.name
= dentry
->d_iname
;
2062 * Both are internal. Just copy target to dentry
2064 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2065 target
->d_name
.len
+ 1);
2066 dentry
->d_name
.len
= target
->d_name
.len
;
2070 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2073 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2076 * XXXX: do we really need to take target->d_lock?
2078 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2079 spin_lock(&target
->d_parent
->d_lock
);
2081 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2082 spin_lock(&dentry
->d_parent
->d_lock
);
2083 spin_lock_nested(&target
->d_parent
->d_lock
,
2084 DENTRY_D_LOCK_NESTED
);
2086 spin_lock(&target
->d_parent
->d_lock
);
2087 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2088 DENTRY_D_LOCK_NESTED
);
2091 if (target
< dentry
) {
2092 spin_lock_nested(&target
->d_lock
, 2);
2093 spin_lock_nested(&dentry
->d_lock
, 3);
2095 spin_lock_nested(&dentry
->d_lock
, 2);
2096 spin_lock_nested(&target
->d_lock
, 3);
2100 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2101 struct dentry
*target
)
2103 if (target
->d_parent
!= dentry
->d_parent
)
2104 spin_unlock(&dentry
->d_parent
->d_lock
);
2105 if (target
->d_parent
!= target
)
2106 spin_unlock(&target
->d_parent
->d_lock
);
2110 * When switching names, the actual string doesn't strictly have to
2111 * be preserved in the target - because we're dropping the target
2112 * anyway. As such, we can just do a simple memcpy() to copy over
2113 * the new name before we switch.
2115 * Note that we have to be a lot more careful about getting the hash
2116 * switched - we have to switch the hash value properly even if it
2117 * then no longer matches the actual (corrupted) string of the target.
2118 * The hash value has to match the hash queue that the dentry is on..
2121 * __d_move - move a dentry
2122 * @dentry: entry to move
2123 * @target: new dentry
2125 * Update the dcache to reflect the move of a file name. Negative
2126 * dcache entries should not be moved in this way. Caller must hold
2127 * rename_lock, the i_mutex of the source and target directories,
2128 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2130 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2132 if (!dentry
->d_inode
)
2133 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2135 BUG_ON(d_ancestor(dentry
, target
));
2136 BUG_ON(d_ancestor(target
, dentry
));
2138 dentry_lock_for_move(dentry
, target
);
2140 write_seqcount_begin(&dentry
->d_seq
);
2141 write_seqcount_begin(&target
->d_seq
);
2143 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2146 * Move the dentry to the target hash queue. Don't bother checking
2147 * for the same hash queue because of how unlikely it is.
2150 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2152 /* Unhash the target: dput() will then get rid of it */
2155 list_del(&dentry
->d_u
.d_child
);
2156 list_del(&target
->d_u
.d_child
);
2158 /* Switch the names.. */
2159 switch_names(dentry
, target
);
2160 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2162 /* ... and switch the parents */
2163 if (IS_ROOT(dentry
)) {
2164 dentry
->d_parent
= target
->d_parent
;
2165 target
->d_parent
= target
;
2166 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2168 swap(dentry
->d_parent
, target
->d_parent
);
2170 /* And add them back to the (new) parent lists */
2171 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2174 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2176 write_seqcount_end(&target
->d_seq
);
2177 write_seqcount_end(&dentry
->d_seq
);
2179 dentry_unlock_parents_for_move(dentry
, target
);
2180 spin_unlock(&target
->d_lock
);
2181 fsnotify_d_move(dentry
);
2182 spin_unlock(&dentry
->d_lock
);
2186 * d_move - move a dentry
2187 * @dentry: entry to move
2188 * @target: new dentry
2190 * Update the dcache to reflect the move of a file name. Negative
2191 * dcache entries should not be moved in this way. See the locking
2192 * requirements for __d_move.
2194 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2196 write_seqlock(&rename_lock
);
2197 __d_move(dentry
, target
);
2198 write_sequnlock(&rename_lock
);
2200 EXPORT_SYMBOL(d_move
);
2203 * d_ancestor - search for an ancestor
2204 * @p1: ancestor dentry
2207 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2208 * an ancestor of p2, else NULL.
2210 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2214 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2215 if (p
->d_parent
== p1
)
2222 * This helper attempts to cope with remotely renamed directories
2224 * It assumes that the caller is already holding
2225 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2227 * Note: If ever the locking in lock_rename() changes, then please
2228 * remember to update this too...
2230 static struct dentry
*__d_unalias(struct inode
*inode
,
2231 struct dentry
*dentry
, struct dentry
*alias
)
2233 struct mutex
*m1
= NULL
, *m2
= NULL
;
2236 /* If alias and dentry share a parent, then no extra locks required */
2237 if (alias
->d_parent
== dentry
->d_parent
)
2240 /* See lock_rename() */
2241 ret
= ERR_PTR(-EBUSY
);
2242 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2244 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2245 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2247 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2249 __d_move(alias
, dentry
);
2252 spin_unlock(&inode
->i_lock
);
2261 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2262 * named dentry in place of the dentry to be replaced.
2263 * returns with anon->d_lock held!
2265 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2267 struct dentry
*dparent
, *aparent
;
2269 dentry_lock_for_move(anon
, dentry
);
2271 write_seqcount_begin(&dentry
->d_seq
);
2272 write_seqcount_begin(&anon
->d_seq
);
2274 dparent
= dentry
->d_parent
;
2275 aparent
= anon
->d_parent
;
2277 switch_names(dentry
, anon
);
2278 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2280 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2281 list_del(&dentry
->d_u
.d_child
);
2282 if (!IS_ROOT(dentry
))
2283 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2285 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2287 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2288 list_del(&anon
->d_u
.d_child
);
2290 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2292 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2294 write_seqcount_end(&dentry
->d_seq
);
2295 write_seqcount_end(&anon
->d_seq
);
2297 dentry_unlock_parents_for_move(anon
, dentry
);
2298 spin_unlock(&dentry
->d_lock
);
2300 /* anon->d_lock still locked, returns locked */
2301 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2305 * d_materialise_unique - introduce an inode into the tree
2306 * @dentry: candidate dentry
2307 * @inode: inode to bind to the dentry, to which aliases may be attached
2309 * Introduces an dentry into the tree, substituting an extant disconnected
2310 * root directory alias in its place if there is one. Caller must hold the
2311 * i_mutex of the parent directory.
2313 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2315 struct dentry
*actual
;
2317 BUG_ON(!d_unhashed(dentry
));
2321 __d_instantiate(dentry
, NULL
);
2326 spin_lock(&inode
->i_lock
);
2328 if (S_ISDIR(inode
->i_mode
)) {
2329 struct dentry
*alias
;
2331 /* Does an aliased dentry already exist? */
2332 alias
= __d_find_alias(inode
, 0);
2335 write_seqlock(&rename_lock
);
2337 if (d_ancestor(alias
, dentry
)) {
2338 /* Check for loops */
2339 actual
= ERR_PTR(-ELOOP
);
2340 } else if (IS_ROOT(alias
)) {
2341 /* Is this an anonymous mountpoint that we
2342 * could splice into our tree? */
2343 __d_materialise_dentry(dentry
, alias
);
2344 write_sequnlock(&rename_lock
);
2348 /* Nope, but we must(!) avoid directory
2350 actual
= __d_unalias(inode
, dentry
, alias
);
2352 write_sequnlock(&rename_lock
);
2359 /* Add a unique reference */
2360 actual
= __d_instantiate_unique(dentry
, inode
);
2364 BUG_ON(!d_unhashed(actual
));
2366 spin_lock(&actual
->d_lock
);
2369 spin_unlock(&actual
->d_lock
);
2370 spin_unlock(&inode
->i_lock
);
2372 if (actual
== dentry
) {
2373 security_d_instantiate(dentry
, inode
);
2380 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2382 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2386 return -ENAMETOOLONG
;
2388 memcpy(*buffer
, str
, namelen
);
2392 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2394 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2398 * prepend_path - Prepend path string to a buffer
2399 * @path: the dentry/vfsmount to report
2400 * @root: root vfsmnt/dentry (may be modified by this function)
2401 * @buffer: pointer to the end of the buffer
2402 * @buflen: pointer to buffer length
2404 * Caller holds the rename_lock.
2406 * If path is not reachable from the supplied root, then the value of
2407 * root is changed (without modifying refcounts).
2409 static int prepend_path(const struct path
*path
, struct path
*root
,
2410 char **buffer
, int *buflen
)
2412 struct dentry
*dentry
= path
->dentry
;
2413 struct vfsmount
*vfsmnt
= path
->mnt
;
2417 br_read_lock(vfsmount_lock
);
2418 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2419 struct dentry
* parent
;
2421 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2423 if (vfsmnt
->mnt_parent
== vfsmnt
) {
2426 dentry
= vfsmnt
->mnt_mountpoint
;
2427 vfsmnt
= vfsmnt
->mnt_parent
;
2430 parent
= dentry
->d_parent
;
2432 spin_lock(&dentry
->d_lock
);
2433 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2434 spin_unlock(&dentry
->d_lock
);
2436 error
= prepend(buffer
, buflen
, "/", 1);
2445 if (!error
&& !slash
)
2446 error
= prepend(buffer
, buflen
, "/", 1);
2448 br_read_unlock(vfsmount_lock
);
2453 * Filesystems needing to implement special "root names"
2454 * should do so with ->d_dname()
2456 if (IS_ROOT(dentry
) &&
2457 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2458 WARN(1, "Root dentry has weird name <%.*s>\n",
2459 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2462 root
->dentry
= dentry
;
2467 * __d_path - return the path of a dentry
2468 * @path: the dentry/vfsmount to report
2469 * @root: root vfsmnt/dentry (may be modified by this function)
2470 * @buf: buffer to return value in
2471 * @buflen: buffer length
2473 * Convert a dentry into an ASCII path name.
2475 * Returns a pointer into the buffer or an error code if the
2476 * path was too long.
2478 * "buflen" should be positive.
2480 * If path is not reachable from the supplied root, then the value of
2481 * root is changed (without modifying refcounts).
2483 char *__d_path(const struct path
*path
, struct path
*root
,
2484 char *buf
, int buflen
)
2486 char *res
= buf
+ buflen
;
2489 prepend(&res
, &buflen
, "\0", 1);
2490 write_seqlock(&rename_lock
);
2491 error
= prepend_path(path
, root
, &res
, &buflen
);
2492 write_sequnlock(&rename_lock
);
2495 return ERR_PTR(error
);
2500 * same as __d_path but appends "(deleted)" for unlinked files.
2502 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2503 char **buf
, int *buflen
)
2505 prepend(buf
, buflen
, "\0", 1);
2506 if (d_unlinked(path
->dentry
)) {
2507 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2512 return prepend_path(path
, root
, buf
, buflen
);
2515 static int prepend_unreachable(char **buffer
, int *buflen
)
2517 return prepend(buffer
, buflen
, "(unreachable)", 13);
2521 * d_path - return the path of a dentry
2522 * @path: path to report
2523 * @buf: buffer to return value in
2524 * @buflen: buffer length
2526 * Convert a dentry into an ASCII path name. If the entry has been deleted
2527 * the string " (deleted)" is appended. Note that this is ambiguous.
2529 * Returns a pointer into the buffer or an error code if the path was
2530 * too long. Note: Callers should use the returned pointer, not the passed
2531 * in buffer, to use the name! The implementation often starts at an offset
2532 * into the buffer, and may leave 0 bytes at the start.
2534 * "buflen" should be positive.
2536 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2538 char *res
= buf
+ buflen
;
2544 * We have various synthetic filesystems that never get mounted. On
2545 * these filesystems dentries are never used for lookup purposes, and
2546 * thus don't need to be hashed. They also don't need a name until a
2547 * user wants to identify the object in /proc/pid/fd/. The little hack
2548 * below allows us to generate a name for these objects on demand:
2550 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2551 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2553 get_fs_root(current
->fs
, &root
);
2554 write_seqlock(&rename_lock
);
2556 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2558 res
= ERR_PTR(error
);
2559 write_sequnlock(&rename_lock
);
2563 EXPORT_SYMBOL(d_path
);
2566 * d_path_with_unreachable - return the path of a dentry
2567 * @path: path to report
2568 * @buf: buffer to return value in
2569 * @buflen: buffer length
2571 * The difference from d_path() is that this prepends "(unreachable)"
2572 * to paths which are unreachable from the current process' root.
2574 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2576 char *res
= buf
+ buflen
;
2581 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2582 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2584 get_fs_root(current
->fs
, &root
);
2585 write_seqlock(&rename_lock
);
2587 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2588 if (!error
&& !path_equal(&tmp
, &root
))
2589 error
= prepend_unreachable(&res
, &buflen
);
2590 write_sequnlock(&rename_lock
);
2593 res
= ERR_PTR(error
);
2599 * Helper function for dentry_operations.d_dname() members
2601 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2602 const char *fmt
, ...)
2608 va_start(args
, fmt
);
2609 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2612 if (sz
> sizeof(temp
) || sz
> buflen
)
2613 return ERR_PTR(-ENAMETOOLONG
);
2615 buffer
+= buflen
- sz
;
2616 return memcpy(buffer
, temp
, sz
);
2620 * Write full pathname from the root of the filesystem into the buffer.
2622 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2624 char *end
= buf
+ buflen
;
2627 prepend(&end
, &buflen
, "\0", 1);
2634 while (!IS_ROOT(dentry
)) {
2635 struct dentry
*parent
= dentry
->d_parent
;
2639 spin_lock(&dentry
->d_lock
);
2640 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2641 spin_unlock(&dentry
->d_lock
);
2642 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2650 return ERR_PTR(-ENAMETOOLONG
);
2653 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2657 write_seqlock(&rename_lock
);
2658 retval
= __dentry_path(dentry
, buf
, buflen
);
2659 write_sequnlock(&rename_lock
);
2663 EXPORT_SYMBOL(dentry_path_raw
);
2665 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2670 write_seqlock(&rename_lock
);
2671 if (d_unlinked(dentry
)) {
2673 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2677 retval
= __dentry_path(dentry
, buf
, buflen
);
2678 write_sequnlock(&rename_lock
);
2679 if (!IS_ERR(retval
) && p
)
2680 *p
= '/'; /* restore '/' overriden with '\0' */
2683 return ERR_PTR(-ENAMETOOLONG
);
2687 * NOTE! The user-level library version returns a
2688 * character pointer. The kernel system call just
2689 * returns the length of the buffer filled (which
2690 * includes the ending '\0' character), or a negative
2691 * error value. So libc would do something like
2693 * char *getcwd(char * buf, size_t size)
2697 * retval = sys_getcwd(buf, size);
2704 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2707 struct path pwd
, root
;
2708 char *page
= (char *) __get_free_page(GFP_USER
);
2713 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2716 write_seqlock(&rename_lock
);
2717 if (!d_unlinked(pwd
.dentry
)) {
2719 struct path tmp
= root
;
2720 char *cwd
= page
+ PAGE_SIZE
;
2721 int buflen
= PAGE_SIZE
;
2723 prepend(&cwd
, &buflen
, "\0", 1);
2724 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2725 write_sequnlock(&rename_lock
);
2730 /* Unreachable from current root */
2731 if (!path_equal(&tmp
, &root
)) {
2732 error
= prepend_unreachable(&cwd
, &buflen
);
2738 len
= PAGE_SIZE
+ page
- cwd
;
2741 if (copy_to_user(buf
, cwd
, len
))
2745 write_sequnlock(&rename_lock
);
2751 free_page((unsigned long) page
);
2756 * Test whether new_dentry is a subdirectory of old_dentry.
2758 * Trivially implemented using the dcache structure
2762 * is_subdir - is new dentry a subdirectory of old_dentry
2763 * @new_dentry: new dentry
2764 * @old_dentry: old dentry
2766 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2767 * Returns 0 otherwise.
2768 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2771 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2776 if (new_dentry
== old_dentry
)
2780 /* for restarting inner loop in case of seq retry */
2781 seq
= read_seqbegin(&rename_lock
);
2783 * Need rcu_readlock to protect against the d_parent trashing
2787 if (d_ancestor(old_dentry
, new_dentry
))
2792 } while (read_seqretry(&rename_lock
, seq
));
2797 int path_is_under(struct path
*path1
, struct path
*path2
)
2799 struct vfsmount
*mnt
= path1
->mnt
;
2800 struct dentry
*dentry
= path1
->dentry
;
2803 br_read_lock(vfsmount_lock
);
2804 if (mnt
!= path2
->mnt
) {
2806 if (mnt
->mnt_parent
== mnt
) {
2807 br_read_unlock(vfsmount_lock
);
2810 if (mnt
->mnt_parent
== path2
->mnt
)
2812 mnt
= mnt
->mnt_parent
;
2814 dentry
= mnt
->mnt_mountpoint
;
2816 res
= is_subdir(dentry
, path2
->dentry
);
2817 br_read_unlock(vfsmount_lock
);
2820 EXPORT_SYMBOL(path_is_under
);
2822 void d_genocide(struct dentry
*root
)
2824 struct dentry
*this_parent
;
2825 struct list_head
*next
;
2829 seq
= read_seqbegin(&rename_lock
);
2832 spin_lock(&this_parent
->d_lock
);
2834 next
= this_parent
->d_subdirs
.next
;
2836 while (next
!= &this_parent
->d_subdirs
) {
2837 struct list_head
*tmp
= next
;
2838 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2841 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2842 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2843 spin_unlock(&dentry
->d_lock
);
2846 if (!list_empty(&dentry
->d_subdirs
)) {
2847 spin_unlock(&this_parent
->d_lock
);
2848 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2849 this_parent
= dentry
;
2850 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2853 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2854 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2857 spin_unlock(&dentry
->d_lock
);
2859 if (this_parent
!= root
) {
2860 struct dentry
*child
= this_parent
;
2861 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2862 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2863 this_parent
->d_count
--;
2865 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2868 next
= child
->d_u
.d_child
.next
;
2871 spin_unlock(&this_parent
->d_lock
);
2872 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2875 write_sequnlock(&rename_lock
);
2880 write_seqlock(&rename_lock
);
2885 * find_inode_number - check for dentry with name
2886 * @dir: directory to check
2887 * @name: Name to find.
2889 * Check whether a dentry already exists for the given name,
2890 * and return the inode number if it has an inode. Otherwise
2893 * This routine is used to post-process directory listings for
2894 * filesystems using synthetic inode numbers, and is necessary
2895 * to keep getcwd() working.
2898 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2900 struct dentry
* dentry
;
2903 dentry
= d_hash_and_lookup(dir
, name
);
2905 if (dentry
->d_inode
)
2906 ino
= dentry
->d_inode
->i_ino
;
2911 EXPORT_SYMBOL(find_inode_number
);
2913 static __initdata
unsigned long dhash_entries
;
2914 static int __init
set_dhash_entries(char *str
)
2918 dhash_entries
= simple_strtoul(str
, &str
, 0);
2921 __setup("dhash_entries=", set_dhash_entries
);
2923 static void __init
dcache_init_early(void)
2927 /* If hashes are distributed across NUMA nodes, defer
2928 * hash allocation until vmalloc space is available.
2934 alloc_large_system_hash("Dentry cache",
2935 sizeof(struct hlist_bl_head
),
2943 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2944 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
2947 static void __init
dcache_init(void)
2952 * A constructor could be added for stable state like the lists,
2953 * but it is probably not worth it because of the cache nature
2956 dentry_cache
= KMEM_CACHE(dentry
,
2957 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2959 /* Hash may have been set up in dcache_init_early */
2964 alloc_large_system_hash("Dentry cache",
2965 sizeof(struct hlist_bl_head
),
2973 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2974 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
2977 /* SLAB cache for __getname() consumers */
2978 struct kmem_cache
*names_cachep __read_mostly
;
2979 EXPORT_SYMBOL(names_cachep
);
2981 EXPORT_SYMBOL(d_genocide
);
2983 void __init
vfs_caches_init_early(void)
2985 dcache_init_early();
2989 void __init
vfs_caches_init(unsigned long mempages
)
2991 unsigned long reserve
;
2993 /* Base hash sizes on available memory, with a reserve equal to
2994 150% of current kernel size */
2996 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2997 mempages
-= reserve
;
2999 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3000 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3004 files_init(mempages
);