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->d_inode->i_lock protects:
43 * - i_dentry, d_alias, d_inode of aliases
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
61 * dentry->d_inode->i_lock
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 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
84 EXPORT_SYMBOL(rename_lock);
86 static struct kmem_cache *dentry_cache __read_mostly;
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
96 #define D_HASHBITS d_hash_shift
97 #define D_HASHMASK d_hash_mask
99 static unsigned int d_hash_mask __read_mostly;
100 static unsigned int d_hash_shift __read_mostly;
102 struct dcache_hash_bucket {
103 struct hlist_bl_head head;
105 static struct dcache_hash_bucket *dentry_hashtable __read_mostly;
107 static inline struct dcache_hash_bucket *d_hash(struct dentry *parent,
110 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
111 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
112 return dentry_hashtable + (hash & D_HASHMASK);
115 static inline void spin_lock_bucket(struct dcache_hash_bucket *b)
117 bit_spin_lock(0, (unsigned long *)&b->head.first);
120 static inline void spin_unlock_bucket(struct dcache_hash_bucket *b)
122 __bit_spin_unlock(0, (unsigned long *)&b->head.first);
125 /* Statistics gathering. */
126 struct dentry_stat_t dentry_stat = {
130 static DEFINE_PER_CPU(unsigned int, nr_dentry);
132 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
133 static int get_nr_dentry(void)
137 for_each_possible_cpu(i)
138 sum += per_cpu(nr_dentry, i);
139 return sum < 0 ? 0 : sum;
142 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
143 size_t *lenp, loff_t *ppos)
145 dentry_stat.nr_dentry = get_nr_dentry();
146 return proc_dointvec(table, write, buffer, lenp, ppos);
150 static void __d_free(struct rcu_head *head)
152 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
154 WARN_ON(!list_empty(&dentry->d_alias));
155 if (dname_external(dentry))
156 kfree(dentry->d_name.name);
157 kmem_cache_free(dentry_cache, dentry);
163 static void d_free(struct dentry *dentry)
165 BUG_ON(dentry->d_count);
166 this_cpu_dec(nr_dentry);
167 if (dentry->d_op && dentry->d_op->d_release)
168 dentry->d_op->d_release(dentry);
170 /* if dentry was never inserted into hash, immediate free is OK */
171 if (hlist_bl_unhashed(&dentry->d_hash))
172 __d_free(&dentry->d_u.d_rcu);
174 call_rcu(&dentry->d_u.d_rcu, __d_free);
178 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
179 * @dentry: the target dentry
180 * After this call, in-progress rcu-walk path lookup will fail. This
181 * should be called after unhashing, and after changing d_inode (if
182 * the dentry has not already been unhashed).
184 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
186 assert_spin_locked(&dentry->d_lock);
187 /* Go through a barrier */
188 write_seqcount_barrier(&dentry->d_seq);
192 * Release the dentry's inode, using the filesystem
193 * d_iput() operation if defined. Dentry has no refcount
196 static void dentry_iput(struct dentry * dentry)
197 __releases(dentry->d_lock)
198 __releases(dentry->d_inode->i_lock)
200 struct inode *inode = dentry->d_inode;
202 dentry->d_inode = NULL;
203 list_del_init(&dentry->d_alias);
204 spin_unlock(&dentry->d_lock);
205 spin_unlock(&inode->i_lock);
207 fsnotify_inoderemove(inode);
208 if (dentry->d_op && dentry->d_op->d_iput)
209 dentry->d_op->d_iput(dentry, inode);
213 spin_unlock(&dentry->d_lock);
218 * Release the dentry's inode, using the filesystem
219 * d_iput() operation if defined. dentry remains in-use.
221 static void dentry_unlink_inode(struct dentry * dentry)
222 __releases(dentry->d_lock)
223 __releases(dentry->d_inode->i_lock)
225 struct inode *inode = dentry->d_inode;
226 dentry->d_inode = NULL;
227 list_del_init(&dentry->d_alias);
228 dentry_rcuwalk_barrier(dentry);
229 spin_unlock(&dentry->d_lock);
230 spin_unlock(&inode->i_lock);
232 fsnotify_inoderemove(inode);
233 if (dentry->d_op && dentry->d_op->d_iput)
234 dentry->d_op->d_iput(dentry, inode);
240 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
242 static void dentry_lru_add(struct dentry *dentry)
244 if (list_empty(&dentry->d_lru)) {
245 spin_lock(&dcache_lru_lock);
246 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
247 dentry->d_sb->s_nr_dentry_unused++;
248 dentry_stat.nr_unused++;
249 spin_unlock(&dcache_lru_lock);
253 static void __dentry_lru_del(struct dentry *dentry)
255 list_del_init(&dentry->d_lru);
256 dentry->d_sb->s_nr_dentry_unused--;
257 dentry_stat.nr_unused--;
260 static void dentry_lru_del(struct dentry *dentry)
262 if (!list_empty(&dentry->d_lru)) {
263 spin_lock(&dcache_lru_lock);
264 __dentry_lru_del(dentry);
265 spin_unlock(&dcache_lru_lock);
269 static void dentry_lru_move_tail(struct dentry *dentry)
271 spin_lock(&dcache_lru_lock);
272 if (list_empty(&dentry->d_lru)) {
273 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
274 dentry->d_sb->s_nr_dentry_unused++;
275 dentry_stat.nr_unused++;
277 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
279 spin_unlock(&dcache_lru_lock);
283 * d_kill - kill dentry and return parent
284 * @dentry: dentry to kill
285 * @parent: parent dentry
287 * The dentry must already be unhashed and removed from the LRU.
289 * If this is the root of the dentry tree, return NULL.
291 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
294 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
295 __releases(dentry->d_lock)
296 __releases(parent->d_lock)
297 __releases(dentry->d_inode->i_lock)
299 list_del(&dentry->d_u.d_child);
301 * Inform try_to_ascend() that we are no longer attached to the
304 dentry->d_flags |= DCACHE_DISCONNECTED;
306 spin_unlock(&parent->d_lock);
309 * dentry_iput drops the locks, at which point nobody (except
310 * transient RCU lookups) can reach this dentry.
317 * d_drop - drop a dentry
318 * @dentry: dentry to drop
320 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
321 * be found through a VFS lookup any more. Note that this is different from
322 * deleting the dentry - d_delete will try to mark the dentry negative if
323 * possible, giving a successful _negative_ lookup, while d_drop will
324 * just make the cache lookup fail.
326 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
327 * reason (NFS timeouts or autofs deletes).
329 * __d_drop requires dentry->d_lock.
331 void __d_drop(struct dentry *dentry)
333 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
334 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) {
336 (unsigned long *)&dentry->d_sb->s_anon.first);
337 dentry->d_flags |= DCACHE_UNHASHED;
338 hlist_bl_del_init(&dentry->d_hash);
340 (unsigned long *)&dentry->d_sb->s_anon.first);
342 struct dcache_hash_bucket *b;
343 b = d_hash(dentry->d_parent, dentry->d_name.hash);
346 * We may not actually need to put DCACHE_UNHASHED
347 * manipulations under the hash lock, but follow
348 * the principle of least surprise.
350 dentry->d_flags |= DCACHE_UNHASHED;
351 hlist_bl_del_rcu(&dentry->d_hash);
352 spin_unlock_bucket(b);
353 dentry_rcuwalk_barrier(dentry);
357 EXPORT_SYMBOL(__d_drop);
359 void d_drop(struct dentry *dentry)
361 spin_lock(&dentry->d_lock);
363 spin_unlock(&dentry->d_lock);
365 EXPORT_SYMBOL(d_drop);
368 * Finish off a dentry we've decided to kill.
369 * dentry->d_lock must be held, returns with it unlocked.
370 * If ref is non-zero, then decrement the refcount too.
371 * Returns dentry requiring refcount drop, or NULL if we're done.
373 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
374 __releases(dentry->d_lock)
377 struct dentry *parent;
379 inode = dentry->d_inode;
380 if (inode && !spin_trylock(&inode->i_lock)) {
382 spin_unlock(&dentry->d_lock);
384 return dentry; /* try again with same dentry */
389 parent = dentry->d_parent;
390 if (parent && !spin_trylock(&parent->d_lock)) {
392 spin_unlock(&inode->i_lock);
398 /* if dentry was on the d_lru list delete it from there */
399 dentry_lru_del(dentry);
400 /* if it was on the hash then remove it */
402 return d_kill(dentry, parent);
408 * This is complicated by the fact that we do not want to put
409 * dentries that are no longer on any hash chain on the unused
410 * list: we'd much rather just get rid of them immediately.
412 * However, that implies that we have to traverse the dentry
413 * tree upwards to the parents which might _also_ now be
414 * scheduled for deletion (it may have been only waiting for
415 * its last child to go away).
417 * This tail recursion is done by hand as we don't want to depend
418 * on the compiler to always get this right (gcc generally doesn't).
419 * Real recursion would eat up our stack space.
423 * dput - release a dentry
424 * @dentry: dentry to release
426 * Release a dentry. This will drop the usage count and if appropriate
427 * call the dentry unlink method as well as removing it from the queues and
428 * releasing its resources. If the parent dentries were scheduled for release
429 * they too may now get deleted.
431 void dput(struct dentry *dentry)
437 if (dentry->d_count == 1)
439 spin_lock(&dentry->d_lock);
440 BUG_ON(!dentry->d_count);
441 if (dentry->d_count > 1) {
443 spin_unlock(&dentry->d_lock);
447 if (dentry->d_flags & DCACHE_OP_DELETE) {
448 if (dentry->d_op->d_delete(dentry))
452 /* Unreachable? Get rid of it */
453 if (d_unhashed(dentry))
456 /* Otherwise leave it cached and ensure it's on the LRU */
457 dentry->d_flags |= DCACHE_REFERENCED;
458 dentry_lru_add(dentry);
461 spin_unlock(&dentry->d_lock);
465 dentry = dentry_kill(dentry, 1);
472 * d_invalidate - invalidate a dentry
473 * @dentry: dentry to invalidate
475 * Try to invalidate the dentry if it turns out to be
476 * possible. If there are other dentries that can be
477 * reached through this one we can't delete it and we
478 * return -EBUSY. On success we return 0.
483 int d_invalidate(struct dentry * dentry)
486 * If it's already been dropped, return OK.
488 spin_lock(&dentry->d_lock);
489 if (d_unhashed(dentry)) {
490 spin_unlock(&dentry->d_lock);
494 * Check whether to do a partial shrink_dcache
495 * to get rid of unused child entries.
497 if (!list_empty(&dentry->d_subdirs)) {
498 spin_unlock(&dentry->d_lock);
499 shrink_dcache_parent(dentry);
500 spin_lock(&dentry->d_lock);
504 * Somebody else still using it?
506 * If it's a directory, we can't drop it
507 * for fear of somebody re-populating it
508 * with children (even though dropping it
509 * would make it unreachable from the root,
510 * we might still populate it if it was a
511 * working directory or similar).
513 if (dentry->d_count > 1) {
514 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
515 spin_unlock(&dentry->d_lock);
521 spin_unlock(&dentry->d_lock);
524 EXPORT_SYMBOL(d_invalidate);
526 /* This must be called with d_lock held */
527 static inline void __dget_dlock(struct dentry *dentry)
532 static inline void __dget(struct dentry *dentry)
534 spin_lock(&dentry->d_lock);
535 __dget_dlock(dentry);
536 spin_unlock(&dentry->d_lock);
539 struct dentry *dget_parent(struct dentry *dentry)
545 * Don't need rcu_dereference because we re-check it was correct under
549 ret = dentry->d_parent;
554 spin_lock(&ret->d_lock);
555 if (unlikely(ret != dentry->d_parent)) {
556 spin_unlock(&ret->d_lock);
561 BUG_ON(!ret->d_count);
563 spin_unlock(&ret->d_lock);
567 EXPORT_SYMBOL(dget_parent);
570 * d_find_alias - grab a hashed alias of inode
571 * @inode: inode in question
572 * @want_discon: flag, used by d_splice_alias, to request
573 * that only a DISCONNECTED alias be returned.
575 * If inode has a hashed alias, or is a directory and has any alias,
576 * acquire the reference to alias and return it. Otherwise return NULL.
577 * Notice that if inode is a directory there can be only one alias and
578 * it can be unhashed only if it has no children, or if it is the root
581 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
582 * any other hashed alias over that one unless @want_discon is set,
583 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
585 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
587 struct dentry *alias, *discon_alias;
591 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
592 spin_lock(&alias->d_lock);
593 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
594 if (IS_ROOT(alias) &&
595 (alias->d_flags & DCACHE_DISCONNECTED)) {
596 discon_alias = alias;
597 } else if (!want_discon) {
599 spin_unlock(&alias->d_lock);
603 spin_unlock(&alias->d_lock);
606 alias = discon_alias;
607 spin_lock(&alias->d_lock);
608 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
609 if (IS_ROOT(alias) &&
610 (alias->d_flags & DCACHE_DISCONNECTED)) {
612 spin_unlock(&alias->d_lock);
616 spin_unlock(&alias->d_lock);
622 struct dentry *d_find_alias(struct inode *inode)
624 struct dentry *de = NULL;
626 if (!list_empty(&inode->i_dentry)) {
627 spin_lock(&inode->i_lock);
628 de = __d_find_alias(inode, 0);
629 spin_unlock(&inode->i_lock);
633 EXPORT_SYMBOL(d_find_alias);
636 * Try to kill dentries associated with this inode.
637 * WARNING: you must own a reference to inode.
639 void d_prune_aliases(struct inode *inode)
641 struct dentry *dentry;
643 spin_lock(&inode->i_lock);
644 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
645 spin_lock(&dentry->d_lock);
646 if (!dentry->d_count) {
647 __dget_dlock(dentry);
649 spin_unlock(&dentry->d_lock);
650 spin_unlock(&inode->i_lock);
654 spin_unlock(&dentry->d_lock);
656 spin_unlock(&inode->i_lock);
658 EXPORT_SYMBOL(d_prune_aliases);
661 * Try to throw away a dentry - free the inode, dput the parent.
662 * Requires dentry->d_lock is held, and dentry->d_count == 0.
663 * Releases dentry->d_lock.
665 * This may fail if locks cannot be acquired no problem, just try again.
667 static void try_prune_one_dentry(struct dentry *dentry)
668 __releases(dentry->d_lock)
670 struct dentry *parent;
672 parent = dentry_kill(dentry, 0);
674 * If dentry_kill returns NULL, we have nothing more to do.
675 * if it returns the same dentry, trylocks failed. In either
676 * case, just loop again.
678 * Otherwise, we need to prune ancestors too. This is necessary
679 * to prevent quadratic behavior of shrink_dcache_parent(), but
680 * is also expected to be beneficial in reducing dentry cache
685 if (parent == dentry)
688 /* Prune ancestors. */
691 spin_lock(&dentry->d_lock);
692 if (dentry->d_count > 1) {
694 spin_unlock(&dentry->d_lock);
697 dentry = dentry_kill(dentry, 1);
701 static void shrink_dentry_list(struct list_head *list)
703 struct dentry *dentry;
707 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
708 if (&dentry->d_lru == list)
710 spin_lock(&dentry->d_lock);
711 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
712 spin_unlock(&dentry->d_lock);
717 * We found an inuse dentry which was not removed from
718 * the LRU because of laziness during lookup. Do not free
719 * it - just keep it off the LRU list.
721 if (dentry->d_count) {
722 dentry_lru_del(dentry);
723 spin_unlock(&dentry->d_lock);
729 try_prune_one_dentry(dentry);
737 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
738 * @sb: superblock to shrink dentry LRU.
739 * @count: number of entries to prune
740 * @flags: flags to control the dentry processing
742 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
744 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
746 /* called from prune_dcache() and shrink_dcache_parent() */
747 struct dentry *dentry;
748 LIST_HEAD(referenced);
753 spin_lock(&dcache_lru_lock);
754 while (!list_empty(&sb->s_dentry_lru)) {
755 dentry = list_entry(sb->s_dentry_lru.prev,
756 struct dentry, d_lru);
757 BUG_ON(dentry->d_sb != sb);
759 if (!spin_trylock(&dentry->d_lock)) {
760 spin_unlock(&dcache_lru_lock);
766 * If we are honouring the DCACHE_REFERENCED flag and the
767 * dentry has this flag set, don't free it. Clear the flag
768 * and put it back on the LRU.
770 if (flags & DCACHE_REFERENCED &&
771 dentry->d_flags & DCACHE_REFERENCED) {
772 dentry->d_flags &= ~DCACHE_REFERENCED;
773 list_move(&dentry->d_lru, &referenced);
774 spin_unlock(&dentry->d_lock);
776 list_move_tail(&dentry->d_lru, &tmp);
777 spin_unlock(&dentry->d_lock);
781 cond_resched_lock(&dcache_lru_lock);
783 if (!list_empty(&referenced))
784 list_splice(&referenced, &sb->s_dentry_lru);
785 spin_unlock(&dcache_lru_lock);
787 shrink_dentry_list(&tmp);
793 * prune_dcache - shrink the dcache
794 * @count: number of entries to try to free
796 * Shrink the dcache. This is done when we need more memory, or simply when we
797 * need to unmount something (at which point we need to unuse all dentries).
799 * This function may fail to free any resources if all the dentries are in use.
801 static void prune_dcache(int count)
803 struct super_block *sb, *p = NULL;
805 int unused = dentry_stat.nr_unused;
809 if (unused == 0 || count == 0)
814 prune_ratio = unused / count;
816 list_for_each_entry(sb, &super_blocks, s_list) {
817 if (list_empty(&sb->s_instances))
819 if (sb->s_nr_dentry_unused == 0)
822 /* Now, we reclaim unused dentrins with fairness.
823 * We reclaim them same percentage from each superblock.
824 * We calculate number of dentries to scan on this sb
825 * as follows, but the implementation is arranged to avoid
827 * number of dentries to scan on this sb =
828 * count * (number of dentries on this sb /
829 * number of dentries in the machine)
831 spin_unlock(&sb_lock);
832 if (prune_ratio != 1)
833 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
835 w_count = sb->s_nr_dentry_unused;
838 * We need to be sure this filesystem isn't being unmounted,
839 * otherwise we could race with generic_shutdown_super(), and
840 * end up holding a reference to an inode while the filesystem
841 * is unmounted. So we try to get s_umount, and make sure
844 if (down_read_trylock(&sb->s_umount)) {
845 if ((sb->s_root != NULL) &&
846 (!list_empty(&sb->s_dentry_lru))) {
847 __shrink_dcache_sb(sb, &w_count,
851 up_read(&sb->s_umount);
858 /* more work left to do? */
864 spin_unlock(&sb_lock);
868 * shrink_dcache_sb - shrink dcache for a superblock
871 * Shrink the dcache for the specified super block. This is used to free
872 * the dcache before unmounting a file system.
874 void shrink_dcache_sb(struct super_block *sb)
878 spin_lock(&dcache_lru_lock);
879 while (!list_empty(&sb->s_dentry_lru)) {
880 list_splice_init(&sb->s_dentry_lru, &tmp);
881 spin_unlock(&dcache_lru_lock);
882 shrink_dentry_list(&tmp);
883 spin_lock(&dcache_lru_lock);
885 spin_unlock(&dcache_lru_lock);
887 EXPORT_SYMBOL(shrink_dcache_sb);
890 * destroy a single subtree of dentries for unmount
891 * - see the comments on shrink_dcache_for_umount() for a description of the
894 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
896 struct dentry *parent;
897 unsigned detached = 0;
899 BUG_ON(!IS_ROOT(dentry));
901 /* detach this root from the system */
902 spin_lock(&dentry->d_lock);
903 dentry_lru_del(dentry);
905 spin_unlock(&dentry->d_lock);
908 /* descend to the first leaf in the current subtree */
909 while (!list_empty(&dentry->d_subdirs)) {
912 /* this is a branch with children - detach all of them
913 * from the system in one go */
914 spin_lock(&dentry->d_lock);
915 list_for_each_entry(loop, &dentry->d_subdirs,
917 spin_lock_nested(&loop->d_lock,
918 DENTRY_D_LOCK_NESTED);
919 dentry_lru_del(loop);
921 spin_unlock(&loop->d_lock);
923 spin_unlock(&dentry->d_lock);
925 /* move to the first child */
926 dentry = list_entry(dentry->d_subdirs.next,
927 struct dentry, d_u.d_child);
930 /* consume the dentries from this leaf up through its parents
931 * until we find one with children or run out altogether */
935 if (dentry->d_count != 0) {
937 "BUG: Dentry %p{i=%lx,n=%s}"
939 " [unmount of %s %s]\n",
942 dentry->d_inode->i_ino : 0UL,
945 dentry->d_sb->s_type->name,
950 if (IS_ROOT(dentry)) {
952 list_del(&dentry->d_u.d_child);
954 parent = dentry->d_parent;
955 spin_lock(&parent->d_lock);
957 list_del(&dentry->d_u.d_child);
958 spin_unlock(&parent->d_lock);
963 inode = dentry->d_inode;
965 dentry->d_inode = NULL;
966 list_del_init(&dentry->d_alias);
967 if (dentry->d_op && dentry->d_op->d_iput)
968 dentry->d_op->d_iput(dentry, inode);
975 /* finished when we fall off the top of the tree,
976 * otherwise we ascend to the parent and move to the
977 * next sibling if there is one */
981 } while (list_empty(&dentry->d_subdirs));
983 dentry = list_entry(dentry->d_subdirs.next,
984 struct dentry, d_u.d_child);
989 * destroy the dentries attached to a superblock on unmounting
990 * - we don't need to use dentry->d_lock because:
991 * - the superblock is detached from all mountings and open files, so the
992 * dentry trees will not be rearranged by the VFS
993 * - s_umount is write-locked, so the memory pressure shrinker will ignore
994 * any dentries belonging to this superblock that it comes across
995 * - the filesystem itself is no longer permitted to rearrange the dentries
998 void shrink_dcache_for_umount(struct super_block *sb)
1000 struct dentry *dentry;
1002 if (down_read_trylock(&sb->s_umount))
1005 dentry = sb->s_root;
1007 spin_lock(&dentry->d_lock);
1009 spin_unlock(&dentry->d_lock);
1010 shrink_dcache_for_umount_subtree(dentry);
1012 while (!hlist_bl_empty(&sb->s_anon)) {
1013 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1014 shrink_dcache_for_umount_subtree(dentry);
1019 * This tries to ascend one level of parenthood, but
1020 * we can race with renaming, so we need to re-check
1021 * the parenthood after dropping the lock and check
1022 * that the sequence number still matches.
1024 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1026 struct dentry *new = old->d_parent;
1029 spin_unlock(&old->d_lock);
1030 spin_lock(&new->d_lock);
1033 * might go back up the wrong parent if we have had a rename
1036 if (new != old->d_parent ||
1037 (old->d_flags & DCACHE_DISCONNECTED) ||
1038 (!locked && read_seqretry(&rename_lock, seq))) {
1039 spin_unlock(&new->d_lock);
1048 * Search for at least 1 mount point in the dentry's subdirs.
1049 * We descend to the next level whenever the d_subdirs
1050 * list is non-empty and continue searching.
1054 * have_submounts - check for mounts over a dentry
1055 * @parent: dentry to check.
1057 * Return true if the parent or its subdirectories contain
1060 int have_submounts(struct dentry *parent)
1062 struct dentry *this_parent;
1063 struct list_head *next;
1067 seq = read_seqbegin(&rename_lock);
1069 this_parent = parent;
1071 if (d_mountpoint(parent))
1073 spin_lock(&this_parent->d_lock);
1075 next = this_parent->d_subdirs.next;
1077 while (next != &this_parent->d_subdirs) {
1078 struct list_head *tmp = next;
1079 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1082 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1083 /* Have we found a mount point ? */
1084 if (d_mountpoint(dentry)) {
1085 spin_unlock(&dentry->d_lock);
1086 spin_unlock(&this_parent->d_lock);
1089 if (!list_empty(&dentry->d_subdirs)) {
1090 spin_unlock(&this_parent->d_lock);
1091 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1092 this_parent = dentry;
1093 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1096 spin_unlock(&dentry->d_lock);
1099 * All done at this level ... ascend and resume the search.
1101 if (this_parent != parent) {
1102 struct dentry *child = this_parent;
1103 this_parent = try_to_ascend(this_parent, locked, seq);
1106 next = child->d_u.d_child.next;
1109 spin_unlock(&this_parent->d_lock);
1110 if (!locked && read_seqretry(&rename_lock, seq))
1113 write_sequnlock(&rename_lock);
1114 return 0; /* No mount points found in tree */
1116 if (!locked && read_seqretry(&rename_lock, seq))
1119 write_sequnlock(&rename_lock);
1124 write_seqlock(&rename_lock);
1127 EXPORT_SYMBOL(have_submounts);
1130 * Search the dentry child list for the specified parent,
1131 * and move any unused dentries to the end of the unused
1132 * list for prune_dcache(). We descend to the next level
1133 * whenever the d_subdirs list is non-empty and continue
1136 * It returns zero iff there are no unused children,
1137 * otherwise it returns the number of children moved to
1138 * the end of the unused list. This may not be the total
1139 * number of unused children, because select_parent can
1140 * drop the lock and return early due to latency
1143 static int select_parent(struct dentry * parent)
1145 struct dentry *this_parent;
1146 struct list_head *next;
1151 seq = read_seqbegin(&rename_lock);
1153 this_parent = parent;
1154 spin_lock(&this_parent->d_lock);
1156 next = this_parent->d_subdirs.next;
1158 while (next != &this_parent->d_subdirs) {
1159 struct list_head *tmp = next;
1160 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1163 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1166 * move only zero ref count dentries to the end
1167 * of the unused list for prune_dcache
1169 if (!dentry->d_count) {
1170 dentry_lru_move_tail(dentry);
1173 dentry_lru_del(dentry);
1177 * We can return to the caller if we have found some (this
1178 * ensures forward progress). We'll be coming back to find
1181 if (found && need_resched()) {
1182 spin_unlock(&dentry->d_lock);
1187 * Descend a level if the d_subdirs list is non-empty.
1189 if (!list_empty(&dentry->d_subdirs)) {
1190 spin_unlock(&this_parent->d_lock);
1191 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1192 this_parent = dentry;
1193 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1197 spin_unlock(&dentry->d_lock);
1200 * All done at this level ... ascend and resume the search.
1202 if (this_parent != parent) {
1203 struct dentry *child = this_parent;
1204 this_parent = try_to_ascend(this_parent, locked, seq);
1207 next = child->d_u.d_child.next;
1211 spin_unlock(&this_parent->d_lock);
1212 if (!locked && read_seqretry(&rename_lock, seq))
1215 write_sequnlock(&rename_lock);
1222 write_seqlock(&rename_lock);
1227 * shrink_dcache_parent - prune dcache
1228 * @parent: parent of entries to prune
1230 * Prune the dcache to remove unused children of the parent dentry.
1233 void shrink_dcache_parent(struct dentry * parent)
1235 struct super_block *sb = parent->d_sb;
1238 while ((found = select_parent(parent)) != 0)
1239 __shrink_dcache_sb(sb, &found, 0);
1241 EXPORT_SYMBOL(shrink_dcache_parent);
1244 * Scan `nr' dentries and return the number which remain.
1246 * We need to avoid reentering the filesystem if the caller is performing a
1247 * GFP_NOFS allocation attempt. One example deadlock is:
1249 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1250 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1251 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1253 * In this case we return -1 to tell the caller that we baled.
1255 static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1258 if (!(gfp_mask & __GFP_FS))
1263 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
1266 static struct shrinker dcache_shrinker = {
1267 .shrink = shrink_dcache_memory,
1268 .seeks = DEFAULT_SEEKS,
1272 * d_alloc - allocate a dcache entry
1273 * @parent: parent of entry to allocate
1274 * @name: qstr of the name
1276 * Allocates a dentry. It returns %NULL if there is insufficient memory
1277 * available. On a success the dentry is returned. The name passed in is
1278 * copied and the copy passed in may be reused after this call.
1281 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1283 struct dentry *dentry;
1286 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1290 if (name->len > DNAME_INLINE_LEN-1) {
1291 dname = kmalloc(name->len + 1, GFP_KERNEL);
1293 kmem_cache_free(dentry_cache, dentry);
1297 dname = dentry->d_iname;
1299 dentry->d_name.name = dname;
1301 dentry->d_name.len = name->len;
1302 dentry->d_name.hash = name->hash;
1303 memcpy(dname, name->name, name->len);
1304 dname[name->len] = 0;
1306 dentry->d_count = 1;
1307 dentry->d_flags = DCACHE_UNHASHED;
1308 spin_lock_init(&dentry->d_lock);
1309 seqcount_init(&dentry->d_seq);
1310 dentry->d_inode = NULL;
1311 dentry->d_parent = NULL;
1312 dentry->d_sb = NULL;
1313 dentry->d_op = NULL;
1314 dentry->d_fsdata = NULL;
1315 INIT_HLIST_BL_NODE(&dentry->d_hash);
1316 INIT_LIST_HEAD(&dentry->d_lru);
1317 INIT_LIST_HEAD(&dentry->d_subdirs);
1318 INIT_LIST_HEAD(&dentry->d_alias);
1319 INIT_LIST_HEAD(&dentry->d_u.d_child);
1322 spin_lock(&parent->d_lock);
1324 * don't need child lock because it is not subject
1325 * to concurrency here
1327 __dget_dlock(parent);
1328 dentry->d_parent = parent;
1329 dentry->d_sb = parent->d_sb;
1330 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1331 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1332 spin_unlock(&parent->d_lock);
1335 this_cpu_inc(nr_dentry);
1339 EXPORT_SYMBOL(d_alloc);
1341 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1343 struct dentry *dentry = d_alloc(NULL, name);
1346 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1347 dentry->d_parent = dentry;
1348 dentry->d_flags |= DCACHE_DISCONNECTED;
1352 EXPORT_SYMBOL(d_alloc_pseudo);
1354 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1359 q.len = strlen(name);
1360 q.hash = full_name_hash(q.name, q.len);
1361 return d_alloc(parent, &q);
1363 EXPORT_SYMBOL(d_alloc_name);
1365 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1367 WARN_ON_ONCE(dentry->d_op);
1368 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1370 DCACHE_OP_REVALIDATE |
1371 DCACHE_OP_DELETE ));
1376 dentry->d_flags |= DCACHE_OP_HASH;
1378 dentry->d_flags |= DCACHE_OP_COMPARE;
1379 if (op->d_revalidate)
1380 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1382 dentry->d_flags |= DCACHE_OP_DELETE;
1385 EXPORT_SYMBOL(d_set_d_op);
1387 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1389 spin_lock(&dentry->d_lock);
1391 if (unlikely(IS_AUTOMOUNT(inode)))
1392 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1393 list_add(&dentry->d_alias, &inode->i_dentry);
1395 dentry->d_inode = inode;
1396 dentry_rcuwalk_barrier(dentry);
1397 spin_unlock(&dentry->d_lock);
1398 fsnotify_d_instantiate(dentry, inode);
1402 * d_instantiate - fill in inode information for a dentry
1403 * @entry: dentry to complete
1404 * @inode: inode to attach to this dentry
1406 * Fill in inode information in the entry.
1408 * This turns negative dentries into productive full members
1411 * NOTE! This assumes that the inode count has been incremented
1412 * (or otherwise set) by the caller to indicate that it is now
1413 * in use by the dcache.
1416 void d_instantiate(struct dentry *entry, struct inode * inode)
1418 BUG_ON(!list_empty(&entry->d_alias));
1420 spin_lock(&inode->i_lock);
1421 __d_instantiate(entry, inode);
1423 spin_unlock(&inode->i_lock);
1424 security_d_instantiate(entry, inode);
1426 EXPORT_SYMBOL(d_instantiate);
1429 * d_instantiate_unique - instantiate a non-aliased dentry
1430 * @entry: dentry to instantiate
1431 * @inode: inode to attach to this dentry
1433 * Fill in inode information in the entry. On success, it returns NULL.
1434 * If an unhashed alias of "entry" already exists, then we return the
1435 * aliased dentry instead and drop one reference to inode.
1437 * Note that in order to avoid conflicts with rename() etc, the caller
1438 * had better be holding the parent directory semaphore.
1440 * This also assumes that the inode count has been incremented
1441 * (or otherwise set) by the caller to indicate that it is now
1442 * in use by the dcache.
1444 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1445 struct inode *inode)
1447 struct dentry *alias;
1448 int len = entry->d_name.len;
1449 const char *name = entry->d_name.name;
1450 unsigned int hash = entry->d_name.hash;
1453 __d_instantiate(entry, NULL);
1457 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1458 struct qstr *qstr = &alias->d_name;
1461 * Don't need alias->d_lock here, because aliases with
1462 * d_parent == entry->d_parent are not subject to name or
1463 * parent changes, because the parent inode i_mutex is held.
1465 if (qstr->hash != hash)
1467 if (alias->d_parent != entry->d_parent)
1469 if (dentry_cmp(qstr->name, qstr->len, name, len))
1475 __d_instantiate(entry, inode);
1479 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1481 struct dentry *result;
1483 BUG_ON(!list_empty(&entry->d_alias));
1486 spin_lock(&inode->i_lock);
1487 result = __d_instantiate_unique(entry, inode);
1489 spin_unlock(&inode->i_lock);
1492 security_d_instantiate(entry, inode);
1496 BUG_ON(!d_unhashed(result));
1501 EXPORT_SYMBOL(d_instantiate_unique);
1504 * d_alloc_root - allocate root dentry
1505 * @root_inode: inode to allocate the root for
1507 * Allocate a root ("/") dentry for the inode given. The inode is
1508 * instantiated and returned. %NULL is returned if there is insufficient
1509 * memory or the inode passed is %NULL.
1512 struct dentry * d_alloc_root(struct inode * root_inode)
1514 struct dentry *res = NULL;
1517 static const struct qstr name = { .name = "/", .len = 1 };
1519 res = d_alloc(NULL, &name);
1521 res->d_sb = root_inode->i_sb;
1522 d_set_d_op(res, res->d_sb->s_d_op);
1523 res->d_parent = res;
1524 d_instantiate(res, root_inode);
1529 EXPORT_SYMBOL(d_alloc_root);
1531 static struct dentry * __d_find_any_alias(struct inode *inode)
1533 struct dentry *alias;
1535 if (list_empty(&inode->i_dentry))
1537 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1542 static struct dentry * d_find_any_alias(struct inode *inode)
1546 spin_lock(&inode->i_lock);
1547 de = __d_find_any_alias(inode);
1548 spin_unlock(&inode->i_lock);
1554 * d_obtain_alias - find or allocate a dentry for a given inode
1555 * @inode: inode to allocate the dentry for
1557 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1558 * similar open by handle operations. The returned dentry may be anonymous,
1559 * or may have a full name (if the inode was already in the cache).
1561 * When called on a directory inode, we must ensure that the inode only ever
1562 * has one dentry. If a dentry is found, that is returned instead of
1563 * allocating a new one.
1565 * On successful return, the reference to the inode has been transferred
1566 * to the dentry. In case of an error the reference on the inode is released.
1567 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1568 * be passed in and will be the error will be propagate to the return value,
1569 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1571 struct dentry *d_obtain_alias(struct inode *inode)
1573 static const struct qstr anonstring = { .name = "" };
1578 return ERR_PTR(-ESTALE);
1580 return ERR_CAST(inode);
1582 res = d_find_any_alias(inode);
1586 tmp = d_alloc(NULL, &anonstring);
1588 res = ERR_PTR(-ENOMEM);
1591 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1594 spin_lock(&inode->i_lock);
1595 res = __d_find_any_alias(inode);
1597 spin_unlock(&inode->i_lock);
1602 /* attach a disconnected dentry */
1603 spin_lock(&tmp->d_lock);
1604 tmp->d_sb = inode->i_sb;
1605 d_set_d_op(tmp, tmp->d_sb->s_d_op);
1606 tmp->d_inode = inode;
1607 tmp->d_flags |= DCACHE_DISCONNECTED;
1608 list_add(&tmp->d_alias, &inode->i_dentry);
1609 bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1610 tmp->d_flags &= ~DCACHE_UNHASHED;
1611 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1612 __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1613 spin_unlock(&tmp->d_lock);
1614 spin_unlock(&inode->i_lock);
1622 EXPORT_SYMBOL(d_obtain_alias);
1625 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1626 * @inode: the inode which may have a disconnected dentry
1627 * @dentry: a negative dentry which we want to point to the inode.
1629 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1630 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1631 * and return it, else simply d_add the inode to the dentry and return NULL.
1633 * This is needed in the lookup routine of any filesystem that is exportable
1634 * (via knfsd) so that we can build dcache paths to directories effectively.
1636 * If a dentry was found and moved, then it is returned. Otherwise NULL
1637 * is returned. This matches the expected return value of ->lookup.
1640 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1642 struct dentry *new = NULL;
1644 if (inode && S_ISDIR(inode->i_mode)) {
1645 spin_lock(&inode->i_lock);
1646 new = __d_find_alias(inode, 1);
1648 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1649 spin_unlock(&inode->i_lock);
1650 security_d_instantiate(new, inode);
1651 d_move(new, dentry);
1654 /* already taking inode->i_lock, so d_add() by hand */
1655 __d_instantiate(dentry, inode);
1656 spin_unlock(&inode->i_lock);
1657 security_d_instantiate(dentry, inode);
1661 d_add(dentry, inode);
1664 EXPORT_SYMBOL(d_splice_alias);
1667 * d_add_ci - lookup or allocate new dentry with case-exact name
1668 * @inode: the inode case-insensitive lookup has found
1669 * @dentry: the negative dentry that was passed to the parent's lookup func
1670 * @name: the case-exact name to be associated with the returned dentry
1672 * This is to avoid filling the dcache with case-insensitive names to the
1673 * same inode, only the actual correct case is stored in the dcache for
1674 * case-insensitive filesystems.
1676 * For a case-insensitive lookup match and if the the case-exact dentry
1677 * already exists in in the dcache, use it and return it.
1679 * If no entry exists with the exact case name, allocate new dentry with
1680 * the exact case, and return the spliced entry.
1682 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1686 struct dentry *found;
1690 * First check if a dentry matching the name already exists,
1691 * if not go ahead and create it now.
1693 found = d_hash_and_lookup(dentry->d_parent, name);
1695 new = d_alloc(dentry->d_parent, name);
1701 found = d_splice_alias(inode, new);
1710 * If a matching dentry exists, and it's not negative use it.
1712 * Decrement the reference count to balance the iget() done
1715 if (found->d_inode) {
1716 if (unlikely(found->d_inode != inode)) {
1717 /* This can't happen because bad inodes are unhashed. */
1718 BUG_ON(!is_bad_inode(inode));
1719 BUG_ON(!is_bad_inode(found->d_inode));
1726 * Negative dentry: instantiate it unless the inode is a directory and
1727 * already has a dentry.
1729 spin_lock(&inode->i_lock);
1730 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1731 __d_instantiate(found, inode);
1732 spin_unlock(&inode->i_lock);
1733 security_d_instantiate(found, inode);
1738 * In case a directory already has a (disconnected) entry grab a
1739 * reference to it, move it in place and use it.
1741 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1743 spin_unlock(&inode->i_lock);
1744 security_d_instantiate(found, inode);
1752 return ERR_PTR(error);
1754 EXPORT_SYMBOL(d_add_ci);
1757 * __d_lookup_rcu - search for a dentry (racy, store-free)
1758 * @parent: parent dentry
1759 * @name: qstr of name we wish to find
1760 * @seq: returns d_seq value at the point where the dentry was found
1761 * @inode: returns dentry->d_inode when the inode was found valid.
1762 * Returns: dentry, or NULL
1764 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1765 * resolution (store-free path walking) design described in
1766 * Documentation/filesystems/path-lookup.txt.
1768 * This is not to be used outside core vfs.
1770 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1771 * held, and rcu_read_lock held. The returned dentry must not be stored into
1772 * without taking d_lock and checking d_seq sequence count against @seq
1775 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1778 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1779 * the returned dentry, so long as its parent's seqlock is checked after the
1780 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1781 * is formed, giving integrity down the path walk.
1783 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1784 unsigned *seq, struct inode **inode)
1786 unsigned int len = name->len;
1787 unsigned int hash = name->hash;
1788 const unsigned char *str = name->name;
1789 struct dcache_hash_bucket *b = d_hash(parent, hash);
1790 struct hlist_bl_node *node;
1791 struct dentry *dentry;
1794 * Note: There is significant duplication with __d_lookup_rcu which is
1795 * required to prevent single threaded performance regressions
1796 * especially on architectures where smp_rmb (in seqcounts) are costly.
1797 * Keep the two functions in sync.
1801 * The hash list is protected using RCU.
1803 * Carefully use d_seq when comparing a candidate dentry, to avoid
1804 * races with d_move().
1806 * It is possible that concurrent renames can mess up our list
1807 * walk here and result in missing our dentry, resulting in the
1808 * false-negative result. d_lookup() protects against concurrent
1809 * renames using rename_lock seqlock.
1811 * See Documentation/vfs/dcache-locking.txt for more details.
1813 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1818 if (dentry->d_name.hash != hash)
1822 *seq = read_seqcount_begin(&dentry->d_seq);
1823 if (dentry->d_parent != parent)
1825 if (d_unhashed(dentry))
1827 tlen = dentry->d_name.len;
1828 tname = dentry->d_name.name;
1829 i = dentry->d_inode;
1834 * This seqcount check is required to ensure name and
1835 * len are loaded atomically, so as not to walk off the
1836 * edge of memory when walking. If we could load this
1837 * atomically some other way, we could drop this check.
1839 if (read_seqcount_retry(&dentry->d_seq, *seq))
1841 if (parent->d_flags & DCACHE_OP_COMPARE) {
1842 if (parent->d_op->d_compare(parent, *inode,
1847 if (dentry_cmp(tname, tlen, str, len))
1851 * No extra seqcount check is required after the name
1852 * compare. The caller must perform a seqcount check in
1853 * order to do anything useful with the returned dentry
1863 * d_lookup - search for a dentry
1864 * @parent: parent dentry
1865 * @name: qstr of name we wish to find
1866 * Returns: dentry, or NULL
1868 * d_lookup searches the children of the parent dentry for the name in
1869 * question. If the dentry is found its reference count is incremented and the
1870 * dentry is returned. The caller must use dput to free the entry when it has
1871 * finished using it. %NULL is returned if the dentry does not exist.
1873 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1875 struct dentry *dentry;
1879 seq = read_seqbegin(&rename_lock);
1880 dentry = __d_lookup(parent, name);
1883 } while (read_seqretry(&rename_lock, seq));
1886 EXPORT_SYMBOL(d_lookup);
1889 * __d_lookup - search for a dentry (racy)
1890 * @parent: parent dentry
1891 * @name: qstr of name we wish to find
1892 * Returns: dentry, or NULL
1894 * __d_lookup is like d_lookup, however it may (rarely) return a
1895 * false-negative result due to unrelated rename activity.
1897 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1898 * however it must be used carefully, eg. with a following d_lookup in
1899 * the case of failure.
1901 * __d_lookup callers must be commented.
1903 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1905 unsigned int len = name->len;
1906 unsigned int hash = name->hash;
1907 const unsigned char *str = name->name;
1908 struct dcache_hash_bucket *b = d_hash(parent, hash);
1909 struct hlist_bl_node *node;
1910 struct dentry *found = NULL;
1911 struct dentry *dentry;
1914 * Note: There is significant duplication with __d_lookup_rcu which is
1915 * required to prevent single threaded performance regressions
1916 * especially on architectures where smp_rmb (in seqcounts) are costly.
1917 * Keep the two functions in sync.
1921 * The hash list is protected using RCU.
1923 * Take d_lock when comparing a candidate dentry, to avoid races
1926 * It is possible that concurrent renames can mess up our list
1927 * walk here and result in missing our dentry, resulting in the
1928 * false-negative result. d_lookup() protects against concurrent
1929 * renames using rename_lock seqlock.
1931 * See Documentation/vfs/dcache-locking.txt for more details.
1935 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1939 if (dentry->d_name.hash != hash)
1942 spin_lock(&dentry->d_lock);
1943 if (dentry->d_parent != parent)
1945 if (d_unhashed(dentry))
1949 * It is safe to compare names since d_move() cannot
1950 * change the qstr (protected by d_lock).
1952 tlen = dentry->d_name.len;
1953 tname = dentry->d_name.name;
1954 if (parent->d_flags & DCACHE_OP_COMPARE) {
1955 if (parent->d_op->d_compare(parent, parent->d_inode,
1956 dentry, dentry->d_inode,
1960 if (dentry_cmp(tname, tlen, str, len))
1966 spin_unlock(&dentry->d_lock);
1969 spin_unlock(&dentry->d_lock);
1977 * d_hash_and_lookup - hash the qstr then search for a dentry
1978 * @dir: Directory to search in
1979 * @name: qstr of name we wish to find
1981 * On hash failure or on lookup failure NULL is returned.
1983 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1985 struct dentry *dentry = NULL;
1988 * Check for a fs-specific hash function. Note that we must
1989 * calculate the standard hash first, as the d_op->d_hash()
1990 * routine may choose to leave the hash value unchanged.
1992 name->hash = full_name_hash(name->name, name->len);
1993 if (dir->d_flags & DCACHE_OP_HASH) {
1994 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1997 dentry = d_lookup(dir, name);
2003 * d_validate - verify dentry provided from insecure source (deprecated)
2004 * @dentry: The dentry alleged to be valid child of @dparent
2005 * @dparent: The parent dentry (known to be valid)
2007 * An insecure source has sent us a dentry, here we verify it and dget() it.
2008 * This is used by ncpfs in its readdir implementation.
2009 * Zero is returned in the dentry is invalid.
2011 * This function is slow for big directories, and deprecated, do not use it.
2013 int d_validate(struct dentry *dentry, struct dentry *dparent)
2015 struct dentry *child;
2017 spin_lock(&dparent->d_lock);
2018 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2019 if (dentry == child) {
2020 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2021 __dget_dlock(dentry);
2022 spin_unlock(&dentry->d_lock);
2023 spin_unlock(&dparent->d_lock);
2027 spin_unlock(&dparent->d_lock);
2031 EXPORT_SYMBOL(d_validate);
2034 * When a file is deleted, we have two options:
2035 * - turn this dentry into a negative dentry
2036 * - unhash this dentry and free it.
2038 * Usually, we want to just turn this into
2039 * a negative dentry, but if anybody else is
2040 * currently using the dentry or the inode
2041 * we can't do that and we fall back on removing
2042 * it from the hash queues and waiting for
2043 * it to be deleted later when it has no users
2047 * d_delete - delete a dentry
2048 * @dentry: The dentry to delete
2050 * Turn the dentry into a negative dentry if possible, otherwise
2051 * remove it from the hash queues so it can be deleted later
2054 void d_delete(struct dentry * dentry)
2056 struct inode *inode;
2059 * Are we the only user?
2062 spin_lock(&dentry->d_lock);
2063 inode = dentry->d_inode;
2064 isdir = S_ISDIR(inode->i_mode);
2065 if (dentry->d_count == 1) {
2066 if (inode && !spin_trylock(&inode->i_lock)) {
2067 spin_unlock(&dentry->d_lock);
2071 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2072 dentry_unlink_inode(dentry);
2073 fsnotify_nameremove(dentry, isdir);
2077 if (!d_unhashed(dentry))
2080 spin_unlock(&dentry->d_lock);
2082 fsnotify_nameremove(dentry, isdir);
2084 EXPORT_SYMBOL(d_delete);
2086 static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b)
2088 BUG_ON(!d_unhashed(entry));
2089 spin_lock_bucket(b);
2090 entry->d_flags &= ~DCACHE_UNHASHED;
2091 hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
2092 spin_unlock_bucket(b);
2095 static void _d_rehash(struct dentry * entry)
2097 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2101 * d_rehash - add an entry back to the hash
2102 * @entry: dentry to add to the hash
2104 * Adds a dentry to the hash according to its name.
2107 void d_rehash(struct dentry * entry)
2109 spin_lock(&entry->d_lock);
2111 spin_unlock(&entry->d_lock);
2113 EXPORT_SYMBOL(d_rehash);
2116 * dentry_update_name_case - update case insensitive dentry with a new name
2117 * @dentry: dentry to be updated
2120 * Update a case insensitive dentry with new case of name.
2122 * dentry must have been returned by d_lookup with name @name. Old and new
2123 * name lengths must match (ie. no d_compare which allows mismatched name
2126 * Parent inode i_mutex must be held over d_lookup and into this call (to
2127 * keep renames and concurrent inserts, and readdir(2) away).
2129 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2131 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2132 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2134 spin_lock(&dentry->d_lock);
2135 write_seqcount_begin(&dentry->d_seq);
2136 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2137 write_seqcount_end(&dentry->d_seq);
2138 spin_unlock(&dentry->d_lock);
2140 EXPORT_SYMBOL(dentry_update_name_case);
2142 static void switch_names(struct dentry *dentry, struct dentry *target)
2144 if (dname_external(target)) {
2145 if (dname_external(dentry)) {
2147 * Both external: swap the pointers
2149 swap(target->d_name.name, dentry->d_name.name);
2152 * dentry:internal, target:external. Steal target's
2153 * storage and make target internal.
2155 memcpy(target->d_iname, dentry->d_name.name,
2156 dentry->d_name.len + 1);
2157 dentry->d_name.name = target->d_name.name;
2158 target->d_name.name = target->d_iname;
2161 if (dname_external(dentry)) {
2163 * dentry:external, target:internal. Give dentry's
2164 * storage to target and make dentry internal
2166 memcpy(dentry->d_iname, target->d_name.name,
2167 target->d_name.len + 1);
2168 target->d_name.name = dentry->d_name.name;
2169 dentry->d_name.name = dentry->d_iname;
2172 * Both are internal. Just copy target to dentry
2174 memcpy(dentry->d_iname, target->d_name.name,
2175 target->d_name.len + 1);
2176 dentry->d_name.len = target->d_name.len;
2180 swap(dentry->d_name.len, target->d_name.len);
2183 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2186 * XXXX: do we really need to take target->d_lock?
2188 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2189 spin_lock(&target->d_parent->d_lock);
2191 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2192 spin_lock(&dentry->d_parent->d_lock);
2193 spin_lock_nested(&target->d_parent->d_lock,
2194 DENTRY_D_LOCK_NESTED);
2196 spin_lock(&target->d_parent->d_lock);
2197 spin_lock_nested(&dentry->d_parent->d_lock,
2198 DENTRY_D_LOCK_NESTED);
2201 if (target < dentry) {
2202 spin_lock_nested(&target->d_lock, 2);
2203 spin_lock_nested(&dentry->d_lock, 3);
2205 spin_lock_nested(&dentry->d_lock, 2);
2206 spin_lock_nested(&target->d_lock, 3);
2210 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2211 struct dentry *target)
2213 if (target->d_parent != dentry->d_parent)
2214 spin_unlock(&dentry->d_parent->d_lock);
2215 if (target->d_parent != target)
2216 spin_unlock(&target->d_parent->d_lock);
2220 * When switching names, the actual string doesn't strictly have to
2221 * be preserved in the target - because we're dropping the target
2222 * anyway. As such, we can just do a simple memcpy() to copy over
2223 * the new name before we switch.
2225 * Note that we have to be a lot more careful about getting the hash
2226 * switched - we have to switch the hash value properly even if it
2227 * then no longer matches the actual (corrupted) string of the target.
2228 * The hash value has to match the hash queue that the dentry is on..
2231 * d_move - move a dentry
2232 * @dentry: entry to move
2233 * @target: new dentry
2235 * Update the dcache to reflect the move of a file name. Negative
2236 * dcache entries should not be moved in this way.
2238 void d_move(struct dentry * dentry, struct dentry * target)
2240 if (!dentry->d_inode)
2241 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2243 BUG_ON(d_ancestor(dentry, target));
2244 BUG_ON(d_ancestor(target, dentry));
2246 write_seqlock(&rename_lock);
2248 dentry_lock_for_move(dentry, target);
2250 write_seqcount_begin(&dentry->d_seq);
2251 write_seqcount_begin(&target->d_seq);
2253 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2256 * Move the dentry to the target hash queue. Don't bother checking
2257 * for the same hash queue because of how unlikely it is.
2260 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2262 /* Unhash the target: dput() will then get rid of it */
2265 list_del(&dentry->d_u.d_child);
2266 list_del(&target->d_u.d_child);
2268 /* Switch the names.. */
2269 switch_names(dentry, target);
2270 swap(dentry->d_name.hash, target->d_name.hash);
2272 /* ... and switch the parents */
2273 if (IS_ROOT(dentry)) {
2274 dentry->d_parent = target->d_parent;
2275 target->d_parent = target;
2276 INIT_LIST_HEAD(&target->d_u.d_child);
2278 swap(dentry->d_parent, target->d_parent);
2280 /* And add them back to the (new) parent lists */
2281 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2284 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2286 write_seqcount_end(&target->d_seq);
2287 write_seqcount_end(&dentry->d_seq);
2289 dentry_unlock_parents_for_move(dentry, target);
2290 spin_unlock(&target->d_lock);
2291 fsnotify_d_move(dentry);
2292 spin_unlock(&dentry->d_lock);
2293 write_sequnlock(&rename_lock);
2295 EXPORT_SYMBOL(d_move);
2298 * d_ancestor - search for an ancestor
2299 * @p1: ancestor dentry
2302 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2303 * an ancestor of p2, else NULL.
2305 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2309 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2310 if (p->d_parent == p1)
2317 * This helper attempts to cope with remotely renamed directories
2319 * It assumes that the caller is already holding
2320 * dentry->d_parent->d_inode->i_mutex and the inode->i_lock
2322 * Note: If ever the locking in lock_rename() changes, then please
2323 * remember to update this too...
2325 static struct dentry *__d_unalias(struct inode *inode,
2326 struct dentry *dentry, struct dentry *alias)
2328 struct mutex *m1 = NULL, *m2 = NULL;
2331 /* If alias and dentry share a parent, then no extra locks required */
2332 if (alias->d_parent == dentry->d_parent)
2335 /* Check for loops */
2336 ret = ERR_PTR(-ELOOP);
2337 if (d_ancestor(alias, dentry))
2340 /* See lock_rename() */
2341 ret = ERR_PTR(-EBUSY);
2342 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2344 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2345 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2347 m2 = &alias->d_parent->d_inode->i_mutex;
2349 d_move(alias, dentry);
2352 spin_unlock(&inode->i_lock);
2361 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2362 * named dentry in place of the dentry to be replaced.
2363 * returns with anon->d_lock held!
2365 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2367 struct dentry *dparent, *aparent;
2369 dentry_lock_for_move(anon, dentry);
2371 write_seqcount_begin(&dentry->d_seq);
2372 write_seqcount_begin(&anon->d_seq);
2374 dparent = dentry->d_parent;
2375 aparent = anon->d_parent;
2377 switch_names(dentry, anon);
2378 swap(dentry->d_name.hash, anon->d_name.hash);
2380 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2381 list_del(&dentry->d_u.d_child);
2382 if (!IS_ROOT(dentry))
2383 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2385 INIT_LIST_HEAD(&dentry->d_u.d_child);
2387 anon->d_parent = (dparent == dentry) ? anon : dparent;
2388 list_del(&anon->d_u.d_child);
2390 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2392 INIT_LIST_HEAD(&anon->d_u.d_child);
2394 write_seqcount_end(&dentry->d_seq);
2395 write_seqcount_end(&anon->d_seq);
2397 dentry_unlock_parents_for_move(anon, dentry);
2398 spin_unlock(&dentry->d_lock);
2400 /* anon->d_lock still locked, returns locked */
2401 anon->d_flags &= ~DCACHE_DISCONNECTED;
2405 * d_materialise_unique - introduce an inode into the tree
2406 * @dentry: candidate dentry
2407 * @inode: inode to bind to the dentry, to which aliases may be attached
2409 * Introduces an dentry into the tree, substituting an extant disconnected
2410 * root directory alias in its place if there is one
2412 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2414 struct dentry *actual;
2416 BUG_ON(!d_unhashed(dentry));
2420 __d_instantiate(dentry, NULL);
2425 spin_lock(&inode->i_lock);
2427 if (S_ISDIR(inode->i_mode)) {
2428 struct dentry *alias;
2430 /* Does an aliased dentry already exist? */
2431 alias = __d_find_alias(inode, 0);
2434 /* Is this an anonymous mountpoint that we could splice
2436 if (IS_ROOT(alias)) {
2437 __d_materialise_dentry(dentry, alias);
2441 /* Nope, but we must(!) avoid directory aliasing */
2442 actual = __d_unalias(inode, dentry, alias);
2449 /* Add a unique reference */
2450 actual = __d_instantiate_unique(dentry, inode);
2454 BUG_ON(!d_unhashed(actual));
2456 spin_lock(&actual->d_lock);
2459 spin_unlock(&actual->d_lock);
2460 spin_unlock(&inode->i_lock);
2462 if (actual == dentry) {
2463 security_d_instantiate(dentry, inode);
2470 EXPORT_SYMBOL_GPL(d_materialise_unique);
2472 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2476 return -ENAMETOOLONG;
2478 memcpy(*buffer, str, namelen);
2482 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2484 return prepend(buffer, buflen, name->name, name->len);
2488 * prepend_path - Prepend path string to a buffer
2489 * @path: the dentry/vfsmount to report
2490 * @root: root vfsmnt/dentry (may be modified by this function)
2491 * @buffer: pointer to the end of the buffer
2492 * @buflen: pointer to buffer length
2494 * Caller holds the rename_lock.
2496 * If path is not reachable from the supplied root, then the value of
2497 * root is changed (without modifying refcounts).
2499 static int prepend_path(const struct path *path, struct path *root,
2500 char **buffer, int *buflen)
2502 struct dentry *dentry = path->dentry;
2503 struct vfsmount *vfsmnt = path->mnt;
2507 br_read_lock(vfsmount_lock);
2508 while (dentry != root->dentry || vfsmnt != root->mnt) {
2509 struct dentry * parent;
2511 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2513 if (vfsmnt->mnt_parent == vfsmnt) {
2516 dentry = vfsmnt->mnt_mountpoint;
2517 vfsmnt = vfsmnt->mnt_parent;
2520 parent = dentry->d_parent;
2522 spin_lock(&dentry->d_lock);
2523 error = prepend_name(buffer, buflen, &dentry->d_name);
2524 spin_unlock(&dentry->d_lock);
2526 error = prepend(buffer, buflen, "/", 1);
2535 if (!error && !slash)
2536 error = prepend(buffer, buflen, "/", 1);
2538 br_read_unlock(vfsmount_lock);
2543 * Filesystems needing to implement special "root names"
2544 * should do so with ->d_dname()
2546 if (IS_ROOT(dentry) &&
2547 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2548 WARN(1, "Root dentry has weird name <%.*s>\n",
2549 (int) dentry->d_name.len, dentry->d_name.name);
2552 root->dentry = dentry;
2557 * __d_path - return the path of a dentry
2558 * @path: the dentry/vfsmount to report
2559 * @root: root vfsmnt/dentry (may be modified by this function)
2560 * @buf: buffer to return value in
2561 * @buflen: buffer length
2563 * Convert a dentry into an ASCII path name.
2565 * Returns a pointer into the buffer or an error code if the
2566 * path was too long.
2568 * "buflen" should be positive.
2570 * If path is not reachable from the supplied root, then the value of
2571 * root is changed (without modifying refcounts).
2573 char *__d_path(const struct path *path, struct path *root,
2574 char *buf, int buflen)
2576 char *res = buf + buflen;
2579 prepend(&res, &buflen, "\0", 1);
2580 write_seqlock(&rename_lock);
2581 error = prepend_path(path, root, &res, &buflen);
2582 write_sequnlock(&rename_lock);
2585 return ERR_PTR(error);
2590 * same as __d_path but appends "(deleted)" for unlinked files.
2592 static int path_with_deleted(const struct path *path, struct path *root,
2593 char **buf, int *buflen)
2595 prepend(buf, buflen, "\0", 1);
2596 if (d_unlinked(path->dentry)) {
2597 int error = prepend(buf, buflen, " (deleted)", 10);
2602 return prepend_path(path, root, buf, buflen);
2605 static int prepend_unreachable(char **buffer, int *buflen)
2607 return prepend(buffer, buflen, "(unreachable)", 13);
2611 * d_path - return the path of a dentry
2612 * @path: path to report
2613 * @buf: buffer to return value in
2614 * @buflen: buffer length
2616 * Convert a dentry into an ASCII path name. If the entry has been deleted
2617 * the string " (deleted)" is appended. Note that this is ambiguous.
2619 * Returns a pointer into the buffer or an error code if the path was
2620 * too long. Note: Callers should use the returned pointer, not the passed
2621 * in buffer, to use the name! The implementation often starts at an offset
2622 * into the buffer, and may leave 0 bytes at the start.
2624 * "buflen" should be positive.
2626 char *d_path(const struct path *path, char *buf, int buflen)
2628 char *res = buf + buflen;
2634 * We have various synthetic filesystems that never get mounted. On
2635 * these filesystems dentries are never used for lookup purposes, and
2636 * thus don't need to be hashed. They also don't need a name until a
2637 * user wants to identify the object in /proc/pid/fd/. The little hack
2638 * below allows us to generate a name for these objects on demand:
2640 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2641 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2643 get_fs_root(current->fs, &root);
2644 write_seqlock(&rename_lock);
2646 error = path_with_deleted(path, &tmp, &res, &buflen);
2648 res = ERR_PTR(error);
2649 write_sequnlock(&rename_lock);
2653 EXPORT_SYMBOL(d_path);
2656 * d_path_with_unreachable - return the path of a dentry
2657 * @path: path to report
2658 * @buf: buffer to return value in
2659 * @buflen: buffer length
2661 * The difference from d_path() is that this prepends "(unreachable)"
2662 * to paths which are unreachable from the current process' root.
2664 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2666 char *res = buf + buflen;
2671 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2672 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2674 get_fs_root(current->fs, &root);
2675 write_seqlock(&rename_lock);
2677 error = path_with_deleted(path, &tmp, &res, &buflen);
2678 if (!error && !path_equal(&tmp, &root))
2679 error = prepend_unreachable(&res, &buflen);
2680 write_sequnlock(&rename_lock);
2683 res = ERR_PTR(error);
2689 * Helper function for dentry_operations.d_dname() members
2691 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2692 const char *fmt, ...)
2698 va_start(args, fmt);
2699 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2702 if (sz > sizeof(temp) || sz > buflen)
2703 return ERR_PTR(-ENAMETOOLONG);
2705 buffer += buflen - sz;
2706 return memcpy(buffer, temp, sz);
2710 * Write full pathname from the root of the filesystem into the buffer.
2712 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2714 char *end = buf + buflen;
2717 prepend(&end, &buflen, "\0", 1);
2724 while (!IS_ROOT(dentry)) {
2725 struct dentry *parent = dentry->d_parent;
2729 spin_lock(&dentry->d_lock);
2730 error = prepend_name(&end, &buflen, &dentry->d_name);
2731 spin_unlock(&dentry->d_lock);
2732 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2740 return ERR_PTR(-ENAMETOOLONG);
2743 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2747 write_seqlock(&rename_lock);
2748 retval = __dentry_path(dentry, buf, buflen);
2749 write_sequnlock(&rename_lock);
2753 EXPORT_SYMBOL(dentry_path_raw);
2755 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2760 write_seqlock(&rename_lock);
2761 if (d_unlinked(dentry)) {
2763 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2767 retval = __dentry_path(dentry, buf, buflen);
2768 write_sequnlock(&rename_lock);
2769 if (!IS_ERR(retval) && p)
2770 *p = '/'; /* restore '/' overriden with '\0' */
2773 return ERR_PTR(-ENAMETOOLONG);
2777 * NOTE! The user-level library version returns a
2778 * character pointer. The kernel system call just
2779 * returns the length of the buffer filled (which
2780 * includes the ending '\0' character), or a negative
2781 * error value. So libc would do something like
2783 * char *getcwd(char * buf, size_t size)
2787 * retval = sys_getcwd(buf, size);
2794 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2797 struct path pwd, root;
2798 char *page = (char *) __get_free_page(GFP_USER);
2803 get_fs_root_and_pwd(current->fs, &root, &pwd);
2806 write_seqlock(&rename_lock);
2807 if (!d_unlinked(pwd.dentry)) {
2809 struct path tmp = root;
2810 char *cwd = page + PAGE_SIZE;
2811 int buflen = PAGE_SIZE;
2813 prepend(&cwd, &buflen, "\0", 1);
2814 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2815 write_sequnlock(&rename_lock);
2820 /* Unreachable from current root */
2821 if (!path_equal(&tmp, &root)) {
2822 error = prepend_unreachable(&cwd, &buflen);
2828 len = PAGE_SIZE + page - cwd;
2831 if (copy_to_user(buf, cwd, len))
2835 write_sequnlock(&rename_lock);
2841 free_page((unsigned long) page);
2846 * Test whether new_dentry is a subdirectory of old_dentry.
2848 * Trivially implemented using the dcache structure
2852 * is_subdir - is new dentry a subdirectory of old_dentry
2853 * @new_dentry: new dentry
2854 * @old_dentry: old dentry
2856 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2857 * Returns 0 otherwise.
2858 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2861 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2866 if (new_dentry == old_dentry)
2870 /* for restarting inner loop in case of seq retry */
2871 seq = read_seqbegin(&rename_lock);
2873 * Need rcu_readlock to protect against the d_parent trashing
2877 if (d_ancestor(old_dentry, new_dentry))
2882 } while (read_seqretry(&rename_lock, seq));
2887 int path_is_under(struct path *path1, struct path *path2)
2889 struct vfsmount *mnt = path1->mnt;
2890 struct dentry *dentry = path1->dentry;
2893 br_read_lock(vfsmount_lock);
2894 if (mnt != path2->mnt) {
2896 if (mnt->mnt_parent == mnt) {
2897 br_read_unlock(vfsmount_lock);
2900 if (mnt->mnt_parent == path2->mnt)
2902 mnt = mnt->mnt_parent;
2904 dentry = mnt->mnt_mountpoint;
2906 res = is_subdir(dentry, path2->dentry);
2907 br_read_unlock(vfsmount_lock);
2910 EXPORT_SYMBOL(path_is_under);
2912 void d_genocide(struct dentry *root)
2914 struct dentry *this_parent;
2915 struct list_head *next;
2919 seq = read_seqbegin(&rename_lock);
2922 spin_lock(&this_parent->d_lock);
2924 next = this_parent->d_subdirs.next;
2926 while (next != &this_parent->d_subdirs) {
2927 struct list_head *tmp = next;
2928 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2931 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2932 if (d_unhashed(dentry) || !dentry->d_inode) {
2933 spin_unlock(&dentry->d_lock);
2936 if (!list_empty(&dentry->d_subdirs)) {
2937 spin_unlock(&this_parent->d_lock);
2938 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2939 this_parent = dentry;
2940 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2943 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2944 dentry->d_flags |= DCACHE_GENOCIDE;
2947 spin_unlock(&dentry->d_lock);
2949 if (this_parent != root) {
2950 struct dentry *child = this_parent;
2951 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2952 this_parent->d_flags |= DCACHE_GENOCIDE;
2953 this_parent->d_count--;
2955 this_parent = try_to_ascend(this_parent, locked, seq);
2958 next = child->d_u.d_child.next;
2961 spin_unlock(&this_parent->d_lock);
2962 if (!locked && read_seqretry(&rename_lock, seq))
2965 write_sequnlock(&rename_lock);
2970 write_seqlock(&rename_lock);
2975 * find_inode_number - check for dentry with name
2976 * @dir: directory to check
2977 * @name: Name to find.
2979 * Check whether a dentry already exists for the given name,
2980 * and return the inode number if it has an inode. Otherwise
2983 * This routine is used to post-process directory listings for
2984 * filesystems using synthetic inode numbers, and is necessary
2985 * to keep getcwd() working.
2988 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2990 struct dentry * dentry;
2993 dentry = d_hash_and_lookup(dir, name);
2995 if (dentry->d_inode)
2996 ino = dentry->d_inode->i_ino;
3001 EXPORT_SYMBOL(find_inode_number);
3003 static __initdata unsigned long dhash_entries;
3004 static int __init set_dhash_entries(char *str)
3008 dhash_entries = simple_strtoul(str, &str, 0);
3011 __setup("dhash_entries=", set_dhash_entries);
3013 static void __init dcache_init_early(void)
3017 /* If hashes are distributed across NUMA nodes, defer
3018 * hash allocation until vmalloc space is available.
3024 alloc_large_system_hash("Dentry cache",
3025 sizeof(struct dcache_hash_bucket),
3033 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3034 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3037 static void __init dcache_init(void)
3042 * A constructor could be added for stable state like the lists,
3043 * but it is probably not worth it because of the cache nature
3046 dentry_cache = KMEM_CACHE(dentry,
3047 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3049 register_shrinker(&dcache_shrinker);
3051 /* Hash may have been set up in dcache_init_early */
3056 alloc_large_system_hash("Dentry cache",
3057 sizeof(struct dcache_hash_bucket),
3065 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3066 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3069 /* SLAB cache for __getname() consumers */
3070 struct kmem_cache *names_cachep __read_mostly;
3071 EXPORT_SYMBOL(names_cachep);
3073 EXPORT_SYMBOL(d_genocide);
3075 void __init vfs_caches_init_early(void)
3077 dcache_init_early();
3081 void __init vfs_caches_init(unsigned long mempages)
3083 unsigned long reserve;
3085 /* Base hash sizes on available memory, with a reserve equal to
3086 150% of current kernel size */
3088 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3089 mempages -= reserve;
3091 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3092 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3096 files_init(mempages);