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>
42 * dcache_inode_lock protects:
43 * - i_dentry, d_alias, d_inode
44 * dcache_hash_bucket lock protects:
45 * - the dcache hash table
46 * s_anon bl list spinlock protects:
47 * - the s_anon list (see __d_drop)
48 * dcache_lru_lock protects:
49 * - the dcache lru lists and counters
56 * - d_parent and d_subdirs
57 * - childrens' d_child and d_parent
64 * dcache_hash_bucket lock
67 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock
70 * dentry->d_parent->d_lock
73 * If no ancestor relationship:
74 * if (dentry1 < dentry2)
78 int sysctl_vfs_cache_pressure __read_mostly = 100;
79 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
81 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_inode_lock);
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);
86 EXPORT_SYMBOL(dcache_inode_lock);
88 static struct kmem_cache *dentry_cache __read_mostly;
91 * This is the single most critical data structure when it comes
92 * to the dcache: the hashtable for lookups. Somebody should try
93 * to make this good - I've just made it work.
95 * This hash-function tries to avoid losing too many bits of hash
96 * information, yet avoid using a prime hash-size or similar.
98 #define D_HASHBITS d_hash_shift
99 #define D_HASHMASK d_hash_mask
101 static unsigned int d_hash_mask __read_mostly;
102 static unsigned int d_hash_shift __read_mostly;
104 struct dcache_hash_bucket {
105 struct hlist_bl_head head;
107 static struct dcache_hash_bucket *dentry_hashtable __read_mostly;
109 static inline struct dcache_hash_bucket *d_hash(struct dentry *parent,
112 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
113 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
114 return dentry_hashtable + (hash & D_HASHMASK);
117 static inline void spin_lock_bucket(struct dcache_hash_bucket *b)
119 bit_spin_lock(0, (unsigned long *)&b->head.first);
122 static inline void spin_unlock_bucket(struct dcache_hash_bucket *b)
124 __bit_spin_unlock(0, (unsigned long *)&b->head.first);
127 /* Statistics gathering. */
128 struct dentry_stat_t dentry_stat = {
132 static DEFINE_PER_CPU(unsigned int, nr_dentry);
134 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
135 static int get_nr_dentry(void)
139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
144 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
145 size_t *lenp, loff_t *ppos)
147 dentry_stat.nr_dentry = get_nr_dentry();
148 return proc_dointvec(table, write, buffer, lenp, ppos);
152 static void __d_free(struct rcu_head *head)
154 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
156 WARN_ON(!list_empty(&dentry->d_alias));
157 if (dname_external(dentry))
158 kfree(dentry->d_name.name);
159 kmem_cache_free(dentry_cache, dentry);
165 static void d_free(struct dentry *dentry)
167 BUG_ON(dentry->d_count);
168 this_cpu_dec(nr_dentry);
169 if (dentry->d_op && dentry->d_op->d_release)
170 dentry->d_op->d_release(dentry);
172 /* if dentry was never inserted into hash, immediate free is OK */
173 if (hlist_bl_unhashed(&dentry->d_hash))
174 __d_free(&dentry->d_u.d_rcu);
176 call_rcu(&dentry->d_u.d_rcu, __d_free);
180 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
181 * After this call, in-progress rcu-walk path lookup will fail. This
182 * should be called after unhashing, and after changing d_inode (if
183 * the dentry has not already been unhashed).
185 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
187 assert_spin_locked(&dentry->d_lock);
188 /* Go through a barrier */
189 write_seqcount_barrier(&dentry->d_seq);
193 * Release the dentry's inode, using the filesystem
194 * d_iput() operation if defined. Dentry has no refcount
197 static void dentry_iput(struct dentry * dentry)
198 __releases(dentry->d_lock)
199 __releases(dcache_inode_lock)
201 struct inode *inode = dentry->d_inode;
203 dentry->d_inode = NULL;
204 list_del_init(&dentry->d_alias);
205 spin_unlock(&dentry->d_lock);
206 spin_unlock(&dcache_inode_lock);
208 fsnotify_inoderemove(inode);
209 if (dentry->d_op && dentry->d_op->d_iput)
210 dentry->d_op->d_iput(dentry, inode);
214 spin_unlock(&dentry->d_lock);
215 spin_unlock(&dcache_inode_lock);
220 * Release the dentry's inode, using the filesystem
221 * d_iput() operation if defined. dentry remains in-use.
223 static void dentry_unlink_inode(struct dentry * dentry)
224 __releases(dentry->d_lock)
225 __releases(dcache_inode_lock)
227 struct inode *inode = dentry->d_inode;
228 dentry->d_inode = NULL;
229 list_del_init(&dentry->d_alias);
230 dentry_rcuwalk_barrier(dentry);
231 spin_unlock(&dentry->d_lock);
232 spin_unlock(&dcache_inode_lock);
234 fsnotify_inoderemove(inode);
235 if (dentry->d_op && dentry->d_op->d_iput)
236 dentry->d_op->d_iput(dentry, inode);
242 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
244 static void dentry_lru_add(struct dentry *dentry)
246 if (list_empty(&dentry->d_lru)) {
247 spin_lock(&dcache_lru_lock);
248 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
249 dentry->d_sb->s_nr_dentry_unused++;
250 dentry_stat.nr_unused++;
251 spin_unlock(&dcache_lru_lock);
255 static void __dentry_lru_del(struct dentry *dentry)
257 list_del_init(&dentry->d_lru);
258 dentry->d_sb->s_nr_dentry_unused--;
259 dentry_stat.nr_unused--;
262 static void dentry_lru_del(struct dentry *dentry)
264 if (!list_empty(&dentry->d_lru)) {
265 spin_lock(&dcache_lru_lock);
266 __dentry_lru_del(dentry);
267 spin_unlock(&dcache_lru_lock);
271 static void dentry_lru_move_tail(struct dentry *dentry)
273 spin_lock(&dcache_lru_lock);
274 if (list_empty(&dentry->d_lru)) {
275 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
276 dentry->d_sb->s_nr_dentry_unused++;
277 dentry_stat.nr_unused++;
279 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
281 spin_unlock(&dcache_lru_lock);
285 * d_kill - kill dentry and return parent
286 * @dentry: dentry to kill
288 * The dentry must already be unhashed and removed from the LRU.
290 * If this is the root of the dentry tree, return NULL.
292 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
295 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
296 __releases(dentry->d_lock)
297 __releases(parent->d_lock)
298 __releases(dcache_inode_lock)
300 dentry->d_parent = NULL;
301 list_del(&dentry->d_u.d_child);
303 spin_unlock(&parent->d_lock);
306 * dentry_iput drops the locks, at which point nobody (except
307 * transient RCU lookups) can reach this dentry.
314 * d_drop - drop a dentry
315 * @dentry: dentry to drop
317 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
318 * be found through a VFS lookup any more. Note that this is different from
319 * deleting the dentry - d_delete will try to mark the dentry negative if
320 * possible, giving a successful _negative_ lookup, while d_drop will
321 * just make the cache lookup fail.
323 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
324 * reason (NFS timeouts or autofs deletes).
326 * __d_drop requires dentry->d_lock.
328 void __d_drop(struct dentry *dentry)
330 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
331 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) {
333 (unsigned long *)&dentry->d_sb->s_anon.first);
334 dentry->d_flags |= DCACHE_UNHASHED;
335 hlist_bl_del_init(&dentry->d_hash);
337 (unsigned long *)&dentry->d_sb->s_anon.first);
339 struct dcache_hash_bucket *b;
340 b = d_hash(dentry->d_parent, dentry->d_name.hash);
343 * We may not actually need to put DCACHE_UNHASHED
344 * manipulations under the hash lock, but follow
345 * the principle of least surprise.
347 dentry->d_flags |= DCACHE_UNHASHED;
348 hlist_bl_del_rcu(&dentry->d_hash);
349 spin_unlock_bucket(b);
350 dentry_rcuwalk_barrier(dentry);
354 EXPORT_SYMBOL(__d_drop);
356 void d_drop(struct dentry *dentry)
358 spin_lock(&dentry->d_lock);
360 spin_unlock(&dentry->d_lock);
362 EXPORT_SYMBOL(d_drop);
365 * Finish off a dentry we've decided to kill.
366 * dentry->d_lock must be held, returns with it unlocked.
367 * If ref is non-zero, then decrement the refcount too.
368 * Returns dentry requiring refcount drop, or NULL if we're done.
370 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
371 __releases(dentry->d_lock)
373 struct dentry *parent;
375 if (!spin_trylock(&dcache_inode_lock)) {
377 spin_unlock(&dentry->d_lock);
379 return dentry; /* try again with same dentry */
384 parent = dentry->d_parent;
385 if (parent && !spin_trylock(&parent->d_lock)) {
386 spin_unlock(&dcache_inode_lock);
392 /* if dentry was on the d_lru list delete it from there */
393 dentry_lru_del(dentry);
394 /* if it was on the hash then remove it */
396 return d_kill(dentry, parent);
402 * This is complicated by the fact that we do not want to put
403 * dentries that are no longer on any hash chain on the unused
404 * list: we'd much rather just get rid of them immediately.
406 * However, that implies that we have to traverse the dentry
407 * tree upwards to the parents which might _also_ now be
408 * scheduled for deletion (it may have been only waiting for
409 * its last child to go away).
411 * This tail recursion is done by hand as we don't want to depend
412 * on the compiler to always get this right (gcc generally doesn't).
413 * Real recursion would eat up our stack space.
417 * dput - release a dentry
418 * @dentry: dentry to release
420 * Release a dentry. This will drop the usage count and if appropriate
421 * call the dentry unlink method as well as removing it from the queues and
422 * releasing its resources. If the parent dentries were scheduled for release
423 * they too may now get deleted.
425 void dput(struct dentry *dentry)
431 if (dentry->d_count == 1)
433 spin_lock(&dentry->d_lock);
434 BUG_ON(!dentry->d_count);
435 if (dentry->d_count > 1) {
437 spin_unlock(&dentry->d_lock);
441 if (dentry->d_flags & DCACHE_OP_DELETE) {
442 if (dentry->d_op->d_delete(dentry))
446 /* Unreachable? Get rid of it */
447 if (d_unhashed(dentry))
450 /* Otherwise leave it cached and ensure it's on the LRU */
451 dentry->d_flags |= DCACHE_REFERENCED;
452 dentry_lru_add(dentry);
455 spin_unlock(&dentry->d_lock);
459 dentry = dentry_kill(dentry, 1);
466 * d_invalidate - invalidate a dentry
467 * @dentry: dentry to invalidate
469 * Try to invalidate the dentry if it turns out to be
470 * possible. If there are other dentries that can be
471 * reached through this one we can't delete it and we
472 * return -EBUSY. On success we return 0.
477 int d_invalidate(struct dentry * dentry)
480 * If it's already been dropped, return OK.
482 spin_lock(&dentry->d_lock);
483 if (d_unhashed(dentry)) {
484 spin_unlock(&dentry->d_lock);
488 * Check whether to do a partial shrink_dcache
489 * to get rid of unused child entries.
491 if (!list_empty(&dentry->d_subdirs)) {
492 spin_unlock(&dentry->d_lock);
493 shrink_dcache_parent(dentry);
494 spin_lock(&dentry->d_lock);
498 * Somebody else still using it?
500 * If it's a directory, we can't drop it
501 * for fear of somebody re-populating it
502 * with children (even though dropping it
503 * would make it unreachable from the root,
504 * we might still populate it if it was a
505 * working directory or similar).
507 if (dentry->d_count > 1) {
508 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
509 spin_unlock(&dentry->d_lock);
515 spin_unlock(&dentry->d_lock);
518 EXPORT_SYMBOL(d_invalidate);
520 /* This must be called with d_lock held */
521 static inline void __dget_dlock(struct dentry *dentry)
526 static inline void __dget(struct dentry *dentry)
528 spin_lock(&dentry->d_lock);
529 __dget_dlock(dentry);
530 spin_unlock(&dentry->d_lock);
533 struct dentry *dget_parent(struct dentry *dentry)
539 * Don't need rcu_dereference because we re-check it was correct under
543 ret = dentry->d_parent;
548 spin_lock(&ret->d_lock);
549 if (unlikely(ret != dentry->d_parent)) {
550 spin_unlock(&ret->d_lock);
555 BUG_ON(!ret->d_count);
557 spin_unlock(&ret->d_lock);
561 EXPORT_SYMBOL(dget_parent);
564 * d_find_alias - grab a hashed alias of inode
565 * @inode: inode in question
566 * @want_discon: flag, used by d_splice_alias, to request
567 * that only a DISCONNECTED alias be returned.
569 * If inode has a hashed alias, or is a directory and has any alias,
570 * acquire the reference to alias and return it. Otherwise return NULL.
571 * Notice that if inode is a directory there can be only one alias and
572 * it can be unhashed only if it has no children, or if it is the root
575 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
576 * any other hashed alias over that one unless @want_discon is set,
577 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
579 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
581 struct dentry *alias, *discon_alias;
585 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
586 spin_lock(&alias->d_lock);
587 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
588 if (IS_ROOT(alias) &&
589 (alias->d_flags & DCACHE_DISCONNECTED)) {
590 discon_alias = alias;
591 } else if (!want_discon) {
593 spin_unlock(&alias->d_lock);
597 spin_unlock(&alias->d_lock);
600 alias = discon_alias;
601 spin_lock(&alias->d_lock);
602 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
603 if (IS_ROOT(alias) &&
604 (alias->d_flags & DCACHE_DISCONNECTED)) {
606 spin_unlock(&alias->d_lock);
610 spin_unlock(&alias->d_lock);
616 struct dentry *d_find_alias(struct inode *inode)
618 struct dentry *de = NULL;
620 if (!list_empty(&inode->i_dentry)) {
621 spin_lock(&dcache_inode_lock);
622 de = __d_find_alias(inode, 0);
623 spin_unlock(&dcache_inode_lock);
627 EXPORT_SYMBOL(d_find_alias);
630 * Try to kill dentries associated with this inode.
631 * WARNING: you must own a reference to inode.
633 void d_prune_aliases(struct inode *inode)
635 struct dentry *dentry;
637 spin_lock(&dcache_inode_lock);
638 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
639 spin_lock(&dentry->d_lock);
640 if (!dentry->d_count) {
641 __dget_dlock(dentry);
643 spin_unlock(&dentry->d_lock);
644 spin_unlock(&dcache_inode_lock);
648 spin_unlock(&dentry->d_lock);
650 spin_unlock(&dcache_inode_lock);
652 EXPORT_SYMBOL(d_prune_aliases);
655 * Try to throw away a dentry - free the inode, dput the parent.
656 * Requires dentry->d_lock is held, and dentry->d_count == 0.
657 * Releases dentry->d_lock.
659 * This may fail if locks cannot be acquired no problem, just try again.
661 static void try_prune_one_dentry(struct dentry *dentry)
662 __releases(dentry->d_lock)
664 struct dentry *parent;
666 parent = dentry_kill(dentry, 0);
668 * If dentry_kill returns NULL, we have nothing more to do.
669 * if it returns the same dentry, trylocks failed. In either
670 * case, just loop again.
672 * Otherwise, we need to prune ancestors too. This is necessary
673 * to prevent quadratic behavior of shrink_dcache_parent(), but
674 * is also expected to be beneficial in reducing dentry cache
679 if (parent == dentry)
682 /* Prune ancestors. */
685 spin_lock(&dentry->d_lock);
686 if (dentry->d_count > 1) {
688 spin_unlock(&dentry->d_lock);
691 dentry = dentry_kill(dentry, 1);
695 static void shrink_dentry_list(struct list_head *list)
697 struct dentry *dentry;
701 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
702 if (&dentry->d_lru == list)
704 spin_lock(&dentry->d_lock);
705 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
706 spin_unlock(&dentry->d_lock);
711 * We found an inuse dentry which was not removed from
712 * the LRU because of laziness during lookup. Do not free
713 * it - just keep it off the LRU list.
715 if (dentry->d_count) {
716 dentry_lru_del(dentry);
717 spin_unlock(&dentry->d_lock);
723 try_prune_one_dentry(dentry);
731 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
732 * @sb: superblock to shrink dentry LRU.
733 * @count: number of entries to prune
734 * @flags: flags to control the dentry processing
736 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
738 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
740 /* called from prune_dcache() and shrink_dcache_parent() */
741 struct dentry *dentry;
742 LIST_HEAD(referenced);
747 spin_lock(&dcache_lru_lock);
748 while (!list_empty(&sb->s_dentry_lru)) {
749 dentry = list_entry(sb->s_dentry_lru.prev,
750 struct dentry, d_lru);
751 BUG_ON(dentry->d_sb != sb);
753 if (!spin_trylock(&dentry->d_lock)) {
754 spin_unlock(&dcache_lru_lock);
760 * If we are honouring the DCACHE_REFERENCED flag and the
761 * dentry has this flag set, don't free it. Clear the flag
762 * and put it back on the LRU.
764 if (flags & DCACHE_REFERENCED &&
765 dentry->d_flags & DCACHE_REFERENCED) {
766 dentry->d_flags &= ~DCACHE_REFERENCED;
767 list_move(&dentry->d_lru, &referenced);
768 spin_unlock(&dentry->d_lock);
770 list_move_tail(&dentry->d_lru, &tmp);
771 spin_unlock(&dentry->d_lock);
775 cond_resched_lock(&dcache_lru_lock);
777 if (!list_empty(&referenced))
778 list_splice(&referenced, &sb->s_dentry_lru);
779 spin_unlock(&dcache_lru_lock);
781 shrink_dentry_list(&tmp);
787 * prune_dcache - shrink the dcache
788 * @count: number of entries to try to free
790 * Shrink the dcache. This is done when we need more memory, or simply when we
791 * need to unmount something (at which point we need to unuse all dentries).
793 * This function may fail to free any resources if all the dentries are in use.
795 static void prune_dcache(int count)
797 struct super_block *sb, *p = NULL;
799 int unused = dentry_stat.nr_unused;
803 if (unused == 0 || count == 0)
808 prune_ratio = unused / count;
810 list_for_each_entry(sb, &super_blocks, s_list) {
811 if (list_empty(&sb->s_instances))
813 if (sb->s_nr_dentry_unused == 0)
816 /* Now, we reclaim unused dentrins with fairness.
817 * We reclaim them same percentage from each superblock.
818 * We calculate number of dentries to scan on this sb
819 * as follows, but the implementation is arranged to avoid
821 * number of dentries to scan on this sb =
822 * count * (number of dentries on this sb /
823 * number of dentries in the machine)
825 spin_unlock(&sb_lock);
826 if (prune_ratio != 1)
827 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
829 w_count = sb->s_nr_dentry_unused;
832 * We need to be sure this filesystem isn't being unmounted,
833 * otherwise we could race with generic_shutdown_super(), and
834 * end up holding a reference to an inode while the filesystem
835 * is unmounted. So we try to get s_umount, and make sure
838 if (down_read_trylock(&sb->s_umount)) {
839 if ((sb->s_root != NULL) &&
840 (!list_empty(&sb->s_dentry_lru))) {
841 __shrink_dcache_sb(sb, &w_count,
845 up_read(&sb->s_umount);
852 /* more work left to do? */
858 spin_unlock(&sb_lock);
862 * shrink_dcache_sb - shrink dcache for a superblock
865 * Shrink the dcache for the specified super block. This is used to free
866 * the dcache before unmounting a file system.
868 void shrink_dcache_sb(struct super_block *sb)
872 spin_lock(&dcache_lru_lock);
873 while (!list_empty(&sb->s_dentry_lru)) {
874 list_splice_init(&sb->s_dentry_lru, &tmp);
875 spin_unlock(&dcache_lru_lock);
876 shrink_dentry_list(&tmp);
877 spin_lock(&dcache_lru_lock);
879 spin_unlock(&dcache_lru_lock);
881 EXPORT_SYMBOL(shrink_dcache_sb);
884 * destroy a single subtree of dentries for unmount
885 * - see the comments on shrink_dcache_for_umount() for a description of the
888 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
890 struct dentry *parent;
891 unsigned detached = 0;
893 BUG_ON(!IS_ROOT(dentry));
895 /* detach this root from the system */
896 spin_lock(&dentry->d_lock);
897 dentry_lru_del(dentry);
899 spin_unlock(&dentry->d_lock);
902 /* descend to the first leaf in the current subtree */
903 while (!list_empty(&dentry->d_subdirs)) {
906 /* this is a branch with children - detach all of them
907 * from the system in one go */
908 spin_lock(&dentry->d_lock);
909 list_for_each_entry(loop, &dentry->d_subdirs,
911 spin_lock_nested(&loop->d_lock,
912 DENTRY_D_LOCK_NESTED);
913 dentry_lru_del(loop);
915 spin_unlock(&loop->d_lock);
917 spin_unlock(&dentry->d_lock);
919 /* move to the first child */
920 dentry = list_entry(dentry->d_subdirs.next,
921 struct dentry, d_u.d_child);
924 /* consume the dentries from this leaf up through its parents
925 * until we find one with children or run out altogether */
929 if (dentry->d_count != 0) {
931 "BUG: Dentry %p{i=%lx,n=%s}"
933 " [unmount of %s %s]\n",
936 dentry->d_inode->i_ino : 0UL,
939 dentry->d_sb->s_type->name,
944 if (IS_ROOT(dentry)) {
946 list_del(&dentry->d_u.d_child);
948 parent = dentry->d_parent;
949 spin_lock(&parent->d_lock);
951 list_del(&dentry->d_u.d_child);
952 spin_unlock(&parent->d_lock);
957 inode = dentry->d_inode;
959 dentry->d_inode = NULL;
960 list_del_init(&dentry->d_alias);
961 if (dentry->d_op && dentry->d_op->d_iput)
962 dentry->d_op->d_iput(dentry, inode);
969 /* finished when we fall off the top of the tree,
970 * otherwise we ascend to the parent and move to the
971 * next sibling if there is one */
975 } while (list_empty(&dentry->d_subdirs));
977 dentry = list_entry(dentry->d_subdirs.next,
978 struct dentry, d_u.d_child);
983 * destroy the dentries attached to a superblock on unmounting
984 * - we don't need to use dentry->d_lock because:
985 * - the superblock is detached from all mountings and open files, so the
986 * dentry trees will not be rearranged by the VFS
987 * - s_umount is write-locked, so the memory pressure shrinker will ignore
988 * any dentries belonging to this superblock that it comes across
989 * - the filesystem itself is no longer permitted to rearrange the dentries
992 void shrink_dcache_for_umount(struct super_block *sb)
994 struct dentry *dentry;
996 if (down_read_trylock(&sb->s_umount))
1001 spin_lock(&dentry->d_lock);
1003 spin_unlock(&dentry->d_lock);
1004 shrink_dcache_for_umount_subtree(dentry);
1006 while (!hlist_bl_empty(&sb->s_anon)) {
1007 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1008 shrink_dcache_for_umount_subtree(dentry);
1013 * Search for at least 1 mount point in the dentry's subdirs.
1014 * We descend to the next level whenever the d_subdirs
1015 * list is non-empty and continue searching.
1019 * have_submounts - check for mounts over a dentry
1020 * @parent: dentry to check.
1022 * Return true if the parent or its subdirectories contain
1025 int have_submounts(struct dentry *parent)
1027 struct dentry *this_parent;
1028 struct list_head *next;
1032 seq = read_seqbegin(&rename_lock);
1034 this_parent = parent;
1036 if (d_mountpoint(parent))
1038 spin_lock(&this_parent->d_lock);
1040 next = this_parent->d_subdirs.next;
1042 while (next != &this_parent->d_subdirs) {
1043 struct list_head *tmp = next;
1044 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1047 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1048 /* Have we found a mount point ? */
1049 if (d_mountpoint(dentry)) {
1050 spin_unlock(&dentry->d_lock);
1051 spin_unlock(&this_parent->d_lock);
1054 if (!list_empty(&dentry->d_subdirs)) {
1055 spin_unlock(&this_parent->d_lock);
1056 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1057 this_parent = dentry;
1058 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1061 spin_unlock(&dentry->d_lock);
1064 * All done at this level ... ascend and resume the search.
1066 if (this_parent != parent) {
1068 struct dentry *child;
1070 tmp = this_parent->d_parent;
1072 spin_unlock(&this_parent->d_lock);
1073 child = this_parent;
1075 spin_lock(&this_parent->d_lock);
1076 /* might go back up the wrong parent if we have had a rename
1078 if (this_parent != child->d_parent ||
1079 (!locked && read_seqretry(&rename_lock, seq))) {
1080 spin_unlock(&this_parent->d_lock);
1085 next = child->d_u.d_child.next;
1088 spin_unlock(&this_parent->d_lock);
1089 if (!locked && read_seqretry(&rename_lock, seq))
1092 write_sequnlock(&rename_lock);
1093 return 0; /* No mount points found in tree */
1095 if (!locked && read_seqretry(&rename_lock, seq))
1098 write_sequnlock(&rename_lock);
1103 write_seqlock(&rename_lock);
1106 EXPORT_SYMBOL(have_submounts);
1109 * Search the dentry child list for the specified parent,
1110 * and move any unused dentries to the end of the unused
1111 * list for prune_dcache(). We descend to the next level
1112 * whenever the d_subdirs list is non-empty and continue
1115 * It returns zero iff there are no unused children,
1116 * otherwise it returns the number of children moved to
1117 * the end of the unused list. This may not be the total
1118 * number of unused children, because select_parent can
1119 * drop the lock and return early due to latency
1122 static int select_parent(struct dentry * parent)
1124 struct dentry *this_parent;
1125 struct list_head *next;
1130 seq = read_seqbegin(&rename_lock);
1132 this_parent = parent;
1133 spin_lock(&this_parent->d_lock);
1135 next = this_parent->d_subdirs.next;
1137 while (next != &this_parent->d_subdirs) {
1138 struct list_head *tmp = next;
1139 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1142 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1145 * move only zero ref count dentries to the end
1146 * of the unused list for prune_dcache
1148 if (!dentry->d_count) {
1149 dentry_lru_move_tail(dentry);
1152 dentry_lru_del(dentry);
1156 * We can return to the caller if we have found some (this
1157 * ensures forward progress). We'll be coming back to find
1160 if (found && need_resched()) {
1161 spin_unlock(&dentry->d_lock);
1166 * Descend a level if the d_subdirs list is non-empty.
1168 if (!list_empty(&dentry->d_subdirs)) {
1169 spin_unlock(&this_parent->d_lock);
1170 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1171 this_parent = dentry;
1172 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1176 spin_unlock(&dentry->d_lock);
1179 * All done at this level ... ascend and resume the search.
1181 if (this_parent != parent) {
1183 struct dentry *child;
1185 tmp = this_parent->d_parent;
1187 spin_unlock(&this_parent->d_lock);
1188 child = this_parent;
1190 spin_lock(&this_parent->d_lock);
1191 /* might go back up the wrong parent if we have had a rename
1193 if (this_parent != child->d_parent ||
1194 (!locked && read_seqretry(&rename_lock, seq))) {
1195 spin_unlock(&this_parent->d_lock);
1200 next = child->d_u.d_child.next;
1204 spin_unlock(&this_parent->d_lock);
1205 if (!locked && read_seqretry(&rename_lock, seq))
1208 write_sequnlock(&rename_lock);
1215 write_seqlock(&rename_lock);
1220 * shrink_dcache_parent - prune dcache
1221 * @parent: parent of entries to prune
1223 * Prune the dcache to remove unused children of the parent dentry.
1226 void shrink_dcache_parent(struct dentry * parent)
1228 struct super_block *sb = parent->d_sb;
1231 while ((found = select_parent(parent)) != 0)
1232 __shrink_dcache_sb(sb, &found, 0);
1234 EXPORT_SYMBOL(shrink_dcache_parent);
1237 * Scan `nr' dentries and return the number which remain.
1239 * We need to avoid reentering the filesystem if the caller is performing a
1240 * GFP_NOFS allocation attempt. One example deadlock is:
1242 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1243 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1244 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1246 * In this case we return -1 to tell the caller that we baled.
1248 static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1251 if (!(gfp_mask & __GFP_FS))
1256 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
1259 static struct shrinker dcache_shrinker = {
1260 .shrink = shrink_dcache_memory,
1261 .seeks = DEFAULT_SEEKS,
1265 * d_alloc - allocate a dcache entry
1266 * @parent: parent of entry to allocate
1267 * @name: qstr of the name
1269 * Allocates a dentry. It returns %NULL if there is insufficient memory
1270 * available. On a success the dentry is returned. The name passed in is
1271 * copied and the copy passed in may be reused after this call.
1274 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1276 struct dentry *dentry;
1279 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1283 if (name->len > DNAME_INLINE_LEN-1) {
1284 dname = kmalloc(name->len + 1, GFP_KERNEL);
1286 kmem_cache_free(dentry_cache, dentry);
1290 dname = dentry->d_iname;
1292 dentry->d_name.name = dname;
1294 dentry->d_name.len = name->len;
1295 dentry->d_name.hash = name->hash;
1296 memcpy(dname, name->name, name->len);
1297 dname[name->len] = 0;
1299 dentry->d_count = 1;
1300 dentry->d_flags = DCACHE_UNHASHED;
1301 spin_lock_init(&dentry->d_lock);
1302 seqcount_init(&dentry->d_seq);
1303 dentry->d_inode = NULL;
1304 dentry->d_parent = NULL;
1305 dentry->d_sb = NULL;
1306 dentry->d_op = NULL;
1307 dentry->d_fsdata = NULL;
1308 INIT_HLIST_BL_NODE(&dentry->d_hash);
1309 INIT_LIST_HEAD(&dentry->d_lru);
1310 INIT_LIST_HEAD(&dentry->d_subdirs);
1311 INIT_LIST_HEAD(&dentry->d_alias);
1312 INIT_LIST_HEAD(&dentry->d_u.d_child);
1315 spin_lock(&parent->d_lock);
1317 * don't need child lock because it is not subject
1318 * to concurrency here
1320 __dget_dlock(parent);
1321 dentry->d_parent = parent;
1322 dentry->d_sb = parent->d_sb;
1323 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1324 spin_unlock(&parent->d_lock);
1327 this_cpu_inc(nr_dentry);
1331 EXPORT_SYMBOL(d_alloc);
1333 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1338 q.len = strlen(name);
1339 q.hash = full_name_hash(q.name, q.len);
1340 return d_alloc(parent, &q);
1342 EXPORT_SYMBOL(d_alloc_name);
1344 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1346 BUG_ON(dentry->d_op);
1347 BUG_ON(dentry->d_flags & (DCACHE_OP_HASH |
1349 DCACHE_OP_REVALIDATE |
1350 DCACHE_OP_DELETE ));
1355 dentry->d_flags |= DCACHE_OP_HASH;
1357 dentry->d_flags |= DCACHE_OP_COMPARE;
1358 if (op->d_revalidate)
1359 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1361 dentry->d_flags |= DCACHE_OP_DELETE;
1364 EXPORT_SYMBOL(d_set_d_op);
1366 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1368 spin_lock(&dentry->d_lock);
1370 list_add(&dentry->d_alias, &inode->i_dentry);
1371 dentry->d_inode = inode;
1372 dentry_rcuwalk_barrier(dentry);
1373 spin_unlock(&dentry->d_lock);
1374 fsnotify_d_instantiate(dentry, inode);
1378 * d_instantiate - fill in inode information for a dentry
1379 * @entry: dentry to complete
1380 * @inode: inode to attach to this dentry
1382 * Fill in inode information in the entry.
1384 * This turns negative dentries into productive full members
1387 * NOTE! This assumes that the inode count has been incremented
1388 * (or otherwise set) by the caller to indicate that it is now
1389 * in use by the dcache.
1392 void d_instantiate(struct dentry *entry, struct inode * inode)
1394 BUG_ON(!list_empty(&entry->d_alias));
1395 spin_lock(&dcache_inode_lock);
1396 __d_instantiate(entry, inode);
1397 spin_unlock(&dcache_inode_lock);
1398 security_d_instantiate(entry, inode);
1400 EXPORT_SYMBOL(d_instantiate);
1403 * d_instantiate_unique - instantiate a non-aliased dentry
1404 * @entry: dentry to instantiate
1405 * @inode: inode to attach to this dentry
1407 * Fill in inode information in the entry. On success, it returns NULL.
1408 * If an unhashed alias of "entry" already exists, then we return the
1409 * aliased dentry instead and drop one reference to inode.
1411 * Note that in order to avoid conflicts with rename() etc, the caller
1412 * had better be holding the parent directory semaphore.
1414 * This also assumes that the inode count has been incremented
1415 * (or otherwise set) by the caller to indicate that it is now
1416 * in use by the dcache.
1418 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1419 struct inode *inode)
1421 struct dentry *alias;
1422 int len = entry->d_name.len;
1423 const char *name = entry->d_name.name;
1424 unsigned int hash = entry->d_name.hash;
1427 __d_instantiate(entry, NULL);
1431 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1432 struct qstr *qstr = &alias->d_name;
1435 * Don't need alias->d_lock here, because aliases with
1436 * d_parent == entry->d_parent are not subject to name or
1437 * parent changes, because the parent inode i_mutex is held.
1439 if (qstr->hash != hash)
1441 if (alias->d_parent != entry->d_parent)
1443 if (qstr->len != len)
1445 if (memcmp(qstr->name, name, len))
1451 __d_instantiate(entry, inode);
1455 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1457 struct dentry *result;
1459 BUG_ON(!list_empty(&entry->d_alias));
1461 spin_lock(&dcache_inode_lock);
1462 result = __d_instantiate_unique(entry, inode);
1463 spin_unlock(&dcache_inode_lock);
1466 security_d_instantiate(entry, inode);
1470 BUG_ON(!d_unhashed(result));
1475 EXPORT_SYMBOL(d_instantiate_unique);
1478 * d_alloc_root - allocate root dentry
1479 * @root_inode: inode to allocate the root for
1481 * Allocate a root ("/") dentry for the inode given. The inode is
1482 * instantiated and returned. %NULL is returned if there is insufficient
1483 * memory or the inode passed is %NULL.
1486 struct dentry * d_alloc_root(struct inode * root_inode)
1488 struct dentry *res = NULL;
1491 static const struct qstr name = { .name = "/", .len = 1 };
1493 res = d_alloc(NULL, &name);
1495 res->d_sb = root_inode->i_sb;
1496 res->d_parent = res;
1497 d_instantiate(res, root_inode);
1502 EXPORT_SYMBOL(d_alloc_root);
1505 * d_obtain_alias - find or allocate a dentry for a given inode
1506 * @inode: inode to allocate the dentry for
1508 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1509 * similar open by handle operations. The returned dentry may be anonymous,
1510 * or may have a full name (if the inode was already in the cache).
1512 * When called on a directory inode, we must ensure that the inode only ever
1513 * has one dentry. If a dentry is found, that is returned instead of
1514 * allocating a new one.
1516 * On successful return, the reference to the inode has been transferred
1517 * to the dentry. In case of an error the reference on the inode is released.
1518 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1519 * be passed in and will be the error will be propagate to the return value,
1520 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1522 struct dentry *d_obtain_alias(struct inode *inode)
1524 static const struct qstr anonstring = { .name = "" };
1529 return ERR_PTR(-ESTALE);
1531 return ERR_CAST(inode);
1533 res = d_find_alias(inode);
1537 tmp = d_alloc(NULL, &anonstring);
1539 res = ERR_PTR(-ENOMEM);
1542 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1545 spin_lock(&dcache_inode_lock);
1546 res = __d_find_alias(inode, 0);
1548 spin_unlock(&dcache_inode_lock);
1553 /* attach a disconnected dentry */
1554 spin_lock(&tmp->d_lock);
1555 tmp->d_sb = inode->i_sb;
1556 tmp->d_inode = inode;
1557 tmp->d_flags |= DCACHE_DISCONNECTED;
1558 list_add(&tmp->d_alias, &inode->i_dentry);
1559 bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1560 tmp->d_flags &= ~DCACHE_UNHASHED;
1561 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1562 __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1563 spin_unlock(&tmp->d_lock);
1564 spin_unlock(&dcache_inode_lock);
1572 EXPORT_SYMBOL(d_obtain_alias);
1575 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1576 * @inode: the inode which may have a disconnected dentry
1577 * @dentry: a negative dentry which we want to point to the inode.
1579 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1580 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1581 * and return it, else simply d_add the inode to the dentry and return NULL.
1583 * This is needed in the lookup routine of any filesystem that is exportable
1584 * (via knfsd) so that we can build dcache paths to directories effectively.
1586 * If a dentry was found and moved, then it is returned. Otherwise NULL
1587 * is returned. This matches the expected return value of ->lookup.
1590 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1592 struct dentry *new = NULL;
1594 if (inode && S_ISDIR(inode->i_mode)) {
1595 spin_lock(&dcache_inode_lock);
1596 new = __d_find_alias(inode, 1);
1598 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1599 spin_unlock(&dcache_inode_lock);
1600 security_d_instantiate(new, inode);
1601 d_move(new, dentry);
1604 /* already got dcache_inode_lock, so d_add() by hand */
1605 __d_instantiate(dentry, inode);
1606 spin_unlock(&dcache_inode_lock);
1607 security_d_instantiate(dentry, inode);
1611 d_add(dentry, inode);
1614 EXPORT_SYMBOL(d_splice_alias);
1617 * d_add_ci - lookup or allocate new dentry with case-exact name
1618 * @inode: the inode case-insensitive lookup has found
1619 * @dentry: the negative dentry that was passed to the parent's lookup func
1620 * @name: the case-exact name to be associated with the returned dentry
1622 * This is to avoid filling the dcache with case-insensitive names to the
1623 * same inode, only the actual correct case is stored in the dcache for
1624 * case-insensitive filesystems.
1626 * For a case-insensitive lookup match and if the the case-exact dentry
1627 * already exists in in the dcache, use it and return it.
1629 * If no entry exists with the exact case name, allocate new dentry with
1630 * the exact case, and return the spliced entry.
1632 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1636 struct dentry *found;
1640 * First check if a dentry matching the name already exists,
1641 * if not go ahead and create it now.
1643 found = d_hash_and_lookup(dentry->d_parent, name);
1645 new = d_alloc(dentry->d_parent, name);
1651 found = d_splice_alias(inode, new);
1660 * If a matching dentry exists, and it's not negative use it.
1662 * Decrement the reference count to balance the iget() done
1665 if (found->d_inode) {
1666 if (unlikely(found->d_inode != inode)) {
1667 /* This can't happen because bad inodes are unhashed. */
1668 BUG_ON(!is_bad_inode(inode));
1669 BUG_ON(!is_bad_inode(found->d_inode));
1676 * Negative dentry: instantiate it unless the inode is a directory and
1677 * already has a dentry.
1679 spin_lock(&dcache_inode_lock);
1680 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1681 __d_instantiate(found, inode);
1682 spin_unlock(&dcache_inode_lock);
1683 security_d_instantiate(found, inode);
1688 * In case a directory already has a (disconnected) entry grab a
1689 * reference to it, move it in place and use it.
1691 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1693 spin_unlock(&dcache_inode_lock);
1694 security_d_instantiate(found, inode);
1702 return ERR_PTR(error);
1704 EXPORT_SYMBOL(d_add_ci);
1707 * __d_lookup_rcu - search for a dentry (racy, store-free)
1708 * @parent: parent dentry
1709 * @name: qstr of name we wish to find
1710 * @seq: returns d_seq value at the point where the dentry was found
1711 * @inode: returns dentry->d_inode when the inode was found valid.
1712 * Returns: dentry, or NULL
1714 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1715 * resolution (store-free path walking) design described in
1716 * Documentation/filesystems/path-lookup.txt.
1718 * This is not to be used outside core vfs.
1720 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1721 * held, and rcu_read_lock held. The returned dentry must not be stored into
1722 * without taking d_lock and checking d_seq sequence count against @seq
1725 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1728 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1729 * the returned dentry, so long as its parent's seqlock is checked after the
1730 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1731 * is formed, giving integrity down the path walk.
1733 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1734 unsigned *seq, struct inode **inode)
1736 unsigned int len = name->len;
1737 unsigned int hash = name->hash;
1738 const unsigned char *str = name->name;
1739 struct dcache_hash_bucket *b = d_hash(parent, hash);
1740 struct hlist_bl_node *node;
1741 struct dentry *dentry;
1744 * Note: There is significant duplication with __d_lookup_rcu which is
1745 * required to prevent single threaded performance regressions
1746 * especially on architectures where smp_rmb (in seqcounts) are costly.
1747 * Keep the two functions in sync.
1751 * The hash list is protected using RCU.
1753 * Carefully use d_seq when comparing a candidate dentry, to avoid
1754 * races with d_move().
1756 * It is possible that concurrent renames can mess up our list
1757 * walk here and result in missing our dentry, resulting in the
1758 * false-negative result. d_lookup() protects against concurrent
1759 * renames using rename_lock seqlock.
1761 * See Documentation/vfs/dcache-locking.txt for more details.
1763 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1768 if (dentry->d_name.hash != hash)
1772 *seq = read_seqcount_begin(&dentry->d_seq);
1773 if (dentry->d_parent != parent)
1775 if (d_unhashed(dentry))
1777 tlen = dentry->d_name.len;
1778 tname = dentry->d_name.name;
1779 i = dentry->d_inode;
1781 * This seqcount check is required to ensure name and
1782 * len are loaded atomically, so as not to walk off the
1783 * edge of memory when walking. If we could load this
1784 * atomically some other way, we could drop this check.
1786 if (read_seqcount_retry(&dentry->d_seq, *seq))
1788 if (parent->d_flags & DCACHE_OP_COMPARE) {
1789 if (parent->d_op->d_compare(parent, *inode,
1796 if (memcmp(tname, str, tlen))
1800 * No extra seqcount check is required after the name
1801 * compare. The caller must perform a seqcount check in
1802 * order to do anything useful with the returned dentry
1812 * d_lookup - search for a dentry
1813 * @parent: parent dentry
1814 * @name: qstr of name we wish to find
1815 * Returns: dentry, or NULL
1817 * d_lookup searches the children of the parent dentry for the name in
1818 * question. If the dentry is found its reference count is incremented and the
1819 * dentry is returned. The caller must use dput to free the entry when it has
1820 * finished using it. %NULL is returned if the dentry does not exist.
1822 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1824 struct dentry *dentry;
1828 seq = read_seqbegin(&rename_lock);
1829 dentry = __d_lookup(parent, name);
1832 } while (read_seqretry(&rename_lock, seq));
1835 EXPORT_SYMBOL(d_lookup);
1838 * __d_lookup - search for a dentry (racy)
1839 * @parent: parent dentry
1840 * @name: qstr of name we wish to find
1841 * Returns: dentry, or NULL
1843 * __d_lookup is like d_lookup, however it may (rarely) return a
1844 * false-negative result due to unrelated rename activity.
1846 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1847 * however it must be used carefully, eg. with a following d_lookup in
1848 * the case of failure.
1850 * __d_lookup callers must be commented.
1852 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1854 unsigned int len = name->len;
1855 unsigned int hash = name->hash;
1856 const unsigned char *str = name->name;
1857 struct dcache_hash_bucket *b = d_hash(parent, hash);
1858 struct hlist_bl_node *node;
1859 struct dentry *found = NULL;
1860 struct dentry *dentry;
1863 * Note: There is significant duplication with __d_lookup_rcu which is
1864 * required to prevent single threaded performance regressions
1865 * especially on architectures where smp_rmb (in seqcounts) are costly.
1866 * Keep the two functions in sync.
1870 * The hash list is protected using RCU.
1872 * Take d_lock when comparing a candidate dentry, to avoid races
1875 * It is possible that concurrent renames can mess up our list
1876 * walk here and result in missing our dentry, resulting in the
1877 * false-negative result. d_lookup() protects against concurrent
1878 * renames using rename_lock seqlock.
1880 * See Documentation/vfs/dcache-locking.txt for more details.
1884 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1888 if (dentry->d_name.hash != hash)
1891 spin_lock(&dentry->d_lock);
1892 if (dentry->d_parent != parent)
1894 if (d_unhashed(dentry))
1898 * It is safe to compare names since d_move() cannot
1899 * change the qstr (protected by d_lock).
1901 tlen = dentry->d_name.len;
1902 tname = dentry->d_name.name;
1903 if (parent->d_flags & DCACHE_OP_COMPARE) {
1904 if (parent->d_op->d_compare(parent, parent->d_inode,
1905 dentry, dentry->d_inode,
1911 if (memcmp(tname, str, tlen))
1917 spin_unlock(&dentry->d_lock);
1920 spin_unlock(&dentry->d_lock);
1928 * d_hash_and_lookup - hash the qstr then search for a dentry
1929 * @dir: Directory to search in
1930 * @name: qstr of name we wish to find
1932 * On hash failure or on lookup failure NULL is returned.
1934 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1936 struct dentry *dentry = NULL;
1939 * Check for a fs-specific hash function. Note that we must
1940 * calculate the standard hash first, as the d_op->d_hash()
1941 * routine may choose to leave the hash value unchanged.
1943 name->hash = full_name_hash(name->name, name->len);
1944 if (dir->d_flags & DCACHE_OP_HASH) {
1945 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1948 dentry = d_lookup(dir, name);
1954 * d_validate - verify dentry provided from insecure source (deprecated)
1955 * @dentry: The dentry alleged to be valid child of @dparent
1956 * @dparent: The parent dentry (known to be valid)
1958 * An insecure source has sent us a dentry, here we verify it and dget() it.
1959 * This is used by ncpfs in its readdir implementation.
1960 * Zero is returned in the dentry is invalid.
1962 * This function is slow for big directories, and deprecated, do not use it.
1964 int d_validate(struct dentry *dentry, struct dentry *dparent)
1966 struct dentry *child;
1968 spin_lock(&dparent->d_lock);
1969 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1970 if (dentry == child) {
1971 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1972 __dget_dlock(dentry);
1973 spin_unlock(&dentry->d_lock);
1974 spin_unlock(&dparent->d_lock);
1978 spin_unlock(&dparent->d_lock);
1982 EXPORT_SYMBOL(d_validate);
1985 * When a file is deleted, we have two options:
1986 * - turn this dentry into a negative dentry
1987 * - unhash this dentry and free it.
1989 * Usually, we want to just turn this into
1990 * a negative dentry, but if anybody else is
1991 * currently using the dentry or the inode
1992 * we can't do that and we fall back on removing
1993 * it from the hash queues and waiting for
1994 * it to be deleted later when it has no users
1998 * d_delete - delete a dentry
1999 * @dentry: The dentry to delete
2001 * Turn the dentry into a negative dentry if possible, otherwise
2002 * remove it from the hash queues so it can be deleted later
2005 void d_delete(struct dentry * dentry)
2009 * Are we the only user?
2012 spin_lock(&dentry->d_lock);
2013 isdir = S_ISDIR(dentry->d_inode->i_mode);
2014 if (dentry->d_count == 1) {
2015 if (!spin_trylock(&dcache_inode_lock)) {
2016 spin_unlock(&dentry->d_lock);
2020 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2021 dentry_unlink_inode(dentry);
2022 fsnotify_nameremove(dentry, isdir);
2026 if (!d_unhashed(dentry))
2029 spin_unlock(&dentry->d_lock);
2031 fsnotify_nameremove(dentry, isdir);
2033 EXPORT_SYMBOL(d_delete);
2035 static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b)
2037 BUG_ON(!d_unhashed(entry));
2038 spin_lock_bucket(b);
2039 entry->d_flags &= ~DCACHE_UNHASHED;
2040 hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
2041 spin_unlock_bucket(b);
2044 static void _d_rehash(struct dentry * entry)
2046 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2050 * d_rehash - add an entry back to the hash
2051 * @entry: dentry to add to the hash
2053 * Adds a dentry to the hash according to its name.
2056 void d_rehash(struct dentry * entry)
2058 spin_lock(&entry->d_lock);
2060 spin_unlock(&entry->d_lock);
2062 EXPORT_SYMBOL(d_rehash);
2065 * dentry_update_name_case - update case insensitive dentry with a new name
2066 * @dentry: dentry to be updated
2069 * Update a case insensitive dentry with new case of name.
2071 * dentry must have been returned by d_lookup with name @name. Old and new
2072 * name lengths must match (ie. no d_compare which allows mismatched name
2075 * Parent inode i_mutex must be held over d_lookup and into this call (to
2076 * keep renames and concurrent inserts, and readdir(2) away).
2078 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2080 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2081 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2083 spin_lock(&dentry->d_lock);
2084 write_seqcount_begin(&dentry->d_seq);
2085 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2086 write_seqcount_end(&dentry->d_seq);
2087 spin_unlock(&dentry->d_lock);
2089 EXPORT_SYMBOL(dentry_update_name_case);
2091 static void switch_names(struct dentry *dentry, struct dentry *target)
2093 if (dname_external(target)) {
2094 if (dname_external(dentry)) {
2096 * Both external: swap the pointers
2098 swap(target->d_name.name, dentry->d_name.name);
2101 * dentry:internal, target:external. Steal target's
2102 * storage and make target internal.
2104 memcpy(target->d_iname, dentry->d_name.name,
2105 dentry->d_name.len + 1);
2106 dentry->d_name.name = target->d_name.name;
2107 target->d_name.name = target->d_iname;
2110 if (dname_external(dentry)) {
2112 * dentry:external, target:internal. Give dentry's
2113 * storage to target and make dentry internal
2115 memcpy(dentry->d_iname, target->d_name.name,
2116 target->d_name.len + 1);
2117 target->d_name.name = dentry->d_name.name;
2118 dentry->d_name.name = dentry->d_iname;
2121 * Both are internal. Just copy target to dentry
2123 memcpy(dentry->d_iname, target->d_name.name,
2124 target->d_name.len + 1);
2125 dentry->d_name.len = target->d_name.len;
2129 swap(dentry->d_name.len, target->d_name.len);
2132 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2135 * XXXX: do we really need to take target->d_lock?
2137 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2138 spin_lock(&target->d_parent->d_lock);
2140 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2141 spin_lock(&dentry->d_parent->d_lock);
2142 spin_lock_nested(&target->d_parent->d_lock,
2143 DENTRY_D_LOCK_NESTED);
2145 spin_lock(&target->d_parent->d_lock);
2146 spin_lock_nested(&dentry->d_parent->d_lock,
2147 DENTRY_D_LOCK_NESTED);
2150 if (target < dentry) {
2151 spin_lock_nested(&target->d_lock, 2);
2152 spin_lock_nested(&dentry->d_lock, 3);
2154 spin_lock_nested(&dentry->d_lock, 2);
2155 spin_lock_nested(&target->d_lock, 3);
2159 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2160 struct dentry *target)
2162 if (target->d_parent != dentry->d_parent)
2163 spin_unlock(&dentry->d_parent->d_lock);
2164 if (target->d_parent != target)
2165 spin_unlock(&target->d_parent->d_lock);
2169 * When switching names, the actual string doesn't strictly have to
2170 * be preserved in the target - because we're dropping the target
2171 * anyway. As such, we can just do a simple memcpy() to copy over
2172 * the new name before we switch.
2174 * Note that we have to be a lot more careful about getting the hash
2175 * switched - we have to switch the hash value properly even if it
2176 * then no longer matches the actual (corrupted) string of the target.
2177 * The hash value has to match the hash queue that the dentry is on..
2180 * d_move - move a dentry
2181 * @dentry: entry to move
2182 * @target: new dentry
2184 * Update the dcache to reflect the move of a file name. Negative
2185 * dcache entries should not be moved in this way.
2187 void d_move(struct dentry * dentry, struct dentry * target)
2189 if (!dentry->d_inode)
2190 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2192 BUG_ON(d_ancestor(dentry, target));
2193 BUG_ON(d_ancestor(target, dentry));
2195 write_seqlock(&rename_lock);
2197 dentry_lock_for_move(dentry, target);
2199 write_seqcount_begin(&dentry->d_seq);
2200 write_seqcount_begin(&target->d_seq);
2202 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2205 * Move the dentry to the target hash queue. Don't bother checking
2206 * for the same hash queue because of how unlikely it is.
2209 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2211 /* Unhash the target: dput() will then get rid of it */
2214 list_del(&dentry->d_u.d_child);
2215 list_del(&target->d_u.d_child);
2217 /* Switch the names.. */
2218 switch_names(dentry, target);
2219 swap(dentry->d_name.hash, target->d_name.hash);
2221 /* ... and switch the parents */
2222 if (IS_ROOT(dentry)) {
2223 dentry->d_parent = target->d_parent;
2224 target->d_parent = target;
2225 INIT_LIST_HEAD(&target->d_u.d_child);
2227 swap(dentry->d_parent, target->d_parent);
2229 /* And add them back to the (new) parent lists */
2230 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2233 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2235 write_seqcount_end(&target->d_seq);
2236 write_seqcount_end(&dentry->d_seq);
2238 dentry_unlock_parents_for_move(dentry, target);
2239 spin_unlock(&target->d_lock);
2240 fsnotify_d_move(dentry);
2241 spin_unlock(&dentry->d_lock);
2242 write_sequnlock(&rename_lock);
2244 EXPORT_SYMBOL(d_move);
2247 * d_ancestor - search for an ancestor
2248 * @p1: ancestor dentry
2251 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2252 * an ancestor of p2, else NULL.
2254 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2258 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2259 if (p->d_parent == p1)
2266 * This helper attempts to cope with remotely renamed directories
2268 * It assumes that the caller is already holding
2269 * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock
2271 * Note: If ever the locking in lock_rename() changes, then please
2272 * remember to update this too...
2274 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
2275 __releases(dcache_inode_lock)
2277 struct mutex *m1 = NULL, *m2 = NULL;
2280 /* If alias and dentry share a parent, then no extra locks required */
2281 if (alias->d_parent == dentry->d_parent)
2284 /* Check for loops */
2285 ret = ERR_PTR(-ELOOP);
2286 if (d_ancestor(alias, dentry))
2289 /* See lock_rename() */
2290 ret = ERR_PTR(-EBUSY);
2291 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2293 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2294 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2296 m2 = &alias->d_parent->d_inode->i_mutex;
2298 d_move(alias, dentry);
2301 spin_unlock(&dcache_inode_lock);
2310 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2311 * named dentry in place of the dentry to be replaced.
2312 * returns with anon->d_lock held!
2314 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2316 struct dentry *dparent, *aparent;
2318 dentry_lock_for_move(anon, dentry);
2320 write_seqcount_begin(&dentry->d_seq);
2321 write_seqcount_begin(&anon->d_seq);
2323 dparent = dentry->d_parent;
2324 aparent = anon->d_parent;
2326 switch_names(dentry, anon);
2327 swap(dentry->d_name.hash, anon->d_name.hash);
2329 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2330 list_del(&dentry->d_u.d_child);
2331 if (!IS_ROOT(dentry))
2332 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2334 INIT_LIST_HEAD(&dentry->d_u.d_child);
2336 anon->d_parent = (dparent == dentry) ? anon : dparent;
2337 list_del(&anon->d_u.d_child);
2339 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2341 INIT_LIST_HEAD(&anon->d_u.d_child);
2343 write_seqcount_end(&dentry->d_seq);
2344 write_seqcount_end(&anon->d_seq);
2346 dentry_unlock_parents_for_move(anon, dentry);
2347 spin_unlock(&dentry->d_lock);
2349 /* anon->d_lock still locked, returns locked */
2350 anon->d_flags &= ~DCACHE_DISCONNECTED;
2354 * d_materialise_unique - introduce an inode into the tree
2355 * @dentry: candidate dentry
2356 * @inode: inode to bind to the dentry, to which aliases may be attached
2358 * Introduces an dentry into the tree, substituting an extant disconnected
2359 * root directory alias in its place if there is one
2361 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2363 struct dentry *actual;
2365 BUG_ON(!d_unhashed(dentry));
2369 __d_instantiate(dentry, NULL);
2374 spin_lock(&dcache_inode_lock);
2376 if (S_ISDIR(inode->i_mode)) {
2377 struct dentry *alias;
2379 /* Does an aliased dentry already exist? */
2380 alias = __d_find_alias(inode, 0);
2383 /* Is this an anonymous mountpoint that we could splice
2385 if (IS_ROOT(alias)) {
2386 __d_materialise_dentry(dentry, alias);
2390 /* Nope, but we must(!) avoid directory aliasing */
2391 actual = __d_unalias(dentry, alias);
2398 /* Add a unique reference */
2399 actual = __d_instantiate_unique(dentry, inode);
2403 BUG_ON(!d_unhashed(actual));
2405 spin_lock(&actual->d_lock);
2408 spin_unlock(&actual->d_lock);
2409 spin_unlock(&dcache_inode_lock);
2411 if (actual == dentry) {
2412 security_d_instantiate(dentry, inode);
2419 EXPORT_SYMBOL_GPL(d_materialise_unique);
2421 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2425 return -ENAMETOOLONG;
2427 memcpy(*buffer, str, namelen);
2431 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2433 return prepend(buffer, buflen, name->name, name->len);
2437 * Prepend path string to a buffer
2439 * @path: the dentry/vfsmount to report
2440 * @root: root vfsmnt/dentry (may be modified by this function)
2441 * @buffer: pointer to the end of the buffer
2442 * @buflen: pointer to buffer length
2444 * Caller holds the rename_lock.
2446 * If path is not reachable from the supplied root, then the value of
2447 * root is changed (without modifying refcounts).
2449 static int prepend_path(const struct path *path, struct path *root,
2450 char **buffer, int *buflen)
2452 struct dentry *dentry = path->dentry;
2453 struct vfsmount *vfsmnt = path->mnt;
2457 br_read_lock(vfsmount_lock);
2458 while (dentry != root->dentry || vfsmnt != root->mnt) {
2459 struct dentry * parent;
2461 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2463 if (vfsmnt->mnt_parent == vfsmnt) {
2466 dentry = vfsmnt->mnt_mountpoint;
2467 vfsmnt = vfsmnt->mnt_parent;
2470 parent = dentry->d_parent;
2472 spin_lock(&dentry->d_lock);
2473 error = prepend_name(buffer, buflen, &dentry->d_name);
2474 spin_unlock(&dentry->d_lock);
2476 error = prepend(buffer, buflen, "/", 1);
2485 if (!error && !slash)
2486 error = prepend(buffer, buflen, "/", 1);
2488 br_read_unlock(vfsmount_lock);
2493 * Filesystems needing to implement special "root names"
2494 * should do so with ->d_dname()
2496 if (IS_ROOT(dentry) &&
2497 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2498 WARN(1, "Root dentry has weird name <%.*s>\n",
2499 (int) dentry->d_name.len, dentry->d_name.name);
2502 root->dentry = dentry;
2507 * __d_path - return the path of a dentry
2508 * @path: the dentry/vfsmount to report
2509 * @root: root vfsmnt/dentry (may be modified by this function)
2510 * @buf: buffer to return value in
2511 * @buflen: buffer length
2513 * Convert a dentry into an ASCII path name.
2515 * Returns a pointer into the buffer or an error code if the
2516 * path was too long.
2518 * "buflen" should be positive.
2520 * If path is not reachable from the supplied root, then the value of
2521 * root is changed (without modifying refcounts).
2523 char *__d_path(const struct path *path, struct path *root,
2524 char *buf, int buflen)
2526 char *res = buf + buflen;
2529 prepend(&res, &buflen, "\0", 1);
2530 write_seqlock(&rename_lock);
2531 error = prepend_path(path, root, &res, &buflen);
2532 write_sequnlock(&rename_lock);
2535 return ERR_PTR(error);
2540 * same as __d_path but appends "(deleted)" for unlinked files.
2542 static int path_with_deleted(const struct path *path, struct path *root,
2543 char **buf, int *buflen)
2545 prepend(buf, buflen, "\0", 1);
2546 if (d_unlinked(path->dentry)) {
2547 int error = prepend(buf, buflen, " (deleted)", 10);
2552 return prepend_path(path, root, buf, buflen);
2555 static int prepend_unreachable(char **buffer, int *buflen)
2557 return prepend(buffer, buflen, "(unreachable)", 13);
2561 * d_path - return the path of a dentry
2562 * @path: path to report
2563 * @buf: buffer to return value in
2564 * @buflen: buffer length
2566 * Convert a dentry into an ASCII path name. If the entry has been deleted
2567 * the string " (deleted)" is appended. Note that this is ambiguous.
2569 * Returns a pointer into the buffer or an error code if the path was
2570 * too long. Note: Callers should use the returned pointer, not the passed
2571 * in buffer, to use the name! The implementation often starts at an offset
2572 * into the buffer, and may leave 0 bytes at the start.
2574 * "buflen" should be positive.
2576 char *d_path(const struct path *path, char *buf, int buflen)
2578 char *res = buf + buflen;
2584 * We have various synthetic filesystems that never get mounted. On
2585 * these filesystems dentries are never used for lookup purposes, and
2586 * thus don't need to be hashed. They also don't need a name until a
2587 * user wants to identify the object in /proc/pid/fd/. The little hack
2588 * below allows us to generate a name for these objects on demand:
2590 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2591 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2593 get_fs_root(current->fs, &root);
2594 write_seqlock(&rename_lock);
2596 error = path_with_deleted(path, &tmp, &res, &buflen);
2598 res = ERR_PTR(error);
2599 write_sequnlock(&rename_lock);
2603 EXPORT_SYMBOL(d_path);
2606 * d_path_with_unreachable - return the path of a dentry
2607 * @path: path to report
2608 * @buf: buffer to return value in
2609 * @buflen: buffer length
2611 * The difference from d_path() is that this prepends "(unreachable)"
2612 * to paths which are unreachable from the current process' root.
2614 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2616 char *res = buf + buflen;
2621 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2622 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2624 get_fs_root(current->fs, &root);
2625 write_seqlock(&rename_lock);
2627 error = path_with_deleted(path, &tmp, &res, &buflen);
2628 if (!error && !path_equal(&tmp, &root))
2629 error = prepend_unreachable(&res, &buflen);
2630 write_sequnlock(&rename_lock);
2633 res = ERR_PTR(error);
2639 * Helper function for dentry_operations.d_dname() members
2641 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2642 const char *fmt, ...)
2648 va_start(args, fmt);
2649 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2652 if (sz > sizeof(temp) || sz > buflen)
2653 return ERR_PTR(-ENAMETOOLONG);
2655 buffer += buflen - sz;
2656 return memcpy(buffer, temp, sz);
2660 * Write full pathname from the root of the filesystem into the buffer.
2662 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2664 char *end = buf + buflen;
2667 prepend(&end, &buflen, "\0", 1);
2674 while (!IS_ROOT(dentry)) {
2675 struct dentry *parent = dentry->d_parent;
2679 spin_lock(&dentry->d_lock);
2680 error = prepend_name(&end, &buflen, &dentry->d_name);
2681 spin_unlock(&dentry->d_lock);
2682 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2690 return ERR_PTR(-ENAMETOOLONG);
2693 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2697 write_seqlock(&rename_lock);
2698 retval = __dentry_path(dentry, buf, buflen);
2699 write_sequnlock(&rename_lock);
2703 EXPORT_SYMBOL(dentry_path_raw);
2705 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2710 write_seqlock(&rename_lock);
2711 if (d_unlinked(dentry)) {
2713 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2717 retval = __dentry_path(dentry, buf, buflen);
2718 write_sequnlock(&rename_lock);
2719 if (!IS_ERR(retval) && p)
2720 *p = '/'; /* restore '/' overriden with '\0' */
2723 return ERR_PTR(-ENAMETOOLONG);
2727 * NOTE! The user-level library version returns a
2728 * character pointer. The kernel system call just
2729 * returns the length of the buffer filled (which
2730 * includes the ending '\0' character), or a negative
2731 * error value. So libc would do something like
2733 * char *getcwd(char * buf, size_t size)
2737 * retval = sys_getcwd(buf, size);
2744 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2747 struct path pwd, root;
2748 char *page = (char *) __get_free_page(GFP_USER);
2753 get_fs_root_and_pwd(current->fs, &root, &pwd);
2756 write_seqlock(&rename_lock);
2757 if (!d_unlinked(pwd.dentry)) {
2759 struct path tmp = root;
2760 char *cwd = page + PAGE_SIZE;
2761 int buflen = PAGE_SIZE;
2763 prepend(&cwd, &buflen, "\0", 1);
2764 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2765 write_sequnlock(&rename_lock);
2770 /* Unreachable from current root */
2771 if (!path_equal(&tmp, &root)) {
2772 error = prepend_unreachable(&cwd, &buflen);
2778 len = PAGE_SIZE + page - cwd;
2781 if (copy_to_user(buf, cwd, len))
2785 write_sequnlock(&rename_lock);
2791 free_page((unsigned long) page);
2796 * Test whether new_dentry is a subdirectory of old_dentry.
2798 * Trivially implemented using the dcache structure
2802 * is_subdir - is new dentry a subdirectory of old_dentry
2803 * @new_dentry: new dentry
2804 * @old_dentry: old dentry
2806 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2807 * Returns 0 otherwise.
2808 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2811 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2816 if (new_dentry == old_dentry)
2820 /* for restarting inner loop in case of seq retry */
2821 seq = read_seqbegin(&rename_lock);
2823 * Need rcu_readlock to protect against the d_parent trashing
2827 if (d_ancestor(old_dentry, new_dentry))
2832 } while (read_seqretry(&rename_lock, seq));
2837 int path_is_under(struct path *path1, struct path *path2)
2839 struct vfsmount *mnt = path1->mnt;
2840 struct dentry *dentry = path1->dentry;
2843 br_read_lock(vfsmount_lock);
2844 if (mnt != path2->mnt) {
2846 if (mnt->mnt_parent == mnt) {
2847 br_read_unlock(vfsmount_lock);
2850 if (mnt->mnt_parent == path2->mnt)
2852 mnt = mnt->mnt_parent;
2854 dentry = mnt->mnt_mountpoint;
2856 res = is_subdir(dentry, path2->dentry);
2857 br_read_unlock(vfsmount_lock);
2860 EXPORT_SYMBOL(path_is_under);
2862 void d_genocide(struct dentry *root)
2864 struct dentry *this_parent;
2865 struct list_head *next;
2869 seq = read_seqbegin(&rename_lock);
2872 spin_lock(&this_parent->d_lock);
2874 next = this_parent->d_subdirs.next;
2876 while (next != &this_parent->d_subdirs) {
2877 struct list_head *tmp = next;
2878 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2881 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2882 if (d_unhashed(dentry) || !dentry->d_inode) {
2883 spin_unlock(&dentry->d_lock);
2886 if (!list_empty(&dentry->d_subdirs)) {
2887 spin_unlock(&this_parent->d_lock);
2888 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2889 this_parent = dentry;
2890 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2893 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2894 dentry->d_flags |= DCACHE_GENOCIDE;
2897 spin_unlock(&dentry->d_lock);
2899 if (this_parent != root) {
2901 struct dentry *child;
2903 tmp = this_parent->d_parent;
2904 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2905 this_parent->d_flags |= DCACHE_GENOCIDE;
2906 this_parent->d_count--;
2909 spin_unlock(&this_parent->d_lock);
2910 child = this_parent;
2912 spin_lock(&this_parent->d_lock);
2913 /* might go back up the wrong parent if we have had a rename
2915 if (this_parent != child->d_parent ||
2916 (!locked && read_seqretry(&rename_lock, seq))) {
2917 spin_unlock(&this_parent->d_lock);
2922 next = child->d_u.d_child.next;
2925 spin_unlock(&this_parent->d_lock);
2926 if (!locked && read_seqretry(&rename_lock, seq))
2929 write_sequnlock(&rename_lock);
2934 write_seqlock(&rename_lock);
2939 * find_inode_number - check for dentry with name
2940 * @dir: directory to check
2941 * @name: Name to find.
2943 * Check whether a dentry already exists for the given name,
2944 * and return the inode number if it has an inode. Otherwise
2947 * This routine is used to post-process directory listings for
2948 * filesystems using synthetic inode numbers, and is necessary
2949 * to keep getcwd() working.
2952 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2954 struct dentry * dentry;
2957 dentry = d_hash_and_lookup(dir, name);
2959 if (dentry->d_inode)
2960 ino = dentry->d_inode->i_ino;
2965 EXPORT_SYMBOL(find_inode_number);
2967 static __initdata unsigned long dhash_entries;
2968 static int __init set_dhash_entries(char *str)
2972 dhash_entries = simple_strtoul(str, &str, 0);
2975 __setup("dhash_entries=", set_dhash_entries);
2977 static void __init dcache_init_early(void)
2981 /* If hashes are distributed across NUMA nodes, defer
2982 * hash allocation until vmalloc space is available.
2988 alloc_large_system_hash("Dentry cache",
2989 sizeof(struct dcache_hash_bucket),
2997 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2998 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3001 static void __init dcache_init(void)
3006 * A constructor could be added for stable state like the lists,
3007 * but it is probably not worth it because of the cache nature
3010 dentry_cache = KMEM_CACHE(dentry,
3011 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3013 register_shrinker(&dcache_shrinker);
3015 /* Hash may have been set up in dcache_init_early */
3020 alloc_large_system_hash("Dentry cache",
3021 sizeof(struct dcache_hash_bucket),
3029 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3030 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3033 /* SLAB cache for __getname() consumers */
3034 struct kmem_cache *names_cachep __read_mostly;
3035 EXPORT_SYMBOL(names_cachep);
3037 EXPORT_SYMBOL(d_genocide);
3039 void __init vfs_caches_init_early(void)
3041 dcache_init_early();
3045 void __init vfs_caches_init(unsigned long mempages)
3047 unsigned long reserve;
3049 /* Base hash sizes on available memory, with a reserve equal to
3050 150% of current kernel size */
3052 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3053 mempages -= reserve;
3055 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3056 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3060 files_init(mempages);