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/config.h>
18 #include <linux/syscalls.h>
19 #include <linux/string.h>
22 #include <linux/fsnotify.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/smp_lock.h>
26 #include <linux/hash.h>
27 #include <linux/cache.h>
28 #include <linux/module.h>
29 #include <linux/mount.h>
30 #include <linux/file.h>
31 #include <asm/uaccess.h>
32 #include <linux/security.h>
33 #include <linux/seqlock.h>
34 #include <linux/swap.h>
35 #include <linux/bootmem.h>
37 /* #define DCACHE_DEBUG 1 */
39 int sysctl_vfs_cache_pressure = 100;
40 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
42 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
43 static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
45 EXPORT_SYMBOL(dcache_lock);
47 static kmem_cache_t *dentry_cache;
49 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
52 * This is the single most critical data structure when it comes
53 * to the dcache: the hashtable for lookups. Somebody should try
54 * to make this good - I've just made it work.
56 * This hash-function tries to avoid losing too many bits of hash
57 * information, yet avoid using a prime hash-size or similar.
59 #define D_HASHBITS d_hash_shift
60 #define D_HASHMASK d_hash_mask
62 static unsigned int d_hash_mask;
63 static unsigned int d_hash_shift;
64 static struct hlist_head *dentry_hashtable;
65 static LIST_HEAD(dentry_unused);
67 /* Statistics gathering. */
68 struct dentry_stat_t dentry_stat = {
72 static void d_callback(struct rcu_head *head)
74 struct dentry * dentry = container_of(head, struct dentry, d_rcu);
76 if (dname_external(dentry))
77 kfree(dentry->d_name.name);
78 kmem_cache_free(dentry_cache, dentry);
82 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
85 static void d_free(struct dentry *dentry)
87 if (dentry->d_op && dentry->d_op->d_release)
88 dentry->d_op->d_release(dentry);
89 call_rcu(&dentry->d_rcu, d_callback);
93 * Release the dentry's inode, using the filesystem
94 * d_iput() operation if defined.
95 * Called with dcache_lock and per dentry lock held, drops both.
97 static inline void dentry_iput(struct dentry * dentry)
99 struct inode *inode = dentry->d_inode;
101 dentry->d_inode = NULL;
102 list_del_init(&dentry->d_alias);
103 spin_unlock(&dentry->d_lock);
104 spin_unlock(&dcache_lock);
105 fsnotify_inoderemove(inode);
106 if (dentry->d_op && dentry->d_op->d_iput)
107 dentry->d_op->d_iput(dentry, inode);
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
119 * This is complicated by the fact that we do not want to put
120 * dentries that are no longer on any hash chain on the unused
121 * list: we'd much rather just get rid of them immediately.
123 * However, that implies that we have to traverse the dentry
124 * tree upwards to the parents which might _also_ now be
125 * scheduled for deletion (it may have been only waiting for
126 * its last child to go away).
128 * This tail recursion is done by hand as we don't want to depend
129 * on the compiler to always get this right (gcc generally doesn't).
130 * Real recursion would eat up our stack space.
134 * dput - release a dentry
135 * @dentry: dentry to release
137 * Release a dentry. This will drop the usage count and if appropriate
138 * call the dentry unlink method as well as removing it from the queues and
139 * releasing its resources. If the parent dentries were scheduled for release
140 * they too may now get deleted.
142 * no dcache lock, please.
145 void dput(struct dentry *dentry)
151 if (atomic_read(&dentry->d_count) == 1)
153 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
156 spin_lock(&dentry->d_lock);
157 if (atomic_read(&dentry->d_count)) {
158 spin_unlock(&dentry->d_lock);
159 spin_unlock(&dcache_lock);
164 * AV: ->d_delete() is _NOT_ allowed to block now.
166 if (dentry->d_op && dentry->d_op->d_delete) {
167 if (dentry->d_op->d_delete(dentry))
170 /* Unreachable? Get rid of it */
171 if (d_unhashed(dentry))
173 if (list_empty(&dentry->d_lru)) {
174 dentry->d_flags |= DCACHE_REFERENCED;
175 list_add(&dentry->d_lru, &dentry_unused);
176 dentry_stat.nr_unused++;
178 spin_unlock(&dentry->d_lock);
179 spin_unlock(&dcache_lock);
186 struct dentry *parent;
188 /* If dentry was on d_lru list
189 * delete it from there
191 if (!list_empty(&dentry->d_lru)) {
192 list_del(&dentry->d_lru);
193 dentry_stat.nr_unused--;
195 list_del(&dentry->d_child);
196 dentry_stat.nr_dentry--; /* For d_free, below */
197 /*drops the locks, at that point nobody can reach this dentry */
199 parent = dentry->d_parent;
201 if (dentry == parent)
209 * d_invalidate - invalidate a dentry
210 * @dentry: dentry to invalidate
212 * Try to invalidate the dentry if it turns out to be
213 * possible. If there are other dentries that can be
214 * reached through this one we can't delete it and we
215 * return -EBUSY. On success we return 0.
220 int d_invalidate(struct dentry * dentry)
223 * If it's already been dropped, return OK.
225 spin_lock(&dcache_lock);
226 if (d_unhashed(dentry)) {
227 spin_unlock(&dcache_lock);
231 * Check whether to do a partial shrink_dcache
232 * to get rid of unused child entries.
234 if (!list_empty(&dentry->d_subdirs)) {
235 spin_unlock(&dcache_lock);
236 shrink_dcache_parent(dentry);
237 spin_lock(&dcache_lock);
241 * Somebody else still using it?
243 * If it's a directory, we can't drop it
244 * for fear of somebody re-populating it
245 * with children (even though dropping it
246 * would make it unreachable from the root,
247 * we might still populate it if it was a
248 * working directory or similar).
250 spin_lock(&dentry->d_lock);
251 if (atomic_read(&dentry->d_count) > 1) {
252 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
253 spin_unlock(&dentry->d_lock);
254 spin_unlock(&dcache_lock);
260 spin_unlock(&dentry->d_lock);
261 spin_unlock(&dcache_lock);
265 /* This should be called _only_ with dcache_lock held */
267 static inline struct dentry * __dget_locked(struct dentry *dentry)
269 atomic_inc(&dentry->d_count);
270 if (!list_empty(&dentry->d_lru)) {
271 dentry_stat.nr_unused--;
272 list_del_init(&dentry->d_lru);
277 struct dentry * dget_locked(struct dentry *dentry)
279 return __dget_locked(dentry);
283 * d_find_alias - grab a hashed alias of inode
284 * @inode: inode in question
285 * @want_discon: flag, used by d_splice_alias, to request
286 * that only a DISCONNECTED alias be returned.
288 * If inode has a hashed alias, or is a directory and has any alias,
289 * acquire the reference to alias and return it. Otherwise return NULL.
290 * Notice that if inode is a directory there can be only one alias and
291 * it can be unhashed only if it has no children, or if it is the root
294 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
295 * any other hashed alias over that one unless @want_discon is set,
296 * in which case only return a DCACHE_DISCONNECTED alias.
299 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
301 struct list_head *head, *next, *tmp;
302 struct dentry *alias, *discon_alias=NULL;
304 head = &inode->i_dentry;
305 next = inode->i_dentry.next;
306 while (next != head) {
310 alias = list_entry(tmp, struct dentry, d_alias);
311 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
312 if (alias->d_flags & DCACHE_DISCONNECTED)
313 discon_alias = alias;
314 else if (!want_discon) {
315 __dget_locked(alias);
321 __dget_locked(discon_alias);
325 struct dentry * d_find_alias(struct inode *inode)
328 spin_lock(&dcache_lock);
329 de = __d_find_alias(inode, 0);
330 spin_unlock(&dcache_lock);
335 * Try to kill dentries associated with this inode.
336 * WARNING: you must own a reference to inode.
338 void d_prune_aliases(struct inode *inode)
340 struct list_head *tmp, *head = &inode->i_dentry;
342 spin_lock(&dcache_lock);
344 while ((tmp = tmp->next) != head) {
345 struct dentry *dentry = list_entry(tmp, struct dentry, d_alias);
346 spin_lock(&dentry->d_lock);
347 if (!atomic_read(&dentry->d_count)) {
348 __dget_locked(dentry);
350 spin_unlock(&dentry->d_lock);
351 spin_unlock(&dcache_lock);
355 spin_unlock(&dentry->d_lock);
357 spin_unlock(&dcache_lock);
361 * Throw away a dentry - free the inode, dput the parent.
362 * This requires that the LRU list has already been
364 * Called with dcache_lock, drops it and then regains.
366 static inline void prune_one_dentry(struct dentry * dentry)
368 struct dentry * parent;
371 list_del(&dentry->d_child);
372 dentry_stat.nr_dentry--; /* For d_free, below */
374 parent = dentry->d_parent;
376 if (parent != dentry)
378 spin_lock(&dcache_lock);
382 * prune_dcache - shrink the dcache
383 * @count: number of entries to try and free
385 * Shrink the dcache. This is done when we need
386 * more memory, or simply when we need to unmount
387 * something (at which point we need to unuse
390 * This function may fail to free any resources if
391 * all the dentries are in use.
394 static void prune_dcache(int count)
396 spin_lock(&dcache_lock);
397 for (; count ; count--) {
398 struct dentry *dentry;
399 struct list_head *tmp;
401 cond_resched_lock(&dcache_lock);
403 tmp = dentry_unused.prev;
404 if (tmp == &dentry_unused)
407 prefetch(dentry_unused.prev);
408 dentry_stat.nr_unused--;
409 dentry = list_entry(tmp, struct dentry, d_lru);
411 spin_lock(&dentry->d_lock);
413 * We found an inuse dentry which was not removed from
414 * dentry_unused because of laziness during lookup. Do not free
415 * it - just keep it off the dentry_unused list.
417 if (atomic_read(&dentry->d_count)) {
418 spin_unlock(&dentry->d_lock);
421 /* If the dentry was recently referenced, don't free it. */
422 if (dentry->d_flags & DCACHE_REFERENCED) {
423 dentry->d_flags &= ~DCACHE_REFERENCED;
424 list_add(&dentry->d_lru, &dentry_unused);
425 dentry_stat.nr_unused++;
426 spin_unlock(&dentry->d_lock);
429 prune_one_dentry(dentry);
431 spin_unlock(&dcache_lock);
435 * Shrink the dcache for the specified super block.
436 * This allows us to unmount a device without disturbing
437 * the dcache for the other devices.
439 * This implementation makes just two traversals of the
440 * unused list. On the first pass we move the selected
441 * dentries to the most recent end, and on the second
442 * pass we free them. The second pass must restart after
443 * each dput(), but since the target dentries are all at
444 * the end, it's really just a single traversal.
448 * shrink_dcache_sb - shrink dcache for a superblock
451 * Shrink the dcache for the specified super block. This
452 * is used to free the dcache before unmounting a file
456 void shrink_dcache_sb(struct super_block * sb)
458 struct list_head *tmp, *next;
459 struct dentry *dentry;
462 * Pass one ... move the dentries for the specified
463 * superblock to the most recent end of the unused list.
465 spin_lock(&dcache_lock);
466 next = dentry_unused.next;
467 while (next != &dentry_unused) {
470 dentry = list_entry(tmp, struct dentry, d_lru);
471 if (dentry->d_sb != sb)
474 list_add(tmp, &dentry_unused);
478 * Pass two ... free the dentries for this superblock.
481 next = dentry_unused.next;
482 while (next != &dentry_unused) {
485 dentry = list_entry(tmp, struct dentry, d_lru);
486 if (dentry->d_sb != sb)
488 dentry_stat.nr_unused--;
490 spin_lock(&dentry->d_lock);
491 if (atomic_read(&dentry->d_count)) {
492 spin_unlock(&dentry->d_lock);
495 prune_one_dentry(dentry);
498 spin_unlock(&dcache_lock);
502 * Search for at least 1 mount point in the dentry's subdirs.
503 * We descend to the next level whenever the d_subdirs
504 * list is non-empty and continue searching.
508 * have_submounts - check for mounts over a dentry
509 * @parent: dentry to check.
511 * Return true if the parent or its subdirectories contain
515 int have_submounts(struct dentry *parent)
517 struct dentry *this_parent = parent;
518 struct list_head *next;
520 spin_lock(&dcache_lock);
521 if (d_mountpoint(parent))
524 next = this_parent->d_subdirs.next;
526 while (next != &this_parent->d_subdirs) {
527 struct list_head *tmp = next;
528 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
530 /* Have we found a mount point ? */
531 if (d_mountpoint(dentry))
533 if (!list_empty(&dentry->d_subdirs)) {
534 this_parent = dentry;
539 * All done at this level ... ascend and resume the search.
541 if (this_parent != parent) {
542 next = this_parent->d_child.next;
543 this_parent = this_parent->d_parent;
546 spin_unlock(&dcache_lock);
547 return 0; /* No mount points found in tree */
549 spin_unlock(&dcache_lock);
554 * Search the dentry child list for the specified parent,
555 * and move any unused dentries to the end of the unused
556 * list for prune_dcache(). We descend to the next level
557 * whenever the d_subdirs list is non-empty and continue
560 * It returns zero iff there are no unused children,
561 * otherwise it returns the number of children moved to
562 * the end of the unused list. This may not be the total
563 * number of unused children, because select_parent can
564 * drop the lock and return early due to latency
567 static int select_parent(struct dentry * parent)
569 struct dentry *this_parent = parent;
570 struct list_head *next;
573 spin_lock(&dcache_lock);
575 next = this_parent->d_subdirs.next;
577 while (next != &this_parent->d_subdirs) {
578 struct list_head *tmp = next;
579 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
582 if (!list_empty(&dentry->d_lru)) {
583 dentry_stat.nr_unused--;
584 list_del_init(&dentry->d_lru);
587 * move only zero ref count dentries to the end
588 * of the unused list for prune_dcache
590 if (!atomic_read(&dentry->d_count)) {
591 list_add(&dentry->d_lru, dentry_unused.prev);
592 dentry_stat.nr_unused++;
597 * We can return to the caller if we have found some (this
598 * ensures forward progress). We'll be coming back to find
601 if (found && need_resched())
605 * Descend a level if the d_subdirs list is non-empty.
607 if (!list_empty(&dentry->d_subdirs)) {
608 this_parent = dentry;
610 printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
611 dentry->d_parent->d_name.name, dentry->d_name.name, found);
617 * All done at this level ... ascend and resume the search.
619 if (this_parent != parent) {
620 next = this_parent->d_child.next;
621 this_parent = this_parent->d_parent;
623 printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
624 this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
629 spin_unlock(&dcache_lock);
634 * shrink_dcache_parent - prune dcache
635 * @parent: parent of entries to prune
637 * Prune the dcache to remove unused children of the parent dentry.
640 void shrink_dcache_parent(struct dentry * parent)
644 while ((found = select_parent(parent)) != 0)
649 * shrink_dcache_anon - further prune the cache
650 * @head: head of d_hash list of dentries to prune
652 * Prune the dentries that are anonymous
654 * parsing d_hash list does not hlist_for_each_rcu() as it
655 * done under dcache_lock.
658 void shrink_dcache_anon(struct hlist_head *head)
660 struct hlist_node *lp;
664 spin_lock(&dcache_lock);
665 hlist_for_each(lp, head) {
666 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
667 if (!list_empty(&this->d_lru)) {
668 dentry_stat.nr_unused--;
669 list_del_init(&this->d_lru);
673 * move only zero ref count dentries to the end
674 * of the unused list for prune_dcache
676 if (!atomic_read(&this->d_count)) {
677 list_add_tail(&this->d_lru, &dentry_unused);
678 dentry_stat.nr_unused++;
682 spin_unlock(&dcache_lock);
688 * Scan `nr' dentries and return the number which remain.
690 * We need to avoid reentering the filesystem if the caller is performing a
691 * GFP_NOFS allocation attempt. One example deadlock is:
693 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
694 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
695 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
697 * In this case we return -1 to tell the caller that we baled.
699 static int shrink_dcache_memory(int nr, unsigned int gfp_mask)
702 if (!(gfp_mask & __GFP_FS))
706 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
710 * d_alloc - allocate a dcache entry
711 * @parent: parent of entry to allocate
712 * @name: qstr of the name
714 * Allocates a dentry. It returns %NULL if there is insufficient memory
715 * available. On a success the dentry is returned. The name passed in is
716 * copied and the copy passed in may be reused after this call.
719 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
721 struct dentry *dentry;
724 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
728 if (name->len > DNAME_INLINE_LEN-1) {
729 dname = kmalloc(name->len + 1, GFP_KERNEL);
731 kmem_cache_free(dentry_cache, dentry);
735 dname = dentry->d_iname;
737 dentry->d_name.name = dname;
739 dentry->d_name.len = name->len;
740 dentry->d_name.hash = name->hash;
741 memcpy(dname, name->name, name->len);
742 dname[name->len] = 0;
744 atomic_set(&dentry->d_count, 1);
745 dentry->d_flags = DCACHE_UNHASHED;
746 spin_lock_init(&dentry->d_lock);
747 dentry->d_inode = NULL;
748 dentry->d_parent = NULL;
751 dentry->d_fsdata = NULL;
752 dentry->d_mounted = 0;
753 dentry->d_cookie = NULL;
754 INIT_HLIST_NODE(&dentry->d_hash);
755 INIT_LIST_HEAD(&dentry->d_lru);
756 INIT_LIST_HEAD(&dentry->d_subdirs);
757 INIT_LIST_HEAD(&dentry->d_alias);
760 dentry->d_parent = dget(parent);
761 dentry->d_sb = parent->d_sb;
763 INIT_LIST_HEAD(&dentry->d_child);
766 spin_lock(&dcache_lock);
768 list_add(&dentry->d_child, &parent->d_subdirs);
769 dentry_stat.nr_dentry++;
770 spin_unlock(&dcache_lock);
775 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
780 q.len = strlen(name);
781 q.hash = full_name_hash(q.name, q.len);
782 return d_alloc(parent, &q);
786 * d_instantiate - fill in inode information for a dentry
787 * @entry: dentry to complete
788 * @inode: inode to attach to this dentry
790 * Fill in inode information in the entry.
792 * This turns negative dentries into productive full members
795 * NOTE! This assumes that the inode count has been incremented
796 * (or otherwise set) by the caller to indicate that it is now
797 * in use by the dcache.
800 void d_instantiate(struct dentry *entry, struct inode * inode)
802 if (!list_empty(&entry->d_alias)) BUG();
803 spin_lock(&dcache_lock);
805 list_add(&entry->d_alias, &inode->i_dentry);
806 entry->d_inode = inode;
807 spin_unlock(&dcache_lock);
808 security_d_instantiate(entry, inode);
812 * d_instantiate_unique - instantiate a non-aliased dentry
813 * @entry: dentry to instantiate
814 * @inode: inode to attach to this dentry
816 * Fill in inode information in the entry. On success, it returns NULL.
817 * If an unhashed alias of "entry" already exists, then we return the
818 * aliased dentry instead.
820 * Note that in order to avoid conflicts with rename() etc, the caller
821 * had better be holding the parent directory semaphore.
823 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
825 struct dentry *alias;
826 int len = entry->d_name.len;
827 const char *name = entry->d_name.name;
828 unsigned int hash = entry->d_name.hash;
830 BUG_ON(!list_empty(&entry->d_alias));
831 spin_lock(&dcache_lock);
834 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
835 struct qstr *qstr = &alias->d_name;
837 if (qstr->hash != hash)
839 if (alias->d_parent != entry->d_parent)
841 if (qstr->len != len)
843 if (memcmp(qstr->name, name, len))
846 spin_unlock(&dcache_lock);
847 BUG_ON(!d_unhashed(alias));
850 list_add(&entry->d_alias, &inode->i_dentry);
852 entry->d_inode = inode;
853 spin_unlock(&dcache_lock);
854 security_d_instantiate(entry, inode);
857 EXPORT_SYMBOL(d_instantiate_unique);
860 * d_alloc_root - allocate root dentry
861 * @root_inode: inode to allocate the root for
863 * Allocate a root ("/") dentry for the inode given. The inode is
864 * instantiated and returned. %NULL is returned if there is insufficient
865 * memory or the inode passed is %NULL.
868 struct dentry * d_alloc_root(struct inode * root_inode)
870 struct dentry *res = NULL;
873 static const struct qstr name = { .name = "/", .len = 1 };
875 res = d_alloc(NULL, &name);
877 res->d_sb = root_inode->i_sb;
879 d_instantiate(res, root_inode);
885 static inline struct hlist_head *d_hash(struct dentry *parent,
888 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
889 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
890 return dentry_hashtable + (hash & D_HASHMASK);
894 * d_alloc_anon - allocate an anonymous dentry
895 * @inode: inode to allocate the dentry for
897 * This is similar to d_alloc_root. It is used by filesystems when
898 * creating a dentry for a given inode, often in the process of
899 * mapping a filehandle to a dentry. The returned dentry may be
900 * anonymous, or may have a full name (if the inode was already
901 * in the cache). The file system may need to make further
902 * efforts to connect this dentry into the dcache properly.
904 * When called on a directory inode, we must ensure that
905 * the inode only ever has one dentry. If a dentry is
906 * found, that is returned instead of allocating a new one.
908 * On successful return, the reference to the inode has been transferred
909 * to the dentry. If %NULL is returned (indicating kmalloc failure),
910 * the reference on the inode has not been released.
913 struct dentry * d_alloc_anon(struct inode *inode)
915 static const struct qstr anonstring = { .name = "" };
919 if ((res = d_find_alias(inode))) {
924 tmp = d_alloc(NULL, &anonstring);
928 tmp->d_parent = tmp; /* make sure dput doesn't croak */
930 spin_lock(&dcache_lock);
931 res = __d_find_alias(inode, 0);
933 /* attach a disconnected dentry */
936 spin_lock(&res->d_lock);
937 res->d_sb = inode->i_sb;
939 res->d_inode = inode;
940 res->d_flags |= DCACHE_DISCONNECTED;
941 res->d_flags &= ~DCACHE_UNHASHED;
942 list_add(&res->d_alias, &inode->i_dentry);
943 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
944 spin_unlock(&res->d_lock);
946 inode = NULL; /* don't drop reference */
948 spin_unlock(&dcache_lock);
959 * d_splice_alias - splice a disconnected dentry into the tree if one exists
960 * @inode: the inode which may have a disconnected dentry
961 * @dentry: a negative dentry which we want to point to the inode.
963 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
964 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
965 * and return it, else simply d_add the inode to the dentry and return NULL.
967 * This is needed in the lookup routine of any filesystem that is exportable
968 * (via knfsd) so that we can build dcache paths to directories effectively.
970 * If a dentry was found and moved, then it is returned. Otherwise NULL
971 * is returned. This matches the expected return value of ->lookup.
974 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
976 struct dentry *new = NULL;
979 spin_lock(&dcache_lock);
980 new = __d_find_alias(inode, 1);
982 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
983 spin_unlock(&dcache_lock);
984 security_d_instantiate(new, inode);
989 /* d_instantiate takes dcache_lock, so we do it by hand */
990 list_add(&dentry->d_alias, &inode->i_dentry);
991 dentry->d_inode = inode;
992 spin_unlock(&dcache_lock);
993 security_d_instantiate(dentry, inode);
997 d_add(dentry, inode);
1003 * d_lookup - search for a dentry
1004 * @parent: parent dentry
1005 * @name: qstr of name we wish to find
1007 * Searches the children of the parent dentry for the name in question. If
1008 * the dentry is found its reference count is incremented and the dentry
1009 * is returned. The caller must use d_put to free the entry when it has
1010 * finished using it. %NULL is returned on failure.
1012 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1013 * Memory barriers are used while updating and doing lockless traversal.
1014 * To avoid races with d_move while rename is happening, d_lock is used.
1016 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1017 * and name pointer in one structure pointed by d_qstr.
1019 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1020 * lookup is going on.
1022 * dentry_unused list is not updated even if lookup finds the required dentry
1023 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1024 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1027 * d_lookup() is protected against the concurrent renames in some unrelated
1028 * directory using the seqlockt_t rename_lock.
1031 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1033 struct dentry * dentry = NULL;
1037 seq = read_seqbegin(&rename_lock);
1038 dentry = __d_lookup(parent, name);
1041 } while (read_seqretry(&rename_lock, seq));
1045 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1047 unsigned int len = name->len;
1048 unsigned int hash = name->hash;
1049 const unsigned char *str = name->name;
1050 struct hlist_head *head = d_hash(parent,hash);
1051 struct dentry *found = NULL;
1052 struct hlist_node *node;
1056 hlist_for_each_rcu(node, head) {
1057 struct dentry *dentry;
1060 dentry = hlist_entry(node, struct dentry, d_hash);
1062 if (dentry->d_name.hash != hash)
1064 if (dentry->d_parent != parent)
1067 spin_lock(&dentry->d_lock);
1070 * Recheck the dentry after taking the lock - d_move may have
1071 * changed things. Don't bother checking the hash because we're
1072 * about to compare the whole name anyway.
1074 if (dentry->d_parent != parent)
1078 * It is safe to compare names since d_move() cannot
1079 * change the qstr (protected by d_lock).
1081 qstr = &dentry->d_name;
1082 if (parent->d_op && parent->d_op->d_compare) {
1083 if (parent->d_op->d_compare(parent, qstr, name))
1086 if (qstr->len != len)
1088 if (memcmp(qstr->name, str, len))
1092 if (!d_unhashed(dentry)) {
1093 atomic_inc(&dentry->d_count);
1096 spin_unlock(&dentry->d_lock);
1099 spin_unlock(&dentry->d_lock);
1107 * d_validate - verify dentry provided from insecure source
1108 * @dentry: The dentry alleged to be valid child of @dparent
1109 * @dparent: The parent dentry (known to be valid)
1110 * @hash: Hash of the dentry
1111 * @len: Length of the name
1113 * An insecure source has sent us a dentry, here we verify it and dget() it.
1114 * This is used by ncpfs in its readdir implementation.
1115 * Zero is returned in the dentry is invalid.
1118 int d_validate(struct dentry *dentry, struct dentry *dparent)
1120 struct hlist_head *base;
1121 struct hlist_node *lhp;
1123 /* Check whether the ptr might be valid at all.. */
1124 if (!kmem_ptr_validate(dentry_cache, dentry))
1127 if (dentry->d_parent != dparent)
1130 spin_lock(&dcache_lock);
1131 base = d_hash(dparent, dentry->d_name.hash);
1132 hlist_for_each(lhp,base) {
1133 /* hlist_for_each_rcu() not required for d_hash list
1134 * as it is parsed under dcache_lock
1136 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1137 __dget_locked(dentry);
1138 spin_unlock(&dcache_lock);
1142 spin_unlock(&dcache_lock);
1148 * When a file is deleted, we have two options:
1149 * - turn this dentry into a negative dentry
1150 * - unhash this dentry and free it.
1152 * Usually, we want to just turn this into
1153 * a negative dentry, but if anybody else is
1154 * currently using the dentry or the inode
1155 * we can't do that and we fall back on removing
1156 * it from the hash queues and waiting for
1157 * it to be deleted later when it has no users
1161 * d_delete - delete a dentry
1162 * @dentry: The dentry to delete
1164 * Turn the dentry into a negative dentry if possible, otherwise
1165 * remove it from the hash queues so it can be deleted later
1168 void d_delete(struct dentry * dentry)
1172 * Are we the only user?
1174 spin_lock(&dcache_lock);
1175 spin_lock(&dentry->d_lock);
1176 isdir = S_ISDIR(dentry->d_inode->i_mode);
1177 if (atomic_read(&dentry->d_count) == 1) {
1178 dentry_iput(dentry);
1179 fsnotify_nameremove(dentry, isdir);
1183 if (!d_unhashed(dentry))
1186 spin_unlock(&dentry->d_lock);
1187 spin_unlock(&dcache_lock);
1189 fsnotify_nameremove(dentry, isdir);
1192 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1195 entry->d_flags &= ~DCACHE_UNHASHED;
1196 hlist_add_head_rcu(&entry->d_hash, list);
1200 * d_rehash - add an entry back to the hash
1201 * @entry: dentry to add to the hash
1203 * Adds a dentry to the hash according to its name.
1206 void d_rehash(struct dentry * entry)
1208 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1210 spin_lock(&dcache_lock);
1211 spin_lock(&entry->d_lock);
1212 __d_rehash(entry, list);
1213 spin_unlock(&entry->d_lock);
1214 spin_unlock(&dcache_lock);
1217 #define do_switch(x,y) do { \
1218 __typeof__ (x) __tmp = x; \
1219 x = y; y = __tmp; } while (0)
1222 * When switching names, the actual string doesn't strictly have to
1223 * be preserved in the target - because we're dropping the target
1224 * anyway. As such, we can just do a simple memcpy() to copy over
1225 * the new name before we switch.
1227 * Note that we have to be a lot more careful about getting the hash
1228 * switched - we have to switch the hash value properly even if it
1229 * then no longer matches the actual (corrupted) string of the target.
1230 * The hash value has to match the hash queue that the dentry is on..
1232 static void switch_names(struct dentry *dentry, struct dentry *target)
1234 if (dname_external(target)) {
1235 if (dname_external(dentry)) {
1237 * Both external: swap the pointers
1239 do_switch(target->d_name.name, dentry->d_name.name);
1242 * dentry:internal, target:external. Steal target's
1243 * storage and make target internal.
1245 dentry->d_name.name = target->d_name.name;
1246 target->d_name.name = target->d_iname;
1249 if (dname_external(dentry)) {
1251 * dentry:external, target:internal. Give dentry's
1252 * storage to target and make dentry internal
1254 memcpy(dentry->d_iname, target->d_name.name,
1255 target->d_name.len + 1);
1256 target->d_name.name = dentry->d_name.name;
1257 dentry->d_name.name = dentry->d_iname;
1260 * Both are internal. Just copy target to dentry
1262 memcpy(dentry->d_iname, target->d_name.name,
1263 target->d_name.len + 1);
1269 * We cannibalize "target" when moving dentry on top of it,
1270 * because it's going to be thrown away anyway. We could be more
1271 * polite about it, though.
1273 * This forceful removal will result in ugly /proc output if
1274 * somebody holds a file open that got deleted due to a rename.
1275 * We could be nicer about the deleted file, and let it show
1276 * up under the name it got deleted rather than the name that
1281 * d_move - move a dentry
1282 * @dentry: entry to move
1283 * @target: new dentry
1285 * Update the dcache to reflect the move of a file name. Negative
1286 * dcache entries should not be moved in this way.
1289 void d_move(struct dentry * dentry, struct dentry * target)
1291 struct hlist_head *list;
1293 if (!dentry->d_inode)
1294 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1296 spin_lock(&dcache_lock);
1297 write_seqlock(&rename_lock);
1299 * XXXX: do we really need to take target->d_lock?
1301 if (target < dentry) {
1302 spin_lock(&target->d_lock);
1303 spin_lock(&dentry->d_lock);
1305 spin_lock(&dentry->d_lock);
1306 spin_lock(&target->d_lock);
1309 /* Move the dentry to the target hash queue, if on different bucket */
1310 if (dentry->d_flags & DCACHE_UNHASHED)
1311 goto already_unhashed;
1313 hlist_del_rcu(&dentry->d_hash);
1316 list = d_hash(target->d_parent, target->d_name.hash);
1317 __d_rehash(dentry, list);
1319 /* Unhash the target: dput() will then get rid of it */
1322 list_del(&dentry->d_child);
1323 list_del(&target->d_child);
1325 /* Switch the names.. */
1326 switch_names(dentry, target);
1327 do_switch(dentry->d_name.len, target->d_name.len);
1328 do_switch(dentry->d_name.hash, target->d_name.hash);
1330 /* ... and switch the parents */
1331 if (IS_ROOT(dentry)) {
1332 dentry->d_parent = target->d_parent;
1333 target->d_parent = target;
1334 INIT_LIST_HEAD(&target->d_child);
1336 do_switch(dentry->d_parent, target->d_parent);
1338 /* And add them back to the (new) parent lists */
1339 list_add(&target->d_child, &target->d_parent->d_subdirs);
1342 list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
1343 spin_unlock(&target->d_lock);
1344 spin_unlock(&dentry->d_lock);
1345 write_sequnlock(&rename_lock);
1346 spin_unlock(&dcache_lock);
1350 * d_path - return the path of a dentry
1351 * @dentry: dentry to report
1352 * @vfsmnt: vfsmnt to which the dentry belongs
1353 * @root: root dentry
1354 * @rootmnt: vfsmnt to which the root dentry belongs
1355 * @buffer: buffer to return value in
1356 * @buflen: buffer length
1358 * Convert a dentry into an ASCII path name. If the entry has been deleted
1359 * the string " (deleted)" is appended. Note that this is ambiguous.
1361 * Returns the buffer or an error code if the path was too long.
1363 * "buflen" should be positive. Caller holds the dcache_lock.
1365 static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1366 struct dentry *root, struct vfsmount *rootmnt,
1367 char *buffer, int buflen)
1369 char * end = buffer+buflen;
1375 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1380 memcpy(end, " (deleted)", 10);
1390 struct dentry * parent;
1392 if (dentry == root && vfsmnt == rootmnt)
1394 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1396 spin_lock(&vfsmount_lock);
1397 if (vfsmnt->mnt_parent == vfsmnt) {
1398 spin_unlock(&vfsmount_lock);
1401 dentry = vfsmnt->mnt_mountpoint;
1402 vfsmnt = vfsmnt->mnt_parent;
1403 spin_unlock(&vfsmount_lock);
1406 parent = dentry->d_parent;
1408 namelen = dentry->d_name.len;
1409 buflen -= namelen + 1;
1413 memcpy(end, dentry->d_name.name, namelen);
1422 namelen = dentry->d_name.len;
1426 retval -= namelen-1; /* hit the slash */
1427 memcpy(retval, dentry->d_name.name, namelen);
1430 return ERR_PTR(-ENAMETOOLONG);
1433 /* write full pathname into buffer and return start of pathname */
1434 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1435 char *buf, int buflen)
1438 struct vfsmount *rootmnt;
1439 struct dentry *root;
1441 read_lock(¤t->fs->lock);
1442 rootmnt = mntget(current->fs->rootmnt);
1443 root = dget(current->fs->root);
1444 read_unlock(¤t->fs->lock);
1445 spin_lock(&dcache_lock);
1446 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1447 spin_unlock(&dcache_lock);
1454 * NOTE! The user-level library version returns a
1455 * character pointer. The kernel system call just
1456 * returns the length of the buffer filled (which
1457 * includes the ending '\0' character), or a negative
1458 * error value. So libc would do something like
1460 * char *getcwd(char * buf, size_t size)
1464 * retval = sys_getcwd(buf, size);
1471 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1474 struct vfsmount *pwdmnt, *rootmnt;
1475 struct dentry *pwd, *root;
1476 char *page = (char *) __get_free_page(GFP_USER);
1481 read_lock(¤t->fs->lock);
1482 pwdmnt = mntget(current->fs->pwdmnt);
1483 pwd = dget(current->fs->pwd);
1484 rootmnt = mntget(current->fs->rootmnt);
1485 root = dget(current->fs->root);
1486 read_unlock(¤t->fs->lock);
1489 /* Has the current directory has been unlinked? */
1490 spin_lock(&dcache_lock);
1491 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1495 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1496 spin_unlock(&dcache_lock);
1498 error = PTR_ERR(cwd);
1503 len = PAGE_SIZE + page - cwd;
1506 if (copy_to_user(buf, cwd, len))
1510 spin_unlock(&dcache_lock);
1517 free_page((unsigned long) page);
1522 * Test whether new_dentry is a subdirectory of old_dentry.
1524 * Trivially implemented using the dcache structure
1528 * is_subdir - is new dentry a subdirectory of old_dentry
1529 * @new_dentry: new dentry
1530 * @old_dentry: old dentry
1532 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1533 * Returns 0 otherwise.
1534 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1537 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1540 struct dentry * saved = new_dentry;
1543 /* need rcu_readlock to protect against the d_parent trashing due to
1548 /* for restarting inner loop in case of seq retry */
1551 seq = read_seqbegin(&rename_lock);
1553 if (new_dentry != old_dentry) {
1554 struct dentry * parent = new_dentry->d_parent;
1555 if (parent == new_dentry)
1557 new_dentry = parent;
1563 } while (read_seqretry(&rename_lock, seq));
1569 void d_genocide(struct dentry *root)
1571 struct dentry *this_parent = root;
1572 struct list_head *next;
1574 spin_lock(&dcache_lock);
1576 next = this_parent->d_subdirs.next;
1578 while (next != &this_parent->d_subdirs) {
1579 struct list_head *tmp = next;
1580 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1582 if (d_unhashed(dentry)||!dentry->d_inode)
1584 if (!list_empty(&dentry->d_subdirs)) {
1585 this_parent = dentry;
1588 atomic_dec(&dentry->d_count);
1590 if (this_parent != root) {
1591 next = this_parent->d_child.next;
1592 atomic_dec(&this_parent->d_count);
1593 this_parent = this_parent->d_parent;
1596 spin_unlock(&dcache_lock);
1600 * find_inode_number - check for dentry with name
1601 * @dir: directory to check
1602 * @name: Name to find.
1604 * Check whether a dentry already exists for the given name,
1605 * and return the inode number if it has an inode. Otherwise
1608 * This routine is used to post-process directory listings for
1609 * filesystems using synthetic inode numbers, and is necessary
1610 * to keep getcwd() working.
1613 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1615 struct dentry * dentry;
1619 * Check for a fs-specific hash function. Note that we must
1620 * calculate the standard hash first, as the d_op->d_hash()
1621 * routine may choose to leave the hash value unchanged.
1623 name->hash = full_name_hash(name->name, name->len);
1624 if (dir->d_op && dir->d_op->d_hash)
1626 if (dir->d_op->d_hash(dir, name) != 0)
1630 dentry = d_lookup(dir, name);
1633 if (dentry->d_inode)
1634 ino = dentry->d_inode->i_ino;
1641 static __initdata unsigned long dhash_entries;
1642 static int __init set_dhash_entries(char *str)
1646 dhash_entries = simple_strtoul(str, &str, 0);
1649 __setup("dhash_entries=", set_dhash_entries);
1651 static void __init dcache_init_early(void)
1655 /* If hashes are distributed across NUMA nodes, defer
1656 * hash allocation until vmalloc space is available.
1662 alloc_large_system_hash("Dentry cache",
1663 sizeof(struct hlist_head),
1671 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1672 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1675 static void __init dcache_init(unsigned long mempages)
1680 * A constructor could be added for stable state like the lists,
1681 * but it is probably not worth it because of the cache nature
1684 dentry_cache = kmem_cache_create("dentry_cache",
1685 sizeof(struct dentry),
1687 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1690 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1692 /* Hash may have been set up in dcache_init_early */
1697 alloc_large_system_hash("Dentry cache",
1698 sizeof(struct hlist_head),
1706 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1707 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1710 /* SLAB cache for __getname() consumers */
1711 kmem_cache_t *names_cachep;
1713 /* SLAB cache for file structures */
1714 kmem_cache_t *filp_cachep;
1716 EXPORT_SYMBOL(d_genocide);
1718 extern void bdev_cache_init(void);
1719 extern void chrdev_init(void);
1721 void __init vfs_caches_init_early(void)
1723 dcache_init_early();
1727 void __init vfs_caches_init(unsigned long mempages)
1729 unsigned long reserve;
1731 /* Base hash sizes on available memory, with a reserve equal to
1732 150% of current kernel size */
1734 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1735 mempages -= reserve;
1737 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1738 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1740 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1741 SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
1743 dcache_init(mempages);
1744 inode_init(mempages);
1745 files_init(mempages);
1751 EXPORT_SYMBOL(d_alloc);
1752 EXPORT_SYMBOL(d_alloc_anon);
1753 EXPORT_SYMBOL(d_alloc_root);
1754 EXPORT_SYMBOL(d_delete);
1755 EXPORT_SYMBOL(d_find_alias);
1756 EXPORT_SYMBOL(d_instantiate);
1757 EXPORT_SYMBOL(d_invalidate);
1758 EXPORT_SYMBOL(d_lookup);
1759 EXPORT_SYMBOL(d_move);
1760 EXPORT_SYMBOL(d_path);
1761 EXPORT_SYMBOL(d_prune_aliases);
1762 EXPORT_SYMBOL(d_rehash);
1763 EXPORT_SYMBOL(d_splice_alias);
1764 EXPORT_SYMBOL(d_validate);
1765 EXPORT_SYMBOL(dget_locked);
1766 EXPORT_SYMBOL(dput);
1767 EXPORT_SYMBOL(find_inode_number);
1768 EXPORT_SYMBOL(have_submounts);
1769 EXPORT_SYMBOL(names_cachep);
1770 EXPORT_SYMBOL(shrink_dcache_parent);
1771 EXPORT_SYMBOL(shrink_dcache_sb);