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 * After this call, in-progress rcu-walk path lookup will fail. This
180 * should be called after unhashing, and after changing d_inode (if
181 * the dentry has not already been unhashed).
183 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
185 assert_spin_locked(&dentry->d_lock);
186 /* Go through a barrier */
187 write_seqcount_barrier(&dentry->d_seq);
191 * Release the dentry's inode, using the filesystem
192 * d_iput() operation if defined. Dentry has no refcount
195 static void dentry_iput(struct dentry * dentry)
196 __releases(dentry->d_lock)
197 __releases(dentry->d_inode->i_lock)
199 struct inode *inode = dentry->d_inode;
201 dentry->d_inode = NULL;
202 list_del_init(&dentry->d_alias);
203 spin_unlock(&dentry->d_lock);
204 spin_unlock(&inode->i_lock);
206 fsnotify_inoderemove(inode);
207 if (dentry->d_op && dentry->d_op->d_iput)
208 dentry->d_op->d_iput(dentry, inode);
212 spin_unlock(&dentry->d_lock);
217 * Release the dentry's inode, using the filesystem
218 * d_iput() operation if defined. dentry remains in-use.
220 static void dentry_unlink_inode(struct dentry * dentry)
221 __releases(dentry->d_lock)
222 __releases(dentry->d_inode->i_lock)
224 struct inode *inode = dentry->d_inode;
225 dentry->d_inode = NULL;
226 list_del_init(&dentry->d_alias);
227 dentry_rcuwalk_barrier(dentry);
228 spin_unlock(&dentry->d_lock);
229 spin_unlock(&inode->i_lock);
231 fsnotify_inoderemove(inode);
232 if (dentry->d_op && dentry->d_op->d_iput)
233 dentry->d_op->d_iput(dentry, inode);
239 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
241 static void dentry_lru_add(struct dentry *dentry)
243 if (list_empty(&dentry->d_lru)) {
244 spin_lock(&dcache_lru_lock);
245 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
246 dentry->d_sb->s_nr_dentry_unused++;
247 dentry_stat.nr_unused++;
248 spin_unlock(&dcache_lru_lock);
252 static void __dentry_lru_del(struct dentry *dentry)
254 list_del_init(&dentry->d_lru);
255 dentry->d_sb->s_nr_dentry_unused--;
256 dentry_stat.nr_unused--;
259 static void dentry_lru_del(struct dentry *dentry)
261 if (!list_empty(&dentry->d_lru)) {
262 spin_lock(&dcache_lru_lock);
263 __dentry_lru_del(dentry);
264 spin_unlock(&dcache_lru_lock);
268 static void dentry_lru_move_tail(struct dentry *dentry)
270 spin_lock(&dcache_lru_lock);
271 if (list_empty(&dentry->d_lru)) {
272 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
273 dentry->d_sb->s_nr_dentry_unused++;
274 dentry_stat.nr_unused++;
276 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
278 spin_unlock(&dcache_lru_lock);
282 * d_kill - kill dentry and return parent
283 * @dentry: dentry to kill
285 * The dentry must already be unhashed and removed from the LRU.
287 * If this is the root of the dentry tree, return NULL.
289 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
292 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
293 __releases(dentry->d_lock)
294 __releases(parent->d_lock)
295 __releases(dentry->d_inode->i_lock)
297 dentry->d_parent = NULL;
298 list_del(&dentry->d_u.d_child);
300 spin_unlock(&parent->d_lock);
303 * dentry_iput drops the locks, at which point nobody (except
304 * transient RCU lookups) can reach this dentry.
311 * d_drop - drop a dentry
312 * @dentry: dentry to drop
314 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
315 * be found through a VFS lookup any more. Note that this is different from
316 * deleting the dentry - d_delete will try to mark the dentry negative if
317 * possible, giving a successful _negative_ lookup, while d_drop will
318 * just make the cache lookup fail.
320 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
321 * reason (NFS timeouts or autofs deletes).
323 * __d_drop requires dentry->d_lock.
325 void __d_drop(struct dentry *dentry)
327 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
328 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) {
330 (unsigned long *)&dentry->d_sb->s_anon.first);
331 dentry->d_flags |= DCACHE_UNHASHED;
332 hlist_bl_del_init(&dentry->d_hash);
334 (unsigned long *)&dentry->d_sb->s_anon.first);
336 struct dcache_hash_bucket *b;
337 b = d_hash(dentry->d_parent, dentry->d_name.hash);
340 * We may not actually need to put DCACHE_UNHASHED
341 * manipulations under the hash lock, but follow
342 * the principle of least surprise.
344 dentry->d_flags |= DCACHE_UNHASHED;
345 hlist_bl_del_rcu(&dentry->d_hash);
346 spin_unlock_bucket(b);
347 dentry_rcuwalk_barrier(dentry);
351 EXPORT_SYMBOL(__d_drop);
353 void d_drop(struct dentry *dentry)
355 spin_lock(&dentry->d_lock);
357 spin_unlock(&dentry->d_lock);
359 EXPORT_SYMBOL(d_drop);
362 * Finish off a dentry we've decided to kill.
363 * dentry->d_lock must be held, returns with it unlocked.
364 * If ref is non-zero, then decrement the refcount too.
365 * Returns dentry requiring refcount drop, or NULL if we're done.
367 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
368 __releases(dentry->d_lock)
371 struct dentry *parent;
373 inode = dentry->d_inode;
374 if (inode && !spin_trylock(&inode->i_lock)) {
376 spin_unlock(&dentry->d_lock);
378 return dentry; /* try again with same dentry */
383 parent = dentry->d_parent;
384 if (parent && !spin_trylock(&parent->d_lock)) {
386 spin_unlock(&inode->i_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(&inode->i_lock);
622 de = __d_find_alias(inode, 0);
623 spin_unlock(&inode->i_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(&inode->i_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(&inode->i_lock);
648 spin_unlock(&dentry->d_lock);
650 spin_unlock(&inode->i_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 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1324 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1325 spin_unlock(&parent->d_lock);
1328 this_cpu_inc(nr_dentry);
1332 EXPORT_SYMBOL(d_alloc);
1334 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1336 struct dentry *dentry = d_alloc(NULL, name);
1339 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1340 dentry->d_parent = dentry;
1341 dentry->d_flags |= DCACHE_DISCONNECTED;
1345 EXPORT_SYMBOL(d_alloc_pseudo);
1347 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1352 q.len = strlen(name);
1353 q.hash = full_name_hash(q.name, q.len);
1354 return d_alloc(parent, &q);
1356 EXPORT_SYMBOL(d_alloc_name);
1358 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1360 WARN_ON_ONCE(dentry->d_op);
1361 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1363 DCACHE_OP_REVALIDATE |
1364 DCACHE_OP_DELETE ));
1369 dentry->d_flags |= DCACHE_OP_HASH;
1371 dentry->d_flags |= DCACHE_OP_COMPARE;
1372 if (op->d_revalidate)
1373 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1375 dentry->d_flags |= DCACHE_OP_DELETE;
1378 EXPORT_SYMBOL(d_set_d_op);
1380 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1382 spin_lock(&dentry->d_lock);
1384 if (unlikely(IS_AUTOMOUNT(inode)))
1385 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1386 list_add(&dentry->d_alias, &inode->i_dentry);
1388 dentry->d_inode = inode;
1389 dentry_rcuwalk_barrier(dentry);
1390 spin_unlock(&dentry->d_lock);
1391 fsnotify_d_instantiate(dentry, inode);
1395 * d_instantiate - fill in inode information for a dentry
1396 * @entry: dentry to complete
1397 * @inode: inode to attach to this dentry
1399 * Fill in inode information in the entry.
1401 * This turns negative dentries into productive full members
1404 * NOTE! This assumes that the inode count has been incremented
1405 * (or otherwise set) by the caller to indicate that it is now
1406 * in use by the dcache.
1409 void d_instantiate(struct dentry *entry, struct inode * inode)
1411 BUG_ON(!list_empty(&entry->d_alias));
1413 spin_lock(&inode->i_lock);
1414 __d_instantiate(entry, inode);
1416 spin_unlock(&inode->i_lock);
1417 security_d_instantiate(entry, inode);
1419 EXPORT_SYMBOL(d_instantiate);
1422 * d_instantiate_unique - instantiate a non-aliased dentry
1423 * @entry: dentry to instantiate
1424 * @inode: inode to attach to this dentry
1426 * Fill in inode information in the entry. On success, it returns NULL.
1427 * If an unhashed alias of "entry" already exists, then we return the
1428 * aliased dentry instead and drop one reference to inode.
1430 * Note that in order to avoid conflicts with rename() etc, the caller
1431 * had better be holding the parent directory semaphore.
1433 * This also assumes that the inode count has been incremented
1434 * (or otherwise set) by the caller to indicate that it is now
1435 * in use by the dcache.
1437 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1438 struct inode *inode)
1440 struct dentry *alias;
1441 int len = entry->d_name.len;
1442 const char *name = entry->d_name.name;
1443 unsigned int hash = entry->d_name.hash;
1446 __d_instantiate(entry, NULL);
1450 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1451 struct qstr *qstr = &alias->d_name;
1454 * Don't need alias->d_lock here, because aliases with
1455 * d_parent == entry->d_parent are not subject to name or
1456 * parent changes, because the parent inode i_mutex is held.
1458 if (qstr->hash != hash)
1460 if (alias->d_parent != entry->d_parent)
1462 if (dentry_cmp(qstr->name, qstr->len, name, len))
1468 __d_instantiate(entry, inode);
1472 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1474 struct dentry *result;
1476 BUG_ON(!list_empty(&entry->d_alias));
1479 spin_lock(&inode->i_lock);
1480 result = __d_instantiate_unique(entry, inode);
1482 spin_unlock(&inode->i_lock);
1485 security_d_instantiate(entry, inode);
1489 BUG_ON(!d_unhashed(result));
1494 EXPORT_SYMBOL(d_instantiate_unique);
1497 * d_alloc_root - allocate root dentry
1498 * @root_inode: inode to allocate the root for
1500 * Allocate a root ("/") dentry for the inode given. The inode is
1501 * instantiated and returned. %NULL is returned if there is insufficient
1502 * memory or the inode passed is %NULL.
1505 struct dentry * d_alloc_root(struct inode * root_inode)
1507 struct dentry *res = NULL;
1510 static const struct qstr name = { .name = "/", .len = 1 };
1512 res = d_alloc(NULL, &name);
1514 res->d_sb = root_inode->i_sb;
1515 d_set_d_op(res, res->d_sb->s_d_op);
1516 res->d_parent = res;
1517 d_instantiate(res, root_inode);
1522 EXPORT_SYMBOL(d_alloc_root);
1525 * d_obtain_alias - find or allocate a dentry for a given inode
1526 * @inode: inode to allocate the dentry for
1528 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1529 * similar open by handle operations. The returned dentry may be anonymous,
1530 * or may have a full name (if the inode was already in the cache).
1532 * When called on a directory inode, we must ensure that the inode only ever
1533 * has one dentry. If a dentry is found, that is returned instead of
1534 * allocating a new one.
1536 * On successful return, the reference to the inode has been transferred
1537 * to the dentry. In case of an error the reference on the inode is released.
1538 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1539 * be passed in and will be the error will be propagate to the return value,
1540 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1542 struct dentry *d_obtain_alias(struct inode *inode)
1544 static const struct qstr anonstring = { .name = "" };
1549 return ERR_PTR(-ESTALE);
1551 return ERR_CAST(inode);
1553 res = d_find_alias(inode);
1557 tmp = d_alloc(NULL, &anonstring);
1559 res = ERR_PTR(-ENOMEM);
1562 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1565 spin_lock(&inode->i_lock);
1566 res = __d_find_alias(inode, 0);
1568 spin_unlock(&inode->i_lock);
1573 /* attach a disconnected dentry */
1574 spin_lock(&tmp->d_lock);
1575 tmp->d_sb = inode->i_sb;
1576 d_set_d_op(tmp, tmp->d_sb->s_d_op);
1577 tmp->d_inode = inode;
1578 tmp->d_flags |= DCACHE_DISCONNECTED;
1579 list_add(&tmp->d_alias, &inode->i_dentry);
1580 bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1581 tmp->d_flags &= ~DCACHE_UNHASHED;
1582 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1583 __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1584 spin_unlock(&tmp->d_lock);
1585 spin_unlock(&inode->i_lock);
1593 EXPORT_SYMBOL(d_obtain_alias);
1596 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1597 * @inode: the inode which may have a disconnected dentry
1598 * @dentry: a negative dentry which we want to point to the inode.
1600 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1601 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1602 * and return it, else simply d_add the inode to the dentry and return NULL.
1604 * This is needed in the lookup routine of any filesystem that is exportable
1605 * (via knfsd) so that we can build dcache paths to directories effectively.
1607 * If a dentry was found and moved, then it is returned. Otherwise NULL
1608 * is returned. This matches the expected return value of ->lookup.
1611 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1613 struct dentry *new = NULL;
1615 if (inode && S_ISDIR(inode->i_mode)) {
1616 spin_lock(&inode->i_lock);
1617 new = __d_find_alias(inode, 1);
1619 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1620 spin_unlock(&inode->i_lock);
1621 security_d_instantiate(new, inode);
1622 d_move(new, dentry);
1625 /* already taking inode->i_lock, so d_add() by hand */
1626 __d_instantiate(dentry, inode);
1627 spin_unlock(&inode->i_lock);
1628 security_d_instantiate(dentry, inode);
1632 d_add(dentry, inode);
1635 EXPORT_SYMBOL(d_splice_alias);
1638 * d_add_ci - lookup or allocate new dentry with case-exact name
1639 * @inode: the inode case-insensitive lookup has found
1640 * @dentry: the negative dentry that was passed to the parent's lookup func
1641 * @name: the case-exact name to be associated with the returned dentry
1643 * This is to avoid filling the dcache with case-insensitive names to the
1644 * same inode, only the actual correct case is stored in the dcache for
1645 * case-insensitive filesystems.
1647 * For a case-insensitive lookup match and if the the case-exact dentry
1648 * already exists in in the dcache, use it and return it.
1650 * If no entry exists with the exact case name, allocate new dentry with
1651 * the exact case, and return the spliced entry.
1653 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1657 struct dentry *found;
1661 * First check if a dentry matching the name already exists,
1662 * if not go ahead and create it now.
1664 found = d_hash_and_lookup(dentry->d_parent, name);
1666 new = d_alloc(dentry->d_parent, name);
1672 found = d_splice_alias(inode, new);
1681 * If a matching dentry exists, and it's not negative use it.
1683 * Decrement the reference count to balance the iget() done
1686 if (found->d_inode) {
1687 if (unlikely(found->d_inode != inode)) {
1688 /* This can't happen because bad inodes are unhashed. */
1689 BUG_ON(!is_bad_inode(inode));
1690 BUG_ON(!is_bad_inode(found->d_inode));
1697 * Negative dentry: instantiate it unless the inode is a directory and
1698 * already has a dentry.
1700 spin_lock(&inode->i_lock);
1701 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1702 __d_instantiate(found, inode);
1703 spin_unlock(&inode->i_lock);
1704 security_d_instantiate(found, inode);
1709 * In case a directory already has a (disconnected) entry grab a
1710 * reference to it, move it in place and use it.
1712 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1714 spin_unlock(&inode->i_lock);
1715 security_d_instantiate(found, inode);
1723 return ERR_PTR(error);
1725 EXPORT_SYMBOL(d_add_ci);
1728 * __d_lookup_rcu - search for a dentry (racy, store-free)
1729 * @parent: parent dentry
1730 * @name: qstr of name we wish to find
1731 * @seq: returns d_seq value at the point where the dentry was found
1732 * @inode: returns dentry->d_inode when the inode was found valid.
1733 * Returns: dentry, or NULL
1735 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1736 * resolution (store-free path walking) design described in
1737 * Documentation/filesystems/path-lookup.txt.
1739 * This is not to be used outside core vfs.
1741 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1742 * held, and rcu_read_lock held. The returned dentry must not be stored into
1743 * without taking d_lock and checking d_seq sequence count against @seq
1746 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1749 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1750 * the returned dentry, so long as its parent's seqlock is checked after the
1751 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1752 * is formed, giving integrity down the path walk.
1754 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1755 unsigned *seq, struct inode **inode)
1757 unsigned int len = name->len;
1758 unsigned int hash = name->hash;
1759 const unsigned char *str = name->name;
1760 struct dcache_hash_bucket *b = d_hash(parent, hash);
1761 struct hlist_bl_node *node;
1762 struct dentry *dentry;
1765 * Note: There is significant duplication with __d_lookup_rcu which is
1766 * required to prevent single threaded performance regressions
1767 * especially on architectures where smp_rmb (in seqcounts) are costly.
1768 * Keep the two functions in sync.
1772 * The hash list is protected using RCU.
1774 * Carefully use d_seq when comparing a candidate dentry, to avoid
1775 * races with d_move().
1777 * It is possible that concurrent renames can mess up our list
1778 * walk here and result in missing our dentry, resulting in the
1779 * false-negative result. d_lookup() protects against concurrent
1780 * renames using rename_lock seqlock.
1782 * See Documentation/vfs/dcache-locking.txt for more details.
1784 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1789 if (dentry->d_name.hash != hash)
1793 *seq = read_seqcount_begin(&dentry->d_seq);
1794 if (dentry->d_parent != parent)
1796 if (d_unhashed(dentry))
1798 tlen = dentry->d_name.len;
1799 tname = dentry->d_name.name;
1800 i = dentry->d_inode;
1805 * This seqcount check is required to ensure name and
1806 * len are loaded atomically, so as not to walk off the
1807 * edge of memory when walking. If we could load this
1808 * atomically some other way, we could drop this check.
1810 if (read_seqcount_retry(&dentry->d_seq, *seq))
1812 if (parent->d_flags & DCACHE_OP_COMPARE) {
1813 if (parent->d_op->d_compare(parent, *inode,
1818 if (dentry_cmp(tname, tlen, str, len))
1822 * No extra seqcount check is required after the name
1823 * compare. The caller must perform a seqcount check in
1824 * order to do anything useful with the returned dentry
1834 * d_lookup - search for a dentry
1835 * @parent: parent dentry
1836 * @name: qstr of name we wish to find
1837 * Returns: dentry, or NULL
1839 * d_lookup searches the children of the parent dentry for the name in
1840 * question. If the dentry is found its reference count is incremented and the
1841 * dentry is returned. The caller must use dput to free the entry when it has
1842 * finished using it. %NULL is returned if the dentry does not exist.
1844 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1846 struct dentry *dentry;
1850 seq = read_seqbegin(&rename_lock);
1851 dentry = __d_lookup(parent, name);
1854 } while (read_seqretry(&rename_lock, seq));
1857 EXPORT_SYMBOL(d_lookup);
1860 * __d_lookup - search for a dentry (racy)
1861 * @parent: parent dentry
1862 * @name: qstr of name we wish to find
1863 * Returns: dentry, or NULL
1865 * __d_lookup is like d_lookup, however it may (rarely) return a
1866 * false-negative result due to unrelated rename activity.
1868 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1869 * however it must be used carefully, eg. with a following d_lookup in
1870 * the case of failure.
1872 * __d_lookup callers must be commented.
1874 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1876 unsigned int len = name->len;
1877 unsigned int hash = name->hash;
1878 const unsigned char *str = name->name;
1879 struct dcache_hash_bucket *b = d_hash(parent, hash);
1880 struct hlist_bl_node *node;
1881 struct dentry *found = NULL;
1882 struct dentry *dentry;
1885 * Note: There is significant duplication with __d_lookup_rcu which is
1886 * required to prevent single threaded performance regressions
1887 * especially on architectures where smp_rmb (in seqcounts) are costly.
1888 * Keep the two functions in sync.
1892 * The hash list is protected using RCU.
1894 * Take d_lock when comparing a candidate dentry, to avoid races
1897 * It is possible that concurrent renames can mess up our list
1898 * walk here and result in missing our dentry, resulting in the
1899 * false-negative result. d_lookup() protects against concurrent
1900 * renames using rename_lock seqlock.
1902 * See Documentation/vfs/dcache-locking.txt for more details.
1906 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1910 if (dentry->d_name.hash != hash)
1913 spin_lock(&dentry->d_lock);
1914 if (dentry->d_parent != parent)
1916 if (d_unhashed(dentry))
1920 * It is safe to compare names since d_move() cannot
1921 * change the qstr (protected by d_lock).
1923 tlen = dentry->d_name.len;
1924 tname = dentry->d_name.name;
1925 if (parent->d_flags & DCACHE_OP_COMPARE) {
1926 if (parent->d_op->d_compare(parent, parent->d_inode,
1927 dentry, dentry->d_inode,
1931 if (dentry_cmp(tname, tlen, str, len))
1937 spin_unlock(&dentry->d_lock);
1940 spin_unlock(&dentry->d_lock);
1948 * d_hash_and_lookup - hash the qstr then search for a dentry
1949 * @dir: Directory to search in
1950 * @name: qstr of name we wish to find
1952 * On hash failure or on lookup failure NULL is returned.
1954 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1956 struct dentry *dentry = NULL;
1959 * Check for a fs-specific hash function. Note that we must
1960 * calculate the standard hash first, as the d_op->d_hash()
1961 * routine may choose to leave the hash value unchanged.
1963 name->hash = full_name_hash(name->name, name->len);
1964 if (dir->d_flags & DCACHE_OP_HASH) {
1965 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1968 dentry = d_lookup(dir, name);
1974 * d_validate - verify dentry provided from insecure source (deprecated)
1975 * @dentry: The dentry alleged to be valid child of @dparent
1976 * @parent: The parent dentry (known to be valid)
1978 * An insecure source has sent us a dentry, here we verify it and dget() it.
1979 * This is used by ncpfs in its readdir implementation.
1980 * Zero is returned in the dentry is invalid.
1982 * This function is slow for big directories, and deprecated, do not use it.
1984 int d_validate(struct dentry *dentry, struct dentry *dparent)
1986 struct dentry *child;
1988 spin_lock(&dparent->d_lock);
1989 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1990 if (dentry == child) {
1991 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1992 __dget_dlock(dentry);
1993 spin_unlock(&dentry->d_lock);
1994 spin_unlock(&dparent->d_lock);
1998 spin_unlock(&dparent->d_lock);
2002 EXPORT_SYMBOL(d_validate);
2005 * When a file is deleted, we have two options:
2006 * - turn this dentry into a negative dentry
2007 * - unhash this dentry and free it.
2009 * Usually, we want to just turn this into
2010 * a negative dentry, but if anybody else is
2011 * currently using the dentry or the inode
2012 * we can't do that and we fall back on removing
2013 * it from the hash queues and waiting for
2014 * it to be deleted later when it has no users
2018 * d_delete - delete a dentry
2019 * @dentry: The dentry to delete
2021 * Turn the dentry into a negative dentry if possible, otherwise
2022 * remove it from the hash queues so it can be deleted later
2025 void d_delete(struct dentry * dentry)
2027 struct inode *inode;
2030 * Are we the only user?
2033 spin_lock(&dentry->d_lock);
2034 inode = dentry->d_inode;
2035 isdir = S_ISDIR(inode->i_mode);
2036 if (dentry->d_count == 1) {
2037 if (inode && !spin_trylock(&inode->i_lock)) {
2038 spin_unlock(&dentry->d_lock);
2042 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2043 dentry_unlink_inode(dentry);
2044 fsnotify_nameremove(dentry, isdir);
2048 if (!d_unhashed(dentry))
2051 spin_unlock(&dentry->d_lock);
2053 fsnotify_nameremove(dentry, isdir);
2055 EXPORT_SYMBOL(d_delete);
2057 static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b)
2059 BUG_ON(!d_unhashed(entry));
2060 spin_lock_bucket(b);
2061 entry->d_flags &= ~DCACHE_UNHASHED;
2062 hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
2063 spin_unlock_bucket(b);
2066 static void _d_rehash(struct dentry * entry)
2068 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2072 * d_rehash - add an entry back to the hash
2073 * @entry: dentry to add to the hash
2075 * Adds a dentry to the hash according to its name.
2078 void d_rehash(struct dentry * entry)
2080 spin_lock(&entry->d_lock);
2082 spin_unlock(&entry->d_lock);
2084 EXPORT_SYMBOL(d_rehash);
2087 * dentry_update_name_case - update case insensitive dentry with a new name
2088 * @dentry: dentry to be updated
2091 * Update a case insensitive dentry with new case of name.
2093 * dentry must have been returned by d_lookup with name @name. Old and new
2094 * name lengths must match (ie. no d_compare which allows mismatched name
2097 * Parent inode i_mutex must be held over d_lookup and into this call (to
2098 * keep renames and concurrent inserts, and readdir(2) away).
2100 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2102 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2103 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2105 spin_lock(&dentry->d_lock);
2106 write_seqcount_begin(&dentry->d_seq);
2107 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2108 write_seqcount_end(&dentry->d_seq);
2109 spin_unlock(&dentry->d_lock);
2111 EXPORT_SYMBOL(dentry_update_name_case);
2113 static void switch_names(struct dentry *dentry, struct dentry *target)
2115 if (dname_external(target)) {
2116 if (dname_external(dentry)) {
2118 * Both external: swap the pointers
2120 swap(target->d_name.name, dentry->d_name.name);
2123 * dentry:internal, target:external. Steal target's
2124 * storage and make target internal.
2126 memcpy(target->d_iname, dentry->d_name.name,
2127 dentry->d_name.len + 1);
2128 dentry->d_name.name = target->d_name.name;
2129 target->d_name.name = target->d_iname;
2132 if (dname_external(dentry)) {
2134 * dentry:external, target:internal. Give dentry's
2135 * storage to target and make dentry internal
2137 memcpy(dentry->d_iname, target->d_name.name,
2138 target->d_name.len + 1);
2139 target->d_name.name = dentry->d_name.name;
2140 dentry->d_name.name = dentry->d_iname;
2143 * Both are internal. Just copy target to dentry
2145 memcpy(dentry->d_iname, target->d_name.name,
2146 target->d_name.len + 1);
2147 dentry->d_name.len = target->d_name.len;
2151 swap(dentry->d_name.len, target->d_name.len);
2154 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2157 * XXXX: do we really need to take target->d_lock?
2159 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2160 spin_lock(&target->d_parent->d_lock);
2162 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2163 spin_lock(&dentry->d_parent->d_lock);
2164 spin_lock_nested(&target->d_parent->d_lock,
2165 DENTRY_D_LOCK_NESTED);
2167 spin_lock(&target->d_parent->d_lock);
2168 spin_lock_nested(&dentry->d_parent->d_lock,
2169 DENTRY_D_LOCK_NESTED);
2172 if (target < dentry) {
2173 spin_lock_nested(&target->d_lock, 2);
2174 spin_lock_nested(&dentry->d_lock, 3);
2176 spin_lock_nested(&dentry->d_lock, 2);
2177 spin_lock_nested(&target->d_lock, 3);
2181 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2182 struct dentry *target)
2184 if (target->d_parent != dentry->d_parent)
2185 spin_unlock(&dentry->d_parent->d_lock);
2186 if (target->d_parent != target)
2187 spin_unlock(&target->d_parent->d_lock);
2191 * When switching names, the actual string doesn't strictly have to
2192 * be preserved in the target - because we're dropping the target
2193 * anyway. As such, we can just do a simple memcpy() to copy over
2194 * the new name before we switch.
2196 * Note that we have to be a lot more careful about getting the hash
2197 * switched - we have to switch the hash value properly even if it
2198 * then no longer matches the actual (corrupted) string of the target.
2199 * The hash value has to match the hash queue that the dentry is on..
2202 * d_move - move a dentry
2203 * @dentry: entry to move
2204 * @target: new dentry
2206 * Update the dcache to reflect the move of a file name. Negative
2207 * dcache entries should not be moved in this way.
2209 void d_move(struct dentry * dentry, struct dentry * target)
2211 if (!dentry->d_inode)
2212 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2214 BUG_ON(d_ancestor(dentry, target));
2215 BUG_ON(d_ancestor(target, dentry));
2217 write_seqlock(&rename_lock);
2219 dentry_lock_for_move(dentry, target);
2221 write_seqcount_begin(&dentry->d_seq);
2222 write_seqcount_begin(&target->d_seq);
2224 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2227 * Move the dentry to the target hash queue. Don't bother checking
2228 * for the same hash queue because of how unlikely it is.
2231 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2233 /* Unhash the target: dput() will then get rid of it */
2236 list_del(&dentry->d_u.d_child);
2237 list_del(&target->d_u.d_child);
2239 /* Switch the names.. */
2240 switch_names(dentry, target);
2241 swap(dentry->d_name.hash, target->d_name.hash);
2243 /* ... and switch the parents */
2244 if (IS_ROOT(dentry)) {
2245 dentry->d_parent = target->d_parent;
2246 target->d_parent = target;
2247 INIT_LIST_HEAD(&target->d_u.d_child);
2249 swap(dentry->d_parent, target->d_parent);
2251 /* And add them back to the (new) parent lists */
2252 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2255 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2257 write_seqcount_end(&target->d_seq);
2258 write_seqcount_end(&dentry->d_seq);
2260 dentry_unlock_parents_for_move(dentry, target);
2261 spin_unlock(&target->d_lock);
2262 fsnotify_d_move(dentry);
2263 spin_unlock(&dentry->d_lock);
2264 write_sequnlock(&rename_lock);
2266 EXPORT_SYMBOL(d_move);
2269 * d_ancestor - search for an ancestor
2270 * @p1: ancestor dentry
2273 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2274 * an ancestor of p2, else NULL.
2276 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2280 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2281 if (p->d_parent == p1)
2288 * This helper attempts to cope with remotely renamed directories
2290 * It assumes that the caller is already holding
2291 * dentry->d_parent->d_inode->i_mutex and the inode->i_lock
2293 * Note: If ever the locking in lock_rename() changes, then please
2294 * remember to update this too...
2296 static struct dentry *__d_unalias(struct inode *inode,
2297 struct dentry *dentry, struct dentry *alias)
2299 struct mutex *m1 = NULL, *m2 = NULL;
2302 /* If alias and dentry share a parent, then no extra locks required */
2303 if (alias->d_parent == dentry->d_parent)
2306 /* Check for loops */
2307 ret = ERR_PTR(-ELOOP);
2308 if (d_ancestor(alias, dentry))
2311 /* See lock_rename() */
2312 ret = ERR_PTR(-EBUSY);
2313 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2315 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2316 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2318 m2 = &alias->d_parent->d_inode->i_mutex;
2320 d_move(alias, dentry);
2323 spin_unlock(&inode->i_lock);
2332 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2333 * named dentry in place of the dentry to be replaced.
2334 * returns with anon->d_lock held!
2336 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2338 struct dentry *dparent, *aparent;
2340 dentry_lock_for_move(anon, dentry);
2342 write_seqcount_begin(&dentry->d_seq);
2343 write_seqcount_begin(&anon->d_seq);
2345 dparent = dentry->d_parent;
2346 aparent = anon->d_parent;
2348 switch_names(dentry, anon);
2349 swap(dentry->d_name.hash, anon->d_name.hash);
2351 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2352 list_del(&dentry->d_u.d_child);
2353 if (!IS_ROOT(dentry))
2354 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2356 INIT_LIST_HEAD(&dentry->d_u.d_child);
2358 anon->d_parent = (dparent == dentry) ? anon : dparent;
2359 list_del(&anon->d_u.d_child);
2361 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2363 INIT_LIST_HEAD(&anon->d_u.d_child);
2365 write_seqcount_end(&dentry->d_seq);
2366 write_seqcount_end(&anon->d_seq);
2368 dentry_unlock_parents_for_move(anon, dentry);
2369 spin_unlock(&dentry->d_lock);
2371 /* anon->d_lock still locked, returns locked */
2372 anon->d_flags &= ~DCACHE_DISCONNECTED;
2376 * d_materialise_unique - introduce an inode into the tree
2377 * @dentry: candidate dentry
2378 * @inode: inode to bind to the dentry, to which aliases may be attached
2380 * Introduces an dentry into the tree, substituting an extant disconnected
2381 * root directory alias in its place if there is one
2383 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2385 struct dentry *actual;
2387 BUG_ON(!d_unhashed(dentry));
2391 __d_instantiate(dentry, NULL);
2396 spin_lock(&inode->i_lock);
2398 if (S_ISDIR(inode->i_mode)) {
2399 struct dentry *alias;
2401 /* Does an aliased dentry already exist? */
2402 alias = __d_find_alias(inode, 0);
2405 /* Is this an anonymous mountpoint that we could splice
2407 if (IS_ROOT(alias)) {
2408 __d_materialise_dentry(dentry, alias);
2412 /* Nope, but we must(!) avoid directory aliasing */
2413 actual = __d_unalias(inode, dentry, alias);
2420 /* Add a unique reference */
2421 actual = __d_instantiate_unique(dentry, inode);
2425 BUG_ON(!d_unhashed(actual));
2427 spin_lock(&actual->d_lock);
2430 spin_unlock(&actual->d_lock);
2431 spin_unlock(&inode->i_lock);
2433 if (actual == dentry) {
2434 security_d_instantiate(dentry, inode);
2441 EXPORT_SYMBOL_GPL(d_materialise_unique);
2443 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2447 return -ENAMETOOLONG;
2449 memcpy(*buffer, str, namelen);
2453 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2455 return prepend(buffer, buflen, name->name, name->len);
2459 * prepend_path - Prepend path string to a buffer
2460 * @path: the dentry/vfsmount to report
2461 * @root: root vfsmnt/dentry (may be modified by this function)
2462 * @buffer: pointer to the end of the buffer
2463 * @buflen: pointer to buffer length
2465 * Caller holds the rename_lock.
2467 * If path is not reachable from the supplied root, then the value of
2468 * root is changed (without modifying refcounts).
2470 static int prepend_path(const struct path *path, struct path *root,
2471 char **buffer, int *buflen)
2473 struct dentry *dentry = path->dentry;
2474 struct vfsmount *vfsmnt = path->mnt;
2478 br_read_lock(vfsmount_lock);
2479 while (dentry != root->dentry || vfsmnt != root->mnt) {
2480 struct dentry * parent;
2482 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2484 if (vfsmnt->mnt_parent == vfsmnt) {
2487 dentry = vfsmnt->mnt_mountpoint;
2488 vfsmnt = vfsmnt->mnt_parent;
2491 parent = dentry->d_parent;
2493 spin_lock(&dentry->d_lock);
2494 error = prepend_name(buffer, buflen, &dentry->d_name);
2495 spin_unlock(&dentry->d_lock);
2497 error = prepend(buffer, buflen, "/", 1);
2506 if (!error && !slash)
2507 error = prepend(buffer, buflen, "/", 1);
2509 br_read_unlock(vfsmount_lock);
2514 * Filesystems needing to implement special "root names"
2515 * should do so with ->d_dname()
2517 if (IS_ROOT(dentry) &&
2518 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2519 WARN(1, "Root dentry has weird name <%.*s>\n",
2520 (int) dentry->d_name.len, dentry->d_name.name);
2523 root->dentry = dentry;
2528 * __d_path - return the path of a dentry
2529 * @path: the dentry/vfsmount to report
2530 * @root: root vfsmnt/dentry (may be modified by this function)
2531 * @buf: buffer to return value in
2532 * @buflen: buffer length
2534 * Convert a dentry into an ASCII path name.
2536 * Returns a pointer into the buffer or an error code if the
2537 * path was too long.
2539 * "buflen" should be positive.
2541 * If path is not reachable from the supplied root, then the value of
2542 * root is changed (without modifying refcounts).
2544 char *__d_path(const struct path *path, struct path *root,
2545 char *buf, int buflen)
2547 char *res = buf + buflen;
2550 prepend(&res, &buflen, "\0", 1);
2551 write_seqlock(&rename_lock);
2552 error = prepend_path(path, root, &res, &buflen);
2553 write_sequnlock(&rename_lock);
2556 return ERR_PTR(error);
2561 * same as __d_path but appends "(deleted)" for unlinked files.
2563 static int path_with_deleted(const struct path *path, struct path *root,
2564 char **buf, int *buflen)
2566 prepend(buf, buflen, "\0", 1);
2567 if (d_unlinked(path->dentry)) {
2568 int error = prepend(buf, buflen, " (deleted)", 10);
2573 return prepend_path(path, root, buf, buflen);
2576 static int prepend_unreachable(char **buffer, int *buflen)
2578 return prepend(buffer, buflen, "(unreachable)", 13);
2582 * d_path - return the path of a dentry
2583 * @path: path to report
2584 * @buf: buffer to return value in
2585 * @buflen: buffer length
2587 * Convert a dentry into an ASCII path name. If the entry has been deleted
2588 * the string " (deleted)" is appended. Note that this is ambiguous.
2590 * Returns a pointer into the buffer or an error code if the path was
2591 * too long. Note: Callers should use the returned pointer, not the passed
2592 * in buffer, to use the name! The implementation often starts at an offset
2593 * into the buffer, and may leave 0 bytes at the start.
2595 * "buflen" should be positive.
2597 char *d_path(const struct path *path, char *buf, int buflen)
2599 char *res = buf + buflen;
2605 * We have various synthetic filesystems that never get mounted. On
2606 * these filesystems dentries are never used for lookup purposes, and
2607 * thus don't need to be hashed. They also don't need a name until a
2608 * user wants to identify the object in /proc/pid/fd/. The little hack
2609 * below allows us to generate a name for these objects on demand:
2611 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2612 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2614 get_fs_root(current->fs, &root);
2615 write_seqlock(&rename_lock);
2617 error = path_with_deleted(path, &tmp, &res, &buflen);
2619 res = ERR_PTR(error);
2620 write_sequnlock(&rename_lock);
2624 EXPORT_SYMBOL(d_path);
2627 * d_path_with_unreachable - return the path of a dentry
2628 * @path: path to report
2629 * @buf: buffer to return value in
2630 * @buflen: buffer length
2632 * The difference from d_path() is that this prepends "(unreachable)"
2633 * to paths which are unreachable from the current process' root.
2635 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2637 char *res = buf + buflen;
2642 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2643 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2645 get_fs_root(current->fs, &root);
2646 write_seqlock(&rename_lock);
2648 error = path_with_deleted(path, &tmp, &res, &buflen);
2649 if (!error && !path_equal(&tmp, &root))
2650 error = prepend_unreachable(&res, &buflen);
2651 write_sequnlock(&rename_lock);
2654 res = ERR_PTR(error);
2660 * Helper function for dentry_operations.d_dname() members
2662 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2663 const char *fmt, ...)
2669 va_start(args, fmt);
2670 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2673 if (sz > sizeof(temp) || sz > buflen)
2674 return ERR_PTR(-ENAMETOOLONG);
2676 buffer += buflen - sz;
2677 return memcpy(buffer, temp, sz);
2681 * Write full pathname from the root of the filesystem into the buffer.
2683 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2685 char *end = buf + buflen;
2688 prepend(&end, &buflen, "\0", 1);
2695 while (!IS_ROOT(dentry)) {
2696 struct dentry *parent = dentry->d_parent;
2700 spin_lock(&dentry->d_lock);
2701 error = prepend_name(&end, &buflen, &dentry->d_name);
2702 spin_unlock(&dentry->d_lock);
2703 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2711 return ERR_PTR(-ENAMETOOLONG);
2714 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2718 write_seqlock(&rename_lock);
2719 retval = __dentry_path(dentry, buf, buflen);
2720 write_sequnlock(&rename_lock);
2724 EXPORT_SYMBOL(dentry_path_raw);
2726 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2731 write_seqlock(&rename_lock);
2732 if (d_unlinked(dentry)) {
2734 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2738 retval = __dentry_path(dentry, buf, buflen);
2739 write_sequnlock(&rename_lock);
2740 if (!IS_ERR(retval) && p)
2741 *p = '/'; /* restore '/' overriden with '\0' */
2744 return ERR_PTR(-ENAMETOOLONG);
2748 * NOTE! The user-level library version returns a
2749 * character pointer. The kernel system call just
2750 * returns the length of the buffer filled (which
2751 * includes the ending '\0' character), or a negative
2752 * error value. So libc would do something like
2754 * char *getcwd(char * buf, size_t size)
2758 * retval = sys_getcwd(buf, size);
2765 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2768 struct path pwd, root;
2769 char *page = (char *) __get_free_page(GFP_USER);
2774 get_fs_root_and_pwd(current->fs, &root, &pwd);
2777 write_seqlock(&rename_lock);
2778 if (!d_unlinked(pwd.dentry)) {
2780 struct path tmp = root;
2781 char *cwd = page + PAGE_SIZE;
2782 int buflen = PAGE_SIZE;
2784 prepend(&cwd, &buflen, "\0", 1);
2785 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2786 write_sequnlock(&rename_lock);
2791 /* Unreachable from current root */
2792 if (!path_equal(&tmp, &root)) {
2793 error = prepend_unreachable(&cwd, &buflen);
2799 len = PAGE_SIZE + page - cwd;
2802 if (copy_to_user(buf, cwd, len))
2806 write_sequnlock(&rename_lock);
2812 free_page((unsigned long) page);
2817 * Test whether new_dentry is a subdirectory of old_dentry.
2819 * Trivially implemented using the dcache structure
2823 * is_subdir - is new dentry a subdirectory of old_dentry
2824 * @new_dentry: new dentry
2825 * @old_dentry: old dentry
2827 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2828 * Returns 0 otherwise.
2829 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2832 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2837 if (new_dentry == old_dentry)
2841 /* for restarting inner loop in case of seq retry */
2842 seq = read_seqbegin(&rename_lock);
2844 * Need rcu_readlock to protect against the d_parent trashing
2848 if (d_ancestor(old_dentry, new_dentry))
2853 } while (read_seqretry(&rename_lock, seq));
2858 int path_is_under(struct path *path1, struct path *path2)
2860 struct vfsmount *mnt = path1->mnt;
2861 struct dentry *dentry = path1->dentry;
2864 br_read_lock(vfsmount_lock);
2865 if (mnt != path2->mnt) {
2867 if (mnt->mnt_parent == mnt) {
2868 br_read_unlock(vfsmount_lock);
2871 if (mnt->mnt_parent == path2->mnt)
2873 mnt = mnt->mnt_parent;
2875 dentry = mnt->mnt_mountpoint;
2877 res = is_subdir(dentry, path2->dentry);
2878 br_read_unlock(vfsmount_lock);
2881 EXPORT_SYMBOL(path_is_under);
2883 void d_genocide(struct dentry *root)
2885 struct dentry *this_parent;
2886 struct list_head *next;
2890 seq = read_seqbegin(&rename_lock);
2893 spin_lock(&this_parent->d_lock);
2895 next = this_parent->d_subdirs.next;
2897 while (next != &this_parent->d_subdirs) {
2898 struct list_head *tmp = next;
2899 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2902 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2903 if (d_unhashed(dentry) || !dentry->d_inode) {
2904 spin_unlock(&dentry->d_lock);
2907 if (!list_empty(&dentry->d_subdirs)) {
2908 spin_unlock(&this_parent->d_lock);
2909 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2910 this_parent = dentry;
2911 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2914 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2915 dentry->d_flags |= DCACHE_GENOCIDE;
2918 spin_unlock(&dentry->d_lock);
2920 if (this_parent != root) {
2922 struct dentry *child;
2924 tmp = this_parent->d_parent;
2925 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2926 this_parent->d_flags |= DCACHE_GENOCIDE;
2927 this_parent->d_count--;
2930 spin_unlock(&this_parent->d_lock);
2931 child = this_parent;
2933 spin_lock(&this_parent->d_lock);
2934 /* might go back up the wrong parent if we have had a rename
2936 if (this_parent != child->d_parent ||
2937 (!locked && read_seqretry(&rename_lock, seq))) {
2938 spin_unlock(&this_parent->d_lock);
2943 next = child->d_u.d_child.next;
2946 spin_unlock(&this_parent->d_lock);
2947 if (!locked && read_seqretry(&rename_lock, seq))
2950 write_sequnlock(&rename_lock);
2955 write_seqlock(&rename_lock);
2960 * find_inode_number - check for dentry with name
2961 * @dir: directory to check
2962 * @name: Name to find.
2964 * Check whether a dentry already exists for the given name,
2965 * and return the inode number if it has an inode. Otherwise
2968 * This routine is used to post-process directory listings for
2969 * filesystems using synthetic inode numbers, and is necessary
2970 * to keep getcwd() working.
2973 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2975 struct dentry * dentry;
2978 dentry = d_hash_and_lookup(dir, name);
2980 if (dentry->d_inode)
2981 ino = dentry->d_inode->i_ino;
2986 EXPORT_SYMBOL(find_inode_number);
2988 static __initdata unsigned long dhash_entries;
2989 static int __init set_dhash_entries(char *str)
2993 dhash_entries = simple_strtoul(str, &str, 0);
2996 __setup("dhash_entries=", set_dhash_entries);
2998 static void __init dcache_init_early(void)
3002 /* If hashes are distributed across NUMA nodes, defer
3003 * hash allocation until vmalloc space is available.
3009 alloc_large_system_hash("Dentry cache",
3010 sizeof(struct dcache_hash_bucket),
3018 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3019 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3022 static void __init dcache_init(void)
3027 * A constructor could be added for stable state like the lists,
3028 * but it is probably not worth it because of the cache nature
3031 dentry_cache = KMEM_CACHE(dentry,
3032 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3034 register_shrinker(&dcache_shrinker);
3036 /* Hash may have been set up in dcache_init_early */
3041 alloc_large_system_hash("Dentry cache",
3042 sizeof(struct dcache_hash_bucket),
3050 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3051 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3054 /* SLAB cache for __getname() consumers */
3055 struct kmem_cache *names_cachep __read_mostly;
3056 EXPORT_SYMBOL(names_cachep);
3058 EXPORT_SYMBOL(d_genocide);
3060 void __init vfs_caches_init_early(void)
3062 dcache_init_early();
3066 void __init vfs_caches_init(unsigned long mempages)
3068 unsigned long reserve;
3070 /* Base hash sizes on available memory, with a reserve equal to
3071 150% of current kernel size */
3073 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3074 mempages -= reserve;
3076 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3077 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3081 files_init(mempages);