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/string.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/smp_lock.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>
35 /* #define DCACHE_DEBUG 1 */
37 int sysctl_vfs_cache_pressure = 100;
39 spinlock_t dcache_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
40 seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
42 EXPORT_SYMBOL(dcache_lock);
44 static kmem_cache_t *dentry_cache;
46 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
49 * This is the single most critical data structure when it comes
50 * to the dcache: the hashtable for lookups. Somebody should try
51 * to make this good - I've just made it work.
53 * This hash-function tries to avoid losing too many bits of hash
54 * information, yet avoid using a prime hash-size or similar.
56 #define D_HASHBITS d_hash_shift
57 #define D_HASHMASK d_hash_mask
59 static unsigned int d_hash_mask;
60 static unsigned int d_hash_shift;
61 static struct hlist_head *dentry_hashtable;
62 static LIST_HEAD(dentry_unused);
64 /* Statistics gathering. */
65 struct dentry_stat_t dentry_stat = {
69 static void d_callback(struct rcu_head *head)
71 struct dentry * dentry = container_of(head, struct dentry, d_rcu);
73 if (dname_external(dentry))
74 kfree(dentry->d_name.name);
75 kmem_cache_free(dentry_cache, dentry);
79 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
82 static void d_free(struct dentry *dentry)
84 if (dentry->d_op && dentry->d_op->d_release)
85 dentry->d_op->d_release(dentry);
86 if (dentry->d_extra_attributes) {
87 kfree(dentry->d_extra_attributes);
88 dentry->d_extra_attributes = NULL;
90 call_rcu(&dentry->d_rcu, d_callback);
94 * Release the dentry's inode, using the filesystem
95 * d_iput() operation if defined.
96 * Called with dcache_lock and per dentry lock held, drops both.
98 static inline void dentry_iput(struct dentry * dentry)
100 struct inode *inode = dentry->d_inode;
102 dentry->d_inode = NULL;
103 list_del_init(&dentry->d_alias);
104 spin_unlock(&dentry->d_lock);
105 spin_unlock(&dcache_lock);
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 if (!atomic_read(&dentry->d_count)) {
347 __dget_locked(dentry);
349 spin_unlock(&dcache_lock);
354 spin_unlock(&dcache_lock);
358 * Throw away a dentry - free the inode, dput the parent.
359 * This requires that the LRU list has already been
361 * Called with dcache_lock, drops it and then regains.
363 static inline void prune_one_dentry(struct dentry * dentry)
365 struct dentry * parent;
368 list_del(&dentry->d_child);
369 dentry_stat.nr_dentry--; /* For d_free, below */
371 parent = dentry->d_parent;
373 if (parent != dentry)
375 spin_lock(&dcache_lock);
379 * prune_dcache - shrink the dcache
380 * @count: number of entries to try and free
382 * Shrink the dcache. This is done when we need
383 * more memory, or simply when we need to unmount
384 * something (at which point we need to unuse
387 * This function may fail to free any resources if
388 * all the dentries are in use.
391 static void prune_dcache(int count)
393 spin_lock(&dcache_lock);
394 for (; count ; count--) {
395 struct dentry *dentry;
396 struct list_head *tmp;
398 cond_resched_lock(&dcache_lock);
400 tmp = dentry_unused.prev;
401 if (tmp == &dentry_unused)
404 prefetch(dentry_unused.prev);
405 dentry_stat.nr_unused--;
406 dentry = list_entry(tmp, struct dentry, d_lru);
408 spin_lock(&dentry->d_lock);
410 * We found an inuse dentry which was not removed from
411 * dentry_unused because of laziness during lookup. Do not free
412 * it - just keep it off the dentry_unused list.
414 if (atomic_read(&dentry->d_count)) {
415 spin_unlock(&dentry->d_lock);
418 /* If the dentry was recently referenced, don't free it. */
419 if (dentry->d_flags & DCACHE_REFERENCED) {
420 dentry->d_flags &= ~DCACHE_REFERENCED;
421 list_add(&dentry->d_lru, &dentry_unused);
422 dentry_stat.nr_unused++;
423 spin_unlock(&dentry->d_lock);
426 prune_one_dentry(dentry);
428 spin_unlock(&dcache_lock);
432 * Shrink the dcache for the specified super block.
433 * This allows us to unmount a device without disturbing
434 * the dcache for the other devices.
436 * This implementation makes just two traversals of the
437 * unused list. On the first pass we move the selected
438 * dentries to the most recent end, and on the second
439 * pass we free them. The second pass must restart after
440 * each dput(), but since the target dentries are all at
441 * the end, it's really just a single traversal.
445 * shrink_dcache_sb - shrink dcache for a superblock
448 * Shrink the dcache for the specified super block. This
449 * is used to free the dcache before unmounting a file
453 void shrink_dcache_sb(struct super_block * sb)
455 struct list_head *tmp, *next;
456 struct dentry *dentry;
459 * Pass one ... move the dentries for the specified
460 * superblock to the most recent end of the unused list.
462 spin_lock(&dcache_lock);
463 next = dentry_unused.next;
464 while (next != &dentry_unused) {
467 dentry = list_entry(tmp, struct dentry, d_lru);
468 if (dentry->d_sb != sb)
471 list_add(tmp, &dentry_unused);
475 * Pass two ... free the dentries for this superblock.
478 next = dentry_unused.next;
479 while (next != &dentry_unused) {
482 dentry = list_entry(tmp, struct dentry, d_lru);
483 if (dentry->d_sb != sb)
485 dentry_stat.nr_unused--;
487 spin_lock(&dentry->d_lock);
488 if (atomic_read(&dentry->d_count)) {
489 spin_unlock(&dentry->d_lock);
492 prune_one_dentry(dentry);
495 spin_unlock(&dcache_lock);
499 * Search for at least 1 mount point in the dentry's subdirs.
500 * We descend to the next level whenever the d_subdirs
501 * list is non-empty and continue searching.
505 * have_submounts - check for mounts over a dentry
506 * @parent: dentry to check.
508 * Return true if the parent or its subdirectories contain
512 int have_submounts(struct dentry *parent)
514 struct dentry *this_parent = parent;
515 struct list_head *next;
517 spin_lock(&dcache_lock);
518 if (d_mountpoint(parent))
521 next = this_parent->d_subdirs.next;
523 while (next != &this_parent->d_subdirs) {
524 struct list_head *tmp = next;
525 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
527 /* Have we found a mount point ? */
528 if (d_mountpoint(dentry))
530 if (!list_empty(&dentry->d_subdirs)) {
531 this_parent = dentry;
536 * All done at this level ... ascend and resume the search.
538 if (this_parent != parent) {
539 next = this_parent->d_child.next;
540 this_parent = this_parent->d_parent;
543 spin_unlock(&dcache_lock);
544 return 0; /* No mount points found in tree */
546 spin_unlock(&dcache_lock);
551 * Search the dentry child list for the specified parent,
552 * and move any unused dentries to the end of the unused
553 * list for prune_dcache(). We descend to the next level
554 * whenever the d_subdirs list is non-empty and continue
557 static int select_parent(struct dentry * parent)
559 struct dentry *this_parent = parent;
560 struct list_head *next;
563 spin_lock(&dcache_lock);
565 next = this_parent->d_subdirs.next;
567 while (next != &this_parent->d_subdirs) {
568 struct list_head *tmp = next;
569 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
572 if (!list_empty(&dentry->d_lru)) {
573 dentry_stat.nr_unused--;
574 list_del_init(&dentry->d_lru);
577 * move only zero ref count dentries to the end
578 * of the unused list for prune_dcache
580 if (!atomic_read(&dentry->d_count)) {
581 list_add(&dentry->d_lru, dentry_unused.prev);
582 dentry_stat.nr_unused++;
586 * Descend a level if the d_subdirs list is non-empty.
588 if (!list_empty(&dentry->d_subdirs)) {
589 this_parent = dentry;
591 printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
592 dentry->d_parent->d_name.name, dentry->d_name.name, found);
598 * All done at this level ... ascend and resume the search.
600 if (this_parent != parent) {
601 next = this_parent->d_child.next;
602 this_parent = this_parent->d_parent;
604 printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
605 this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
609 spin_unlock(&dcache_lock);
614 * shrink_dcache_parent - prune dcache
615 * @parent: parent of entries to prune
617 * Prune the dcache to remove unused children of the parent dentry.
620 void shrink_dcache_parent(struct dentry * parent)
624 while ((found = select_parent(parent)) != 0)
629 * shrink_dcache_anon - further prune the cache
630 * @head: head of d_hash list of dentries to prune
632 * Prune the dentries that are anonymous
634 * parsing d_hash list does not hlist_for_each_rcu() as it
635 * done under dcache_lock.
638 void shrink_dcache_anon(struct hlist_head *head)
640 struct hlist_node *lp;
644 spin_lock(&dcache_lock);
645 hlist_for_each(lp, head) {
646 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
647 if (!list_empty(&this->d_lru)) {
648 dentry_stat.nr_unused--;
649 list_del_init(&this->d_lru);
653 * move only zero ref count dentries to the end
654 * of the unused list for prune_dcache
656 if (!atomic_read(&this->d_count)) {
657 list_add_tail(&this->d_lru, &dentry_unused);
658 dentry_stat.nr_unused++;
662 spin_unlock(&dcache_lock);
668 * Scan `nr' dentries and return the number which remain.
670 * We need to avoid reentering the filesystem if the caller is performing a
671 * GFP_NOFS allocation attempt. One example deadlock is:
673 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
674 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
675 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
677 * In this case we return -1 to tell the caller that we baled.
679 static int shrink_dcache_memory(int nr, unsigned int gfp_mask)
682 if (!(gfp_mask & __GFP_FS))
686 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
690 * d_alloc - allocate a dcache entry
691 * @parent: parent of entry to allocate
692 * @name: qstr of the name
694 * Allocates a dentry. It returns %NULL if there is insufficient memory
695 * available. On a success the dentry is returned. The name passed in is
696 * copied and the copy passed in may be reused after this call.
699 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
701 struct dentry *dentry;
704 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
708 if (name->len > DNAME_INLINE_LEN-1) {
709 dname = kmalloc(name->len + 1, GFP_KERNEL);
711 kmem_cache_free(dentry_cache, dentry);
715 dname = dentry->d_iname;
717 dentry->d_name.name = dname;
719 dentry->d_name.len = name->len;
720 dentry->d_name.hash = name->hash;
721 memcpy(dname, name->name, name->len);
722 dname[name->len] = 0;
724 atomic_set(&dentry->d_count, 1);
725 dentry->d_flags = DCACHE_UNHASHED;
726 dentry->d_lock = SPIN_LOCK_UNLOCKED;
727 dentry->d_inode = NULL;
728 dentry->d_parent = NULL;
731 dentry->d_fsdata = NULL;
732 dentry->d_extra_attributes = NULL;
733 dentry->d_mounted = 0;
734 dentry->d_cookie = NULL;
735 dentry->d_bucket = NULL;
736 INIT_HLIST_NODE(&dentry->d_hash);
737 INIT_LIST_HEAD(&dentry->d_lru);
738 INIT_LIST_HEAD(&dentry->d_subdirs);
739 INIT_LIST_HEAD(&dentry->d_alias);
742 dentry->d_parent = dget(parent);
743 dentry->d_sb = parent->d_sb;
745 INIT_LIST_HEAD(&dentry->d_child);
748 spin_lock(&dcache_lock);
750 list_add(&dentry->d_child, &parent->d_subdirs);
751 dentry_stat.nr_dentry++;
752 spin_unlock(&dcache_lock);
758 * d_instantiate - fill in inode information for a dentry
759 * @entry: dentry to complete
760 * @inode: inode to attach to this dentry
762 * Fill in inode information in the entry.
764 * This turns negative dentries into productive full members
767 * NOTE! This assumes that the inode count has been incremented
768 * (or otherwise set) by the caller to indicate that it is now
769 * in use by the dcache.
772 void d_instantiate(struct dentry *entry, struct inode * inode)
774 if (!list_empty(&entry->d_alias)) BUG();
775 spin_lock(&dcache_lock);
777 list_add(&entry->d_alias, &inode->i_dentry);
778 entry->d_inode = inode;
779 spin_unlock(&dcache_lock);
780 security_d_instantiate(entry, inode);
784 * d_alloc_root - allocate root dentry
785 * @root_inode: inode to allocate the root for
787 * Allocate a root ("/") dentry for the inode given. The inode is
788 * instantiated and returned. %NULL is returned if there is insufficient
789 * memory or the inode passed is %NULL.
792 struct dentry * d_alloc_root(struct inode * root_inode)
794 struct dentry *res = NULL;
797 static const struct qstr name = { .name = "/", .len = 1 };
799 res = d_alloc(NULL, &name);
801 res->d_sb = root_inode->i_sb;
803 d_instantiate(res, root_inode);
809 static inline struct hlist_head *d_hash(struct dentry *parent,
812 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
813 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
814 return dentry_hashtable + (hash & D_HASHMASK);
818 * d_alloc_anon - allocate an anonymous dentry
819 * @inode: inode to allocate the dentry for
821 * This is similar to d_alloc_root. It is used by filesystems when
822 * creating a dentry for a given inode, often in the process of
823 * mapping a filehandle to a dentry. The returned dentry may be
824 * anonymous, or may have a full name (if the inode was already
825 * in the cache). The file system may need to make further
826 * efforts to connect this dentry into the dcache properly.
828 * When called on a directory inode, we must ensure that
829 * the inode only ever has one dentry. If a dentry is
830 * found, that is returned instead of allocating a new one.
832 * On successful return, the reference to the inode has been transferred
833 * to the dentry. If %NULL is returned (indicating kmalloc failure),
834 * the reference on the inode has not been released.
837 struct dentry * d_alloc_anon(struct inode *inode)
839 static const struct qstr anonstring = { .name = "" };
843 if ((res = d_find_alias(inode))) {
848 tmp = d_alloc(NULL, &anonstring);
852 tmp->d_parent = tmp; /* make sure dput doesn't croak */
854 spin_lock(&dcache_lock);
855 res = __d_find_alias(inode, 0);
857 /* attach a disconnected dentry */
860 spin_lock(&res->d_lock);
861 res->d_sb = inode->i_sb;
863 res->d_inode = inode;
866 * Set d_bucket to an "impossible" bucket address so
867 * that d_move() doesn't get a false positive
869 res->d_bucket = NULL;
870 res->d_flags |= DCACHE_DISCONNECTED;
871 res->d_flags &= ~DCACHE_UNHASHED;
872 list_add(&res->d_alias, &inode->i_dentry);
873 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
874 spin_unlock(&res->d_lock);
876 inode = NULL; /* don't drop reference */
878 spin_unlock(&dcache_lock);
889 * d_splice_alias - splice a disconnected dentry into the tree if one exists
890 * @inode: the inode which may have a disconnected dentry
891 * @dentry: a negative dentry which we want to point to the inode.
893 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
894 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
895 * and return it, else simply d_add the inode to the dentry and return NULL.
897 * This is needed in the lookup routine of any filesystem that is exportable
898 * (via knfsd) so that we can build dcache paths to directories effectively.
900 * If a dentry was found and moved, then it is returned. Otherwise NULL
901 * is returned. This matches the expected return value of ->lookup.
904 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
906 struct dentry *new = NULL;
909 spin_lock(&dcache_lock);
910 new = __d_find_alias(inode, 1);
912 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
913 spin_unlock(&dcache_lock);
914 security_d_instantiate(new, inode);
919 /* d_instantiate takes dcache_lock, so we do it by hand */
920 list_add(&dentry->d_alias, &inode->i_dentry);
921 dentry->d_inode = inode;
922 spin_unlock(&dcache_lock);
923 security_d_instantiate(dentry, inode);
927 d_add(dentry, inode);
933 * d_lookup - search for a dentry
934 * @parent: parent dentry
935 * @name: qstr of name we wish to find
937 * Searches the children of the parent dentry for the name in question. If
938 * the dentry is found its reference count is incremented and the dentry
939 * is returned. The caller must use d_put to free the entry when it has
940 * finished using it. %NULL is returned on failure.
942 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
943 * Memory barriers are used while updating and doing lockless traversal.
944 * To avoid races with d_move while rename is happening, d_lock is used.
946 * Overflows in memcmp(), while d_move, are avoided by keeping the length
947 * and name pointer in one structure pointed by d_qstr.
949 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
950 * lookup is going on.
952 * dentry_unused list is not updated even if lookup finds the required dentry
953 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
954 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
957 * d_lookup() is protected against the concurrent renames in some unrelated
958 * directory using the seqlockt_t rename_lock.
961 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
963 struct dentry * dentry = NULL;
967 seq = read_seqbegin(&rename_lock);
968 dentry = __d_lookup(parent, name);
971 } while (read_seqretry(&rename_lock, seq));
975 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
977 unsigned int len = name->len;
978 unsigned int hash = name->hash;
979 const unsigned char *str = name->name;
980 struct hlist_head *head = d_hash(parent,hash);
981 struct dentry *found = NULL;
982 struct hlist_node *node;
986 hlist_for_each_rcu(node, head) {
987 struct dentry *dentry;
990 dentry = hlist_entry(node, struct dentry, d_hash);
994 if (dentry->d_name.hash != hash)
996 if (dentry->d_parent != parent)
999 spin_lock(&dentry->d_lock);
1002 * If lookup ends up in a different bucket due to concurrent
1005 if (unlikely(dentry->d_bucket != head))
1009 * Recheck the dentry after taking the lock - d_move may have
1010 * changed things. Don't bother checking the hash because we're
1011 * about to compare the whole name anyway.
1013 if (dentry->d_parent != parent)
1016 qstr = rcu_dereference(&dentry->d_name);
1017 if (parent->d_op && parent->d_op->d_compare) {
1018 if (parent->d_op->d_compare(parent, qstr, name))
1021 if (qstr->len != len)
1023 if (memcmp(qstr->name, str, len))
1027 if (!d_unhashed(dentry)) {
1028 atomic_inc(&dentry->d_count);
1032 spin_unlock(&dentry->d_lock);
1035 spin_unlock(&dentry->d_lock);
1043 * d_validate - verify dentry provided from insecure source
1044 * @dentry: The dentry alleged to be valid child of @dparent
1045 * @dparent: The parent dentry (known to be valid)
1046 * @hash: Hash of the dentry
1047 * @len: Length of the name
1049 * An insecure source has sent us a dentry, here we verify it and dget() it.
1050 * This is used by ncpfs in its readdir implementation.
1051 * Zero is returned in the dentry is invalid.
1054 int d_validate(struct dentry *dentry, struct dentry *dparent)
1056 struct hlist_head *base;
1057 struct hlist_node *lhp;
1059 /* Check whether the ptr might be valid at all.. */
1060 if (!kmem_ptr_validate(dentry_cache, dentry))
1063 if (dentry->d_parent != dparent)
1066 spin_lock(&dcache_lock);
1067 base = d_hash(dparent, dentry->d_name.hash);
1068 hlist_for_each(lhp,base) {
1069 /* hlist_for_each_rcu() not required for d_hash list
1070 * as it is parsed under dcache_lock
1072 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1073 __dget_locked(dentry);
1074 spin_unlock(&dcache_lock);
1078 spin_unlock(&dcache_lock);
1084 * When a file is deleted, we have two options:
1085 * - turn this dentry into a negative dentry
1086 * - unhash this dentry and free it.
1088 * Usually, we want to just turn this into
1089 * a negative dentry, but if anybody else is
1090 * currently using the dentry or the inode
1091 * we can't do that and we fall back on removing
1092 * it from the hash queues and waiting for
1093 * it to be deleted later when it has no users
1097 * d_delete - delete a dentry
1098 * @dentry: The dentry to delete
1100 * Turn the dentry into a negative dentry if possible, otherwise
1101 * remove it from the hash queues so it can be deleted later
1104 void d_delete(struct dentry * dentry)
1107 * Are we the only user?
1109 spin_lock(&dcache_lock);
1110 spin_lock(&dentry->d_lock);
1111 if (atomic_read(&dentry->d_count) == 1) {
1112 dentry_iput(dentry);
1116 if (!d_unhashed(dentry))
1119 spin_unlock(&dentry->d_lock);
1120 spin_unlock(&dcache_lock);
1124 * d_rehash - add an entry back to the hash
1125 * @entry: dentry to add to the hash
1127 * Adds a dentry to the hash according to its name.
1130 void d_rehash(struct dentry * entry)
1132 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1134 spin_lock(&dcache_lock);
1135 spin_lock(&entry->d_lock);
1136 entry->d_flags &= ~DCACHE_UNHASHED;
1137 spin_unlock(&entry->d_lock);
1138 entry->d_bucket = list;
1139 hlist_add_head_rcu(&entry->d_hash, list);
1140 spin_unlock(&dcache_lock);
1143 #define do_switch(x,y) do { \
1144 __typeof__ (x) __tmp = x; \
1145 x = y; y = __tmp; } while (0)
1148 * When switching names, the actual string doesn't strictly have to
1149 * be preserved in the target - because we're dropping the target
1150 * anyway. As such, we can just do a simple memcpy() to copy over
1151 * the new name before we switch.
1153 * Note that we have to be a lot more careful about getting the hash
1154 * switched - we have to switch the hash value properly even if it
1155 * then no longer matches the actual (corrupted) string of the target.
1156 * The hash value has to match the hash queue that the dentry is on..
1158 static void switch_names(struct dentry *dentry, struct dentry *target)
1160 if (dname_external(target)) {
1161 if (dname_external(dentry)) {
1163 * Both external: swap the pointers
1165 do_switch(target->d_name.name, dentry->d_name.name);
1168 * dentry:internal, target:external. Steal target's
1169 * storage and make target internal.
1171 dentry->d_name.name = target->d_name.name;
1172 target->d_name.name = target->d_iname;
1175 if (dname_external(dentry)) {
1177 * dentry:external, target:internal. Give dentry's
1178 * storage to target and make dentry internal
1180 memcpy(dentry->d_iname, target->d_name.name,
1181 target->d_name.len + 1);
1182 target->d_name.name = dentry->d_name.name;
1183 dentry->d_name.name = dentry->d_iname;
1186 * Both are internal. Just copy target to dentry
1188 memcpy(dentry->d_iname, target->d_name.name,
1189 target->d_name.len + 1);
1195 * We cannibalize "target" when moving dentry on top of it,
1196 * because it's going to be thrown away anyway. We could be more
1197 * polite about it, though.
1199 * This forceful removal will result in ugly /proc output if
1200 * somebody holds a file open that got deleted due to a rename.
1201 * We could be nicer about the deleted file, and let it show
1202 * up under the name it got deleted rather than the name that
1207 * d_move - move a dentry
1208 * @dentry: entry to move
1209 * @target: new dentry
1211 * Update the dcache to reflect the move of a file name. Negative
1212 * dcache entries should not be moved in this way.
1215 void d_move(struct dentry * dentry, struct dentry * target)
1217 if (!dentry->d_inode)
1218 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1220 spin_lock(&dcache_lock);
1221 write_seqlock(&rename_lock);
1223 * XXXX: do we really need to take target->d_lock?
1225 if (target < dentry) {
1226 spin_lock(&target->d_lock);
1227 spin_lock(&dentry->d_lock);
1229 spin_lock(&dentry->d_lock);
1230 spin_lock(&target->d_lock);
1233 /* Move the dentry to the target hash queue, if on different bucket */
1234 if (dentry->d_flags & DCACHE_UNHASHED)
1235 goto already_unhashed;
1236 if (dentry->d_bucket != target->d_bucket) {
1237 hlist_del_rcu(&dentry->d_hash);
1239 dentry->d_bucket = target->d_bucket;
1240 hlist_add_head_rcu(&dentry->d_hash, target->d_bucket);
1241 dentry->d_flags &= ~DCACHE_UNHASHED;
1244 /* Unhash the target: dput() will then get rid of it */
1247 /* flush any possible attributes */
1248 if (dentry->d_extra_attributes) {
1249 kfree(dentry->d_extra_attributes);
1250 dentry->d_extra_attributes = NULL;
1252 if (target->d_extra_attributes) {
1253 kfree(target->d_extra_attributes);
1254 target->d_extra_attributes = NULL;
1257 list_del(&dentry->d_child);
1258 list_del(&target->d_child);
1260 /* Switch the names.. */
1261 switch_names(dentry, target);
1263 do_switch(dentry->d_name.len, target->d_name.len);
1264 do_switch(dentry->d_name.hash, target->d_name.hash);
1266 /* ... and switch the parents */
1267 if (IS_ROOT(dentry)) {
1268 dentry->d_parent = target->d_parent;
1269 target->d_parent = target;
1270 INIT_LIST_HEAD(&target->d_child);
1272 do_switch(dentry->d_parent, target->d_parent);
1274 /* And add them back to the (new) parent lists */
1275 list_add(&target->d_child, &target->d_parent->d_subdirs);
1278 list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
1279 spin_unlock(&target->d_lock);
1280 spin_unlock(&dentry->d_lock);
1281 write_sequnlock(&rename_lock);
1282 spin_unlock(&dcache_lock);
1286 * d_path - return the path of a dentry
1287 * @dentry: dentry to report
1288 * @vfsmnt: vfsmnt to which the dentry belongs
1289 * @root: root dentry
1290 * @rootmnt: vfsmnt to which the root dentry belongs
1291 * @buffer: buffer to return value in
1292 * @buflen: buffer length
1294 * Convert a dentry into an ASCII path name. If the entry has been deleted
1295 * the string " (deleted)" is appended. Note that this is ambiguous.
1297 * Returns the buffer or an error code if the path was too long.
1299 * "buflen" should be positive. Caller holds the dcache_lock.
1301 char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1302 struct dentry *root, struct vfsmount *rootmnt,
1303 char *buffer, int buflen)
1305 char * end = buffer+buflen;
1311 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1316 memcpy(end, " (deleted)", 10);
1326 struct dentry * parent;
1328 if (dentry == root && vfsmnt == rootmnt)
1330 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1332 spin_lock(&vfsmount_lock);
1333 if (vfsmnt->mnt_parent == vfsmnt) {
1334 spin_unlock(&vfsmount_lock);
1337 dentry = vfsmnt->mnt_mountpoint;
1338 vfsmnt = vfsmnt->mnt_parent;
1339 spin_unlock(&vfsmount_lock);
1342 parent = dentry->d_parent;
1344 namelen = dentry->d_name.len;
1345 buflen -= namelen + 1;
1349 memcpy(end, dentry->d_name.name, namelen);
1358 namelen = dentry->d_name.len;
1362 retval -= namelen-1; /* hit the slash */
1363 memcpy(retval, dentry->d_name.name, namelen);
1366 return ERR_PTR(-ENAMETOOLONG);
1369 EXPORT_SYMBOL_GPL(__d_path);
1371 /* write full pathname into buffer and return start of pathname */
1372 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1373 char *buf, int buflen)
1376 struct vfsmount *rootmnt;
1377 struct dentry *root;
1379 read_lock(¤t->fs->lock);
1380 rootmnt = mntget(current->fs->rootmnt);
1381 root = dget(current->fs->root);
1382 read_unlock(¤t->fs->lock);
1383 spin_lock(&dcache_lock);
1384 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1385 spin_unlock(&dcache_lock);
1392 * NOTE! The user-level library version returns a
1393 * character pointer. The kernel system call just
1394 * returns the length of the buffer filled (which
1395 * includes the ending '\0' character), or a negative
1396 * error value. So libc would do something like
1398 * char *getcwd(char * buf, size_t size)
1402 * retval = sys_getcwd(buf, size);
1409 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1412 struct vfsmount *pwdmnt, *rootmnt;
1413 struct dentry *pwd, *root;
1414 char *page = (char *) __get_free_page(GFP_USER);
1419 read_lock(¤t->fs->lock);
1420 pwdmnt = mntget(current->fs->pwdmnt);
1421 pwd = dget(current->fs->pwd);
1422 rootmnt = mntget(current->fs->rootmnt);
1423 root = dget(current->fs->root);
1424 read_unlock(¤t->fs->lock);
1427 /* Has the current directory has been unlinked? */
1428 spin_lock(&dcache_lock);
1429 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1433 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1434 spin_unlock(&dcache_lock);
1436 error = PTR_ERR(cwd);
1441 len = PAGE_SIZE + page - cwd;
1444 if (copy_to_user(buf, cwd, len))
1448 spin_unlock(&dcache_lock);
1455 free_page((unsigned long) page);
1460 * Test whether new_dentry is a subdirectory of old_dentry.
1462 * Trivially implemented using the dcache structure
1466 * is_subdir - is new dentry a subdirectory of old_dentry
1467 * @new_dentry: new dentry
1468 * @old_dentry: old dentry
1470 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1471 * Returns 0 otherwise.
1472 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1475 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1478 struct dentry * saved = new_dentry;
1482 /* need rcu_readlock to protect against the d_parent trashing due to
1487 /* for restarting inner loop in case of seq retry */
1489 seq = read_seqbegin(&rename_lock);
1491 if (new_dentry != old_dentry) {
1492 struct dentry * parent = new_dentry->d_parent;
1493 if (parent == new_dentry)
1495 new_dentry = parent;
1501 } while (read_seqretry(&rename_lock, seq));
1507 void d_genocide(struct dentry *root)
1509 struct dentry *this_parent = root;
1510 struct list_head *next;
1512 spin_lock(&dcache_lock);
1514 next = this_parent->d_subdirs.next;
1516 while (next != &this_parent->d_subdirs) {
1517 struct list_head *tmp = next;
1518 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1520 if (d_unhashed(dentry)||!dentry->d_inode)
1522 if (!list_empty(&dentry->d_subdirs)) {
1523 this_parent = dentry;
1526 atomic_dec(&dentry->d_count);
1528 if (this_parent != root) {
1529 next = this_parent->d_child.next;
1530 atomic_dec(&this_parent->d_count);
1531 this_parent = this_parent->d_parent;
1534 spin_unlock(&dcache_lock);
1538 * find_inode_number - check for dentry with name
1539 * @dir: directory to check
1540 * @name: Name to find.
1542 * Check whether a dentry already exists for the given name,
1543 * and return the inode number if it has an inode. Otherwise
1546 * This routine is used to post-process directory listings for
1547 * filesystems using synthetic inode numbers, and is necessary
1548 * to keep getcwd() working.
1551 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1553 struct dentry * dentry;
1557 * Check for a fs-specific hash function. Note that we must
1558 * calculate the standard hash first, as the d_op->d_hash()
1559 * routine may choose to leave the hash value unchanged.
1561 name->hash = full_name_hash(name->name, name->len);
1562 if (dir->d_op && dir->d_op->d_hash)
1564 if (dir->d_op->d_hash(dir, name) != 0)
1568 dentry = d_lookup(dir, name);
1571 if (dentry->d_inode)
1572 ino = dentry->d_inode->i_ino;
1579 static __initdata unsigned long dhash_entries;
1580 static int __init set_dhash_entries(char *str)
1584 dhash_entries = simple_strtoul(str, &str, 0);
1587 __setup("dhash_entries=", set_dhash_entries);
1589 static void __init dcache_init_early(void)
1594 alloc_large_system_hash("Dentry cache",
1595 sizeof(struct hlist_head),
1602 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1603 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1606 void flush_dentry_attributes (void)
1608 struct hlist_node *tmp;
1609 struct dentry *dentry;
1612 spin_lock(&dcache_lock);
1613 for (i = 0; i <= d_hash_mask; i++)
1614 hlist_for_each_entry(dentry, tmp, dentry_hashtable+i, d_hash) {
1615 kfree(dentry->d_extra_attributes);
1616 dentry->d_extra_attributes = NULL;
1618 spin_unlock(&dcache_lock);
1621 EXPORT_SYMBOL_GPL(flush_dentry_attributes);
1623 static void __init dcache_init(unsigned long mempages)
1626 * A constructor could be added for stable state like the lists,
1627 * but it is probably not worth it because of the cache nature
1630 dentry_cache = kmem_cache_create("dentry_cache",
1631 sizeof(struct dentry),
1633 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1636 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1639 /* SLAB cache for __getname() consumers */
1640 kmem_cache_t *names_cachep;
1642 /* SLAB cache for file structures */
1643 kmem_cache_t *filp_cachep;
1645 EXPORT_SYMBOL(d_genocide);
1647 extern void bdev_cache_init(void);
1648 extern void chrdev_init(void);
1650 void __init vfs_caches_init_early(void)
1652 dcache_init_early();
1656 void __init vfs_caches_init(unsigned long mempages)
1658 unsigned long reserve;
1660 /* Base hash sizes on available memory, with a reserve equal to
1661 150% of current kernel size */
1663 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1664 mempages -= reserve;
1666 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1667 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1669 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1670 SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
1672 dcache_init(mempages);
1673 inode_init(mempages);
1674 files_init(mempages);
1680 EXPORT_SYMBOL(d_alloc);
1681 EXPORT_SYMBOL(d_alloc_anon);
1682 EXPORT_SYMBOL(d_alloc_root);
1683 EXPORT_SYMBOL(d_delete);
1684 EXPORT_SYMBOL(d_find_alias);
1685 EXPORT_SYMBOL(d_instantiate);
1686 EXPORT_SYMBOL(d_invalidate);
1687 EXPORT_SYMBOL(d_lookup);
1688 EXPORT_SYMBOL(d_move);
1689 EXPORT_SYMBOL(d_path);
1690 EXPORT_SYMBOL(d_prune_aliases);
1691 EXPORT_SYMBOL(d_rehash);
1692 EXPORT_SYMBOL(d_splice_alias);
1693 EXPORT_SYMBOL(d_validate);
1694 EXPORT_SYMBOL(dget_locked);
1695 EXPORT_SYMBOL(dput);
1696 EXPORT_SYMBOL(find_inode_number);
1697 EXPORT_SYMBOL(have_submounts);
1698 EXPORT_SYMBOL(names_cachep);
1699 EXPORT_SYMBOL(shrink_dcache_parent);
1700 EXPORT_SYMBOL(shrink_dcache_sb);