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>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
44 * dcache->d_inode->i_lock protects:
45 * - i_dentry, d_alias, d_inode of aliases
46 * dcache_hash_bucket lock protects:
47 * - the dcache hash table
48 * s_anon bl list spinlock protects:
49 * - the s_anon list (see __d_drop)
50 * dcache_lru_lock protects:
51 * - the dcache lru lists and counters
58 * - d_parent and d_subdirs
59 * - childrens' d_child and d_parent
63 * dentry->d_inode->i_lock
66 * dcache_hash_bucket lock
69 * If there is an ancestor relationship:
70 * dentry->d_parent->...->d_parent->d_lock
72 * dentry->d_parent->d_lock
75 * If no ancestor relationship:
76 * if (dentry1 < dentry2)
80 int sysctl_vfs_cache_pressure __read_mostly = 100;
81 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
84 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
86 EXPORT_SYMBOL(rename_lock);
88 static struct kmem_cache *dentry_cache __read_mostly;
91 * This is the single most critical data structure when it comes
92 * to the dcache: the hashtable for lookups. Somebody should try
93 * to make this good - I've just made it work.
95 * This hash-function tries to avoid losing too many bits of hash
96 * information, yet avoid using a prime hash-size or similar.
98 #define D_HASHBITS d_hash_shift
99 #define D_HASHMASK d_hash_mask
101 static unsigned int d_hash_mask __read_mostly;
102 static unsigned int d_hash_shift __read_mostly;
104 static struct hlist_bl_head *dentry_hashtable __read_mostly;
106 static inline struct hlist_bl_head *d_hash(struct dentry *parent,
109 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
110 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
111 return dentry_hashtable + (hash & D_HASHMASK);
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat = {
119 static DEFINE_PER_CPU(unsigned int, nr_dentry);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
122 static int get_nr_dentry(void)
126 for_each_possible_cpu(i)
127 sum += per_cpu(nr_dentry, i);
128 return sum < 0 ? 0 : sum;
131 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
132 size_t *lenp, loff_t *ppos)
134 dentry_stat.nr_dentry = get_nr_dentry();
135 return proc_dointvec(table, write, buffer, lenp, ppos);
139 static void __d_free(struct rcu_head *head)
141 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
143 WARN_ON(!list_empty(&dentry->d_alias));
144 if (dname_external(dentry))
145 kfree(dentry->d_name.name);
146 kmem_cache_free(dentry_cache, dentry);
152 static void d_free(struct dentry *dentry)
154 BUG_ON(dentry->d_count);
155 this_cpu_dec(nr_dentry);
156 if (dentry->d_op && dentry->d_op->d_release)
157 dentry->d_op->d_release(dentry);
159 /* if dentry was never visible to RCU, immediate free is OK */
160 if (!(dentry->d_flags & DCACHE_RCUACCESS))
161 __d_free(&dentry->d_u.d_rcu);
163 call_rcu(&dentry->d_u.d_rcu, __d_free);
167 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
168 * @dentry: the target dentry
169 * After this call, in-progress rcu-walk path lookup will fail. This
170 * should be called after unhashing, and after changing d_inode (if
171 * the dentry has not already been unhashed).
173 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
175 assert_spin_locked(&dentry->d_lock);
176 /* Go through a barrier */
177 write_seqcount_barrier(&dentry->d_seq);
181 * Release the dentry's inode, using the filesystem
182 * d_iput() operation if defined. Dentry has no refcount
185 static void dentry_iput(struct dentry * dentry)
186 __releases(dentry->d_lock)
187 __releases(dentry->d_inode->i_lock)
189 struct inode *inode = dentry->d_inode;
191 dentry->d_inode = NULL;
192 list_del_init(&dentry->d_alias);
193 spin_unlock(&dentry->d_lock);
194 spin_unlock(&inode->i_lock);
196 fsnotify_inoderemove(inode);
197 if (dentry->d_op && dentry->d_op->d_iput)
198 dentry->d_op->d_iput(dentry, inode);
202 spin_unlock(&dentry->d_lock);
207 * Release the dentry's inode, using the filesystem
208 * d_iput() operation if defined. dentry remains in-use.
210 static void dentry_unlink_inode(struct dentry * dentry)
211 __releases(dentry->d_lock)
212 __releases(dentry->d_inode->i_lock)
214 struct inode *inode = dentry->d_inode;
215 dentry->d_inode = NULL;
216 list_del_init(&dentry->d_alias);
217 dentry_rcuwalk_barrier(dentry);
218 spin_unlock(&dentry->d_lock);
219 spin_unlock(&inode->i_lock);
221 fsnotify_inoderemove(inode);
222 if (dentry->d_op && dentry->d_op->d_iput)
223 dentry->d_op->d_iput(dentry, inode);
229 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
231 static void dentry_lru_add(struct dentry *dentry)
233 if (list_empty(&dentry->d_lru)) {
234 spin_lock(&dcache_lru_lock);
235 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
236 dentry->d_sb->s_nr_dentry_unused++;
237 dentry_stat.nr_unused++;
238 spin_unlock(&dcache_lru_lock);
242 static void __dentry_lru_del(struct dentry *dentry)
244 list_del_init(&dentry->d_lru);
245 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
246 dentry->d_sb->s_nr_dentry_unused--;
247 dentry_stat.nr_unused--;
251 * Remove a dentry with references from the LRU.
253 static void dentry_lru_del(struct dentry *dentry)
255 if (!list_empty(&dentry->d_lru)) {
256 spin_lock(&dcache_lru_lock);
257 __dentry_lru_del(dentry);
258 spin_unlock(&dcache_lru_lock);
263 * Remove a dentry that is unreferenced and about to be pruned
264 * (unhashed and destroyed) from the LRU, and inform the file system.
265 * This wrapper should be called _prior_ to unhashing a victim dentry.
267 static void dentry_lru_prune(struct dentry *dentry)
269 if (!list_empty(&dentry->d_lru)) {
270 if (dentry->d_flags & DCACHE_OP_PRUNE)
271 dentry->d_op->d_prune(dentry);
273 spin_lock(&dcache_lru_lock);
274 __dentry_lru_del(dentry);
275 spin_unlock(&dcache_lru_lock);
279 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
281 spin_lock(&dcache_lru_lock);
282 if (list_empty(&dentry->d_lru)) {
283 list_add_tail(&dentry->d_lru, list);
284 dentry->d_sb->s_nr_dentry_unused++;
285 dentry_stat.nr_unused++;
287 list_move_tail(&dentry->d_lru, list);
289 spin_unlock(&dcache_lru_lock);
293 * d_kill - kill dentry and return parent
294 * @dentry: dentry to kill
295 * @parent: parent dentry
297 * The dentry must already be unhashed and removed from the LRU.
299 * If this is the root of the dentry tree, return NULL.
301 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
304 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
305 __releases(dentry->d_lock)
306 __releases(parent->d_lock)
307 __releases(dentry->d_inode->i_lock)
309 list_del(&dentry->d_u.d_child);
311 * Inform try_to_ascend() that we are no longer attached to the
314 dentry->d_flags |= DCACHE_DENTRY_KILLED;
316 spin_unlock(&parent->d_lock);
319 * dentry_iput drops the locks, at which point nobody (except
320 * transient RCU lookups) can reach this dentry.
327 * Unhash a dentry without inserting an RCU walk barrier or checking that
328 * dentry->d_lock is locked. The caller must take care of that, if
331 static void __d_shrink(struct dentry *dentry)
333 if (!d_unhashed(dentry)) {
334 struct hlist_bl_head *b;
335 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
336 b = &dentry->d_sb->s_anon;
338 b = d_hash(dentry->d_parent, dentry->d_name.hash);
341 __hlist_bl_del(&dentry->d_hash);
342 dentry->d_hash.pprev = NULL;
348 * d_drop - drop a dentry
349 * @dentry: dentry to drop
351 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
352 * be found through a VFS lookup any more. Note that this is different from
353 * deleting the dentry - d_delete will try to mark the dentry negative if
354 * possible, giving a successful _negative_ lookup, while d_drop will
355 * just make the cache lookup fail.
357 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
358 * reason (NFS timeouts or autofs deletes).
360 * __d_drop requires dentry->d_lock.
362 void __d_drop(struct dentry *dentry)
364 if (!d_unhashed(dentry)) {
366 dentry_rcuwalk_barrier(dentry);
369 EXPORT_SYMBOL(__d_drop);
371 void d_drop(struct dentry *dentry)
373 spin_lock(&dentry->d_lock);
375 spin_unlock(&dentry->d_lock);
377 EXPORT_SYMBOL(d_drop);
380 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
381 * @dentry: dentry to drop
383 * This is called when we do a lookup on a placeholder dentry that needed to be
384 * looked up. The dentry should have been hashed in order for it to be found by
385 * the lookup code, but now needs to be unhashed while we do the actual lookup
386 * and clear the DCACHE_NEED_LOOKUP flag.
388 void d_clear_need_lookup(struct dentry *dentry)
390 spin_lock(&dentry->d_lock);
392 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
393 spin_unlock(&dentry->d_lock);
395 EXPORT_SYMBOL(d_clear_need_lookup);
398 * Finish off a dentry we've decided to kill.
399 * dentry->d_lock must be held, returns with it unlocked.
400 * If ref is non-zero, then decrement the refcount too.
401 * Returns dentry requiring refcount drop, or NULL if we're done.
403 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
404 __releases(dentry->d_lock)
407 struct dentry *parent;
409 inode = dentry->d_inode;
410 if (inode && !spin_trylock(&inode->i_lock)) {
412 spin_unlock(&dentry->d_lock);
414 return dentry; /* try again with same dentry */
419 parent = dentry->d_parent;
420 if (parent && !spin_trylock(&parent->d_lock)) {
422 spin_unlock(&inode->i_lock);
429 * if dentry was on the d_lru list delete it from there.
430 * inform the fs via d_prune that this dentry is about to be
431 * unhashed and destroyed.
433 dentry_lru_prune(dentry);
434 /* if it was on the hash then remove it */
436 return d_kill(dentry, parent);
442 * This is complicated by the fact that we do not want to put
443 * dentries that are no longer on any hash chain on the unused
444 * list: we'd much rather just get rid of them immediately.
446 * However, that implies that we have to traverse the dentry
447 * tree upwards to the parents which might _also_ now be
448 * scheduled for deletion (it may have been only waiting for
449 * its last child to go away).
451 * This tail recursion is done by hand as we don't want to depend
452 * on the compiler to always get this right (gcc generally doesn't).
453 * Real recursion would eat up our stack space.
457 * dput - release a dentry
458 * @dentry: dentry to release
460 * Release a dentry. This will drop the usage count and if appropriate
461 * call the dentry unlink method as well as removing it from the queues and
462 * releasing its resources. If the parent dentries were scheduled for release
463 * they too may now get deleted.
465 void dput(struct dentry *dentry)
471 if (dentry->d_count == 1)
473 spin_lock(&dentry->d_lock);
474 BUG_ON(!dentry->d_count);
475 if (dentry->d_count > 1) {
477 spin_unlock(&dentry->d_lock);
481 if (dentry->d_flags & DCACHE_OP_DELETE) {
482 if (dentry->d_op->d_delete(dentry))
486 /* Unreachable? Get rid of it */
487 if (d_unhashed(dentry))
491 * If this dentry needs lookup, don't set the referenced flag so that it
492 * is more likely to be cleaned up by the dcache shrinker in case of
495 if (!d_need_lookup(dentry))
496 dentry->d_flags |= DCACHE_REFERENCED;
497 dentry_lru_add(dentry);
500 spin_unlock(&dentry->d_lock);
504 dentry = dentry_kill(dentry, 1);
511 * d_invalidate - invalidate a dentry
512 * @dentry: dentry to invalidate
514 * Try to invalidate the dentry if it turns out to be
515 * possible. If there are other dentries that can be
516 * reached through this one we can't delete it and we
517 * return -EBUSY. On success we return 0.
522 int d_invalidate(struct dentry * dentry)
525 * If it's already been dropped, return OK.
527 spin_lock(&dentry->d_lock);
528 if (d_unhashed(dentry)) {
529 spin_unlock(&dentry->d_lock);
533 * Check whether to do a partial shrink_dcache
534 * to get rid of unused child entries.
536 if (!list_empty(&dentry->d_subdirs)) {
537 spin_unlock(&dentry->d_lock);
538 shrink_dcache_parent(dentry);
539 spin_lock(&dentry->d_lock);
543 * Somebody else still using it?
545 * If it's a directory, we can't drop it
546 * for fear of somebody re-populating it
547 * with children (even though dropping it
548 * would make it unreachable from the root,
549 * we might still populate it if it was a
550 * working directory or similar).
551 * We also need to leave mountpoints alone,
554 if (dentry->d_count > 1 && dentry->d_inode) {
555 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
556 spin_unlock(&dentry->d_lock);
562 spin_unlock(&dentry->d_lock);
565 EXPORT_SYMBOL(d_invalidate);
567 /* This must be called with d_lock held */
568 static inline void __dget_dlock(struct dentry *dentry)
573 static inline void __dget(struct dentry *dentry)
575 spin_lock(&dentry->d_lock);
576 __dget_dlock(dentry);
577 spin_unlock(&dentry->d_lock);
580 struct dentry *dget_parent(struct dentry *dentry)
586 * Don't need rcu_dereference because we re-check it was correct under
590 ret = dentry->d_parent;
591 spin_lock(&ret->d_lock);
592 if (unlikely(ret != dentry->d_parent)) {
593 spin_unlock(&ret->d_lock);
598 BUG_ON(!ret->d_count);
600 spin_unlock(&ret->d_lock);
603 EXPORT_SYMBOL(dget_parent);
606 * d_find_alias - grab a hashed alias of inode
607 * @inode: inode in question
608 * @want_discon: flag, used by d_splice_alias, to request
609 * that only a DISCONNECTED alias be returned.
611 * If inode has a hashed alias, or is a directory and has any alias,
612 * acquire the reference to alias and return it. Otherwise return NULL.
613 * Notice that if inode is a directory there can be only one alias and
614 * it can be unhashed only if it has no children, or if it is the root
617 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
618 * any other hashed alias over that one unless @want_discon is set,
619 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
621 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
623 struct dentry *alias, *discon_alias;
627 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
628 spin_lock(&alias->d_lock);
629 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
630 if (IS_ROOT(alias) &&
631 (alias->d_flags & DCACHE_DISCONNECTED)) {
632 discon_alias = alias;
633 } else if (!want_discon) {
635 spin_unlock(&alias->d_lock);
639 spin_unlock(&alias->d_lock);
642 alias = discon_alias;
643 spin_lock(&alias->d_lock);
644 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
645 if (IS_ROOT(alias) &&
646 (alias->d_flags & DCACHE_DISCONNECTED)) {
648 spin_unlock(&alias->d_lock);
652 spin_unlock(&alias->d_lock);
658 struct dentry *d_find_alias(struct inode *inode)
660 struct dentry *de = NULL;
662 if (!list_empty(&inode->i_dentry)) {
663 spin_lock(&inode->i_lock);
664 de = __d_find_alias(inode, 0);
665 spin_unlock(&inode->i_lock);
669 EXPORT_SYMBOL(d_find_alias);
672 * Try to kill dentries associated with this inode.
673 * WARNING: you must own a reference to inode.
675 void d_prune_aliases(struct inode *inode)
677 struct dentry *dentry;
679 spin_lock(&inode->i_lock);
680 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
681 spin_lock(&dentry->d_lock);
682 if (!dentry->d_count) {
683 __dget_dlock(dentry);
685 spin_unlock(&dentry->d_lock);
686 spin_unlock(&inode->i_lock);
690 spin_unlock(&dentry->d_lock);
692 spin_unlock(&inode->i_lock);
694 EXPORT_SYMBOL(d_prune_aliases);
697 * Try to throw away a dentry - free the inode, dput the parent.
698 * Requires dentry->d_lock is held, and dentry->d_count == 0.
699 * Releases dentry->d_lock.
701 * This may fail if locks cannot be acquired no problem, just try again.
703 static void try_prune_one_dentry(struct dentry *dentry)
704 __releases(dentry->d_lock)
706 struct dentry *parent;
708 parent = dentry_kill(dentry, 0);
710 * If dentry_kill returns NULL, we have nothing more to do.
711 * if it returns the same dentry, trylocks failed. In either
712 * case, just loop again.
714 * Otherwise, we need to prune ancestors too. This is necessary
715 * to prevent quadratic behavior of shrink_dcache_parent(), but
716 * is also expected to be beneficial in reducing dentry cache
721 if (parent == dentry)
724 /* Prune ancestors. */
727 spin_lock(&dentry->d_lock);
728 if (dentry->d_count > 1) {
730 spin_unlock(&dentry->d_lock);
733 dentry = dentry_kill(dentry, 1);
737 static void shrink_dentry_list(struct list_head *list)
739 struct dentry *dentry;
743 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
744 if (&dentry->d_lru == list)
746 spin_lock(&dentry->d_lock);
747 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
748 spin_unlock(&dentry->d_lock);
753 * We found an inuse dentry which was not removed from
754 * the LRU because of laziness during lookup. Do not free
755 * it - just keep it off the LRU list.
757 if (dentry->d_count) {
758 dentry_lru_del(dentry);
759 spin_unlock(&dentry->d_lock);
765 try_prune_one_dentry(dentry);
773 * prune_dcache_sb - shrink the dcache
775 * @count: number of entries to try to free
777 * Attempt to shrink the superblock dcache LRU by @count entries. This is
778 * done when we need more memory an called from the superblock shrinker
781 * This function may fail to free any resources if all the dentries are in
784 void prune_dcache_sb(struct super_block *sb, int count)
786 struct dentry *dentry;
787 LIST_HEAD(referenced);
791 spin_lock(&dcache_lru_lock);
792 while (!list_empty(&sb->s_dentry_lru)) {
793 dentry = list_entry(sb->s_dentry_lru.prev,
794 struct dentry, d_lru);
795 BUG_ON(dentry->d_sb != sb);
797 if (!spin_trylock(&dentry->d_lock)) {
798 spin_unlock(&dcache_lru_lock);
803 if (dentry->d_flags & DCACHE_REFERENCED) {
804 dentry->d_flags &= ~DCACHE_REFERENCED;
805 list_move(&dentry->d_lru, &referenced);
806 spin_unlock(&dentry->d_lock);
808 list_move_tail(&dentry->d_lru, &tmp);
809 dentry->d_flags |= DCACHE_SHRINK_LIST;
810 spin_unlock(&dentry->d_lock);
814 cond_resched_lock(&dcache_lru_lock);
816 if (!list_empty(&referenced))
817 list_splice(&referenced, &sb->s_dentry_lru);
818 spin_unlock(&dcache_lru_lock);
820 shrink_dentry_list(&tmp);
824 * shrink_dcache_sb - shrink dcache for a superblock
827 * Shrink the dcache for the specified super block. This is used to free
828 * the dcache before unmounting a file system.
830 void shrink_dcache_sb(struct super_block *sb)
834 spin_lock(&dcache_lru_lock);
835 while (!list_empty(&sb->s_dentry_lru)) {
836 list_splice_init(&sb->s_dentry_lru, &tmp);
837 spin_unlock(&dcache_lru_lock);
838 shrink_dentry_list(&tmp);
839 spin_lock(&dcache_lru_lock);
841 spin_unlock(&dcache_lru_lock);
843 EXPORT_SYMBOL(shrink_dcache_sb);
846 * destroy a single subtree of dentries for unmount
847 * - see the comments on shrink_dcache_for_umount() for a description of the
850 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
852 struct dentry *parent;
854 BUG_ON(!IS_ROOT(dentry));
857 /* descend to the first leaf in the current subtree */
858 while (!list_empty(&dentry->d_subdirs))
859 dentry = list_entry(dentry->d_subdirs.next,
860 struct dentry, d_u.d_child);
862 /* consume the dentries from this leaf up through its parents
863 * until we find one with children or run out altogether */
868 * remove the dentry from the lru, and inform
869 * the fs that this dentry is about to be
870 * unhashed and destroyed.
872 dentry_lru_prune(dentry);
875 if (dentry->d_count != 0) {
877 "BUG: Dentry %p{i=%lx,n=%s}"
879 " [unmount of %s %s]\n",
882 dentry->d_inode->i_ino : 0UL,
885 dentry->d_sb->s_type->name,
890 if (IS_ROOT(dentry)) {
892 list_del(&dentry->d_u.d_child);
894 parent = dentry->d_parent;
896 list_del(&dentry->d_u.d_child);
899 inode = dentry->d_inode;
901 dentry->d_inode = NULL;
902 list_del_init(&dentry->d_alias);
903 if (dentry->d_op && dentry->d_op->d_iput)
904 dentry->d_op->d_iput(dentry, inode);
911 /* finished when we fall off the top of the tree,
912 * otherwise we ascend to the parent and move to the
913 * next sibling if there is one */
917 } while (list_empty(&dentry->d_subdirs));
919 dentry = list_entry(dentry->d_subdirs.next,
920 struct dentry, d_u.d_child);
925 * destroy the dentries attached to a superblock on unmounting
926 * - we don't need to use dentry->d_lock because:
927 * - the superblock is detached from all mountings and open files, so the
928 * dentry trees will not be rearranged by the VFS
929 * - s_umount is write-locked, so the memory pressure shrinker will ignore
930 * any dentries belonging to this superblock that it comes across
931 * - the filesystem itself is no longer permitted to rearrange the dentries
934 void shrink_dcache_for_umount(struct super_block *sb)
936 struct dentry *dentry;
938 if (down_read_trylock(&sb->s_umount))
944 shrink_dcache_for_umount_subtree(dentry);
946 while (!hlist_bl_empty(&sb->s_anon)) {
947 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
948 shrink_dcache_for_umount_subtree(dentry);
953 * This tries to ascend one level of parenthood, but
954 * we can race with renaming, so we need to re-check
955 * the parenthood after dropping the lock and check
956 * that the sequence number still matches.
958 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
960 struct dentry *new = old->d_parent;
963 spin_unlock(&old->d_lock);
964 spin_lock(&new->d_lock);
967 * might go back up the wrong parent if we have had a rename
970 if (new != old->d_parent ||
971 (old->d_flags & DCACHE_DENTRY_KILLED) ||
972 (!locked && read_seqretry(&rename_lock, seq))) {
973 spin_unlock(&new->d_lock);
982 * Search for at least 1 mount point in the dentry's subdirs.
983 * We descend to the next level whenever the d_subdirs
984 * list is non-empty and continue searching.
988 * have_submounts - check for mounts over a dentry
989 * @parent: dentry to check.
991 * Return true if the parent or its subdirectories contain
994 int have_submounts(struct dentry *parent)
996 struct dentry *this_parent;
997 struct list_head *next;
1001 seq = read_seqbegin(&rename_lock);
1003 this_parent = parent;
1005 if (d_mountpoint(parent))
1007 spin_lock(&this_parent->d_lock);
1009 next = this_parent->d_subdirs.next;
1011 while (next != &this_parent->d_subdirs) {
1012 struct list_head *tmp = next;
1013 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1016 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1017 /* Have we found a mount point ? */
1018 if (d_mountpoint(dentry)) {
1019 spin_unlock(&dentry->d_lock);
1020 spin_unlock(&this_parent->d_lock);
1023 if (!list_empty(&dentry->d_subdirs)) {
1024 spin_unlock(&this_parent->d_lock);
1025 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1026 this_parent = dentry;
1027 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1030 spin_unlock(&dentry->d_lock);
1033 * All done at this level ... ascend and resume the search.
1035 if (this_parent != parent) {
1036 struct dentry *child = this_parent;
1037 this_parent = try_to_ascend(this_parent, locked, seq);
1040 next = child->d_u.d_child.next;
1043 spin_unlock(&this_parent->d_lock);
1044 if (!locked && read_seqretry(&rename_lock, seq))
1047 write_sequnlock(&rename_lock);
1048 return 0; /* No mount points found in tree */
1050 if (!locked && read_seqretry(&rename_lock, seq))
1053 write_sequnlock(&rename_lock);
1058 write_seqlock(&rename_lock);
1061 EXPORT_SYMBOL(have_submounts);
1064 * Search the dentry child list for the specified parent,
1065 * and move any unused dentries to the end of the unused
1066 * list for prune_dcache(). We descend to the next level
1067 * whenever the d_subdirs list is non-empty and continue
1070 * It returns zero iff there are no unused children,
1071 * otherwise it returns the number of children moved to
1072 * the end of the unused list. This may not be the total
1073 * number of unused children, because select_parent can
1074 * drop the lock and return early due to latency
1077 static int select_parent(struct dentry *parent, struct list_head *dispose)
1079 struct dentry *this_parent;
1080 struct list_head *next;
1085 seq = read_seqbegin(&rename_lock);
1087 this_parent = parent;
1088 spin_lock(&this_parent->d_lock);
1090 next = this_parent->d_subdirs.next;
1092 while (next != &this_parent->d_subdirs) {
1093 struct list_head *tmp = next;
1094 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1097 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1100 * move only zero ref count dentries to the dispose list.
1102 * Those which are presently on the shrink list, being processed
1103 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1104 * loop in shrink_dcache_parent() might not make any progress
1107 if (dentry->d_count) {
1108 dentry_lru_del(dentry);
1109 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1110 dentry_lru_move_list(dentry, dispose);
1111 dentry->d_flags |= DCACHE_SHRINK_LIST;
1115 * We can return to the caller if we have found some (this
1116 * ensures forward progress). We'll be coming back to find
1119 if (found && need_resched()) {
1120 spin_unlock(&dentry->d_lock);
1125 * Descend a level if the d_subdirs list is non-empty.
1127 if (!list_empty(&dentry->d_subdirs)) {
1128 spin_unlock(&this_parent->d_lock);
1129 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1130 this_parent = dentry;
1131 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1135 spin_unlock(&dentry->d_lock);
1138 * All done at this level ... ascend and resume the search.
1140 if (this_parent != parent) {
1141 struct dentry *child = this_parent;
1142 this_parent = try_to_ascend(this_parent, locked, seq);
1145 next = child->d_u.d_child.next;
1149 spin_unlock(&this_parent->d_lock);
1150 if (!locked && read_seqretry(&rename_lock, seq))
1153 write_sequnlock(&rename_lock);
1160 write_seqlock(&rename_lock);
1165 * shrink_dcache_parent - prune dcache
1166 * @parent: parent of entries to prune
1168 * Prune the dcache to remove unused children of the parent dentry.
1170 void shrink_dcache_parent(struct dentry * parent)
1175 while ((found = select_parent(parent, &dispose)) != 0)
1176 shrink_dentry_list(&dispose);
1178 EXPORT_SYMBOL(shrink_dcache_parent);
1181 * __d_alloc - allocate a dcache entry
1182 * @sb: filesystem it will belong to
1183 * @name: qstr of the name
1185 * Allocates a dentry. It returns %NULL if there is insufficient memory
1186 * available. On a success the dentry is returned. The name passed in is
1187 * copied and the copy passed in may be reused after this call.
1190 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1192 struct dentry *dentry;
1195 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1199 if (name->len > DNAME_INLINE_LEN-1) {
1200 dname = kmalloc(name->len + 1, GFP_KERNEL);
1202 kmem_cache_free(dentry_cache, dentry);
1206 dname = dentry->d_iname;
1208 dentry->d_name.name = dname;
1210 dentry->d_name.len = name->len;
1211 dentry->d_name.hash = name->hash;
1212 memcpy(dname, name->name, name->len);
1213 dname[name->len] = 0;
1215 dentry->d_count = 1;
1216 dentry->d_flags = 0;
1217 spin_lock_init(&dentry->d_lock);
1218 seqcount_init(&dentry->d_seq);
1219 dentry->d_inode = NULL;
1220 dentry->d_parent = dentry;
1222 dentry->d_op = NULL;
1223 dentry->d_fsdata = NULL;
1224 INIT_HLIST_BL_NODE(&dentry->d_hash);
1225 INIT_LIST_HEAD(&dentry->d_lru);
1226 INIT_LIST_HEAD(&dentry->d_subdirs);
1227 INIT_LIST_HEAD(&dentry->d_alias);
1228 INIT_LIST_HEAD(&dentry->d_u.d_child);
1229 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1231 this_cpu_inc(nr_dentry);
1237 * d_alloc - allocate a dcache entry
1238 * @parent: parent of entry to allocate
1239 * @name: qstr of the name
1241 * Allocates a dentry. It returns %NULL if there is insufficient memory
1242 * available. On a success the dentry is returned. The name passed in is
1243 * copied and the copy passed in may be reused after this call.
1245 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1247 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1251 spin_lock(&parent->d_lock);
1253 * don't need child lock because it is not subject
1254 * to concurrency here
1256 __dget_dlock(parent);
1257 dentry->d_parent = parent;
1258 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1259 spin_unlock(&parent->d_lock);
1263 EXPORT_SYMBOL(d_alloc);
1265 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1267 struct dentry *dentry = __d_alloc(sb, name);
1269 dentry->d_flags |= DCACHE_DISCONNECTED;
1272 EXPORT_SYMBOL(d_alloc_pseudo);
1274 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1279 q.len = strlen(name);
1280 q.hash = full_name_hash(q.name, q.len);
1281 return d_alloc(parent, &q);
1283 EXPORT_SYMBOL(d_alloc_name);
1285 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1287 WARN_ON_ONCE(dentry->d_op);
1288 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1290 DCACHE_OP_REVALIDATE |
1291 DCACHE_OP_DELETE ));
1296 dentry->d_flags |= DCACHE_OP_HASH;
1298 dentry->d_flags |= DCACHE_OP_COMPARE;
1299 if (op->d_revalidate)
1300 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1302 dentry->d_flags |= DCACHE_OP_DELETE;
1304 dentry->d_flags |= DCACHE_OP_PRUNE;
1307 EXPORT_SYMBOL(d_set_d_op);
1309 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1311 spin_lock(&dentry->d_lock);
1313 if (unlikely(IS_AUTOMOUNT(inode)))
1314 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1315 list_add(&dentry->d_alias, &inode->i_dentry);
1317 dentry->d_inode = inode;
1318 dentry_rcuwalk_barrier(dentry);
1319 spin_unlock(&dentry->d_lock);
1320 fsnotify_d_instantiate(dentry, inode);
1324 * d_instantiate - fill in inode information for a dentry
1325 * @entry: dentry to complete
1326 * @inode: inode to attach to this dentry
1328 * Fill in inode information in the entry.
1330 * This turns negative dentries into productive full members
1333 * NOTE! This assumes that the inode count has been incremented
1334 * (or otherwise set) by the caller to indicate that it is now
1335 * in use by the dcache.
1338 void d_instantiate(struct dentry *entry, struct inode * inode)
1340 BUG_ON(!list_empty(&entry->d_alias));
1342 spin_lock(&inode->i_lock);
1343 __d_instantiate(entry, inode);
1345 spin_unlock(&inode->i_lock);
1346 security_d_instantiate(entry, inode);
1348 EXPORT_SYMBOL(d_instantiate);
1351 * d_instantiate_unique - instantiate a non-aliased dentry
1352 * @entry: dentry to instantiate
1353 * @inode: inode to attach to this dentry
1355 * Fill in inode information in the entry. On success, it returns NULL.
1356 * If an unhashed alias of "entry" already exists, then we return the
1357 * aliased dentry instead and drop one reference to inode.
1359 * Note that in order to avoid conflicts with rename() etc, the caller
1360 * had better be holding the parent directory semaphore.
1362 * This also assumes that the inode count has been incremented
1363 * (or otherwise set) by the caller to indicate that it is now
1364 * in use by the dcache.
1366 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1367 struct inode *inode)
1369 struct dentry *alias;
1370 int len = entry->d_name.len;
1371 const char *name = entry->d_name.name;
1372 unsigned int hash = entry->d_name.hash;
1375 __d_instantiate(entry, NULL);
1379 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1380 struct qstr *qstr = &alias->d_name;
1383 * Don't need alias->d_lock here, because aliases with
1384 * d_parent == entry->d_parent are not subject to name or
1385 * parent changes, because the parent inode i_mutex is held.
1387 if (qstr->hash != hash)
1389 if (alias->d_parent != entry->d_parent)
1391 if (dentry_cmp(qstr->name, qstr->len, name, len))
1397 __d_instantiate(entry, inode);
1401 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1403 struct dentry *result;
1405 BUG_ON(!list_empty(&entry->d_alias));
1408 spin_lock(&inode->i_lock);
1409 result = __d_instantiate_unique(entry, inode);
1411 spin_unlock(&inode->i_lock);
1414 security_d_instantiate(entry, inode);
1418 BUG_ON(!d_unhashed(result));
1423 EXPORT_SYMBOL(d_instantiate_unique);
1426 * d_alloc_root - allocate root dentry
1427 * @root_inode: inode to allocate the root for
1429 * Allocate a root ("/") dentry for the inode given. The inode is
1430 * instantiated and returned. %NULL is returned if there is insufficient
1431 * memory or the inode passed is %NULL.
1434 struct dentry * d_alloc_root(struct inode * root_inode)
1436 struct dentry *res = NULL;
1439 static const struct qstr name = { .name = "/", .len = 1 };
1441 res = __d_alloc(root_inode->i_sb, &name);
1443 d_instantiate(res, root_inode);
1447 EXPORT_SYMBOL(d_alloc_root);
1449 static struct dentry * __d_find_any_alias(struct inode *inode)
1451 struct dentry *alias;
1453 if (list_empty(&inode->i_dentry))
1455 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1460 static struct dentry * d_find_any_alias(struct inode *inode)
1464 spin_lock(&inode->i_lock);
1465 de = __d_find_any_alias(inode);
1466 spin_unlock(&inode->i_lock);
1472 * d_obtain_alias - find or allocate a dentry for a given inode
1473 * @inode: inode to allocate the dentry for
1475 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1476 * similar open by handle operations. The returned dentry may be anonymous,
1477 * or may have a full name (if the inode was already in the cache).
1479 * When called on a directory inode, we must ensure that the inode only ever
1480 * has one dentry. If a dentry is found, that is returned instead of
1481 * allocating a new one.
1483 * On successful return, the reference to the inode has been transferred
1484 * to the dentry. In case of an error the reference on the inode is released.
1485 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1486 * be passed in and will be the error will be propagate to the return value,
1487 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1489 struct dentry *d_obtain_alias(struct inode *inode)
1491 static const struct qstr anonstring = { .name = "" };
1496 return ERR_PTR(-ESTALE);
1498 return ERR_CAST(inode);
1500 res = d_find_any_alias(inode);
1504 tmp = __d_alloc(inode->i_sb, &anonstring);
1506 res = ERR_PTR(-ENOMEM);
1510 spin_lock(&inode->i_lock);
1511 res = __d_find_any_alias(inode);
1513 spin_unlock(&inode->i_lock);
1518 /* attach a disconnected dentry */
1519 spin_lock(&tmp->d_lock);
1520 tmp->d_inode = inode;
1521 tmp->d_flags |= DCACHE_DISCONNECTED;
1522 list_add(&tmp->d_alias, &inode->i_dentry);
1523 hlist_bl_lock(&tmp->d_sb->s_anon);
1524 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1525 hlist_bl_unlock(&tmp->d_sb->s_anon);
1526 spin_unlock(&tmp->d_lock);
1527 spin_unlock(&inode->i_lock);
1528 security_d_instantiate(tmp, inode);
1533 if (res && !IS_ERR(res))
1534 security_d_instantiate(res, inode);
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;
1561 return ERR_CAST(inode);
1563 if (inode && S_ISDIR(inode->i_mode)) {
1564 spin_lock(&inode->i_lock);
1565 new = __d_find_alias(inode, 1);
1567 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1568 spin_unlock(&inode->i_lock);
1569 security_d_instantiate(new, inode);
1570 d_move(new, dentry);
1573 /* already taking inode->i_lock, so d_add() by hand */
1574 __d_instantiate(dentry, inode);
1575 spin_unlock(&inode->i_lock);
1576 security_d_instantiate(dentry, inode);
1580 d_add(dentry, inode);
1583 EXPORT_SYMBOL(d_splice_alias);
1586 * d_add_ci - lookup or allocate new dentry with case-exact name
1587 * @inode: the inode case-insensitive lookup has found
1588 * @dentry: the negative dentry that was passed to the parent's lookup func
1589 * @name: the case-exact name to be associated with the returned dentry
1591 * This is to avoid filling the dcache with case-insensitive names to the
1592 * same inode, only the actual correct case is stored in the dcache for
1593 * case-insensitive filesystems.
1595 * For a case-insensitive lookup match and if the the case-exact dentry
1596 * already exists in in the dcache, use it and return it.
1598 * If no entry exists with the exact case name, allocate new dentry with
1599 * the exact case, and return the spliced entry.
1601 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1605 struct dentry *found;
1609 * First check if a dentry matching the name already exists,
1610 * if not go ahead and create it now.
1612 found = d_hash_and_lookup(dentry->d_parent, name);
1614 new = d_alloc(dentry->d_parent, name);
1620 found = d_splice_alias(inode, new);
1629 * If a matching dentry exists, and it's not negative use it.
1631 * Decrement the reference count to balance the iget() done
1634 if (found->d_inode) {
1635 if (unlikely(found->d_inode != inode)) {
1636 /* This can't happen because bad inodes are unhashed. */
1637 BUG_ON(!is_bad_inode(inode));
1638 BUG_ON(!is_bad_inode(found->d_inode));
1645 * We are going to instantiate this dentry, unhash it and clear the
1646 * lookup flag so we can do that.
1648 if (unlikely(d_need_lookup(found)))
1649 d_clear_need_lookup(found);
1652 * Negative dentry: instantiate it unless the inode is a directory and
1653 * already has a dentry.
1655 new = d_splice_alias(inode, found);
1664 return ERR_PTR(error);
1666 EXPORT_SYMBOL(d_add_ci);
1669 * __d_lookup_rcu - search for a dentry (racy, store-free)
1670 * @parent: parent dentry
1671 * @name: qstr of name we wish to find
1672 * @seq: returns d_seq value at the point where the dentry was found
1673 * @inode: returns dentry->d_inode when the inode was found valid.
1674 * Returns: dentry, or NULL
1676 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1677 * resolution (store-free path walking) design described in
1678 * Documentation/filesystems/path-lookup.txt.
1680 * This is not to be used outside core vfs.
1682 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1683 * held, and rcu_read_lock held. The returned dentry must not be stored into
1684 * without taking d_lock and checking d_seq sequence count against @seq
1687 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1690 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1691 * the returned dentry, so long as its parent's seqlock is checked after the
1692 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1693 * is formed, giving integrity down the path walk.
1695 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1696 unsigned *seq, struct inode **inode)
1698 unsigned int len = name->len;
1699 unsigned int hash = name->hash;
1700 const unsigned char *str = name->name;
1701 struct hlist_bl_head *b = d_hash(parent, hash);
1702 struct hlist_bl_node *node;
1703 struct dentry *dentry;
1706 * Note: There is significant duplication with __d_lookup_rcu which is
1707 * required to prevent single threaded performance regressions
1708 * especially on architectures where smp_rmb (in seqcounts) are costly.
1709 * Keep the two functions in sync.
1713 * The hash list is protected using RCU.
1715 * Carefully use d_seq when comparing a candidate dentry, to avoid
1716 * races with d_move().
1718 * It is possible that concurrent renames can mess up our list
1719 * walk here and result in missing our dentry, resulting in the
1720 * false-negative result. d_lookup() protects against concurrent
1721 * renames using rename_lock seqlock.
1723 * See Documentation/filesystems/path-lookup.txt for more details.
1725 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1730 if (dentry->d_name.hash != hash)
1734 *seq = read_seqcount_begin(&dentry->d_seq);
1735 if (dentry->d_parent != parent)
1737 if (d_unhashed(dentry))
1739 tlen = dentry->d_name.len;
1740 tname = dentry->d_name.name;
1741 i = dentry->d_inode;
1744 * This seqcount check is required to ensure name and
1745 * len are loaded atomically, so as not to walk off the
1746 * edge of memory when walking. If we could load this
1747 * atomically some other way, we could drop this check.
1749 if (read_seqcount_retry(&dentry->d_seq, *seq))
1751 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1752 if (parent->d_op->d_compare(parent, *inode,
1757 if (dentry_cmp(tname, tlen, str, len))
1761 * No extra seqcount check is required after the name
1762 * compare. The caller must perform a seqcount check in
1763 * order to do anything useful with the returned dentry
1773 * d_lookup - search for a dentry
1774 * @parent: parent dentry
1775 * @name: qstr of name we wish to find
1776 * Returns: dentry, or NULL
1778 * d_lookup searches the children of the parent dentry for the name in
1779 * question. If the dentry is found its reference count is incremented and the
1780 * dentry is returned. The caller must use dput to free the entry when it has
1781 * finished using it. %NULL is returned if the dentry does not exist.
1783 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1785 struct dentry *dentry;
1789 seq = read_seqbegin(&rename_lock);
1790 dentry = __d_lookup(parent, name);
1793 } while (read_seqretry(&rename_lock, seq));
1796 EXPORT_SYMBOL(d_lookup);
1799 * __d_lookup - search for a dentry (racy)
1800 * @parent: parent dentry
1801 * @name: qstr of name we wish to find
1802 * Returns: dentry, or NULL
1804 * __d_lookup is like d_lookup, however it may (rarely) return a
1805 * false-negative result due to unrelated rename activity.
1807 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1808 * however it must be used carefully, eg. with a following d_lookup in
1809 * the case of failure.
1811 * __d_lookup callers must be commented.
1813 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1815 unsigned int len = name->len;
1816 unsigned int hash = name->hash;
1817 const unsigned char *str = name->name;
1818 struct hlist_bl_head *b = d_hash(parent, hash);
1819 struct hlist_bl_node *node;
1820 struct dentry *found = NULL;
1821 struct dentry *dentry;
1824 * Note: There is significant duplication with __d_lookup_rcu which is
1825 * required to prevent single threaded performance regressions
1826 * especially on architectures where smp_rmb (in seqcounts) are costly.
1827 * Keep the two functions in sync.
1831 * The hash list is protected using RCU.
1833 * Take d_lock when comparing a candidate dentry, to avoid races
1836 * It is possible that concurrent renames can mess up our list
1837 * walk here and result in missing our dentry, resulting in the
1838 * false-negative result. d_lookup() protects against concurrent
1839 * renames using rename_lock seqlock.
1841 * See Documentation/filesystems/path-lookup.txt for more details.
1845 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1849 if (dentry->d_name.hash != hash)
1852 spin_lock(&dentry->d_lock);
1853 if (dentry->d_parent != parent)
1855 if (d_unhashed(dentry))
1859 * It is safe to compare names since d_move() cannot
1860 * change the qstr (protected by d_lock).
1862 tlen = dentry->d_name.len;
1863 tname = dentry->d_name.name;
1864 if (parent->d_flags & DCACHE_OP_COMPARE) {
1865 if (parent->d_op->d_compare(parent, parent->d_inode,
1866 dentry, dentry->d_inode,
1870 if (dentry_cmp(tname, tlen, str, len))
1876 spin_unlock(&dentry->d_lock);
1879 spin_unlock(&dentry->d_lock);
1887 * d_hash_and_lookup - hash the qstr then search for a dentry
1888 * @dir: Directory to search in
1889 * @name: qstr of name we wish to find
1891 * On hash failure or on lookup failure NULL is returned.
1893 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1895 struct dentry *dentry = NULL;
1898 * Check for a fs-specific hash function. Note that we must
1899 * calculate the standard hash first, as the d_op->d_hash()
1900 * routine may choose to leave the hash value unchanged.
1902 name->hash = full_name_hash(name->name, name->len);
1903 if (dir->d_flags & DCACHE_OP_HASH) {
1904 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1907 dentry = d_lookup(dir, name);
1913 * d_validate - verify dentry provided from insecure source (deprecated)
1914 * @dentry: The dentry alleged to be valid child of @dparent
1915 * @dparent: The parent dentry (known to be valid)
1917 * An insecure source has sent us a dentry, here we verify it and dget() it.
1918 * This is used by ncpfs in its readdir implementation.
1919 * Zero is returned in the dentry is invalid.
1921 * This function is slow for big directories, and deprecated, do not use it.
1923 int d_validate(struct dentry *dentry, struct dentry *dparent)
1925 struct dentry *child;
1927 spin_lock(&dparent->d_lock);
1928 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1929 if (dentry == child) {
1930 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1931 __dget_dlock(dentry);
1932 spin_unlock(&dentry->d_lock);
1933 spin_unlock(&dparent->d_lock);
1937 spin_unlock(&dparent->d_lock);
1941 EXPORT_SYMBOL(d_validate);
1944 * When a file is deleted, we have two options:
1945 * - turn this dentry into a negative dentry
1946 * - unhash this dentry and free it.
1948 * Usually, we want to just turn this into
1949 * a negative dentry, but if anybody else is
1950 * currently using the dentry or the inode
1951 * we can't do that and we fall back on removing
1952 * it from the hash queues and waiting for
1953 * it to be deleted later when it has no users
1957 * d_delete - delete a dentry
1958 * @dentry: The dentry to delete
1960 * Turn the dentry into a negative dentry if possible, otherwise
1961 * remove it from the hash queues so it can be deleted later
1964 void d_delete(struct dentry * dentry)
1966 struct inode *inode;
1969 * Are we the only user?
1972 spin_lock(&dentry->d_lock);
1973 inode = dentry->d_inode;
1974 isdir = S_ISDIR(inode->i_mode);
1975 if (dentry->d_count == 1) {
1976 if (inode && !spin_trylock(&inode->i_lock)) {
1977 spin_unlock(&dentry->d_lock);
1981 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1982 dentry_unlink_inode(dentry);
1983 fsnotify_nameremove(dentry, isdir);
1987 if (!d_unhashed(dentry))
1990 spin_unlock(&dentry->d_lock);
1992 fsnotify_nameremove(dentry, isdir);
1994 EXPORT_SYMBOL(d_delete);
1996 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
1998 BUG_ON(!d_unhashed(entry));
2000 entry->d_flags |= DCACHE_RCUACCESS;
2001 hlist_bl_add_head_rcu(&entry->d_hash, b);
2005 static void _d_rehash(struct dentry * entry)
2007 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2011 * d_rehash - add an entry back to the hash
2012 * @entry: dentry to add to the hash
2014 * Adds a dentry to the hash according to its name.
2017 void d_rehash(struct dentry * entry)
2019 spin_lock(&entry->d_lock);
2021 spin_unlock(&entry->d_lock);
2023 EXPORT_SYMBOL(d_rehash);
2026 * dentry_update_name_case - update case insensitive dentry with a new name
2027 * @dentry: dentry to be updated
2030 * Update a case insensitive dentry with new case of name.
2032 * dentry must have been returned by d_lookup with name @name. Old and new
2033 * name lengths must match (ie. no d_compare which allows mismatched name
2036 * Parent inode i_mutex must be held over d_lookup and into this call (to
2037 * keep renames and concurrent inserts, and readdir(2) away).
2039 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2041 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2042 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2044 spin_lock(&dentry->d_lock);
2045 write_seqcount_begin(&dentry->d_seq);
2046 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2047 write_seqcount_end(&dentry->d_seq);
2048 spin_unlock(&dentry->d_lock);
2050 EXPORT_SYMBOL(dentry_update_name_case);
2052 static void switch_names(struct dentry *dentry, struct dentry *target)
2054 if (dname_external(target)) {
2055 if (dname_external(dentry)) {
2057 * Both external: swap the pointers
2059 swap(target->d_name.name, dentry->d_name.name);
2062 * dentry:internal, target:external. Steal target's
2063 * storage and make target internal.
2065 memcpy(target->d_iname, dentry->d_name.name,
2066 dentry->d_name.len + 1);
2067 dentry->d_name.name = target->d_name.name;
2068 target->d_name.name = target->d_iname;
2071 if (dname_external(dentry)) {
2073 * dentry:external, target:internal. Give dentry's
2074 * storage to target and make dentry internal
2076 memcpy(dentry->d_iname, target->d_name.name,
2077 target->d_name.len + 1);
2078 target->d_name.name = dentry->d_name.name;
2079 dentry->d_name.name = dentry->d_iname;
2082 * Both are internal. Just copy target to dentry
2084 memcpy(dentry->d_iname, target->d_name.name,
2085 target->d_name.len + 1);
2086 dentry->d_name.len = target->d_name.len;
2090 swap(dentry->d_name.len, target->d_name.len);
2093 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2096 * XXXX: do we really need to take target->d_lock?
2098 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2099 spin_lock(&target->d_parent->d_lock);
2101 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2102 spin_lock(&dentry->d_parent->d_lock);
2103 spin_lock_nested(&target->d_parent->d_lock,
2104 DENTRY_D_LOCK_NESTED);
2106 spin_lock(&target->d_parent->d_lock);
2107 spin_lock_nested(&dentry->d_parent->d_lock,
2108 DENTRY_D_LOCK_NESTED);
2111 if (target < dentry) {
2112 spin_lock_nested(&target->d_lock, 2);
2113 spin_lock_nested(&dentry->d_lock, 3);
2115 spin_lock_nested(&dentry->d_lock, 2);
2116 spin_lock_nested(&target->d_lock, 3);
2120 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2121 struct dentry *target)
2123 if (target->d_parent != dentry->d_parent)
2124 spin_unlock(&dentry->d_parent->d_lock);
2125 if (target->d_parent != target)
2126 spin_unlock(&target->d_parent->d_lock);
2130 * When switching names, the actual string doesn't strictly have to
2131 * be preserved in the target - because we're dropping the target
2132 * anyway. As such, we can just do a simple memcpy() to copy over
2133 * the new name before we switch.
2135 * Note that we have to be a lot more careful about getting the hash
2136 * switched - we have to switch the hash value properly even if it
2137 * then no longer matches the actual (corrupted) string of the target.
2138 * The hash value has to match the hash queue that the dentry is on..
2141 * __d_move - move a dentry
2142 * @dentry: entry to move
2143 * @target: new dentry
2145 * Update the dcache to reflect the move of a file name. Negative
2146 * dcache entries should not be moved in this way. Caller must hold
2147 * rename_lock, the i_mutex of the source and target directories,
2148 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2150 static void __d_move(struct dentry * dentry, struct dentry * target)
2152 if (!dentry->d_inode)
2153 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2155 BUG_ON(d_ancestor(dentry, target));
2156 BUG_ON(d_ancestor(target, dentry));
2158 dentry_lock_for_move(dentry, target);
2160 write_seqcount_begin(&dentry->d_seq);
2161 write_seqcount_begin(&target->d_seq);
2163 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2166 * Move the dentry to the target hash queue. Don't bother checking
2167 * for the same hash queue because of how unlikely it is.
2170 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2172 /* Unhash the target: dput() will then get rid of it */
2175 list_del(&dentry->d_u.d_child);
2176 list_del(&target->d_u.d_child);
2178 /* Switch the names.. */
2179 switch_names(dentry, target);
2180 swap(dentry->d_name.hash, target->d_name.hash);
2182 /* ... and switch the parents */
2183 if (IS_ROOT(dentry)) {
2184 dentry->d_parent = target->d_parent;
2185 target->d_parent = target;
2186 INIT_LIST_HEAD(&target->d_u.d_child);
2188 swap(dentry->d_parent, target->d_parent);
2190 /* And add them back to the (new) parent lists */
2191 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2194 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2196 write_seqcount_end(&target->d_seq);
2197 write_seqcount_end(&dentry->d_seq);
2199 dentry_unlock_parents_for_move(dentry, target);
2200 spin_unlock(&target->d_lock);
2201 fsnotify_d_move(dentry);
2202 spin_unlock(&dentry->d_lock);
2206 * d_move - move a dentry
2207 * @dentry: entry to move
2208 * @target: new dentry
2210 * Update the dcache to reflect the move of a file name. Negative
2211 * dcache entries should not be moved in this way. See the locking
2212 * requirements for __d_move.
2214 void d_move(struct dentry *dentry, struct dentry *target)
2216 write_seqlock(&rename_lock);
2217 __d_move(dentry, target);
2218 write_sequnlock(&rename_lock);
2220 EXPORT_SYMBOL(d_move);
2223 * d_ancestor - search for an ancestor
2224 * @p1: ancestor dentry
2227 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2228 * an ancestor of p2, else NULL.
2230 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2234 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2235 if (p->d_parent == p1)
2242 * This helper attempts to cope with remotely renamed directories
2244 * It assumes that the caller is already holding
2245 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2247 * Note: If ever the locking in lock_rename() changes, then please
2248 * remember to update this too...
2250 static struct dentry *__d_unalias(struct inode *inode,
2251 struct dentry *dentry, struct dentry *alias)
2253 struct mutex *m1 = NULL, *m2 = NULL;
2256 /* If alias and dentry share a parent, then no extra locks required */
2257 if (alias->d_parent == dentry->d_parent)
2260 /* See lock_rename() */
2261 ret = ERR_PTR(-EBUSY);
2262 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2264 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2265 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2267 m2 = &alias->d_parent->d_inode->i_mutex;
2269 __d_move(alias, dentry);
2272 spin_unlock(&inode->i_lock);
2281 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2282 * named dentry in place of the dentry to be replaced.
2283 * returns with anon->d_lock held!
2285 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2287 struct dentry *dparent, *aparent;
2289 dentry_lock_for_move(anon, dentry);
2291 write_seqcount_begin(&dentry->d_seq);
2292 write_seqcount_begin(&anon->d_seq);
2294 dparent = dentry->d_parent;
2295 aparent = anon->d_parent;
2297 switch_names(dentry, anon);
2298 swap(dentry->d_name.hash, anon->d_name.hash);
2300 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2301 list_del(&dentry->d_u.d_child);
2302 if (!IS_ROOT(dentry))
2303 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2305 INIT_LIST_HEAD(&dentry->d_u.d_child);
2307 anon->d_parent = (dparent == dentry) ? anon : dparent;
2308 list_del(&anon->d_u.d_child);
2310 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2312 INIT_LIST_HEAD(&anon->d_u.d_child);
2314 write_seqcount_end(&dentry->d_seq);
2315 write_seqcount_end(&anon->d_seq);
2317 dentry_unlock_parents_for_move(anon, dentry);
2318 spin_unlock(&dentry->d_lock);
2320 /* anon->d_lock still locked, returns locked */
2321 anon->d_flags &= ~DCACHE_DISCONNECTED;
2325 * d_materialise_unique - introduce an inode into the tree
2326 * @dentry: candidate dentry
2327 * @inode: inode to bind to the dentry, to which aliases may be attached
2329 * Introduces an dentry into the tree, substituting an extant disconnected
2330 * root directory alias in its place if there is one. Caller must hold the
2331 * i_mutex of the parent directory.
2333 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2335 struct dentry *actual;
2337 BUG_ON(!d_unhashed(dentry));
2341 __d_instantiate(dentry, NULL);
2346 spin_lock(&inode->i_lock);
2348 if (S_ISDIR(inode->i_mode)) {
2349 struct dentry *alias;
2351 /* Does an aliased dentry already exist? */
2352 alias = __d_find_alias(inode, 0);
2355 write_seqlock(&rename_lock);
2357 if (d_ancestor(alias, dentry)) {
2358 /* Check for loops */
2359 actual = ERR_PTR(-ELOOP);
2360 spin_unlock(&inode->i_lock);
2361 } else if (IS_ROOT(alias)) {
2362 /* Is this an anonymous mountpoint that we
2363 * could splice into our tree? */
2364 __d_materialise_dentry(dentry, alias);
2365 write_sequnlock(&rename_lock);
2369 /* Nope, but we must(!) avoid directory
2370 * aliasing. This drops inode->i_lock */
2371 actual = __d_unalias(inode, dentry, alias);
2373 write_sequnlock(&rename_lock);
2374 if (IS_ERR(actual)) {
2375 if (PTR_ERR(actual) == -ELOOP)
2376 pr_warn_ratelimited(
2377 "VFS: Lookup of '%s' in %s %s"
2378 " would have caused loop\n",
2379 dentry->d_name.name,
2380 inode->i_sb->s_type->name,
2388 /* Add a unique reference */
2389 actual = __d_instantiate_unique(dentry, inode);
2393 BUG_ON(!d_unhashed(actual));
2395 spin_lock(&actual->d_lock);
2398 spin_unlock(&actual->d_lock);
2399 spin_unlock(&inode->i_lock);
2401 if (actual == dentry) {
2402 security_d_instantiate(dentry, inode);
2409 EXPORT_SYMBOL_GPL(d_materialise_unique);
2411 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2415 return -ENAMETOOLONG;
2417 memcpy(*buffer, str, namelen);
2421 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2423 return prepend(buffer, buflen, name->name, name->len);
2427 * prepend_path - Prepend path string to a buffer
2428 * @path: the dentry/vfsmount to report
2429 * @root: root vfsmnt/dentry
2430 * @buffer: pointer to the end of the buffer
2431 * @buflen: pointer to buffer length
2433 * Caller holds the rename_lock.
2435 static int prepend_path(const struct path *path,
2436 const struct path *root,
2437 char **buffer, int *buflen)
2439 struct dentry *dentry = path->dentry;
2440 struct vfsmount *vfsmnt = path->mnt;
2444 br_read_lock(vfsmount_lock);
2445 while (dentry != root->dentry || vfsmnt != root->mnt) {
2446 struct dentry * parent;
2448 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2450 if (vfsmnt->mnt_parent == vfsmnt) {
2453 dentry = vfsmnt->mnt_mountpoint;
2454 vfsmnt = vfsmnt->mnt_parent;
2457 parent = dentry->d_parent;
2459 spin_lock(&dentry->d_lock);
2460 error = prepend_name(buffer, buflen, &dentry->d_name);
2461 spin_unlock(&dentry->d_lock);
2463 error = prepend(buffer, buflen, "/", 1);
2471 if (!error && !slash)
2472 error = prepend(buffer, buflen, "/", 1);
2475 br_read_unlock(vfsmount_lock);
2480 * Filesystems needing to implement special "root names"
2481 * should do so with ->d_dname()
2483 if (IS_ROOT(dentry) &&
2484 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2485 WARN(1, "Root dentry has weird name <%.*s>\n",
2486 (int) dentry->d_name.len, dentry->d_name.name);
2489 error = prepend(buffer, buflen, "/", 1);
2491 error = vfsmnt->mnt_ns ? 1 : 2;
2496 * __d_path - return the path of a dentry
2497 * @path: the dentry/vfsmount to report
2498 * @root: root vfsmnt/dentry
2499 * @buf: buffer to return value in
2500 * @buflen: buffer length
2502 * Convert a dentry into an ASCII path name.
2504 * Returns a pointer into the buffer or an error code if the
2505 * path was too long.
2507 * "buflen" should be positive.
2509 * If the path is not reachable from the supplied root, return %NULL.
2511 char *__d_path(const struct path *path,
2512 const struct path *root,
2513 char *buf, int buflen)
2515 char *res = buf + buflen;
2518 prepend(&res, &buflen, "\0", 1);
2519 write_seqlock(&rename_lock);
2520 error = prepend_path(path, root, &res, &buflen);
2521 write_sequnlock(&rename_lock);
2524 return ERR_PTR(error);
2530 char *d_absolute_path(const struct path *path,
2531 char *buf, int buflen)
2533 struct path root = {};
2534 char *res = buf + buflen;
2537 prepend(&res, &buflen, "\0", 1);
2538 write_seqlock(&rename_lock);
2539 error = prepend_path(path, &root, &res, &buflen);
2540 write_sequnlock(&rename_lock);
2545 return ERR_PTR(error);
2550 * same as __d_path but appends "(deleted)" for unlinked files.
2552 static int path_with_deleted(const struct path *path,
2553 const struct path *root,
2554 char **buf, int *buflen)
2556 prepend(buf, buflen, "\0", 1);
2557 if (d_unlinked(path->dentry)) {
2558 int error = prepend(buf, buflen, " (deleted)", 10);
2563 return prepend_path(path, root, buf, buflen);
2566 static int prepend_unreachable(char **buffer, int *buflen)
2568 return prepend(buffer, buflen, "(unreachable)", 13);
2572 * d_path - return the path of a dentry
2573 * @path: path to report
2574 * @buf: buffer to return value in
2575 * @buflen: buffer length
2577 * Convert a dentry into an ASCII path name. If the entry has been deleted
2578 * the string " (deleted)" is appended. Note that this is ambiguous.
2580 * Returns a pointer into the buffer or an error code if the path was
2581 * too long. Note: Callers should use the returned pointer, not the passed
2582 * in buffer, to use the name! The implementation often starts at an offset
2583 * into the buffer, and may leave 0 bytes at the start.
2585 * "buflen" should be positive.
2587 char *d_path(const struct path *path, char *buf, int buflen)
2589 char *res = buf + buflen;
2594 * We have various synthetic filesystems that never get mounted. On
2595 * these filesystems dentries are never used for lookup purposes, and
2596 * thus don't need to be hashed. They also don't need a name until a
2597 * user wants to identify the object in /proc/pid/fd/. The little hack
2598 * below allows us to generate a name for these objects on demand:
2600 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2601 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2603 get_fs_root(current->fs, &root);
2604 write_seqlock(&rename_lock);
2605 error = path_with_deleted(path, &root, &res, &buflen);
2607 res = ERR_PTR(error);
2608 write_sequnlock(&rename_lock);
2612 EXPORT_SYMBOL(d_path);
2615 * d_path_with_unreachable - return the path of a dentry
2616 * @path: path to report
2617 * @buf: buffer to return value in
2618 * @buflen: buffer length
2620 * The difference from d_path() is that this prepends "(unreachable)"
2621 * to paths which are unreachable from the current process' root.
2623 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2625 char *res = buf + buflen;
2629 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2630 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2632 get_fs_root(current->fs, &root);
2633 write_seqlock(&rename_lock);
2634 error = path_with_deleted(path, &root, &res, &buflen);
2636 error = prepend_unreachable(&res, &buflen);
2637 write_sequnlock(&rename_lock);
2640 res = ERR_PTR(error);
2646 * Helper function for dentry_operations.d_dname() members
2648 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2649 const char *fmt, ...)
2655 va_start(args, fmt);
2656 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2659 if (sz > sizeof(temp) || sz > buflen)
2660 return ERR_PTR(-ENAMETOOLONG);
2662 buffer += buflen - sz;
2663 return memcpy(buffer, temp, sz);
2667 * Write full pathname from the root of the filesystem into the buffer.
2669 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2671 char *end = buf + buflen;
2674 prepend(&end, &buflen, "\0", 1);
2681 while (!IS_ROOT(dentry)) {
2682 struct dentry *parent = dentry->d_parent;
2686 spin_lock(&dentry->d_lock);
2687 error = prepend_name(&end, &buflen, &dentry->d_name);
2688 spin_unlock(&dentry->d_lock);
2689 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2697 return ERR_PTR(-ENAMETOOLONG);
2700 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2704 write_seqlock(&rename_lock);
2705 retval = __dentry_path(dentry, buf, buflen);
2706 write_sequnlock(&rename_lock);
2710 EXPORT_SYMBOL(dentry_path_raw);
2712 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2717 write_seqlock(&rename_lock);
2718 if (d_unlinked(dentry)) {
2720 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2724 retval = __dentry_path(dentry, buf, buflen);
2725 write_sequnlock(&rename_lock);
2726 if (!IS_ERR(retval) && p)
2727 *p = '/'; /* restore '/' overriden with '\0' */
2730 return ERR_PTR(-ENAMETOOLONG);
2734 * NOTE! The user-level library version returns a
2735 * character pointer. The kernel system call just
2736 * returns the length of the buffer filled (which
2737 * includes the ending '\0' character), or a negative
2738 * error value. So libc would do something like
2740 * char *getcwd(char * buf, size_t size)
2744 * retval = sys_getcwd(buf, size);
2751 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2754 struct path pwd, root;
2755 char *page = (char *) __get_free_page(GFP_USER);
2760 get_fs_root_and_pwd(current->fs, &root, &pwd);
2763 write_seqlock(&rename_lock);
2764 if (!d_unlinked(pwd.dentry)) {
2766 char *cwd = page + PAGE_SIZE;
2767 int buflen = PAGE_SIZE;
2769 prepend(&cwd, &buflen, "\0", 1);
2770 error = prepend_path(&pwd, &root, &cwd, &buflen);
2771 write_sequnlock(&rename_lock);
2776 /* Unreachable from current root */
2778 error = prepend_unreachable(&cwd, &buflen);
2784 len = PAGE_SIZE + page - cwd;
2787 if (copy_to_user(buf, cwd, len))
2791 write_sequnlock(&rename_lock);
2797 free_page((unsigned long) page);
2802 * Test whether new_dentry is a subdirectory of old_dentry.
2804 * Trivially implemented using the dcache structure
2808 * is_subdir - is new dentry a subdirectory of old_dentry
2809 * @new_dentry: new dentry
2810 * @old_dentry: old dentry
2812 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2813 * Returns 0 otherwise.
2814 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2817 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2822 if (new_dentry == old_dentry)
2826 /* for restarting inner loop in case of seq retry */
2827 seq = read_seqbegin(&rename_lock);
2829 * Need rcu_readlock to protect against the d_parent trashing
2833 if (d_ancestor(old_dentry, new_dentry))
2838 } while (read_seqretry(&rename_lock, seq));
2843 int path_is_under(struct path *path1, struct path *path2)
2845 struct vfsmount *mnt = path1->mnt;
2846 struct dentry *dentry = path1->dentry;
2849 br_read_lock(vfsmount_lock);
2850 if (mnt != path2->mnt) {
2852 if (mnt->mnt_parent == mnt) {
2853 br_read_unlock(vfsmount_lock);
2856 if (mnt->mnt_parent == path2->mnt)
2858 mnt = mnt->mnt_parent;
2860 dentry = mnt->mnt_mountpoint;
2862 res = is_subdir(dentry, path2->dentry);
2863 br_read_unlock(vfsmount_lock);
2866 EXPORT_SYMBOL(path_is_under);
2868 void d_genocide(struct dentry *root)
2870 struct dentry *this_parent;
2871 struct list_head *next;
2875 seq = read_seqbegin(&rename_lock);
2878 spin_lock(&this_parent->d_lock);
2880 next = this_parent->d_subdirs.next;
2882 while (next != &this_parent->d_subdirs) {
2883 struct list_head *tmp = next;
2884 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2887 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2888 if (d_unhashed(dentry) || !dentry->d_inode) {
2889 spin_unlock(&dentry->d_lock);
2892 if (!list_empty(&dentry->d_subdirs)) {
2893 spin_unlock(&this_parent->d_lock);
2894 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2895 this_parent = dentry;
2896 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2899 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2900 dentry->d_flags |= DCACHE_GENOCIDE;
2903 spin_unlock(&dentry->d_lock);
2905 if (this_parent != root) {
2906 struct dentry *child = this_parent;
2907 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2908 this_parent->d_flags |= DCACHE_GENOCIDE;
2909 this_parent->d_count--;
2911 this_parent = try_to_ascend(this_parent, locked, seq);
2914 next = child->d_u.d_child.next;
2917 spin_unlock(&this_parent->d_lock);
2918 if (!locked && read_seqretry(&rename_lock, seq))
2921 write_sequnlock(&rename_lock);
2926 write_seqlock(&rename_lock);
2931 * find_inode_number - check for dentry with name
2932 * @dir: directory to check
2933 * @name: Name to find.
2935 * Check whether a dentry already exists for the given name,
2936 * and return the inode number if it has an inode. Otherwise
2939 * This routine is used to post-process directory listings for
2940 * filesystems using synthetic inode numbers, and is necessary
2941 * to keep getcwd() working.
2944 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2946 struct dentry * dentry;
2949 dentry = d_hash_and_lookup(dir, name);
2951 if (dentry->d_inode)
2952 ino = dentry->d_inode->i_ino;
2957 EXPORT_SYMBOL(find_inode_number);
2959 static __initdata unsigned long dhash_entries;
2960 static int __init set_dhash_entries(char *str)
2964 dhash_entries = simple_strtoul(str, &str, 0);
2967 __setup("dhash_entries=", set_dhash_entries);
2969 static void __init dcache_init_early(void)
2973 /* If hashes are distributed across NUMA nodes, defer
2974 * hash allocation until vmalloc space is available.
2980 alloc_large_system_hash("Dentry cache",
2981 sizeof(struct hlist_bl_head),
2989 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2990 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
2993 static void __init dcache_init(void)
2998 * A constructor could be added for stable state like the lists,
2999 * but it is probably not worth it because of the cache nature
3002 dentry_cache = KMEM_CACHE(dentry,
3003 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3005 /* Hash may have been set up in dcache_init_early */
3010 alloc_large_system_hash("Dentry cache",
3011 sizeof(struct hlist_bl_head),
3019 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3020 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3023 /* SLAB cache for __getname() consumers */
3024 struct kmem_cache *names_cachep __read_mostly;
3025 EXPORT_SYMBOL(names_cachep);
3027 EXPORT_SYMBOL(d_genocide);
3029 void __init vfs_caches_init_early(void)
3031 dcache_init_early();
3035 void __init vfs_caches_init(unsigned long mempages)
3037 unsigned long reserve;
3039 /* Base hash sizes on available memory, with a reserve equal to
3040 150% of current kernel size */
3042 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3043 mempages -= reserve;
3045 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3046 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3050 files_init(mempages);