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>
43 * dcache->d_inode->i_lock protects:
44 * - i_dentry, d_alias, d_inode of aliases
45 * dcache_hash_bucket lock protects:
46 * - the dcache hash table
47 * s_anon bl list spinlock protects:
48 * - the s_anon list (see __d_drop)
49 * dcache_lru_lock protects:
50 * - the dcache lru lists and counters
57 * - d_parent and d_subdirs
58 * - childrens' d_child and d_parent
62 * dentry->d_inode->i_lock
65 * dcache_hash_bucket lock
68 * If there is an ancestor relationship:
69 * dentry->d_parent->...->d_parent->d_lock
71 * dentry->d_parent->d_lock
74 * If no ancestor relationship:
75 * if (dentry1 < dentry2)
79 int sysctl_vfs_cache_pressure __read_mostly = 100;
80 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
82 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
83 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
85 EXPORT_SYMBOL(rename_lock);
87 static struct kmem_cache *dentry_cache __read_mostly;
90 * This is the single most critical data structure when it comes
91 * to the dcache: the hashtable for lookups. Somebody should try
92 * to make this good - I've just made it work.
94 * This hash-function tries to avoid losing too many bits of hash
95 * information, yet avoid using a prime hash-size or similar.
97 #define D_HASHBITS d_hash_shift
98 #define D_HASHMASK d_hash_mask
100 static unsigned int d_hash_mask __read_mostly;
101 static unsigned int d_hash_shift __read_mostly;
103 static struct hlist_bl_head *dentry_hashtable __read_mostly;
105 static inline struct hlist_bl_head *d_hash(struct dentry *parent,
108 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
109 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
110 return dentry_hashtable + (hash & D_HASHMASK);
113 /* Statistics gathering. */
114 struct dentry_stat_t dentry_stat = {
118 static DEFINE_PER_CPU(unsigned int, nr_dentry);
120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
121 static int get_nr_dentry(void)
125 for_each_possible_cpu(i)
126 sum += per_cpu(nr_dentry, i);
127 return sum < 0 ? 0 : sum;
130 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
131 size_t *lenp, loff_t *ppos)
133 dentry_stat.nr_dentry = get_nr_dentry();
134 return proc_dointvec(table, write, buffer, lenp, ppos);
138 static void __d_free(struct rcu_head *head)
140 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
142 WARN_ON(!list_empty(&dentry->d_alias));
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 BUG_ON(dentry->d_count);
154 this_cpu_dec(nr_dentry);
155 if (dentry->d_op && dentry->d_op->d_release)
156 dentry->d_op->d_release(dentry);
158 /* if dentry was never visible to RCU, immediate free is OK */
159 if (!(dentry->d_flags & DCACHE_RCUACCESS))
160 __d_free(&dentry->d_u.d_rcu);
162 call_rcu(&dentry->d_u.d_rcu, __d_free);
166 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
167 * @dentry: the target dentry
168 * After this call, in-progress rcu-walk path lookup will fail. This
169 * should be called after unhashing, and after changing d_inode (if
170 * the dentry has not already been unhashed).
172 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
174 assert_spin_locked(&dentry->d_lock);
175 /* Go through a barrier */
176 write_seqcount_barrier(&dentry->d_seq);
180 * Release the dentry's inode, using the filesystem
181 * d_iput() operation if defined. Dentry has no refcount
184 static void dentry_iput(struct dentry * dentry)
185 __releases(dentry->d_lock)
186 __releases(dentry->d_inode->i_lock)
188 struct inode *inode = dentry->d_inode;
190 dentry->d_inode = NULL;
191 list_del_init(&dentry->d_alias);
192 spin_unlock(&dentry->d_lock);
193 spin_unlock(&inode->i_lock);
195 fsnotify_inoderemove(inode);
196 if (dentry->d_op && dentry->d_op->d_iput)
197 dentry->d_op->d_iput(dentry, inode);
201 spin_unlock(&dentry->d_lock);
206 * Release the dentry's inode, using the filesystem
207 * d_iput() operation if defined. dentry remains in-use.
209 static void dentry_unlink_inode(struct dentry * dentry)
210 __releases(dentry->d_lock)
211 __releases(dentry->d_inode->i_lock)
213 struct inode *inode = dentry->d_inode;
214 dentry->d_inode = NULL;
215 list_del_init(&dentry->d_alias);
216 dentry_rcuwalk_barrier(dentry);
217 spin_unlock(&dentry->d_lock);
218 spin_unlock(&inode->i_lock);
220 fsnotify_inoderemove(inode);
221 if (dentry->d_op && dentry->d_op->d_iput)
222 dentry->d_op->d_iput(dentry, inode);
228 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
230 static void dentry_lru_add(struct dentry *dentry)
232 if (list_empty(&dentry->d_lru)) {
233 spin_lock(&dcache_lru_lock);
234 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
235 dentry->d_sb->s_nr_dentry_unused++;
236 dentry_stat.nr_unused++;
237 spin_unlock(&dcache_lru_lock);
241 static void __dentry_lru_del(struct dentry *dentry)
243 list_del_init(&dentry->d_lru);
244 dentry->d_sb->s_nr_dentry_unused--;
245 dentry_stat.nr_unused--;
248 static void dentry_lru_del(struct dentry *dentry)
250 if (!list_empty(&dentry->d_lru)) {
251 spin_lock(&dcache_lru_lock);
252 __dentry_lru_del(dentry);
253 spin_unlock(&dcache_lru_lock);
257 static void dentry_lru_move_tail(struct dentry *dentry)
259 spin_lock(&dcache_lru_lock);
260 if (list_empty(&dentry->d_lru)) {
261 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
262 dentry->d_sb->s_nr_dentry_unused++;
263 dentry_stat.nr_unused++;
265 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
267 spin_unlock(&dcache_lru_lock);
271 * d_kill - kill dentry and return parent
272 * @dentry: dentry to kill
273 * @parent: parent dentry
275 * The dentry must already be unhashed and removed from the LRU.
277 * If this is the root of the dentry tree, return NULL.
279 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
282 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
283 __releases(dentry->d_lock)
284 __releases(parent->d_lock)
285 __releases(dentry->d_inode->i_lock)
287 list_del(&dentry->d_u.d_child);
289 * Inform try_to_ascend() that we are no longer attached to the
292 dentry->d_flags |= DCACHE_DISCONNECTED;
294 spin_unlock(&parent->d_lock);
297 * dentry_iput drops the locks, at which point nobody (except
298 * transient RCU lookups) can reach this dentry.
305 * d_drop - drop a dentry
306 * @dentry: dentry to drop
308 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
309 * be found through a VFS lookup any more. Note that this is different from
310 * deleting the dentry - d_delete will try to mark the dentry negative if
311 * possible, giving a successful _negative_ lookup, while d_drop will
312 * just make the cache lookup fail.
314 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
315 * reason (NFS timeouts or autofs deletes).
317 * __d_drop requires dentry->d_lock.
319 void __d_drop(struct dentry *dentry)
321 if (!d_unhashed(dentry)) {
322 struct hlist_bl_head *b;
323 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
324 b = &dentry->d_sb->s_anon;
326 b = d_hash(dentry->d_parent, dentry->d_name.hash);
329 __hlist_bl_del(&dentry->d_hash);
330 dentry->d_hash.pprev = NULL;
333 dentry_rcuwalk_barrier(dentry);
336 EXPORT_SYMBOL(__d_drop);
338 void d_drop(struct dentry *dentry)
340 spin_lock(&dentry->d_lock);
342 spin_unlock(&dentry->d_lock);
344 EXPORT_SYMBOL(d_drop);
347 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
348 * @dentry: dentry to drop
350 * This is called when we do a lookup on a placeholder dentry that needed to be
351 * looked up. The dentry should have been hashed in order for it to be found by
352 * the lookup code, but now needs to be unhashed while we do the actual lookup
353 * and clear the DCACHE_NEED_LOOKUP flag.
355 void d_clear_need_lookup(struct dentry *dentry)
357 spin_lock(&dentry->d_lock);
359 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
360 spin_unlock(&dentry->d_lock);
362 EXPORT_SYMBOL(d_clear_need_lookup);
365 * Finish off a dentry we've decided to kill.
366 * dentry->d_lock must be held, returns with it unlocked.
367 * If ref is non-zero, then decrement the refcount too.
368 * Returns dentry requiring refcount drop, or NULL if we're done.
370 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
371 __releases(dentry->d_lock)
374 struct dentry *parent;
376 inode = dentry->d_inode;
377 if (inode && !spin_trylock(&inode->i_lock)) {
379 spin_unlock(&dentry->d_lock);
381 return dentry; /* try again with same dentry */
386 parent = dentry->d_parent;
387 if (parent && !spin_trylock(&parent->d_lock)) {
389 spin_unlock(&inode->i_lock);
395 /* if dentry was on the d_lru list delete it from there */
396 dentry_lru_del(dentry);
397 /* if it was on the hash then remove it */
399 return d_kill(dentry, parent);
405 * This is complicated by the fact that we do not want to put
406 * dentries that are no longer on any hash chain on the unused
407 * list: we'd much rather just get rid of them immediately.
409 * However, that implies that we have to traverse the dentry
410 * tree upwards to the parents which might _also_ now be
411 * scheduled for deletion (it may have been only waiting for
412 * its last child to go away).
414 * This tail recursion is done by hand as we don't want to depend
415 * on the compiler to always get this right (gcc generally doesn't).
416 * Real recursion would eat up our stack space.
420 * dput - release a dentry
421 * @dentry: dentry to release
423 * Release a dentry. This will drop the usage count and if appropriate
424 * call the dentry unlink method as well as removing it from the queues and
425 * releasing its resources. If the parent dentries were scheduled for release
426 * they too may now get deleted.
428 void dput(struct dentry *dentry)
434 if (dentry->d_count == 1)
436 spin_lock(&dentry->d_lock);
437 BUG_ON(!dentry->d_count);
438 if (dentry->d_count > 1) {
440 spin_unlock(&dentry->d_lock);
444 if (dentry->d_flags & DCACHE_OP_DELETE) {
445 if (dentry->d_op->d_delete(dentry))
449 /* Unreachable? Get rid of it */
450 if (d_unhashed(dentry))
454 * If this dentry needs lookup, don't set the referenced flag so that it
455 * is more likely to be cleaned up by the dcache shrinker in case of
458 if (!d_need_lookup(dentry))
459 dentry->d_flags |= DCACHE_REFERENCED;
460 dentry_lru_add(dentry);
463 spin_unlock(&dentry->d_lock);
467 dentry = dentry_kill(dentry, 1);
474 * d_invalidate - invalidate a dentry
475 * @dentry: dentry to invalidate
477 * Try to invalidate the dentry if it turns out to be
478 * possible. If there are other dentries that can be
479 * reached through this one we can't delete it and we
480 * return -EBUSY. On success we return 0.
485 int d_invalidate(struct dentry * dentry)
488 * If it's already been dropped, return OK.
490 spin_lock(&dentry->d_lock);
491 if (d_unhashed(dentry)) {
492 spin_unlock(&dentry->d_lock);
496 * Check whether to do a partial shrink_dcache
497 * to get rid of unused child entries.
499 if (!list_empty(&dentry->d_subdirs)) {
500 spin_unlock(&dentry->d_lock);
501 shrink_dcache_parent(dentry);
502 spin_lock(&dentry->d_lock);
506 * Somebody else still using it?
508 * If it's a directory, we can't drop it
509 * for fear of somebody re-populating it
510 * with children (even though dropping it
511 * would make it unreachable from the root,
512 * we might still populate it if it was a
513 * working directory or similar).
515 if (dentry->d_count > 1) {
516 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
517 spin_unlock(&dentry->d_lock);
523 spin_unlock(&dentry->d_lock);
526 EXPORT_SYMBOL(d_invalidate);
528 /* This must be called with d_lock held */
529 static inline void __dget_dlock(struct dentry *dentry)
534 static inline void __dget(struct dentry *dentry)
536 spin_lock(&dentry->d_lock);
537 __dget_dlock(dentry);
538 spin_unlock(&dentry->d_lock);
541 struct dentry *dget_parent(struct dentry *dentry)
547 * Don't need rcu_dereference because we re-check it was correct under
551 ret = dentry->d_parent;
556 spin_lock(&ret->d_lock);
557 if (unlikely(ret != dentry->d_parent)) {
558 spin_unlock(&ret->d_lock);
563 BUG_ON(!ret->d_count);
565 spin_unlock(&ret->d_lock);
569 EXPORT_SYMBOL(dget_parent);
572 * d_find_alias - grab a hashed alias of inode
573 * @inode: inode in question
574 * @want_discon: flag, used by d_splice_alias, to request
575 * that only a DISCONNECTED alias be returned.
577 * If inode has a hashed alias, or is a directory and has any alias,
578 * acquire the reference to alias and return it. Otherwise return NULL.
579 * Notice that if inode is a directory there can be only one alias and
580 * it can be unhashed only if it has no children, or if it is the root
583 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
584 * any other hashed alias over that one unless @want_discon is set,
585 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
587 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
589 struct dentry *alias, *discon_alias;
593 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
594 spin_lock(&alias->d_lock);
595 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
596 if (IS_ROOT(alias) &&
597 (alias->d_flags & DCACHE_DISCONNECTED)) {
598 discon_alias = alias;
599 } else if (!want_discon) {
601 spin_unlock(&alias->d_lock);
605 spin_unlock(&alias->d_lock);
608 alias = discon_alias;
609 spin_lock(&alias->d_lock);
610 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
611 if (IS_ROOT(alias) &&
612 (alias->d_flags & DCACHE_DISCONNECTED)) {
614 spin_unlock(&alias->d_lock);
618 spin_unlock(&alias->d_lock);
624 struct dentry *d_find_alias(struct inode *inode)
626 struct dentry *de = NULL;
628 if (!list_empty(&inode->i_dentry)) {
629 spin_lock(&inode->i_lock);
630 de = __d_find_alias(inode, 0);
631 spin_unlock(&inode->i_lock);
635 EXPORT_SYMBOL(d_find_alias);
638 * Try to kill dentries associated with this inode.
639 * WARNING: you must own a reference to inode.
641 void d_prune_aliases(struct inode *inode)
643 struct dentry *dentry;
645 spin_lock(&inode->i_lock);
646 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
647 spin_lock(&dentry->d_lock);
648 if (!dentry->d_count) {
649 __dget_dlock(dentry);
651 spin_unlock(&dentry->d_lock);
652 spin_unlock(&inode->i_lock);
656 spin_unlock(&dentry->d_lock);
658 spin_unlock(&inode->i_lock);
660 EXPORT_SYMBOL(d_prune_aliases);
663 * Try to throw away a dentry - free the inode, dput the parent.
664 * Requires dentry->d_lock is held, and dentry->d_count == 0.
665 * Releases dentry->d_lock.
667 * This may fail if locks cannot be acquired no problem, just try again.
669 static void try_prune_one_dentry(struct dentry *dentry)
670 __releases(dentry->d_lock)
672 struct dentry *parent;
674 parent = dentry_kill(dentry, 0);
676 * If dentry_kill returns NULL, we have nothing more to do.
677 * if it returns the same dentry, trylocks failed. In either
678 * case, just loop again.
680 * Otherwise, we need to prune ancestors too. This is necessary
681 * to prevent quadratic behavior of shrink_dcache_parent(), but
682 * is also expected to be beneficial in reducing dentry cache
687 if (parent == dentry)
690 /* Prune ancestors. */
693 spin_lock(&dentry->d_lock);
694 if (dentry->d_count > 1) {
696 spin_unlock(&dentry->d_lock);
699 dentry = dentry_kill(dentry, 1);
703 static void shrink_dentry_list(struct list_head *list)
705 struct dentry *dentry;
709 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
710 if (&dentry->d_lru == list)
712 spin_lock(&dentry->d_lock);
713 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
714 spin_unlock(&dentry->d_lock);
719 * We found an inuse dentry which was not removed from
720 * the LRU because of laziness during lookup. Do not free
721 * it - just keep it off the LRU list.
723 if (dentry->d_count) {
724 dentry_lru_del(dentry);
725 spin_unlock(&dentry->d_lock);
731 try_prune_one_dentry(dentry);
739 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
740 * @sb: superblock to shrink dentry LRU.
741 * @count: number of entries to prune
742 * @flags: flags to control the dentry processing
744 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
746 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
748 /* called from prune_dcache() and shrink_dcache_parent() */
749 struct dentry *dentry;
750 LIST_HEAD(referenced);
755 spin_lock(&dcache_lru_lock);
756 while (!list_empty(&sb->s_dentry_lru)) {
757 dentry = list_entry(sb->s_dentry_lru.prev,
758 struct dentry, d_lru);
759 BUG_ON(dentry->d_sb != sb);
761 if (!spin_trylock(&dentry->d_lock)) {
762 spin_unlock(&dcache_lru_lock);
768 * If we are honouring the DCACHE_REFERENCED flag and the
769 * dentry has this flag set, don't free it. Clear the flag
770 * and put it back on the LRU.
772 if (flags & DCACHE_REFERENCED &&
773 dentry->d_flags & DCACHE_REFERENCED) {
774 dentry->d_flags &= ~DCACHE_REFERENCED;
775 list_move(&dentry->d_lru, &referenced);
776 spin_unlock(&dentry->d_lock);
778 list_move_tail(&dentry->d_lru, &tmp);
779 spin_unlock(&dentry->d_lock);
783 cond_resched_lock(&dcache_lru_lock);
785 if (!list_empty(&referenced))
786 list_splice(&referenced, &sb->s_dentry_lru);
787 spin_unlock(&dcache_lru_lock);
789 shrink_dentry_list(&tmp);
795 * prune_dcache - shrink the dcache
796 * @count: number of entries to try to free
798 * Shrink the dcache. This is done when we need more memory, or simply when we
799 * need to unmount something (at which point we need to unuse all dentries).
801 * This function may fail to free any resources if all the dentries are in use.
803 static void prune_dcache(int count)
805 struct super_block *sb, *p = NULL;
807 int unused = dentry_stat.nr_unused;
811 if (unused == 0 || count == 0)
816 prune_ratio = unused / count;
818 list_for_each_entry(sb, &super_blocks, s_list) {
819 if (list_empty(&sb->s_instances))
821 if (sb->s_nr_dentry_unused == 0)
824 /* Now, we reclaim unused dentrins with fairness.
825 * We reclaim them same percentage from each superblock.
826 * We calculate number of dentries to scan on this sb
827 * as follows, but the implementation is arranged to avoid
829 * number of dentries to scan on this sb =
830 * count * (number of dentries on this sb /
831 * number of dentries in the machine)
833 spin_unlock(&sb_lock);
834 if (prune_ratio != 1)
835 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
837 w_count = sb->s_nr_dentry_unused;
840 * We need to be sure this filesystem isn't being unmounted,
841 * otherwise we could race with generic_shutdown_super(), and
842 * end up holding a reference to an inode while the filesystem
843 * is unmounted. So we try to get s_umount, and make sure
846 if (down_read_trylock(&sb->s_umount)) {
847 if ((sb->s_root != NULL) &&
848 (!list_empty(&sb->s_dentry_lru))) {
849 __shrink_dcache_sb(sb, &w_count,
853 up_read(&sb->s_umount);
860 /* more work left to do? */
866 spin_unlock(&sb_lock);
870 * shrink_dcache_sb - shrink dcache for a superblock
873 * Shrink the dcache for the specified super block. This is used to free
874 * the dcache before unmounting a file system.
876 void shrink_dcache_sb(struct super_block *sb)
880 spin_lock(&dcache_lru_lock);
881 while (!list_empty(&sb->s_dentry_lru)) {
882 list_splice_init(&sb->s_dentry_lru, &tmp);
883 spin_unlock(&dcache_lru_lock);
884 shrink_dentry_list(&tmp);
885 spin_lock(&dcache_lru_lock);
887 spin_unlock(&dcache_lru_lock);
889 EXPORT_SYMBOL(shrink_dcache_sb);
892 * destroy a single subtree of dentries for unmount
893 * - see the comments on shrink_dcache_for_umount() for a description of the
896 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
898 struct dentry *parent;
899 unsigned detached = 0;
901 BUG_ON(!IS_ROOT(dentry));
903 /* detach this root from the system */
904 spin_lock(&dentry->d_lock);
905 dentry_lru_del(dentry);
907 spin_unlock(&dentry->d_lock);
910 /* descend to the first leaf in the current subtree */
911 while (!list_empty(&dentry->d_subdirs)) {
914 /* this is a branch with children - detach all of them
915 * from the system in one go */
916 spin_lock(&dentry->d_lock);
917 list_for_each_entry(loop, &dentry->d_subdirs,
919 spin_lock_nested(&loop->d_lock,
920 DENTRY_D_LOCK_NESTED);
921 dentry_lru_del(loop);
923 spin_unlock(&loop->d_lock);
925 spin_unlock(&dentry->d_lock);
927 /* move to the first child */
928 dentry = list_entry(dentry->d_subdirs.next,
929 struct dentry, d_u.d_child);
932 /* consume the dentries from this leaf up through its parents
933 * until we find one with children or run out altogether */
937 if (dentry->d_count != 0) {
939 "BUG: Dentry %p{i=%lx,n=%s}"
941 " [unmount of %s %s]\n",
944 dentry->d_inode->i_ino : 0UL,
947 dentry->d_sb->s_type->name,
952 if (IS_ROOT(dentry)) {
954 list_del(&dentry->d_u.d_child);
956 parent = dentry->d_parent;
957 spin_lock(&parent->d_lock);
959 list_del(&dentry->d_u.d_child);
960 spin_unlock(&parent->d_lock);
965 inode = dentry->d_inode;
967 dentry->d_inode = NULL;
968 list_del_init(&dentry->d_alias);
969 if (dentry->d_op && dentry->d_op->d_iput)
970 dentry->d_op->d_iput(dentry, inode);
977 /* finished when we fall off the top of the tree,
978 * otherwise we ascend to the parent and move to the
979 * next sibling if there is one */
983 } while (list_empty(&dentry->d_subdirs));
985 dentry = list_entry(dentry->d_subdirs.next,
986 struct dentry, d_u.d_child);
991 * destroy the dentries attached to a superblock on unmounting
992 * - we don't need to use dentry->d_lock because:
993 * - the superblock is detached from all mountings and open files, so the
994 * dentry trees will not be rearranged by the VFS
995 * - s_umount is write-locked, so the memory pressure shrinker will ignore
996 * any dentries belonging to this superblock that it comes across
997 * - the filesystem itself is no longer permitted to rearrange the dentries
1000 void shrink_dcache_for_umount(struct super_block *sb)
1002 struct dentry *dentry;
1004 if (down_read_trylock(&sb->s_umount))
1007 dentry = sb->s_root;
1009 spin_lock(&dentry->d_lock);
1011 spin_unlock(&dentry->d_lock);
1012 shrink_dcache_for_umount_subtree(dentry);
1014 while (!hlist_bl_empty(&sb->s_anon)) {
1015 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1016 shrink_dcache_for_umount_subtree(dentry);
1021 * This tries to ascend one level of parenthood, but
1022 * we can race with renaming, so we need to re-check
1023 * the parenthood after dropping the lock and check
1024 * that the sequence number still matches.
1026 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1028 struct dentry *new = old->d_parent;
1031 spin_unlock(&old->d_lock);
1032 spin_lock(&new->d_lock);
1035 * might go back up the wrong parent if we have had a rename
1038 if (new != old->d_parent ||
1039 (old->d_flags & DCACHE_DISCONNECTED) ||
1040 (!locked && read_seqretry(&rename_lock, seq))) {
1041 spin_unlock(&new->d_lock);
1050 * Search for at least 1 mount point in the dentry's subdirs.
1051 * We descend to the next level whenever the d_subdirs
1052 * list is non-empty and continue searching.
1056 * have_submounts - check for mounts over a dentry
1057 * @parent: dentry to check.
1059 * Return true if the parent or its subdirectories contain
1062 int have_submounts(struct dentry *parent)
1064 struct dentry *this_parent;
1065 struct list_head *next;
1069 seq = read_seqbegin(&rename_lock);
1071 this_parent = parent;
1073 if (d_mountpoint(parent))
1075 spin_lock(&this_parent->d_lock);
1077 next = this_parent->d_subdirs.next;
1079 while (next != &this_parent->d_subdirs) {
1080 struct list_head *tmp = next;
1081 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1084 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1085 /* Have we found a mount point ? */
1086 if (d_mountpoint(dentry)) {
1087 spin_unlock(&dentry->d_lock);
1088 spin_unlock(&this_parent->d_lock);
1091 if (!list_empty(&dentry->d_subdirs)) {
1092 spin_unlock(&this_parent->d_lock);
1093 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1094 this_parent = dentry;
1095 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1098 spin_unlock(&dentry->d_lock);
1101 * All done at this level ... ascend and resume the search.
1103 if (this_parent != parent) {
1104 struct dentry *child = this_parent;
1105 this_parent = try_to_ascend(this_parent, locked, seq);
1108 next = child->d_u.d_child.next;
1111 spin_unlock(&this_parent->d_lock);
1112 if (!locked && read_seqretry(&rename_lock, seq))
1115 write_sequnlock(&rename_lock);
1116 return 0; /* No mount points found in tree */
1118 if (!locked && read_seqretry(&rename_lock, seq))
1121 write_sequnlock(&rename_lock);
1126 write_seqlock(&rename_lock);
1129 EXPORT_SYMBOL(have_submounts);
1132 * Search the dentry child list for the specified parent,
1133 * and move any unused dentries to the end of the unused
1134 * list for prune_dcache(). We descend to the next level
1135 * whenever the d_subdirs list is non-empty and continue
1138 * It returns zero iff there are no unused children,
1139 * otherwise it returns the number of children moved to
1140 * the end of the unused list. This may not be the total
1141 * number of unused children, because select_parent can
1142 * drop the lock and return early due to latency
1145 static int select_parent(struct dentry * parent)
1147 struct dentry *this_parent;
1148 struct list_head *next;
1153 seq = read_seqbegin(&rename_lock);
1155 this_parent = parent;
1156 spin_lock(&this_parent->d_lock);
1158 next = this_parent->d_subdirs.next;
1160 while (next != &this_parent->d_subdirs) {
1161 struct list_head *tmp = next;
1162 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1165 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1168 * move only zero ref count dentries to the end
1169 * of the unused list for prune_dcache
1171 if (!dentry->d_count) {
1172 dentry_lru_move_tail(dentry);
1175 dentry_lru_del(dentry);
1179 * We can return to the caller if we have found some (this
1180 * ensures forward progress). We'll be coming back to find
1183 if (found && need_resched()) {
1184 spin_unlock(&dentry->d_lock);
1189 * Descend a level if the d_subdirs list is non-empty.
1191 if (!list_empty(&dentry->d_subdirs)) {
1192 spin_unlock(&this_parent->d_lock);
1193 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1194 this_parent = dentry;
1195 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1199 spin_unlock(&dentry->d_lock);
1202 * All done at this level ... ascend and resume the search.
1204 if (this_parent != parent) {
1205 struct dentry *child = this_parent;
1206 this_parent = try_to_ascend(this_parent, locked, seq);
1209 next = child->d_u.d_child.next;
1213 spin_unlock(&this_parent->d_lock);
1214 if (!locked && read_seqretry(&rename_lock, seq))
1217 write_sequnlock(&rename_lock);
1224 write_seqlock(&rename_lock);
1229 * shrink_dcache_parent - prune dcache
1230 * @parent: parent of entries to prune
1232 * Prune the dcache to remove unused children of the parent dentry.
1235 void shrink_dcache_parent(struct dentry * parent)
1237 struct super_block *sb = parent->d_sb;
1240 while ((found = select_parent(parent)) != 0)
1241 __shrink_dcache_sb(sb, &found, 0);
1243 EXPORT_SYMBOL(shrink_dcache_parent);
1246 * Scan `sc->nr_slab_to_reclaim' dentries and return the number which remain.
1248 * We need to avoid reentering the filesystem if the caller is performing a
1249 * GFP_NOFS allocation attempt. One example deadlock is:
1251 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1252 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1253 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1255 * In this case we return -1 to tell the caller that we baled.
1257 static int shrink_dcache_memory(struct shrinker *shrink,
1258 struct shrink_control *sc)
1260 int nr = sc->nr_to_scan;
1261 gfp_t gfp_mask = sc->gfp_mask;
1264 if (!(gfp_mask & __GFP_FS))
1269 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
1272 static struct shrinker dcache_shrinker = {
1273 .shrink = shrink_dcache_memory,
1274 .seeks = DEFAULT_SEEKS,
1278 * d_alloc - allocate a dcache entry
1279 * @parent: parent of entry to allocate
1280 * @name: qstr of the name
1282 * Allocates a dentry. It returns %NULL if there is insufficient memory
1283 * available. On a success the dentry is returned. The name passed in is
1284 * copied and the copy passed in may be reused after this call.
1287 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1289 struct dentry *dentry;
1292 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1296 if (name->len > DNAME_INLINE_LEN-1) {
1297 dname = kmalloc(name->len + 1, GFP_KERNEL);
1299 kmem_cache_free(dentry_cache, dentry);
1303 dname = dentry->d_iname;
1305 dentry->d_name.name = dname;
1307 dentry->d_name.len = name->len;
1308 dentry->d_name.hash = name->hash;
1309 memcpy(dname, name->name, name->len);
1310 dname[name->len] = 0;
1312 dentry->d_count = 1;
1313 dentry->d_flags = 0;
1314 spin_lock_init(&dentry->d_lock);
1315 seqcount_init(&dentry->d_seq);
1316 dentry->d_inode = NULL;
1317 dentry->d_parent = NULL;
1318 dentry->d_sb = NULL;
1319 dentry->d_op = NULL;
1320 dentry->d_fsdata = NULL;
1321 INIT_HLIST_BL_NODE(&dentry->d_hash);
1322 INIT_LIST_HEAD(&dentry->d_lru);
1323 INIT_LIST_HEAD(&dentry->d_subdirs);
1324 INIT_LIST_HEAD(&dentry->d_alias);
1325 INIT_LIST_HEAD(&dentry->d_u.d_child);
1328 spin_lock(&parent->d_lock);
1330 * don't need child lock because it is not subject
1331 * to concurrency here
1333 __dget_dlock(parent);
1334 dentry->d_parent = parent;
1335 dentry->d_sb = parent->d_sb;
1336 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1337 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1338 spin_unlock(&parent->d_lock);
1341 this_cpu_inc(nr_dentry);
1345 EXPORT_SYMBOL(d_alloc);
1347 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1349 struct dentry *dentry = d_alloc(NULL, name);
1352 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1353 dentry->d_parent = dentry;
1354 dentry->d_flags |= DCACHE_DISCONNECTED;
1358 EXPORT_SYMBOL(d_alloc_pseudo);
1360 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1365 q.len = strlen(name);
1366 q.hash = full_name_hash(q.name, q.len);
1367 return d_alloc(parent, &q);
1369 EXPORT_SYMBOL(d_alloc_name);
1371 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1373 WARN_ON_ONCE(dentry->d_op);
1374 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1376 DCACHE_OP_REVALIDATE |
1377 DCACHE_OP_DELETE ));
1382 dentry->d_flags |= DCACHE_OP_HASH;
1384 dentry->d_flags |= DCACHE_OP_COMPARE;
1385 if (op->d_revalidate)
1386 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1388 dentry->d_flags |= DCACHE_OP_DELETE;
1391 EXPORT_SYMBOL(d_set_d_op);
1393 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1395 spin_lock(&dentry->d_lock);
1397 if (unlikely(IS_AUTOMOUNT(inode)))
1398 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1399 list_add(&dentry->d_alias, &inode->i_dentry);
1401 dentry->d_inode = inode;
1402 dentry_rcuwalk_barrier(dentry);
1403 spin_unlock(&dentry->d_lock);
1404 fsnotify_d_instantiate(dentry, inode);
1408 * d_instantiate - fill in inode information for a dentry
1409 * @entry: dentry to complete
1410 * @inode: inode to attach to this dentry
1412 * Fill in inode information in the entry.
1414 * This turns negative dentries into productive full members
1417 * NOTE! This assumes that the inode count has been incremented
1418 * (or otherwise set) by the caller to indicate that it is now
1419 * in use by the dcache.
1422 void d_instantiate(struct dentry *entry, struct inode * inode)
1424 BUG_ON(!list_empty(&entry->d_alias));
1426 spin_lock(&inode->i_lock);
1427 __d_instantiate(entry, inode);
1429 spin_unlock(&inode->i_lock);
1430 security_d_instantiate(entry, inode);
1432 EXPORT_SYMBOL(d_instantiate);
1435 * d_instantiate_unique - instantiate a non-aliased dentry
1436 * @entry: dentry to instantiate
1437 * @inode: inode to attach to this dentry
1439 * Fill in inode information in the entry. On success, it returns NULL.
1440 * If an unhashed alias of "entry" already exists, then we return the
1441 * aliased dentry instead and drop one reference to inode.
1443 * Note that in order to avoid conflicts with rename() etc, the caller
1444 * had better be holding the parent directory semaphore.
1446 * This also assumes that the inode count has been incremented
1447 * (or otherwise set) by the caller to indicate that it is now
1448 * in use by the dcache.
1450 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1451 struct inode *inode)
1453 struct dentry *alias;
1454 int len = entry->d_name.len;
1455 const char *name = entry->d_name.name;
1456 unsigned int hash = entry->d_name.hash;
1459 __d_instantiate(entry, NULL);
1463 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1464 struct qstr *qstr = &alias->d_name;
1467 * Don't need alias->d_lock here, because aliases with
1468 * d_parent == entry->d_parent are not subject to name or
1469 * parent changes, because the parent inode i_mutex is held.
1471 if (qstr->hash != hash)
1473 if (alias->d_parent != entry->d_parent)
1475 if (dentry_cmp(qstr->name, qstr->len, name, len))
1481 __d_instantiate(entry, inode);
1485 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1487 struct dentry *result;
1489 BUG_ON(!list_empty(&entry->d_alias));
1492 spin_lock(&inode->i_lock);
1493 result = __d_instantiate_unique(entry, inode);
1495 spin_unlock(&inode->i_lock);
1498 security_d_instantiate(entry, inode);
1502 BUG_ON(!d_unhashed(result));
1507 EXPORT_SYMBOL(d_instantiate_unique);
1510 * d_alloc_root - allocate root dentry
1511 * @root_inode: inode to allocate the root for
1513 * Allocate a root ("/") dentry for the inode given. The inode is
1514 * instantiated and returned. %NULL is returned if there is insufficient
1515 * memory or the inode passed is %NULL.
1518 struct dentry * d_alloc_root(struct inode * root_inode)
1520 struct dentry *res = NULL;
1523 static const struct qstr name = { .name = "/", .len = 1 };
1525 res = d_alloc(NULL, &name);
1527 res->d_sb = root_inode->i_sb;
1528 d_set_d_op(res, res->d_sb->s_d_op);
1529 res->d_parent = res;
1530 d_instantiate(res, root_inode);
1535 EXPORT_SYMBOL(d_alloc_root);
1537 static struct dentry * __d_find_any_alias(struct inode *inode)
1539 struct dentry *alias;
1541 if (list_empty(&inode->i_dentry))
1543 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1548 static struct dentry * d_find_any_alias(struct inode *inode)
1552 spin_lock(&inode->i_lock);
1553 de = __d_find_any_alias(inode);
1554 spin_unlock(&inode->i_lock);
1560 * d_obtain_alias - find or allocate a dentry for a given inode
1561 * @inode: inode to allocate the dentry for
1563 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1564 * similar open by handle operations. The returned dentry may be anonymous,
1565 * or may have a full name (if the inode was already in the cache).
1567 * When called on a directory inode, we must ensure that the inode only ever
1568 * has one dentry. If a dentry is found, that is returned instead of
1569 * allocating a new one.
1571 * On successful return, the reference to the inode has been transferred
1572 * to the dentry. In case of an error the reference on the inode is released.
1573 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1574 * be passed in and will be the error will be propagate to the return value,
1575 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1577 struct dentry *d_obtain_alias(struct inode *inode)
1579 static const struct qstr anonstring = { .name = "" };
1584 return ERR_PTR(-ESTALE);
1586 return ERR_CAST(inode);
1588 res = d_find_any_alias(inode);
1592 tmp = d_alloc(NULL, &anonstring);
1594 res = ERR_PTR(-ENOMEM);
1597 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1600 spin_lock(&inode->i_lock);
1601 res = __d_find_any_alias(inode);
1603 spin_unlock(&inode->i_lock);
1608 /* attach a disconnected dentry */
1609 spin_lock(&tmp->d_lock);
1610 tmp->d_sb = inode->i_sb;
1611 d_set_d_op(tmp, tmp->d_sb->s_d_op);
1612 tmp->d_inode = inode;
1613 tmp->d_flags |= DCACHE_DISCONNECTED;
1614 list_add(&tmp->d_alias, &inode->i_dentry);
1615 hlist_bl_lock(&tmp->d_sb->s_anon);
1616 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1617 hlist_bl_unlock(&tmp->d_sb->s_anon);
1618 spin_unlock(&tmp->d_lock);
1619 spin_unlock(&inode->i_lock);
1620 security_d_instantiate(tmp, inode);
1625 if (res && !IS_ERR(res))
1626 security_d_instantiate(res, inode);
1630 EXPORT_SYMBOL(d_obtain_alias);
1633 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1634 * @inode: the inode which may have a disconnected dentry
1635 * @dentry: a negative dentry which we want to point to the inode.
1637 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1638 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1639 * and return it, else simply d_add the inode to the dentry and return NULL.
1641 * This is needed in the lookup routine of any filesystem that is exportable
1642 * (via knfsd) so that we can build dcache paths to directories effectively.
1644 * If a dentry was found and moved, then it is returned. Otherwise NULL
1645 * is returned. This matches the expected return value of ->lookup.
1648 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1650 struct dentry *new = NULL;
1652 if (inode && S_ISDIR(inode->i_mode)) {
1653 spin_lock(&inode->i_lock);
1654 new = __d_find_alias(inode, 1);
1656 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1657 spin_unlock(&inode->i_lock);
1658 security_d_instantiate(new, inode);
1659 d_move(new, dentry);
1662 /* already taking inode->i_lock, so d_add() by hand */
1663 __d_instantiate(dentry, inode);
1664 spin_unlock(&inode->i_lock);
1665 security_d_instantiate(dentry, inode);
1669 d_add(dentry, inode);
1672 EXPORT_SYMBOL(d_splice_alias);
1675 * d_add_ci - lookup or allocate new dentry with case-exact name
1676 * @inode: the inode case-insensitive lookup has found
1677 * @dentry: the negative dentry that was passed to the parent's lookup func
1678 * @name: the case-exact name to be associated with the returned dentry
1680 * This is to avoid filling the dcache with case-insensitive names to the
1681 * same inode, only the actual correct case is stored in the dcache for
1682 * case-insensitive filesystems.
1684 * For a case-insensitive lookup match and if the the case-exact dentry
1685 * already exists in in the dcache, use it and return it.
1687 * If no entry exists with the exact case name, allocate new dentry with
1688 * the exact case, and return the spliced entry.
1690 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1694 struct dentry *found;
1698 * First check if a dentry matching the name already exists,
1699 * if not go ahead and create it now.
1701 found = d_hash_and_lookup(dentry->d_parent, name);
1703 new = d_alloc(dentry->d_parent, name);
1709 found = d_splice_alias(inode, new);
1718 * If a matching dentry exists, and it's not negative use it.
1720 * Decrement the reference count to balance the iget() done
1723 if (found->d_inode) {
1724 if (unlikely(found->d_inode != inode)) {
1725 /* This can't happen because bad inodes are unhashed. */
1726 BUG_ON(!is_bad_inode(inode));
1727 BUG_ON(!is_bad_inode(found->d_inode));
1734 * We are going to instantiate this dentry, unhash it and clear the
1735 * lookup flag so we can do that.
1737 if (unlikely(d_need_lookup(found)))
1738 d_clear_need_lookup(found);
1741 * Negative dentry: instantiate it unless the inode is a directory and
1742 * already has a dentry.
1744 spin_lock(&inode->i_lock);
1745 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1746 __d_instantiate(found, inode);
1747 spin_unlock(&inode->i_lock);
1748 security_d_instantiate(found, inode);
1753 * In case a directory already has a (disconnected) entry grab a
1754 * reference to it, move it in place and use it.
1756 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1758 spin_unlock(&inode->i_lock);
1759 security_d_instantiate(found, inode);
1767 return ERR_PTR(error);
1769 EXPORT_SYMBOL(d_add_ci);
1772 * __d_lookup_rcu - search for a dentry (racy, store-free)
1773 * @parent: parent dentry
1774 * @name: qstr of name we wish to find
1775 * @seq: returns d_seq value at the point where the dentry was found
1776 * @inode: returns dentry->d_inode when the inode was found valid.
1777 * Returns: dentry, or NULL
1779 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1780 * resolution (store-free path walking) design described in
1781 * Documentation/filesystems/path-lookup.txt.
1783 * This is not to be used outside core vfs.
1785 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1786 * held, and rcu_read_lock held. The returned dentry must not be stored into
1787 * without taking d_lock and checking d_seq sequence count against @seq
1790 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1793 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1794 * the returned dentry, so long as its parent's seqlock is checked after the
1795 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1796 * is formed, giving integrity down the path walk.
1798 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1799 unsigned *seq, struct inode **inode)
1801 unsigned int len = name->len;
1802 unsigned int hash = name->hash;
1803 const unsigned char *str = name->name;
1804 struct hlist_bl_head *b = d_hash(parent, hash);
1805 struct hlist_bl_node *node;
1806 struct dentry *dentry;
1809 * Note: There is significant duplication with __d_lookup_rcu which is
1810 * required to prevent single threaded performance regressions
1811 * especially on architectures where smp_rmb (in seqcounts) are costly.
1812 * Keep the two functions in sync.
1816 * The hash list is protected using RCU.
1818 * Carefully use d_seq when comparing a candidate dentry, to avoid
1819 * races with d_move().
1821 * It is possible that concurrent renames can mess up our list
1822 * walk here and result in missing our dentry, resulting in the
1823 * false-negative result. d_lookup() protects against concurrent
1824 * renames using rename_lock seqlock.
1826 * See Documentation/filesystems/path-lookup.txt for more details.
1828 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1833 if (dentry->d_name.hash != hash)
1837 *seq = read_seqcount_begin(&dentry->d_seq);
1838 if (dentry->d_parent != parent)
1840 if (d_unhashed(dentry))
1842 tlen = dentry->d_name.len;
1843 tname = dentry->d_name.name;
1844 i = dentry->d_inode;
1849 * This seqcount check is required to ensure name and
1850 * len are loaded atomically, so as not to walk off the
1851 * edge of memory when walking. If we could load this
1852 * atomically some other way, we could drop this check.
1854 if (read_seqcount_retry(&dentry->d_seq, *seq))
1856 if (parent->d_flags & DCACHE_OP_COMPARE) {
1857 if (parent->d_op->d_compare(parent, *inode,
1862 if (dentry_cmp(tname, tlen, str, len))
1866 * No extra seqcount check is required after the name
1867 * compare. The caller must perform a seqcount check in
1868 * order to do anything useful with the returned dentry
1878 * d_lookup - search for a dentry
1879 * @parent: parent dentry
1880 * @name: qstr of name we wish to find
1881 * Returns: dentry, or NULL
1883 * d_lookup searches the children of the parent dentry for the name in
1884 * question. If the dentry is found its reference count is incremented and the
1885 * dentry is returned. The caller must use dput to free the entry when it has
1886 * finished using it. %NULL is returned if the dentry does not exist.
1888 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1890 struct dentry *dentry;
1894 seq = read_seqbegin(&rename_lock);
1895 dentry = __d_lookup(parent, name);
1898 } while (read_seqretry(&rename_lock, seq));
1901 EXPORT_SYMBOL(d_lookup);
1904 * __d_lookup - search for a dentry (racy)
1905 * @parent: parent dentry
1906 * @name: qstr of name we wish to find
1907 * Returns: dentry, or NULL
1909 * __d_lookup is like d_lookup, however it may (rarely) return a
1910 * false-negative result due to unrelated rename activity.
1912 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1913 * however it must be used carefully, eg. with a following d_lookup in
1914 * the case of failure.
1916 * __d_lookup callers must be commented.
1918 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1920 unsigned int len = name->len;
1921 unsigned int hash = name->hash;
1922 const unsigned char *str = name->name;
1923 struct hlist_bl_head *b = d_hash(parent, hash);
1924 struct hlist_bl_node *node;
1925 struct dentry *found = NULL;
1926 struct dentry *dentry;
1929 * Note: There is significant duplication with __d_lookup_rcu which is
1930 * required to prevent single threaded performance regressions
1931 * especially on architectures where smp_rmb (in seqcounts) are costly.
1932 * Keep the two functions in sync.
1936 * The hash list is protected using RCU.
1938 * Take d_lock when comparing a candidate dentry, to avoid races
1941 * It is possible that concurrent renames can mess up our list
1942 * walk here and result in missing our dentry, resulting in the
1943 * false-negative result. d_lookup() protects against concurrent
1944 * renames using rename_lock seqlock.
1946 * See Documentation/filesystems/path-lookup.txt for more details.
1950 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1954 if (dentry->d_name.hash != hash)
1957 spin_lock(&dentry->d_lock);
1958 if (dentry->d_parent != parent)
1960 if (d_unhashed(dentry))
1964 * It is safe to compare names since d_move() cannot
1965 * change the qstr (protected by d_lock).
1967 tlen = dentry->d_name.len;
1968 tname = dentry->d_name.name;
1969 if (parent->d_flags & DCACHE_OP_COMPARE) {
1970 if (parent->d_op->d_compare(parent, parent->d_inode,
1971 dentry, dentry->d_inode,
1975 if (dentry_cmp(tname, tlen, str, len))
1981 spin_unlock(&dentry->d_lock);
1984 spin_unlock(&dentry->d_lock);
1992 * d_hash_and_lookup - hash the qstr then search for a dentry
1993 * @dir: Directory to search in
1994 * @name: qstr of name we wish to find
1996 * On hash failure or on lookup failure NULL is returned.
1998 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2000 struct dentry *dentry = NULL;
2003 * Check for a fs-specific hash function. Note that we must
2004 * calculate the standard hash first, as the d_op->d_hash()
2005 * routine may choose to leave the hash value unchanged.
2007 name->hash = full_name_hash(name->name, name->len);
2008 if (dir->d_flags & DCACHE_OP_HASH) {
2009 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
2012 dentry = d_lookup(dir, name);
2018 * d_validate - verify dentry provided from insecure source (deprecated)
2019 * @dentry: The dentry alleged to be valid child of @dparent
2020 * @dparent: The parent dentry (known to be valid)
2022 * An insecure source has sent us a dentry, here we verify it and dget() it.
2023 * This is used by ncpfs in its readdir implementation.
2024 * Zero is returned in the dentry is invalid.
2026 * This function is slow for big directories, and deprecated, do not use it.
2028 int d_validate(struct dentry *dentry, struct dentry *dparent)
2030 struct dentry *child;
2032 spin_lock(&dparent->d_lock);
2033 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2034 if (dentry == child) {
2035 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2036 __dget_dlock(dentry);
2037 spin_unlock(&dentry->d_lock);
2038 spin_unlock(&dparent->d_lock);
2042 spin_unlock(&dparent->d_lock);
2046 EXPORT_SYMBOL(d_validate);
2049 * When a file is deleted, we have two options:
2050 * - turn this dentry into a negative dentry
2051 * - unhash this dentry and free it.
2053 * Usually, we want to just turn this into
2054 * a negative dentry, but if anybody else is
2055 * currently using the dentry or the inode
2056 * we can't do that and we fall back on removing
2057 * it from the hash queues and waiting for
2058 * it to be deleted later when it has no users
2062 * d_delete - delete a dentry
2063 * @dentry: The dentry to delete
2065 * Turn the dentry into a negative dentry if possible, otherwise
2066 * remove it from the hash queues so it can be deleted later
2069 void d_delete(struct dentry * dentry)
2071 struct inode *inode;
2074 * Are we the only user?
2077 spin_lock(&dentry->d_lock);
2078 inode = dentry->d_inode;
2079 isdir = S_ISDIR(inode->i_mode);
2080 if (dentry->d_count == 1) {
2081 if (inode && !spin_trylock(&inode->i_lock)) {
2082 spin_unlock(&dentry->d_lock);
2086 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2087 dentry_unlink_inode(dentry);
2088 fsnotify_nameremove(dentry, isdir);
2092 if (!d_unhashed(dentry))
2095 spin_unlock(&dentry->d_lock);
2097 fsnotify_nameremove(dentry, isdir);
2099 EXPORT_SYMBOL(d_delete);
2101 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2103 BUG_ON(!d_unhashed(entry));
2105 entry->d_flags |= DCACHE_RCUACCESS;
2106 hlist_bl_add_head_rcu(&entry->d_hash, b);
2110 static void _d_rehash(struct dentry * entry)
2112 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2116 * d_rehash - add an entry back to the hash
2117 * @entry: dentry to add to the hash
2119 * Adds a dentry to the hash according to its name.
2122 void d_rehash(struct dentry * entry)
2124 spin_lock(&entry->d_lock);
2126 spin_unlock(&entry->d_lock);
2128 EXPORT_SYMBOL(d_rehash);
2131 * dentry_update_name_case - update case insensitive dentry with a new name
2132 * @dentry: dentry to be updated
2135 * Update a case insensitive dentry with new case of name.
2137 * dentry must have been returned by d_lookup with name @name. Old and new
2138 * name lengths must match (ie. no d_compare which allows mismatched name
2141 * Parent inode i_mutex must be held over d_lookup and into this call (to
2142 * keep renames and concurrent inserts, and readdir(2) away).
2144 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2146 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2147 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2149 spin_lock(&dentry->d_lock);
2150 write_seqcount_begin(&dentry->d_seq);
2151 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2152 write_seqcount_end(&dentry->d_seq);
2153 spin_unlock(&dentry->d_lock);
2155 EXPORT_SYMBOL(dentry_update_name_case);
2157 static void switch_names(struct dentry *dentry, struct dentry *target)
2159 if (dname_external(target)) {
2160 if (dname_external(dentry)) {
2162 * Both external: swap the pointers
2164 swap(target->d_name.name, dentry->d_name.name);
2167 * dentry:internal, target:external. Steal target's
2168 * storage and make target internal.
2170 memcpy(target->d_iname, dentry->d_name.name,
2171 dentry->d_name.len + 1);
2172 dentry->d_name.name = target->d_name.name;
2173 target->d_name.name = target->d_iname;
2176 if (dname_external(dentry)) {
2178 * dentry:external, target:internal. Give dentry's
2179 * storage to target and make dentry internal
2181 memcpy(dentry->d_iname, target->d_name.name,
2182 target->d_name.len + 1);
2183 target->d_name.name = dentry->d_name.name;
2184 dentry->d_name.name = dentry->d_iname;
2187 * Both are internal. Just copy target to dentry
2189 memcpy(dentry->d_iname, target->d_name.name,
2190 target->d_name.len + 1);
2191 dentry->d_name.len = target->d_name.len;
2195 swap(dentry->d_name.len, target->d_name.len);
2198 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2201 * XXXX: do we really need to take target->d_lock?
2203 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2204 spin_lock(&target->d_parent->d_lock);
2206 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2207 spin_lock(&dentry->d_parent->d_lock);
2208 spin_lock_nested(&target->d_parent->d_lock,
2209 DENTRY_D_LOCK_NESTED);
2211 spin_lock(&target->d_parent->d_lock);
2212 spin_lock_nested(&dentry->d_parent->d_lock,
2213 DENTRY_D_LOCK_NESTED);
2216 if (target < dentry) {
2217 spin_lock_nested(&target->d_lock, 2);
2218 spin_lock_nested(&dentry->d_lock, 3);
2220 spin_lock_nested(&dentry->d_lock, 2);
2221 spin_lock_nested(&target->d_lock, 3);
2225 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2226 struct dentry *target)
2228 if (target->d_parent != dentry->d_parent)
2229 spin_unlock(&dentry->d_parent->d_lock);
2230 if (target->d_parent != target)
2231 spin_unlock(&target->d_parent->d_lock);
2235 * When switching names, the actual string doesn't strictly have to
2236 * be preserved in the target - because we're dropping the target
2237 * anyway. As such, we can just do a simple memcpy() to copy over
2238 * the new name before we switch.
2240 * Note that we have to be a lot more careful about getting the hash
2241 * switched - we have to switch the hash value properly even if it
2242 * then no longer matches the actual (corrupted) string of the target.
2243 * The hash value has to match the hash queue that the dentry is on..
2246 * __d_move - move a dentry
2247 * @dentry: entry to move
2248 * @target: new dentry
2250 * Update the dcache to reflect the move of a file name. Negative
2251 * dcache entries should not be moved in this way. Caller hold
2254 static void __d_move(struct dentry * dentry, struct dentry * target)
2256 if (!dentry->d_inode)
2257 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2259 BUG_ON(d_ancestor(dentry, target));
2260 BUG_ON(d_ancestor(target, dentry));
2262 dentry_lock_for_move(dentry, target);
2264 write_seqcount_begin(&dentry->d_seq);
2265 write_seqcount_begin(&target->d_seq);
2267 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2270 * Move the dentry to the target hash queue. Don't bother checking
2271 * for the same hash queue because of how unlikely it is.
2274 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2276 /* Unhash the target: dput() will then get rid of it */
2279 list_del(&dentry->d_u.d_child);
2280 list_del(&target->d_u.d_child);
2282 /* Switch the names.. */
2283 switch_names(dentry, target);
2284 swap(dentry->d_name.hash, target->d_name.hash);
2286 /* ... and switch the parents */
2287 if (IS_ROOT(dentry)) {
2288 dentry->d_parent = target->d_parent;
2289 target->d_parent = target;
2290 INIT_LIST_HEAD(&target->d_u.d_child);
2292 swap(dentry->d_parent, target->d_parent);
2294 /* And add them back to the (new) parent lists */
2295 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2298 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2300 write_seqcount_end(&target->d_seq);
2301 write_seqcount_end(&dentry->d_seq);
2303 dentry_unlock_parents_for_move(dentry, target);
2304 spin_unlock(&target->d_lock);
2305 fsnotify_d_move(dentry);
2306 spin_unlock(&dentry->d_lock);
2310 * d_move - move a dentry
2311 * @dentry: entry to move
2312 * @target: new dentry
2314 * Update the dcache to reflect the move of a file name. Negative
2315 * dcache entries should not be moved in this way.
2317 void d_move(struct dentry *dentry, struct dentry *target)
2319 write_seqlock(&rename_lock);
2320 __d_move(dentry, target);
2321 write_sequnlock(&rename_lock);
2323 EXPORT_SYMBOL(d_move);
2326 * d_ancestor - search for an ancestor
2327 * @p1: ancestor dentry
2330 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2331 * an ancestor of p2, else NULL.
2333 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2337 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2338 if (p->d_parent == p1)
2345 * This helper attempts to cope with remotely renamed directories
2347 * It assumes that the caller is already holding
2348 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2350 * Note: If ever the locking in lock_rename() changes, then please
2351 * remember to update this too...
2353 static struct dentry *__d_unalias(struct inode *inode,
2354 struct dentry *dentry, struct dentry *alias)
2356 struct mutex *m1 = NULL, *m2 = NULL;
2359 /* If alias and dentry share a parent, then no extra locks required */
2360 if (alias->d_parent == dentry->d_parent)
2363 /* See lock_rename() */
2364 ret = ERR_PTR(-EBUSY);
2365 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2367 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2368 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2370 m2 = &alias->d_parent->d_inode->i_mutex;
2372 __d_move(alias, dentry);
2375 spin_unlock(&inode->i_lock);
2384 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2385 * named dentry in place of the dentry to be replaced.
2386 * returns with anon->d_lock held!
2388 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2390 struct dentry *dparent, *aparent;
2392 dentry_lock_for_move(anon, dentry);
2394 write_seqcount_begin(&dentry->d_seq);
2395 write_seqcount_begin(&anon->d_seq);
2397 dparent = dentry->d_parent;
2398 aparent = anon->d_parent;
2400 switch_names(dentry, anon);
2401 swap(dentry->d_name.hash, anon->d_name.hash);
2403 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2404 list_del(&dentry->d_u.d_child);
2405 if (!IS_ROOT(dentry))
2406 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2408 INIT_LIST_HEAD(&dentry->d_u.d_child);
2410 anon->d_parent = (dparent == dentry) ? anon : dparent;
2411 list_del(&anon->d_u.d_child);
2413 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2415 INIT_LIST_HEAD(&anon->d_u.d_child);
2417 write_seqcount_end(&dentry->d_seq);
2418 write_seqcount_end(&anon->d_seq);
2420 dentry_unlock_parents_for_move(anon, dentry);
2421 spin_unlock(&dentry->d_lock);
2423 /* anon->d_lock still locked, returns locked */
2424 anon->d_flags &= ~DCACHE_DISCONNECTED;
2428 * d_materialise_unique - introduce an inode into the tree
2429 * @dentry: candidate dentry
2430 * @inode: inode to bind to the dentry, to which aliases may be attached
2432 * Introduces an dentry into the tree, substituting an extant disconnected
2433 * root directory alias in its place if there is one
2435 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2437 struct dentry *actual;
2439 BUG_ON(!d_unhashed(dentry));
2443 __d_instantiate(dentry, NULL);
2448 spin_lock(&inode->i_lock);
2450 if (S_ISDIR(inode->i_mode)) {
2451 struct dentry *alias;
2453 /* Does an aliased dentry already exist? */
2454 alias = __d_find_alias(inode, 0);
2457 write_seqlock(&rename_lock);
2459 if (d_ancestor(alias, dentry)) {
2460 /* Check for loops */
2461 actual = ERR_PTR(-ELOOP);
2462 } else if (IS_ROOT(alias)) {
2463 /* Is this an anonymous mountpoint that we
2464 * could splice into our tree? */
2465 __d_materialise_dentry(dentry, alias);
2466 write_sequnlock(&rename_lock);
2470 /* Nope, but we must(!) avoid directory
2472 actual = __d_unalias(inode, dentry, alias);
2474 write_sequnlock(&rename_lock);
2481 /* Add a unique reference */
2482 actual = __d_instantiate_unique(dentry, inode);
2486 BUG_ON(!d_unhashed(actual));
2488 spin_lock(&actual->d_lock);
2491 spin_unlock(&actual->d_lock);
2492 spin_unlock(&inode->i_lock);
2494 if (actual == dentry) {
2495 security_d_instantiate(dentry, inode);
2502 EXPORT_SYMBOL_GPL(d_materialise_unique);
2504 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2508 return -ENAMETOOLONG;
2510 memcpy(*buffer, str, namelen);
2514 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2516 return prepend(buffer, buflen, name->name, name->len);
2520 * prepend_path - Prepend path string to a buffer
2521 * @path: the dentry/vfsmount to report
2522 * @root: root vfsmnt/dentry (may be modified by this function)
2523 * @buffer: pointer to the end of the buffer
2524 * @buflen: pointer to buffer length
2526 * Caller holds the rename_lock.
2528 * If path is not reachable from the supplied root, then the value of
2529 * root is changed (without modifying refcounts).
2531 static int prepend_path(const struct path *path, struct path *root,
2532 char **buffer, int *buflen)
2534 struct dentry *dentry = path->dentry;
2535 struct vfsmount *vfsmnt = path->mnt;
2539 br_read_lock(vfsmount_lock);
2540 while (dentry != root->dentry || vfsmnt != root->mnt) {
2541 struct dentry * parent;
2543 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2545 if (vfsmnt->mnt_parent == vfsmnt) {
2548 dentry = vfsmnt->mnt_mountpoint;
2549 vfsmnt = vfsmnt->mnt_parent;
2552 parent = dentry->d_parent;
2554 spin_lock(&dentry->d_lock);
2555 error = prepend_name(buffer, buflen, &dentry->d_name);
2556 spin_unlock(&dentry->d_lock);
2558 error = prepend(buffer, buflen, "/", 1);
2567 if (!error && !slash)
2568 error = prepend(buffer, buflen, "/", 1);
2570 br_read_unlock(vfsmount_lock);
2575 * Filesystems needing to implement special "root names"
2576 * should do so with ->d_dname()
2578 if (IS_ROOT(dentry) &&
2579 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2580 WARN(1, "Root dentry has weird name <%.*s>\n",
2581 (int) dentry->d_name.len, dentry->d_name.name);
2584 root->dentry = dentry;
2589 * __d_path - return the path of a dentry
2590 * @path: the dentry/vfsmount to report
2591 * @root: root vfsmnt/dentry (may be modified by this function)
2592 * @buf: buffer to return value in
2593 * @buflen: buffer length
2595 * Convert a dentry into an ASCII path name.
2597 * Returns a pointer into the buffer or an error code if the
2598 * path was too long.
2600 * "buflen" should be positive.
2602 * If path is not reachable from the supplied root, then the value of
2603 * root is changed (without modifying refcounts).
2605 char *__d_path(const struct path *path, struct path *root,
2606 char *buf, int buflen)
2608 char *res = buf + buflen;
2611 prepend(&res, &buflen, "\0", 1);
2612 write_seqlock(&rename_lock);
2613 error = prepend_path(path, root, &res, &buflen);
2614 write_sequnlock(&rename_lock);
2617 return ERR_PTR(error);
2622 * same as __d_path but appends "(deleted)" for unlinked files.
2624 static int path_with_deleted(const struct path *path, struct path *root,
2625 char **buf, int *buflen)
2627 prepend(buf, buflen, "\0", 1);
2628 if (d_unlinked(path->dentry)) {
2629 int error = prepend(buf, buflen, " (deleted)", 10);
2634 return prepend_path(path, root, buf, buflen);
2637 static int prepend_unreachable(char **buffer, int *buflen)
2639 return prepend(buffer, buflen, "(unreachable)", 13);
2643 * d_path - return the path of a dentry
2644 * @path: path to report
2645 * @buf: buffer to return value in
2646 * @buflen: buffer length
2648 * Convert a dentry into an ASCII path name. If the entry has been deleted
2649 * the string " (deleted)" is appended. Note that this is ambiguous.
2651 * Returns a pointer into the buffer or an error code if the path was
2652 * too long. Note: Callers should use the returned pointer, not the passed
2653 * in buffer, to use the name! The implementation often starts at an offset
2654 * into the buffer, and may leave 0 bytes at the start.
2656 * "buflen" should be positive.
2658 char *d_path(const struct path *path, char *buf, int buflen)
2660 char *res = buf + buflen;
2666 * We have various synthetic filesystems that never get mounted. On
2667 * these filesystems dentries are never used for lookup purposes, and
2668 * thus don't need to be hashed. They also don't need a name until a
2669 * user wants to identify the object in /proc/pid/fd/. The little hack
2670 * below allows us to generate a name for these objects on demand:
2672 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2673 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2675 get_fs_root(current->fs, &root);
2676 write_seqlock(&rename_lock);
2678 error = path_with_deleted(path, &tmp, &res, &buflen);
2680 res = ERR_PTR(error);
2681 write_sequnlock(&rename_lock);
2685 EXPORT_SYMBOL(d_path);
2688 * d_path_with_unreachable - return the path of a dentry
2689 * @path: path to report
2690 * @buf: buffer to return value in
2691 * @buflen: buffer length
2693 * The difference from d_path() is that this prepends "(unreachable)"
2694 * to paths which are unreachable from the current process' root.
2696 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2698 char *res = buf + buflen;
2703 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2704 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2706 get_fs_root(current->fs, &root);
2707 write_seqlock(&rename_lock);
2709 error = path_with_deleted(path, &tmp, &res, &buflen);
2710 if (!error && !path_equal(&tmp, &root))
2711 error = prepend_unreachable(&res, &buflen);
2712 write_sequnlock(&rename_lock);
2715 res = ERR_PTR(error);
2721 * Helper function for dentry_operations.d_dname() members
2723 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2724 const char *fmt, ...)
2730 va_start(args, fmt);
2731 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2734 if (sz > sizeof(temp) || sz > buflen)
2735 return ERR_PTR(-ENAMETOOLONG);
2737 buffer += buflen - sz;
2738 return memcpy(buffer, temp, sz);
2742 * Write full pathname from the root of the filesystem into the buffer.
2744 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2746 char *end = buf + buflen;
2749 prepend(&end, &buflen, "\0", 1);
2756 while (!IS_ROOT(dentry)) {
2757 struct dentry *parent = dentry->d_parent;
2761 spin_lock(&dentry->d_lock);
2762 error = prepend_name(&end, &buflen, &dentry->d_name);
2763 spin_unlock(&dentry->d_lock);
2764 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2772 return ERR_PTR(-ENAMETOOLONG);
2775 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2779 write_seqlock(&rename_lock);
2780 retval = __dentry_path(dentry, buf, buflen);
2781 write_sequnlock(&rename_lock);
2785 EXPORT_SYMBOL(dentry_path_raw);
2787 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2792 write_seqlock(&rename_lock);
2793 if (d_unlinked(dentry)) {
2795 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2799 retval = __dentry_path(dentry, buf, buflen);
2800 write_sequnlock(&rename_lock);
2801 if (!IS_ERR(retval) && p)
2802 *p = '/'; /* restore '/' overriden with '\0' */
2805 return ERR_PTR(-ENAMETOOLONG);
2809 * NOTE! The user-level library version returns a
2810 * character pointer. The kernel system call just
2811 * returns the length of the buffer filled (which
2812 * includes the ending '\0' character), or a negative
2813 * error value. So libc would do something like
2815 * char *getcwd(char * buf, size_t size)
2819 * retval = sys_getcwd(buf, size);
2826 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2829 struct path pwd, root;
2830 char *page = (char *) __get_free_page(GFP_USER);
2835 get_fs_root_and_pwd(current->fs, &root, &pwd);
2838 write_seqlock(&rename_lock);
2839 if (!d_unlinked(pwd.dentry)) {
2841 struct path tmp = root;
2842 char *cwd = page + PAGE_SIZE;
2843 int buflen = PAGE_SIZE;
2845 prepend(&cwd, &buflen, "\0", 1);
2846 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2847 write_sequnlock(&rename_lock);
2852 /* Unreachable from current root */
2853 if (!path_equal(&tmp, &root)) {
2854 error = prepend_unreachable(&cwd, &buflen);
2860 len = PAGE_SIZE + page - cwd;
2863 if (copy_to_user(buf, cwd, len))
2867 write_sequnlock(&rename_lock);
2873 free_page((unsigned long) page);
2878 * Test whether new_dentry is a subdirectory of old_dentry.
2880 * Trivially implemented using the dcache structure
2884 * is_subdir - is new dentry a subdirectory of old_dentry
2885 * @new_dentry: new dentry
2886 * @old_dentry: old dentry
2888 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2889 * Returns 0 otherwise.
2890 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2893 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2898 if (new_dentry == old_dentry)
2902 /* for restarting inner loop in case of seq retry */
2903 seq = read_seqbegin(&rename_lock);
2905 * Need rcu_readlock to protect against the d_parent trashing
2909 if (d_ancestor(old_dentry, new_dentry))
2914 } while (read_seqretry(&rename_lock, seq));
2919 int path_is_under(struct path *path1, struct path *path2)
2921 struct vfsmount *mnt = path1->mnt;
2922 struct dentry *dentry = path1->dentry;
2925 br_read_lock(vfsmount_lock);
2926 if (mnt != path2->mnt) {
2928 if (mnt->mnt_parent == mnt) {
2929 br_read_unlock(vfsmount_lock);
2932 if (mnt->mnt_parent == path2->mnt)
2934 mnt = mnt->mnt_parent;
2936 dentry = mnt->mnt_mountpoint;
2938 res = is_subdir(dentry, path2->dentry);
2939 br_read_unlock(vfsmount_lock);
2942 EXPORT_SYMBOL(path_is_under);
2944 void d_genocide(struct dentry *root)
2946 struct dentry *this_parent;
2947 struct list_head *next;
2951 seq = read_seqbegin(&rename_lock);
2954 spin_lock(&this_parent->d_lock);
2956 next = this_parent->d_subdirs.next;
2958 while (next != &this_parent->d_subdirs) {
2959 struct list_head *tmp = next;
2960 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2963 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2964 if (d_unhashed(dentry) || !dentry->d_inode) {
2965 spin_unlock(&dentry->d_lock);
2968 if (!list_empty(&dentry->d_subdirs)) {
2969 spin_unlock(&this_parent->d_lock);
2970 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2971 this_parent = dentry;
2972 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2975 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2976 dentry->d_flags |= DCACHE_GENOCIDE;
2979 spin_unlock(&dentry->d_lock);
2981 if (this_parent != root) {
2982 struct dentry *child = this_parent;
2983 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2984 this_parent->d_flags |= DCACHE_GENOCIDE;
2985 this_parent->d_count--;
2987 this_parent = try_to_ascend(this_parent, locked, seq);
2990 next = child->d_u.d_child.next;
2993 spin_unlock(&this_parent->d_lock);
2994 if (!locked && read_seqretry(&rename_lock, seq))
2997 write_sequnlock(&rename_lock);
3002 write_seqlock(&rename_lock);
3007 * find_inode_number - check for dentry with name
3008 * @dir: directory to check
3009 * @name: Name to find.
3011 * Check whether a dentry already exists for the given name,
3012 * and return the inode number if it has an inode. Otherwise
3015 * This routine is used to post-process directory listings for
3016 * filesystems using synthetic inode numbers, and is necessary
3017 * to keep getcwd() working.
3020 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3022 struct dentry * dentry;
3025 dentry = d_hash_and_lookup(dir, name);
3027 if (dentry->d_inode)
3028 ino = dentry->d_inode->i_ino;
3033 EXPORT_SYMBOL(find_inode_number);
3035 static __initdata unsigned long dhash_entries;
3036 static int __init set_dhash_entries(char *str)
3040 dhash_entries = simple_strtoul(str, &str, 0);
3043 __setup("dhash_entries=", set_dhash_entries);
3045 static void __init dcache_init_early(void)
3049 /* If hashes are distributed across NUMA nodes, defer
3050 * hash allocation until vmalloc space is available.
3056 alloc_large_system_hash("Dentry cache",
3057 sizeof(struct hlist_bl_head),
3065 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3066 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3069 static void __init dcache_init(void)
3074 * A constructor could be added for stable state like the lists,
3075 * but it is probably not worth it because of the cache nature
3078 dentry_cache = KMEM_CACHE(dentry,
3079 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3081 register_shrinker(&dcache_shrinker);
3083 /* Hash may have been set up in dcache_init_early */
3088 alloc_large_system_hash("Dentry cache",
3089 sizeof(struct hlist_bl_head),
3097 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3098 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3101 /* SLAB cache for __getname() consumers */
3102 struct kmem_cache *names_cachep __read_mostly;
3103 EXPORT_SYMBOL(names_cachep);
3105 EXPORT_SYMBOL(d_genocide);
3107 void __init vfs_caches_init_early(void)
3109 dcache_init_early();
3113 void __init vfs_caches_init(unsigned long mempages)
3115 unsigned long reserve;
3117 /* Base hash sizes on available memory, with a reserve equal to
3118 150% of current kernel size */
3120 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3121 mempages -= reserve;
3123 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3124 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3128 files_init(mempages);