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_u.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
60 * - d_u.d_alias, d_inode
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 if (dname_external(dentry))
144 kfree(dentry->d_name.name);
145 kmem_cache_free(dentry_cache, dentry);
151 static void d_free(struct dentry *dentry)
153 WARN_ON(!list_empty(&dentry->d_u.d_alias));
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_u.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_u.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_entry(&dentry->d_child);
311 * Inform ascending readers 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_u.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_u.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_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_child);
894 parent = dentry->d_parent;
896 list_del(&dentry->d_child);
899 inode = dentry->d_inode;
901 dentry->d_inode = NULL;
902 list_del_init(&dentry->d_u.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_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);
954 * Search for at least 1 mount point in the dentry's subdirs.
955 * We descend to the next level whenever the d_subdirs
956 * list is non-empty and continue searching.
960 * have_submounts - check for mounts over a dentry
961 * @parent: dentry to check.
963 * Return true if the parent or its subdirectories contain
966 int have_submounts(struct dentry *parent)
968 struct dentry *this_parent;
969 struct list_head *next;
973 seq = read_seqbegin(&rename_lock);
975 this_parent = parent;
977 if (d_mountpoint(parent))
979 spin_lock(&this_parent->d_lock);
981 next = this_parent->d_subdirs.next;
983 while (next != &this_parent->d_subdirs) {
984 struct list_head *tmp = next;
985 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
988 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
989 /* Have we found a mount point ? */
990 if (d_mountpoint(dentry)) {
991 spin_unlock(&dentry->d_lock);
992 spin_unlock(&this_parent->d_lock);
995 if (!list_empty(&dentry->d_subdirs)) {
996 spin_unlock(&this_parent->d_lock);
997 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
998 this_parent = dentry;
999 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1002 spin_unlock(&dentry->d_lock);
1005 * All done at this level ... ascend and resume the search.
1009 if (this_parent != parent) {
1010 struct dentry *child = this_parent;
1011 this_parent = child->d_parent;
1013 spin_unlock(&child->d_lock);
1014 spin_lock(&this_parent->d_lock);
1016 /* might go back up the wrong parent if we have had a rename */
1017 if (!locked && read_seqretry(&rename_lock, seq))
1019 next = child->d_child.next;
1020 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1021 if (next == &this_parent->d_subdirs)
1023 child = list_entry(next, struct dentry, d_child);
1029 if (!locked && read_seqretry(&rename_lock, seq))
1031 spin_unlock(&this_parent->d_lock);
1034 write_sequnlock(&rename_lock);
1035 return 0; /* No mount points found in tree */
1037 if (!locked && read_seqretry(&rename_lock, seq))
1040 write_sequnlock(&rename_lock);
1044 spin_unlock(&this_parent->d_lock);
1049 write_seqlock(&rename_lock);
1052 EXPORT_SYMBOL(have_submounts);
1055 * Search the dentry child list for the specified parent,
1056 * and move any unused dentries to the end of the unused
1057 * list for prune_dcache(). We descend to the next level
1058 * whenever the d_subdirs list is non-empty and continue
1061 * It returns zero iff there are no unused children,
1062 * otherwise it returns the number of children moved to
1063 * the end of the unused list. This may not be the total
1064 * number of unused children, because select_parent can
1065 * drop the lock and return early due to latency
1068 static int select_parent(struct dentry *parent, struct list_head *dispose)
1070 struct dentry *this_parent;
1071 struct list_head *next;
1076 seq = read_seqbegin(&rename_lock);
1078 this_parent = parent;
1079 spin_lock(&this_parent->d_lock);
1081 next = this_parent->d_subdirs.next;
1083 while (next != &this_parent->d_subdirs) {
1084 struct list_head *tmp = next;
1085 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1088 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1091 * move only zero ref count dentries to the dispose list.
1093 * Those which are presently on the shrink list, being processed
1094 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1095 * loop in shrink_dcache_parent() might not make any progress
1098 if (dentry->d_count) {
1099 dentry_lru_del(dentry);
1100 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1101 dentry_lru_move_list(dentry, dispose);
1102 dentry->d_flags |= DCACHE_SHRINK_LIST;
1106 * We can return to the caller if we have found some (this
1107 * ensures forward progress). We'll be coming back to find
1110 if (found && need_resched()) {
1111 spin_unlock(&dentry->d_lock);
1116 * Descend a level if the d_subdirs list is non-empty.
1118 if (!list_empty(&dentry->d_subdirs)) {
1119 spin_unlock(&this_parent->d_lock);
1120 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1121 this_parent = dentry;
1122 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1126 spin_unlock(&dentry->d_lock);
1129 * All done at this level ... ascend and resume the search.
1133 if (this_parent != parent) {
1134 struct dentry *child = this_parent;
1135 this_parent = child->d_parent;
1137 spin_unlock(&child->d_lock);
1138 spin_lock(&this_parent->d_lock);
1140 /* might go back up the wrong parent if we have had a rename */
1141 if (!locked && read_seqretry(&rename_lock, seq))
1143 next = child->d_child.next;
1144 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1145 if (next == &this_parent->d_subdirs)
1147 child = list_entry(next, struct dentry, d_child);
1154 if (!locked && read_seqretry(&rename_lock, seq))
1156 spin_unlock(&this_parent->d_lock);
1159 write_sequnlock(&rename_lock);
1163 spin_unlock(&this_parent->d_lock);
1170 write_seqlock(&rename_lock);
1175 * shrink_dcache_parent - prune dcache
1176 * @parent: parent of entries to prune
1178 * Prune the dcache to remove unused children of the parent dentry.
1180 void shrink_dcache_parent(struct dentry * parent)
1185 while ((found = select_parent(parent, &dispose)) != 0) {
1186 shrink_dentry_list(&dispose);
1190 EXPORT_SYMBOL(shrink_dcache_parent);
1193 * __d_alloc - allocate a dcache entry
1194 * @sb: filesystem it will belong to
1195 * @name: qstr of the name
1197 * Allocates a dentry. It returns %NULL if there is insufficient memory
1198 * available. On a success the dentry is returned. The name passed in is
1199 * copied and the copy passed in may be reused after this call.
1202 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1204 struct dentry *dentry;
1207 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1211 if (name->len > DNAME_INLINE_LEN-1) {
1212 dname = kmalloc(name->len + 1, GFP_KERNEL);
1214 kmem_cache_free(dentry_cache, dentry);
1218 dname = dentry->d_iname;
1220 dentry->d_name.name = dname;
1222 dentry->d_name.len = name->len;
1223 dentry->d_name.hash = name->hash;
1224 memcpy(dname, name->name, name->len);
1225 dname[name->len] = 0;
1227 dentry->d_count = 1;
1228 dentry->d_flags = 0;
1229 spin_lock_init(&dentry->d_lock);
1230 seqcount_init(&dentry->d_seq);
1231 dentry->d_inode = NULL;
1232 dentry->d_parent = dentry;
1234 dentry->d_op = NULL;
1235 dentry->d_fsdata = NULL;
1236 INIT_HLIST_BL_NODE(&dentry->d_hash);
1237 INIT_LIST_HEAD(&dentry->d_lru);
1238 INIT_LIST_HEAD(&dentry->d_subdirs);
1239 INIT_LIST_HEAD(&dentry->d_u.d_alias);
1240 INIT_LIST_HEAD(&dentry->d_child);
1241 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1243 this_cpu_inc(nr_dentry);
1249 * d_alloc - allocate a dcache entry
1250 * @parent: parent of entry to allocate
1251 * @name: qstr of the name
1253 * Allocates a dentry. It returns %NULL if there is insufficient memory
1254 * available. On a success the dentry is returned. The name passed in is
1255 * copied and the copy passed in may be reused after this call.
1257 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1259 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1263 spin_lock(&parent->d_lock);
1265 * don't need child lock because it is not subject
1266 * to concurrency here
1268 __dget_dlock(parent);
1269 dentry->d_parent = parent;
1270 list_add(&dentry->d_child, &parent->d_subdirs);
1271 spin_unlock(&parent->d_lock);
1275 EXPORT_SYMBOL(d_alloc);
1277 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1279 struct dentry *dentry = __d_alloc(sb, name);
1281 dentry->d_flags |= DCACHE_DISCONNECTED;
1284 EXPORT_SYMBOL(d_alloc_pseudo);
1286 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1291 q.len = strlen(name);
1292 q.hash = full_name_hash(q.name, q.len);
1293 return d_alloc(parent, &q);
1295 EXPORT_SYMBOL(d_alloc_name);
1297 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1299 WARN_ON_ONCE(dentry->d_op);
1300 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1302 DCACHE_OP_REVALIDATE |
1303 DCACHE_OP_DELETE ));
1308 dentry->d_flags |= DCACHE_OP_HASH;
1310 dentry->d_flags |= DCACHE_OP_COMPARE;
1311 if (op->d_revalidate)
1312 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1314 dentry->d_flags |= DCACHE_OP_DELETE;
1316 dentry->d_flags |= DCACHE_OP_PRUNE;
1319 EXPORT_SYMBOL(d_set_d_op);
1321 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1323 spin_lock(&dentry->d_lock);
1325 if (unlikely(IS_AUTOMOUNT(inode)))
1326 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1327 list_add(&dentry->d_u.d_alias, &inode->i_dentry);
1329 dentry->d_inode = inode;
1330 dentry_rcuwalk_barrier(dentry);
1331 spin_unlock(&dentry->d_lock);
1332 fsnotify_d_instantiate(dentry, inode);
1336 * d_instantiate - fill in inode information for a dentry
1337 * @entry: dentry to complete
1338 * @inode: inode to attach to this dentry
1340 * Fill in inode information in the entry.
1342 * This turns negative dentries into productive full members
1345 * NOTE! This assumes that the inode count has been incremented
1346 * (or otherwise set) by the caller to indicate that it is now
1347 * in use by the dcache.
1350 void d_instantiate(struct dentry *entry, struct inode * inode)
1352 BUG_ON(!list_empty(&entry->d_u.d_alias));
1354 spin_lock(&inode->i_lock);
1355 __d_instantiate(entry, inode);
1357 spin_unlock(&inode->i_lock);
1358 security_d_instantiate(entry, inode);
1360 EXPORT_SYMBOL(d_instantiate);
1363 * d_instantiate_unique - instantiate a non-aliased dentry
1364 * @entry: dentry to instantiate
1365 * @inode: inode to attach to this dentry
1367 * Fill in inode information in the entry. On success, it returns NULL.
1368 * If an unhashed alias of "entry" already exists, then we return the
1369 * aliased dentry instead and drop one reference to inode.
1371 * Note that in order to avoid conflicts with rename() etc, the caller
1372 * had better be holding the parent directory semaphore.
1374 * This also assumes that the inode count has been incremented
1375 * (or otherwise set) by the caller to indicate that it is now
1376 * in use by the dcache.
1378 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1379 struct inode *inode)
1381 struct dentry *alias;
1382 int len = entry->d_name.len;
1383 const char *name = entry->d_name.name;
1384 unsigned int hash = entry->d_name.hash;
1387 __d_instantiate(entry, NULL);
1391 list_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1392 struct qstr *qstr = &alias->d_name;
1395 * Don't need alias->d_lock here, because aliases with
1396 * d_parent == entry->d_parent are not subject to name or
1397 * parent changes, because the parent inode i_mutex is held.
1399 if (qstr->hash != hash)
1401 if (alias->d_parent != entry->d_parent)
1403 if (dentry_cmp(qstr->name, qstr->len, name, len))
1409 __d_instantiate(entry, inode);
1413 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1415 struct dentry *result;
1417 BUG_ON(!list_empty(&entry->d_u.d_alias));
1420 spin_lock(&inode->i_lock);
1421 result = __d_instantiate_unique(entry, inode);
1423 spin_unlock(&inode->i_lock);
1426 security_d_instantiate(entry, inode);
1430 BUG_ON(!d_unhashed(result));
1435 EXPORT_SYMBOL(d_instantiate_unique);
1438 * d_alloc_root - allocate root dentry
1439 * @root_inode: inode to allocate the root for
1441 * Allocate a root ("/") dentry for the inode given. The inode is
1442 * instantiated and returned. %NULL is returned if there is insufficient
1443 * memory or the inode passed is %NULL.
1446 struct dentry * d_alloc_root(struct inode * root_inode)
1448 struct dentry *res = NULL;
1451 static const struct qstr name = { .name = "/", .len = 1 };
1453 res = __d_alloc(root_inode->i_sb, &name);
1455 d_instantiate(res, root_inode);
1459 EXPORT_SYMBOL(d_alloc_root);
1461 static struct dentry * __d_find_any_alias(struct inode *inode)
1463 struct dentry *alias;
1465 if (list_empty(&inode->i_dentry))
1467 alias = list_first_entry(&inode->i_dentry, struct dentry, d_u.d_alias);
1472 static struct dentry * d_find_any_alias(struct inode *inode)
1476 spin_lock(&inode->i_lock);
1477 de = __d_find_any_alias(inode);
1478 spin_unlock(&inode->i_lock);
1484 * d_obtain_alias - find or allocate a dentry for a given inode
1485 * @inode: inode to allocate the dentry for
1487 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1488 * similar open by handle operations. The returned dentry may be anonymous,
1489 * or may have a full name (if the inode was already in the cache).
1491 * When called on a directory inode, we must ensure that the inode only ever
1492 * has one dentry. If a dentry is found, that is returned instead of
1493 * allocating a new one.
1495 * On successful return, the reference to the inode has been transferred
1496 * to the dentry. In case of an error the reference on the inode is released.
1497 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1498 * be passed in and will be the error will be propagate to the return value,
1499 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1501 struct dentry *d_obtain_alias(struct inode *inode)
1503 static const struct qstr anonstring = { .name = "/", .len = 1 };
1508 return ERR_PTR(-ESTALE);
1510 return ERR_CAST(inode);
1512 res = d_find_any_alias(inode);
1516 tmp = __d_alloc(inode->i_sb, &anonstring);
1518 res = ERR_PTR(-ENOMEM);
1522 spin_lock(&inode->i_lock);
1523 res = __d_find_any_alias(inode);
1525 spin_unlock(&inode->i_lock);
1530 /* attach a disconnected dentry */
1531 spin_lock(&tmp->d_lock);
1532 tmp->d_inode = inode;
1533 tmp->d_flags |= DCACHE_DISCONNECTED;
1534 list_add(&tmp->d_u.d_alias, &inode->i_dentry);
1535 hlist_bl_lock(&tmp->d_sb->s_anon);
1536 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1537 hlist_bl_unlock(&tmp->d_sb->s_anon);
1538 spin_unlock(&tmp->d_lock);
1539 spin_unlock(&inode->i_lock);
1540 security_d_instantiate(tmp, inode);
1545 if (res && !IS_ERR(res))
1546 security_d_instantiate(res, inode);
1550 EXPORT_SYMBOL(d_obtain_alias);
1553 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1554 * @inode: the inode which may have a disconnected dentry
1555 * @dentry: a negative dentry which we want to point to the inode.
1557 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1558 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1559 * and return it, else simply d_add the inode to the dentry and return NULL.
1561 * This is needed in the lookup routine of any filesystem that is exportable
1562 * (via knfsd) so that we can build dcache paths to directories effectively.
1564 * If a dentry was found and moved, then it is returned. Otherwise NULL
1565 * is returned. This matches the expected return value of ->lookup.
1568 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1570 struct dentry *new = NULL;
1573 return ERR_CAST(inode);
1575 if (inode && S_ISDIR(inode->i_mode)) {
1576 spin_lock(&inode->i_lock);
1577 new = __d_find_alias(inode, 1);
1579 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1580 spin_unlock(&inode->i_lock);
1581 security_d_instantiate(new, inode);
1582 d_move(new, dentry);
1585 /* already taking inode->i_lock, so d_add() by hand */
1586 __d_instantiate(dentry, inode);
1587 spin_unlock(&inode->i_lock);
1588 security_d_instantiate(dentry, inode);
1592 d_add(dentry, inode);
1595 EXPORT_SYMBOL(d_splice_alias);
1598 * d_add_ci - lookup or allocate new dentry with case-exact name
1599 * @inode: the inode case-insensitive lookup has found
1600 * @dentry: the negative dentry that was passed to the parent's lookup func
1601 * @name: the case-exact name to be associated with the returned dentry
1603 * This is to avoid filling the dcache with case-insensitive names to the
1604 * same inode, only the actual correct case is stored in the dcache for
1605 * case-insensitive filesystems.
1607 * For a case-insensitive lookup match and if the the case-exact dentry
1608 * already exists in in the dcache, use it and return it.
1610 * If no entry exists with the exact case name, allocate new dentry with
1611 * the exact case, and return the spliced entry.
1613 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1617 struct dentry *found;
1621 * First check if a dentry matching the name already exists,
1622 * if not go ahead and create it now.
1624 found = d_hash_and_lookup(dentry->d_parent, name);
1626 new = d_alloc(dentry->d_parent, name);
1632 found = d_splice_alias(inode, new);
1641 * If a matching dentry exists, and it's not negative use it.
1643 * Decrement the reference count to balance the iget() done
1646 if (found->d_inode) {
1647 if (unlikely(found->d_inode != inode)) {
1648 /* This can't happen because bad inodes are unhashed. */
1649 BUG_ON(!is_bad_inode(inode));
1650 BUG_ON(!is_bad_inode(found->d_inode));
1657 * We are going to instantiate this dentry, unhash it and clear the
1658 * lookup flag so we can do that.
1660 if (unlikely(d_need_lookup(found)))
1661 d_clear_need_lookup(found);
1664 * Negative dentry: instantiate it unless the inode is a directory and
1665 * already has a dentry.
1667 new = d_splice_alias(inode, found);
1676 return ERR_PTR(error);
1678 EXPORT_SYMBOL(d_add_ci);
1681 * __d_lookup_rcu - search for a dentry (racy, store-free)
1682 * @parent: parent dentry
1683 * @name: qstr of name we wish to find
1684 * @seq: returns d_seq value at the point where the dentry was found
1685 * @inode: returns dentry->d_inode when the inode was found valid.
1686 * Returns: dentry, or NULL
1688 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1689 * resolution (store-free path walking) design described in
1690 * Documentation/filesystems/path-lookup.txt.
1692 * This is not to be used outside core vfs.
1694 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1695 * held, and rcu_read_lock held. The returned dentry must not be stored into
1696 * without taking d_lock and checking d_seq sequence count against @seq
1699 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1702 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1703 * the returned dentry, so long as its parent's seqlock is checked after the
1704 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1705 * is formed, giving integrity down the path walk.
1707 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1708 unsigned *seq, struct inode **inode)
1710 unsigned int len = name->len;
1711 unsigned int hash = name->hash;
1712 const unsigned char *str = name->name;
1713 struct hlist_bl_head *b = d_hash(parent, hash);
1714 struct hlist_bl_node *node;
1715 struct dentry *dentry;
1718 * Note: There is significant duplication with __d_lookup_rcu which is
1719 * required to prevent single threaded performance regressions
1720 * especially on architectures where smp_rmb (in seqcounts) are costly.
1721 * Keep the two functions in sync.
1725 * The hash list is protected using RCU.
1727 * Carefully use d_seq when comparing a candidate dentry, to avoid
1728 * races with d_move().
1730 * It is possible that concurrent renames can mess up our list
1731 * walk here and result in missing our dentry, resulting in the
1732 * false-negative result. d_lookup() protects against concurrent
1733 * renames using rename_lock seqlock.
1735 * See Documentation/filesystems/path-lookup.txt for more details.
1737 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1742 if (dentry->d_name.hash != hash)
1746 *seq = read_seqcount_begin(&dentry->d_seq);
1747 if (dentry->d_parent != parent)
1749 if (d_unhashed(dentry))
1751 tlen = dentry->d_name.len;
1752 tname = dentry->d_name.name;
1753 i = dentry->d_inode;
1756 * This seqcount check is required to ensure name and
1757 * len are loaded atomically, so as not to walk off the
1758 * edge of memory when walking. If we could load this
1759 * atomically some other way, we could drop this check.
1761 if (read_seqcount_retry(&dentry->d_seq, *seq))
1763 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1764 if (parent->d_op->d_compare(parent, *inode,
1769 if (dentry_cmp(tname, tlen, str, len))
1773 * No extra seqcount check is required after the name
1774 * compare. The caller must perform a seqcount check in
1775 * order to do anything useful with the returned dentry
1785 * d_lookup - search for a dentry
1786 * @parent: parent dentry
1787 * @name: qstr of name we wish to find
1788 * Returns: dentry, or NULL
1790 * d_lookup searches the children of the parent dentry for the name in
1791 * question. If the dentry is found its reference count is incremented and the
1792 * dentry is returned. The caller must use dput to free the entry when it has
1793 * finished using it. %NULL is returned if the dentry does not exist.
1795 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1797 struct dentry *dentry;
1801 seq = read_seqbegin(&rename_lock);
1802 dentry = __d_lookup(parent, name);
1805 } while (read_seqretry(&rename_lock, seq));
1808 EXPORT_SYMBOL(d_lookup);
1811 * __d_lookup - search for a dentry (racy)
1812 * @parent: parent dentry
1813 * @name: qstr of name we wish to find
1814 * Returns: dentry, or NULL
1816 * __d_lookup is like d_lookup, however it may (rarely) return a
1817 * false-negative result due to unrelated rename activity.
1819 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1820 * however it must be used carefully, eg. with a following d_lookup in
1821 * the case of failure.
1823 * __d_lookup callers must be commented.
1825 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1827 unsigned int len = name->len;
1828 unsigned int hash = name->hash;
1829 const unsigned char *str = name->name;
1830 struct hlist_bl_head *b = d_hash(parent, hash);
1831 struct hlist_bl_node *node;
1832 struct dentry *found = NULL;
1833 struct dentry *dentry;
1836 * Note: There is significant duplication with __d_lookup_rcu which is
1837 * required to prevent single threaded performance regressions
1838 * especially on architectures where smp_rmb (in seqcounts) are costly.
1839 * Keep the two functions in sync.
1843 * The hash list is protected using RCU.
1845 * Take d_lock when comparing a candidate dentry, to avoid races
1848 * It is possible that concurrent renames can mess up our list
1849 * walk here and result in missing our dentry, resulting in the
1850 * false-negative result. d_lookup() protects against concurrent
1851 * renames using rename_lock seqlock.
1853 * See Documentation/filesystems/path-lookup.txt for more details.
1857 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1861 if (dentry->d_name.hash != hash)
1864 spin_lock(&dentry->d_lock);
1865 if (dentry->d_parent != parent)
1867 if (d_unhashed(dentry))
1871 * It is safe to compare names since d_move() cannot
1872 * change the qstr (protected by d_lock).
1874 tlen = dentry->d_name.len;
1875 tname = dentry->d_name.name;
1876 if (parent->d_flags & DCACHE_OP_COMPARE) {
1877 if (parent->d_op->d_compare(parent, parent->d_inode,
1878 dentry, dentry->d_inode,
1882 if (dentry_cmp(tname, tlen, str, len))
1888 spin_unlock(&dentry->d_lock);
1891 spin_unlock(&dentry->d_lock);
1899 * d_hash_and_lookup - hash the qstr then search for a dentry
1900 * @dir: Directory to search in
1901 * @name: qstr of name we wish to find
1903 * On hash failure or on lookup failure NULL is returned.
1905 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1907 struct dentry *dentry = NULL;
1910 * Check for a fs-specific hash function. Note that we must
1911 * calculate the standard hash first, as the d_op->d_hash()
1912 * routine may choose to leave the hash value unchanged.
1914 name->hash = full_name_hash(name->name, name->len);
1915 if (dir->d_flags & DCACHE_OP_HASH) {
1916 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1919 dentry = d_lookup(dir, name);
1925 * d_validate - verify dentry provided from insecure source (deprecated)
1926 * @dentry: The dentry alleged to be valid child of @dparent
1927 * @dparent: The parent dentry (known to be valid)
1929 * An insecure source has sent us a dentry, here we verify it and dget() it.
1930 * This is used by ncpfs in its readdir implementation.
1931 * Zero is returned in the dentry is invalid.
1933 * This function is slow for big directories, and deprecated, do not use it.
1935 int d_validate(struct dentry *dentry, struct dentry *dparent)
1937 struct dentry *child;
1939 spin_lock(&dparent->d_lock);
1940 list_for_each_entry(child, &dparent->d_subdirs, d_child) {
1941 if (dentry == child) {
1942 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1943 __dget_dlock(dentry);
1944 spin_unlock(&dentry->d_lock);
1945 spin_unlock(&dparent->d_lock);
1949 spin_unlock(&dparent->d_lock);
1953 EXPORT_SYMBOL(d_validate);
1956 * When a file is deleted, we have two options:
1957 * - turn this dentry into a negative dentry
1958 * - unhash this dentry and free it.
1960 * Usually, we want to just turn this into
1961 * a negative dentry, but if anybody else is
1962 * currently using the dentry or the inode
1963 * we can't do that and we fall back on removing
1964 * it from the hash queues and waiting for
1965 * it to be deleted later when it has no users
1969 * d_delete - delete a dentry
1970 * @dentry: The dentry to delete
1972 * Turn the dentry into a negative dentry if possible, otherwise
1973 * remove it from the hash queues so it can be deleted later
1976 void d_delete(struct dentry * dentry)
1978 struct inode *inode;
1981 * Are we the only user?
1984 spin_lock(&dentry->d_lock);
1985 inode = dentry->d_inode;
1986 isdir = S_ISDIR(inode->i_mode);
1987 if (dentry->d_count == 1) {
1988 if (inode && !spin_trylock(&inode->i_lock)) {
1989 spin_unlock(&dentry->d_lock);
1993 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1994 dentry_unlink_inode(dentry);
1995 fsnotify_nameremove(dentry, isdir);
1999 if (!d_unhashed(dentry))
2002 spin_unlock(&dentry->d_lock);
2004 fsnotify_nameremove(dentry, isdir);
2006 EXPORT_SYMBOL(d_delete);
2008 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2010 BUG_ON(!d_unhashed(entry));
2012 entry->d_flags |= DCACHE_RCUACCESS;
2013 hlist_bl_add_head_rcu(&entry->d_hash, b);
2017 static void _d_rehash(struct dentry * entry)
2019 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2023 * d_rehash - add an entry back to the hash
2024 * @entry: dentry to add to the hash
2026 * Adds a dentry to the hash according to its name.
2029 void d_rehash(struct dentry * entry)
2031 spin_lock(&entry->d_lock);
2033 spin_unlock(&entry->d_lock);
2035 EXPORT_SYMBOL(d_rehash);
2038 * dentry_update_name_case - update case insensitive dentry with a new name
2039 * @dentry: dentry to be updated
2042 * Update a case insensitive dentry with new case of name.
2044 * dentry must have been returned by d_lookup with name @name. Old and new
2045 * name lengths must match (ie. no d_compare which allows mismatched name
2048 * Parent inode i_mutex must be held over d_lookup and into this call (to
2049 * keep renames and concurrent inserts, and readdir(2) away).
2051 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2053 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2054 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2056 spin_lock(&dentry->d_lock);
2057 write_seqcount_begin(&dentry->d_seq);
2058 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2059 write_seqcount_end(&dentry->d_seq);
2060 spin_unlock(&dentry->d_lock);
2062 EXPORT_SYMBOL(dentry_update_name_case);
2064 static void switch_names(struct dentry *dentry, struct dentry *target)
2066 if (dname_external(target)) {
2067 if (dname_external(dentry)) {
2069 * Both external: swap the pointers
2071 swap(target->d_name.name, dentry->d_name.name);
2074 * dentry:internal, target:external. Steal target's
2075 * storage and make target internal.
2077 memcpy(target->d_iname, dentry->d_name.name,
2078 dentry->d_name.len + 1);
2079 dentry->d_name.name = target->d_name.name;
2080 target->d_name.name = target->d_iname;
2083 if (dname_external(dentry)) {
2085 * dentry:external, target:internal. Give dentry's
2086 * storage to target and make dentry internal
2088 memcpy(dentry->d_iname, target->d_name.name,
2089 target->d_name.len + 1);
2090 target->d_name.name = dentry->d_name.name;
2091 dentry->d_name.name = dentry->d_iname;
2094 * Both are internal. Just copy target to dentry
2096 memcpy(dentry->d_iname, target->d_name.name,
2097 target->d_name.len + 1);
2098 dentry->d_name.len = target->d_name.len;
2102 swap(dentry->d_name.len, target->d_name.len);
2105 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2108 * XXXX: do we really need to take target->d_lock?
2110 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2111 spin_lock(&target->d_parent->d_lock);
2113 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2114 spin_lock(&dentry->d_parent->d_lock);
2115 spin_lock_nested(&target->d_parent->d_lock,
2116 DENTRY_D_LOCK_NESTED);
2118 spin_lock(&target->d_parent->d_lock);
2119 spin_lock_nested(&dentry->d_parent->d_lock,
2120 DENTRY_D_LOCK_NESTED);
2123 if (target < dentry) {
2124 spin_lock_nested(&target->d_lock, 2);
2125 spin_lock_nested(&dentry->d_lock, 3);
2127 spin_lock_nested(&dentry->d_lock, 2);
2128 spin_lock_nested(&target->d_lock, 3);
2132 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2133 struct dentry *target)
2135 if (target->d_parent != dentry->d_parent)
2136 spin_unlock(&dentry->d_parent->d_lock);
2137 if (target->d_parent != target)
2138 spin_unlock(&target->d_parent->d_lock);
2142 * When switching names, the actual string doesn't strictly have to
2143 * be preserved in the target - because we're dropping the target
2144 * anyway. As such, we can just do a simple memcpy() to copy over
2145 * the new name before we switch.
2147 * Note that we have to be a lot more careful about getting the hash
2148 * switched - we have to switch the hash value properly even if it
2149 * then no longer matches the actual (corrupted) string of the target.
2150 * The hash value has to match the hash queue that the dentry is on..
2153 * __d_move - move a dentry
2154 * @dentry: entry to move
2155 * @target: new dentry
2157 * Update the dcache to reflect the move of a file name. Negative
2158 * dcache entries should not be moved in this way. Caller must hold
2159 * rename_lock, the i_mutex of the source and target directories,
2160 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2162 static void __d_move(struct dentry * dentry, struct dentry * target)
2164 if (!dentry->d_inode)
2165 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2167 BUG_ON(d_ancestor(dentry, target));
2168 BUG_ON(d_ancestor(target, dentry));
2170 dentry_lock_for_move(dentry, target);
2172 write_seqcount_begin(&dentry->d_seq);
2173 write_seqcount_begin(&target->d_seq);
2175 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2178 * Move the dentry to the target hash queue. Don't bother checking
2179 * for the same hash queue because of how unlikely it is.
2182 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2184 /* Unhash the target: dput() will then get rid of it */
2187 list_del(&dentry->d_child);
2188 list_del(&target->d_child);
2190 /* Switch the names.. */
2191 switch_names(dentry, target);
2192 swap(dentry->d_name.hash, target->d_name.hash);
2194 /* ... and switch the parents */
2195 if (IS_ROOT(dentry)) {
2196 dentry->d_parent = target->d_parent;
2197 target->d_parent = target;
2198 INIT_LIST_HEAD(&target->d_child);
2200 swap(dentry->d_parent, target->d_parent);
2202 /* And add them back to the (new) parent lists */
2203 list_add(&target->d_child, &target->d_parent->d_subdirs);
2206 list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
2208 write_seqcount_end(&target->d_seq);
2209 write_seqcount_end(&dentry->d_seq);
2211 dentry_unlock_parents_for_move(dentry, target);
2212 spin_unlock(&target->d_lock);
2213 fsnotify_d_move(dentry);
2214 spin_unlock(&dentry->d_lock);
2218 * d_move - move a dentry
2219 * @dentry: entry to move
2220 * @target: new dentry
2222 * Update the dcache to reflect the move of a file name. Negative
2223 * dcache entries should not be moved in this way. See the locking
2224 * requirements for __d_move.
2226 void d_move(struct dentry *dentry, struct dentry *target)
2228 write_seqlock(&rename_lock);
2229 __d_move(dentry, target);
2230 write_sequnlock(&rename_lock);
2232 EXPORT_SYMBOL(d_move);
2235 * d_ancestor - search for an ancestor
2236 * @p1: ancestor dentry
2239 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2240 * an ancestor of p2, else NULL.
2242 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2246 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2247 if (p->d_parent == p1)
2254 * This helper attempts to cope with remotely renamed directories
2256 * It assumes that the caller is already holding
2257 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2259 * Note: If ever the locking in lock_rename() changes, then please
2260 * remember to update this too...
2262 static struct dentry *__d_unalias(struct inode *inode,
2263 struct dentry *dentry, struct dentry *alias)
2265 struct mutex *m1 = NULL, *m2 = NULL;
2268 /* If alias and dentry share a parent, then no extra locks required */
2269 if (alias->d_parent == dentry->d_parent)
2272 /* See lock_rename() */
2273 ret = ERR_PTR(-EBUSY);
2274 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2276 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2277 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2279 m2 = &alias->d_parent->d_inode->i_mutex;
2281 __d_move(alias, dentry);
2284 spin_unlock(&inode->i_lock);
2293 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2294 * named dentry in place of the dentry to be replaced.
2295 * returns with anon->d_lock held!
2297 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2299 struct dentry *dparent, *aparent;
2301 dentry_lock_for_move(anon, dentry);
2303 write_seqcount_begin(&dentry->d_seq);
2304 write_seqcount_begin(&anon->d_seq);
2306 dparent = dentry->d_parent;
2307 aparent = anon->d_parent;
2309 switch_names(dentry, anon);
2310 swap(dentry->d_name.hash, anon->d_name.hash);
2312 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2313 list_del(&dentry->d_child);
2314 if (!IS_ROOT(dentry))
2315 list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
2317 INIT_LIST_HEAD(&dentry->d_child);
2319 anon->d_parent = (dparent == dentry) ? anon : dparent;
2320 list_del(&anon->d_child);
2322 list_add(&anon->d_child, &anon->d_parent->d_subdirs);
2324 INIT_LIST_HEAD(&anon->d_child);
2326 write_seqcount_end(&dentry->d_seq);
2327 write_seqcount_end(&anon->d_seq);
2329 dentry_unlock_parents_for_move(anon, dentry);
2330 spin_unlock(&dentry->d_lock);
2332 /* anon->d_lock still locked, returns locked */
2333 anon->d_flags &= ~DCACHE_DISCONNECTED;
2337 * d_materialise_unique - introduce an inode into the tree
2338 * @dentry: candidate dentry
2339 * @inode: inode to bind to the dentry, to which aliases may be attached
2341 * Introduces an dentry into the tree, substituting an extant disconnected
2342 * root directory alias in its place if there is one. Caller must hold the
2343 * i_mutex of the parent directory.
2345 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2347 struct dentry *actual;
2349 BUG_ON(!d_unhashed(dentry));
2353 __d_instantiate(dentry, NULL);
2358 spin_lock(&inode->i_lock);
2360 if (S_ISDIR(inode->i_mode)) {
2361 struct dentry *alias;
2363 /* Does an aliased dentry already exist? */
2364 alias = __d_find_alias(inode, 0);
2367 write_seqlock(&rename_lock);
2369 if (d_ancestor(alias, dentry)) {
2370 /* Check for loops */
2371 actual = ERR_PTR(-ELOOP);
2372 spin_unlock(&inode->i_lock);
2373 } else if (IS_ROOT(alias)) {
2374 /* Is this an anonymous mountpoint that we
2375 * could splice into our tree? */
2376 __d_materialise_dentry(dentry, alias);
2377 write_sequnlock(&rename_lock);
2381 /* Nope, but we must(!) avoid directory
2382 * aliasing. This drops inode->i_lock */
2383 actual = __d_unalias(inode, dentry, alias);
2385 write_sequnlock(&rename_lock);
2386 if (IS_ERR(actual)) {
2387 if (PTR_ERR(actual) == -ELOOP)
2388 pr_warn_ratelimited(
2389 "VFS: Lookup of '%s' in %s %s"
2390 " would have caused loop\n",
2391 dentry->d_name.name,
2392 inode->i_sb->s_type->name,
2400 /* Add a unique reference */
2401 actual = __d_instantiate_unique(dentry, inode);
2405 BUG_ON(!d_unhashed(actual));
2407 spin_lock(&actual->d_lock);
2410 spin_unlock(&actual->d_lock);
2411 spin_unlock(&inode->i_lock);
2413 if (actual == dentry) {
2414 security_d_instantiate(dentry, inode);
2421 EXPORT_SYMBOL_GPL(d_materialise_unique);
2423 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2427 return -ENAMETOOLONG;
2429 memcpy(*buffer, str, namelen);
2433 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2435 return prepend(buffer, buflen, name->name, name->len);
2439 * prepend_path - Prepend path string to a buffer
2440 * @path: the dentry/vfsmount to report
2441 * @root: root vfsmnt/dentry
2442 * @buffer: pointer to the end of the buffer
2443 * @buflen: pointer to buffer length
2445 * Caller holds the rename_lock.
2447 static int prepend_path(const struct path *path,
2448 const struct path *root,
2449 char **buffer, int *buflen)
2451 struct dentry *dentry = path->dentry;
2452 struct vfsmount *vfsmnt = path->mnt;
2456 while (dentry != root->dentry || vfsmnt != root->mnt) {
2457 struct dentry * parent;
2459 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2461 if (vfsmnt->mnt_parent == vfsmnt) {
2464 dentry = vfsmnt->mnt_mountpoint;
2465 vfsmnt = vfsmnt->mnt_parent;
2468 parent = dentry->d_parent;
2470 spin_lock(&dentry->d_lock);
2471 error = prepend_name(buffer, buflen, &dentry->d_name);
2472 spin_unlock(&dentry->d_lock);
2474 error = prepend(buffer, buflen, "/", 1);
2482 if (!error && !slash)
2483 error = prepend(buffer, buflen, "/", 1);
2489 * Filesystems needing to implement special "root names"
2490 * should do so with ->d_dname()
2492 if (IS_ROOT(dentry) &&
2493 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2494 WARN(1, "Root dentry has weird name <%.*s>\n",
2495 (int) dentry->d_name.len, dentry->d_name.name);
2498 error = prepend(buffer, buflen, "/", 1);
2500 error = vfsmnt->mnt_ns ? 1 : 2;
2505 * __d_path - return the path of a dentry
2506 * @path: the dentry/vfsmount to report
2507 * @root: root vfsmnt/dentry
2508 * @buf: buffer to return value in
2509 * @buflen: buffer length
2511 * Convert a dentry into an ASCII path name.
2513 * Returns a pointer into the buffer or an error code if the
2514 * path was too long.
2516 * "buflen" should be positive.
2518 * If the path is not reachable from the supplied root, return %NULL.
2520 char *__d_path(const struct path *path,
2521 const struct path *root,
2522 char *buf, int buflen)
2524 char *res = buf + buflen;
2527 prepend(&res, &buflen, "\0", 1);
2528 br_read_lock(vfsmount_lock);
2529 write_seqlock(&rename_lock);
2530 error = prepend_path(path, root, &res, &buflen);
2531 write_sequnlock(&rename_lock);
2532 br_read_unlock(vfsmount_lock);
2535 return ERR_PTR(error);
2541 char *d_absolute_path(const struct path *path,
2542 char *buf, int buflen)
2544 struct path root = {};
2545 char *res = buf + buflen;
2548 prepend(&res, &buflen, "\0", 1);
2549 br_read_lock(vfsmount_lock);
2550 write_seqlock(&rename_lock);
2551 error = prepend_path(path, &root, &res, &buflen);
2552 write_sequnlock(&rename_lock);
2553 br_read_unlock(vfsmount_lock);
2558 return ERR_PTR(error);
2563 * same as __d_path but appends "(deleted)" for unlinked files.
2565 static int path_with_deleted(const struct path *path,
2566 const struct path *root,
2567 char **buf, int *buflen)
2569 prepend(buf, buflen, "\0", 1);
2570 if (d_unlinked(path->dentry)) {
2571 int error = prepend(buf, buflen, " (deleted)", 10);
2576 return prepend_path(path, root, buf, buflen);
2579 static int prepend_unreachable(char **buffer, int *buflen)
2581 return prepend(buffer, buflen, "(unreachable)", 13);
2585 * d_path - return the path of a dentry
2586 * @path: path to report
2587 * @buf: buffer to return value in
2588 * @buflen: buffer length
2590 * Convert a dentry into an ASCII path name. If the entry has been deleted
2591 * the string " (deleted)" is appended. Note that this is ambiguous.
2593 * Returns a pointer into the buffer or an error code if the path was
2594 * too long. Note: Callers should use the returned pointer, not the passed
2595 * in buffer, to use the name! The implementation often starts at an offset
2596 * into the buffer, and may leave 0 bytes at the start.
2598 * "buflen" should be positive.
2600 char *d_path(const struct path *path, char *buf, int buflen)
2602 char *res = buf + buflen;
2607 * We have various synthetic filesystems that never get mounted. On
2608 * these filesystems dentries are never used for lookup purposes, and
2609 * thus don't need to be hashed. They also don't need a name until a
2610 * user wants to identify the object in /proc/pid/fd/. The little hack
2611 * below allows us to generate a name for these objects on demand:
2613 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2614 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2616 get_fs_root(current->fs, &root);
2617 br_read_lock(vfsmount_lock);
2618 write_seqlock(&rename_lock);
2619 error = path_with_deleted(path, &root, &res, &buflen);
2620 write_sequnlock(&rename_lock);
2621 br_read_unlock(vfsmount_lock);
2623 res = ERR_PTR(error);
2627 EXPORT_SYMBOL(d_path);
2630 * d_path_with_unreachable - return the path of a dentry
2631 * @path: path to report
2632 * @buf: buffer to return value in
2633 * @buflen: buffer length
2635 * The difference from d_path() is that this prepends "(unreachable)"
2636 * to paths which are unreachable from the current process' root.
2638 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2640 char *res = buf + buflen;
2644 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2645 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2647 get_fs_root(current->fs, &root);
2648 write_seqlock(&rename_lock);
2649 error = path_with_deleted(path, &root, &res, &buflen);
2651 error = prepend_unreachable(&res, &buflen);
2652 write_sequnlock(&rename_lock);
2655 res = ERR_PTR(error);
2661 * Helper function for dentry_operations.d_dname() members
2663 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2664 const char *fmt, ...)
2670 va_start(args, fmt);
2671 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2674 if (sz > sizeof(temp) || sz > buflen)
2675 return ERR_PTR(-ENAMETOOLONG);
2677 buffer += buflen - sz;
2678 return memcpy(buffer, temp, sz);
2682 * Write full pathname from the root of the filesystem into the buffer.
2684 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2686 char *end = buf + buflen;
2689 prepend(&end, &buflen, "\0", 1);
2696 while (!IS_ROOT(dentry)) {
2697 struct dentry *parent = dentry->d_parent;
2701 spin_lock(&dentry->d_lock);
2702 error = prepend_name(&end, &buflen, &dentry->d_name);
2703 spin_unlock(&dentry->d_lock);
2704 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2712 return ERR_PTR(-ENAMETOOLONG);
2715 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2719 write_seqlock(&rename_lock);
2720 retval = __dentry_path(dentry, buf, buflen);
2721 write_sequnlock(&rename_lock);
2725 EXPORT_SYMBOL(dentry_path_raw);
2727 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2732 write_seqlock(&rename_lock);
2733 if (d_unlinked(dentry)) {
2735 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2739 retval = __dentry_path(dentry, buf, buflen);
2740 write_sequnlock(&rename_lock);
2741 if (!IS_ERR(retval) && p)
2742 *p = '/'; /* restore '/' overriden with '\0' */
2745 return ERR_PTR(-ENAMETOOLONG);
2749 * NOTE! The user-level library version returns a
2750 * character pointer. The kernel system call just
2751 * returns the length of the buffer filled (which
2752 * includes the ending '\0' character), or a negative
2753 * error value. So libc would do something like
2755 * char *getcwd(char * buf, size_t size)
2759 * retval = sys_getcwd(buf, size);
2766 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2769 struct path pwd, root;
2770 char *page = (char *) __get_free_page(GFP_USER);
2775 get_fs_root_and_pwd(current->fs, &root, &pwd);
2778 br_read_lock(vfsmount_lock);
2779 write_seqlock(&rename_lock);
2780 if (!d_unlinked(pwd.dentry)) {
2782 char *cwd = page + PAGE_SIZE;
2783 int buflen = PAGE_SIZE;
2785 prepend(&cwd, &buflen, "\0", 1);
2786 error = prepend_path(&pwd, &root, &cwd, &buflen);
2787 write_sequnlock(&rename_lock);
2788 br_read_unlock(vfsmount_lock);
2793 /* Unreachable from current root */
2795 error = prepend_unreachable(&cwd, &buflen);
2801 len = PAGE_SIZE + page - cwd;
2804 if (copy_to_user(buf, cwd, len))
2808 write_sequnlock(&rename_lock);
2809 br_read_unlock(vfsmount_lock);
2815 free_page((unsigned long) page);
2820 * Test whether new_dentry is a subdirectory of old_dentry.
2822 * Trivially implemented using the dcache structure
2826 * is_subdir - is new dentry a subdirectory of old_dentry
2827 * @new_dentry: new dentry
2828 * @old_dentry: old dentry
2830 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2831 * Returns 0 otherwise.
2832 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2835 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2840 if (new_dentry == old_dentry)
2844 /* for restarting inner loop in case of seq retry */
2845 seq = read_seqbegin(&rename_lock);
2847 * Need rcu_readlock to protect against the d_parent trashing
2851 if (d_ancestor(old_dentry, new_dentry))
2856 } while (read_seqretry(&rename_lock, seq));
2861 int path_is_under(struct path *path1, struct path *path2)
2863 struct vfsmount *mnt = path1->mnt;
2864 struct dentry *dentry = path1->dentry;
2867 br_read_lock(vfsmount_lock);
2868 if (mnt != path2->mnt) {
2870 if (mnt->mnt_parent == mnt) {
2871 br_read_unlock(vfsmount_lock);
2874 if (mnt->mnt_parent == path2->mnt)
2876 mnt = mnt->mnt_parent;
2878 dentry = mnt->mnt_mountpoint;
2880 res = is_subdir(dentry, path2->dentry);
2881 br_read_unlock(vfsmount_lock);
2884 EXPORT_SYMBOL(path_is_under);
2886 void d_genocide(struct dentry *root)
2888 struct dentry *this_parent;
2889 struct list_head *next;
2893 seq = read_seqbegin(&rename_lock);
2896 spin_lock(&this_parent->d_lock);
2898 next = this_parent->d_subdirs.next;
2900 while (next != &this_parent->d_subdirs) {
2901 struct list_head *tmp = next;
2902 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
2905 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2906 if (d_unhashed(dentry) || !dentry->d_inode) {
2907 spin_unlock(&dentry->d_lock);
2910 if (!list_empty(&dentry->d_subdirs)) {
2911 spin_unlock(&this_parent->d_lock);
2912 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2913 this_parent = dentry;
2914 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2917 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2918 dentry->d_flags |= DCACHE_GENOCIDE;
2921 spin_unlock(&dentry->d_lock);
2925 if (this_parent != root) {
2926 struct dentry *child = this_parent;
2927 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2928 this_parent->d_flags |= DCACHE_GENOCIDE;
2929 this_parent->d_count--;
2931 this_parent = child->d_parent;
2933 spin_unlock(&child->d_lock);
2934 spin_lock(&this_parent->d_lock);
2936 /* might go back up the wrong parent if we have had a rename */
2937 if (!locked && read_seqretry(&rename_lock, seq))
2939 next = child->d_child.next;
2940 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
2941 if (next == &this_parent->d_subdirs)
2943 child = list_entry(next, struct dentry, d_child);
2949 if (!locked && read_seqretry(&rename_lock, seq))
2951 spin_unlock(&this_parent->d_lock);
2954 write_sequnlock(&rename_lock);
2958 spin_unlock(&this_parent->d_lock);
2963 write_seqlock(&rename_lock);
2968 * find_inode_number - check for dentry with name
2969 * @dir: directory to check
2970 * @name: Name to find.
2972 * Check whether a dentry already exists for the given name,
2973 * and return the inode number if it has an inode. Otherwise
2976 * This routine is used to post-process directory listings for
2977 * filesystems using synthetic inode numbers, and is necessary
2978 * to keep getcwd() working.
2981 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2983 struct dentry * dentry;
2986 dentry = d_hash_and_lookup(dir, name);
2988 if (dentry->d_inode)
2989 ino = dentry->d_inode->i_ino;
2994 EXPORT_SYMBOL(find_inode_number);
2996 static __initdata unsigned long dhash_entries;
2997 static int __init set_dhash_entries(char *str)
3001 dhash_entries = simple_strtoul(str, &str, 0);
3004 __setup("dhash_entries=", set_dhash_entries);
3006 static void __init dcache_init_early(void)
3010 /* If hashes are distributed across NUMA nodes, defer
3011 * hash allocation until vmalloc space is available.
3017 alloc_large_system_hash("Dentry cache",
3018 sizeof(struct hlist_bl_head),
3026 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3027 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3030 static void __init dcache_init(void)
3035 * A constructor could be added for stable state like the lists,
3036 * but it is probably not worth it because of the cache nature
3039 dentry_cache = KMEM_CACHE(dentry,
3040 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3042 /* Hash may have been set up in dcache_init_early */
3047 alloc_large_system_hash("Dentry cache",
3048 sizeof(struct hlist_bl_head),
3056 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3057 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3060 /* SLAB cache for __getname() consumers */
3061 struct kmem_cache *names_cachep __read_mostly;
3062 EXPORT_SYMBOL(names_cachep);
3064 EXPORT_SYMBOL(d_genocide);
3066 void __init vfs_caches_init_early(void)
3068 dcache_init_early();
3072 void __init vfs_caches_init(unsigned long mempages)
3074 unsigned long reserve;
3076 /* Base hash sizes on available memory, with a reserve equal to
3077 150% of current kernel size */
3079 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3080 mempages -= reserve;
3082 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3083 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3087 files_init(mempages);