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_sb->s_nr_dentry_unused--;
246 dentry_stat.nr_unused--;
250 * Remove a dentry with references from the LRU.
252 static void dentry_lru_del(struct dentry *dentry)
254 if (!list_empty(&dentry->d_lru)) {
255 spin_lock(&dcache_lru_lock);
256 __dentry_lru_del(dentry);
257 spin_unlock(&dcache_lru_lock);
262 * Remove a dentry that is unreferenced and about to be pruned
263 * (unhashed and destroyed) from the LRU, and inform the file system.
264 * This wrapper should be called _prior_ to unhashing a victim dentry.
266 static void dentry_lru_prune(struct dentry *dentry)
268 if (!list_empty(&dentry->d_lru)) {
269 if (dentry->d_flags & DCACHE_OP_PRUNE)
270 dentry->d_op->d_prune(dentry);
272 spin_lock(&dcache_lru_lock);
273 __dentry_lru_del(dentry);
274 spin_unlock(&dcache_lru_lock);
278 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
280 spin_lock(&dcache_lru_lock);
281 if (list_empty(&dentry->d_lru)) {
282 list_add_tail(&dentry->d_lru, list);
283 dentry->d_sb->s_nr_dentry_unused++;
284 dentry_stat.nr_unused++;
286 list_move_tail(&dentry->d_lru, list);
288 spin_unlock(&dcache_lru_lock);
292 * d_kill - kill dentry and return parent
293 * @dentry: dentry to kill
294 * @parent: parent dentry
296 * The dentry must already be unhashed and removed from the LRU.
298 * If this is the root of the dentry tree, return NULL.
300 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
303 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
304 __releases(dentry->d_lock)
305 __releases(parent->d_lock)
306 __releases(dentry->d_inode->i_lock)
308 list_del(&dentry->d_u.d_child);
310 * Inform try_to_ascend() that we are no longer attached to the
313 dentry->d_flags |= DCACHE_DISCONNECTED;
315 spin_unlock(&parent->d_lock);
318 * dentry_iput drops the locks, at which point nobody (except
319 * transient RCU lookups) can reach this dentry.
326 * Unhash a dentry without inserting an RCU walk barrier or checking that
327 * dentry->d_lock is locked. The caller must take care of that, if
330 static void __d_shrink(struct dentry *dentry)
332 if (!d_unhashed(dentry)) {
333 struct hlist_bl_head *b;
334 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
335 b = &dentry->d_sb->s_anon;
337 b = d_hash(dentry->d_parent, dentry->d_name.hash);
340 __hlist_bl_del(&dentry->d_hash);
341 dentry->d_hash.pprev = NULL;
347 * d_drop - drop a dentry
348 * @dentry: dentry to drop
350 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
351 * be found through a VFS lookup any more. Note that this is different from
352 * deleting the dentry - d_delete will try to mark the dentry negative if
353 * possible, giving a successful _negative_ lookup, while d_drop will
354 * just make the cache lookup fail.
356 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
357 * reason (NFS timeouts or autofs deletes).
359 * __d_drop requires dentry->d_lock.
361 void __d_drop(struct dentry *dentry)
363 if (!d_unhashed(dentry)) {
365 dentry_rcuwalk_barrier(dentry);
368 EXPORT_SYMBOL(__d_drop);
370 void d_drop(struct dentry *dentry)
372 spin_lock(&dentry->d_lock);
374 spin_unlock(&dentry->d_lock);
376 EXPORT_SYMBOL(d_drop);
379 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
380 * @dentry: dentry to drop
382 * This is called when we do a lookup on a placeholder dentry that needed to be
383 * looked up. The dentry should have been hashed in order for it to be found by
384 * the lookup code, but now needs to be unhashed while we do the actual lookup
385 * and clear the DCACHE_NEED_LOOKUP flag.
387 void d_clear_need_lookup(struct dentry *dentry)
389 spin_lock(&dentry->d_lock);
391 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
392 spin_unlock(&dentry->d_lock);
394 EXPORT_SYMBOL(d_clear_need_lookup);
397 * Finish off a dentry we've decided to kill.
398 * dentry->d_lock must be held, returns with it unlocked.
399 * If ref is non-zero, then decrement the refcount too.
400 * Returns dentry requiring refcount drop, or NULL if we're done.
402 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
403 __releases(dentry->d_lock)
406 struct dentry *parent;
408 inode = dentry->d_inode;
409 if (inode && !spin_trylock(&inode->i_lock)) {
411 spin_unlock(&dentry->d_lock);
413 return dentry; /* try again with same dentry */
418 parent = dentry->d_parent;
419 if (parent && !spin_trylock(&parent->d_lock)) {
421 spin_unlock(&inode->i_lock);
428 * if dentry was on the d_lru list delete it from there.
429 * inform the fs via d_prune that this dentry is about to be
430 * unhashed and destroyed.
432 dentry_lru_prune(dentry);
433 /* if it was on the hash then remove it */
435 return d_kill(dentry, parent);
441 * This is complicated by the fact that we do not want to put
442 * dentries that are no longer on any hash chain on the unused
443 * list: we'd much rather just get rid of them immediately.
445 * However, that implies that we have to traverse the dentry
446 * tree upwards to the parents which might _also_ now be
447 * scheduled for deletion (it may have been only waiting for
448 * its last child to go away).
450 * This tail recursion is done by hand as we don't want to depend
451 * on the compiler to always get this right (gcc generally doesn't).
452 * Real recursion would eat up our stack space.
456 * dput - release a dentry
457 * @dentry: dentry to release
459 * Release a dentry. This will drop the usage count and if appropriate
460 * call the dentry unlink method as well as removing it from the queues and
461 * releasing its resources. If the parent dentries were scheduled for release
462 * they too may now get deleted.
464 void dput(struct dentry *dentry)
470 if (dentry->d_count == 1)
472 spin_lock(&dentry->d_lock);
473 BUG_ON(!dentry->d_count);
474 if (dentry->d_count > 1) {
476 spin_unlock(&dentry->d_lock);
480 if (dentry->d_flags & DCACHE_OP_DELETE) {
481 if (dentry->d_op->d_delete(dentry))
485 /* Unreachable? Get rid of it */
486 if (d_unhashed(dentry))
490 * If this dentry needs lookup, don't set the referenced flag so that it
491 * is more likely to be cleaned up by the dcache shrinker in case of
494 if (!d_need_lookup(dentry))
495 dentry->d_flags |= DCACHE_REFERENCED;
496 dentry_lru_add(dentry);
499 spin_unlock(&dentry->d_lock);
503 dentry = dentry_kill(dentry, 1);
510 * d_invalidate - invalidate a dentry
511 * @dentry: dentry to invalidate
513 * Try to invalidate the dentry if it turns out to be
514 * possible. If there are other dentries that can be
515 * reached through this one we can't delete it and we
516 * return -EBUSY. On success we return 0.
521 int d_invalidate(struct dentry * dentry)
524 * If it's already been dropped, return OK.
526 spin_lock(&dentry->d_lock);
527 if (d_unhashed(dentry)) {
528 spin_unlock(&dentry->d_lock);
532 * Check whether to do a partial shrink_dcache
533 * to get rid of unused child entries.
535 if (!list_empty(&dentry->d_subdirs)) {
536 spin_unlock(&dentry->d_lock);
537 shrink_dcache_parent(dentry);
538 spin_lock(&dentry->d_lock);
542 * Somebody else still using it?
544 * If it's a directory, we can't drop it
545 * for fear of somebody re-populating it
546 * with children (even though dropping it
547 * would make it unreachable from the root,
548 * we might still populate it if it was a
549 * working directory or similar).
550 * We also need to leave mountpoints alone,
553 if (dentry->d_count > 1 && dentry->d_inode) {
554 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
555 spin_unlock(&dentry->d_lock);
561 spin_unlock(&dentry->d_lock);
564 EXPORT_SYMBOL(d_invalidate);
566 /* This must be called with d_lock held */
567 static inline void __dget_dlock(struct dentry *dentry)
572 static inline void __dget(struct dentry *dentry)
574 spin_lock(&dentry->d_lock);
575 __dget_dlock(dentry);
576 spin_unlock(&dentry->d_lock);
579 struct dentry *dget_parent(struct dentry *dentry)
585 * Don't need rcu_dereference because we re-check it was correct under
589 ret = dentry->d_parent;
590 spin_lock(&ret->d_lock);
591 if (unlikely(ret != dentry->d_parent)) {
592 spin_unlock(&ret->d_lock);
597 BUG_ON(!ret->d_count);
599 spin_unlock(&ret->d_lock);
602 EXPORT_SYMBOL(dget_parent);
605 * d_find_alias - grab a hashed alias of inode
606 * @inode: inode in question
607 * @want_discon: flag, used by d_splice_alias, to request
608 * that only a DISCONNECTED alias be returned.
610 * If inode has a hashed alias, or is a directory and has any alias,
611 * acquire the reference to alias and return it. Otherwise return NULL.
612 * Notice that if inode is a directory there can be only one alias and
613 * it can be unhashed only if it has no children, or if it is the root
616 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
617 * any other hashed alias over that one unless @want_discon is set,
618 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
620 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
622 struct dentry *alias, *discon_alias;
626 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
627 spin_lock(&alias->d_lock);
628 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
629 if (IS_ROOT(alias) &&
630 (alias->d_flags & DCACHE_DISCONNECTED)) {
631 discon_alias = alias;
632 } else if (!want_discon) {
634 spin_unlock(&alias->d_lock);
638 spin_unlock(&alias->d_lock);
641 alias = discon_alias;
642 spin_lock(&alias->d_lock);
643 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
644 if (IS_ROOT(alias) &&
645 (alias->d_flags & DCACHE_DISCONNECTED)) {
647 spin_unlock(&alias->d_lock);
651 spin_unlock(&alias->d_lock);
657 struct dentry *d_find_alias(struct inode *inode)
659 struct dentry *de = NULL;
661 if (!list_empty(&inode->i_dentry)) {
662 spin_lock(&inode->i_lock);
663 de = __d_find_alias(inode, 0);
664 spin_unlock(&inode->i_lock);
668 EXPORT_SYMBOL(d_find_alias);
671 * Try to kill dentries associated with this inode.
672 * WARNING: you must own a reference to inode.
674 void d_prune_aliases(struct inode *inode)
676 struct dentry *dentry;
678 spin_lock(&inode->i_lock);
679 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
680 spin_lock(&dentry->d_lock);
681 if (!dentry->d_count) {
682 __dget_dlock(dentry);
684 spin_unlock(&dentry->d_lock);
685 spin_unlock(&inode->i_lock);
689 spin_unlock(&dentry->d_lock);
691 spin_unlock(&inode->i_lock);
693 EXPORT_SYMBOL(d_prune_aliases);
696 * Try to throw away a dentry - free the inode, dput the parent.
697 * Requires dentry->d_lock is held, and dentry->d_count == 0.
698 * Releases dentry->d_lock.
700 * This may fail if locks cannot be acquired no problem, just try again.
702 static void try_prune_one_dentry(struct dentry *dentry)
703 __releases(dentry->d_lock)
705 struct dentry *parent;
707 parent = dentry_kill(dentry, 0);
709 * If dentry_kill returns NULL, we have nothing more to do.
710 * if it returns the same dentry, trylocks failed. In either
711 * case, just loop again.
713 * Otherwise, we need to prune ancestors too. This is necessary
714 * to prevent quadratic behavior of shrink_dcache_parent(), but
715 * is also expected to be beneficial in reducing dentry cache
720 if (parent == dentry)
723 /* Prune ancestors. */
726 spin_lock(&dentry->d_lock);
727 if (dentry->d_count > 1) {
729 spin_unlock(&dentry->d_lock);
732 dentry = dentry_kill(dentry, 1);
736 static void shrink_dentry_list(struct list_head *list)
738 struct dentry *dentry;
742 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
743 if (&dentry->d_lru == list)
745 spin_lock(&dentry->d_lock);
746 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
747 spin_unlock(&dentry->d_lock);
752 * We found an inuse dentry which was not removed from
753 * the LRU because of laziness during lookup. Do not free
754 * it - just keep it off the LRU list.
756 if (dentry->d_count) {
757 dentry_lru_del(dentry);
758 spin_unlock(&dentry->d_lock);
764 try_prune_one_dentry(dentry);
772 * prune_dcache_sb - shrink the dcache
774 * @count: number of entries to try to free
776 * Attempt to shrink the superblock dcache LRU by @count entries. This is
777 * done when we need more memory an called from the superblock shrinker
780 * This function may fail to free any resources if all the dentries are in
783 void prune_dcache_sb(struct super_block *sb, int count)
785 struct dentry *dentry;
786 LIST_HEAD(referenced);
790 spin_lock(&dcache_lru_lock);
791 while (!list_empty(&sb->s_dentry_lru)) {
792 dentry = list_entry(sb->s_dentry_lru.prev,
793 struct dentry, d_lru);
794 BUG_ON(dentry->d_sb != sb);
796 if (!spin_trylock(&dentry->d_lock)) {
797 spin_unlock(&dcache_lru_lock);
802 if (dentry->d_flags & DCACHE_REFERENCED) {
803 dentry->d_flags &= ~DCACHE_REFERENCED;
804 list_move(&dentry->d_lru, &referenced);
805 spin_unlock(&dentry->d_lock);
807 list_move_tail(&dentry->d_lru, &tmp);
808 spin_unlock(&dentry->d_lock);
812 cond_resched_lock(&dcache_lru_lock);
814 if (!list_empty(&referenced))
815 list_splice(&referenced, &sb->s_dentry_lru);
816 spin_unlock(&dcache_lru_lock);
818 shrink_dentry_list(&tmp);
822 * shrink_dcache_sb - shrink dcache for a superblock
825 * Shrink the dcache for the specified super block. This is used to free
826 * the dcache before unmounting a file system.
828 void shrink_dcache_sb(struct super_block *sb)
832 spin_lock(&dcache_lru_lock);
833 while (!list_empty(&sb->s_dentry_lru)) {
834 list_splice_init(&sb->s_dentry_lru, &tmp);
835 spin_unlock(&dcache_lru_lock);
836 shrink_dentry_list(&tmp);
837 spin_lock(&dcache_lru_lock);
839 spin_unlock(&dcache_lru_lock);
841 EXPORT_SYMBOL(shrink_dcache_sb);
844 * destroy a single subtree of dentries for unmount
845 * - see the comments on shrink_dcache_for_umount() for a description of the
848 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
850 struct dentry *parent;
852 BUG_ON(!IS_ROOT(dentry));
855 /* descend to the first leaf in the current subtree */
856 while (!list_empty(&dentry->d_subdirs))
857 dentry = list_entry(dentry->d_subdirs.next,
858 struct dentry, d_u.d_child);
860 /* consume the dentries from this leaf up through its parents
861 * until we find one with children or run out altogether */
866 * remove the dentry from the lru, and inform
867 * the fs that this dentry is about to be
868 * unhashed and destroyed.
870 dentry_lru_prune(dentry);
873 if (dentry->d_count != 0) {
875 "BUG: Dentry %p{i=%lx,n=%s}"
877 " [unmount of %s %s]\n",
880 dentry->d_inode->i_ino : 0UL,
883 dentry->d_sb->s_type->name,
888 if (IS_ROOT(dentry)) {
890 list_del(&dentry->d_u.d_child);
892 parent = dentry->d_parent;
894 list_del(&dentry->d_u.d_child);
897 inode = dentry->d_inode;
899 dentry->d_inode = NULL;
900 list_del_init(&dentry->d_alias);
901 if (dentry->d_op && dentry->d_op->d_iput)
902 dentry->d_op->d_iput(dentry, inode);
909 /* finished when we fall off the top of the tree,
910 * otherwise we ascend to the parent and move to the
911 * next sibling if there is one */
915 } while (list_empty(&dentry->d_subdirs));
917 dentry = list_entry(dentry->d_subdirs.next,
918 struct dentry, d_u.d_child);
923 * destroy the dentries attached to a superblock on unmounting
924 * - we don't need to use dentry->d_lock because:
925 * - the superblock is detached from all mountings and open files, so the
926 * dentry trees will not be rearranged by the VFS
927 * - s_umount is write-locked, so the memory pressure shrinker will ignore
928 * any dentries belonging to this superblock that it comes across
929 * - the filesystem itself is no longer permitted to rearrange the dentries
932 void shrink_dcache_for_umount(struct super_block *sb)
934 struct dentry *dentry;
936 if (down_read_trylock(&sb->s_umount))
942 shrink_dcache_for_umount_subtree(dentry);
944 while (!hlist_bl_empty(&sb->s_anon)) {
945 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
946 shrink_dcache_for_umount_subtree(dentry);
951 * This tries to ascend one level of parenthood, but
952 * we can race with renaming, so we need to re-check
953 * the parenthood after dropping the lock and check
954 * that the sequence number still matches.
956 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
958 struct dentry *new = old->d_parent;
961 spin_unlock(&old->d_lock);
962 spin_lock(&new->d_lock);
965 * might go back up the wrong parent if we have had a rename
968 if (new != old->d_parent ||
969 (old->d_flags & DCACHE_DISCONNECTED) ||
970 (!locked && read_seqretry(&rename_lock, seq))) {
971 spin_unlock(&new->d_lock);
980 * Search for at least 1 mount point in the dentry's subdirs.
981 * We descend to the next level whenever the d_subdirs
982 * list is non-empty and continue searching.
986 * have_submounts - check for mounts over a dentry
987 * @parent: dentry to check.
989 * Return true if the parent or its subdirectories contain
992 int have_submounts(struct dentry *parent)
994 struct dentry *this_parent;
995 struct list_head *next;
999 seq = read_seqbegin(&rename_lock);
1001 this_parent = parent;
1003 if (d_mountpoint(parent))
1005 spin_lock(&this_parent->d_lock);
1007 next = this_parent->d_subdirs.next;
1009 while (next != &this_parent->d_subdirs) {
1010 struct list_head *tmp = next;
1011 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1014 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1015 /* Have we found a mount point ? */
1016 if (d_mountpoint(dentry)) {
1017 spin_unlock(&dentry->d_lock);
1018 spin_unlock(&this_parent->d_lock);
1021 if (!list_empty(&dentry->d_subdirs)) {
1022 spin_unlock(&this_parent->d_lock);
1023 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1024 this_parent = dentry;
1025 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1028 spin_unlock(&dentry->d_lock);
1031 * All done at this level ... ascend and resume the search.
1033 if (this_parent != parent) {
1034 struct dentry *child = this_parent;
1035 this_parent = try_to_ascend(this_parent, locked, seq);
1038 next = child->d_u.d_child.next;
1041 spin_unlock(&this_parent->d_lock);
1042 if (!locked && read_seqretry(&rename_lock, seq))
1045 write_sequnlock(&rename_lock);
1046 return 0; /* No mount points found in tree */
1048 if (!locked && read_seqretry(&rename_lock, seq))
1051 write_sequnlock(&rename_lock);
1056 write_seqlock(&rename_lock);
1059 EXPORT_SYMBOL(have_submounts);
1062 * Search the dentry child list for the specified parent,
1063 * and move any unused dentries to the end of the unused
1064 * list for prune_dcache(). We descend to the next level
1065 * whenever the d_subdirs list is non-empty and continue
1068 * It returns zero iff there are no unused children,
1069 * otherwise it returns the number of children moved to
1070 * the end of the unused list. This may not be the total
1071 * number of unused children, because select_parent can
1072 * drop the lock and return early due to latency
1075 static int select_parent(struct dentry *parent, struct list_head *dispose)
1077 struct dentry *this_parent;
1078 struct list_head *next;
1083 seq = read_seqbegin(&rename_lock);
1085 this_parent = parent;
1086 spin_lock(&this_parent->d_lock);
1088 next = this_parent->d_subdirs.next;
1090 while (next != &this_parent->d_subdirs) {
1091 struct list_head *tmp = next;
1092 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1095 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1098 * move only zero ref count dentries to the dispose list.
1100 if (!dentry->d_count) {
1101 dentry_lru_move_list(dentry, dispose);
1104 dentry_lru_del(dentry);
1108 * We can return to the caller if we have found some (this
1109 * ensures forward progress). We'll be coming back to find
1112 if (found && need_resched()) {
1113 spin_unlock(&dentry->d_lock);
1118 * Descend a level if the d_subdirs list is non-empty.
1120 if (!list_empty(&dentry->d_subdirs)) {
1121 spin_unlock(&this_parent->d_lock);
1122 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1123 this_parent = dentry;
1124 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1128 spin_unlock(&dentry->d_lock);
1131 * All done at this level ... ascend and resume the search.
1133 if (this_parent != parent) {
1134 struct dentry *child = this_parent;
1135 this_parent = try_to_ascend(this_parent, locked, seq);
1138 next = child->d_u.d_child.next;
1142 spin_unlock(&this_parent->d_lock);
1143 if (!locked && read_seqretry(&rename_lock, seq))
1146 write_sequnlock(&rename_lock);
1153 write_seqlock(&rename_lock);
1158 * shrink_dcache_parent - prune dcache
1159 * @parent: parent of entries to prune
1161 * Prune the dcache to remove unused children of the parent dentry.
1163 void shrink_dcache_parent(struct dentry * parent)
1168 while ((found = select_parent(parent, &dispose)) != 0)
1169 shrink_dentry_list(&dispose);
1171 EXPORT_SYMBOL(shrink_dcache_parent);
1174 * __d_alloc - allocate a dcache entry
1175 * @sb: filesystem it will belong to
1176 * @name: qstr of the name
1178 * Allocates a dentry. It returns %NULL if there is insufficient memory
1179 * available. On a success the dentry is returned. The name passed in is
1180 * copied and the copy passed in may be reused after this call.
1183 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1185 struct dentry *dentry;
1188 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1192 if (name->len > DNAME_INLINE_LEN-1) {
1193 dname = kmalloc(name->len + 1, GFP_KERNEL);
1195 kmem_cache_free(dentry_cache, dentry);
1199 dname = dentry->d_iname;
1201 dentry->d_name.name = dname;
1203 dentry->d_name.len = name->len;
1204 dentry->d_name.hash = name->hash;
1205 memcpy(dname, name->name, name->len);
1206 dname[name->len] = 0;
1208 dentry->d_count = 1;
1209 dentry->d_flags = 0;
1210 spin_lock_init(&dentry->d_lock);
1211 seqcount_init(&dentry->d_seq);
1212 dentry->d_inode = NULL;
1213 dentry->d_parent = dentry;
1215 dentry->d_op = NULL;
1216 dentry->d_fsdata = NULL;
1217 INIT_HLIST_BL_NODE(&dentry->d_hash);
1218 INIT_LIST_HEAD(&dentry->d_lru);
1219 INIT_LIST_HEAD(&dentry->d_subdirs);
1220 INIT_LIST_HEAD(&dentry->d_alias);
1221 INIT_LIST_HEAD(&dentry->d_u.d_child);
1222 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1224 this_cpu_inc(nr_dentry);
1230 * d_alloc - allocate a dcache entry
1231 * @parent: parent of entry to allocate
1232 * @name: qstr of the name
1234 * Allocates a dentry. It returns %NULL if there is insufficient memory
1235 * available. On a success the dentry is returned. The name passed in is
1236 * copied and the copy passed in may be reused after this call.
1238 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1240 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1244 spin_lock(&parent->d_lock);
1246 * don't need child lock because it is not subject
1247 * to concurrency here
1249 __dget_dlock(parent);
1250 dentry->d_parent = parent;
1251 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1252 spin_unlock(&parent->d_lock);
1256 EXPORT_SYMBOL(d_alloc);
1258 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1260 struct dentry *dentry = __d_alloc(sb, name);
1262 dentry->d_flags |= DCACHE_DISCONNECTED;
1265 EXPORT_SYMBOL(d_alloc_pseudo);
1267 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1272 q.len = strlen(name);
1273 q.hash = full_name_hash(q.name, q.len);
1274 return d_alloc(parent, &q);
1276 EXPORT_SYMBOL(d_alloc_name);
1278 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1280 WARN_ON_ONCE(dentry->d_op);
1281 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1283 DCACHE_OP_REVALIDATE |
1284 DCACHE_OP_DELETE ));
1289 dentry->d_flags |= DCACHE_OP_HASH;
1291 dentry->d_flags |= DCACHE_OP_COMPARE;
1292 if (op->d_revalidate)
1293 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1295 dentry->d_flags |= DCACHE_OP_DELETE;
1297 dentry->d_flags |= DCACHE_OP_PRUNE;
1300 EXPORT_SYMBOL(d_set_d_op);
1302 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1304 spin_lock(&dentry->d_lock);
1306 if (unlikely(IS_AUTOMOUNT(inode)))
1307 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1308 list_add(&dentry->d_alias, &inode->i_dentry);
1310 dentry->d_inode = inode;
1311 dentry_rcuwalk_barrier(dentry);
1312 spin_unlock(&dentry->d_lock);
1313 fsnotify_d_instantiate(dentry, inode);
1317 * d_instantiate - fill in inode information for a dentry
1318 * @entry: dentry to complete
1319 * @inode: inode to attach to this dentry
1321 * Fill in inode information in the entry.
1323 * This turns negative dentries into productive full members
1326 * NOTE! This assumes that the inode count has been incremented
1327 * (or otherwise set) by the caller to indicate that it is now
1328 * in use by the dcache.
1331 void d_instantiate(struct dentry *entry, struct inode * inode)
1333 BUG_ON(!list_empty(&entry->d_alias));
1335 spin_lock(&inode->i_lock);
1336 __d_instantiate(entry, inode);
1338 spin_unlock(&inode->i_lock);
1339 security_d_instantiate(entry, inode);
1341 EXPORT_SYMBOL(d_instantiate);
1344 * d_instantiate_unique - instantiate a non-aliased dentry
1345 * @entry: dentry to instantiate
1346 * @inode: inode to attach to this dentry
1348 * Fill in inode information in the entry. On success, it returns NULL.
1349 * If an unhashed alias of "entry" already exists, then we return the
1350 * aliased dentry instead and drop one reference to inode.
1352 * Note that in order to avoid conflicts with rename() etc, the caller
1353 * had better be holding the parent directory semaphore.
1355 * This also assumes that the inode count has been incremented
1356 * (or otherwise set) by the caller to indicate that it is now
1357 * in use by the dcache.
1359 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1360 struct inode *inode)
1362 struct dentry *alias;
1363 int len = entry->d_name.len;
1364 const char *name = entry->d_name.name;
1365 unsigned int hash = entry->d_name.hash;
1368 __d_instantiate(entry, NULL);
1372 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1373 struct qstr *qstr = &alias->d_name;
1376 * Don't need alias->d_lock here, because aliases with
1377 * d_parent == entry->d_parent are not subject to name or
1378 * parent changes, because the parent inode i_mutex is held.
1380 if (qstr->hash != hash)
1382 if (alias->d_parent != entry->d_parent)
1384 if (dentry_cmp(qstr->name, qstr->len, name, len))
1390 __d_instantiate(entry, inode);
1394 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1396 struct dentry *result;
1398 BUG_ON(!list_empty(&entry->d_alias));
1401 spin_lock(&inode->i_lock);
1402 result = __d_instantiate_unique(entry, inode);
1404 spin_unlock(&inode->i_lock);
1407 security_d_instantiate(entry, inode);
1411 BUG_ON(!d_unhashed(result));
1416 EXPORT_SYMBOL(d_instantiate_unique);
1419 * d_alloc_root - allocate root dentry
1420 * @root_inode: inode to allocate the root for
1422 * Allocate a root ("/") dentry for the inode given. The inode is
1423 * instantiated and returned. %NULL is returned if there is insufficient
1424 * memory or the inode passed is %NULL.
1427 struct dentry * d_alloc_root(struct inode * root_inode)
1429 struct dentry *res = NULL;
1432 static const struct qstr name = { .name = "/", .len = 1 };
1434 res = __d_alloc(root_inode->i_sb, &name);
1436 d_instantiate(res, root_inode);
1440 EXPORT_SYMBOL(d_alloc_root);
1442 static struct dentry * __d_find_any_alias(struct inode *inode)
1444 struct dentry *alias;
1446 if (list_empty(&inode->i_dentry))
1448 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1453 static struct dentry * d_find_any_alias(struct inode *inode)
1457 spin_lock(&inode->i_lock);
1458 de = __d_find_any_alias(inode);
1459 spin_unlock(&inode->i_lock);
1465 * d_obtain_alias - find or allocate a dentry for a given inode
1466 * @inode: inode to allocate the dentry for
1468 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1469 * similar open by handle operations. The returned dentry may be anonymous,
1470 * or may have a full name (if the inode was already in the cache).
1472 * When called on a directory inode, we must ensure that the inode only ever
1473 * has one dentry. If a dentry is found, that is returned instead of
1474 * allocating a new one.
1476 * On successful return, the reference to the inode has been transferred
1477 * to the dentry. In case of an error the reference on the inode is released.
1478 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1479 * be passed in and will be the error will be propagate to the return value,
1480 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1482 struct dentry *d_obtain_alias(struct inode *inode)
1484 static const struct qstr anonstring = { .name = "" };
1489 return ERR_PTR(-ESTALE);
1491 return ERR_CAST(inode);
1493 res = d_find_any_alias(inode);
1497 tmp = __d_alloc(inode->i_sb, &anonstring);
1499 res = ERR_PTR(-ENOMEM);
1503 spin_lock(&inode->i_lock);
1504 res = __d_find_any_alias(inode);
1506 spin_unlock(&inode->i_lock);
1511 /* attach a disconnected dentry */
1512 spin_lock(&tmp->d_lock);
1513 tmp->d_inode = inode;
1514 tmp->d_flags |= DCACHE_DISCONNECTED;
1515 list_add(&tmp->d_alias, &inode->i_dentry);
1516 hlist_bl_lock(&tmp->d_sb->s_anon);
1517 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1518 hlist_bl_unlock(&tmp->d_sb->s_anon);
1519 spin_unlock(&tmp->d_lock);
1520 spin_unlock(&inode->i_lock);
1521 security_d_instantiate(tmp, inode);
1526 if (res && !IS_ERR(res))
1527 security_d_instantiate(res, inode);
1531 EXPORT_SYMBOL(d_obtain_alias);
1534 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1535 * @inode: the inode which may have a disconnected dentry
1536 * @dentry: a negative dentry which we want to point to the inode.
1538 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1539 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1540 * and return it, else simply d_add the inode to the dentry and return NULL.
1542 * This is needed in the lookup routine of any filesystem that is exportable
1543 * (via knfsd) so that we can build dcache paths to directories effectively.
1545 * If a dentry was found and moved, then it is returned. Otherwise NULL
1546 * is returned. This matches the expected return value of ->lookup.
1549 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1551 struct dentry *new = NULL;
1554 return ERR_CAST(inode);
1556 if (inode && S_ISDIR(inode->i_mode)) {
1557 spin_lock(&inode->i_lock);
1558 new = __d_find_alias(inode, 1);
1560 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1561 spin_unlock(&inode->i_lock);
1562 security_d_instantiate(new, inode);
1563 d_move(new, dentry);
1566 /* already taking inode->i_lock, so d_add() by hand */
1567 __d_instantiate(dentry, inode);
1568 spin_unlock(&inode->i_lock);
1569 security_d_instantiate(dentry, inode);
1573 d_add(dentry, inode);
1576 EXPORT_SYMBOL(d_splice_alias);
1579 * d_add_ci - lookup or allocate new dentry with case-exact name
1580 * @inode: the inode case-insensitive lookup has found
1581 * @dentry: the negative dentry that was passed to the parent's lookup func
1582 * @name: the case-exact name to be associated with the returned dentry
1584 * This is to avoid filling the dcache with case-insensitive names to the
1585 * same inode, only the actual correct case is stored in the dcache for
1586 * case-insensitive filesystems.
1588 * For a case-insensitive lookup match and if the the case-exact dentry
1589 * already exists in in the dcache, use it and return it.
1591 * If no entry exists with the exact case name, allocate new dentry with
1592 * the exact case, and return the spliced entry.
1594 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1598 struct dentry *found;
1602 * First check if a dentry matching the name already exists,
1603 * if not go ahead and create it now.
1605 found = d_hash_and_lookup(dentry->d_parent, name);
1607 new = d_alloc(dentry->d_parent, name);
1613 found = d_splice_alias(inode, new);
1622 * If a matching dentry exists, and it's not negative use it.
1624 * Decrement the reference count to balance the iget() done
1627 if (found->d_inode) {
1628 if (unlikely(found->d_inode != inode)) {
1629 /* This can't happen because bad inodes are unhashed. */
1630 BUG_ON(!is_bad_inode(inode));
1631 BUG_ON(!is_bad_inode(found->d_inode));
1638 * We are going to instantiate this dentry, unhash it and clear the
1639 * lookup flag so we can do that.
1641 if (unlikely(d_need_lookup(found)))
1642 d_clear_need_lookup(found);
1645 * Negative dentry: instantiate it unless the inode is a directory and
1646 * already has a dentry.
1648 new = d_splice_alias(inode, found);
1657 return ERR_PTR(error);
1659 EXPORT_SYMBOL(d_add_ci);
1662 * __d_lookup_rcu - search for a dentry (racy, store-free)
1663 * @parent: parent dentry
1664 * @name: qstr of name we wish to find
1665 * @seq: returns d_seq value at the point where the dentry was found
1666 * @inode: returns dentry->d_inode when the inode was found valid.
1667 * Returns: dentry, or NULL
1669 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1670 * resolution (store-free path walking) design described in
1671 * Documentation/filesystems/path-lookup.txt.
1673 * This is not to be used outside core vfs.
1675 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1676 * held, and rcu_read_lock held. The returned dentry must not be stored into
1677 * without taking d_lock and checking d_seq sequence count against @seq
1680 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1683 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1684 * the returned dentry, so long as its parent's seqlock is checked after the
1685 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1686 * is formed, giving integrity down the path walk.
1688 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1689 unsigned *seq, struct inode **inode)
1691 unsigned int len = name->len;
1692 unsigned int hash = name->hash;
1693 const unsigned char *str = name->name;
1694 struct hlist_bl_head *b = d_hash(parent, hash);
1695 struct hlist_bl_node *node;
1696 struct dentry *dentry;
1699 * Note: There is significant duplication with __d_lookup_rcu which is
1700 * required to prevent single threaded performance regressions
1701 * especially on architectures where smp_rmb (in seqcounts) are costly.
1702 * Keep the two functions in sync.
1706 * The hash list is protected using RCU.
1708 * Carefully use d_seq when comparing a candidate dentry, to avoid
1709 * races with d_move().
1711 * It is possible that concurrent renames can mess up our list
1712 * walk here and result in missing our dentry, resulting in the
1713 * false-negative result. d_lookup() protects against concurrent
1714 * renames using rename_lock seqlock.
1716 * See Documentation/filesystems/path-lookup.txt for more details.
1718 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1723 if (dentry->d_name.hash != hash)
1727 *seq = read_seqcount_begin(&dentry->d_seq);
1728 if (dentry->d_parent != parent)
1730 if (d_unhashed(dentry))
1732 tlen = dentry->d_name.len;
1733 tname = dentry->d_name.name;
1734 i = dentry->d_inode;
1737 * This seqcount check is required to ensure name and
1738 * len are loaded atomically, so as not to walk off the
1739 * edge of memory when walking. If we could load this
1740 * atomically some other way, we could drop this check.
1742 if (read_seqcount_retry(&dentry->d_seq, *seq))
1744 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1745 if (parent->d_op->d_compare(parent, *inode,
1750 if (dentry_cmp(tname, tlen, str, len))
1754 * No extra seqcount check is required after the name
1755 * compare. The caller must perform a seqcount check in
1756 * order to do anything useful with the returned dentry
1766 * d_lookup - search for a dentry
1767 * @parent: parent dentry
1768 * @name: qstr of name we wish to find
1769 * Returns: dentry, or NULL
1771 * d_lookup searches the children of the parent dentry for the name in
1772 * question. If the dentry is found its reference count is incremented and the
1773 * dentry is returned. The caller must use dput to free the entry when it has
1774 * finished using it. %NULL is returned if the dentry does not exist.
1776 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1778 struct dentry *dentry;
1782 seq = read_seqbegin(&rename_lock);
1783 dentry = __d_lookup(parent, name);
1786 } while (read_seqretry(&rename_lock, seq));
1789 EXPORT_SYMBOL(d_lookup);
1792 * __d_lookup - search for a dentry (racy)
1793 * @parent: parent dentry
1794 * @name: qstr of name we wish to find
1795 * Returns: dentry, or NULL
1797 * __d_lookup is like d_lookup, however it may (rarely) return a
1798 * false-negative result due to unrelated rename activity.
1800 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1801 * however it must be used carefully, eg. with a following d_lookup in
1802 * the case of failure.
1804 * __d_lookup callers must be commented.
1806 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1808 unsigned int len = name->len;
1809 unsigned int hash = name->hash;
1810 const unsigned char *str = name->name;
1811 struct hlist_bl_head *b = d_hash(parent, hash);
1812 struct hlist_bl_node *node;
1813 struct dentry *found = NULL;
1814 struct dentry *dentry;
1817 * Note: There is significant duplication with __d_lookup_rcu which is
1818 * required to prevent single threaded performance regressions
1819 * especially on architectures where smp_rmb (in seqcounts) are costly.
1820 * Keep the two functions in sync.
1824 * The hash list is protected using RCU.
1826 * Take d_lock when comparing a candidate dentry, to avoid races
1829 * It is possible that concurrent renames can mess up our list
1830 * walk here and result in missing our dentry, resulting in the
1831 * false-negative result. d_lookup() protects against concurrent
1832 * renames using rename_lock seqlock.
1834 * See Documentation/filesystems/path-lookup.txt for more details.
1838 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1842 if (dentry->d_name.hash != hash)
1845 spin_lock(&dentry->d_lock);
1846 if (dentry->d_parent != parent)
1848 if (d_unhashed(dentry))
1852 * It is safe to compare names since d_move() cannot
1853 * change the qstr (protected by d_lock).
1855 tlen = dentry->d_name.len;
1856 tname = dentry->d_name.name;
1857 if (parent->d_flags & DCACHE_OP_COMPARE) {
1858 if (parent->d_op->d_compare(parent, parent->d_inode,
1859 dentry, dentry->d_inode,
1863 if (dentry_cmp(tname, tlen, str, len))
1869 spin_unlock(&dentry->d_lock);
1872 spin_unlock(&dentry->d_lock);
1880 * d_hash_and_lookup - hash the qstr then search for a dentry
1881 * @dir: Directory to search in
1882 * @name: qstr of name we wish to find
1884 * On hash failure or on lookup failure NULL is returned.
1886 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1888 struct dentry *dentry = NULL;
1891 * Check for a fs-specific hash function. Note that we must
1892 * calculate the standard hash first, as the d_op->d_hash()
1893 * routine may choose to leave the hash value unchanged.
1895 name->hash = full_name_hash(name->name, name->len);
1896 if (dir->d_flags & DCACHE_OP_HASH) {
1897 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1900 dentry = d_lookup(dir, name);
1906 * d_validate - verify dentry provided from insecure source (deprecated)
1907 * @dentry: The dentry alleged to be valid child of @dparent
1908 * @dparent: The parent dentry (known to be valid)
1910 * An insecure source has sent us a dentry, here we verify it and dget() it.
1911 * This is used by ncpfs in its readdir implementation.
1912 * Zero is returned in the dentry is invalid.
1914 * This function is slow for big directories, and deprecated, do not use it.
1916 int d_validate(struct dentry *dentry, struct dentry *dparent)
1918 struct dentry *child;
1920 spin_lock(&dparent->d_lock);
1921 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1922 if (dentry == child) {
1923 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1924 __dget_dlock(dentry);
1925 spin_unlock(&dentry->d_lock);
1926 spin_unlock(&dparent->d_lock);
1930 spin_unlock(&dparent->d_lock);
1934 EXPORT_SYMBOL(d_validate);
1937 * When a file is deleted, we have two options:
1938 * - turn this dentry into a negative dentry
1939 * - unhash this dentry and free it.
1941 * Usually, we want to just turn this into
1942 * a negative dentry, but if anybody else is
1943 * currently using the dentry or the inode
1944 * we can't do that and we fall back on removing
1945 * it from the hash queues and waiting for
1946 * it to be deleted later when it has no users
1950 * d_delete - delete a dentry
1951 * @dentry: The dentry to delete
1953 * Turn the dentry into a negative dentry if possible, otherwise
1954 * remove it from the hash queues so it can be deleted later
1957 void d_delete(struct dentry * dentry)
1959 struct inode *inode;
1962 * Are we the only user?
1965 spin_lock(&dentry->d_lock);
1966 inode = dentry->d_inode;
1967 isdir = S_ISDIR(inode->i_mode);
1968 if (dentry->d_count == 1) {
1969 if (inode && !spin_trylock(&inode->i_lock)) {
1970 spin_unlock(&dentry->d_lock);
1974 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1975 dentry_unlink_inode(dentry);
1976 fsnotify_nameremove(dentry, isdir);
1980 if (!d_unhashed(dentry))
1983 spin_unlock(&dentry->d_lock);
1985 fsnotify_nameremove(dentry, isdir);
1987 EXPORT_SYMBOL(d_delete);
1989 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
1991 BUG_ON(!d_unhashed(entry));
1993 entry->d_flags |= DCACHE_RCUACCESS;
1994 hlist_bl_add_head_rcu(&entry->d_hash, b);
1998 static void _d_rehash(struct dentry * entry)
2000 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2004 * d_rehash - add an entry back to the hash
2005 * @entry: dentry to add to the hash
2007 * Adds a dentry to the hash according to its name.
2010 void d_rehash(struct dentry * entry)
2012 spin_lock(&entry->d_lock);
2014 spin_unlock(&entry->d_lock);
2016 EXPORT_SYMBOL(d_rehash);
2019 * dentry_update_name_case - update case insensitive dentry with a new name
2020 * @dentry: dentry to be updated
2023 * Update a case insensitive dentry with new case of name.
2025 * dentry must have been returned by d_lookup with name @name. Old and new
2026 * name lengths must match (ie. no d_compare which allows mismatched name
2029 * Parent inode i_mutex must be held over d_lookup and into this call (to
2030 * keep renames and concurrent inserts, and readdir(2) away).
2032 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2034 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2035 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2037 spin_lock(&dentry->d_lock);
2038 write_seqcount_begin(&dentry->d_seq);
2039 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2040 write_seqcount_end(&dentry->d_seq);
2041 spin_unlock(&dentry->d_lock);
2043 EXPORT_SYMBOL(dentry_update_name_case);
2045 static void switch_names(struct dentry *dentry, struct dentry *target)
2047 if (dname_external(target)) {
2048 if (dname_external(dentry)) {
2050 * Both external: swap the pointers
2052 swap(target->d_name.name, dentry->d_name.name);
2055 * dentry:internal, target:external. Steal target's
2056 * storage and make target internal.
2058 memcpy(target->d_iname, dentry->d_name.name,
2059 dentry->d_name.len + 1);
2060 dentry->d_name.name = target->d_name.name;
2061 target->d_name.name = target->d_iname;
2064 if (dname_external(dentry)) {
2066 * dentry:external, target:internal. Give dentry's
2067 * storage to target and make dentry internal
2069 memcpy(dentry->d_iname, target->d_name.name,
2070 target->d_name.len + 1);
2071 target->d_name.name = dentry->d_name.name;
2072 dentry->d_name.name = dentry->d_iname;
2075 * Both are internal. Just copy target to dentry
2077 memcpy(dentry->d_iname, target->d_name.name,
2078 target->d_name.len + 1);
2079 dentry->d_name.len = target->d_name.len;
2083 swap(dentry->d_name.len, target->d_name.len);
2086 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2089 * XXXX: do we really need to take target->d_lock?
2091 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2092 spin_lock(&target->d_parent->d_lock);
2094 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2095 spin_lock(&dentry->d_parent->d_lock);
2096 spin_lock_nested(&target->d_parent->d_lock,
2097 DENTRY_D_LOCK_NESTED);
2099 spin_lock(&target->d_parent->d_lock);
2100 spin_lock_nested(&dentry->d_parent->d_lock,
2101 DENTRY_D_LOCK_NESTED);
2104 if (target < dentry) {
2105 spin_lock_nested(&target->d_lock, 2);
2106 spin_lock_nested(&dentry->d_lock, 3);
2108 spin_lock_nested(&dentry->d_lock, 2);
2109 spin_lock_nested(&target->d_lock, 3);
2113 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2114 struct dentry *target)
2116 if (target->d_parent != dentry->d_parent)
2117 spin_unlock(&dentry->d_parent->d_lock);
2118 if (target->d_parent != target)
2119 spin_unlock(&target->d_parent->d_lock);
2123 * When switching names, the actual string doesn't strictly have to
2124 * be preserved in the target - because we're dropping the target
2125 * anyway. As such, we can just do a simple memcpy() to copy over
2126 * the new name before we switch.
2128 * Note that we have to be a lot more careful about getting the hash
2129 * switched - we have to switch the hash value properly even if it
2130 * then no longer matches the actual (corrupted) string of the target.
2131 * The hash value has to match the hash queue that the dentry is on..
2134 * __d_move - move a dentry
2135 * @dentry: entry to move
2136 * @target: new dentry
2138 * Update the dcache to reflect the move of a file name. Negative
2139 * dcache entries should not be moved in this way. Caller must hold
2140 * rename_lock, the i_mutex of the source and target directories,
2141 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2143 static void __d_move(struct dentry * dentry, struct dentry * target)
2145 if (!dentry->d_inode)
2146 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2148 BUG_ON(d_ancestor(dentry, target));
2149 BUG_ON(d_ancestor(target, dentry));
2151 dentry_lock_for_move(dentry, target);
2153 write_seqcount_begin(&dentry->d_seq);
2154 write_seqcount_begin(&target->d_seq);
2156 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2159 * Move the dentry to the target hash queue. Don't bother checking
2160 * for the same hash queue because of how unlikely it is.
2163 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2165 /* Unhash the target: dput() will then get rid of it */
2168 list_del(&dentry->d_u.d_child);
2169 list_del(&target->d_u.d_child);
2171 /* Switch the names.. */
2172 switch_names(dentry, target);
2173 swap(dentry->d_name.hash, target->d_name.hash);
2175 /* ... and switch the parents */
2176 if (IS_ROOT(dentry)) {
2177 dentry->d_parent = target->d_parent;
2178 target->d_parent = target;
2179 INIT_LIST_HEAD(&target->d_u.d_child);
2181 swap(dentry->d_parent, target->d_parent);
2183 /* And add them back to the (new) parent lists */
2184 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2187 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2189 write_seqcount_end(&target->d_seq);
2190 write_seqcount_end(&dentry->d_seq);
2192 dentry_unlock_parents_for_move(dentry, target);
2193 spin_unlock(&target->d_lock);
2194 fsnotify_d_move(dentry);
2195 spin_unlock(&dentry->d_lock);
2199 * d_move - move a dentry
2200 * @dentry: entry to move
2201 * @target: new dentry
2203 * Update the dcache to reflect the move of a file name. Negative
2204 * dcache entries should not be moved in this way. See the locking
2205 * requirements for __d_move.
2207 void d_move(struct dentry *dentry, struct dentry *target)
2209 write_seqlock(&rename_lock);
2210 __d_move(dentry, target);
2211 write_sequnlock(&rename_lock);
2213 EXPORT_SYMBOL(d_move);
2216 * d_ancestor - search for an ancestor
2217 * @p1: ancestor dentry
2220 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2221 * an ancestor of p2, else NULL.
2223 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2227 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2228 if (p->d_parent == p1)
2235 * This helper attempts to cope with remotely renamed directories
2237 * It assumes that the caller is already holding
2238 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2240 * Note: If ever the locking in lock_rename() changes, then please
2241 * remember to update this too...
2243 static struct dentry *__d_unalias(struct inode *inode,
2244 struct dentry *dentry, struct dentry *alias)
2246 struct mutex *m1 = NULL, *m2 = NULL;
2249 /* If alias and dentry share a parent, then no extra locks required */
2250 if (alias->d_parent == dentry->d_parent)
2253 /* See lock_rename() */
2254 ret = ERR_PTR(-EBUSY);
2255 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2257 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2258 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2260 m2 = &alias->d_parent->d_inode->i_mutex;
2262 __d_move(alias, dentry);
2265 spin_unlock(&inode->i_lock);
2274 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2275 * named dentry in place of the dentry to be replaced.
2276 * returns with anon->d_lock held!
2278 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2280 struct dentry *dparent, *aparent;
2282 dentry_lock_for_move(anon, dentry);
2284 write_seqcount_begin(&dentry->d_seq);
2285 write_seqcount_begin(&anon->d_seq);
2287 dparent = dentry->d_parent;
2288 aparent = anon->d_parent;
2290 switch_names(dentry, anon);
2291 swap(dentry->d_name.hash, anon->d_name.hash);
2293 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2294 list_del(&dentry->d_u.d_child);
2295 if (!IS_ROOT(dentry))
2296 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2298 INIT_LIST_HEAD(&dentry->d_u.d_child);
2300 anon->d_parent = (dparent == dentry) ? anon : dparent;
2301 list_del(&anon->d_u.d_child);
2303 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2305 INIT_LIST_HEAD(&anon->d_u.d_child);
2307 write_seqcount_end(&dentry->d_seq);
2308 write_seqcount_end(&anon->d_seq);
2310 dentry_unlock_parents_for_move(anon, dentry);
2311 spin_unlock(&dentry->d_lock);
2313 /* anon->d_lock still locked, returns locked */
2314 anon->d_flags &= ~DCACHE_DISCONNECTED;
2318 * d_materialise_unique - introduce an inode into the tree
2319 * @dentry: candidate dentry
2320 * @inode: inode to bind to the dentry, to which aliases may be attached
2322 * Introduces an dentry into the tree, substituting an extant disconnected
2323 * root directory alias in its place if there is one. Caller must hold the
2324 * i_mutex of the parent directory.
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(&inode->i_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 write_seqlock(&rename_lock);
2350 if (d_ancestor(alias, dentry)) {
2351 /* Check for loops */
2352 actual = ERR_PTR(-ELOOP);
2353 } else if (IS_ROOT(alias)) {
2354 /* Is this an anonymous mountpoint that we
2355 * could splice into our tree? */
2356 __d_materialise_dentry(dentry, alias);
2357 write_sequnlock(&rename_lock);
2361 /* Nope, but we must(!) avoid directory
2363 actual = __d_unalias(inode, dentry, alias);
2365 write_sequnlock(&rename_lock);
2366 if (IS_ERR(actual)) {
2367 if (PTR_ERR(actual) == -ELOOP)
2368 pr_warn_ratelimited(
2369 "VFS: Lookup of '%s' in %s %s"
2370 " would have caused loop\n",
2371 dentry->d_name.name,
2372 inode->i_sb->s_type->name,
2380 /* Add a unique reference */
2381 actual = __d_instantiate_unique(dentry, inode);
2385 BUG_ON(!d_unhashed(actual));
2387 spin_lock(&actual->d_lock);
2390 spin_unlock(&actual->d_lock);
2391 spin_unlock(&inode->i_lock);
2393 if (actual == dentry) {
2394 security_d_instantiate(dentry, inode);
2401 EXPORT_SYMBOL_GPL(d_materialise_unique);
2403 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2407 return -ENAMETOOLONG;
2409 memcpy(*buffer, str, namelen);
2413 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2415 return prepend(buffer, buflen, name->name, name->len);
2419 * prepend_path - Prepend path string to a buffer
2420 * @path: the dentry/vfsmount to report
2421 * @root: root vfsmnt/dentry
2422 * @buffer: pointer to the end of the buffer
2423 * @buflen: pointer to buffer length
2425 * Caller holds the rename_lock.
2427 static int prepend_path(const struct path *path,
2428 const struct path *root,
2429 char **buffer, int *buflen)
2431 struct dentry *dentry = path->dentry;
2432 struct vfsmount *vfsmnt = path->mnt;
2436 br_read_lock(vfsmount_lock);
2437 while (dentry != root->dentry || vfsmnt != root->mnt) {
2438 struct dentry * parent;
2440 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2442 if (vfsmnt->mnt_parent == vfsmnt) {
2445 dentry = vfsmnt->mnt_mountpoint;
2446 vfsmnt = vfsmnt->mnt_parent;
2449 parent = dentry->d_parent;
2451 spin_lock(&dentry->d_lock);
2452 error = prepend_name(buffer, buflen, &dentry->d_name);
2453 spin_unlock(&dentry->d_lock);
2455 error = prepend(buffer, buflen, "/", 1);
2463 if (!error && !slash)
2464 error = prepend(buffer, buflen, "/", 1);
2467 br_read_unlock(vfsmount_lock);
2472 * Filesystems needing to implement special "root names"
2473 * should do so with ->d_dname()
2475 if (IS_ROOT(dentry) &&
2476 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2477 WARN(1, "Root dentry has weird name <%.*s>\n",
2478 (int) dentry->d_name.len, dentry->d_name.name);
2481 error = prepend(buffer, buflen, "/", 1);
2483 error = vfsmnt->mnt_ns ? 1 : 2;
2488 * __d_path - return the path of a dentry
2489 * @path: the dentry/vfsmount to report
2490 * @root: root vfsmnt/dentry
2491 * @buf: buffer to return value in
2492 * @buflen: buffer length
2494 * Convert a dentry into an ASCII path name.
2496 * Returns a pointer into the buffer or an error code if the
2497 * path was too long.
2499 * "buflen" should be positive.
2501 * If the path is not reachable from the supplied root, return %NULL.
2503 char *__d_path(const struct path *path,
2504 const struct path *root,
2505 char *buf, int buflen)
2507 char *res = buf + buflen;
2510 prepend(&res, &buflen, "\0", 1);
2511 write_seqlock(&rename_lock);
2512 error = prepend_path(path, root, &res, &buflen);
2513 write_sequnlock(&rename_lock);
2516 return ERR_PTR(error);
2522 char *d_absolute_path(const struct path *path,
2523 char *buf, int buflen)
2525 struct path root = {};
2526 char *res = buf + buflen;
2529 prepend(&res, &buflen, "\0", 1);
2530 write_seqlock(&rename_lock);
2531 error = prepend_path(path, &root, &res, &buflen);
2532 write_sequnlock(&rename_lock);
2537 return ERR_PTR(error);
2542 * same as __d_path but appends "(deleted)" for unlinked files.
2544 static int path_with_deleted(const struct path *path,
2545 const struct path *root,
2546 char **buf, int *buflen)
2548 prepend(buf, buflen, "\0", 1);
2549 if (d_unlinked(path->dentry)) {
2550 int error = prepend(buf, buflen, " (deleted)", 10);
2555 return prepend_path(path, root, buf, buflen);
2558 static int prepend_unreachable(char **buffer, int *buflen)
2560 return prepend(buffer, buflen, "(unreachable)", 13);
2564 * d_path - return the path of a dentry
2565 * @path: path to report
2566 * @buf: buffer to return value in
2567 * @buflen: buffer length
2569 * Convert a dentry into an ASCII path name. If the entry has been deleted
2570 * the string " (deleted)" is appended. Note that this is ambiguous.
2572 * Returns a pointer into the buffer or an error code if the path was
2573 * too long. Note: Callers should use the returned pointer, not the passed
2574 * in buffer, to use the name! The implementation often starts at an offset
2575 * into the buffer, and may leave 0 bytes at the start.
2577 * "buflen" should be positive.
2579 char *d_path(const struct path *path, char *buf, int buflen)
2581 char *res = buf + buflen;
2586 * We have various synthetic filesystems that never get mounted. On
2587 * these filesystems dentries are never used for lookup purposes, and
2588 * thus don't need to be hashed. They also don't need a name until a
2589 * user wants to identify the object in /proc/pid/fd/. The little hack
2590 * below allows us to generate a name for these objects on demand:
2592 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2593 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2595 get_fs_root(current->fs, &root);
2596 write_seqlock(&rename_lock);
2597 error = path_with_deleted(path, &root, &res, &buflen);
2599 res = ERR_PTR(error);
2600 write_sequnlock(&rename_lock);
2604 EXPORT_SYMBOL(d_path);
2607 * d_path_with_unreachable - return the path of a dentry
2608 * @path: path to report
2609 * @buf: buffer to return value in
2610 * @buflen: buffer length
2612 * The difference from d_path() is that this prepends "(unreachable)"
2613 * to paths which are unreachable from the current process' root.
2615 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2617 char *res = buf + buflen;
2621 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2622 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2624 get_fs_root(current->fs, &root);
2625 write_seqlock(&rename_lock);
2626 error = path_with_deleted(path, &root, &res, &buflen);
2628 error = prepend_unreachable(&res, &buflen);
2629 write_sequnlock(&rename_lock);
2632 res = ERR_PTR(error);
2638 * Helper function for dentry_operations.d_dname() members
2640 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2641 const char *fmt, ...)
2647 va_start(args, fmt);
2648 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2651 if (sz > sizeof(temp) || sz > buflen)
2652 return ERR_PTR(-ENAMETOOLONG);
2654 buffer += buflen - sz;
2655 return memcpy(buffer, temp, sz);
2659 * Write full pathname from the root of the filesystem into the buffer.
2661 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2663 char *end = buf + buflen;
2666 prepend(&end, &buflen, "\0", 1);
2673 while (!IS_ROOT(dentry)) {
2674 struct dentry *parent = dentry->d_parent;
2678 spin_lock(&dentry->d_lock);
2679 error = prepend_name(&end, &buflen, &dentry->d_name);
2680 spin_unlock(&dentry->d_lock);
2681 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2689 return ERR_PTR(-ENAMETOOLONG);
2692 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2696 write_seqlock(&rename_lock);
2697 retval = __dentry_path(dentry, buf, buflen);
2698 write_sequnlock(&rename_lock);
2702 EXPORT_SYMBOL(dentry_path_raw);
2704 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2709 write_seqlock(&rename_lock);
2710 if (d_unlinked(dentry)) {
2712 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2716 retval = __dentry_path(dentry, buf, buflen);
2717 write_sequnlock(&rename_lock);
2718 if (!IS_ERR(retval) && p)
2719 *p = '/'; /* restore '/' overriden with '\0' */
2722 return ERR_PTR(-ENAMETOOLONG);
2726 * NOTE! The user-level library version returns a
2727 * character pointer. The kernel system call just
2728 * returns the length of the buffer filled (which
2729 * includes the ending '\0' character), or a negative
2730 * error value. So libc would do something like
2732 * char *getcwd(char * buf, size_t size)
2736 * retval = sys_getcwd(buf, size);
2743 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2746 struct path pwd, root;
2747 char *page = (char *) __get_free_page(GFP_USER);
2752 get_fs_root_and_pwd(current->fs, &root, &pwd);
2755 write_seqlock(&rename_lock);
2756 if (!d_unlinked(pwd.dentry)) {
2758 char *cwd = page + PAGE_SIZE;
2759 int buflen = PAGE_SIZE;
2761 prepend(&cwd, &buflen, "\0", 1);
2762 error = prepend_path(&pwd, &root, &cwd, &buflen);
2763 write_sequnlock(&rename_lock);
2768 /* Unreachable from current root */
2770 error = prepend_unreachable(&cwd, &buflen);
2776 len = PAGE_SIZE + page - cwd;
2779 if (copy_to_user(buf, cwd, len))
2783 write_sequnlock(&rename_lock);
2789 free_page((unsigned long) page);
2794 * Test whether new_dentry is a subdirectory of old_dentry.
2796 * Trivially implemented using the dcache structure
2800 * is_subdir - is new dentry a subdirectory of old_dentry
2801 * @new_dentry: new dentry
2802 * @old_dentry: old dentry
2804 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2805 * Returns 0 otherwise.
2806 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2809 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2814 if (new_dentry == old_dentry)
2818 /* for restarting inner loop in case of seq retry */
2819 seq = read_seqbegin(&rename_lock);
2821 * Need rcu_readlock to protect against the d_parent trashing
2825 if (d_ancestor(old_dentry, new_dentry))
2830 } while (read_seqretry(&rename_lock, seq));
2835 int path_is_under(struct path *path1, struct path *path2)
2837 struct vfsmount *mnt = path1->mnt;
2838 struct dentry *dentry = path1->dentry;
2841 br_read_lock(vfsmount_lock);
2842 if (mnt != path2->mnt) {
2844 if (mnt->mnt_parent == mnt) {
2845 br_read_unlock(vfsmount_lock);
2848 if (mnt->mnt_parent == path2->mnt)
2850 mnt = mnt->mnt_parent;
2852 dentry = mnt->mnt_mountpoint;
2854 res = is_subdir(dentry, path2->dentry);
2855 br_read_unlock(vfsmount_lock);
2858 EXPORT_SYMBOL(path_is_under);
2860 void d_genocide(struct dentry *root)
2862 struct dentry *this_parent;
2863 struct list_head *next;
2867 seq = read_seqbegin(&rename_lock);
2870 spin_lock(&this_parent->d_lock);
2872 next = this_parent->d_subdirs.next;
2874 while (next != &this_parent->d_subdirs) {
2875 struct list_head *tmp = next;
2876 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2879 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2880 if (d_unhashed(dentry) || !dentry->d_inode) {
2881 spin_unlock(&dentry->d_lock);
2884 if (!list_empty(&dentry->d_subdirs)) {
2885 spin_unlock(&this_parent->d_lock);
2886 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2887 this_parent = dentry;
2888 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2891 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2892 dentry->d_flags |= DCACHE_GENOCIDE;
2895 spin_unlock(&dentry->d_lock);
2897 if (this_parent != root) {
2898 struct dentry *child = this_parent;
2899 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2900 this_parent->d_flags |= DCACHE_GENOCIDE;
2901 this_parent->d_count--;
2903 this_parent = try_to_ascend(this_parent, locked, seq);
2906 next = child->d_u.d_child.next;
2909 spin_unlock(&this_parent->d_lock);
2910 if (!locked && read_seqretry(&rename_lock, seq))
2913 write_sequnlock(&rename_lock);
2918 write_seqlock(&rename_lock);
2923 * find_inode_number - check for dentry with name
2924 * @dir: directory to check
2925 * @name: Name to find.
2927 * Check whether a dentry already exists for the given name,
2928 * and return the inode number if it has an inode. Otherwise
2931 * This routine is used to post-process directory listings for
2932 * filesystems using synthetic inode numbers, and is necessary
2933 * to keep getcwd() working.
2936 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2938 struct dentry * dentry;
2941 dentry = d_hash_and_lookup(dir, name);
2943 if (dentry->d_inode)
2944 ino = dentry->d_inode->i_ino;
2949 EXPORT_SYMBOL(find_inode_number);
2951 static __initdata unsigned long dhash_entries;
2952 static int __init set_dhash_entries(char *str)
2956 dhash_entries = simple_strtoul(str, &str, 0);
2959 __setup("dhash_entries=", set_dhash_entries);
2961 static void __init dcache_init_early(void)
2965 /* If hashes are distributed across NUMA nodes, defer
2966 * hash allocation until vmalloc space is available.
2972 alloc_large_system_hash("Dentry cache",
2973 sizeof(struct hlist_bl_head),
2981 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2982 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
2985 static void __init dcache_init(void)
2990 * A constructor could be added for stable state like the lists,
2991 * but it is probably not worth it because of the cache nature
2994 dentry_cache = KMEM_CACHE(dentry,
2995 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2997 /* Hash may have been set up in dcache_init_early */
3002 alloc_large_system_hash("Dentry cache",
3003 sizeof(struct hlist_bl_head),
3011 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3012 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3015 /* SLAB cache for __getname() consumers */
3016 struct kmem_cache *names_cachep __read_mostly;
3017 EXPORT_SYMBOL(names_cachep);
3019 EXPORT_SYMBOL(d_genocide);
3021 void __init vfs_caches_init_early(void)
3023 dcache_init_early();
3027 void __init vfs_caches_init(unsigned long mempages)
3029 unsigned long reserve;
3031 /* Base hash sizes on available memory, with a reserve equal to
3032 150% of current kernel size */
3034 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3035 mempages -= reserve;
3037 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3038 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3042 files_init(mempages);