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>
40 * dcache_inode_lock protects:
41 * - i_dentry, d_alias, d_inode
42 * dcache_hash_lock protects:
43 * - the dcache hash table, s_anon lists
44 * dcache_lru_lock protects:
45 * - the dcache lru lists and counters
52 * - d_parent and d_subdirs
53 * - childrens' d_child and d_parent
62 * If there is an ancestor relationship:
63 * dentry->d_parent->...->d_parent->d_lock
65 * dentry->d_parent->d_lock
68 * If no ancestor relationship:
69 * if (dentry1 < dentry2)
73 int sysctl_vfs_cache_pressure __read_mostly = 100;
74 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
76 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_inode_lock);
77 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_hash_lock);
78 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
81 EXPORT_SYMBOL(rename_lock);
82 EXPORT_SYMBOL(dcache_inode_lock);
84 static struct kmem_cache *dentry_cache __read_mostly;
87 * This is the single most critical data structure when it comes
88 * to the dcache: the hashtable for lookups. Somebody should try
89 * to make this good - I've just made it work.
91 * This hash-function tries to avoid losing too many bits of hash
92 * information, yet avoid using a prime hash-size or similar.
94 #define D_HASHBITS d_hash_shift
95 #define D_HASHMASK d_hash_mask
97 static unsigned int d_hash_mask __read_mostly;
98 static unsigned int d_hash_shift __read_mostly;
99 static struct hlist_head *dentry_hashtable __read_mostly;
101 /* Statistics gathering. */
102 struct dentry_stat_t dentry_stat = {
106 static DEFINE_PER_CPU(unsigned int, nr_dentry);
108 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
109 static int get_nr_dentry(void)
113 for_each_possible_cpu(i)
114 sum += per_cpu(nr_dentry, i);
115 return sum < 0 ? 0 : sum;
118 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
119 size_t *lenp, loff_t *ppos)
121 dentry_stat.nr_dentry = get_nr_dentry();
122 return proc_dointvec(table, write, buffer, lenp, ppos);
126 static void __d_free(struct rcu_head *head)
128 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
130 WARN_ON(!list_empty(&dentry->d_alias));
131 if (dname_external(dentry))
132 kfree(dentry->d_name.name);
133 kmem_cache_free(dentry_cache, dentry);
139 static void d_free(struct dentry *dentry)
141 BUG_ON(dentry->d_count);
142 this_cpu_dec(nr_dentry);
143 if (dentry->d_op && dentry->d_op->d_release)
144 dentry->d_op->d_release(dentry);
146 /* if dentry was never inserted into hash, immediate free is OK */
147 if (hlist_unhashed(&dentry->d_hash))
148 __d_free(&dentry->d_u.d_rcu);
150 call_rcu(&dentry->d_u.d_rcu, __d_free);
154 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
155 * After this call, in-progress rcu-walk path lookup will fail. This
156 * should be called after unhashing, and after changing d_inode (if
157 * the dentry has not already been unhashed).
159 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
161 assert_spin_locked(&dentry->d_lock);
162 /* Go through a barrier */
163 write_seqcount_barrier(&dentry->d_seq);
167 * Release the dentry's inode, using the filesystem
168 * d_iput() operation if defined. Dentry has no refcount
171 static void dentry_iput(struct dentry * dentry)
172 __releases(dentry->d_lock)
173 __releases(dcache_inode_lock)
175 struct inode *inode = dentry->d_inode;
177 dentry->d_inode = NULL;
178 list_del_init(&dentry->d_alias);
179 spin_unlock(&dentry->d_lock);
180 spin_unlock(&dcache_inode_lock);
182 fsnotify_inoderemove(inode);
183 if (dentry->d_op && dentry->d_op->d_iput)
184 dentry->d_op->d_iput(dentry, inode);
188 spin_unlock(&dentry->d_lock);
189 spin_unlock(&dcache_inode_lock);
194 * Release the dentry's inode, using the filesystem
195 * d_iput() operation if defined. dentry remains in-use.
197 static void dentry_unlink_inode(struct dentry * dentry)
198 __releases(dentry->d_lock)
199 __releases(dcache_inode_lock)
201 struct inode *inode = dentry->d_inode;
202 dentry->d_inode = NULL;
203 list_del_init(&dentry->d_alias);
204 dentry_rcuwalk_barrier(dentry);
205 spin_unlock(&dentry->d_lock);
206 spin_unlock(&dcache_inode_lock);
208 fsnotify_inoderemove(inode);
209 if (dentry->d_op && dentry->d_op->d_iput)
210 dentry->d_op->d_iput(dentry, inode);
216 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
218 static void dentry_lru_add(struct dentry *dentry)
220 if (list_empty(&dentry->d_lru)) {
221 spin_lock(&dcache_lru_lock);
222 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
223 dentry->d_sb->s_nr_dentry_unused++;
224 dentry_stat.nr_unused++;
225 spin_unlock(&dcache_lru_lock);
229 static void __dentry_lru_del(struct dentry *dentry)
231 list_del_init(&dentry->d_lru);
232 dentry->d_sb->s_nr_dentry_unused--;
233 dentry_stat.nr_unused--;
236 static void dentry_lru_del(struct dentry *dentry)
238 if (!list_empty(&dentry->d_lru)) {
239 spin_lock(&dcache_lru_lock);
240 __dentry_lru_del(dentry);
241 spin_unlock(&dcache_lru_lock);
245 static void dentry_lru_move_tail(struct dentry *dentry)
247 spin_lock(&dcache_lru_lock);
248 if (list_empty(&dentry->d_lru)) {
249 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
250 dentry->d_sb->s_nr_dentry_unused++;
251 dentry_stat.nr_unused++;
253 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
255 spin_unlock(&dcache_lru_lock);
259 * d_kill - kill dentry and return parent
260 * @dentry: dentry to kill
262 * The dentry must already be unhashed and removed from the LRU.
264 * If this is the root of the dentry tree, return NULL.
266 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
269 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
270 __releases(dentry->d_lock)
271 __releases(parent->d_lock)
272 __releases(dcache_inode_lock)
274 dentry->d_parent = NULL;
275 list_del(&dentry->d_u.d_child);
277 spin_unlock(&parent->d_lock);
280 * dentry_iput drops the locks, at which point nobody (except
281 * transient RCU lookups) can reach this dentry.
288 * d_drop - drop a dentry
289 * @dentry: dentry to drop
291 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
292 * be found through a VFS lookup any more. Note that this is different from
293 * deleting the dentry - d_delete will try to mark the dentry negative if
294 * possible, giving a successful _negative_ lookup, while d_drop will
295 * just make the cache lookup fail.
297 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
298 * reason (NFS timeouts or autofs deletes).
300 * __d_drop requires dentry->d_lock.
302 void __d_drop(struct dentry *dentry)
304 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
305 dentry->d_flags |= DCACHE_UNHASHED;
306 spin_lock(&dcache_hash_lock);
307 hlist_del_rcu(&dentry->d_hash);
308 spin_unlock(&dcache_hash_lock);
309 dentry_rcuwalk_barrier(dentry);
312 EXPORT_SYMBOL(__d_drop);
314 void d_drop(struct dentry *dentry)
316 spin_lock(&dentry->d_lock);
318 spin_unlock(&dentry->d_lock);
320 EXPORT_SYMBOL(d_drop);
323 * Finish off a dentry we've decided to kill.
324 * dentry->d_lock must be held, returns with it unlocked.
325 * If ref is non-zero, then decrement the refcount too.
326 * Returns dentry requiring refcount drop, or NULL if we're done.
328 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
329 __releases(dentry->d_lock)
331 struct dentry *parent;
333 if (!spin_trylock(&dcache_inode_lock)) {
335 spin_unlock(&dentry->d_lock);
337 return dentry; /* try again with same dentry */
342 parent = dentry->d_parent;
343 if (parent && !spin_trylock(&parent->d_lock)) {
344 spin_unlock(&dcache_inode_lock);
350 /* if dentry was on the d_lru list delete it from there */
351 dentry_lru_del(dentry);
352 /* if it was on the hash then remove it */
354 return d_kill(dentry, parent);
360 * This is complicated by the fact that we do not want to put
361 * dentries that are no longer on any hash chain on the unused
362 * list: we'd much rather just get rid of them immediately.
364 * However, that implies that we have to traverse the dentry
365 * tree upwards to the parents which might _also_ now be
366 * scheduled for deletion (it may have been only waiting for
367 * its last child to go away).
369 * This tail recursion is done by hand as we don't want to depend
370 * on the compiler to always get this right (gcc generally doesn't).
371 * Real recursion would eat up our stack space.
375 * dput - release a dentry
376 * @dentry: dentry to release
378 * Release a dentry. This will drop the usage count and if appropriate
379 * call the dentry unlink method as well as removing it from the queues and
380 * releasing its resources. If the parent dentries were scheduled for release
381 * they too may now get deleted.
383 void dput(struct dentry *dentry)
389 if (dentry->d_count == 1)
391 spin_lock(&dentry->d_lock);
392 BUG_ON(!dentry->d_count);
393 if (dentry->d_count > 1) {
395 spin_unlock(&dentry->d_lock);
399 if (dentry->d_flags & DCACHE_OP_DELETE) {
400 if (dentry->d_op->d_delete(dentry))
404 /* Unreachable? Get rid of it */
405 if (d_unhashed(dentry))
408 /* Otherwise leave it cached and ensure it's on the LRU */
409 dentry->d_flags |= DCACHE_REFERENCED;
410 dentry_lru_add(dentry);
413 spin_unlock(&dentry->d_lock);
417 dentry = dentry_kill(dentry, 1);
424 * d_invalidate - invalidate a dentry
425 * @dentry: dentry to invalidate
427 * Try to invalidate the dentry if it turns out to be
428 * possible. If there are other dentries that can be
429 * reached through this one we can't delete it and we
430 * return -EBUSY. On success we return 0.
435 int d_invalidate(struct dentry * dentry)
438 * If it's already been dropped, return OK.
440 spin_lock(&dentry->d_lock);
441 if (d_unhashed(dentry)) {
442 spin_unlock(&dentry->d_lock);
446 * Check whether to do a partial shrink_dcache
447 * to get rid of unused child entries.
449 if (!list_empty(&dentry->d_subdirs)) {
450 spin_unlock(&dentry->d_lock);
451 shrink_dcache_parent(dentry);
452 spin_lock(&dentry->d_lock);
456 * Somebody else still using it?
458 * If it's a directory, we can't drop it
459 * for fear of somebody re-populating it
460 * with children (even though dropping it
461 * would make it unreachable from the root,
462 * we might still populate it if it was a
463 * working directory or similar).
465 if (dentry->d_count > 1) {
466 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
467 spin_unlock(&dentry->d_lock);
473 spin_unlock(&dentry->d_lock);
476 EXPORT_SYMBOL(d_invalidate);
478 /* This must be called with d_lock held */
479 static inline void __dget_dlock(struct dentry *dentry)
484 static inline void __dget(struct dentry *dentry)
486 spin_lock(&dentry->d_lock);
487 __dget_dlock(dentry);
488 spin_unlock(&dentry->d_lock);
491 struct dentry *dget_parent(struct dentry *dentry)
497 * Don't need rcu_dereference because we re-check it was correct under
501 ret = dentry->d_parent;
506 spin_lock(&ret->d_lock);
507 if (unlikely(ret != dentry->d_parent)) {
508 spin_unlock(&ret->d_lock);
513 BUG_ON(!ret->d_count);
515 spin_unlock(&ret->d_lock);
519 EXPORT_SYMBOL(dget_parent);
522 * d_find_alias - grab a hashed alias of inode
523 * @inode: inode in question
524 * @want_discon: flag, used by d_splice_alias, to request
525 * that only a DISCONNECTED alias be returned.
527 * If inode has a hashed alias, or is a directory and has any alias,
528 * acquire the reference to alias and return it. Otherwise return NULL.
529 * Notice that if inode is a directory there can be only one alias and
530 * it can be unhashed only if it has no children, or if it is the root
533 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
534 * any other hashed alias over that one unless @want_discon is set,
535 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
537 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
539 struct dentry *alias, *discon_alias;
543 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
544 spin_lock(&alias->d_lock);
545 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
546 if (IS_ROOT(alias) &&
547 (alias->d_flags & DCACHE_DISCONNECTED)) {
548 discon_alias = alias;
549 } else if (!want_discon) {
551 spin_unlock(&alias->d_lock);
555 spin_unlock(&alias->d_lock);
558 alias = discon_alias;
559 spin_lock(&alias->d_lock);
560 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
561 if (IS_ROOT(alias) &&
562 (alias->d_flags & DCACHE_DISCONNECTED)) {
564 spin_unlock(&alias->d_lock);
568 spin_unlock(&alias->d_lock);
574 struct dentry *d_find_alias(struct inode *inode)
576 struct dentry *de = NULL;
578 if (!list_empty(&inode->i_dentry)) {
579 spin_lock(&dcache_inode_lock);
580 de = __d_find_alias(inode, 0);
581 spin_unlock(&dcache_inode_lock);
585 EXPORT_SYMBOL(d_find_alias);
588 * Try to kill dentries associated with this inode.
589 * WARNING: you must own a reference to inode.
591 void d_prune_aliases(struct inode *inode)
593 struct dentry *dentry;
595 spin_lock(&dcache_inode_lock);
596 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
597 spin_lock(&dentry->d_lock);
598 if (!dentry->d_count) {
599 __dget_dlock(dentry);
601 spin_unlock(&dentry->d_lock);
602 spin_unlock(&dcache_inode_lock);
606 spin_unlock(&dentry->d_lock);
608 spin_unlock(&dcache_inode_lock);
610 EXPORT_SYMBOL(d_prune_aliases);
613 * Try to throw away a dentry - free the inode, dput the parent.
614 * Requires dentry->d_lock is held, and dentry->d_count == 0.
615 * Releases dentry->d_lock.
617 * This may fail if locks cannot be acquired no problem, just try again.
619 static void try_prune_one_dentry(struct dentry *dentry)
620 __releases(dentry->d_lock)
622 struct dentry *parent;
624 parent = dentry_kill(dentry, 0);
626 * If dentry_kill returns NULL, we have nothing more to do.
627 * if it returns the same dentry, trylocks failed. In either
628 * case, just loop again.
630 * Otherwise, we need to prune ancestors too. This is necessary
631 * to prevent quadratic behavior of shrink_dcache_parent(), but
632 * is also expected to be beneficial in reducing dentry cache
637 if (parent == dentry)
640 /* Prune ancestors. */
643 spin_lock(&dentry->d_lock);
644 if (dentry->d_count > 1) {
646 spin_unlock(&dentry->d_lock);
649 dentry = dentry_kill(dentry, 1);
653 static void shrink_dentry_list(struct list_head *list)
655 struct dentry *dentry;
659 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
660 if (&dentry->d_lru == list)
662 spin_lock(&dentry->d_lock);
663 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
664 spin_unlock(&dentry->d_lock);
669 * We found an inuse dentry which was not removed from
670 * the LRU because of laziness during lookup. Do not free
671 * it - just keep it off the LRU list.
673 if (dentry->d_count) {
674 dentry_lru_del(dentry);
675 spin_unlock(&dentry->d_lock);
681 try_prune_one_dentry(dentry);
689 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
690 * @sb: superblock to shrink dentry LRU.
691 * @count: number of entries to prune
692 * @flags: flags to control the dentry processing
694 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
696 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
698 /* called from prune_dcache() and shrink_dcache_parent() */
699 struct dentry *dentry;
700 LIST_HEAD(referenced);
705 spin_lock(&dcache_lru_lock);
706 while (!list_empty(&sb->s_dentry_lru)) {
707 dentry = list_entry(sb->s_dentry_lru.prev,
708 struct dentry, d_lru);
709 BUG_ON(dentry->d_sb != sb);
711 if (!spin_trylock(&dentry->d_lock)) {
712 spin_unlock(&dcache_lru_lock);
718 * If we are honouring the DCACHE_REFERENCED flag and the
719 * dentry has this flag set, don't free it. Clear the flag
720 * and put it back on the LRU.
722 if (flags & DCACHE_REFERENCED &&
723 dentry->d_flags & DCACHE_REFERENCED) {
724 dentry->d_flags &= ~DCACHE_REFERENCED;
725 list_move(&dentry->d_lru, &referenced);
726 spin_unlock(&dentry->d_lock);
728 list_move_tail(&dentry->d_lru, &tmp);
729 spin_unlock(&dentry->d_lock);
733 cond_resched_lock(&dcache_lru_lock);
735 if (!list_empty(&referenced))
736 list_splice(&referenced, &sb->s_dentry_lru);
737 spin_unlock(&dcache_lru_lock);
739 shrink_dentry_list(&tmp);
745 * prune_dcache - shrink the dcache
746 * @count: number of entries to try to free
748 * Shrink the dcache. This is done when we need more memory, or simply when we
749 * need to unmount something (at which point we need to unuse all dentries).
751 * This function may fail to free any resources if all the dentries are in use.
753 static void prune_dcache(int count)
755 struct super_block *sb, *p = NULL;
757 int unused = dentry_stat.nr_unused;
761 if (unused == 0 || count == 0)
766 prune_ratio = unused / count;
768 list_for_each_entry(sb, &super_blocks, s_list) {
769 if (list_empty(&sb->s_instances))
771 if (sb->s_nr_dentry_unused == 0)
774 /* Now, we reclaim unused dentrins with fairness.
775 * We reclaim them same percentage from each superblock.
776 * We calculate number of dentries to scan on this sb
777 * as follows, but the implementation is arranged to avoid
779 * number of dentries to scan on this sb =
780 * count * (number of dentries on this sb /
781 * number of dentries in the machine)
783 spin_unlock(&sb_lock);
784 if (prune_ratio != 1)
785 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
787 w_count = sb->s_nr_dentry_unused;
790 * We need to be sure this filesystem isn't being unmounted,
791 * otherwise we could race with generic_shutdown_super(), and
792 * end up holding a reference to an inode while the filesystem
793 * is unmounted. So we try to get s_umount, and make sure
796 if (down_read_trylock(&sb->s_umount)) {
797 if ((sb->s_root != NULL) &&
798 (!list_empty(&sb->s_dentry_lru))) {
799 __shrink_dcache_sb(sb, &w_count,
803 up_read(&sb->s_umount);
810 /* more work left to do? */
816 spin_unlock(&sb_lock);
820 * shrink_dcache_sb - shrink dcache for a superblock
823 * Shrink the dcache for the specified super block. This is used to free
824 * the dcache before unmounting a file system.
826 void shrink_dcache_sb(struct super_block *sb)
830 spin_lock(&dcache_lru_lock);
831 while (!list_empty(&sb->s_dentry_lru)) {
832 list_splice_init(&sb->s_dentry_lru, &tmp);
833 spin_unlock(&dcache_lru_lock);
834 shrink_dentry_list(&tmp);
835 spin_lock(&dcache_lru_lock);
837 spin_unlock(&dcache_lru_lock);
839 EXPORT_SYMBOL(shrink_dcache_sb);
842 * destroy a single subtree of dentries for unmount
843 * - see the comments on shrink_dcache_for_umount() for a description of the
846 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
848 struct dentry *parent;
849 unsigned detached = 0;
851 BUG_ON(!IS_ROOT(dentry));
853 /* detach this root from the system */
854 spin_lock(&dentry->d_lock);
855 dentry_lru_del(dentry);
857 spin_unlock(&dentry->d_lock);
860 /* descend to the first leaf in the current subtree */
861 while (!list_empty(&dentry->d_subdirs)) {
864 /* this is a branch with children - detach all of them
865 * from the system in one go */
866 spin_lock(&dentry->d_lock);
867 list_for_each_entry(loop, &dentry->d_subdirs,
869 spin_lock_nested(&loop->d_lock,
870 DENTRY_D_LOCK_NESTED);
871 dentry_lru_del(loop);
873 spin_unlock(&loop->d_lock);
875 spin_unlock(&dentry->d_lock);
877 /* move to the first child */
878 dentry = list_entry(dentry->d_subdirs.next,
879 struct dentry, d_u.d_child);
882 /* consume the dentries from this leaf up through its parents
883 * until we find one with children or run out altogether */
887 if (dentry->d_count != 0) {
889 "BUG: Dentry %p{i=%lx,n=%s}"
891 " [unmount of %s %s]\n",
894 dentry->d_inode->i_ino : 0UL,
897 dentry->d_sb->s_type->name,
902 if (IS_ROOT(dentry)) {
904 list_del(&dentry->d_u.d_child);
906 parent = dentry->d_parent;
907 spin_lock(&parent->d_lock);
909 list_del(&dentry->d_u.d_child);
910 spin_unlock(&parent->d_lock);
915 inode = dentry->d_inode;
917 dentry->d_inode = NULL;
918 list_del_init(&dentry->d_alias);
919 if (dentry->d_op && dentry->d_op->d_iput)
920 dentry->d_op->d_iput(dentry, inode);
927 /* finished when we fall off the top of the tree,
928 * otherwise we ascend to the parent and move to the
929 * next sibling if there is one */
933 } while (list_empty(&dentry->d_subdirs));
935 dentry = list_entry(dentry->d_subdirs.next,
936 struct dentry, d_u.d_child);
941 * destroy the dentries attached to a superblock on unmounting
942 * - we don't need to use dentry->d_lock because:
943 * - the superblock is detached from all mountings and open files, so the
944 * dentry trees will not be rearranged by the VFS
945 * - s_umount is write-locked, so the memory pressure shrinker will ignore
946 * any dentries belonging to this superblock that it comes across
947 * - the filesystem itself is no longer permitted to rearrange the dentries
950 void shrink_dcache_for_umount(struct super_block *sb)
952 struct dentry *dentry;
954 if (down_read_trylock(&sb->s_umount))
959 spin_lock(&dentry->d_lock);
961 spin_unlock(&dentry->d_lock);
962 shrink_dcache_for_umount_subtree(dentry);
964 while (!hlist_empty(&sb->s_anon)) {
965 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
966 shrink_dcache_for_umount_subtree(dentry);
971 * Search for at least 1 mount point in the dentry's subdirs.
972 * We descend to the next level whenever the d_subdirs
973 * list is non-empty and continue searching.
977 * have_submounts - check for mounts over a dentry
978 * @parent: dentry to check.
980 * Return true if the parent or its subdirectories contain
983 int have_submounts(struct dentry *parent)
985 struct dentry *this_parent;
986 struct list_head *next;
990 seq = read_seqbegin(&rename_lock);
992 this_parent = parent;
994 if (d_mountpoint(parent))
996 spin_lock(&this_parent->d_lock);
998 next = this_parent->d_subdirs.next;
1000 while (next != &this_parent->d_subdirs) {
1001 struct list_head *tmp = next;
1002 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1005 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1006 /* Have we found a mount point ? */
1007 if (d_mountpoint(dentry)) {
1008 spin_unlock(&dentry->d_lock);
1009 spin_unlock(&this_parent->d_lock);
1012 if (!list_empty(&dentry->d_subdirs)) {
1013 spin_unlock(&this_parent->d_lock);
1014 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1015 this_parent = dentry;
1016 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1019 spin_unlock(&dentry->d_lock);
1022 * All done at this level ... ascend and resume the search.
1024 if (this_parent != parent) {
1026 struct dentry *child;
1028 tmp = this_parent->d_parent;
1030 spin_unlock(&this_parent->d_lock);
1031 child = this_parent;
1033 spin_lock(&this_parent->d_lock);
1034 /* might go back up the wrong parent if we have had a rename
1036 if (this_parent != child->d_parent ||
1037 (!locked && read_seqretry(&rename_lock, seq))) {
1038 spin_unlock(&this_parent->d_lock);
1043 next = child->d_u.d_child.next;
1046 spin_unlock(&this_parent->d_lock);
1047 if (!locked && read_seqretry(&rename_lock, seq))
1050 write_sequnlock(&rename_lock);
1051 return 0; /* No mount points found in tree */
1053 if (!locked && read_seqretry(&rename_lock, seq))
1056 write_sequnlock(&rename_lock);
1061 write_seqlock(&rename_lock);
1064 EXPORT_SYMBOL(have_submounts);
1067 * Search the dentry child list for the specified parent,
1068 * and move any unused dentries to the end of the unused
1069 * list for prune_dcache(). We descend to the next level
1070 * whenever the d_subdirs list is non-empty and continue
1073 * It returns zero iff there are no unused children,
1074 * otherwise it returns the number of children moved to
1075 * the end of the unused list. This may not be the total
1076 * number of unused children, because select_parent can
1077 * drop the lock and return early due to latency
1080 static int select_parent(struct dentry * parent)
1082 struct dentry *this_parent;
1083 struct list_head *next;
1088 seq = read_seqbegin(&rename_lock);
1090 this_parent = parent;
1091 spin_lock(&this_parent->d_lock);
1093 next = this_parent->d_subdirs.next;
1095 while (next != &this_parent->d_subdirs) {
1096 struct list_head *tmp = next;
1097 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1100 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1103 * move only zero ref count dentries to the end
1104 * of the unused list for prune_dcache
1106 if (!dentry->d_count) {
1107 dentry_lru_move_tail(dentry);
1110 dentry_lru_del(dentry);
1114 * We can return to the caller if we have found some (this
1115 * ensures forward progress). We'll be coming back to find
1118 if (found && need_resched()) {
1119 spin_unlock(&dentry->d_lock);
1124 * Descend a level if the d_subdirs list is non-empty.
1126 if (!list_empty(&dentry->d_subdirs)) {
1127 spin_unlock(&this_parent->d_lock);
1128 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1129 this_parent = dentry;
1130 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1134 spin_unlock(&dentry->d_lock);
1137 * All done at this level ... ascend and resume the search.
1139 if (this_parent != parent) {
1141 struct dentry *child;
1143 tmp = this_parent->d_parent;
1145 spin_unlock(&this_parent->d_lock);
1146 child = this_parent;
1148 spin_lock(&this_parent->d_lock);
1149 /* might go back up the wrong parent if we have had a rename
1151 if (this_parent != child->d_parent ||
1152 (!locked && read_seqretry(&rename_lock, seq))) {
1153 spin_unlock(&this_parent->d_lock);
1158 next = child->d_u.d_child.next;
1162 spin_unlock(&this_parent->d_lock);
1163 if (!locked && read_seqretry(&rename_lock, seq))
1166 write_sequnlock(&rename_lock);
1173 write_seqlock(&rename_lock);
1178 * shrink_dcache_parent - prune dcache
1179 * @parent: parent of entries to prune
1181 * Prune the dcache to remove unused children of the parent dentry.
1184 void shrink_dcache_parent(struct dentry * parent)
1186 struct super_block *sb = parent->d_sb;
1189 while ((found = select_parent(parent)) != 0)
1190 __shrink_dcache_sb(sb, &found, 0);
1192 EXPORT_SYMBOL(shrink_dcache_parent);
1195 * Scan `nr' dentries and return the number which remain.
1197 * We need to avoid reentering the filesystem if the caller is performing a
1198 * GFP_NOFS allocation attempt. One example deadlock is:
1200 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1201 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1202 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1204 * In this case we return -1 to tell the caller that we baled.
1206 static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1209 if (!(gfp_mask & __GFP_FS))
1214 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
1217 static struct shrinker dcache_shrinker = {
1218 .shrink = shrink_dcache_memory,
1219 .seeks = DEFAULT_SEEKS,
1223 * d_alloc - allocate a dcache entry
1224 * @parent: parent of entry to allocate
1225 * @name: qstr of the name
1227 * Allocates a dentry. It returns %NULL if there is insufficient memory
1228 * available. On a success the dentry is returned. The name passed in is
1229 * copied and the copy passed in may be reused after this call.
1232 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1234 struct dentry *dentry;
1237 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1241 if (name->len > DNAME_INLINE_LEN-1) {
1242 dname = kmalloc(name->len + 1, GFP_KERNEL);
1244 kmem_cache_free(dentry_cache, dentry);
1248 dname = dentry->d_iname;
1250 dentry->d_name.name = dname;
1252 dentry->d_name.len = name->len;
1253 dentry->d_name.hash = name->hash;
1254 memcpy(dname, name->name, name->len);
1255 dname[name->len] = 0;
1257 dentry->d_count = 1;
1258 dentry->d_flags = DCACHE_UNHASHED;
1259 spin_lock_init(&dentry->d_lock);
1260 seqcount_init(&dentry->d_seq);
1261 dentry->d_inode = NULL;
1262 dentry->d_parent = NULL;
1263 dentry->d_sb = NULL;
1264 dentry->d_op = NULL;
1265 dentry->d_fsdata = NULL;
1266 INIT_HLIST_NODE(&dentry->d_hash);
1267 INIT_LIST_HEAD(&dentry->d_lru);
1268 INIT_LIST_HEAD(&dentry->d_subdirs);
1269 INIT_LIST_HEAD(&dentry->d_alias);
1270 INIT_LIST_HEAD(&dentry->d_u.d_child);
1273 spin_lock(&parent->d_lock);
1275 * don't need child lock because it is not subject
1276 * to concurrency here
1278 __dget_dlock(parent);
1279 dentry->d_parent = parent;
1280 dentry->d_sb = parent->d_sb;
1281 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1282 spin_unlock(&parent->d_lock);
1285 this_cpu_inc(nr_dentry);
1289 EXPORT_SYMBOL(d_alloc);
1291 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1296 q.len = strlen(name);
1297 q.hash = full_name_hash(q.name, q.len);
1298 return d_alloc(parent, &q);
1300 EXPORT_SYMBOL(d_alloc_name);
1302 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1304 BUG_ON(dentry->d_op);
1305 BUG_ON(dentry->d_flags & (DCACHE_OP_HASH |
1307 DCACHE_OP_REVALIDATE |
1308 DCACHE_OP_DELETE ));
1313 dentry->d_flags |= DCACHE_OP_HASH;
1315 dentry->d_flags |= DCACHE_OP_COMPARE;
1316 if (op->d_revalidate)
1317 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1319 dentry->d_flags |= DCACHE_OP_DELETE;
1322 EXPORT_SYMBOL(d_set_d_op);
1324 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1326 spin_lock(&dentry->d_lock);
1328 list_add(&dentry->d_alias, &inode->i_dentry);
1329 dentry->d_inode = inode;
1330 dentry_rcuwalk_barrier(dentry);
1331 spin_unlock(&dentry->d_lock);
1332 fsnotify_d_instantiate(dentry, inode);
1336 * d_instantiate - fill in inode information for a dentry
1337 * @entry: dentry to complete
1338 * @inode: inode to attach to this dentry
1340 * Fill in inode information in the entry.
1342 * This turns negative dentries into productive full members
1345 * NOTE! This assumes that the inode count has been incremented
1346 * (or otherwise set) by the caller to indicate that it is now
1347 * in use by the dcache.
1350 void d_instantiate(struct dentry *entry, struct inode * inode)
1352 BUG_ON(!list_empty(&entry->d_alias));
1353 spin_lock(&dcache_inode_lock);
1354 __d_instantiate(entry, inode);
1355 spin_unlock(&dcache_inode_lock);
1356 security_d_instantiate(entry, inode);
1358 EXPORT_SYMBOL(d_instantiate);
1361 * d_instantiate_unique - instantiate a non-aliased dentry
1362 * @entry: dentry to instantiate
1363 * @inode: inode to attach to this dentry
1365 * Fill in inode information in the entry. On success, it returns NULL.
1366 * If an unhashed alias of "entry" already exists, then we return the
1367 * aliased dentry instead and drop one reference to inode.
1369 * Note that in order to avoid conflicts with rename() etc, the caller
1370 * had better be holding the parent directory semaphore.
1372 * This also assumes that the inode count has been incremented
1373 * (or otherwise set) by the caller to indicate that it is now
1374 * in use by the dcache.
1376 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1377 struct inode *inode)
1379 struct dentry *alias;
1380 int len = entry->d_name.len;
1381 const char *name = entry->d_name.name;
1382 unsigned int hash = entry->d_name.hash;
1385 __d_instantiate(entry, NULL);
1389 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1390 struct qstr *qstr = &alias->d_name;
1393 * Don't need alias->d_lock here, because aliases with
1394 * d_parent == entry->d_parent are not subject to name or
1395 * parent changes, because the parent inode i_mutex is held.
1397 if (qstr->hash != hash)
1399 if (alias->d_parent != entry->d_parent)
1401 if (qstr->len != len)
1403 if (memcmp(qstr->name, name, len))
1409 __d_instantiate(entry, inode);
1413 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1415 struct dentry *result;
1417 BUG_ON(!list_empty(&entry->d_alias));
1419 spin_lock(&dcache_inode_lock);
1420 result = __d_instantiate_unique(entry, inode);
1421 spin_unlock(&dcache_inode_lock);
1424 security_d_instantiate(entry, inode);
1428 BUG_ON(!d_unhashed(result));
1433 EXPORT_SYMBOL(d_instantiate_unique);
1436 * d_alloc_root - allocate root dentry
1437 * @root_inode: inode to allocate the root for
1439 * Allocate a root ("/") dentry for the inode given. The inode is
1440 * instantiated and returned. %NULL is returned if there is insufficient
1441 * memory or the inode passed is %NULL.
1444 struct dentry * d_alloc_root(struct inode * root_inode)
1446 struct dentry *res = NULL;
1449 static const struct qstr name = { .name = "/", .len = 1 };
1451 res = d_alloc(NULL, &name);
1453 res->d_sb = root_inode->i_sb;
1454 res->d_parent = res;
1455 d_instantiate(res, root_inode);
1460 EXPORT_SYMBOL(d_alloc_root);
1462 static inline struct hlist_head *d_hash(struct dentry *parent,
1465 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1466 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1467 return dentry_hashtable + (hash & D_HASHMASK);
1471 * d_obtain_alias - find or allocate a dentry for a given inode
1472 * @inode: inode to allocate the dentry for
1474 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1475 * similar open by handle operations. The returned dentry may be anonymous,
1476 * or may have a full name (if the inode was already in the cache).
1478 * When called on a directory inode, we must ensure that the inode only ever
1479 * has one dentry. If a dentry is found, that is returned instead of
1480 * allocating a new one.
1482 * On successful return, the reference to the inode has been transferred
1483 * to the dentry. In case of an error the reference on the inode is released.
1484 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1485 * be passed in and will be the error will be propagate to the return value,
1486 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1488 struct dentry *d_obtain_alias(struct inode *inode)
1490 static const struct qstr anonstring = { .name = "" };
1495 return ERR_PTR(-ESTALE);
1497 return ERR_CAST(inode);
1499 res = d_find_alias(inode);
1503 tmp = d_alloc(NULL, &anonstring);
1505 res = ERR_PTR(-ENOMEM);
1508 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1511 spin_lock(&dcache_inode_lock);
1512 res = __d_find_alias(inode, 0);
1514 spin_unlock(&dcache_inode_lock);
1519 /* attach a disconnected dentry */
1520 spin_lock(&tmp->d_lock);
1521 tmp->d_sb = inode->i_sb;
1522 tmp->d_inode = inode;
1523 tmp->d_flags |= DCACHE_DISCONNECTED;
1524 tmp->d_flags &= ~DCACHE_UNHASHED;
1525 list_add(&tmp->d_alias, &inode->i_dentry);
1526 spin_lock(&dcache_hash_lock);
1527 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1528 spin_unlock(&dcache_hash_lock);
1529 spin_unlock(&tmp->d_lock);
1530 spin_unlock(&dcache_inode_lock);
1538 EXPORT_SYMBOL(d_obtain_alias);
1541 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1542 * @inode: the inode which may have a disconnected dentry
1543 * @dentry: a negative dentry which we want to point to the inode.
1545 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1546 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1547 * and return it, else simply d_add the inode to the dentry and return NULL.
1549 * This is needed in the lookup routine of any filesystem that is exportable
1550 * (via knfsd) so that we can build dcache paths to directories effectively.
1552 * If a dentry was found and moved, then it is returned. Otherwise NULL
1553 * is returned. This matches the expected return value of ->lookup.
1556 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1558 struct dentry *new = NULL;
1560 if (inode && S_ISDIR(inode->i_mode)) {
1561 spin_lock(&dcache_inode_lock);
1562 new = __d_find_alias(inode, 1);
1564 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1565 spin_unlock(&dcache_inode_lock);
1566 security_d_instantiate(new, inode);
1567 d_move(new, dentry);
1570 /* already taking dcache_inode_lock, so d_add() by hand */
1571 __d_instantiate(dentry, inode);
1572 spin_unlock(&dcache_inode_lock);
1573 security_d_instantiate(dentry, inode);
1577 d_add(dentry, inode);
1580 EXPORT_SYMBOL(d_splice_alias);
1583 * d_add_ci - lookup or allocate new dentry with case-exact name
1584 * @inode: the inode case-insensitive lookup has found
1585 * @dentry: the negative dentry that was passed to the parent's lookup func
1586 * @name: the case-exact name to be associated with the returned dentry
1588 * This is to avoid filling the dcache with case-insensitive names to the
1589 * same inode, only the actual correct case is stored in the dcache for
1590 * case-insensitive filesystems.
1592 * For a case-insensitive lookup match and if the the case-exact dentry
1593 * already exists in in the dcache, use it and return it.
1595 * If no entry exists with the exact case name, allocate new dentry with
1596 * the exact case, and return the spliced entry.
1598 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1602 struct dentry *found;
1606 * First check if a dentry matching the name already exists,
1607 * if not go ahead and create it now.
1609 found = d_hash_and_lookup(dentry->d_parent, name);
1611 new = d_alloc(dentry->d_parent, name);
1617 found = d_splice_alias(inode, new);
1626 * If a matching dentry exists, and it's not negative use it.
1628 * Decrement the reference count to balance the iget() done
1631 if (found->d_inode) {
1632 if (unlikely(found->d_inode != inode)) {
1633 /* This can't happen because bad inodes are unhashed. */
1634 BUG_ON(!is_bad_inode(inode));
1635 BUG_ON(!is_bad_inode(found->d_inode));
1642 * Negative dentry: instantiate it unless the inode is a directory and
1643 * already has a dentry.
1645 spin_lock(&dcache_inode_lock);
1646 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1647 __d_instantiate(found, inode);
1648 spin_unlock(&dcache_inode_lock);
1649 security_d_instantiate(found, inode);
1654 * In case a directory already has a (disconnected) entry grab a
1655 * reference to it, move it in place and use it.
1657 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1659 spin_unlock(&dcache_inode_lock);
1660 security_d_instantiate(found, inode);
1668 return ERR_PTR(error);
1670 EXPORT_SYMBOL(d_add_ci);
1673 * __d_lookup_rcu - search for a dentry (racy, store-free)
1674 * @parent: parent dentry
1675 * @name: qstr of name we wish to find
1676 * @seq: returns d_seq value at the point where the dentry was found
1677 * @inode: returns dentry->d_inode when the inode was found valid.
1678 * Returns: dentry, or NULL
1680 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1681 * resolution (store-free path walking) design described in
1682 * Documentation/filesystems/path-lookup.txt.
1684 * This is not to be used outside core vfs.
1686 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1687 * held, and rcu_read_lock held. The returned dentry must not be stored into
1688 * without taking d_lock and checking d_seq sequence count against @seq
1691 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1694 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1695 * the returned dentry, so long as its parent's seqlock is checked after the
1696 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1697 * is formed, giving integrity down the path walk.
1699 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1700 unsigned *seq, struct inode **inode)
1702 unsigned int len = name->len;
1703 unsigned int hash = name->hash;
1704 const unsigned char *str = name->name;
1705 struct hlist_head *head = d_hash(parent, hash);
1706 struct hlist_node *node;
1707 struct dentry *dentry;
1710 * Note: There is significant duplication with __d_lookup_rcu which is
1711 * required to prevent single threaded performance regressions
1712 * especially on architectures where smp_rmb (in seqcounts) are costly.
1713 * Keep the two functions in sync.
1717 * The hash list is protected using RCU.
1719 * Carefully use d_seq when comparing a candidate dentry, to avoid
1720 * races with d_move().
1722 * It is possible that concurrent renames can mess up our list
1723 * walk here and result in missing our dentry, resulting in the
1724 * false-negative result. d_lookup() protects against concurrent
1725 * renames using rename_lock seqlock.
1727 * See Documentation/vfs/dcache-locking.txt for more details.
1729 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1734 if (dentry->d_name.hash != hash)
1738 *seq = read_seqcount_begin(&dentry->d_seq);
1739 if (dentry->d_parent != parent)
1741 if (d_unhashed(dentry))
1743 tlen = dentry->d_name.len;
1744 tname = dentry->d_name.name;
1745 i = dentry->d_inode;
1747 * This seqcount check is required to ensure name and
1748 * len are loaded atomically, so as not to walk off the
1749 * edge of memory when walking. If we could load this
1750 * atomically some other way, we could drop this check.
1752 if (read_seqcount_retry(&dentry->d_seq, *seq))
1754 if (parent->d_flags & DCACHE_OP_COMPARE) {
1755 if (parent->d_op->d_compare(parent, *inode,
1762 if (memcmp(tname, str, tlen))
1766 * No extra seqcount check is required after the name
1767 * compare. The caller must perform a seqcount check in
1768 * order to do anything useful with the returned dentry
1778 * d_lookup - search for a dentry
1779 * @parent: parent dentry
1780 * @name: qstr of name we wish to find
1781 * Returns: dentry, or NULL
1783 * d_lookup searches the children of the parent dentry for the name in
1784 * question. If the dentry is found its reference count is incremented and the
1785 * dentry is returned. The caller must use dput to free the entry when it has
1786 * finished using it. %NULL is returned if the dentry does not exist.
1788 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1790 struct dentry *dentry;
1794 seq = read_seqbegin(&rename_lock);
1795 dentry = __d_lookup(parent, name);
1798 } while (read_seqretry(&rename_lock, seq));
1801 EXPORT_SYMBOL(d_lookup);
1804 * __d_lookup - search for a dentry (racy)
1805 * @parent: parent dentry
1806 * @name: qstr of name we wish to find
1807 * Returns: dentry, or NULL
1809 * __d_lookup is like d_lookup, however it may (rarely) return a
1810 * false-negative result due to unrelated rename activity.
1812 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1813 * however it must be used carefully, eg. with a following d_lookup in
1814 * the case of failure.
1816 * __d_lookup callers must be commented.
1818 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1820 unsigned int len = name->len;
1821 unsigned int hash = name->hash;
1822 const unsigned char *str = name->name;
1823 struct hlist_head *head = d_hash(parent,hash);
1824 struct hlist_node *node;
1825 struct dentry *found = NULL;
1826 struct dentry *dentry;
1829 * Note: There is significant duplication with __d_lookup_rcu which is
1830 * required to prevent single threaded performance regressions
1831 * especially on architectures where smp_rmb (in seqcounts) are costly.
1832 * Keep the two functions in sync.
1836 * The hash list is protected using RCU.
1838 * Take d_lock when comparing a candidate dentry, to avoid races
1841 * It is possible that concurrent renames can mess up our list
1842 * walk here and result in missing our dentry, resulting in the
1843 * false-negative result. d_lookup() protects against concurrent
1844 * renames using rename_lock seqlock.
1846 * See Documentation/vfs/dcache-locking.txt for more details.
1850 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1854 if (dentry->d_name.hash != hash)
1857 spin_lock(&dentry->d_lock);
1858 if (dentry->d_parent != parent)
1860 if (d_unhashed(dentry))
1864 * It is safe to compare names since d_move() cannot
1865 * change the qstr (protected by d_lock).
1867 tlen = dentry->d_name.len;
1868 tname = dentry->d_name.name;
1869 if (parent->d_flags & DCACHE_OP_COMPARE) {
1870 if (parent->d_op->d_compare(parent, parent->d_inode,
1871 dentry, dentry->d_inode,
1877 if (memcmp(tname, str, tlen))
1883 spin_unlock(&dentry->d_lock);
1886 spin_unlock(&dentry->d_lock);
1894 * d_hash_and_lookup - hash the qstr then search for a dentry
1895 * @dir: Directory to search in
1896 * @name: qstr of name we wish to find
1898 * On hash failure or on lookup failure NULL is returned.
1900 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1902 struct dentry *dentry = NULL;
1905 * Check for a fs-specific hash function. Note that we must
1906 * calculate the standard hash first, as the d_op->d_hash()
1907 * routine may choose to leave the hash value unchanged.
1909 name->hash = full_name_hash(name->name, name->len);
1910 if (dir->d_flags & DCACHE_OP_HASH) {
1911 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1914 dentry = d_lookup(dir, name);
1920 * d_validate - verify dentry provided from insecure source (deprecated)
1921 * @dentry: The dentry alleged to be valid child of @dparent
1922 * @dparent: The parent dentry (known to be valid)
1924 * An insecure source has sent us a dentry, here we verify it and dget() it.
1925 * This is used by ncpfs in its readdir implementation.
1926 * Zero is returned in the dentry is invalid.
1928 * This function is slow for big directories, and deprecated, do not use it.
1930 int d_validate(struct dentry *dentry, struct dentry *dparent)
1932 struct dentry *child;
1934 spin_lock(&dparent->d_lock);
1935 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1936 if (dentry == child) {
1937 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1938 __dget_dlock(dentry);
1939 spin_unlock(&dentry->d_lock);
1940 spin_unlock(&dparent->d_lock);
1944 spin_unlock(&dparent->d_lock);
1948 EXPORT_SYMBOL(d_validate);
1951 * When a file is deleted, we have two options:
1952 * - turn this dentry into a negative dentry
1953 * - unhash this dentry and free it.
1955 * Usually, we want to just turn this into
1956 * a negative dentry, but if anybody else is
1957 * currently using the dentry or the inode
1958 * we can't do that and we fall back on removing
1959 * it from the hash queues and waiting for
1960 * it to be deleted later when it has no users
1964 * d_delete - delete a dentry
1965 * @dentry: The dentry to delete
1967 * Turn the dentry into a negative dentry if possible, otherwise
1968 * remove it from the hash queues so it can be deleted later
1971 void d_delete(struct dentry * dentry)
1975 * Are we the only user?
1978 spin_lock(&dentry->d_lock);
1979 isdir = S_ISDIR(dentry->d_inode->i_mode);
1980 if (dentry->d_count == 1) {
1981 if (!spin_trylock(&dcache_inode_lock)) {
1982 spin_unlock(&dentry->d_lock);
1986 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1987 dentry_unlink_inode(dentry);
1988 fsnotify_nameremove(dentry, isdir);
1992 if (!d_unhashed(dentry))
1995 spin_unlock(&dentry->d_lock);
1997 fsnotify_nameremove(dentry, isdir);
1999 EXPORT_SYMBOL(d_delete);
2001 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
2004 entry->d_flags &= ~DCACHE_UNHASHED;
2005 hlist_add_head_rcu(&entry->d_hash, list);
2008 static void _d_rehash(struct dentry * entry)
2010 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2014 * d_rehash - add an entry back to the hash
2015 * @entry: dentry to add to the hash
2017 * Adds a dentry to the hash according to its name.
2020 void d_rehash(struct dentry * entry)
2022 spin_lock(&entry->d_lock);
2023 spin_lock(&dcache_hash_lock);
2025 spin_unlock(&dcache_hash_lock);
2026 spin_unlock(&entry->d_lock);
2028 EXPORT_SYMBOL(d_rehash);
2031 * dentry_update_name_case - update case insensitive dentry with a new name
2032 * @dentry: dentry to be updated
2035 * Update a case insensitive dentry with new case of name.
2037 * dentry must have been returned by d_lookup with name @name. Old and new
2038 * name lengths must match (ie. no d_compare which allows mismatched name
2041 * Parent inode i_mutex must be held over d_lookup and into this call (to
2042 * keep renames and concurrent inserts, and readdir(2) away).
2044 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2046 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2047 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2049 spin_lock(&dentry->d_lock);
2050 write_seqcount_begin(&dentry->d_seq);
2051 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2052 write_seqcount_end(&dentry->d_seq);
2053 spin_unlock(&dentry->d_lock);
2055 EXPORT_SYMBOL(dentry_update_name_case);
2057 static void switch_names(struct dentry *dentry, struct dentry *target)
2059 if (dname_external(target)) {
2060 if (dname_external(dentry)) {
2062 * Both external: swap the pointers
2064 swap(target->d_name.name, dentry->d_name.name);
2067 * dentry:internal, target:external. Steal target's
2068 * storage and make target internal.
2070 memcpy(target->d_iname, dentry->d_name.name,
2071 dentry->d_name.len + 1);
2072 dentry->d_name.name = target->d_name.name;
2073 target->d_name.name = target->d_iname;
2076 if (dname_external(dentry)) {
2078 * dentry:external, target:internal. Give dentry's
2079 * storage to target and make dentry internal
2081 memcpy(dentry->d_iname, target->d_name.name,
2082 target->d_name.len + 1);
2083 target->d_name.name = dentry->d_name.name;
2084 dentry->d_name.name = dentry->d_iname;
2087 * Both are internal. Just copy target to dentry
2089 memcpy(dentry->d_iname, target->d_name.name,
2090 target->d_name.len + 1);
2091 dentry->d_name.len = target->d_name.len;
2095 swap(dentry->d_name.len, target->d_name.len);
2098 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2101 * XXXX: do we really need to take target->d_lock?
2103 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2104 spin_lock(&target->d_parent->d_lock);
2106 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2107 spin_lock(&dentry->d_parent->d_lock);
2108 spin_lock_nested(&target->d_parent->d_lock,
2109 DENTRY_D_LOCK_NESTED);
2111 spin_lock(&target->d_parent->d_lock);
2112 spin_lock_nested(&dentry->d_parent->d_lock,
2113 DENTRY_D_LOCK_NESTED);
2116 if (target < dentry) {
2117 spin_lock_nested(&target->d_lock, 2);
2118 spin_lock_nested(&dentry->d_lock, 3);
2120 spin_lock_nested(&dentry->d_lock, 2);
2121 spin_lock_nested(&target->d_lock, 3);
2125 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2126 struct dentry *target)
2128 if (target->d_parent != dentry->d_parent)
2129 spin_unlock(&dentry->d_parent->d_lock);
2130 if (target->d_parent != target)
2131 spin_unlock(&target->d_parent->d_lock);
2135 * When switching names, the actual string doesn't strictly have to
2136 * be preserved in the target - because we're dropping the target
2137 * anyway. As such, we can just do a simple memcpy() to copy over
2138 * the new name before we switch.
2140 * Note that we have to be a lot more careful about getting the hash
2141 * switched - we have to switch the hash value properly even if it
2142 * then no longer matches the actual (corrupted) string of the target.
2143 * The hash value has to match the hash queue that the dentry is on..
2146 * d_move - move a dentry
2147 * @dentry: entry to move
2148 * @target: new dentry
2150 * Update the dcache to reflect the move of a file name. Negative
2151 * dcache entries should not be moved in this way.
2153 void d_move(struct dentry * dentry, struct dentry * target)
2155 if (!dentry->d_inode)
2156 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2158 BUG_ON(d_ancestor(dentry, target));
2159 BUG_ON(d_ancestor(target, dentry));
2161 write_seqlock(&rename_lock);
2163 dentry_lock_for_move(dentry, target);
2165 write_seqcount_begin(&dentry->d_seq);
2166 write_seqcount_begin(&target->d_seq);
2168 /* Move the dentry to the target hash queue, if on different bucket */
2169 spin_lock(&dcache_hash_lock);
2170 if (!d_unhashed(dentry))
2171 hlist_del_rcu(&dentry->d_hash);
2172 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2173 spin_unlock(&dcache_hash_lock);
2175 /* Unhash the target: dput() will then get rid of it */
2176 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2179 list_del(&dentry->d_u.d_child);
2180 list_del(&target->d_u.d_child);
2182 /* Switch the names.. */
2183 switch_names(dentry, target);
2184 swap(dentry->d_name.hash, target->d_name.hash);
2186 /* ... and switch the parents */
2187 if (IS_ROOT(dentry)) {
2188 dentry->d_parent = target->d_parent;
2189 target->d_parent = target;
2190 INIT_LIST_HEAD(&target->d_u.d_child);
2192 swap(dentry->d_parent, target->d_parent);
2194 /* And add them back to the (new) parent lists */
2195 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2198 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2200 write_seqcount_end(&target->d_seq);
2201 write_seqcount_end(&dentry->d_seq);
2203 dentry_unlock_parents_for_move(dentry, target);
2204 spin_unlock(&target->d_lock);
2205 fsnotify_d_move(dentry);
2206 spin_unlock(&dentry->d_lock);
2207 write_sequnlock(&rename_lock);
2209 EXPORT_SYMBOL(d_move);
2212 * d_ancestor - search for an ancestor
2213 * @p1: ancestor dentry
2216 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2217 * an ancestor of p2, else NULL.
2219 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2223 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2224 if (p->d_parent == p1)
2231 * This helper attempts to cope with remotely renamed directories
2233 * It assumes that the caller is already holding
2234 * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock
2236 * Note: If ever the locking in lock_rename() changes, then please
2237 * remember to update this too...
2239 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
2240 __releases(dcache_inode_lock)
2242 struct mutex *m1 = NULL, *m2 = NULL;
2245 /* If alias and dentry share a parent, then no extra locks required */
2246 if (alias->d_parent == dentry->d_parent)
2249 /* Check for loops */
2250 ret = ERR_PTR(-ELOOP);
2251 if (d_ancestor(alias, dentry))
2254 /* See lock_rename() */
2255 ret = ERR_PTR(-EBUSY);
2256 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2258 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2259 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2261 m2 = &alias->d_parent->d_inode->i_mutex;
2263 d_move(alias, dentry);
2266 spin_unlock(&dcache_inode_lock);
2275 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2276 * named dentry in place of the dentry to be replaced.
2277 * returns with anon->d_lock held!
2279 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2281 struct dentry *dparent, *aparent;
2283 dentry_lock_for_move(anon, dentry);
2285 write_seqcount_begin(&dentry->d_seq);
2286 write_seqcount_begin(&anon->d_seq);
2288 dparent = dentry->d_parent;
2289 aparent = anon->d_parent;
2291 switch_names(dentry, anon);
2292 swap(dentry->d_name.hash, anon->d_name.hash);
2294 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2295 list_del(&dentry->d_u.d_child);
2296 if (!IS_ROOT(dentry))
2297 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2299 INIT_LIST_HEAD(&dentry->d_u.d_child);
2301 anon->d_parent = (dparent == dentry) ? anon : dparent;
2302 list_del(&anon->d_u.d_child);
2304 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2306 INIT_LIST_HEAD(&anon->d_u.d_child);
2308 write_seqcount_end(&dentry->d_seq);
2309 write_seqcount_end(&anon->d_seq);
2311 dentry_unlock_parents_for_move(anon, dentry);
2312 spin_unlock(&dentry->d_lock);
2314 /* anon->d_lock still locked, returns locked */
2315 anon->d_flags &= ~DCACHE_DISCONNECTED;
2319 * d_materialise_unique - introduce an inode into the tree
2320 * @dentry: candidate dentry
2321 * @inode: inode to bind to the dentry, to which aliases may be attached
2323 * Introduces an dentry into the tree, substituting an extant disconnected
2324 * root directory alias in its place if there is one
2326 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2328 struct dentry *actual;
2330 BUG_ON(!d_unhashed(dentry));
2334 __d_instantiate(dentry, NULL);
2339 spin_lock(&dcache_inode_lock);
2341 if (S_ISDIR(inode->i_mode)) {
2342 struct dentry *alias;
2344 /* Does an aliased dentry already exist? */
2345 alias = __d_find_alias(inode, 0);
2348 /* Is this an anonymous mountpoint that we could splice
2350 if (IS_ROOT(alias)) {
2351 __d_materialise_dentry(dentry, alias);
2355 /* Nope, but we must(!) avoid directory aliasing */
2356 actual = __d_unalias(dentry, alias);
2363 /* Add a unique reference */
2364 actual = __d_instantiate_unique(dentry, inode);
2368 BUG_ON(!d_unhashed(actual));
2370 spin_lock(&actual->d_lock);
2372 spin_lock(&dcache_hash_lock);
2374 spin_unlock(&dcache_hash_lock);
2375 spin_unlock(&actual->d_lock);
2376 spin_unlock(&dcache_inode_lock);
2378 if (actual == dentry) {
2379 security_d_instantiate(dentry, inode);
2386 EXPORT_SYMBOL_GPL(d_materialise_unique);
2388 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2392 return -ENAMETOOLONG;
2394 memcpy(*buffer, str, namelen);
2398 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2400 return prepend(buffer, buflen, name->name, name->len);
2404 * Prepend path string to a buffer
2406 * @path: the dentry/vfsmount to report
2407 * @root: root vfsmnt/dentry (may be modified by this function)
2408 * @buffer: pointer to the end of the buffer
2409 * @buflen: pointer to buffer length
2411 * Caller holds the rename_lock.
2413 * If path is not reachable from the supplied root, then the value of
2414 * root is changed (without modifying refcounts).
2416 static int prepend_path(const struct path *path, struct path *root,
2417 char **buffer, int *buflen)
2419 struct dentry *dentry = path->dentry;
2420 struct vfsmount *vfsmnt = path->mnt;
2424 br_read_lock(vfsmount_lock);
2425 while (dentry != root->dentry || vfsmnt != root->mnt) {
2426 struct dentry * parent;
2428 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2430 if (vfsmnt->mnt_parent == vfsmnt) {
2433 dentry = vfsmnt->mnt_mountpoint;
2434 vfsmnt = vfsmnt->mnt_parent;
2437 parent = dentry->d_parent;
2439 spin_lock(&dentry->d_lock);
2440 error = prepend_name(buffer, buflen, &dentry->d_name);
2441 spin_unlock(&dentry->d_lock);
2443 error = prepend(buffer, buflen, "/", 1);
2452 if (!error && !slash)
2453 error = prepend(buffer, buflen, "/", 1);
2455 br_read_unlock(vfsmount_lock);
2460 * Filesystems needing to implement special "root names"
2461 * should do so with ->d_dname()
2463 if (IS_ROOT(dentry) &&
2464 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2465 WARN(1, "Root dentry has weird name <%.*s>\n",
2466 (int) dentry->d_name.len, dentry->d_name.name);
2469 root->dentry = dentry;
2474 * __d_path - return the path of a dentry
2475 * @path: the dentry/vfsmount to report
2476 * @root: root vfsmnt/dentry (may be modified by this function)
2477 * @buf: buffer to return value in
2478 * @buflen: buffer length
2480 * Convert a dentry into an ASCII path name.
2482 * Returns a pointer into the buffer or an error code if the
2483 * path was too long.
2485 * "buflen" should be positive.
2487 * If path is not reachable from the supplied root, then the value of
2488 * root is changed (without modifying refcounts).
2490 char *__d_path(const struct path *path, struct path *root,
2491 char *buf, int buflen)
2493 char *res = buf + buflen;
2496 prepend(&res, &buflen, "\0", 1);
2497 write_seqlock(&rename_lock);
2498 error = prepend_path(path, root, &res, &buflen);
2499 write_sequnlock(&rename_lock);
2502 return ERR_PTR(error);
2507 * same as __d_path but appends "(deleted)" for unlinked files.
2509 static int path_with_deleted(const struct path *path, struct path *root,
2510 char **buf, int *buflen)
2512 prepend(buf, buflen, "\0", 1);
2513 if (d_unlinked(path->dentry)) {
2514 int error = prepend(buf, buflen, " (deleted)", 10);
2519 return prepend_path(path, root, buf, buflen);
2522 static int prepend_unreachable(char **buffer, int *buflen)
2524 return prepend(buffer, buflen, "(unreachable)", 13);
2528 * d_path - return the path of a dentry
2529 * @path: path to report
2530 * @buf: buffer to return value in
2531 * @buflen: buffer length
2533 * Convert a dentry into an ASCII path name. If the entry has been deleted
2534 * the string " (deleted)" is appended. Note that this is ambiguous.
2536 * Returns a pointer into the buffer or an error code if the path was
2537 * too long. Note: Callers should use the returned pointer, not the passed
2538 * in buffer, to use the name! The implementation often starts at an offset
2539 * into the buffer, and may leave 0 bytes at the start.
2541 * "buflen" should be positive.
2543 char *d_path(const struct path *path, char *buf, int buflen)
2545 char *res = buf + buflen;
2551 * We have various synthetic filesystems that never get mounted. On
2552 * these filesystems dentries are never used for lookup purposes, and
2553 * thus don't need to be hashed. They also don't need a name until a
2554 * user wants to identify the object in /proc/pid/fd/. The little hack
2555 * below allows us to generate a name for these objects on demand:
2557 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2558 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2560 get_fs_root(current->fs, &root);
2561 write_seqlock(&rename_lock);
2563 error = path_with_deleted(path, &tmp, &res, &buflen);
2565 res = ERR_PTR(error);
2566 write_sequnlock(&rename_lock);
2570 EXPORT_SYMBOL(d_path);
2573 * d_path_with_unreachable - return the path of a dentry
2574 * @path: path to report
2575 * @buf: buffer to return value in
2576 * @buflen: buffer length
2578 * The difference from d_path() is that this prepends "(unreachable)"
2579 * to paths which are unreachable from the current process' root.
2581 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2583 char *res = buf + buflen;
2588 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2589 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2591 get_fs_root(current->fs, &root);
2592 write_seqlock(&rename_lock);
2594 error = path_with_deleted(path, &tmp, &res, &buflen);
2595 if (!error && !path_equal(&tmp, &root))
2596 error = prepend_unreachable(&res, &buflen);
2597 write_sequnlock(&rename_lock);
2600 res = ERR_PTR(error);
2606 * Helper function for dentry_operations.d_dname() members
2608 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2609 const char *fmt, ...)
2615 va_start(args, fmt);
2616 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2619 if (sz > sizeof(temp) || sz > buflen)
2620 return ERR_PTR(-ENAMETOOLONG);
2622 buffer += buflen - sz;
2623 return memcpy(buffer, temp, sz);
2627 * Write full pathname from the root of the filesystem into the buffer.
2629 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2631 char *end = buf + buflen;
2634 prepend(&end, &buflen, "\0", 1);
2641 while (!IS_ROOT(dentry)) {
2642 struct dentry *parent = dentry->d_parent;
2646 spin_lock(&dentry->d_lock);
2647 error = prepend_name(&end, &buflen, &dentry->d_name);
2648 spin_unlock(&dentry->d_lock);
2649 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2657 return ERR_PTR(-ENAMETOOLONG);
2660 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2664 write_seqlock(&rename_lock);
2665 retval = __dentry_path(dentry, buf, buflen);
2666 write_sequnlock(&rename_lock);
2670 EXPORT_SYMBOL(dentry_path_raw);
2672 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2677 write_seqlock(&rename_lock);
2678 if (d_unlinked(dentry)) {
2680 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2684 retval = __dentry_path(dentry, buf, buflen);
2685 write_sequnlock(&rename_lock);
2686 if (!IS_ERR(retval) && p)
2687 *p = '/'; /* restore '/' overriden with '\0' */
2690 return ERR_PTR(-ENAMETOOLONG);
2694 * NOTE! The user-level library version returns a
2695 * character pointer. The kernel system call just
2696 * returns the length of the buffer filled (which
2697 * includes the ending '\0' character), or a negative
2698 * error value. So libc would do something like
2700 * char *getcwd(char * buf, size_t size)
2704 * retval = sys_getcwd(buf, size);
2711 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2714 struct path pwd, root;
2715 char *page = (char *) __get_free_page(GFP_USER);
2720 get_fs_root_and_pwd(current->fs, &root, &pwd);
2723 write_seqlock(&rename_lock);
2724 if (!d_unlinked(pwd.dentry)) {
2726 struct path tmp = root;
2727 char *cwd = page + PAGE_SIZE;
2728 int buflen = PAGE_SIZE;
2730 prepend(&cwd, &buflen, "\0", 1);
2731 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2732 write_sequnlock(&rename_lock);
2737 /* Unreachable from current root */
2738 if (!path_equal(&tmp, &root)) {
2739 error = prepend_unreachable(&cwd, &buflen);
2745 len = PAGE_SIZE + page - cwd;
2748 if (copy_to_user(buf, cwd, len))
2752 write_sequnlock(&rename_lock);
2758 free_page((unsigned long) page);
2763 * Test whether new_dentry is a subdirectory of old_dentry.
2765 * Trivially implemented using the dcache structure
2769 * is_subdir - is new dentry a subdirectory of old_dentry
2770 * @new_dentry: new dentry
2771 * @old_dentry: old dentry
2773 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2774 * Returns 0 otherwise.
2775 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2778 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2783 if (new_dentry == old_dentry)
2787 /* for restarting inner loop in case of seq retry */
2788 seq = read_seqbegin(&rename_lock);
2790 * Need rcu_readlock to protect against the d_parent trashing
2794 if (d_ancestor(old_dentry, new_dentry))
2799 } while (read_seqretry(&rename_lock, seq));
2804 int path_is_under(struct path *path1, struct path *path2)
2806 struct vfsmount *mnt = path1->mnt;
2807 struct dentry *dentry = path1->dentry;
2810 br_read_lock(vfsmount_lock);
2811 if (mnt != path2->mnt) {
2813 if (mnt->mnt_parent == mnt) {
2814 br_read_unlock(vfsmount_lock);
2817 if (mnt->mnt_parent == path2->mnt)
2819 mnt = mnt->mnt_parent;
2821 dentry = mnt->mnt_mountpoint;
2823 res = is_subdir(dentry, path2->dentry);
2824 br_read_unlock(vfsmount_lock);
2827 EXPORT_SYMBOL(path_is_under);
2829 void d_genocide(struct dentry *root)
2831 struct dentry *this_parent;
2832 struct list_head *next;
2836 seq = read_seqbegin(&rename_lock);
2839 spin_lock(&this_parent->d_lock);
2841 next = this_parent->d_subdirs.next;
2843 while (next != &this_parent->d_subdirs) {
2844 struct list_head *tmp = next;
2845 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2848 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2849 if (d_unhashed(dentry) || !dentry->d_inode) {
2850 spin_unlock(&dentry->d_lock);
2853 if (!list_empty(&dentry->d_subdirs)) {
2854 spin_unlock(&this_parent->d_lock);
2855 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2856 this_parent = dentry;
2857 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2860 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2861 dentry->d_flags |= DCACHE_GENOCIDE;
2864 spin_unlock(&dentry->d_lock);
2866 if (this_parent != root) {
2868 struct dentry *child;
2870 tmp = this_parent->d_parent;
2871 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2872 this_parent->d_flags |= DCACHE_GENOCIDE;
2873 this_parent->d_count--;
2876 spin_unlock(&this_parent->d_lock);
2877 child = this_parent;
2879 spin_lock(&this_parent->d_lock);
2880 /* might go back up the wrong parent if we have had a rename
2882 if (this_parent != child->d_parent ||
2883 (!locked && read_seqretry(&rename_lock, seq))) {
2884 spin_unlock(&this_parent->d_lock);
2889 next = child->d_u.d_child.next;
2892 spin_unlock(&this_parent->d_lock);
2893 if (!locked && read_seqretry(&rename_lock, seq))
2896 write_sequnlock(&rename_lock);
2901 write_seqlock(&rename_lock);
2906 * find_inode_number - check for dentry with name
2907 * @dir: directory to check
2908 * @name: Name to find.
2910 * Check whether a dentry already exists for the given name,
2911 * and return the inode number if it has an inode. Otherwise
2914 * This routine is used to post-process directory listings for
2915 * filesystems using synthetic inode numbers, and is necessary
2916 * to keep getcwd() working.
2919 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2921 struct dentry * dentry;
2924 dentry = d_hash_and_lookup(dir, name);
2926 if (dentry->d_inode)
2927 ino = dentry->d_inode->i_ino;
2932 EXPORT_SYMBOL(find_inode_number);
2934 static __initdata unsigned long dhash_entries;
2935 static int __init set_dhash_entries(char *str)
2939 dhash_entries = simple_strtoul(str, &str, 0);
2942 __setup("dhash_entries=", set_dhash_entries);
2944 static void __init dcache_init_early(void)
2948 /* If hashes are distributed across NUMA nodes, defer
2949 * hash allocation until vmalloc space is available.
2955 alloc_large_system_hash("Dentry cache",
2956 sizeof(struct hlist_head),
2964 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2965 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2968 static void __init dcache_init(void)
2973 * A constructor could be added for stable state like the lists,
2974 * but it is probably not worth it because of the cache nature
2977 dentry_cache = KMEM_CACHE(dentry,
2978 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2980 register_shrinker(&dcache_shrinker);
2982 /* Hash may have been set up in dcache_init_early */
2987 alloc_large_system_hash("Dentry cache",
2988 sizeof(struct hlist_head),
2996 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2997 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
3000 /* SLAB cache for __getname() consumers */
3001 struct kmem_cache *names_cachep __read_mostly;
3002 EXPORT_SYMBOL(names_cachep);
3004 EXPORT_SYMBOL(d_genocide);
3006 void __init vfs_caches_init_early(void)
3008 dcache_init_early();
3012 void __init vfs_caches_init(unsigned long mempages)
3014 unsigned long reserve;
3016 /* Base hash sizes on available memory, with a reserve equal to
3017 150% of current kernel size */
3019 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3020 mempages -= reserve;
3022 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3023 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3027 files_init(mempages);