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>
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
64 * dentry->d_inode->i_lock
67 * dcache_hash_bucket lock
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
73 * dentry->d_parent->d_lock
76 * If no ancestor relationship:
77 * if (dentry1 < dentry2)
81 int sysctl_vfs_cache_pressure __read_mostly = 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
87 EXPORT_SYMBOL(rename_lock);
89 static struct kmem_cache *dentry_cache __read_mostly;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
99 #define D_HASHBITS d_hash_shift
100 #define D_HASHMASK d_hash_mask
102 static unsigned int d_hash_mask __read_mostly;
103 static unsigned int d_hash_shift __read_mostly;
105 static struct hlist_bl_head *dentry_hashtable __read_mostly;
107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
110 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
111 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
112 return dentry_hashtable + (hash & D_HASHMASK);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat = {
120 static DEFINE_PER_CPU(unsigned int, nr_dentry);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 static int get_nr_dentry(void)
127 for_each_possible_cpu(i)
128 sum += per_cpu(nr_dentry, i);
129 return sum < 0 ? 0 : sum;
132 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
133 size_t *lenp, loff_t *ppos)
135 dentry_stat.nr_dentry = get_nr_dentry();
136 return proc_dointvec(table, write, buffer, lenp, ppos);
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
144 static inline int dentry_cmp(const unsigned char *cs, size_t scount,
145 const unsigned char *ct, size_t tcount)
147 if (scount != tcount)
160 static void __d_free(struct rcu_head *head)
162 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
164 WARN_ON(!list_empty(&dentry->d_alias));
165 if (dname_external(dentry))
166 kfree(dentry->d_name.name);
167 kmem_cache_free(dentry_cache, dentry);
173 static void d_free(struct dentry *dentry)
175 BUG_ON(dentry->d_count);
176 this_cpu_dec(nr_dentry);
177 if (dentry->d_op && dentry->d_op->d_release)
178 dentry->d_op->d_release(dentry);
180 /* if dentry was never visible to RCU, immediate free is OK */
181 if (!(dentry->d_flags & DCACHE_RCUACCESS))
182 __d_free(&dentry->d_u.d_rcu);
184 call_rcu(&dentry->d_u.d_rcu, __d_free);
188 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
189 * @dentry: the target dentry
190 * After this call, in-progress rcu-walk path lookup will fail. This
191 * should be called after unhashing, and after changing d_inode (if
192 * the dentry has not already been unhashed).
194 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
196 assert_spin_locked(&dentry->d_lock);
197 /* Go through a barrier */
198 write_seqcount_barrier(&dentry->d_seq);
202 * Release the dentry's inode, using the filesystem
203 * d_iput() operation if defined. Dentry has no refcount
206 static void dentry_iput(struct dentry * dentry)
207 __releases(dentry->d_lock)
208 __releases(dentry->d_inode->i_lock)
210 struct inode *inode = dentry->d_inode;
212 dentry->d_inode = NULL;
213 list_del_init(&dentry->d_alias);
214 spin_unlock(&dentry->d_lock);
215 spin_unlock(&inode->i_lock);
217 fsnotify_inoderemove(inode);
218 if (dentry->d_op && dentry->d_op->d_iput)
219 dentry->d_op->d_iput(dentry, inode);
223 spin_unlock(&dentry->d_lock);
228 * Release the dentry's inode, using the filesystem
229 * d_iput() operation if defined. dentry remains in-use.
231 static void dentry_unlink_inode(struct dentry * dentry)
232 __releases(dentry->d_lock)
233 __releases(dentry->d_inode->i_lock)
235 struct inode *inode = dentry->d_inode;
236 dentry->d_inode = NULL;
237 list_del_init(&dentry->d_alias);
238 dentry_rcuwalk_barrier(dentry);
239 spin_unlock(&dentry->d_lock);
240 spin_unlock(&inode->i_lock);
242 fsnotify_inoderemove(inode);
243 if (dentry->d_op && dentry->d_op->d_iput)
244 dentry->d_op->d_iput(dentry, inode);
250 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
252 static void dentry_lru_add(struct dentry *dentry)
254 if (list_empty(&dentry->d_lru)) {
255 spin_lock(&dcache_lru_lock);
256 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
257 dentry->d_sb->s_nr_dentry_unused++;
258 dentry_stat.nr_unused++;
259 spin_unlock(&dcache_lru_lock);
263 static void __dentry_lru_del(struct dentry *dentry)
265 list_del_init(&dentry->d_lru);
266 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
267 dentry->d_sb->s_nr_dentry_unused--;
268 dentry_stat.nr_unused--;
272 * Remove a dentry with references from the LRU.
274 static void dentry_lru_del(struct dentry *dentry)
276 if (!list_empty(&dentry->d_lru)) {
277 spin_lock(&dcache_lru_lock);
278 __dentry_lru_del(dentry);
279 spin_unlock(&dcache_lru_lock);
284 * Remove a dentry that is unreferenced and about to be pruned
285 * (unhashed and destroyed) from the LRU, and inform the file system.
286 * This wrapper should be called _prior_ to unhashing a victim dentry.
288 static void dentry_lru_prune(struct dentry *dentry)
290 if (!list_empty(&dentry->d_lru)) {
291 if (dentry->d_flags & DCACHE_OP_PRUNE)
292 dentry->d_op->d_prune(dentry);
294 spin_lock(&dcache_lru_lock);
295 __dentry_lru_del(dentry);
296 spin_unlock(&dcache_lru_lock);
300 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
302 spin_lock(&dcache_lru_lock);
303 if (list_empty(&dentry->d_lru)) {
304 list_add_tail(&dentry->d_lru, list);
305 dentry->d_sb->s_nr_dentry_unused++;
306 dentry_stat.nr_unused++;
308 list_move_tail(&dentry->d_lru, list);
310 spin_unlock(&dcache_lru_lock);
314 * d_kill - kill dentry and return parent
315 * @dentry: dentry to kill
316 * @parent: parent dentry
318 * The dentry must already be unhashed and removed from the LRU.
320 * If this is the root of the dentry tree, return NULL.
322 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
325 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
326 __releases(dentry->d_lock)
327 __releases(parent->d_lock)
328 __releases(dentry->d_inode->i_lock)
330 list_del(&dentry->d_u.d_child);
332 * Inform try_to_ascend() that we are no longer attached to the
335 dentry->d_flags |= DCACHE_DISCONNECTED;
337 spin_unlock(&parent->d_lock);
340 * dentry_iput drops the locks, at which point nobody (except
341 * transient RCU lookups) can reach this dentry.
348 * Unhash a dentry without inserting an RCU walk barrier or checking that
349 * dentry->d_lock is locked. The caller must take care of that, if
352 static void __d_shrink(struct dentry *dentry)
354 if (!d_unhashed(dentry)) {
355 struct hlist_bl_head *b;
356 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
357 b = &dentry->d_sb->s_anon;
359 b = d_hash(dentry->d_parent, dentry->d_name.hash);
362 __hlist_bl_del(&dentry->d_hash);
363 dentry->d_hash.pprev = NULL;
369 * d_drop - drop a dentry
370 * @dentry: dentry to drop
372 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
373 * be found through a VFS lookup any more. Note that this is different from
374 * deleting the dentry - d_delete will try to mark the dentry negative if
375 * possible, giving a successful _negative_ lookup, while d_drop will
376 * just make the cache lookup fail.
378 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
379 * reason (NFS timeouts or autofs deletes).
381 * __d_drop requires dentry->d_lock.
383 void __d_drop(struct dentry *dentry)
385 if (!d_unhashed(dentry)) {
387 dentry_rcuwalk_barrier(dentry);
390 EXPORT_SYMBOL(__d_drop);
392 void d_drop(struct dentry *dentry)
394 spin_lock(&dentry->d_lock);
396 spin_unlock(&dentry->d_lock);
398 EXPORT_SYMBOL(d_drop);
401 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
402 * @dentry: dentry to drop
404 * This is called when we do a lookup on a placeholder dentry that needed to be
405 * looked up. The dentry should have been hashed in order for it to be found by
406 * the lookup code, but now needs to be unhashed while we do the actual lookup
407 * and clear the DCACHE_NEED_LOOKUP flag.
409 void d_clear_need_lookup(struct dentry *dentry)
411 spin_lock(&dentry->d_lock);
413 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
414 spin_unlock(&dentry->d_lock);
416 EXPORT_SYMBOL(d_clear_need_lookup);
419 * Finish off a dentry we've decided to kill.
420 * dentry->d_lock must be held, returns with it unlocked.
421 * If ref is non-zero, then decrement the refcount too.
422 * Returns dentry requiring refcount drop, or NULL if we're done.
424 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
425 __releases(dentry->d_lock)
428 struct dentry *parent;
430 inode = dentry->d_inode;
431 if (inode && !spin_trylock(&inode->i_lock)) {
433 spin_unlock(&dentry->d_lock);
435 return dentry; /* try again with same dentry */
440 parent = dentry->d_parent;
441 if (parent && !spin_trylock(&parent->d_lock)) {
443 spin_unlock(&inode->i_lock);
450 * if dentry was on the d_lru list delete it from there.
451 * inform the fs via d_prune that this dentry is about to be
452 * unhashed and destroyed.
454 dentry_lru_prune(dentry);
455 /* if it was on the hash then remove it */
457 return d_kill(dentry, parent);
463 * This is complicated by the fact that we do not want to put
464 * dentries that are no longer on any hash chain on the unused
465 * list: we'd much rather just get rid of them immediately.
467 * However, that implies that we have to traverse the dentry
468 * tree upwards to the parents which might _also_ now be
469 * scheduled for deletion (it may have been only waiting for
470 * its last child to go away).
472 * This tail recursion is done by hand as we don't want to depend
473 * on the compiler to always get this right (gcc generally doesn't).
474 * Real recursion would eat up our stack space.
478 * dput - release a dentry
479 * @dentry: dentry to release
481 * Release a dentry. This will drop the usage count and if appropriate
482 * call the dentry unlink method as well as removing it from the queues and
483 * releasing its resources. If the parent dentries were scheduled for release
484 * they too may now get deleted.
486 void dput(struct dentry *dentry)
492 if (dentry->d_count == 1)
494 spin_lock(&dentry->d_lock);
495 BUG_ON(!dentry->d_count);
496 if (dentry->d_count > 1) {
498 spin_unlock(&dentry->d_lock);
502 if (dentry->d_flags & DCACHE_OP_DELETE) {
503 if (dentry->d_op->d_delete(dentry))
507 /* Unreachable? Get rid of it */
508 if (d_unhashed(dentry))
512 * If this dentry needs lookup, don't set the referenced flag so that it
513 * is more likely to be cleaned up by the dcache shrinker in case of
516 if (!d_need_lookup(dentry))
517 dentry->d_flags |= DCACHE_REFERENCED;
518 dentry_lru_add(dentry);
521 spin_unlock(&dentry->d_lock);
525 dentry = dentry_kill(dentry, 1);
532 * d_invalidate - invalidate a dentry
533 * @dentry: dentry to invalidate
535 * Try to invalidate the dentry if it turns out to be
536 * possible. If there are other dentries that can be
537 * reached through this one we can't delete it and we
538 * return -EBUSY. On success we return 0.
543 int d_invalidate(struct dentry * dentry)
546 * If it's already been dropped, return OK.
548 spin_lock(&dentry->d_lock);
549 if (d_unhashed(dentry)) {
550 spin_unlock(&dentry->d_lock);
554 * Check whether to do a partial shrink_dcache
555 * to get rid of unused child entries.
557 if (!list_empty(&dentry->d_subdirs)) {
558 spin_unlock(&dentry->d_lock);
559 shrink_dcache_parent(dentry);
560 spin_lock(&dentry->d_lock);
564 * Somebody else still using it?
566 * If it's a directory, we can't drop it
567 * for fear of somebody re-populating it
568 * with children (even though dropping it
569 * would make it unreachable from the root,
570 * we might still populate it if it was a
571 * working directory or similar).
572 * We also need to leave mountpoints alone,
575 if (dentry->d_count > 1 && dentry->d_inode) {
576 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
577 spin_unlock(&dentry->d_lock);
583 spin_unlock(&dentry->d_lock);
586 EXPORT_SYMBOL(d_invalidate);
588 /* This must be called with d_lock held */
589 static inline void __dget_dlock(struct dentry *dentry)
594 static inline void __dget(struct dentry *dentry)
596 spin_lock(&dentry->d_lock);
597 __dget_dlock(dentry);
598 spin_unlock(&dentry->d_lock);
601 struct dentry *dget_parent(struct dentry *dentry)
607 * Don't need rcu_dereference because we re-check it was correct under
611 ret = dentry->d_parent;
612 spin_lock(&ret->d_lock);
613 if (unlikely(ret != dentry->d_parent)) {
614 spin_unlock(&ret->d_lock);
619 BUG_ON(!ret->d_count);
621 spin_unlock(&ret->d_lock);
624 EXPORT_SYMBOL(dget_parent);
627 * d_find_alias - grab a hashed alias of inode
628 * @inode: inode in question
629 * @want_discon: flag, used by d_splice_alias, to request
630 * that only a DISCONNECTED alias be returned.
632 * If inode has a hashed alias, or is a directory and has any alias,
633 * acquire the reference to alias and return it. Otherwise return NULL.
634 * Notice that if inode is a directory there can be only one alias and
635 * it can be unhashed only if it has no children, or if it is the root
638 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
639 * any other hashed alias over that one unless @want_discon is set,
640 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
642 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
644 struct dentry *alias, *discon_alias;
648 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
649 spin_lock(&alias->d_lock);
650 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
651 if (IS_ROOT(alias) &&
652 (alias->d_flags & DCACHE_DISCONNECTED)) {
653 discon_alias = alias;
654 } else if (!want_discon) {
656 spin_unlock(&alias->d_lock);
660 spin_unlock(&alias->d_lock);
663 alias = discon_alias;
664 spin_lock(&alias->d_lock);
665 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
666 if (IS_ROOT(alias) &&
667 (alias->d_flags & DCACHE_DISCONNECTED)) {
669 spin_unlock(&alias->d_lock);
673 spin_unlock(&alias->d_lock);
679 struct dentry *d_find_alias(struct inode *inode)
681 struct dentry *de = NULL;
683 if (!list_empty(&inode->i_dentry)) {
684 spin_lock(&inode->i_lock);
685 de = __d_find_alias(inode, 0);
686 spin_unlock(&inode->i_lock);
690 EXPORT_SYMBOL(d_find_alias);
693 * Try to kill dentries associated with this inode.
694 * WARNING: you must own a reference to inode.
696 void d_prune_aliases(struct inode *inode)
698 struct dentry *dentry;
700 spin_lock(&inode->i_lock);
701 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
702 spin_lock(&dentry->d_lock);
703 if (!dentry->d_count) {
704 __dget_dlock(dentry);
706 spin_unlock(&dentry->d_lock);
707 spin_unlock(&inode->i_lock);
711 spin_unlock(&dentry->d_lock);
713 spin_unlock(&inode->i_lock);
715 EXPORT_SYMBOL(d_prune_aliases);
718 * Try to throw away a dentry - free the inode, dput the parent.
719 * Requires dentry->d_lock is held, and dentry->d_count == 0.
720 * Releases dentry->d_lock.
722 * This may fail if locks cannot be acquired no problem, just try again.
724 static void try_prune_one_dentry(struct dentry *dentry)
725 __releases(dentry->d_lock)
727 struct dentry *parent;
729 parent = dentry_kill(dentry, 0);
731 * If dentry_kill returns NULL, we have nothing more to do.
732 * if it returns the same dentry, trylocks failed. In either
733 * case, just loop again.
735 * Otherwise, we need to prune ancestors too. This is necessary
736 * to prevent quadratic behavior of shrink_dcache_parent(), but
737 * is also expected to be beneficial in reducing dentry cache
742 if (parent == dentry)
745 /* Prune ancestors. */
748 spin_lock(&dentry->d_lock);
749 if (dentry->d_count > 1) {
751 spin_unlock(&dentry->d_lock);
754 dentry = dentry_kill(dentry, 1);
758 static void shrink_dentry_list(struct list_head *list)
760 struct dentry *dentry;
764 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
765 if (&dentry->d_lru == list)
767 spin_lock(&dentry->d_lock);
768 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
769 spin_unlock(&dentry->d_lock);
774 * We found an inuse dentry which was not removed from
775 * the LRU because of laziness during lookup. Do not free
776 * it - just keep it off the LRU list.
778 if (dentry->d_count) {
779 dentry_lru_del(dentry);
780 spin_unlock(&dentry->d_lock);
786 try_prune_one_dentry(dentry);
794 * prune_dcache_sb - shrink the dcache
796 * @count: number of entries to try to free
798 * Attempt to shrink the superblock dcache LRU by @count entries. This is
799 * done when we need more memory an called from the superblock shrinker
802 * This function may fail to free any resources if all the dentries are in
805 void prune_dcache_sb(struct super_block *sb, int count)
807 struct dentry *dentry;
808 LIST_HEAD(referenced);
812 spin_lock(&dcache_lru_lock);
813 while (!list_empty(&sb->s_dentry_lru)) {
814 dentry = list_entry(sb->s_dentry_lru.prev,
815 struct dentry, d_lru);
816 BUG_ON(dentry->d_sb != sb);
818 if (!spin_trylock(&dentry->d_lock)) {
819 spin_unlock(&dcache_lru_lock);
824 if (dentry->d_flags & DCACHE_REFERENCED) {
825 dentry->d_flags &= ~DCACHE_REFERENCED;
826 list_move(&dentry->d_lru, &referenced);
827 spin_unlock(&dentry->d_lock);
829 list_move_tail(&dentry->d_lru, &tmp);
830 dentry->d_flags |= DCACHE_SHRINK_LIST;
831 spin_unlock(&dentry->d_lock);
835 cond_resched_lock(&dcache_lru_lock);
837 if (!list_empty(&referenced))
838 list_splice(&referenced, &sb->s_dentry_lru);
839 spin_unlock(&dcache_lru_lock);
841 shrink_dentry_list(&tmp);
845 * shrink_dcache_sb - shrink dcache for a superblock
848 * Shrink the dcache for the specified super block. This is used to free
849 * the dcache before unmounting a file system.
851 void shrink_dcache_sb(struct super_block *sb)
855 spin_lock(&dcache_lru_lock);
856 while (!list_empty(&sb->s_dentry_lru)) {
857 list_splice_init(&sb->s_dentry_lru, &tmp);
858 spin_unlock(&dcache_lru_lock);
859 shrink_dentry_list(&tmp);
860 spin_lock(&dcache_lru_lock);
862 spin_unlock(&dcache_lru_lock);
864 EXPORT_SYMBOL(shrink_dcache_sb);
867 * destroy a single subtree of dentries for unmount
868 * - see the comments on shrink_dcache_for_umount() for a description of the
871 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
873 struct dentry *parent;
875 BUG_ON(!IS_ROOT(dentry));
878 /* descend to the first leaf in the current subtree */
879 while (!list_empty(&dentry->d_subdirs))
880 dentry = list_entry(dentry->d_subdirs.next,
881 struct dentry, d_u.d_child);
883 /* consume the dentries from this leaf up through its parents
884 * until we find one with children or run out altogether */
889 * remove the dentry from the lru, and inform
890 * the fs that this dentry is about to be
891 * unhashed and destroyed.
893 dentry_lru_prune(dentry);
896 if (dentry->d_count != 0) {
898 "BUG: Dentry %p{i=%lx,n=%s}"
900 " [unmount of %s %s]\n",
903 dentry->d_inode->i_ino : 0UL,
906 dentry->d_sb->s_type->name,
911 if (IS_ROOT(dentry)) {
913 list_del(&dentry->d_u.d_child);
915 parent = dentry->d_parent;
917 list_del(&dentry->d_u.d_child);
920 inode = dentry->d_inode;
922 dentry->d_inode = NULL;
923 list_del_init(&dentry->d_alias);
924 if (dentry->d_op && dentry->d_op->d_iput)
925 dentry->d_op->d_iput(dentry, inode);
932 /* finished when we fall off the top of the tree,
933 * otherwise we ascend to the parent and move to the
934 * next sibling if there is one */
938 } while (list_empty(&dentry->d_subdirs));
940 dentry = list_entry(dentry->d_subdirs.next,
941 struct dentry, d_u.d_child);
946 * destroy the dentries attached to a superblock on unmounting
947 * - we don't need to use dentry->d_lock because:
948 * - the superblock is detached from all mountings and open files, so the
949 * dentry trees will not be rearranged by the VFS
950 * - s_umount is write-locked, so the memory pressure shrinker will ignore
951 * any dentries belonging to this superblock that it comes across
952 * - the filesystem itself is no longer permitted to rearrange the dentries
955 void shrink_dcache_for_umount(struct super_block *sb)
957 struct dentry *dentry;
959 if (down_read_trylock(&sb->s_umount))
965 shrink_dcache_for_umount_subtree(dentry);
967 while (!hlist_bl_empty(&sb->s_anon)) {
968 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
969 shrink_dcache_for_umount_subtree(dentry);
974 * This tries to ascend one level of parenthood, but
975 * we can race with renaming, so we need to re-check
976 * the parenthood after dropping the lock and check
977 * that the sequence number still matches.
979 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
981 struct dentry *new = old->d_parent;
984 spin_unlock(&old->d_lock);
985 spin_lock(&new->d_lock);
988 * might go back up the wrong parent if we have had a rename
991 if (new != old->d_parent ||
992 (old->d_flags & DCACHE_DISCONNECTED) ||
993 (!locked && read_seqretry(&rename_lock, seq))) {
994 spin_unlock(&new->d_lock);
1003 * Search for at least 1 mount point in the dentry's subdirs.
1004 * We descend to the next level whenever the d_subdirs
1005 * list is non-empty and continue searching.
1009 * have_submounts - check for mounts over a dentry
1010 * @parent: dentry to check.
1012 * Return true if the parent or its subdirectories contain
1015 int have_submounts(struct dentry *parent)
1017 struct dentry *this_parent;
1018 struct list_head *next;
1022 seq = read_seqbegin(&rename_lock);
1024 this_parent = parent;
1026 if (d_mountpoint(parent))
1028 spin_lock(&this_parent->d_lock);
1030 next = this_parent->d_subdirs.next;
1032 while (next != &this_parent->d_subdirs) {
1033 struct list_head *tmp = next;
1034 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1037 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1038 /* Have we found a mount point ? */
1039 if (d_mountpoint(dentry)) {
1040 spin_unlock(&dentry->d_lock);
1041 spin_unlock(&this_parent->d_lock);
1044 if (!list_empty(&dentry->d_subdirs)) {
1045 spin_unlock(&this_parent->d_lock);
1046 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1047 this_parent = dentry;
1048 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1051 spin_unlock(&dentry->d_lock);
1054 * All done at this level ... ascend and resume the search.
1056 if (this_parent != parent) {
1057 struct dentry *child = this_parent;
1058 this_parent = try_to_ascend(this_parent, locked, seq);
1061 next = child->d_u.d_child.next;
1064 spin_unlock(&this_parent->d_lock);
1065 if (!locked && read_seqretry(&rename_lock, seq))
1068 write_sequnlock(&rename_lock);
1069 return 0; /* No mount points found in tree */
1071 if (!locked && read_seqretry(&rename_lock, seq))
1074 write_sequnlock(&rename_lock);
1079 write_seqlock(&rename_lock);
1082 EXPORT_SYMBOL(have_submounts);
1085 * Search the dentry child list for the specified parent,
1086 * and move any unused dentries to the end of the unused
1087 * list for prune_dcache(). We descend to the next level
1088 * whenever the d_subdirs list is non-empty and continue
1091 * It returns zero iff there are no unused children,
1092 * otherwise it returns the number of children moved to
1093 * the end of the unused list. This may not be the total
1094 * number of unused children, because select_parent can
1095 * drop the lock and return early due to latency
1098 static int select_parent(struct dentry *parent, struct list_head *dispose)
1100 struct dentry *this_parent;
1101 struct list_head *next;
1106 seq = read_seqbegin(&rename_lock);
1108 this_parent = parent;
1109 spin_lock(&this_parent->d_lock);
1111 next = this_parent->d_subdirs.next;
1113 while (next != &this_parent->d_subdirs) {
1114 struct list_head *tmp = next;
1115 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1118 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1121 * move only zero ref count dentries to the dispose list.
1123 * Those which are presently on the shrink list, being processed
1124 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1125 * loop in shrink_dcache_parent() might not make any progress
1128 if (dentry->d_count) {
1129 dentry_lru_del(dentry);
1130 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1131 dentry_lru_move_list(dentry, dispose);
1132 dentry->d_flags |= DCACHE_SHRINK_LIST;
1136 * We can return to the caller if we have found some (this
1137 * ensures forward progress). We'll be coming back to find
1140 if (found && need_resched()) {
1141 spin_unlock(&dentry->d_lock);
1146 * Descend a level if the d_subdirs list is non-empty.
1148 if (!list_empty(&dentry->d_subdirs)) {
1149 spin_unlock(&this_parent->d_lock);
1150 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1151 this_parent = dentry;
1152 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1156 spin_unlock(&dentry->d_lock);
1159 * All done at this level ... ascend and resume the search.
1161 if (this_parent != parent) {
1162 struct dentry *child = this_parent;
1163 this_parent = try_to_ascend(this_parent, locked, seq);
1166 next = child->d_u.d_child.next;
1170 spin_unlock(&this_parent->d_lock);
1171 if (!locked && read_seqretry(&rename_lock, seq))
1174 write_sequnlock(&rename_lock);
1181 write_seqlock(&rename_lock);
1186 * shrink_dcache_parent - prune dcache
1187 * @parent: parent of entries to prune
1189 * Prune the dcache to remove unused children of the parent dentry.
1191 void shrink_dcache_parent(struct dentry * parent)
1196 while ((found = select_parent(parent, &dispose)) != 0)
1197 shrink_dentry_list(&dispose);
1199 EXPORT_SYMBOL(shrink_dcache_parent);
1202 * __d_alloc - allocate a dcache entry
1203 * @sb: filesystem it will belong to
1204 * @name: qstr of the name
1206 * Allocates a dentry. It returns %NULL if there is insufficient memory
1207 * available. On a success the dentry is returned. The name passed in is
1208 * copied and the copy passed in may be reused after this call.
1211 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1213 struct dentry *dentry;
1216 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1220 if (name->len > DNAME_INLINE_LEN-1) {
1221 dname = kmalloc(name->len + 1, GFP_KERNEL);
1223 kmem_cache_free(dentry_cache, dentry);
1227 dname = dentry->d_iname;
1229 dentry->d_name.name = dname;
1231 dentry->d_name.len = name->len;
1232 dentry->d_name.hash = name->hash;
1233 memcpy(dname, name->name, name->len);
1234 dname[name->len] = 0;
1236 dentry->d_count = 1;
1237 dentry->d_flags = 0;
1238 spin_lock_init(&dentry->d_lock);
1239 seqcount_init(&dentry->d_seq);
1240 dentry->d_inode = NULL;
1241 dentry->d_parent = dentry;
1243 dentry->d_op = NULL;
1244 dentry->d_fsdata = NULL;
1245 INIT_HLIST_BL_NODE(&dentry->d_hash);
1246 INIT_LIST_HEAD(&dentry->d_lru);
1247 INIT_LIST_HEAD(&dentry->d_subdirs);
1248 INIT_LIST_HEAD(&dentry->d_alias);
1249 INIT_LIST_HEAD(&dentry->d_u.d_child);
1250 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1252 this_cpu_inc(nr_dentry);
1258 * d_alloc - allocate a dcache entry
1259 * @parent: parent of entry to allocate
1260 * @name: qstr of the name
1262 * Allocates a dentry. It returns %NULL if there is insufficient memory
1263 * available. On a success the dentry is returned. The name passed in is
1264 * copied and the copy passed in may be reused after this call.
1266 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1268 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1272 spin_lock(&parent->d_lock);
1274 * don't need child lock because it is not subject
1275 * to concurrency here
1277 __dget_dlock(parent);
1278 dentry->d_parent = parent;
1279 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1280 spin_unlock(&parent->d_lock);
1284 EXPORT_SYMBOL(d_alloc);
1286 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1288 struct dentry *dentry = __d_alloc(sb, name);
1290 dentry->d_flags |= DCACHE_DISCONNECTED;
1293 EXPORT_SYMBOL(d_alloc_pseudo);
1295 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1300 q.len = strlen(name);
1301 q.hash = full_name_hash(q.name, q.len);
1302 return d_alloc(parent, &q);
1304 EXPORT_SYMBOL(d_alloc_name);
1306 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1308 WARN_ON_ONCE(dentry->d_op);
1309 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1311 DCACHE_OP_REVALIDATE |
1312 DCACHE_OP_DELETE ));
1317 dentry->d_flags |= DCACHE_OP_HASH;
1319 dentry->d_flags |= DCACHE_OP_COMPARE;
1320 if (op->d_revalidate)
1321 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1323 dentry->d_flags |= DCACHE_OP_DELETE;
1325 dentry->d_flags |= DCACHE_OP_PRUNE;
1328 EXPORT_SYMBOL(d_set_d_op);
1330 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1332 spin_lock(&dentry->d_lock);
1334 if (unlikely(IS_AUTOMOUNT(inode)))
1335 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1336 list_add(&dentry->d_alias, &inode->i_dentry);
1338 dentry->d_inode = inode;
1339 dentry_rcuwalk_barrier(dentry);
1340 spin_unlock(&dentry->d_lock);
1341 fsnotify_d_instantiate(dentry, inode);
1345 * d_instantiate - fill in inode information for a dentry
1346 * @entry: dentry to complete
1347 * @inode: inode to attach to this dentry
1349 * Fill in inode information in the entry.
1351 * This turns negative dentries into productive full members
1354 * NOTE! This assumes that the inode count has been incremented
1355 * (or otherwise set) by the caller to indicate that it is now
1356 * in use by the dcache.
1359 void d_instantiate(struct dentry *entry, struct inode * inode)
1361 BUG_ON(!list_empty(&entry->d_alias));
1363 spin_lock(&inode->i_lock);
1364 __d_instantiate(entry, inode);
1366 spin_unlock(&inode->i_lock);
1367 security_d_instantiate(entry, inode);
1369 EXPORT_SYMBOL(d_instantiate);
1372 * d_instantiate_unique - instantiate a non-aliased dentry
1373 * @entry: dentry to instantiate
1374 * @inode: inode to attach to this dentry
1376 * Fill in inode information in the entry. On success, it returns NULL.
1377 * If an unhashed alias of "entry" already exists, then we return the
1378 * aliased dentry instead and drop one reference to inode.
1380 * Note that in order to avoid conflicts with rename() etc, the caller
1381 * had better be holding the parent directory semaphore.
1383 * This also assumes that the inode count has been incremented
1384 * (or otherwise set) by the caller to indicate that it is now
1385 * in use by the dcache.
1387 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1388 struct inode *inode)
1390 struct dentry *alias;
1391 int len = entry->d_name.len;
1392 const char *name = entry->d_name.name;
1393 unsigned int hash = entry->d_name.hash;
1396 __d_instantiate(entry, NULL);
1400 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1401 struct qstr *qstr = &alias->d_name;
1404 * Don't need alias->d_lock here, because aliases with
1405 * d_parent == entry->d_parent are not subject to name or
1406 * parent changes, because the parent inode i_mutex is held.
1408 if (qstr->hash != hash)
1410 if (alias->d_parent != entry->d_parent)
1412 if (dentry_cmp(qstr->name, qstr->len, name, len))
1418 __d_instantiate(entry, inode);
1422 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1424 struct dentry *result;
1426 BUG_ON(!list_empty(&entry->d_alias));
1429 spin_lock(&inode->i_lock);
1430 result = __d_instantiate_unique(entry, inode);
1432 spin_unlock(&inode->i_lock);
1435 security_d_instantiate(entry, inode);
1439 BUG_ON(!d_unhashed(result));
1444 EXPORT_SYMBOL(d_instantiate_unique);
1446 struct dentry *d_make_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);
1461 EXPORT_SYMBOL(d_make_root);
1463 static struct dentry * __d_find_any_alias(struct inode *inode)
1465 struct dentry *alias;
1467 if (list_empty(&inode->i_dentry))
1469 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1475 * d_find_any_alias - find any alias for a given inode
1476 * @inode: inode to find an alias for
1478 * If any aliases exist for the given inode, take and return a
1479 * reference for one of them. If no aliases exist, return %NULL.
1481 struct dentry *d_find_any_alias(struct inode *inode)
1485 spin_lock(&inode->i_lock);
1486 de = __d_find_any_alias(inode);
1487 spin_unlock(&inode->i_lock);
1490 EXPORT_SYMBOL(d_find_any_alias);
1493 * d_obtain_alias - find or allocate a dentry for a given inode
1494 * @inode: inode to allocate the dentry for
1496 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1497 * similar open by handle operations. The returned dentry may be anonymous,
1498 * or may have a full name (if the inode was already in the cache).
1500 * When called on a directory inode, we must ensure that the inode only ever
1501 * has one dentry. If a dentry is found, that is returned instead of
1502 * allocating a new one.
1504 * On successful return, the reference to the inode has been transferred
1505 * to the dentry. In case of an error the reference on the inode is released.
1506 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1507 * be passed in and will be the error will be propagate to the return value,
1508 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1510 struct dentry *d_obtain_alias(struct inode *inode)
1512 static const struct qstr anonstring = { .name = "" };
1517 return ERR_PTR(-ESTALE);
1519 return ERR_CAST(inode);
1521 res = d_find_any_alias(inode);
1525 tmp = __d_alloc(inode->i_sb, &anonstring);
1527 res = ERR_PTR(-ENOMEM);
1531 spin_lock(&inode->i_lock);
1532 res = __d_find_any_alias(inode);
1534 spin_unlock(&inode->i_lock);
1539 /* attach a disconnected dentry */
1540 spin_lock(&tmp->d_lock);
1541 tmp->d_inode = inode;
1542 tmp->d_flags |= DCACHE_DISCONNECTED;
1543 list_add(&tmp->d_alias, &inode->i_dentry);
1544 hlist_bl_lock(&tmp->d_sb->s_anon);
1545 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1546 hlist_bl_unlock(&tmp->d_sb->s_anon);
1547 spin_unlock(&tmp->d_lock);
1548 spin_unlock(&inode->i_lock);
1549 security_d_instantiate(tmp, inode);
1554 if (res && !IS_ERR(res))
1555 security_d_instantiate(res, inode);
1559 EXPORT_SYMBOL(d_obtain_alias);
1562 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1563 * @inode: the inode which may have a disconnected dentry
1564 * @dentry: a negative dentry which we want to point to the inode.
1566 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1567 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1568 * and return it, else simply d_add the inode to the dentry and return NULL.
1570 * This is needed in the lookup routine of any filesystem that is exportable
1571 * (via knfsd) so that we can build dcache paths to directories effectively.
1573 * If a dentry was found and moved, then it is returned. Otherwise NULL
1574 * is returned. This matches the expected return value of ->lookup.
1577 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1579 struct dentry *new = NULL;
1582 return ERR_CAST(inode);
1584 if (inode && S_ISDIR(inode->i_mode)) {
1585 spin_lock(&inode->i_lock);
1586 new = __d_find_alias(inode, 1);
1588 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1589 spin_unlock(&inode->i_lock);
1590 security_d_instantiate(new, inode);
1591 d_move(new, dentry);
1594 /* already taking inode->i_lock, so d_add() by hand */
1595 __d_instantiate(dentry, inode);
1596 spin_unlock(&inode->i_lock);
1597 security_d_instantiate(dentry, inode);
1601 d_add(dentry, inode);
1604 EXPORT_SYMBOL(d_splice_alias);
1607 * d_add_ci - lookup or allocate new dentry with case-exact name
1608 * @inode: the inode case-insensitive lookup has found
1609 * @dentry: the negative dentry that was passed to the parent's lookup func
1610 * @name: the case-exact name to be associated with the returned dentry
1612 * This is to avoid filling the dcache with case-insensitive names to the
1613 * same inode, only the actual correct case is stored in the dcache for
1614 * case-insensitive filesystems.
1616 * For a case-insensitive lookup match and if the the case-exact dentry
1617 * already exists in in the dcache, use it and return it.
1619 * If no entry exists with the exact case name, allocate new dentry with
1620 * the exact case, and return the spliced entry.
1622 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1626 struct dentry *found;
1630 * First check if a dentry matching the name already exists,
1631 * if not go ahead and create it now.
1633 found = d_hash_and_lookup(dentry->d_parent, name);
1635 new = d_alloc(dentry->d_parent, name);
1641 found = d_splice_alias(inode, new);
1650 * If a matching dentry exists, and it's not negative use it.
1652 * Decrement the reference count to balance the iget() done
1655 if (found->d_inode) {
1656 if (unlikely(found->d_inode != inode)) {
1657 /* This can't happen because bad inodes are unhashed. */
1658 BUG_ON(!is_bad_inode(inode));
1659 BUG_ON(!is_bad_inode(found->d_inode));
1666 * We are going to instantiate this dentry, unhash it and clear the
1667 * lookup flag so we can do that.
1669 if (unlikely(d_need_lookup(found)))
1670 d_clear_need_lookup(found);
1673 * Negative dentry: instantiate it unless the inode is a directory and
1674 * already has a dentry.
1676 new = d_splice_alias(inode, found);
1685 return ERR_PTR(error);
1687 EXPORT_SYMBOL(d_add_ci);
1690 * __d_lookup_rcu - search for a dentry (racy, store-free)
1691 * @parent: parent dentry
1692 * @name: qstr of name we wish to find
1693 * @seq: returns d_seq value at the point where the dentry was found
1694 * @inode: returns dentry->d_inode when the inode was found valid.
1695 * Returns: dentry, or NULL
1697 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1698 * resolution (store-free path walking) design described in
1699 * Documentation/filesystems/path-lookup.txt.
1701 * This is not to be used outside core vfs.
1703 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1704 * held, and rcu_read_lock held. The returned dentry must not be stored into
1705 * without taking d_lock and checking d_seq sequence count against @seq
1708 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1711 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1712 * the returned dentry, so long as its parent's seqlock is checked after the
1713 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1714 * is formed, giving integrity down the path walk.
1716 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1717 const struct qstr *name,
1718 unsigned *seqp, struct inode **inode)
1720 unsigned int len = name->len;
1721 unsigned int hash = name->hash;
1722 const unsigned char *str = name->name;
1723 struct hlist_bl_head *b = d_hash(parent, hash);
1724 struct hlist_bl_node *node;
1725 struct dentry *dentry;
1728 * Note: There is significant duplication with __d_lookup_rcu which is
1729 * required to prevent single threaded performance regressions
1730 * especially on architectures where smp_rmb (in seqcounts) are costly.
1731 * Keep the two functions in sync.
1735 * The hash list is protected using RCU.
1737 * Carefully use d_seq when comparing a candidate dentry, to avoid
1738 * races with d_move().
1740 * It is possible that concurrent renames can mess up our list
1741 * walk here and result in missing our dentry, resulting in the
1742 * false-negative result. d_lookup() protects against concurrent
1743 * renames using rename_lock seqlock.
1745 * See Documentation/filesystems/path-lookup.txt for more details.
1747 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1753 if (dentry->d_name.hash != hash)
1757 seq = read_seqcount_begin(&dentry->d_seq);
1758 if (dentry->d_parent != parent)
1760 if (d_unhashed(dentry))
1762 tlen = dentry->d_name.len;
1763 tname = dentry->d_name.name;
1764 i = dentry->d_inode;
1767 * This seqcount check is required to ensure name and
1768 * len are loaded atomically, so as not to walk off the
1769 * edge of memory when walking. If we could load this
1770 * atomically some other way, we could drop this check.
1772 if (read_seqcount_retry(&dentry->d_seq, seq))
1774 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1775 if (parent->d_op->d_compare(parent, *inode,
1780 if (dentry_cmp(tname, tlen, str, len))
1784 * No extra seqcount check is required after the name
1785 * compare. The caller must perform a seqcount check in
1786 * order to do anything useful with the returned dentry
1797 * d_lookup - search for a dentry
1798 * @parent: parent dentry
1799 * @name: qstr of name we wish to find
1800 * Returns: dentry, or NULL
1802 * d_lookup searches the children of the parent dentry for the name in
1803 * question. If the dentry is found its reference count is incremented and the
1804 * dentry is returned. The caller must use dput to free the entry when it has
1805 * finished using it. %NULL is returned if the dentry does not exist.
1807 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1809 struct dentry *dentry;
1813 seq = read_seqbegin(&rename_lock);
1814 dentry = __d_lookup(parent, name);
1817 } while (read_seqretry(&rename_lock, seq));
1820 EXPORT_SYMBOL(d_lookup);
1823 * __d_lookup - search for a dentry (racy)
1824 * @parent: parent dentry
1825 * @name: qstr of name we wish to find
1826 * Returns: dentry, or NULL
1828 * __d_lookup is like d_lookup, however it may (rarely) return a
1829 * false-negative result due to unrelated rename activity.
1831 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1832 * however it must be used carefully, eg. with a following d_lookup in
1833 * the case of failure.
1835 * __d_lookup callers must be commented.
1837 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1839 unsigned int len = name->len;
1840 unsigned int hash = name->hash;
1841 const unsigned char *str = name->name;
1842 struct hlist_bl_head *b = d_hash(parent, hash);
1843 struct hlist_bl_node *node;
1844 struct dentry *found = NULL;
1845 struct dentry *dentry;
1848 * Note: There is significant duplication with __d_lookup_rcu which is
1849 * required to prevent single threaded performance regressions
1850 * especially on architectures where smp_rmb (in seqcounts) are costly.
1851 * Keep the two functions in sync.
1855 * The hash list is protected using RCU.
1857 * Take d_lock when comparing a candidate dentry, to avoid races
1860 * It is possible that concurrent renames can mess up our list
1861 * walk here and result in missing our dentry, resulting in the
1862 * false-negative result. d_lookup() protects against concurrent
1863 * renames using rename_lock seqlock.
1865 * See Documentation/filesystems/path-lookup.txt for more details.
1869 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1873 if (dentry->d_name.hash != hash)
1876 spin_lock(&dentry->d_lock);
1877 if (dentry->d_parent != parent)
1879 if (d_unhashed(dentry))
1883 * It is safe to compare names since d_move() cannot
1884 * change the qstr (protected by d_lock).
1886 tlen = dentry->d_name.len;
1887 tname = dentry->d_name.name;
1888 if (parent->d_flags & DCACHE_OP_COMPARE) {
1889 if (parent->d_op->d_compare(parent, parent->d_inode,
1890 dentry, dentry->d_inode,
1894 if (dentry_cmp(tname, tlen, str, len))
1900 spin_unlock(&dentry->d_lock);
1903 spin_unlock(&dentry->d_lock);
1911 * d_hash_and_lookup - hash the qstr then search for a dentry
1912 * @dir: Directory to search in
1913 * @name: qstr of name we wish to find
1915 * On hash failure or on lookup failure NULL is returned.
1917 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1919 struct dentry *dentry = NULL;
1922 * Check for a fs-specific hash function. Note that we must
1923 * calculate the standard hash first, as the d_op->d_hash()
1924 * routine may choose to leave the hash value unchanged.
1926 name->hash = full_name_hash(name->name, name->len);
1927 if (dir->d_flags & DCACHE_OP_HASH) {
1928 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1931 dentry = d_lookup(dir, name);
1937 * d_validate - verify dentry provided from insecure source (deprecated)
1938 * @dentry: The dentry alleged to be valid child of @dparent
1939 * @dparent: The parent dentry (known to be valid)
1941 * An insecure source has sent us a dentry, here we verify it and dget() it.
1942 * This is used by ncpfs in its readdir implementation.
1943 * Zero is returned in the dentry is invalid.
1945 * This function is slow for big directories, and deprecated, do not use it.
1947 int d_validate(struct dentry *dentry, struct dentry *dparent)
1949 struct dentry *child;
1951 spin_lock(&dparent->d_lock);
1952 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1953 if (dentry == child) {
1954 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1955 __dget_dlock(dentry);
1956 spin_unlock(&dentry->d_lock);
1957 spin_unlock(&dparent->d_lock);
1961 spin_unlock(&dparent->d_lock);
1965 EXPORT_SYMBOL(d_validate);
1968 * When a file is deleted, we have two options:
1969 * - turn this dentry into a negative dentry
1970 * - unhash this dentry and free it.
1972 * Usually, we want to just turn this into
1973 * a negative dentry, but if anybody else is
1974 * currently using the dentry or the inode
1975 * we can't do that and we fall back on removing
1976 * it from the hash queues and waiting for
1977 * it to be deleted later when it has no users
1981 * d_delete - delete a dentry
1982 * @dentry: The dentry to delete
1984 * Turn the dentry into a negative dentry if possible, otherwise
1985 * remove it from the hash queues so it can be deleted later
1988 void d_delete(struct dentry * dentry)
1990 struct inode *inode;
1993 * Are we the only user?
1996 spin_lock(&dentry->d_lock);
1997 inode = dentry->d_inode;
1998 isdir = S_ISDIR(inode->i_mode);
1999 if (dentry->d_count == 1) {
2000 if (inode && !spin_trylock(&inode->i_lock)) {
2001 spin_unlock(&dentry->d_lock);
2005 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2006 dentry_unlink_inode(dentry);
2007 fsnotify_nameremove(dentry, isdir);
2011 if (!d_unhashed(dentry))
2014 spin_unlock(&dentry->d_lock);
2016 fsnotify_nameremove(dentry, isdir);
2018 EXPORT_SYMBOL(d_delete);
2020 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2022 BUG_ON(!d_unhashed(entry));
2024 entry->d_flags |= DCACHE_RCUACCESS;
2025 hlist_bl_add_head_rcu(&entry->d_hash, b);
2029 static void _d_rehash(struct dentry * entry)
2031 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2035 * d_rehash - add an entry back to the hash
2036 * @entry: dentry to add to the hash
2038 * Adds a dentry to the hash according to its name.
2041 void d_rehash(struct dentry * entry)
2043 spin_lock(&entry->d_lock);
2045 spin_unlock(&entry->d_lock);
2047 EXPORT_SYMBOL(d_rehash);
2050 * dentry_update_name_case - update case insensitive dentry with a new name
2051 * @dentry: dentry to be updated
2054 * Update a case insensitive dentry with new case of name.
2056 * dentry must have been returned by d_lookup with name @name. Old and new
2057 * name lengths must match (ie. no d_compare which allows mismatched name
2060 * Parent inode i_mutex must be held over d_lookup and into this call (to
2061 * keep renames and concurrent inserts, and readdir(2) away).
2063 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2065 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2066 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2068 spin_lock(&dentry->d_lock);
2069 write_seqcount_begin(&dentry->d_seq);
2070 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2071 write_seqcount_end(&dentry->d_seq);
2072 spin_unlock(&dentry->d_lock);
2074 EXPORT_SYMBOL(dentry_update_name_case);
2076 static void switch_names(struct dentry *dentry, struct dentry *target)
2078 if (dname_external(target)) {
2079 if (dname_external(dentry)) {
2081 * Both external: swap the pointers
2083 swap(target->d_name.name, dentry->d_name.name);
2086 * dentry:internal, target:external. Steal target's
2087 * storage and make target internal.
2089 memcpy(target->d_iname, dentry->d_name.name,
2090 dentry->d_name.len + 1);
2091 dentry->d_name.name = target->d_name.name;
2092 target->d_name.name = target->d_iname;
2095 if (dname_external(dentry)) {
2097 * dentry:external, target:internal. Give dentry's
2098 * storage to target and make dentry internal
2100 memcpy(dentry->d_iname, target->d_name.name,
2101 target->d_name.len + 1);
2102 target->d_name.name = dentry->d_name.name;
2103 dentry->d_name.name = dentry->d_iname;
2106 * Both are internal. Just copy target to dentry
2108 memcpy(dentry->d_iname, target->d_name.name,
2109 target->d_name.len + 1);
2110 dentry->d_name.len = target->d_name.len;
2114 swap(dentry->d_name.len, target->d_name.len);
2117 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2120 * XXXX: do we really need to take target->d_lock?
2122 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2123 spin_lock(&target->d_parent->d_lock);
2125 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2126 spin_lock(&dentry->d_parent->d_lock);
2127 spin_lock_nested(&target->d_parent->d_lock,
2128 DENTRY_D_LOCK_NESTED);
2130 spin_lock(&target->d_parent->d_lock);
2131 spin_lock_nested(&dentry->d_parent->d_lock,
2132 DENTRY_D_LOCK_NESTED);
2135 if (target < dentry) {
2136 spin_lock_nested(&target->d_lock, 2);
2137 spin_lock_nested(&dentry->d_lock, 3);
2139 spin_lock_nested(&dentry->d_lock, 2);
2140 spin_lock_nested(&target->d_lock, 3);
2144 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2145 struct dentry *target)
2147 if (target->d_parent != dentry->d_parent)
2148 spin_unlock(&dentry->d_parent->d_lock);
2149 if (target->d_parent != target)
2150 spin_unlock(&target->d_parent->d_lock);
2154 * When switching names, the actual string doesn't strictly have to
2155 * be preserved in the target - because we're dropping the target
2156 * anyway. As such, we can just do a simple memcpy() to copy over
2157 * the new name before we switch.
2159 * Note that we have to be a lot more careful about getting the hash
2160 * switched - we have to switch the hash value properly even if it
2161 * then no longer matches the actual (corrupted) string of the target.
2162 * The hash value has to match the hash queue that the dentry is on..
2165 * __d_move - move a dentry
2166 * @dentry: entry to move
2167 * @target: new dentry
2169 * Update the dcache to reflect the move of a file name. Negative
2170 * dcache entries should not be moved in this way. Caller must hold
2171 * rename_lock, the i_mutex of the source and target directories,
2172 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2174 static void __d_move(struct dentry * dentry, struct dentry * target)
2176 if (!dentry->d_inode)
2177 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2179 BUG_ON(d_ancestor(dentry, target));
2180 BUG_ON(d_ancestor(target, dentry));
2182 dentry_lock_for_move(dentry, target);
2184 write_seqcount_begin(&dentry->d_seq);
2185 write_seqcount_begin(&target->d_seq);
2187 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2190 * Move the dentry to the target hash queue. Don't bother checking
2191 * for the same hash queue because of how unlikely it is.
2194 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2196 /* Unhash the target: dput() will then get rid of it */
2199 list_del(&dentry->d_u.d_child);
2200 list_del(&target->d_u.d_child);
2202 /* Switch the names.. */
2203 switch_names(dentry, target);
2204 swap(dentry->d_name.hash, target->d_name.hash);
2206 /* ... and switch the parents */
2207 if (IS_ROOT(dentry)) {
2208 dentry->d_parent = target->d_parent;
2209 target->d_parent = target;
2210 INIT_LIST_HEAD(&target->d_u.d_child);
2212 swap(dentry->d_parent, target->d_parent);
2214 /* And add them back to the (new) parent lists */
2215 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2218 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2220 write_seqcount_end(&target->d_seq);
2221 write_seqcount_end(&dentry->d_seq);
2223 dentry_unlock_parents_for_move(dentry, target);
2224 spin_unlock(&target->d_lock);
2225 fsnotify_d_move(dentry);
2226 spin_unlock(&dentry->d_lock);
2230 * d_move - move a dentry
2231 * @dentry: entry to move
2232 * @target: new dentry
2234 * Update the dcache to reflect the move of a file name. Negative
2235 * dcache entries should not be moved in this way. See the locking
2236 * requirements for __d_move.
2238 void d_move(struct dentry *dentry, struct dentry *target)
2240 write_seqlock(&rename_lock);
2241 __d_move(dentry, target);
2242 write_sequnlock(&rename_lock);
2244 EXPORT_SYMBOL(d_move);
2247 * d_ancestor - search for an ancestor
2248 * @p1: ancestor dentry
2251 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2252 * an ancestor of p2, else NULL.
2254 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2258 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2259 if (p->d_parent == p1)
2266 * This helper attempts to cope with remotely renamed directories
2268 * It assumes that the caller is already holding
2269 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2271 * Note: If ever the locking in lock_rename() changes, then please
2272 * remember to update this too...
2274 static struct dentry *__d_unalias(struct inode *inode,
2275 struct dentry *dentry, struct dentry *alias)
2277 struct mutex *m1 = NULL, *m2 = NULL;
2280 /* If alias and dentry share a parent, then no extra locks required */
2281 if (alias->d_parent == dentry->d_parent)
2284 /* See lock_rename() */
2285 ret = ERR_PTR(-EBUSY);
2286 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2288 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2289 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2291 m2 = &alias->d_parent->d_inode->i_mutex;
2293 __d_move(alias, dentry);
2296 spin_unlock(&inode->i_lock);
2305 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2306 * named dentry in place of the dentry to be replaced.
2307 * returns with anon->d_lock held!
2309 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2311 struct dentry *dparent, *aparent;
2313 dentry_lock_for_move(anon, dentry);
2315 write_seqcount_begin(&dentry->d_seq);
2316 write_seqcount_begin(&anon->d_seq);
2318 dparent = dentry->d_parent;
2319 aparent = anon->d_parent;
2321 switch_names(dentry, anon);
2322 swap(dentry->d_name.hash, anon->d_name.hash);
2324 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2325 list_del(&dentry->d_u.d_child);
2326 if (!IS_ROOT(dentry))
2327 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2329 INIT_LIST_HEAD(&dentry->d_u.d_child);
2331 anon->d_parent = (dparent == dentry) ? anon : dparent;
2332 list_del(&anon->d_u.d_child);
2334 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2336 INIT_LIST_HEAD(&anon->d_u.d_child);
2338 write_seqcount_end(&dentry->d_seq);
2339 write_seqcount_end(&anon->d_seq);
2341 dentry_unlock_parents_for_move(anon, dentry);
2342 spin_unlock(&dentry->d_lock);
2344 /* anon->d_lock still locked, returns locked */
2345 anon->d_flags &= ~DCACHE_DISCONNECTED;
2349 * d_materialise_unique - introduce an inode into the tree
2350 * @dentry: candidate dentry
2351 * @inode: inode to bind to the dentry, to which aliases may be attached
2353 * Introduces an dentry into the tree, substituting an extant disconnected
2354 * root directory alias in its place if there is one. Caller must hold the
2355 * i_mutex of the parent directory.
2357 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2359 struct dentry *actual;
2361 BUG_ON(!d_unhashed(dentry));
2365 __d_instantiate(dentry, NULL);
2370 spin_lock(&inode->i_lock);
2372 if (S_ISDIR(inode->i_mode)) {
2373 struct dentry *alias;
2375 /* Does an aliased dentry already exist? */
2376 alias = __d_find_alias(inode, 0);
2379 write_seqlock(&rename_lock);
2381 if (d_ancestor(alias, dentry)) {
2382 /* Check for loops */
2383 actual = ERR_PTR(-ELOOP);
2384 } else if (IS_ROOT(alias)) {
2385 /* Is this an anonymous mountpoint that we
2386 * could splice into our tree? */
2387 __d_materialise_dentry(dentry, alias);
2388 write_sequnlock(&rename_lock);
2392 /* Nope, but we must(!) avoid directory
2394 actual = __d_unalias(inode, dentry, alias);
2396 write_sequnlock(&rename_lock);
2397 if (IS_ERR(actual)) {
2398 if (PTR_ERR(actual) == -ELOOP)
2399 pr_warn_ratelimited(
2400 "VFS: Lookup of '%s' in %s %s"
2401 " would have caused loop\n",
2402 dentry->d_name.name,
2403 inode->i_sb->s_type->name,
2411 /* Add a unique reference */
2412 actual = __d_instantiate_unique(dentry, inode);
2416 BUG_ON(!d_unhashed(actual));
2418 spin_lock(&actual->d_lock);
2421 spin_unlock(&actual->d_lock);
2422 spin_unlock(&inode->i_lock);
2424 if (actual == dentry) {
2425 security_d_instantiate(dentry, inode);
2432 EXPORT_SYMBOL_GPL(d_materialise_unique);
2434 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2438 return -ENAMETOOLONG;
2440 memcpy(*buffer, str, namelen);
2444 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2446 return prepend(buffer, buflen, name->name, name->len);
2450 * prepend_path - Prepend path string to a buffer
2451 * @path: the dentry/vfsmount to report
2452 * @root: root vfsmnt/dentry
2453 * @buffer: pointer to the end of the buffer
2454 * @buflen: pointer to buffer length
2456 * Caller holds the rename_lock.
2458 static int prepend_path(const struct path *path,
2459 const struct path *root,
2460 char **buffer, int *buflen)
2462 struct dentry *dentry = path->dentry;
2463 struct vfsmount *vfsmnt = path->mnt;
2464 struct mount *mnt = real_mount(vfsmnt);
2468 br_read_lock(vfsmount_lock);
2469 while (dentry != root->dentry || vfsmnt != root->mnt) {
2470 struct dentry * parent;
2472 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2474 if (!mnt_has_parent(mnt))
2476 dentry = mnt->mnt_mountpoint;
2477 mnt = mnt->mnt_parent;
2481 parent = dentry->d_parent;
2483 spin_lock(&dentry->d_lock);
2484 error = prepend_name(buffer, buflen, &dentry->d_name);
2485 spin_unlock(&dentry->d_lock);
2487 error = prepend(buffer, buflen, "/", 1);
2495 if (!error && !slash)
2496 error = prepend(buffer, buflen, "/", 1);
2499 br_read_unlock(vfsmount_lock);
2504 * Filesystems needing to implement special "root names"
2505 * should do so with ->d_dname()
2507 if (IS_ROOT(dentry) &&
2508 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2509 WARN(1, "Root dentry has weird name <%.*s>\n",
2510 (int) dentry->d_name.len, dentry->d_name.name);
2513 error = prepend(buffer, buflen, "/", 1);
2515 error = real_mount(vfsmnt)->mnt_ns ? 1 : 2;
2520 * __d_path - return the path of a dentry
2521 * @path: the dentry/vfsmount to report
2522 * @root: root vfsmnt/dentry
2523 * @buf: buffer to return value in
2524 * @buflen: buffer length
2526 * Convert a dentry into an ASCII path name.
2528 * Returns a pointer into the buffer or an error code if the
2529 * path was too long.
2531 * "buflen" should be positive.
2533 * If the path is not reachable from the supplied root, return %NULL.
2535 char *__d_path(const struct path *path,
2536 const struct path *root,
2537 char *buf, int buflen)
2539 char *res = buf + buflen;
2542 prepend(&res, &buflen, "\0", 1);
2543 write_seqlock(&rename_lock);
2544 error = prepend_path(path, root, &res, &buflen);
2545 write_sequnlock(&rename_lock);
2548 return ERR_PTR(error);
2554 char *d_absolute_path(const struct path *path,
2555 char *buf, int buflen)
2557 struct path root = {};
2558 char *res = buf + buflen;
2561 prepend(&res, &buflen, "\0", 1);
2562 write_seqlock(&rename_lock);
2563 error = prepend_path(path, &root, &res, &buflen);
2564 write_sequnlock(&rename_lock);
2569 return ERR_PTR(error);
2574 * same as __d_path but appends "(deleted)" for unlinked files.
2576 static int path_with_deleted(const struct path *path,
2577 const struct path *root,
2578 char **buf, int *buflen)
2580 prepend(buf, buflen, "\0", 1);
2581 if (d_unlinked(path->dentry)) {
2582 int error = prepend(buf, buflen, " (deleted)", 10);
2587 return prepend_path(path, root, buf, buflen);
2590 static int prepend_unreachable(char **buffer, int *buflen)
2592 return prepend(buffer, buflen, "(unreachable)", 13);
2596 * d_path - return the path of a dentry
2597 * @path: path to report
2598 * @buf: buffer to return value in
2599 * @buflen: buffer length
2601 * Convert a dentry into an ASCII path name. If the entry has been deleted
2602 * the string " (deleted)" is appended. Note that this is ambiguous.
2604 * Returns a pointer into the buffer or an error code if the path was
2605 * too long. Note: Callers should use the returned pointer, not the passed
2606 * in buffer, to use the name! The implementation often starts at an offset
2607 * into the buffer, and may leave 0 bytes at the start.
2609 * "buflen" should be positive.
2611 char *d_path(const struct path *path, char *buf, int buflen)
2613 char *res = buf + buflen;
2618 * We have various synthetic filesystems that never get mounted. On
2619 * these filesystems dentries are never used for lookup purposes, and
2620 * thus don't need to be hashed. They also don't need a name until a
2621 * user wants to identify the object in /proc/pid/fd/. The little hack
2622 * below allows us to generate a name for these objects on demand:
2624 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2625 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2627 get_fs_root(current->fs, &root);
2628 write_seqlock(&rename_lock);
2629 error = path_with_deleted(path, &root, &res, &buflen);
2631 res = ERR_PTR(error);
2632 write_sequnlock(&rename_lock);
2636 EXPORT_SYMBOL(d_path);
2639 * d_path_with_unreachable - return the path of a dentry
2640 * @path: path to report
2641 * @buf: buffer to return value in
2642 * @buflen: buffer length
2644 * The difference from d_path() is that this prepends "(unreachable)"
2645 * to paths which are unreachable from the current process' root.
2647 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2649 char *res = buf + buflen;
2653 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2654 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2656 get_fs_root(current->fs, &root);
2657 write_seqlock(&rename_lock);
2658 error = path_with_deleted(path, &root, &res, &buflen);
2660 error = prepend_unreachable(&res, &buflen);
2661 write_sequnlock(&rename_lock);
2664 res = ERR_PTR(error);
2670 * Helper function for dentry_operations.d_dname() members
2672 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2673 const char *fmt, ...)
2679 va_start(args, fmt);
2680 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2683 if (sz > sizeof(temp) || sz > buflen)
2684 return ERR_PTR(-ENAMETOOLONG);
2686 buffer += buflen - sz;
2687 return memcpy(buffer, temp, sz);
2691 * Write full pathname from the root of the filesystem into the buffer.
2693 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2695 char *end = buf + buflen;
2698 prepend(&end, &buflen, "\0", 1);
2705 while (!IS_ROOT(dentry)) {
2706 struct dentry *parent = dentry->d_parent;
2710 spin_lock(&dentry->d_lock);
2711 error = prepend_name(&end, &buflen, &dentry->d_name);
2712 spin_unlock(&dentry->d_lock);
2713 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2721 return ERR_PTR(-ENAMETOOLONG);
2724 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2728 write_seqlock(&rename_lock);
2729 retval = __dentry_path(dentry, buf, buflen);
2730 write_sequnlock(&rename_lock);
2734 EXPORT_SYMBOL(dentry_path_raw);
2736 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2741 write_seqlock(&rename_lock);
2742 if (d_unlinked(dentry)) {
2744 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2748 retval = __dentry_path(dentry, buf, buflen);
2749 write_sequnlock(&rename_lock);
2750 if (!IS_ERR(retval) && p)
2751 *p = '/'; /* restore '/' overriden with '\0' */
2754 return ERR_PTR(-ENAMETOOLONG);
2758 * NOTE! The user-level library version returns a
2759 * character pointer. The kernel system call just
2760 * returns the length of the buffer filled (which
2761 * includes the ending '\0' character), or a negative
2762 * error value. So libc would do something like
2764 * char *getcwd(char * buf, size_t size)
2768 * retval = sys_getcwd(buf, size);
2775 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2778 struct path pwd, root;
2779 char *page = (char *) __get_free_page(GFP_USER);
2784 get_fs_root_and_pwd(current->fs, &root, &pwd);
2787 write_seqlock(&rename_lock);
2788 if (!d_unlinked(pwd.dentry)) {
2790 char *cwd = page + PAGE_SIZE;
2791 int buflen = PAGE_SIZE;
2793 prepend(&cwd, &buflen, "\0", 1);
2794 error = prepend_path(&pwd, &root, &cwd, &buflen);
2795 write_sequnlock(&rename_lock);
2800 /* Unreachable from current root */
2802 error = prepend_unreachable(&cwd, &buflen);
2808 len = PAGE_SIZE + page - cwd;
2811 if (copy_to_user(buf, cwd, len))
2815 write_sequnlock(&rename_lock);
2821 free_page((unsigned long) page);
2826 * Test whether new_dentry is a subdirectory of old_dentry.
2828 * Trivially implemented using the dcache structure
2832 * is_subdir - is new dentry a subdirectory of old_dentry
2833 * @new_dentry: new dentry
2834 * @old_dentry: old dentry
2836 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2837 * Returns 0 otherwise.
2838 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2841 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2846 if (new_dentry == old_dentry)
2850 /* for restarting inner loop in case of seq retry */
2851 seq = read_seqbegin(&rename_lock);
2853 * Need rcu_readlock to protect against the d_parent trashing
2857 if (d_ancestor(old_dentry, new_dentry))
2862 } while (read_seqretry(&rename_lock, seq));
2867 void d_genocide(struct dentry *root)
2869 struct dentry *this_parent;
2870 struct list_head *next;
2874 seq = read_seqbegin(&rename_lock);
2877 spin_lock(&this_parent->d_lock);
2879 next = this_parent->d_subdirs.next;
2881 while (next != &this_parent->d_subdirs) {
2882 struct list_head *tmp = next;
2883 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2886 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2887 if (d_unhashed(dentry) || !dentry->d_inode) {
2888 spin_unlock(&dentry->d_lock);
2891 if (!list_empty(&dentry->d_subdirs)) {
2892 spin_unlock(&this_parent->d_lock);
2893 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2894 this_parent = dentry;
2895 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2898 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2899 dentry->d_flags |= DCACHE_GENOCIDE;
2902 spin_unlock(&dentry->d_lock);
2904 if (this_parent != root) {
2905 struct dentry *child = this_parent;
2906 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2907 this_parent->d_flags |= DCACHE_GENOCIDE;
2908 this_parent->d_count--;
2910 this_parent = try_to_ascend(this_parent, locked, seq);
2913 next = child->d_u.d_child.next;
2916 spin_unlock(&this_parent->d_lock);
2917 if (!locked && read_seqretry(&rename_lock, seq))
2920 write_sequnlock(&rename_lock);
2925 write_seqlock(&rename_lock);
2930 * find_inode_number - check for dentry with name
2931 * @dir: directory to check
2932 * @name: Name to find.
2934 * Check whether a dentry already exists for the given name,
2935 * and return the inode number if it has an inode. Otherwise
2938 * This routine is used to post-process directory listings for
2939 * filesystems using synthetic inode numbers, and is necessary
2940 * to keep getcwd() working.
2943 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2945 struct dentry * dentry;
2948 dentry = d_hash_and_lookup(dir, name);
2950 if (dentry->d_inode)
2951 ino = dentry->d_inode->i_ino;
2956 EXPORT_SYMBOL(find_inode_number);
2958 static __initdata unsigned long dhash_entries;
2959 static int __init set_dhash_entries(char *str)
2963 dhash_entries = simple_strtoul(str, &str, 0);
2966 __setup("dhash_entries=", set_dhash_entries);
2968 static void __init dcache_init_early(void)
2972 /* If hashes are distributed across NUMA nodes, defer
2973 * hash allocation until vmalloc space is available.
2979 alloc_large_system_hash("Dentry cache",
2980 sizeof(struct hlist_bl_head),
2988 for (loop = 0; loop < (1U << d_hash_shift); loop++)
2989 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
2992 static void __init dcache_init(void)
2997 * A constructor could be added for stable state like the lists,
2998 * but it is probably not worth it because of the cache nature
3001 dentry_cache = KMEM_CACHE(dentry,
3002 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3004 /* Hash may have been set up in dcache_init_early */
3009 alloc_large_system_hash("Dentry cache",
3010 sizeof(struct hlist_bl_head),
3018 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3019 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3022 /* SLAB cache for __getname() consumers */
3023 struct kmem_cache *names_cachep __read_mostly;
3024 EXPORT_SYMBOL(names_cachep);
3026 EXPORT_SYMBOL(d_genocide);
3028 void __init vfs_caches_init_early(void)
3030 dcache_init_early();
3034 void __init vfs_caches_init(unsigned long mempages)
3036 unsigned long reserve;
3038 /* Base hash sizes on available memory, with a reserve equal to
3039 150% of current kernel size */
3041 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3042 mempages -= reserve;
3044 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3045 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3049 files_init(mempages);