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/export.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 / L1_CACHE_BYTES;
111 hash = hash + (hash >> 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 #ifdef CONFIG_DCACHE_WORD_ACCESS
146 #include <asm/word-at-a-time.h>
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
156 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
158 unsigned long a,b,mask;
161 a = *(unsigned long *)cs;
162 b = load_unaligned_zeropad(ct);
163 if (tcount < sizeof(unsigned long))
165 if (unlikely(a != b))
167 cs += sizeof(unsigned long);
168 ct += sizeof(unsigned long);
169 tcount -= sizeof(unsigned long);
173 mask = ~(~0ul << tcount*8);
174 return unlikely(!!((a ^ b) & mask));
179 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
193 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
196 * Be careful about RCU walk racing with rename:
197 * use ACCESS_ONCE to fetch the name pointer.
199 * NOTE! Even if a rename will mean that the length
200 * was not loaded atomically, we don't care. The
201 * RCU walk will check the sequence count eventually,
202 * and catch it. And we won't overrun the buffer,
203 * because we're reading the name pointer atomically,
204 * and a dentry name is guaranteed to be properly
205 * terminated with a NUL byte.
207 * End result: even if 'len' is wrong, we'll exit
208 * early because the data cannot match (there can
209 * be no NUL in the ct/tcount data)
211 return dentry_string_cmp(ACCESS_ONCE(dentry->d_name.name), ct, tcount);
214 static void __d_free(struct rcu_head *head)
216 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
218 WARN_ON(!list_empty(&dentry->d_alias));
219 if (dname_external(dentry))
220 kfree(dentry->d_name.name);
221 kmem_cache_free(dentry_cache, dentry);
227 static void d_free(struct dentry *dentry)
229 BUG_ON(dentry->d_count);
230 this_cpu_dec(nr_dentry);
231 if (dentry->d_op && dentry->d_op->d_release)
232 dentry->d_op->d_release(dentry);
234 /* if dentry was never visible to RCU, immediate free is OK */
235 if (!(dentry->d_flags & DCACHE_RCUACCESS))
236 __d_free(&dentry->d_u.d_rcu);
238 call_rcu(&dentry->d_u.d_rcu, __d_free);
242 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
243 * @dentry: the target dentry
244 * After this call, in-progress rcu-walk path lookup will fail. This
245 * should be called after unhashing, and after changing d_inode (if
246 * the dentry has not already been unhashed).
248 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
250 assert_spin_locked(&dentry->d_lock);
251 /* Go through a barrier */
252 write_seqcount_barrier(&dentry->d_seq);
256 * Release the dentry's inode, using the filesystem
257 * d_iput() operation if defined. Dentry has no refcount
260 static void dentry_iput(struct dentry * dentry)
261 __releases(dentry->d_lock)
262 __releases(dentry->d_inode->i_lock)
264 struct inode *inode = dentry->d_inode;
266 dentry->d_inode = NULL;
267 list_del_init(&dentry->d_alias);
268 spin_unlock(&dentry->d_lock);
269 spin_unlock(&inode->i_lock);
271 fsnotify_inoderemove(inode);
272 if (dentry->d_op && dentry->d_op->d_iput)
273 dentry->d_op->d_iput(dentry, inode);
277 spin_unlock(&dentry->d_lock);
282 * Release the dentry's inode, using the filesystem
283 * d_iput() operation if defined. dentry remains in-use.
285 static void dentry_unlink_inode(struct dentry * dentry)
286 __releases(dentry->d_lock)
287 __releases(dentry->d_inode->i_lock)
289 struct inode *inode = dentry->d_inode;
290 dentry->d_inode = NULL;
291 list_del_init(&dentry->d_alias);
292 dentry_rcuwalk_barrier(dentry);
293 spin_unlock(&dentry->d_lock);
294 spin_unlock(&inode->i_lock);
296 fsnotify_inoderemove(inode);
297 if (dentry->d_op && dentry->d_op->d_iput)
298 dentry->d_op->d_iput(dentry, inode);
304 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
306 static void dentry_lru_add(struct dentry *dentry)
308 if (list_empty(&dentry->d_lru)) {
309 spin_lock(&dcache_lru_lock);
310 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
311 dentry->d_sb->s_nr_dentry_unused++;
312 dentry_stat.nr_unused++;
313 spin_unlock(&dcache_lru_lock);
317 static void __dentry_lru_del(struct dentry *dentry)
319 list_del_init(&dentry->d_lru);
320 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
321 dentry->d_sb->s_nr_dentry_unused--;
322 dentry_stat.nr_unused--;
326 * Remove a dentry with references from the LRU.
328 static void dentry_lru_del(struct dentry *dentry)
330 if (!list_empty(&dentry->d_lru)) {
331 spin_lock(&dcache_lru_lock);
332 __dentry_lru_del(dentry);
333 spin_unlock(&dcache_lru_lock);
338 * Remove a dentry that is unreferenced and about to be pruned
339 * (unhashed and destroyed) from the LRU, and inform the file system.
340 * This wrapper should be called _prior_ to unhashing a victim dentry.
342 static void dentry_lru_prune(struct dentry *dentry)
344 if (!list_empty(&dentry->d_lru)) {
345 if (dentry->d_flags & DCACHE_OP_PRUNE)
346 dentry->d_op->d_prune(dentry);
348 spin_lock(&dcache_lru_lock);
349 __dentry_lru_del(dentry);
350 spin_unlock(&dcache_lru_lock);
354 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
356 spin_lock(&dcache_lru_lock);
357 if (list_empty(&dentry->d_lru)) {
358 list_add_tail(&dentry->d_lru, list);
359 dentry->d_sb->s_nr_dentry_unused++;
360 dentry_stat.nr_unused++;
362 list_move_tail(&dentry->d_lru, list);
364 spin_unlock(&dcache_lru_lock);
368 * d_kill - kill dentry and return parent
369 * @dentry: dentry to kill
370 * @parent: parent dentry
372 * The dentry must already be unhashed and removed from the LRU.
374 * If this is the root of the dentry tree, return NULL.
376 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
379 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
380 __releases(dentry->d_lock)
381 __releases(parent->d_lock)
382 __releases(dentry->d_inode->i_lock)
384 list_del(&dentry->d_u.d_child);
386 * Inform try_to_ascend() that we are no longer attached to the
389 dentry->d_flags |= DCACHE_DISCONNECTED;
391 spin_unlock(&parent->d_lock);
394 * dentry_iput drops the locks, at which point nobody (except
395 * transient RCU lookups) can reach this dentry.
402 * Unhash a dentry without inserting an RCU walk barrier or checking that
403 * dentry->d_lock is locked. The caller must take care of that, if
406 static void __d_shrink(struct dentry *dentry)
408 if (!d_unhashed(dentry)) {
409 struct hlist_bl_head *b;
410 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
411 b = &dentry->d_sb->s_anon;
413 b = d_hash(dentry->d_parent, dentry->d_name.hash);
416 __hlist_bl_del(&dentry->d_hash);
417 dentry->d_hash.pprev = NULL;
423 * d_drop - drop a dentry
424 * @dentry: dentry to drop
426 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
427 * be found through a VFS lookup any more. Note that this is different from
428 * deleting the dentry - d_delete will try to mark the dentry negative if
429 * possible, giving a successful _negative_ lookup, while d_drop will
430 * just make the cache lookup fail.
432 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
433 * reason (NFS timeouts or autofs deletes).
435 * __d_drop requires dentry->d_lock.
437 void __d_drop(struct dentry *dentry)
439 if (!d_unhashed(dentry)) {
441 dentry_rcuwalk_barrier(dentry);
444 EXPORT_SYMBOL(__d_drop);
446 void d_drop(struct dentry *dentry)
448 spin_lock(&dentry->d_lock);
450 spin_unlock(&dentry->d_lock);
452 EXPORT_SYMBOL(d_drop);
455 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
456 * @dentry: dentry to drop
458 * This is called when we do a lookup on a placeholder dentry that needed to be
459 * looked up. The dentry should have been hashed in order for it to be found by
460 * the lookup code, but now needs to be unhashed while we do the actual lookup
461 * and clear the DCACHE_NEED_LOOKUP flag.
463 void d_clear_need_lookup(struct dentry *dentry)
465 spin_lock(&dentry->d_lock);
467 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
468 spin_unlock(&dentry->d_lock);
470 EXPORT_SYMBOL(d_clear_need_lookup);
473 * Finish off a dentry we've decided to kill.
474 * dentry->d_lock must be held, returns with it unlocked.
475 * If ref is non-zero, then decrement the refcount too.
476 * Returns dentry requiring refcount drop, or NULL if we're done.
478 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
479 __releases(dentry->d_lock)
482 struct dentry *parent;
484 inode = dentry->d_inode;
485 if (inode && !spin_trylock(&inode->i_lock)) {
487 spin_unlock(&dentry->d_lock);
489 return dentry; /* try again with same dentry */
494 parent = dentry->d_parent;
495 if (parent && !spin_trylock(&parent->d_lock)) {
497 spin_unlock(&inode->i_lock);
504 * if dentry was on the d_lru list delete it from there.
505 * inform the fs via d_prune that this dentry is about to be
506 * unhashed and destroyed.
508 dentry_lru_prune(dentry);
509 /* if it was on the hash then remove it */
511 return d_kill(dentry, parent);
517 * This is complicated by the fact that we do not want to put
518 * dentries that are no longer on any hash chain on the unused
519 * list: we'd much rather just get rid of them immediately.
521 * However, that implies that we have to traverse the dentry
522 * tree upwards to the parents which might _also_ now be
523 * scheduled for deletion (it may have been only waiting for
524 * its last child to go away).
526 * This tail recursion is done by hand as we don't want to depend
527 * on the compiler to always get this right (gcc generally doesn't).
528 * Real recursion would eat up our stack space.
532 * dput - release a dentry
533 * @dentry: dentry to release
535 * Release a dentry. This will drop the usage count and if appropriate
536 * call the dentry unlink method as well as removing it from the queues and
537 * releasing its resources. If the parent dentries were scheduled for release
538 * they too may now get deleted.
540 void dput(struct dentry *dentry)
546 if (dentry->d_count == 1)
548 spin_lock(&dentry->d_lock);
549 BUG_ON(!dentry->d_count);
550 if (dentry->d_count > 1) {
552 spin_unlock(&dentry->d_lock);
556 if (dentry->d_flags & DCACHE_OP_DELETE) {
557 if (dentry->d_op->d_delete(dentry))
561 /* Unreachable? Get rid of it */
562 if (d_unhashed(dentry))
566 * If this dentry needs lookup, don't set the referenced flag so that it
567 * is more likely to be cleaned up by the dcache shrinker in case of
570 if (!d_need_lookup(dentry))
571 dentry->d_flags |= DCACHE_REFERENCED;
572 dentry_lru_add(dentry);
575 spin_unlock(&dentry->d_lock);
579 dentry = dentry_kill(dentry, 1);
586 * d_invalidate - invalidate a dentry
587 * @dentry: dentry to invalidate
589 * Try to invalidate the dentry if it turns out to be
590 * possible. If there are other dentries that can be
591 * reached through this one we can't delete it and we
592 * return -EBUSY. On success we return 0.
597 int d_invalidate(struct dentry * dentry)
600 * If it's already been dropped, return OK.
602 spin_lock(&dentry->d_lock);
603 if (d_unhashed(dentry)) {
604 spin_unlock(&dentry->d_lock);
608 * Check whether to do a partial shrink_dcache
609 * to get rid of unused child entries.
611 if (!list_empty(&dentry->d_subdirs)) {
612 spin_unlock(&dentry->d_lock);
613 shrink_dcache_parent(dentry);
614 spin_lock(&dentry->d_lock);
618 * Somebody else still using it?
620 * If it's a directory, we can't drop it
621 * for fear of somebody re-populating it
622 * with children (even though dropping it
623 * would make it unreachable from the root,
624 * we might still populate it if it was a
625 * working directory or similar).
626 * We also need to leave mountpoints alone,
629 if (dentry->d_count > 1 && dentry->d_inode) {
630 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
631 spin_unlock(&dentry->d_lock);
637 spin_unlock(&dentry->d_lock);
640 EXPORT_SYMBOL(d_invalidate);
642 /* This must be called with d_lock held */
643 static inline void __dget_dlock(struct dentry *dentry)
648 static inline void __dget(struct dentry *dentry)
650 spin_lock(&dentry->d_lock);
651 __dget_dlock(dentry);
652 spin_unlock(&dentry->d_lock);
655 struct dentry *dget_parent(struct dentry *dentry)
661 * Don't need rcu_dereference because we re-check it was correct under
665 ret = dentry->d_parent;
666 spin_lock(&ret->d_lock);
667 if (unlikely(ret != dentry->d_parent)) {
668 spin_unlock(&ret->d_lock);
673 BUG_ON(!ret->d_count);
675 spin_unlock(&ret->d_lock);
678 EXPORT_SYMBOL(dget_parent);
681 * d_find_alias - grab a hashed alias of inode
682 * @inode: inode in question
683 * @want_discon: flag, used by d_splice_alias, to request
684 * that only a DISCONNECTED alias be returned.
686 * If inode has a hashed alias, or is a directory and has any alias,
687 * acquire the reference to alias and return it. Otherwise return NULL.
688 * Notice that if inode is a directory there can be only one alias and
689 * it can be unhashed only if it has no children, or if it is the root
692 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
693 * any other hashed alias over that one unless @want_discon is set,
694 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
696 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
698 struct dentry *alias, *discon_alias;
702 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
703 spin_lock(&alias->d_lock);
704 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
705 if (IS_ROOT(alias) &&
706 (alias->d_flags & DCACHE_DISCONNECTED)) {
707 discon_alias = alias;
708 } else if (!want_discon) {
710 spin_unlock(&alias->d_lock);
714 spin_unlock(&alias->d_lock);
717 alias = discon_alias;
718 spin_lock(&alias->d_lock);
719 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
720 if (IS_ROOT(alias) &&
721 (alias->d_flags & DCACHE_DISCONNECTED)) {
723 spin_unlock(&alias->d_lock);
727 spin_unlock(&alias->d_lock);
733 struct dentry *d_find_alias(struct inode *inode)
735 struct dentry *de = NULL;
737 if (!list_empty(&inode->i_dentry)) {
738 spin_lock(&inode->i_lock);
739 de = __d_find_alias(inode, 0);
740 spin_unlock(&inode->i_lock);
744 EXPORT_SYMBOL(d_find_alias);
747 * Try to kill dentries associated with this inode.
748 * WARNING: you must own a reference to inode.
750 void d_prune_aliases(struct inode *inode)
752 struct dentry *dentry;
754 spin_lock(&inode->i_lock);
755 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
756 spin_lock(&dentry->d_lock);
757 if (!dentry->d_count) {
758 __dget_dlock(dentry);
760 spin_unlock(&dentry->d_lock);
761 spin_unlock(&inode->i_lock);
765 spin_unlock(&dentry->d_lock);
767 spin_unlock(&inode->i_lock);
769 EXPORT_SYMBOL(d_prune_aliases);
772 * Try to throw away a dentry - free the inode, dput the parent.
773 * Requires dentry->d_lock is held, and dentry->d_count == 0.
774 * Releases dentry->d_lock.
776 * This may fail if locks cannot be acquired no problem, just try again.
778 static void try_prune_one_dentry(struct dentry *dentry)
779 __releases(dentry->d_lock)
781 struct dentry *parent;
783 parent = dentry_kill(dentry, 0);
785 * If dentry_kill returns NULL, we have nothing more to do.
786 * if it returns the same dentry, trylocks failed. In either
787 * case, just loop again.
789 * Otherwise, we need to prune ancestors too. This is necessary
790 * to prevent quadratic behavior of shrink_dcache_parent(), but
791 * is also expected to be beneficial in reducing dentry cache
796 if (parent == dentry)
799 /* Prune ancestors. */
802 spin_lock(&dentry->d_lock);
803 if (dentry->d_count > 1) {
805 spin_unlock(&dentry->d_lock);
808 dentry = dentry_kill(dentry, 1);
812 static void shrink_dentry_list(struct list_head *list)
814 struct dentry *dentry;
818 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
819 if (&dentry->d_lru == list)
821 spin_lock(&dentry->d_lock);
822 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
823 spin_unlock(&dentry->d_lock);
828 * We found an inuse dentry which was not removed from
829 * the LRU because of laziness during lookup. Do not free
830 * it - just keep it off the LRU list.
832 if (dentry->d_count) {
833 dentry_lru_del(dentry);
834 spin_unlock(&dentry->d_lock);
840 try_prune_one_dentry(dentry);
848 * prune_dcache_sb - shrink the dcache
850 * @count: number of entries to try to free
852 * Attempt to shrink the superblock dcache LRU by @count entries. This is
853 * done when we need more memory an called from the superblock shrinker
856 * This function may fail to free any resources if all the dentries are in
859 void prune_dcache_sb(struct super_block *sb, int count)
861 struct dentry *dentry;
862 LIST_HEAD(referenced);
866 spin_lock(&dcache_lru_lock);
867 while (!list_empty(&sb->s_dentry_lru)) {
868 dentry = list_entry(sb->s_dentry_lru.prev,
869 struct dentry, d_lru);
870 BUG_ON(dentry->d_sb != sb);
872 if (!spin_trylock(&dentry->d_lock)) {
873 spin_unlock(&dcache_lru_lock);
878 if (dentry->d_flags & DCACHE_REFERENCED) {
879 dentry->d_flags &= ~DCACHE_REFERENCED;
880 list_move(&dentry->d_lru, &referenced);
881 spin_unlock(&dentry->d_lock);
883 list_move_tail(&dentry->d_lru, &tmp);
884 dentry->d_flags |= DCACHE_SHRINK_LIST;
885 spin_unlock(&dentry->d_lock);
889 cond_resched_lock(&dcache_lru_lock);
891 if (!list_empty(&referenced))
892 list_splice(&referenced, &sb->s_dentry_lru);
893 spin_unlock(&dcache_lru_lock);
895 shrink_dentry_list(&tmp);
899 * shrink_dcache_sb - shrink dcache for a superblock
902 * Shrink the dcache for the specified super block. This is used to free
903 * the dcache before unmounting a file system.
905 void shrink_dcache_sb(struct super_block *sb)
909 spin_lock(&dcache_lru_lock);
910 while (!list_empty(&sb->s_dentry_lru)) {
911 list_splice_init(&sb->s_dentry_lru, &tmp);
912 spin_unlock(&dcache_lru_lock);
913 shrink_dentry_list(&tmp);
914 spin_lock(&dcache_lru_lock);
916 spin_unlock(&dcache_lru_lock);
918 EXPORT_SYMBOL(shrink_dcache_sb);
921 * destroy a single subtree of dentries for unmount
922 * - see the comments on shrink_dcache_for_umount() for a description of the
925 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
927 struct dentry *parent;
929 BUG_ON(!IS_ROOT(dentry));
932 /* descend to the first leaf in the current subtree */
933 while (!list_empty(&dentry->d_subdirs))
934 dentry = list_entry(dentry->d_subdirs.next,
935 struct dentry, d_u.d_child);
937 /* consume the dentries from this leaf up through its parents
938 * until we find one with children or run out altogether */
943 * remove the dentry from the lru, and inform
944 * the fs that this dentry is about to be
945 * unhashed and destroyed.
947 dentry_lru_prune(dentry);
950 if (dentry->d_count != 0) {
952 "BUG: Dentry %p{i=%lx,n=%s}"
954 " [unmount of %s %s]\n",
957 dentry->d_inode->i_ino : 0UL,
960 dentry->d_sb->s_type->name,
965 if (IS_ROOT(dentry)) {
967 list_del(&dentry->d_u.d_child);
969 parent = dentry->d_parent;
971 list_del(&dentry->d_u.d_child);
974 inode = dentry->d_inode;
976 dentry->d_inode = NULL;
977 list_del_init(&dentry->d_alias);
978 if (dentry->d_op && dentry->d_op->d_iput)
979 dentry->d_op->d_iput(dentry, inode);
986 /* finished when we fall off the top of the tree,
987 * otherwise we ascend to the parent and move to the
988 * next sibling if there is one */
992 } while (list_empty(&dentry->d_subdirs));
994 dentry = list_entry(dentry->d_subdirs.next,
995 struct dentry, d_u.d_child);
1000 * destroy the dentries attached to a superblock on unmounting
1001 * - we don't need to use dentry->d_lock because:
1002 * - the superblock is detached from all mountings and open files, so the
1003 * dentry trees will not be rearranged by the VFS
1004 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1005 * any dentries belonging to this superblock that it comes across
1006 * - the filesystem itself is no longer permitted to rearrange the dentries
1007 * in this superblock
1009 void shrink_dcache_for_umount(struct super_block *sb)
1011 struct dentry *dentry;
1013 if (down_read_trylock(&sb->s_umount))
1016 dentry = sb->s_root;
1019 shrink_dcache_for_umount_subtree(dentry);
1021 while (!hlist_bl_empty(&sb->s_anon)) {
1022 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1023 shrink_dcache_for_umount_subtree(dentry);
1028 * This tries to ascend one level of parenthood, but
1029 * we can race with renaming, so we need to re-check
1030 * the parenthood after dropping the lock and check
1031 * that the sequence number still matches.
1033 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1035 struct dentry *new = old->d_parent;
1038 spin_unlock(&old->d_lock);
1039 spin_lock(&new->d_lock);
1042 * might go back up the wrong parent if we have had a rename
1045 if (new != old->d_parent ||
1046 (old->d_flags & DCACHE_DISCONNECTED) ||
1047 (!locked && read_seqretry(&rename_lock, seq))) {
1048 spin_unlock(&new->d_lock);
1057 * Search for at least 1 mount point in the dentry's subdirs.
1058 * We descend to the next level whenever the d_subdirs
1059 * list is non-empty and continue searching.
1063 * have_submounts - check for mounts over a dentry
1064 * @parent: dentry to check.
1066 * Return true if the parent or its subdirectories contain
1069 int have_submounts(struct dentry *parent)
1071 struct dentry *this_parent;
1072 struct list_head *next;
1076 seq = read_seqbegin(&rename_lock);
1078 this_parent = parent;
1080 if (d_mountpoint(parent))
1082 spin_lock(&this_parent->d_lock);
1084 next = this_parent->d_subdirs.next;
1086 while (next != &this_parent->d_subdirs) {
1087 struct list_head *tmp = next;
1088 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1091 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1092 /* Have we found a mount point ? */
1093 if (d_mountpoint(dentry)) {
1094 spin_unlock(&dentry->d_lock);
1095 spin_unlock(&this_parent->d_lock);
1098 if (!list_empty(&dentry->d_subdirs)) {
1099 spin_unlock(&this_parent->d_lock);
1100 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1101 this_parent = dentry;
1102 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1105 spin_unlock(&dentry->d_lock);
1108 * All done at this level ... ascend and resume the search.
1110 if (this_parent != parent) {
1111 struct dentry *child = this_parent;
1112 this_parent = try_to_ascend(this_parent, locked, seq);
1115 next = child->d_u.d_child.next;
1118 spin_unlock(&this_parent->d_lock);
1119 if (!locked && read_seqretry(&rename_lock, seq))
1122 write_sequnlock(&rename_lock);
1123 return 0; /* No mount points found in tree */
1125 if (!locked && read_seqretry(&rename_lock, seq))
1128 write_sequnlock(&rename_lock);
1133 write_seqlock(&rename_lock);
1136 EXPORT_SYMBOL(have_submounts);
1139 * Search the dentry child list for the specified parent,
1140 * and move any unused dentries to the end of the unused
1141 * list for prune_dcache(). We descend to the next level
1142 * whenever the d_subdirs list is non-empty and continue
1145 * It returns zero iff there are no unused children,
1146 * otherwise it returns the number of children moved to
1147 * the end of the unused list. This may not be the total
1148 * number of unused children, because select_parent can
1149 * drop the lock and return early due to latency
1152 static int select_parent(struct dentry *parent, struct list_head *dispose)
1154 struct dentry *this_parent;
1155 struct list_head *next;
1160 seq = read_seqbegin(&rename_lock);
1162 this_parent = parent;
1163 spin_lock(&this_parent->d_lock);
1165 next = this_parent->d_subdirs.next;
1167 while (next != &this_parent->d_subdirs) {
1168 struct list_head *tmp = next;
1169 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1172 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1175 * move only zero ref count dentries to the dispose list.
1177 * Those which are presently on the shrink list, being processed
1178 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1179 * loop in shrink_dcache_parent() might not make any progress
1182 if (dentry->d_count) {
1183 dentry_lru_del(dentry);
1184 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1185 dentry_lru_move_list(dentry, dispose);
1186 dentry->d_flags |= DCACHE_SHRINK_LIST;
1190 * We can return to the caller if we have found some (this
1191 * ensures forward progress). We'll be coming back to find
1194 if (found && need_resched()) {
1195 spin_unlock(&dentry->d_lock);
1200 * Descend a level if the d_subdirs list is non-empty.
1202 if (!list_empty(&dentry->d_subdirs)) {
1203 spin_unlock(&this_parent->d_lock);
1204 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1205 this_parent = dentry;
1206 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1210 spin_unlock(&dentry->d_lock);
1213 * All done at this level ... ascend and resume the search.
1215 if (this_parent != parent) {
1216 struct dentry *child = this_parent;
1217 this_parent = try_to_ascend(this_parent, locked, seq);
1220 next = child->d_u.d_child.next;
1224 spin_unlock(&this_parent->d_lock);
1225 if (!locked && read_seqretry(&rename_lock, seq))
1228 write_sequnlock(&rename_lock);
1235 write_seqlock(&rename_lock);
1240 * shrink_dcache_parent - prune dcache
1241 * @parent: parent of entries to prune
1243 * Prune the dcache to remove unused children of the parent dentry.
1245 void shrink_dcache_parent(struct dentry * parent)
1250 while ((found = select_parent(parent, &dispose)) != 0)
1251 shrink_dentry_list(&dispose);
1253 EXPORT_SYMBOL(shrink_dcache_parent);
1256 * __d_alloc - allocate a dcache entry
1257 * @sb: filesystem it will belong to
1258 * @name: qstr of the name
1260 * Allocates a dentry. It returns %NULL if there is insufficient memory
1261 * available. On a success the dentry is returned. The name passed in is
1262 * copied and the copy passed in may be reused after this call.
1265 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1267 struct dentry *dentry;
1270 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1274 if (name->len > DNAME_INLINE_LEN-1) {
1275 dname = kmalloc(name->len + 1, GFP_KERNEL);
1277 kmem_cache_free(dentry_cache, dentry);
1281 dname = dentry->d_iname;
1283 dentry->d_name.name = dname;
1285 dentry->d_name.len = name->len;
1286 dentry->d_name.hash = name->hash;
1287 memcpy(dname, name->name, name->len);
1288 dname[name->len] = 0;
1290 dentry->d_count = 1;
1291 dentry->d_flags = 0;
1292 spin_lock_init(&dentry->d_lock);
1293 seqcount_init(&dentry->d_seq);
1294 dentry->d_inode = NULL;
1295 dentry->d_parent = dentry;
1297 dentry->d_op = NULL;
1298 dentry->d_fsdata = NULL;
1299 INIT_HLIST_BL_NODE(&dentry->d_hash);
1300 INIT_LIST_HEAD(&dentry->d_lru);
1301 INIT_LIST_HEAD(&dentry->d_subdirs);
1302 INIT_LIST_HEAD(&dentry->d_alias);
1303 INIT_LIST_HEAD(&dentry->d_u.d_child);
1304 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1306 this_cpu_inc(nr_dentry);
1312 * d_alloc - allocate a dcache entry
1313 * @parent: parent of entry to allocate
1314 * @name: qstr of the name
1316 * Allocates a dentry. It returns %NULL if there is insufficient memory
1317 * available. On a success the dentry is returned. The name passed in is
1318 * copied and the copy passed in may be reused after this call.
1320 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1322 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1326 spin_lock(&parent->d_lock);
1328 * don't need child lock because it is not subject
1329 * to concurrency here
1331 __dget_dlock(parent);
1332 dentry->d_parent = parent;
1333 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1334 spin_unlock(&parent->d_lock);
1338 EXPORT_SYMBOL(d_alloc);
1340 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1342 struct dentry *dentry = __d_alloc(sb, name);
1344 dentry->d_flags |= DCACHE_DISCONNECTED;
1347 EXPORT_SYMBOL(d_alloc_pseudo);
1349 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1354 q.len = strlen(name);
1355 q.hash = full_name_hash(q.name, q.len);
1356 return d_alloc(parent, &q);
1358 EXPORT_SYMBOL(d_alloc_name);
1360 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1362 WARN_ON_ONCE(dentry->d_op);
1363 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1365 DCACHE_OP_REVALIDATE |
1366 DCACHE_OP_DELETE ));
1371 dentry->d_flags |= DCACHE_OP_HASH;
1373 dentry->d_flags |= DCACHE_OP_COMPARE;
1374 if (op->d_revalidate)
1375 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1377 dentry->d_flags |= DCACHE_OP_DELETE;
1379 dentry->d_flags |= DCACHE_OP_PRUNE;
1382 EXPORT_SYMBOL(d_set_d_op);
1384 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1386 spin_lock(&dentry->d_lock);
1388 if (unlikely(IS_AUTOMOUNT(inode)))
1389 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1390 list_add(&dentry->d_alias, &inode->i_dentry);
1392 dentry->d_inode = inode;
1393 dentry_rcuwalk_barrier(dentry);
1394 spin_unlock(&dentry->d_lock);
1395 fsnotify_d_instantiate(dentry, inode);
1399 * d_instantiate - fill in inode information for a dentry
1400 * @entry: dentry to complete
1401 * @inode: inode to attach to this dentry
1403 * Fill in inode information in the entry.
1405 * This turns negative dentries into productive full members
1408 * NOTE! This assumes that the inode count has been incremented
1409 * (or otherwise set) by the caller to indicate that it is now
1410 * in use by the dcache.
1413 void d_instantiate(struct dentry *entry, struct inode * inode)
1415 BUG_ON(!list_empty(&entry->d_alias));
1417 spin_lock(&inode->i_lock);
1418 __d_instantiate(entry, inode);
1420 spin_unlock(&inode->i_lock);
1421 security_d_instantiate(entry, inode);
1423 EXPORT_SYMBOL(d_instantiate);
1426 * d_instantiate_unique - instantiate a non-aliased dentry
1427 * @entry: dentry to instantiate
1428 * @inode: inode to attach to this dentry
1430 * Fill in inode information in the entry. On success, it returns NULL.
1431 * If an unhashed alias of "entry" already exists, then we return the
1432 * aliased dentry instead and drop one reference to inode.
1434 * Note that in order to avoid conflicts with rename() etc, the caller
1435 * had better be holding the parent directory semaphore.
1437 * This also assumes that the inode count has been incremented
1438 * (or otherwise set) by the caller to indicate that it is now
1439 * in use by the dcache.
1441 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1442 struct inode *inode)
1444 struct dentry *alias;
1445 int len = entry->d_name.len;
1446 const char *name = entry->d_name.name;
1447 unsigned int hash = entry->d_name.hash;
1450 __d_instantiate(entry, NULL);
1454 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1456 * Don't need alias->d_lock here, because aliases with
1457 * d_parent == entry->d_parent are not subject to name or
1458 * parent changes, because the parent inode i_mutex is held.
1460 if (alias->d_name.hash != hash)
1462 if (alias->d_parent != entry->d_parent)
1464 if (alias->d_name.len != len)
1466 if (dentry_cmp(alias, name, len))
1472 __d_instantiate(entry, inode);
1476 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1478 struct dentry *result;
1480 BUG_ON(!list_empty(&entry->d_alias));
1483 spin_lock(&inode->i_lock);
1484 result = __d_instantiate_unique(entry, inode);
1486 spin_unlock(&inode->i_lock);
1489 security_d_instantiate(entry, inode);
1493 BUG_ON(!d_unhashed(result));
1498 EXPORT_SYMBOL(d_instantiate_unique);
1500 struct dentry *d_make_root(struct inode *root_inode)
1502 struct dentry *res = NULL;
1505 static const struct qstr name = QSTR_INIT("/", 1);
1507 res = __d_alloc(root_inode->i_sb, &name);
1509 d_instantiate(res, root_inode);
1515 EXPORT_SYMBOL(d_make_root);
1517 static struct dentry * __d_find_any_alias(struct inode *inode)
1519 struct dentry *alias;
1521 if (list_empty(&inode->i_dentry))
1523 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1529 * d_find_any_alias - find any alias for a given inode
1530 * @inode: inode to find an alias for
1532 * If any aliases exist for the given inode, take and return a
1533 * reference for one of them. If no aliases exist, return %NULL.
1535 struct dentry *d_find_any_alias(struct inode *inode)
1539 spin_lock(&inode->i_lock);
1540 de = __d_find_any_alias(inode);
1541 spin_unlock(&inode->i_lock);
1544 EXPORT_SYMBOL(d_find_any_alias);
1547 * d_obtain_alias - find or allocate a dentry for a given inode
1548 * @inode: inode to allocate the dentry for
1550 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1551 * similar open by handle operations. The returned dentry may be anonymous,
1552 * or may have a full name (if the inode was already in the cache).
1554 * When called on a directory inode, we must ensure that the inode only ever
1555 * has one dentry. If a dentry is found, that is returned instead of
1556 * allocating a new one.
1558 * On successful return, the reference to the inode has been transferred
1559 * to the dentry. In case of an error the reference on the inode is released.
1560 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1561 * be passed in and will be the error will be propagate to the return value,
1562 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1564 struct dentry *d_obtain_alias(struct inode *inode)
1566 static const struct qstr anonstring = { .name = "" };
1571 return ERR_PTR(-ESTALE);
1573 return ERR_CAST(inode);
1575 res = d_find_any_alias(inode);
1579 tmp = __d_alloc(inode->i_sb, &anonstring);
1581 res = ERR_PTR(-ENOMEM);
1585 spin_lock(&inode->i_lock);
1586 res = __d_find_any_alias(inode);
1588 spin_unlock(&inode->i_lock);
1593 /* attach a disconnected dentry */
1594 spin_lock(&tmp->d_lock);
1595 tmp->d_inode = inode;
1596 tmp->d_flags |= DCACHE_DISCONNECTED;
1597 list_add(&tmp->d_alias, &inode->i_dentry);
1598 hlist_bl_lock(&tmp->d_sb->s_anon);
1599 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1600 hlist_bl_unlock(&tmp->d_sb->s_anon);
1601 spin_unlock(&tmp->d_lock);
1602 spin_unlock(&inode->i_lock);
1603 security_d_instantiate(tmp, inode);
1608 if (res && !IS_ERR(res))
1609 security_d_instantiate(res, inode);
1613 EXPORT_SYMBOL(d_obtain_alias);
1616 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1617 * @inode: the inode which may have a disconnected dentry
1618 * @dentry: a negative dentry which we want to point to the inode.
1620 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1621 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1622 * and return it, else simply d_add the inode to the dentry and return NULL.
1624 * This is needed in the lookup routine of any filesystem that is exportable
1625 * (via knfsd) so that we can build dcache paths to directories effectively.
1627 * If a dentry was found and moved, then it is returned. Otherwise NULL
1628 * is returned. This matches the expected return value of ->lookup.
1631 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1633 struct dentry *new = NULL;
1636 return ERR_CAST(inode);
1638 if (inode && S_ISDIR(inode->i_mode)) {
1639 spin_lock(&inode->i_lock);
1640 new = __d_find_alias(inode, 1);
1642 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1643 spin_unlock(&inode->i_lock);
1644 security_d_instantiate(new, inode);
1645 d_move(new, dentry);
1648 /* already taking inode->i_lock, so d_add() by hand */
1649 __d_instantiate(dentry, inode);
1650 spin_unlock(&inode->i_lock);
1651 security_d_instantiate(dentry, inode);
1655 d_add(dentry, inode);
1658 EXPORT_SYMBOL(d_splice_alias);
1661 * d_add_ci - lookup or allocate new dentry with case-exact name
1662 * @inode: the inode case-insensitive lookup has found
1663 * @dentry: the negative dentry that was passed to the parent's lookup func
1664 * @name: the case-exact name to be associated with the returned dentry
1666 * This is to avoid filling the dcache with case-insensitive names to the
1667 * same inode, only the actual correct case is stored in the dcache for
1668 * case-insensitive filesystems.
1670 * For a case-insensitive lookup match and if the the case-exact dentry
1671 * already exists in in the dcache, use it and return it.
1673 * If no entry exists with the exact case name, allocate new dentry with
1674 * the exact case, and return the spliced entry.
1676 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1680 struct dentry *found;
1684 * First check if a dentry matching the name already exists,
1685 * if not go ahead and create it now.
1687 found = d_hash_and_lookup(dentry->d_parent, name);
1689 new = d_alloc(dentry->d_parent, name);
1695 found = d_splice_alias(inode, new);
1704 * If a matching dentry exists, and it's not negative use it.
1706 * Decrement the reference count to balance the iget() done
1709 if (found->d_inode) {
1710 if (unlikely(found->d_inode != inode)) {
1711 /* This can't happen because bad inodes are unhashed. */
1712 BUG_ON(!is_bad_inode(inode));
1713 BUG_ON(!is_bad_inode(found->d_inode));
1720 * We are going to instantiate this dentry, unhash it and clear the
1721 * lookup flag so we can do that.
1723 if (unlikely(d_need_lookup(found)))
1724 d_clear_need_lookup(found);
1727 * Negative dentry: instantiate it unless the inode is a directory and
1728 * already has a dentry.
1730 new = d_splice_alias(inode, found);
1739 return ERR_PTR(error);
1741 EXPORT_SYMBOL(d_add_ci);
1744 * Do the slow-case of the dentry name compare.
1746 * Unlike the dentry_cmp() function, we need to atomically
1747 * load the name, length and inode information, so that the
1748 * filesystem can rely on them, and can use the 'name' and
1749 * 'len' information without worrying about walking off the
1750 * end of memory etc.
1752 * Thus the read_seqcount_retry() and the "duplicate" info
1753 * in arguments (the low-level filesystem should not look
1754 * at the dentry inode or name contents directly, since
1755 * rename can change them while we're in RCU mode).
1757 enum slow_d_compare {
1763 static noinline enum slow_d_compare slow_dentry_cmp(
1764 const struct dentry *parent,
1765 struct inode *inode,
1766 struct dentry *dentry,
1768 const struct qstr *name)
1770 int tlen = dentry->d_name.len;
1771 const char *tname = dentry->d_name.name;
1772 struct inode *i = dentry->d_inode;
1774 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1776 return D_COMP_SEQRETRY;
1778 if (parent->d_op->d_compare(parent, inode,
1781 return D_COMP_NOMATCH;
1786 * __d_lookup_rcu - search for a dentry (racy, store-free)
1787 * @parent: parent dentry
1788 * @name: qstr of name we wish to find
1789 * @seqp: returns d_seq value at the point where the dentry was found
1790 * @inode: returns dentry->d_inode when the inode was found valid.
1791 * Returns: dentry, or NULL
1793 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1794 * resolution (store-free path walking) design described in
1795 * Documentation/filesystems/path-lookup.txt.
1797 * This is not to be used outside core vfs.
1799 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1800 * held, and rcu_read_lock held. The returned dentry must not be stored into
1801 * without taking d_lock and checking d_seq sequence count against @seq
1804 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1807 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1808 * the returned dentry, so long as its parent's seqlock is checked after the
1809 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1810 * is formed, giving integrity down the path walk.
1812 * NOTE! The caller *has* to check the resulting dentry against the sequence
1813 * number we've returned before using any of the resulting dentry state!
1815 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1816 const struct qstr *name,
1817 unsigned *seqp, struct inode *inode)
1819 u64 hashlen = name->hash_len;
1820 const unsigned char *str = name->name;
1821 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1822 struct hlist_bl_node *node;
1823 struct dentry *dentry;
1826 * Note: There is significant duplication with __d_lookup_rcu which is
1827 * required to prevent single threaded performance regressions
1828 * especially on architectures where smp_rmb (in seqcounts) are costly.
1829 * Keep the two functions in sync.
1833 * The hash list is protected using RCU.
1835 * Carefully use d_seq when comparing a candidate dentry, to avoid
1836 * races with d_move().
1838 * It is possible that concurrent renames can mess up our list
1839 * walk here and result in missing our dentry, resulting in the
1840 * false-negative result. d_lookup() protects against concurrent
1841 * renames using rename_lock seqlock.
1843 * See Documentation/filesystems/path-lookup.txt for more details.
1845 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1850 * The dentry sequence count protects us from concurrent
1851 * renames, and thus protects inode, parent and name fields.
1853 * The caller must perform a seqcount check in order
1854 * to do anything useful with the returned dentry,
1855 * including using the 'd_inode' pointer.
1857 * NOTE! We do a "raw" seqcount_begin here. That means that
1858 * we don't wait for the sequence count to stabilize if it
1859 * is in the middle of a sequence change. If we do the slow
1860 * dentry compare, we will do seqretries until it is stable,
1861 * and if we end up with a successful lookup, we actually
1862 * want to exit RCU lookup anyway.
1864 seq = raw_seqcount_begin(&dentry->d_seq);
1865 if (dentry->d_parent != parent)
1869 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1870 if (dentry->d_name.hash != hashlen_hash(hashlen))
1872 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1875 case D_COMP_NOMATCH:
1882 if (dentry->d_name.hash_len != hashlen)
1884 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
1891 * d_lookup - search for a dentry
1892 * @parent: parent dentry
1893 * @name: qstr of name we wish to find
1894 * Returns: dentry, or NULL
1896 * d_lookup searches the children of the parent dentry for the name in
1897 * question. If the dentry is found its reference count is incremented and the
1898 * dentry is returned. The caller must use dput to free the entry when it has
1899 * finished using it. %NULL is returned if the dentry does not exist.
1901 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1903 struct dentry *dentry;
1907 seq = read_seqbegin(&rename_lock);
1908 dentry = __d_lookup(parent, name);
1911 } while (read_seqretry(&rename_lock, seq));
1914 EXPORT_SYMBOL(d_lookup);
1917 * __d_lookup - search for a dentry (racy)
1918 * @parent: parent dentry
1919 * @name: qstr of name we wish to find
1920 * Returns: dentry, or NULL
1922 * __d_lookup is like d_lookup, however it may (rarely) return a
1923 * false-negative result due to unrelated rename activity.
1925 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1926 * however it must be used carefully, eg. with a following d_lookup in
1927 * the case of failure.
1929 * __d_lookup callers must be commented.
1931 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1933 unsigned int len = name->len;
1934 unsigned int hash = name->hash;
1935 const unsigned char *str = name->name;
1936 struct hlist_bl_head *b = d_hash(parent, hash);
1937 struct hlist_bl_node *node;
1938 struct dentry *found = NULL;
1939 struct dentry *dentry;
1942 * Note: There is significant duplication with __d_lookup_rcu which is
1943 * required to prevent single threaded performance regressions
1944 * especially on architectures where smp_rmb (in seqcounts) are costly.
1945 * Keep the two functions in sync.
1949 * The hash list is protected using RCU.
1951 * Take d_lock when comparing a candidate dentry, to avoid races
1954 * It is possible that concurrent renames can mess up our list
1955 * walk here and result in missing our dentry, resulting in the
1956 * false-negative result. d_lookup() protects against concurrent
1957 * renames using rename_lock seqlock.
1959 * See Documentation/filesystems/path-lookup.txt for more details.
1963 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1965 if (dentry->d_name.hash != hash)
1968 spin_lock(&dentry->d_lock);
1969 if (dentry->d_parent != parent)
1971 if (d_unhashed(dentry))
1975 * It is safe to compare names since d_move() cannot
1976 * change the qstr (protected by d_lock).
1978 if (parent->d_flags & DCACHE_OP_COMPARE) {
1979 int tlen = dentry->d_name.len;
1980 const char *tname = dentry->d_name.name;
1981 if (parent->d_op->d_compare(parent, parent->d_inode,
1982 dentry, dentry->d_inode,
1986 if (dentry->d_name.len != len)
1988 if (dentry_cmp(dentry, str, len))
1994 spin_unlock(&dentry->d_lock);
1997 spin_unlock(&dentry->d_lock);
2005 * d_hash_and_lookup - hash the qstr then search for a dentry
2006 * @dir: Directory to search in
2007 * @name: qstr of name we wish to find
2009 * On hash failure or on lookup failure NULL is returned.
2011 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2013 struct dentry *dentry = NULL;
2016 * Check for a fs-specific hash function. Note that we must
2017 * calculate the standard hash first, as the d_op->d_hash()
2018 * routine may choose to leave the hash value unchanged.
2020 name->hash = full_name_hash(name->name, name->len);
2021 if (dir->d_flags & DCACHE_OP_HASH) {
2022 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
2025 dentry = d_lookup(dir, name);
2031 * d_validate - verify dentry provided from insecure source (deprecated)
2032 * @dentry: The dentry alleged to be valid child of @dparent
2033 * @dparent: The parent dentry (known to be valid)
2035 * An insecure source has sent us a dentry, here we verify it and dget() it.
2036 * This is used by ncpfs in its readdir implementation.
2037 * Zero is returned in the dentry is invalid.
2039 * This function is slow for big directories, and deprecated, do not use it.
2041 int d_validate(struct dentry *dentry, struct dentry *dparent)
2043 struct dentry *child;
2045 spin_lock(&dparent->d_lock);
2046 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2047 if (dentry == child) {
2048 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2049 __dget_dlock(dentry);
2050 spin_unlock(&dentry->d_lock);
2051 spin_unlock(&dparent->d_lock);
2055 spin_unlock(&dparent->d_lock);
2059 EXPORT_SYMBOL(d_validate);
2062 * When a file is deleted, we have two options:
2063 * - turn this dentry into a negative dentry
2064 * - unhash this dentry and free it.
2066 * Usually, we want to just turn this into
2067 * a negative dentry, but if anybody else is
2068 * currently using the dentry or the inode
2069 * we can't do that and we fall back on removing
2070 * it from the hash queues and waiting for
2071 * it to be deleted later when it has no users
2075 * d_delete - delete a dentry
2076 * @dentry: The dentry to delete
2078 * Turn the dentry into a negative dentry if possible, otherwise
2079 * remove it from the hash queues so it can be deleted later
2082 void d_delete(struct dentry * dentry)
2084 struct inode *inode;
2087 * Are we the only user?
2090 spin_lock(&dentry->d_lock);
2091 inode = dentry->d_inode;
2092 isdir = S_ISDIR(inode->i_mode);
2093 if (dentry->d_count == 1) {
2094 if (inode && !spin_trylock(&inode->i_lock)) {
2095 spin_unlock(&dentry->d_lock);
2099 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2100 dentry_unlink_inode(dentry);
2101 fsnotify_nameremove(dentry, isdir);
2105 if (!d_unhashed(dentry))
2108 spin_unlock(&dentry->d_lock);
2110 fsnotify_nameremove(dentry, isdir);
2112 EXPORT_SYMBOL(d_delete);
2114 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2116 BUG_ON(!d_unhashed(entry));
2118 entry->d_flags |= DCACHE_RCUACCESS;
2119 hlist_bl_add_head_rcu(&entry->d_hash, b);
2123 static void _d_rehash(struct dentry * entry)
2125 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2129 * d_rehash - add an entry back to the hash
2130 * @entry: dentry to add to the hash
2132 * Adds a dentry to the hash according to its name.
2135 void d_rehash(struct dentry * entry)
2137 spin_lock(&entry->d_lock);
2139 spin_unlock(&entry->d_lock);
2141 EXPORT_SYMBOL(d_rehash);
2144 * dentry_update_name_case - update case insensitive dentry with a new name
2145 * @dentry: dentry to be updated
2148 * Update a case insensitive dentry with new case of name.
2150 * dentry must have been returned by d_lookup with name @name. Old and new
2151 * name lengths must match (ie. no d_compare which allows mismatched name
2154 * Parent inode i_mutex must be held over d_lookup and into this call (to
2155 * keep renames and concurrent inserts, and readdir(2) away).
2157 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2159 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2160 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2162 spin_lock(&dentry->d_lock);
2163 write_seqcount_begin(&dentry->d_seq);
2164 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2165 write_seqcount_end(&dentry->d_seq);
2166 spin_unlock(&dentry->d_lock);
2168 EXPORT_SYMBOL(dentry_update_name_case);
2170 static void switch_names(struct dentry *dentry, struct dentry *target)
2172 if (dname_external(target)) {
2173 if (dname_external(dentry)) {
2175 * Both external: swap the pointers
2177 swap(target->d_name.name, dentry->d_name.name);
2180 * dentry:internal, target:external. Steal target's
2181 * storage and make target internal.
2183 memcpy(target->d_iname, dentry->d_name.name,
2184 dentry->d_name.len + 1);
2185 dentry->d_name.name = target->d_name.name;
2186 target->d_name.name = target->d_iname;
2189 if (dname_external(dentry)) {
2191 * dentry:external, target:internal. Give dentry's
2192 * storage to target and make dentry internal
2194 memcpy(dentry->d_iname, target->d_name.name,
2195 target->d_name.len + 1);
2196 target->d_name.name = dentry->d_name.name;
2197 dentry->d_name.name = dentry->d_iname;
2200 * Both are internal. Just copy target to dentry
2202 memcpy(dentry->d_iname, target->d_name.name,
2203 target->d_name.len + 1);
2204 dentry->d_name.len = target->d_name.len;
2208 swap(dentry->d_name.len, target->d_name.len);
2211 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2214 * XXXX: do we really need to take target->d_lock?
2216 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2217 spin_lock(&target->d_parent->d_lock);
2219 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2220 spin_lock(&dentry->d_parent->d_lock);
2221 spin_lock_nested(&target->d_parent->d_lock,
2222 DENTRY_D_LOCK_NESTED);
2224 spin_lock(&target->d_parent->d_lock);
2225 spin_lock_nested(&dentry->d_parent->d_lock,
2226 DENTRY_D_LOCK_NESTED);
2229 if (target < dentry) {
2230 spin_lock_nested(&target->d_lock, 2);
2231 spin_lock_nested(&dentry->d_lock, 3);
2233 spin_lock_nested(&dentry->d_lock, 2);
2234 spin_lock_nested(&target->d_lock, 3);
2238 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2239 struct dentry *target)
2241 if (target->d_parent != dentry->d_parent)
2242 spin_unlock(&dentry->d_parent->d_lock);
2243 if (target->d_parent != target)
2244 spin_unlock(&target->d_parent->d_lock);
2248 * When switching names, the actual string doesn't strictly have to
2249 * be preserved in the target - because we're dropping the target
2250 * anyway. As such, we can just do a simple memcpy() to copy over
2251 * the new name before we switch.
2253 * Note that we have to be a lot more careful about getting the hash
2254 * switched - we have to switch the hash value properly even if it
2255 * then no longer matches the actual (corrupted) string of the target.
2256 * The hash value has to match the hash queue that the dentry is on..
2259 * __d_move - move a dentry
2260 * @dentry: entry to move
2261 * @target: new dentry
2263 * Update the dcache to reflect the move of a file name. Negative
2264 * dcache entries should not be moved in this way. Caller must hold
2265 * rename_lock, the i_mutex of the source and target directories,
2266 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2268 static void __d_move(struct dentry * dentry, struct dentry * target)
2270 if (!dentry->d_inode)
2271 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2273 BUG_ON(d_ancestor(dentry, target));
2274 BUG_ON(d_ancestor(target, dentry));
2276 dentry_lock_for_move(dentry, target);
2278 write_seqcount_begin(&dentry->d_seq);
2279 write_seqcount_begin(&target->d_seq);
2281 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2284 * Move the dentry to the target hash queue. Don't bother checking
2285 * for the same hash queue because of how unlikely it is.
2288 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2290 /* Unhash the target: dput() will then get rid of it */
2293 list_del(&dentry->d_u.d_child);
2294 list_del(&target->d_u.d_child);
2296 /* Switch the names.. */
2297 switch_names(dentry, target);
2298 swap(dentry->d_name.hash, target->d_name.hash);
2300 /* ... and switch the parents */
2301 if (IS_ROOT(dentry)) {
2302 dentry->d_parent = target->d_parent;
2303 target->d_parent = target;
2304 INIT_LIST_HEAD(&target->d_u.d_child);
2306 swap(dentry->d_parent, target->d_parent);
2308 /* And add them back to the (new) parent lists */
2309 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2312 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2314 write_seqcount_end(&target->d_seq);
2315 write_seqcount_end(&dentry->d_seq);
2317 dentry_unlock_parents_for_move(dentry, target);
2318 spin_unlock(&target->d_lock);
2319 fsnotify_d_move(dentry);
2320 spin_unlock(&dentry->d_lock);
2324 * d_move - move a dentry
2325 * @dentry: entry to move
2326 * @target: new dentry
2328 * Update the dcache to reflect the move of a file name. Negative
2329 * dcache entries should not be moved in this way. See the locking
2330 * requirements for __d_move.
2332 void d_move(struct dentry *dentry, struct dentry *target)
2334 write_seqlock(&rename_lock);
2335 __d_move(dentry, target);
2336 write_sequnlock(&rename_lock);
2338 EXPORT_SYMBOL(d_move);
2341 * d_ancestor - search for an ancestor
2342 * @p1: ancestor dentry
2345 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2346 * an ancestor of p2, else NULL.
2348 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2352 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2353 if (p->d_parent == p1)
2360 * This helper attempts to cope with remotely renamed directories
2362 * It assumes that the caller is already holding
2363 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2365 * Note: If ever the locking in lock_rename() changes, then please
2366 * remember to update this too...
2368 static struct dentry *__d_unalias(struct inode *inode,
2369 struct dentry *dentry, struct dentry *alias)
2371 struct mutex *m1 = NULL, *m2 = NULL;
2374 /* If alias and dentry share a parent, then no extra locks required */
2375 if (alias->d_parent == dentry->d_parent)
2378 /* See lock_rename() */
2379 ret = ERR_PTR(-EBUSY);
2380 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2382 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2383 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2385 m2 = &alias->d_parent->d_inode->i_mutex;
2387 __d_move(alias, dentry);
2390 spin_unlock(&inode->i_lock);
2399 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2400 * named dentry in place of the dentry to be replaced.
2401 * returns with anon->d_lock held!
2403 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2405 struct dentry *dparent, *aparent;
2407 dentry_lock_for_move(anon, dentry);
2409 write_seqcount_begin(&dentry->d_seq);
2410 write_seqcount_begin(&anon->d_seq);
2412 dparent = dentry->d_parent;
2413 aparent = anon->d_parent;
2415 switch_names(dentry, anon);
2416 swap(dentry->d_name.hash, anon->d_name.hash);
2418 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2419 list_del(&dentry->d_u.d_child);
2420 if (!IS_ROOT(dentry))
2421 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2423 INIT_LIST_HEAD(&dentry->d_u.d_child);
2425 anon->d_parent = (dparent == dentry) ? anon : dparent;
2426 list_del(&anon->d_u.d_child);
2428 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2430 INIT_LIST_HEAD(&anon->d_u.d_child);
2432 write_seqcount_end(&dentry->d_seq);
2433 write_seqcount_end(&anon->d_seq);
2435 dentry_unlock_parents_for_move(anon, dentry);
2436 spin_unlock(&dentry->d_lock);
2438 /* anon->d_lock still locked, returns locked */
2439 anon->d_flags &= ~DCACHE_DISCONNECTED;
2443 * d_materialise_unique - introduce an inode into the tree
2444 * @dentry: candidate dentry
2445 * @inode: inode to bind to the dentry, to which aliases may be attached
2447 * Introduces an dentry into the tree, substituting an extant disconnected
2448 * root directory alias in its place if there is one. Caller must hold the
2449 * i_mutex of the parent directory.
2451 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2453 struct dentry *actual;
2455 BUG_ON(!d_unhashed(dentry));
2459 __d_instantiate(dentry, NULL);
2464 spin_lock(&inode->i_lock);
2466 if (S_ISDIR(inode->i_mode)) {
2467 struct dentry *alias;
2469 /* Does an aliased dentry already exist? */
2470 alias = __d_find_alias(inode, 0);
2473 write_seqlock(&rename_lock);
2475 if (d_ancestor(alias, dentry)) {
2476 /* Check for loops */
2477 actual = ERR_PTR(-ELOOP);
2478 spin_unlock(&inode->i_lock);
2479 } else if (IS_ROOT(alias)) {
2480 /* Is this an anonymous mountpoint that we
2481 * could splice into our tree? */
2482 __d_materialise_dentry(dentry, alias);
2483 write_sequnlock(&rename_lock);
2487 /* Nope, but we must(!) avoid directory
2488 * aliasing. This drops inode->i_lock */
2489 actual = __d_unalias(inode, dentry, alias);
2491 write_sequnlock(&rename_lock);
2492 if (IS_ERR(actual)) {
2493 if (PTR_ERR(actual) == -ELOOP)
2494 pr_warn_ratelimited(
2495 "VFS: Lookup of '%s' in %s %s"
2496 " would have caused loop\n",
2497 dentry->d_name.name,
2498 inode->i_sb->s_type->name,
2506 /* Add a unique reference */
2507 actual = __d_instantiate_unique(dentry, inode);
2511 BUG_ON(!d_unhashed(actual));
2513 spin_lock(&actual->d_lock);
2516 spin_unlock(&actual->d_lock);
2517 spin_unlock(&inode->i_lock);
2519 if (actual == dentry) {
2520 security_d_instantiate(dentry, inode);
2527 EXPORT_SYMBOL_GPL(d_materialise_unique);
2529 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2533 return -ENAMETOOLONG;
2535 memcpy(*buffer, str, namelen);
2539 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2541 return prepend(buffer, buflen, name->name, name->len);
2545 * prepend_path - Prepend path string to a buffer
2546 * @path: the dentry/vfsmount to report
2547 * @root: root vfsmnt/dentry
2548 * @buffer: pointer to the end of the buffer
2549 * @buflen: pointer to buffer length
2551 * Caller holds the rename_lock.
2553 static int prepend_path(const struct path *path,
2554 const struct path *root,
2555 char **buffer, int *buflen)
2557 struct dentry *dentry = path->dentry;
2558 struct vfsmount *vfsmnt = path->mnt;
2559 struct mount *mnt = real_mount(vfsmnt);
2563 br_read_lock(vfsmount_lock);
2564 while (dentry != root->dentry || vfsmnt != root->mnt) {
2565 struct dentry * parent;
2567 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2569 if (!mnt_has_parent(mnt))
2571 dentry = mnt->mnt_mountpoint;
2572 mnt = mnt->mnt_parent;
2576 parent = dentry->d_parent;
2578 spin_lock(&dentry->d_lock);
2579 error = prepend_name(buffer, buflen, &dentry->d_name);
2580 spin_unlock(&dentry->d_lock);
2582 error = prepend(buffer, buflen, "/", 1);
2590 if (!error && !slash)
2591 error = prepend(buffer, buflen, "/", 1);
2594 br_read_unlock(vfsmount_lock);
2599 * Filesystems needing to implement special "root names"
2600 * should do so with ->d_dname()
2602 if (IS_ROOT(dentry) &&
2603 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2604 WARN(1, "Root dentry has weird name <%.*s>\n",
2605 (int) dentry->d_name.len, dentry->d_name.name);
2608 error = prepend(buffer, buflen, "/", 1);
2610 error = real_mount(vfsmnt)->mnt_ns ? 1 : 2;
2615 * __d_path - return the path of a dentry
2616 * @path: the dentry/vfsmount to report
2617 * @root: root vfsmnt/dentry
2618 * @buf: buffer to return value in
2619 * @buflen: buffer length
2621 * Convert a dentry into an ASCII path name.
2623 * Returns a pointer into the buffer or an error code if the
2624 * path was too long.
2626 * "buflen" should be positive.
2628 * If the path is not reachable from the supplied root, return %NULL.
2630 char *__d_path(const struct path *path,
2631 const struct path *root,
2632 char *buf, int buflen)
2634 char *res = buf + buflen;
2637 prepend(&res, &buflen, "\0", 1);
2638 write_seqlock(&rename_lock);
2639 error = prepend_path(path, root, &res, &buflen);
2640 write_sequnlock(&rename_lock);
2643 return ERR_PTR(error);
2649 char *d_absolute_path(const struct path *path,
2650 char *buf, int buflen)
2652 struct path root = {};
2653 char *res = buf + buflen;
2656 prepend(&res, &buflen, "\0", 1);
2657 write_seqlock(&rename_lock);
2658 error = prepend_path(path, &root, &res, &buflen);
2659 write_sequnlock(&rename_lock);
2664 return ERR_PTR(error);
2669 * same as __d_path but appends "(deleted)" for unlinked files.
2671 static int path_with_deleted(const struct path *path,
2672 const struct path *root,
2673 char **buf, int *buflen)
2675 prepend(buf, buflen, "\0", 1);
2676 if (d_unlinked(path->dentry)) {
2677 int error = prepend(buf, buflen, " (deleted)", 10);
2682 return prepend_path(path, root, buf, buflen);
2685 static int prepend_unreachable(char **buffer, int *buflen)
2687 return prepend(buffer, buflen, "(unreachable)", 13);
2691 * d_path - return the path of a dentry
2692 * @path: path to report
2693 * @buf: buffer to return value in
2694 * @buflen: buffer length
2696 * Convert a dentry into an ASCII path name. If the entry has been deleted
2697 * the string " (deleted)" is appended. Note that this is ambiguous.
2699 * Returns a pointer into the buffer or an error code if the path was
2700 * too long. Note: Callers should use the returned pointer, not the passed
2701 * in buffer, to use the name! The implementation often starts at an offset
2702 * into the buffer, and may leave 0 bytes at the start.
2704 * "buflen" should be positive.
2706 char *d_path(const struct path *path, char *buf, int buflen)
2708 char *res = buf + buflen;
2713 * We have various synthetic filesystems that never get mounted. On
2714 * these filesystems dentries are never used for lookup purposes, and
2715 * thus don't need to be hashed. They also don't need a name until a
2716 * user wants to identify the object in /proc/pid/fd/. The little hack
2717 * below allows us to generate a name for these objects on demand:
2719 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2720 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2722 get_fs_root(current->fs, &root);
2723 write_seqlock(&rename_lock);
2724 error = path_with_deleted(path, &root, &res, &buflen);
2726 res = ERR_PTR(error);
2727 write_sequnlock(&rename_lock);
2731 EXPORT_SYMBOL(d_path);
2734 * d_path_with_unreachable - return the path of a dentry
2735 * @path: path to report
2736 * @buf: buffer to return value in
2737 * @buflen: buffer length
2739 * The difference from d_path() is that this prepends "(unreachable)"
2740 * to paths which are unreachable from the current process' root.
2742 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2744 char *res = buf + buflen;
2748 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2749 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2751 get_fs_root(current->fs, &root);
2752 write_seqlock(&rename_lock);
2753 error = path_with_deleted(path, &root, &res, &buflen);
2755 error = prepend_unreachable(&res, &buflen);
2756 write_sequnlock(&rename_lock);
2759 res = ERR_PTR(error);
2765 * Helper function for dentry_operations.d_dname() members
2767 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2768 const char *fmt, ...)
2774 va_start(args, fmt);
2775 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2778 if (sz > sizeof(temp) || sz > buflen)
2779 return ERR_PTR(-ENAMETOOLONG);
2781 buffer += buflen - sz;
2782 return memcpy(buffer, temp, sz);
2786 * Write full pathname from the root of the filesystem into the buffer.
2788 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2790 char *end = buf + buflen;
2793 prepend(&end, &buflen, "\0", 1);
2800 while (!IS_ROOT(dentry)) {
2801 struct dentry *parent = dentry->d_parent;
2805 spin_lock(&dentry->d_lock);
2806 error = prepend_name(&end, &buflen, &dentry->d_name);
2807 spin_unlock(&dentry->d_lock);
2808 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2816 return ERR_PTR(-ENAMETOOLONG);
2819 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2823 write_seqlock(&rename_lock);
2824 retval = __dentry_path(dentry, buf, buflen);
2825 write_sequnlock(&rename_lock);
2829 EXPORT_SYMBOL(dentry_path_raw);
2831 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2836 write_seqlock(&rename_lock);
2837 if (d_unlinked(dentry)) {
2839 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2843 retval = __dentry_path(dentry, buf, buflen);
2844 write_sequnlock(&rename_lock);
2845 if (!IS_ERR(retval) && p)
2846 *p = '/'; /* restore '/' overriden with '\0' */
2849 return ERR_PTR(-ENAMETOOLONG);
2853 * NOTE! The user-level library version returns a
2854 * character pointer. The kernel system call just
2855 * returns the length of the buffer filled (which
2856 * includes the ending '\0' character), or a negative
2857 * error value. So libc would do something like
2859 * char *getcwd(char * buf, size_t size)
2863 * retval = sys_getcwd(buf, size);
2870 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2873 struct path pwd, root;
2874 char *page = (char *) __get_free_page(GFP_USER);
2879 get_fs_root_and_pwd(current->fs, &root, &pwd);
2882 write_seqlock(&rename_lock);
2883 if (!d_unlinked(pwd.dentry)) {
2885 char *cwd = page + PAGE_SIZE;
2886 int buflen = PAGE_SIZE;
2888 prepend(&cwd, &buflen, "\0", 1);
2889 error = prepend_path(&pwd, &root, &cwd, &buflen);
2890 write_sequnlock(&rename_lock);
2895 /* Unreachable from current root */
2897 error = prepend_unreachable(&cwd, &buflen);
2903 len = PAGE_SIZE + page - cwd;
2906 if (copy_to_user(buf, cwd, len))
2910 write_sequnlock(&rename_lock);
2916 free_page((unsigned long) page);
2921 * Test whether new_dentry is a subdirectory of old_dentry.
2923 * Trivially implemented using the dcache structure
2927 * is_subdir - is new dentry a subdirectory of old_dentry
2928 * @new_dentry: new dentry
2929 * @old_dentry: old dentry
2931 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2932 * Returns 0 otherwise.
2933 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2936 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2941 if (new_dentry == old_dentry)
2945 /* for restarting inner loop in case of seq retry */
2946 seq = read_seqbegin(&rename_lock);
2948 * Need rcu_readlock to protect against the d_parent trashing
2952 if (d_ancestor(old_dentry, new_dentry))
2957 } while (read_seqretry(&rename_lock, seq));
2962 void d_genocide(struct dentry *root)
2964 struct dentry *this_parent;
2965 struct list_head *next;
2969 seq = read_seqbegin(&rename_lock);
2972 spin_lock(&this_parent->d_lock);
2974 next = this_parent->d_subdirs.next;
2976 while (next != &this_parent->d_subdirs) {
2977 struct list_head *tmp = next;
2978 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2981 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2982 if (d_unhashed(dentry) || !dentry->d_inode) {
2983 spin_unlock(&dentry->d_lock);
2986 if (!list_empty(&dentry->d_subdirs)) {
2987 spin_unlock(&this_parent->d_lock);
2988 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2989 this_parent = dentry;
2990 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2993 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2994 dentry->d_flags |= DCACHE_GENOCIDE;
2997 spin_unlock(&dentry->d_lock);
2999 if (this_parent != root) {
3000 struct dentry *child = this_parent;
3001 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
3002 this_parent->d_flags |= DCACHE_GENOCIDE;
3003 this_parent->d_count--;
3005 this_parent = try_to_ascend(this_parent, locked, seq);
3008 next = child->d_u.d_child.next;
3011 spin_unlock(&this_parent->d_lock);
3012 if (!locked && read_seqretry(&rename_lock, seq))
3015 write_sequnlock(&rename_lock);
3020 write_seqlock(&rename_lock);
3025 * find_inode_number - check for dentry with name
3026 * @dir: directory to check
3027 * @name: Name to find.
3029 * Check whether a dentry already exists for the given name,
3030 * and return the inode number if it has an inode. Otherwise
3033 * This routine is used to post-process directory listings for
3034 * filesystems using synthetic inode numbers, and is necessary
3035 * to keep getcwd() working.
3038 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3040 struct dentry * dentry;
3043 dentry = d_hash_and_lookup(dir, name);
3045 if (dentry->d_inode)
3046 ino = dentry->d_inode->i_ino;
3051 EXPORT_SYMBOL(find_inode_number);
3053 static __initdata unsigned long dhash_entries;
3054 static int __init set_dhash_entries(char *str)
3058 dhash_entries = simple_strtoul(str, &str, 0);
3061 __setup("dhash_entries=", set_dhash_entries);
3063 static void __init dcache_init_early(void)
3067 /* If hashes are distributed across NUMA nodes, defer
3068 * hash allocation until vmalloc space is available.
3074 alloc_large_system_hash("Dentry cache",
3075 sizeof(struct hlist_bl_head),
3083 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3084 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3087 static void __init dcache_init(void)
3092 * A constructor could be added for stable state like the lists,
3093 * but it is probably not worth it because of the cache nature
3096 dentry_cache = KMEM_CACHE(dentry,
3097 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3099 /* Hash may have been set up in dcache_init_early */
3104 alloc_large_system_hash("Dentry cache",
3105 sizeof(struct hlist_bl_head),
3113 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3114 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3117 /* SLAB cache for __getname() consumers */
3118 struct kmem_cache *names_cachep __read_mostly;
3119 EXPORT_SYMBOL(names_cachep);
3121 EXPORT_SYMBOL(d_genocide);
3123 void __init vfs_caches_init_early(void)
3125 dcache_init_early();
3129 void __init vfs_caches_init(unsigned long mempages)
3131 unsigned long reserve;
3133 /* Base hash sizes on available memory, with a reserve equal to
3134 150% of current kernel size */
3136 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3137 mempages -= reserve;
3139 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3140 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3144 files_init(mempages);