Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target...
[linux-btrfs-devel.git] / fs / dcache.c
bloba88948b8bd17d8b16c42af20ee55bcf47f358923
1 /*
2 * fs/dcache.c
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
9 /*
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>
19 #include <linux/mm.h>
20 #include <linux/fs.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/module.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include "internal.h"
42 * Usage:
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
51 * d_lock protects:
52 * - d_flags
53 * - d_name
54 * - d_lru
55 * - d_count
56 * - d_unhashed()
57 * - d_parent and d_subdirs
58 * - childrens' d_child and d_parent
59 * - d_alias, d_inode
61 * Ordering:
62 * dentry->d_inode->i_lock
63 * dentry->d_lock
64 * dcache_lru_lock
65 * dcache_hash_bucket lock
66 * s_anon lock
68 * If there is an ancestor relationship:
69 * dentry->d_parent->...->d_parent->d_lock
70 * ...
71 * dentry->d_parent->d_lock
72 * dentry->d_lock
74 * If no ancestor relationship:
75 * if (dentry1 < dentry2)
76 * dentry1->d_lock
77 * dentry2->d_lock
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,
106 unsigned long hash)
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 = {
115 .age_limit = 45,
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)
123 int i;
124 int sum = 0;
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);
136 #endif
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);
149 * no locks, please.
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);
161 else
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
182 * and is unhashed.
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;
189 if (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);
194 if (!inode->i_nlink)
195 fsnotify_inoderemove(inode);
196 if (dentry->d_op && dentry->d_op->d_iput)
197 dentry->d_op->d_iput(dentry, inode);
198 else
199 iput(inode);
200 } else {
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);
219 if (!inode->i_nlink)
220 fsnotify_inoderemove(inode);
221 if (dentry->d_op && dentry->d_op->d_iput)
222 dentry->d_op->d_iput(dentry, inode);
223 else
224 iput(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++;
264 } else {
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
280 * d_kill.
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
290 * dentry tree
292 dentry->d_flags |= DCACHE_DISCONNECTED;
293 if (parent)
294 spin_unlock(&parent->d_lock);
295 dentry_iput(dentry);
297 * dentry_iput drops the locks, at which point nobody (except
298 * transient RCU lookups) can reach this dentry.
300 d_free(dentry);
301 return parent;
305 * Unhash a dentry without inserting an RCU walk barrier or checking that
306 * dentry->d_lock is locked. The caller must take care of that, if
307 * appropriate.
309 static void __d_shrink(struct dentry *dentry)
311 if (!d_unhashed(dentry)) {
312 struct hlist_bl_head *b;
313 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
314 b = &dentry->d_sb->s_anon;
315 else
316 b = d_hash(dentry->d_parent, dentry->d_name.hash);
318 hlist_bl_lock(b);
319 __hlist_bl_del(&dentry->d_hash);
320 dentry->d_hash.pprev = NULL;
321 hlist_bl_unlock(b);
326 * d_drop - drop a dentry
327 * @dentry: dentry to drop
329 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
330 * be found through a VFS lookup any more. Note that this is different from
331 * deleting the dentry - d_delete will try to mark the dentry negative if
332 * possible, giving a successful _negative_ lookup, while d_drop will
333 * just make the cache lookup fail.
335 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
336 * reason (NFS timeouts or autofs deletes).
338 * __d_drop requires dentry->d_lock.
340 void __d_drop(struct dentry *dentry)
342 if (!d_unhashed(dentry)) {
343 __d_shrink(dentry);
344 dentry_rcuwalk_barrier(dentry);
347 EXPORT_SYMBOL(__d_drop);
349 void d_drop(struct dentry *dentry)
351 spin_lock(&dentry->d_lock);
352 __d_drop(dentry);
353 spin_unlock(&dentry->d_lock);
355 EXPORT_SYMBOL(d_drop);
358 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
359 * @dentry: dentry to drop
361 * This is called when we do a lookup on a placeholder dentry that needed to be
362 * looked up. The dentry should have been hashed in order for it to be found by
363 * the lookup code, but now needs to be unhashed while we do the actual lookup
364 * and clear the DCACHE_NEED_LOOKUP flag.
366 void d_clear_need_lookup(struct dentry *dentry)
368 spin_lock(&dentry->d_lock);
369 __d_drop(dentry);
370 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
371 spin_unlock(&dentry->d_lock);
373 EXPORT_SYMBOL(d_clear_need_lookup);
376 * Finish off a dentry we've decided to kill.
377 * dentry->d_lock must be held, returns with it unlocked.
378 * If ref is non-zero, then decrement the refcount too.
379 * Returns dentry requiring refcount drop, or NULL if we're done.
381 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
382 __releases(dentry->d_lock)
384 struct inode *inode;
385 struct dentry *parent;
387 inode = dentry->d_inode;
388 if (inode && !spin_trylock(&inode->i_lock)) {
389 relock:
390 spin_unlock(&dentry->d_lock);
391 cpu_relax();
392 return dentry; /* try again with same dentry */
394 if (IS_ROOT(dentry))
395 parent = NULL;
396 else
397 parent = dentry->d_parent;
398 if (parent && !spin_trylock(&parent->d_lock)) {
399 if (inode)
400 spin_unlock(&inode->i_lock);
401 goto relock;
404 if (ref)
405 dentry->d_count--;
406 /* if dentry was on the d_lru list delete it from there */
407 dentry_lru_del(dentry);
408 /* if it was on the hash then remove it */
409 __d_drop(dentry);
410 return d_kill(dentry, parent);
414 * This is dput
416 * This is complicated by the fact that we do not want to put
417 * dentries that are no longer on any hash chain on the unused
418 * list: we'd much rather just get rid of them immediately.
420 * However, that implies that we have to traverse the dentry
421 * tree upwards to the parents which might _also_ now be
422 * scheduled for deletion (it may have been only waiting for
423 * its last child to go away).
425 * This tail recursion is done by hand as we don't want to depend
426 * on the compiler to always get this right (gcc generally doesn't).
427 * Real recursion would eat up our stack space.
431 * dput - release a dentry
432 * @dentry: dentry to release
434 * Release a dentry. This will drop the usage count and if appropriate
435 * call the dentry unlink method as well as removing it from the queues and
436 * releasing its resources. If the parent dentries were scheduled for release
437 * they too may now get deleted.
439 void dput(struct dentry *dentry)
441 if (!dentry)
442 return;
444 repeat:
445 if (dentry->d_count == 1)
446 might_sleep();
447 spin_lock(&dentry->d_lock);
448 BUG_ON(!dentry->d_count);
449 if (dentry->d_count > 1) {
450 dentry->d_count--;
451 spin_unlock(&dentry->d_lock);
452 return;
455 if (dentry->d_flags & DCACHE_OP_DELETE) {
456 if (dentry->d_op->d_delete(dentry))
457 goto kill_it;
460 /* Unreachable? Get rid of it */
461 if (d_unhashed(dentry))
462 goto kill_it;
465 * If this dentry needs lookup, don't set the referenced flag so that it
466 * is more likely to be cleaned up by the dcache shrinker in case of
467 * memory pressure.
469 if (!d_need_lookup(dentry))
470 dentry->d_flags |= DCACHE_REFERENCED;
471 dentry_lru_add(dentry);
473 dentry->d_count--;
474 spin_unlock(&dentry->d_lock);
475 return;
477 kill_it:
478 dentry = dentry_kill(dentry, 1);
479 if (dentry)
480 goto repeat;
482 EXPORT_SYMBOL(dput);
485 * d_invalidate - invalidate a dentry
486 * @dentry: dentry to invalidate
488 * Try to invalidate the dentry if it turns out to be
489 * possible. If there are other dentries that can be
490 * reached through this one we can't delete it and we
491 * return -EBUSY. On success we return 0.
493 * no dcache lock.
496 int d_invalidate(struct dentry * dentry)
499 * If it's already been dropped, return OK.
501 spin_lock(&dentry->d_lock);
502 if (d_unhashed(dentry)) {
503 spin_unlock(&dentry->d_lock);
504 return 0;
507 * Check whether to do a partial shrink_dcache
508 * to get rid of unused child entries.
510 if (!list_empty(&dentry->d_subdirs)) {
511 spin_unlock(&dentry->d_lock);
512 shrink_dcache_parent(dentry);
513 spin_lock(&dentry->d_lock);
517 * Somebody else still using it?
519 * If it's a directory, we can't drop it
520 * for fear of somebody re-populating it
521 * with children (even though dropping it
522 * would make it unreachable from the root,
523 * we might still populate it if it was a
524 * working directory or similar).
526 if (dentry->d_count > 1) {
527 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
528 spin_unlock(&dentry->d_lock);
529 return -EBUSY;
533 __d_drop(dentry);
534 spin_unlock(&dentry->d_lock);
535 return 0;
537 EXPORT_SYMBOL(d_invalidate);
539 /* This must be called with d_lock held */
540 static inline void __dget_dlock(struct dentry *dentry)
542 dentry->d_count++;
545 static inline void __dget(struct dentry *dentry)
547 spin_lock(&dentry->d_lock);
548 __dget_dlock(dentry);
549 spin_unlock(&dentry->d_lock);
552 struct dentry *dget_parent(struct dentry *dentry)
554 struct dentry *ret;
556 repeat:
558 * Don't need rcu_dereference because we re-check it was correct under
559 * the lock.
561 rcu_read_lock();
562 ret = dentry->d_parent;
563 spin_lock(&ret->d_lock);
564 if (unlikely(ret != dentry->d_parent)) {
565 spin_unlock(&ret->d_lock);
566 rcu_read_unlock();
567 goto repeat;
569 rcu_read_unlock();
570 BUG_ON(!ret->d_count);
571 ret->d_count++;
572 spin_unlock(&ret->d_lock);
573 return ret;
575 EXPORT_SYMBOL(dget_parent);
578 * d_find_alias - grab a hashed alias of inode
579 * @inode: inode in question
580 * @want_discon: flag, used by d_splice_alias, to request
581 * that only a DISCONNECTED alias be returned.
583 * If inode has a hashed alias, or is a directory and has any alias,
584 * acquire the reference to alias and return it. Otherwise return NULL.
585 * Notice that if inode is a directory there can be only one alias and
586 * it can be unhashed only if it has no children, or if it is the root
587 * of a filesystem.
589 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
590 * any other hashed alias over that one unless @want_discon is set,
591 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
593 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
595 struct dentry *alias, *discon_alias;
597 again:
598 discon_alias = NULL;
599 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
600 spin_lock(&alias->d_lock);
601 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
602 if (IS_ROOT(alias) &&
603 (alias->d_flags & DCACHE_DISCONNECTED)) {
604 discon_alias = alias;
605 } else if (!want_discon) {
606 __dget_dlock(alias);
607 spin_unlock(&alias->d_lock);
608 return alias;
611 spin_unlock(&alias->d_lock);
613 if (discon_alias) {
614 alias = discon_alias;
615 spin_lock(&alias->d_lock);
616 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
617 if (IS_ROOT(alias) &&
618 (alias->d_flags & DCACHE_DISCONNECTED)) {
619 __dget_dlock(alias);
620 spin_unlock(&alias->d_lock);
621 return alias;
624 spin_unlock(&alias->d_lock);
625 goto again;
627 return NULL;
630 struct dentry *d_find_alias(struct inode *inode)
632 struct dentry *de = NULL;
634 if (!list_empty(&inode->i_dentry)) {
635 spin_lock(&inode->i_lock);
636 de = __d_find_alias(inode, 0);
637 spin_unlock(&inode->i_lock);
639 return de;
641 EXPORT_SYMBOL(d_find_alias);
644 * Try to kill dentries associated with this inode.
645 * WARNING: you must own a reference to inode.
647 void d_prune_aliases(struct inode *inode)
649 struct dentry *dentry;
650 restart:
651 spin_lock(&inode->i_lock);
652 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
653 spin_lock(&dentry->d_lock);
654 if (!dentry->d_count) {
655 __dget_dlock(dentry);
656 __d_drop(dentry);
657 spin_unlock(&dentry->d_lock);
658 spin_unlock(&inode->i_lock);
659 dput(dentry);
660 goto restart;
662 spin_unlock(&dentry->d_lock);
664 spin_unlock(&inode->i_lock);
666 EXPORT_SYMBOL(d_prune_aliases);
669 * Try to throw away a dentry - free the inode, dput the parent.
670 * Requires dentry->d_lock is held, and dentry->d_count == 0.
671 * Releases dentry->d_lock.
673 * This may fail if locks cannot be acquired no problem, just try again.
675 static void try_prune_one_dentry(struct dentry *dentry)
676 __releases(dentry->d_lock)
678 struct dentry *parent;
680 parent = dentry_kill(dentry, 0);
682 * If dentry_kill returns NULL, we have nothing more to do.
683 * if it returns the same dentry, trylocks failed. In either
684 * case, just loop again.
686 * Otherwise, we need to prune ancestors too. This is necessary
687 * to prevent quadratic behavior of shrink_dcache_parent(), but
688 * is also expected to be beneficial in reducing dentry cache
689 * fragmentation.
691 if (!parent)
692 return;
693 if (parent == dentry)
694 return;
696 /* Prune ancestors. */
697 dentry = parent;
698 while (dentry) {
699 spin_lock(&dentry->d_lock);
700 if (dentry->d_count > 1) {
701 dentry->d_count--;
702 spin_unlock(&dentry->d_lock);
703 return;
705 dentry = dentry_kill(dentry, 1);
709 static void shrink_dentry_list(struct list_head *list)
711 struct dentry *dentry;
713 rcu_read_lock();
714 for (;;) {
715 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
716 if (&dentry->d_lru == list)
717 break; /* empty */
718 spin_lock(&dentry->d_lock);
719 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
720 spin_unlock(&dentry->d_lock);
721 continue;
725 * We found an inuse dentry which was not removed from
726 * the LRU because of laziness during lookup. Do not free
727 * it - just keep it off the LRU list.
729 if (dentry->d_count) {
730 dentry_lru_del(dentry);
731 spin_unlock(&dentry->d_lock);
732 continue;
735 rcu_read_unlock();
737 try_prune_one_dentry(dentry);
739 rcu_read_lock();
741 rcu_read_unlock();
745 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
746 * @sb: superblock to shrink dentry LRU.
747 * @count: number of entries to prune
748 * @flags: flags to control the dentry processing
750 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
752 static void __shrink_dcache_sb(struct super_block *sb, int count, int flags)
754 struct dentry *dentry;
755 LIST_HEAD(referenced);
756 LIST_HEAD(tmp);
758 relock:
759 spin_lock(&dcache_lru_lock);
760 while (!list_empty(&sb->s_dentry_lru)) {
761 dentry = list_entry(sb->s_dentry_lru.prev,
762 struct dentry, d_lru);
763 BUG_ON(dentry->d_sb != sb);
765 if (!spin_trylock(&dentry->d_lock)) {
766 spin_unlock(&dcache_lru_lock);
767 cpu_relax();
768 goto relock;
772 * If we are honouring the DCACHE_REFERENCED flag and the
773 * dentry has this flag set, don't free it. Clear the flag
774 * and put it back on the LRU.
776 if (flags & DCACHE_REFERENCED &&
777 dentry->d_flags & DCACHE_REFERENCED) {
778 dentry->d_flags &= ~DCACHE_REFERENCED;
779 list_move(&dentry->d_lru, &referenced);
780 spin_unlock(&dentry->d_lock);
781 } else {
782 list_move_tail(&dentry->d_lru, &tmp);
783 spin_unlock(&dentry->d_lock);
784 if (!--count)
785 break;
787 cond_resched_lock(&dcache_lru_lock);
789 if (!list_empty(&referenced))
790 list_splice(&referenced, &sb->s_dentry_lru);
791 spin_unlock(&dcache_lru_lock);
793 shrink_dentry_list(&tmp);
797 * prune_dcache_sb - shrink the dcache
798 * @sb: superblock
799 * @nr_to_scan: number of entries to try to free
801 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
802 * done when we need more memory an called from the superblock shrinker
803 * function.
805 * This function may fail to free any resources if all the dentries are in
806 * use.
808 void prune_dcache_sb(struct super_block *sb, int nr_to_scan)
810 __shrink_dcache_sb(sb, nr_to_scan, DCACHE_REFERENCED);
814 * shrink_dcache_sb - shrink dcache for a superblock
815 * @sb: superblock
817 * Shrink the dcache for the specified super block. This is used to free
818 * the dcache before unmounting a file system.
820 void shrink_dcache_sb(struct super_block *sb)
822 LIST_HEAD(tmp);
824 spin_lock(&dcache_lru_lock);
825 while (!list_empty(&sb->s_dentry_lru)) {
826 list_splice_init(&sb->s_dentry_lru, &tmp);
827 spin_unlock(&dcache_lru_lock);
828 shrink_dentry_list(&tmp);
829 spin_lock(&dcache_lru_lock);
831 spin_unlock(&dcache_lru_lock);
833 EXPORT_SYMBOL(shrink_dcache_sb);
836 * destroy a single subtree of dentries for unmount
837 * - see the comments on shrink_dcache_for_umount() for a description of the
838 * locking
840 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
842 struct dentry *parent;
844 BUG_ON(!IS_ROOT(dentry));
846 for (;;) {
847 /* descend to the first leaf in the current subtree */
848 while (!list_empty(&dentry->d_subdirs))
849 dentry = list_entry(dentry->d_subdirs.next,
850 struct dentry, d_u.d_child);
852 /* consume the dentries from this leaf up through its parents
853 * until we find one with children or run out altogether */
854 do {
855 struct inode *inode;
857 /* detach from the system */
858 dentry_lru_del(dentry);
859 __d_shrink(dentry);
861 if (dentry->d_count != 0) {
862 printk(KERN_ERR
863 "BUG: Dentry %p{i=%lx,n=%s}"
864 " still in use (%d)"
865 " [unmount of %s %s]\n",
866 dentry,
867 dentry->d_inode ?
868 dentry->d_inode->i_ino : 0UL,
869 dentry->d_name.name,
870 dentry->d_count,
871 dentry->d_sb->s_type->name,
872 dentry->d_sb->s_id);
873 BUG();
876 if (IS_ROOT(dentry)) {
877 parent = NULL;
878 list_del(&dentry->d_u.d_child);
879 } else {
880 parent = dentry->d_parent;
881 parent->d_count--;
882 list_del(&dentry->d_u.d_child);
885 inode = dentry->d_inode;
886 if (inode) {
887 dentry->d_inode = NULL;
888 list_del_init(&dentry->d_alias);
889 if (dentry->d_op && dentry->d_op->d_iput)
890 dentry->d_op->d_iput(dentry, inode);
891 else
892 iput(inode);
895 d_free(dentry);
897 /* finished when we fall off the top of the tree,
898 * otherwise we ascend to the parent and move to the
899 * next sibling if there is one */
900 if (!parent)
901 return;
902 dentry = parent;
903 } while (list_empty(&dentry->d_subdirs));
905 dentry = list_entry(dentry->d_subdirs.next,
906 struct dentry, d_u.d_child);
911 * destroy the dentries attached to a superblock on unmounting
912 * - we don't need to use dentry->d_lock because:
913 * - the superblock is detached from all mountings and open files, so the
914 * dentry trees will not be rearranged by the VFS
915 * - s_umount is write-locked, so the memory pressure shrinker will ignore
916 * any dentries belonging to this superblock that it comes across
917 * - the filesystem itself is no longer permitted to rearrange the dentries
918 * in this superblock
920 void shrink_dcache_for_umount(struct super_block *sb)
922 struct dentry *dentry;
924 if (down_read_trylock(&sb->s_umount))
925 BUG();
927 dentry = sb->s_root;
928 sb->s_root = NULL;
929 dentry->d_count--;
930 shrink_dcache_for_umount_subtree(dentry);
932 while (!hlist_bl_empty(&sb->s_anon)) {
933 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
934 shrink_dcache_for_umount_subtree(dentry);
939 * This tries to ascend one level of parenthood, but
940 * we can race with renaming, so we need to re-check
941 * the parenthood after dropping the lock and check
942 * that the sequence number still matches.
944 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
946 struct dentry *new = old->d_parent;
948 rcu_read_lock();
949 spin_unlock(&old->d_lock);
950 spin_lock(&new->d_lock);
953 * might go back up the wrong parent if we have had a rename
954 * or deletion
956 if (new != old->d_parent ||
957 (old->d_flags & DCACHE_DISCONNECTED) ||
958 (!locked && read_seqretry(&rename_lock, seq))) {
959 spin_unlock(&new->d_lock);
960 new = NULL;
962 rcu_read_unlock();
963 return new;
968 * Search for at least 1 mount point in the dentry's subdirs.
969 * We descend to the next level whenever the d_subdirs
970 * list is non-empty and continue searching.
974 * have_submounts - check for mounts over a dentry
975 * @parent: dentry to check.
977 * Return true if the parent or its subdirectories contain
978 * a mount point
980 int have_submounts(struct dentry *parent)
982 struct dentry *this_parent;
983 struct list_head *next;
984 unsigned seq;
985 int locked = 0;
987 seq = read_seqbegin(&rename_lock);
988 again:
989 this_parent = parent;
991 if (d_mountpoint(parent))
992 goto positive;
993 spin_lock(&this_parent->d_lock);
994 repeat:
995 next = this_parent->d_subdirs.next;
996 resume:
997 while (next != &this_parent->d_subdirs) {
998 struct list_head *tmp = next;
999 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1000 next = tmp->next;
1002 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1003 /* Have we found a mount point ? */
1004 if (d_mountpoint(dentry)) {
1005 spin_unlock(&dentry->d_lock);
1006 spin_unlock(&this_parent->d_lock);
1007 goto positive;
1009 if (!list_empty(&dentry->d_subdirs)) {
1010 spin_unlock(&this_parent->d_lock);
1011 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1012 this_parent = dentry;
1013 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1014 goto repeat;
1016 spin_unlock(&dentry->d_lock);
1019 * All done at this level ... ascend and resume the search.
1021 if (this_parent != parent) {
1022 struct dentry *child = this_parent;
1023 this_parent = try_to_ascend(this_parent, locked, seq);
1024 if (!this_parent)
1025 goto rename_retry;
1026 next = child->d_u.d_child.next;
1027 goto resume;
1029 spin_unlock(&this_parent->d_lock);
1030 if (!locked && read_seqretry(&rename_lock, seq))
1031 goto rename_retry;
1032 if (locked)
1033 write_sequnlock(&rename_lock);
1034 return 0; /* No mount points found in tree */
1035 positive:
1036 if (!locked && read_seqretry(&rename_lock, seq))
1037 goto rename_retry;
1038 if (locked)
1039 write_sequnlock(&rename_lock);
1040 return 1;
1042 rename_retry:
1043 locked = 1;
1044 write_seqlock(&rename_lock);
1045 goto again;
1047 EXPORT_SYMBOL(have_submounts);
1050 * Search the dentry child list for the specified parent,
1051 * and move any unused dentries to the end of the unused
1052 * list for prune_dcache(). We descend to the next level
1053 * whenever the d_subdirs list is non-empty and continue
1054 * searching.
1056 * It returns zero iff there are no unused children,
1057 * otherwise it returns the number of children moved to
1058 * the end of the unused list. This may not be the total
1059 * number of unused children, because select_parent can
1060 * drop the lock and return early due to latency
1061 * constraints.
1063 static int select_parent(struct dentry * parent)
1065 struct dentry *this_parent;
1066 struct list_head *next;
1067 unsigned seq;
1068 int found = 0;
1069 int locked = 0;
1071 seq = read_seqbegin(&rename_lock);
1072 again:
1073 this_parent = parent;
1074 spin_lock(&this_parent->d_lock);
1075 repeat:
1076 next = this_parent->d_subdirs.next;
1077 resume:
1078 while (next != &this_parent->d_subdirs) {
1079 struct list_head *tmp = next;
1080 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1081 next = tmp->next;
1083 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1086 * move only zero ref count dentries to the end
1087 * of the unused list for prune_dcache
1089 if (!dentry->d_count) {
1090 dentry_lru_move_tail(dentry);
1091 found++;
1092 } else {
1093 dentry_lru_del(dentry);
1097 * We can return to the caller if we have found some (this
1098 * ensures forward progress). We'll be coming back to find
1099 * the rest.
1101 if (found && need_resched()) {
1102 spin_unlock(&dentry->d_lock);
1103 goto out;
1107 * Descend a level if the d_subdirs list is non-empty.
1109 if (!list_empty(&dentry->d_subdirs)) {
1110 spin_unlock(&this_parent->d_lock);
1111 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1112 this_parent = dentry;
1113 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1114 goto repeat;
1117 spin_unlock(&dentry->d_lock);
1120 * All done at this level ... ascend and resume the search.
1122 if (this_parent != parent) {
1123 struct dentry *child = this_parent;
1124 this_parent = try_to_ascend(this_parent, locked, seq);
1125 if (!this_parent)
1126 goto rename_retry;
1127 next = child->d_u.d_child.next;
1128 goto resume;
1130 out:
1131 spin_unlock(&this_parent->d_lock);
1132 if (!locked && read_seqretry(&rename_lock, seq))
1133 goto rename_retry;
1134 if (locked)
1135 write_sequnlock(&rename_lock);
1136 return found;
1138 rename_retry:
1139 if (found)
1140 return found;
1141 locked = 1;
1142 write_seqlock(&rename_lock);
1143 goto again;
1147 * shrink_dcache_parent - prune dcache
1148 * @parent: parent of entries to prune
1150 * Prune the dcache to remove unused children of the parent dentry.
1153 void shrink_dcache_parent(struct dentry * parent)
1155 struct super_block *sb = parent->d_sb;
1156 int found;
1158 while ((found = select_parent(parent)) != 0)
1159 __shrink_dcache_sb(sb, found, 0);
1161 EXPORT_SYMBOL(shrink_dcache_parent);
1164 * __d_alloc - allocate a dcache entry
1165 * @sb: filesystem it will belong to
1166 * @name: qstr of the name
1168 * Allocates a dentry. It returns %NULL if there is insufficient memory
1169 * available. On a success the dentry is returned. The name passed in is
1170 * copied and the copy passed in may be reused after this call.
1173 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1175 struct dentry *dentry;
1176 char *dname;
1178 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1179 if (!dentry)
1180 return NULL;
1182 if (name->len > DNAME_INLINE_LEN-1) {
1183 dname = kmalloc(name->len + 1, GFP_KERNEL);
1184 if (!dname) {
1185 kmem_cache_free(dentry_cache, dentry);
1186 return NULL;
1188 } else {
1189 dname = dentry->d_iname;
1191 dentry->d_name.name = dname;
1193 dentry->d_name.len = name->len;
1194 dentry->d_name.hash = name->hash;
1195 memcpy(dname, name->name, name->len);
1196 dname[name->len] = 0;
1198 dentry->d_count = 1;
1199 dentry->d_flags = 0;
1200 spin_lock_init(&dentry->d_lock);
1201 seqcount_init(&dentry->d_seq);
1202 dentry->d_inode = NULL;
1203 dentry->d_parent = dentry;
1204 dentry->d_sb = sb;
1205 dentry->d_op = NULL;
1206 dentry->d_fsdata = NULL;
1207 INIT_HLIST_BL_NODE(&dentry->d_hash);
1208 INIT_LIST_HEAD(&dentry->d_lru);
1209 INIT_LIST_HEAD(&dentry->d_subdirs);
1210 INIT_LIST_HEAD(&dentry->d_alias);
1211 INIT_LIST_HEAD(&dentry->d_u.d_child);
1212 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1214 this_cpu_inc(nr_dentry);
1216 return dentry;
1220 * d_alloc - allocate a dcache entry
1221 * @parent: parent of entry to allocate
1222 * @name: qstr of the name
1224 * Allocates a dentry. It returns %NULL if there is insufficient memory
1225 * available. On a success the dentry is returned. The name passed in is
1226 * copied and the copy passed in may be reused after this call.
1228 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1230 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1231 if (!dentry)
1232 return NULL;
1234 spin_lock(&parent->d_lock);
1236 * don't need child lock because it is not subject
1237 * to concurrency here
1239 __dget_dlock(parent);
1240 dentry->d_parent = parent;
1241 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1242 spin_unlock(&parent->d_lock);
1244 return dentry;
1246 EXPORT_SYMBOL(d_alloc);
1248 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1250 struct dentry *dentry = __d_alloc(sb, name);
1251 if (dentry)
1252 dentry->d_flags |= DCACHE_DISCONNECTED;
1253 return dentry;
1255 EXPORT_SYMBOL(d_alloc_pseudo);
1257 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1259 struct qstr q;
1261 q.name = name;
1262 q.len = strlen(name);
1263 q.hash = full_name_hash(q.name, q.len);
1264 return d_alloc(parent, &q);
1266 EXPORT_SYMBOL(d_alloc_name);
1268 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1270 WARN_ON_ONCE(dentry->d_op);
1271 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1272 DCACHE_OP_COMPARE |
1273 DCACHE_OP_REVALIDATE |
1274 DCACHE_OP_DELETE ));
1275 dentry->d_op = op;
1276 if (!op)
1277 return;
1278 if (op->d_hash)
1279 dentry->d_flags |= DCACHE_OP_HASH;
1280 if (op->d_compare)
1281 dentry->d_flags |= DCACHE_OP_COMPARE;
1282 if (op->d_revalidate)
1283 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1284 if (op->d_delete)
1285 dentry->d_flags |= DCACHE_OP_DELETE;
1288 EXPORT_SYMBOL(d_set_d_op);
1290 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1292 spin_lock(&dentry->d_lock);
1293 if (inode) {
1294 if (unlikely(IS_AUTOMOUNT(inode)))
1295 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1296 list_add(&dentry->d_alias, &inode->i_dentry);
1298 dentry->d_inode = inode;
1299 dentry_rcuwalk_barrier(dentry);
1300 spin_unlock(&dentry->d_lock);
1301 fsnotify_d_instantiate(dentry, inode);
1305 * d_instantiate - fill in inode information for a dentry
1306 * @entry: dentry to complete
1307 * @inode: inode to attach to this dentry
1309 * Fill in inode information in the entry.
1311 * This turns negative dentries into productive full members
1312 * of society.
1314 * NOTE! This assumes that the inode count has been incremented
1315 * (or otherwise set) by the caller to indicate that it is now
1316 * in use by the dcache.
1319 void d_instantiate(struct dentry *entry, struct inode * inode)
1321 BUG_ON(!list_empty(&entry->d_alias));
1322 if (inode)
1323 spin_lock(&inode->i_lock);
1324 __d_instantiate(entry, inode);
1325 if (inode)
1326 spin_unlock(&inode->i_lock);
1327 security_d_instantiate(entry, inode);
1329 EXPORT_SYMBOL(d_instantiate);
1332 * d_instantiate_unique - instantiate a non-aliased dentry
1333 * @entry: dentry to instantiate
1334 * @inode: inode to attach to this dentry
1336 * Fill in inode information in the entry. On success, it returns NULL.
1337 * If an unhashed alias of "entry" already exists, then we return the
1338 * aliased dentry instead and drop one reference to inode.
1340 * Note that in order to avoid conflicts with rename() etc, the caller
1341 * had better be holding the parent directory semaphore.
1343 * This also assumes that the inode count has been incremented
1344 * (or otherwise set) by the caller to indicate that it is now
1345 * in use by the dcache.
1347 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1348 struct inode *inode)
1350 struct dentry *alias;
1351 int len = entry->d_name.len;
1352 const char *name = entry->d_name.name;
1353 unsigned int hash = entry->d_name.hash;
1355 if (!inode) {
1356 __d_instantiate(entry, NULL);
1357 return NULL;
1360 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1361 struct qstr *qstr = &alias->d_name;
1364 * Don't need alias->d_lock here, because aliases with
1365 * d_parent == entry->d_parent are not subject to name or
1366 * parent changes, because the parent inode i_mutex is held.
1368 if (qstr->hash != hash)
1369 continue;
1370 if (alias->d_parent != entry->d_parent)
1371 continue;
1372 if (dentry_cmp(qstr->name, qstr->len, name, len))
1373 continue;
1374 __dget(alias);
1375 return alias;
1378 __d_instantiate(entry, inode);
1379 return NULL;
1382 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1384 struct dentry *result;
1386 BUG_ON(!list_empty(&entry->d_alias));
1388 if (inode)
1389 spin_lock(&inode->i_lock);
1390 result = __d_instantiate_unique(entry, inode);
1391 if (inode)
1392 spin_unlock(&inode->i_lock);
1394 if (!result) {
1395 security_d_instantiate(entry, inode);
1396 return NULL;
1399 BUG_ON(!d_unhashed(result));
1400 iput(inode);
1401 return result;
1404 EXPORT_SYMBOL(d_instantiate_unique);
1407 * d_alloc_root - allocate root dentry
1408 * @root_inode: inode to allocate the root for
1410 * Allocate a root ("/") dentry for the inode given. The inode is
1411 * instantiated and returned. %NULL is returned if there is insufficient
1412 * memory or the inode passed is %NULL.
1415 struct dentry * d_alloc_root(struct inode * root_inode)
1417 struct dentry *res = NULL;
1419 if (root_inode) {
1420 static const struct qstr name = { .name = "/", .len = 1 };
1422 res = __d_alloc(root_inode->i_sb, &name);
1423 if (res)
1424 d_instantiate(res, root_inode);
1426 return res;
1428 EXPORT_SYMBOL(d_alloc_root);
1430 static struct dentry * __d_find_any_alias(struct inode *inode)
1432 struct dentry *alias;
1434 if (list_empty(&inode->i_dentry))
1435 return NULL;
1436 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1437 __dget(alias);
1438 return alias;
1441 static struct dentry * d_find_any_alias(struct inode *inode)
1443 struct dentry *de;
1445 spin_lock(&inode->i_lock);
1446 de = __d_find_any_alias(inode);
1447 spin_unlock(&inode->i_lock);
1448 return de;
1453 * d_obtain_alias - find or allocate a dentry for a given inode
1454 * @inode: inode to allocate the dentry for
1456 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1457 * similar open by handle operations. The returned dentry may be anonymous,
1458 * or may have a full name (if the inode was already in the cache).
1460 * When called on a directory inode, we must ensure that the inode only ever
1461 * has one dentry. If a dentry is found, that is returned instead of
1462 * allocating a new one.
1464 * On successful return, the reference to the inode has been transferred
1465 * to the dentry. In case of an error the reference on the inode is released.
1466 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1467 * be passed in and will be the error will be propagate to the return value,
1468 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1470 struct dentry *d_obtain_alias(struct inode *inode)
1472 static const struct qstr anonstring = { .name = "" };
1473 struct dentry *tmp;
1474 struct dentry *res;
1476 if (!inode)
1477 return ERR_PTR(-ESTALE);
1478 if (IS_ERR(inode))
1479 return ERR_CAST(inode);
1481 res = d_find_any_alias(inode);
1482 if (res)
1483 goto out_iput;
1485 tmp = __d_alloc(inode->i_sb, &anonstring);
1486 if (!tmp) {
1487 res = ERR_PTR(-ENOMEM);
1488 goto out_iput;
1491 spin_lock(&inode->i_lock);
1492 res = __d_find_any_alias(inode);
1493 if (res) {
1494 spin_unlock(&inode->i_lock);
1495 dput(tmp);
1496 goto out_iput;
1499 /* attach a disconnected dentry */
1500 spin_lock(&tmp->d_lock);
1501 tmp->d_inode = inode;
1502 tmp->d_flags |= DCACHE_DISCONNECTED;
1503 list_add(&tmp->d_alias, &inode->i_dentry);
1504 hlist_bl_lock(&tmp->d_sb->s_anon);
1505 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1506 hlist_bl_unlock(&tmp->d_sb->s_anon);
1507 spin_unlock(&tmp->d_lock);
1508 spin_unlock(&inode->i_lock);
1509 security_d_instantiate(tmp, inode);
1511 return tmp;
1513 out_iput:
1514 if (res && !IS_ERR(res))
1515 security_d_instantiate(res, inode);
1516 iput(inode);
1517 return res;
1519 EXPORT_SYMBOL(d_obtain_alias);
1522 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1523 * @inode: the inode which may have a disconnected dentry
1524 * @dentry: a negative dentry which we want to point to the inode.
1526 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1527 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1528 * and return it, else simply d_add the inode to the dentry and return NULL.
1530 * This is needed in the lookup routine of any filesystem that is exportable
1531 * (via knfsd) so that we can build dcache paths to directories effectively.
1533 * If a dentry was found and moved, then it is returned. Otherwise NULL
1534 * is returned. This matches the expected return value of ->lookup.
1537 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1539 struct dentry *new = NULL;
1541 if (IS_ERR(inode))
1542 return ERR_CAST(inode);
1544 if (inode && S_ISDIR(inode->i_mode)) {
1545 spin_lock(&inode->i_lock);
1546 new = __d_find_alias(inode, 1);
1547 if (new) {
1548 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1549 spin_unlock(&inode->i_lock);
1550 security_d_instantiate(new, inode);
1551 d_move(new, dentry);
1552 iput(inode);
1553 } else {
1554 /* already taking inode->i_lock, so d_add() by hand */
1555 __d_instantiate(dentry, inode);
1556 spin_unlock(&inode->i_lock);
1557 security_d_instantiate(dentry, inode);
1558 d_rehash(dentry);
1560 } else
1561 d_add(dentry, inode);
1562 return new;
1564 EXPORT_SYMBOL(d_splice_alias);
1567 * d_add_ci - lookup or allocate new dentry with case-exact name
1568 * @inode: the inode case-insensitive lookup has found
1569 * @dentry: the negative dentry that was passed to the parent's lookup func
1570 * @name: the case-exact name to be associated with the returned dentry
1572 * This is to avoid filling the dcache with case-insensitive names to the
1573 * same inode, only the actual correct case is stored in the dcache for
1574 * case-insensitive filesystems.
1576 * For a case-insensitive lookup match and if the the case-exact dentry
1577 * already exists in in the dcache, use it and return it.
1579 * If no entry exists with the exact case name, allocate new dentry with
1580 * the exact case, and return the spliced entry.
1582 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1583 struct qstr *name)
1585 int error;
1586 struct dentry *found;
1587 struct dentry *new;
1590 * First check if a dentry matching the name already exists,
1591 * if not go ahead and create it now.
1593 found = d_hash_and_lookup(dentry->d_parent, name);
1594 if (!found) {
1595 new = d_alloc(dentry->d_parent, name);
1596 if (!new) {
1597 error = -ENOMEM;
1598 goto err_out;
1601 found = d_splice_alias(inode, new);
1602 if (found) {
1603 dput(new);
1604 return found;
1606 return new;
1610 * If a matching dentry exists, and it's not negative use it.
1612 * Decrement the reference count to balance the iget() done
1613 * earlier on.
1615 if (found->d_inode) {
1616 if (unlikely(found->d_inode != inode)) {
1617 /* This can't happen because bad inodes are unhashed. */
1618 BUG_ON(!is_bad_inode(inode));
1619 BUG_ON(!is_bad_inode(found->d_inode));
1621 iput(inode);
1622 return found;
1626 * We are going to instantiate this dentry, unhash it and clear the
1627 * lookup flag so we can do that.
1629 if (unlikely(d_need_lookup(found)))
1630 d_clear_need_lookup(found);
1633 * Negative dentry: instantiate it unless the inode is a directory and
1634 * already has a dentry.
1636 new = d_splice_alias(inode, found);
1637 if (new) {
1638 dput(found);
1639 found = new;
1641 return found;
1643 err_out:
1644 iput(inode);
1645 return ERR_PTR(error);
1647 EXPORT_SYMBOL(d_add_ci);
1650 * __d_lookup_rcu - search for a dentry (racy, store-free)
1651 * @parent: parent dentry
1652 * @name: qstr of name we wish to find
1653 * @seq: returns d_seq value at the point where the dentry was found
1654 * @inode: returns dentry->d_inode when the inode was found valid.
1655 * Returns: dentry, or NULL
1657 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1658 * resolution (store-free path walking) design described in
1659 * Documentation/filesystems/path-lookup.txt.
1661 * This is not to be used outside core vfs.
1663 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1664 * held, and rcu_read_lock held. The returned dentry must not be stored into
1665 * without taking d_lock and checking d_seq sequence count against @seq
1666 * returned here.
1668 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1669 * function.
1671 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1672 * the returned dentry, so long as its parent's seqlock is checked after the
1673 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1674 * is formed, giving integrity down the path walk.
1676 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1677 unsigned *seq, struct inode **inode)
1679 unsigned int len = name->len;
1680 unsigned int hash = name->hash;
1681 const unsigned char *str = name->name;
1682 struct hlist_bl_head *b = d_hash(parent, hash);
1683 struct hlist_bl_node *node;
1684 struct dentry *dentry;
1687 * Note: There is significant duplication with __d_lookup_rcu which is
1688 * required to prevent single threaded performance regressions
1689 * especially on architectures where smp_rmb (in seqcounts) are costly.
1690 * Keep the two functions in sync.
1694 * The hash list is protected using RCU.
1696 * Carefully use d_seq when comparing a candidate dentry, to avoid
1697 * races with d_move().
1699 * It is possible that concurrent renames can mess up our list
1700 * walk here and result in missing our dentry, resulting in the
1701 * false-negative result. d_lookup() protects against concurrent
1702 * renames using rename_lock seqlock.
1704 * See Documentation/filesystems/path-lookup.txt for more details.
1706 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1707 struct inode *i;
1708 const char *tname;
1709 int tlen;
1711 if (dentry->d_name.hash != hash)
1712 continue;
1714 seqretry:
1715 *seq = read_seqcount_begin(&dentry->d_seq);
1716 if (dentry->d_parent != parent)
1717 continue;
1718 if (d_unhashed(dentry))
1719 continue;
1720 tlen = dentry->d_name.len;
1721 tname = dentry->d_name.name;
1722 i = dentry->d_inode;
1723 prefetch(tname);
1725 * This seqcount check is required to ensure name and
1726 * len are loaded atomically, so as not to walk off the
1727 * edge of memory when walking. If we could load this
1728 * atomically some other way, we could drop this check.
1730 if (read_seqcount_retry(&dentry->d_seq, *seq))
1731 goto seqretry;
1732 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1733 if (parent->d_op->d_compare(parent, *inode,
1734 dentry, i,
1735 tlen, tname, name))
1736 continue;
1737 } else {
1738 if (dentry_cmp(tname, tlen, str, len))
1739 continue;
1742 * No extra seqcount check is required after the name
1743 * compare. The caller must perform a seqcount check in
1744 * order to do anything useful with the returned dentry
1745 * anyway.
1747 *inode = i;
1748 return dentry;
1750 return NULL;
1754 * d_lookup - search for a dentry
1755 * @parent: parent dentry
1756 * @name: qstr of name we wish to find
1757 * Returns: dentry, or NULL
1759 * d_lookup searches the children of the parent dentry for the name in
1760 * question. If the dentry is found its reference count is incremented and the
1761 * dentry is returned. The caller must use dput to free the entry when it has
1762 * finished using it. %NULL is returned if the dentry does not exist.
1764 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1766 struct dentry *dentry;
1767 unsigned seq;
1769 do {
1770 seq = read_seqbegin(&rename_lock);
1771 dentry = __d_lookup(parent, name);
1772 if (dentry)
1773 break;
1774 } while (read_seqretry(&rename_lock, seq));
1775 return dentry;
1777 EXPORT_SYMBOL(d_lookup);
1780 * __d_lookup - search for a dentry (racy)
1781 * @parent: parent dentry
1782 * @name: qstr of name we wish to find
1783 * Returns: dentry, or NULL
1785 * __d_lookup is like d_lookup, however it may (rarely) return a
1786 * false-negative result due to unrelated rename activity.
1788 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1789 * however it must be used carefully, eg. with a following d_lookup in
1790 * the case of failure.
1792 * __d_lookup callers must be commented.
1794 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1796 unsigned int len = name->len;
1797 unsigned int hash = name->hash;
1798 const unsigned char *str = name->name;
1799 struct hlist_bl_head *b = d_hash(parent, hash);
1800 struct hlist_bl_node *node;
1801 struct dentry *found = NULL;
1802 struct dentry *dentry;
1805 * Note: There is significant duplication with __d_lookup_rcu which is
1806 * required to prevent single threaded performance regressions
1807 * especially on architectures where smp_rmb (in seqcounts) are costly.
1808 * Keep the two functions in sync.
1812 * The hash list is protected using RCU.
1814 * Take d_lock when comparing a candidate dentry, to avoid races
1815 * with d_move().
1817 * It is possible that concurrent renames can mess up our list
1818 * walk here and result in missing our dentry, resulting in the
1819 * false-negative result. d_lookup() protects against concurrent
1820 * renames using rename_lock seqlock.
1822 * See Documentation/filesystems/path-lookup.txt for more details.
1824 rcu_read_lock();
1826 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1827 const char *tname;
1828 int tlen;
1830 if (dentry->d_name.hash != hash)
1831 continue;
1833 spin_lock(&dentry->d_lock);
1834 if (dentry->d_parent != parent)
1835 goto next;
1836 if (d_unhashed(dentry))
1837 goto next;
1840 * It is safe to compare names since d_move() cannot
1841 * change the qstr (protected by d_lock).
1843 tlen = dentry->d_name.len;
1844 tname = dentry->d_name.name;
1845 if (parent->d_flags & DCACHE_OP_COMPARE) {
1846 if (parent->d_op->d_compare(parent, parent->d_inode,
1847 dentry, dentry->d_inode,
1848 tlen, tname, name))
1849 goto next;
1850 } else {
1851 if (dentry_cmp(tname, tlen, str, len))
1852 goto next;
1855 dentry->d_count++;
1856 found = dentry;
1857 spin_unlock(&dentry->d_lock);
1858 break;
1859 next:
1860 spin_unlock(&dentry->d_lock);
1862 rcu_read_unlock();
1864 return found;
1868 * d_hash_and_lookup - hash the qstr then search for a dentry
1869 * @dir: Directory to search in
1870 * @name: qstr of name we wish to find
1872 * On hash failure or on lookup failure NULL is returned.
1874 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1876 struct dentry *dentry = NULL;
1879 * Check for a fs-specific hash function. Note that we must
1880 * calculate the standard hash first, as the d_op->d_hash()
1881 * routine may choose to leave the hash value unchanged.
1883 name->hash = full_name_hash(name->name, name->len);
1884 if (dir->d_flags & DCACHE_OP_HASH) {
1885 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1886 goto out;
1888 dentry = d_lookup(dir, name);
1889 out:
1890 return dentry;
1894 * d_validate - verify dentry provided from insecure source (deprecated)
1895 * @dentry: The dentry alleged to be valid child of @dparent
1896 * @dparent: The parent dentry (known to be valid)
1898 * An insecure source has sent us a dentry, here we verify it and dget() it.
1899 * This is used by ncpfs in its readdir implementation.
1900 * Zero is returned in the dentry is invalid.
1902 * This function is slow for big directories, and deprecated, do not use it.
1904 int d_validate(struct dentry *dentry, struct dentry *dparent)
1906 struct dentry *child;
1908 spin_lock(&dparent->d_lock);
1909 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1910 if (dentry == child) {
1911 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1912 __dget_dlock(dentry);
1913 spin_unlock(&dentry->d_lock);
1914 spin_unlock(&dparent->d_lock);
1915 return 1;
1918 spin_unlock(&dparent->d_lock);
1920 return 0;
1922 EXPORT_SYMBOL(d_validate);
1925 * When a file is deleted, we have two options:
1926 * - turn this dentry into a negative dentry
1927 * - unhash this dentry and free it.
1929 * Usually, we want to just turn this into
1930 * a negative dentry, but if anybody else is
1931 * currently using the dentry or the inode
1932 * we can't do that and we fall back on removing
1933 * it from the hash queues and waiting for
1934 * it to be deleted later when it has no users
1938 * d_delete - delete a dentry
1939 * @dentry: The dentry to delete
1941 * Turn the dentry into a negative dentry if possible, otherwise
1942 * remove it from the hash queues so it can be deleted later
1945 void d_delete(struct dentry * dentry)
1947 struct inode *inode;
1948 int isdir = 0;
1950 * Are we the only user?
1952 again:
1953 spin_lock(&dentry->d_lock);
1954 inode = dentry->d_inode;
1955 isdir = S_ISDIR(inode->i_mode);
1956 if (dentry->d_count == 1) {
1957 if (inode && !spin_trylock(&inode->i_lock)) {
1958 spin_unlock(&dentry->d_lock);
1959 cpu_relax();
1960 goto again;
1962 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1963 dentry_unlink_inode(dentry);
1964 fsnotify_nameremove(dentry, isdir);
1965 return;
1968 if (!d_unhashed(dentry))
1969 __d_drop(dentry);
1971 spin_unlock(&dentry->d_lock);
1973 fsnotify_nameremove(dentry, isdir);
1975 EXPORT_SYMBOL(d_delete);
1977 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
1979 BUG_ON(!d_unhashed(entry));
1980 hlist_bl_lock(b);
1981 entry->d_flags |= DCACHE_RCUACCESS;
1982 hlist_bl_add_head_rcu(&entry->d_hash, b);
1983 hlist_bl_unlock(b);
1986 static void _d_rehash(struct dentry * entry)
1988 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1992 * d_rehash - add an entry back to the hash
1993 * @entry: dentry to add to the hash
1995 * Adds a dentry to the hash according to its name.
1998 void d_rehash(struct dentry * entry)
2000 spin_lock(&entry->d_lock);
2001 _d_rehash(entry);
2002 spin_unlock(&entry->d_lock);
2004 EXPORT_SYMBOL(d_rehash);
2007 * dentry_update_name_case - update case insensitive dentry with a new name
2008 * @dentry: dentry to be updated
2009 * @name: new name
2011 * Update a case insensitive dentry with new case of name.
2013 * dentry must have been returned by d_lookup with name @name. Old and new
2014 * name lengths must match (ie. no d_compare which allows mismatched name
2015 * lengths).
2017 * Parent inode i_mutex must be held over d_lookup and into this call (to
2018 * keep renames and concurrent inserts, and readdir(2) away).
2020 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2022 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2023 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2025 spin_lock(&dentry->d_lock);
2026 write_seqcount_begin(&dentry->d_seq);
2027 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2028 write_seqcount_end(&dentry->d_seq);
2029 spin_unlock(&dentry->d_lock);
2031 EXPORT_SYMBOL(dentry_update_name_case);
2033 static void switch_names(struct dentry *dentry, struct dentry *target)
2035 if (dname_external(target)) {
2036 if (dname_external(dentry)) {
2038 * Both external: swap the pointers
2040 swap(target->d_name.name, dentry->d_name.name);
2041 } else {
2043 * dentry:internal, target:external. Steal target's
2044 * storage and make target internal.
2046 memcpy(target->d_iname, dentry->d_name.name,
2047 dentry->d_name.len + 1);
2048 dentry->d_name.name = target->d_name.name;
2049 target->d_name.name = target->d_iname;
2051 } else {
2052 if (dname_external(dentry)) {
2054 * dentry:external, target:internal. Give dentry's
2055 * storage to target and make dentry internal
2057 memcpy(dentry->d_iname, target->d_name.name,
2058 target->d_name.len + 1);
2059 target->d_name.name = dentry->d_name.name;
2060 dentry->d_name.name = dentry->d_iname;
2061 } else {
2063 * Both are internal. Just copy target to dentry
2065 memcpy(dentry->d_iname, target->d_name.name,
2066 target->d_name.len + 1);
2067 dentry->d_name.len = target->d_name.len;
2068 return;
2071 swap(dentry->d_name.len, target->d_name.len);
2074 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2077 * XXXX: do we really need to take target->d_lock?
2079 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2080 spin_lock(&target->d_parent->d_lock);
2081 else {
2082 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2083 spin_lock(&dentry->d_parent->d_lock);
2084 spin_lock_nested(&target->d_parent->d_lock,
2085 DENTRY_D_LOCK_NESTED);
2086 } else {
2087 spin_lock(&target->d_parent->d_lock);
2088 spin_lock_nested(&dentry->d_parent->d_lock,
2089 DENTRY_D_LOCK_NESTED);
2092 if (target < dentry) {
2093 spin_lock_nested(&target->d_lock, 2);
2094 spin_lock_nested(&dentry->d_lock, 3);
2095 } else {
2096 spin_lock_nested(&dentry->d_lock, 2);
2097 spin_lock_nested(&target->d_lock, 3);
2101 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2102 struct dentry *target)
2104 if (target->d_parent != dentry->d_parent)
2105 spin_unlock(&dentry->d_parent->d_lock);
2106 if (target->d_parent != target)
2107 spin_unlock(&target->d_parent->d_lock);
2111 * When switching names, the actual string doesn't strictly have to
2112 * be preserved in the target - because we're dropping the target
2113 * anyway. As such, we can just do a simple memcpy() to copy over
2114 * the new name before we switch.
2116 * Note that we have to be a lot more careful about getting the hash
2117 * switched - we have to switch the hash value properly even if it
2118 * then no longer matches the actual (corrupted) string of the target.
2119 * The hash value has to match the hash queue that the dentry is on..
2122 * __d_move - move a dentry
2123 * @dentry: entry to move
2124 * @target: new dentry
2126 * Update the dcache to reflect the move of a file name. Negative
2127 * dcache entries should not be moved in this way. Caller must hold
2128 * rename_lock, the i_mutex of the source and target directories,
2129 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2131 static void __d_move(struct dentry * dentry, struct dentry * target)
2133 if (!dentry->d_inode)
2134 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2136 BUG_ON(d_ancestor(dentry, target));
2137 BUG_ON(d_ancestor(target, dentry));
2139 dentry_lock_for_move(dentry, target);
2141 write_seqcount_begin(&dentry->d_seq);
2142 write_seqcount_begin(&target->d_seq);
2144 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2147 * Move the dentry to the target hash queue. Don't bother checking
2148 * for the same hash queue because of how unlikely it is.
2150 __d_drop(dentry);
2151 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2153 /* Unhash the target: dput() will then get rid of it */
2154 __d_drop(target);
2156 list_del(&dentry->d_u.d_child);
2157 list_del(&target->d_u.d_child);
2159 /* Switch the names.. */
2160 switch_names(dentry, target);
2161 swap(dentry->d_name.hash, target->d_name.hash);
2163 /* ... and switch the parents */
2164 if (IS_ROOT(dentry)) {
2165 dentry->d_parent = target->d_parent;
2166 target->d_parent = target;
2167 INIT_LIST_HEAD(&target->d_u.d_child);
2168 } else {
2169 swap(dentry->d_parent, target->d_parent);
2171 /* And add them back to the (new) parent lists */
2172 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2175 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2177 write_seqcount_end(&target->d_seq);
2178 write_seqcount_end(&dentry->d_seq);
2180 dentry_unlock_parents_for_move(dentry, target);
2181 spin_unlock(&target->d_lock);
2182 fsnotify_d_move(dentry);
2183 spin_unlock(&dentry->d_lock);
2187 * d_move - move a dentry
2188 * @dentry: entry to move
2189 * @target: new dentry
2191 * Update the dcache to reflect the move of a file name. Negative
2192 * dcache entries should not be moved in this way. See the locking
2193 * requirements for __d_move.
2195 void d_move(struct dentry *dentry, struct dentry *target)
2197 write_seqlock(&rename_lock);
2198 __d_move(dentry, target);
2199 write_sequnlock(&rename_lock);
2201 EXPORT_SYMBOL(d_move);
2204 * d_ancestor - search for an ancestor
2205 * @p1: ancestor dentry
2206 * @p2: child dentry
2208 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2209 * an ancestor of p2, else NULL.
2211 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2213 struct dentry *p;
2215 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2216 if (p->d_parent == p1)
2217 return p;
2219 return NULL;
2223 * This helper attempts to cope with remotely renamed directories
2225 * It assumes that the caller is already holding
2226 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2228 * Note: If ever the locking in lock_rename() changes, then please
2229 * remember to update this too...
2231 static struct dentry *__d_unalias(struct inode *inode,
2232 struct dentry *dentry, struct dentry *alias)
2234 struct mutex *m1 = NULL, *m2 = NULL;
2235 struct dentry *ret;
2237 /* If alias and dentry share a parent, then no extra locks required */
2238 if (alias->d_parent == dentry->d_parent)
2239 goto out_unalias;
2241 /* See lock_rename() */
2242 ret = ERR_PTR(-EBUSY);
2243 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2244 goto out_err;
2245 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2246 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2247 goto out_err;
2248 m2 = &alias->d_parent->d_inode->i_mutex;
2249 out_unalias:
2250 __d_move(alias, dentry);
2251 ret = alias;
2252 out_err:
2253 spin_unlock(&inode->i_lock);
2254 if (m2)
2255 mutex_unlock(m2);
2256 if (m1)
2257 mutex_unlock(m1);
2258 return ret;
2262 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2263 * named dentry in place of the dentry to be replaced.
2264 * returns with anon->d_lock held!
2266 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2268 struct dentry *dparent, *aparent;
2270 dentry_lock_for_move(anon, dentry);
2272 write_seqcount_begin(&dentry->d_seq);
2273 write_seqcount_begin(&anon->d_seq);
2275 dparent = dentry->d_parent;
2276 aparent = anon->d_parent;
2278 switch_names(dentry, anon);
2279 swap(dentry->d_name.hash, anon->d_name.hash);
2281 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2282 list_del(&dentry->d_u.d_child);
2283 if (!IS_ROOT(dentry))
2284 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2285 else
2286 INIT_LIST_HEAD(&dentry->d_u.d_child);
2288 anon->d_parent = (dparent == dentry) ? anon : dparent;
2289 list_del(&anon->d_u.d_child);
2290 if (!IS_ROOT(anon))
2291 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2292 else
2293 INIT_LIST_HEAD(&anon->d_u.d_child);
2295 write_seqcount_end(&dentry->d_seq);
2296 write_seqcount_end(&anon->d_seq);
2298 dentry_unlock_parents_for_move(anon, dentry);
2299 spin_unlock(&dentry->d_lock);
2301 /* anon->d_lock still locked, returns locked */
2302 anon->d_flags &= ~DCACHE_DISCONNECTED;
2306 * d_materialise_unique - introduce an inode into the tree
2307 * @dentry: candidate dentry
2308 * @inode: inode to bind to the dentry, to which aliases may be attached
2310 * Introduces an dentry into the tree, substituting an extant disconnected
2311 * root directory alias in its place if there is one. Caller must hold the
2312 * i_mutex of the parent directory.
2314 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2316 struct dentry *actual;
2318 BUG_ON(!d_unhashed(dentry));
2320 if (!inode) {
2321 actual = dentry;
2322 __d_instantiate(dentry, NULL);
2323 d_rehash(actual);
2324 goto out_nolock;
2327 spin_lock(&inode->i_lock);
2329 if (S_ISDIR(inode->i_mode)) {
2330 struct dentry *alias;
2332 /* Does an aliased dentry already exist? */
2333 alias = __d_find_alias(inode, 0);
2334 if (alias) {
2335 actual = alias;
2336 write_seqlock(&rename_lock);
2338 if (d_ancestor(alias, dentry)) {
2339 /* Check for loops */
2340 actual = ERR_PTR(-ELOOP);
2341 } else if (IS_ROOT(alias)) {
2342 /* Is this an anonymous mountpoint that we
2343 * could splice into our tree? */
2344 __d_materialise_dentry(dentry, alias);
2345 write_sequnlock(&rename_lock);
2346 __d_drop(alias);
2347 goto found;
2348 } else {
2349 /* Nope, but we must(!) avoid directory
2350 * aliasing */
2351 actual = __d_unalias(inode, dentry, alias);
2353 write_sequnlock(&rename_lock);
2354 if (IS_ERR(actual))
2355 dput(alias);
2356 goto out_nolock;
2360 /* Add a unique reference */
2361 actual = __d_instantiate_unique(dentry, inode);
2362 if (!actual)
2363 actual = dentry;
2364 else
2365 BUG_ON(!d_unhashed(actual));
2367 spin_lock(&actual->d_lock);
2368 found:
2369 _d_rehash(actual);
2370 spin_unlock(&actual->d_lock);
2371 spin_unlock(&inode->i_lock);
2372 out_nolock:
2373 if (actual == dentry) {
2374 security_d_instantiate(dentry, inode);
2375 return NULL;
2378 iput(inode);
2379 return actual;
2381 EXPORT_SYMBOL_GPL(d_materialise_unique);
2383 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2385 *buflen -= namelen;
2386 if (*buflen < 0)
2387 return -ENAMETOOLONG;
2388 *buffer -= namelen;
2389 memcpy(*buffer, str, namelen);
2390 return 0;
2393 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2395 return prepend(buffer, buflen, name->name, name->len);
2399 * prepend_path - Prepend path string to a buffer
2400 * @path: the dentry/vfsmount to report
2401 * @root: root vfsmnt/dentry (may be modified by this function)
2402 * @buffer: pointer to the end of the buffer
2403 * @buflen: pointer to buffer length
2405 * Caller holds the rename_lock.
2407 * If path is not reachable from the supplied root, then the value of
2408 * root is changed (without modifying refcounts).
2410 static int prepend_path(const struct path *path, struct path *root,
2411 char **buffer, int *buflen)
2413 struct dentry *dentry = path->dentry;
2414 struct vfsmount *vfsmnt = path->mnt;
2415 bool slash = false;
2416 int error = 0;
2418 br_read_lock(vfsmount_lock);
2419 while (dentry != root->dentry || vfsmnt != root->mnt) {
2420 struct dentry * parent;
2422 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2423 /* Global root? */
2424 if (vfsmnt->mnt_parent == vfsmnt) {
2425 goto global_root;
2427 dentry = vfsmnt->mnt_mountpoint;
2428 vfsmnt = vfsmnt->mnt_parent;
2429 continue;
2431 parent = dentry->d_parent;
2432 prefetch(parent);
2433 spin_lock(&dentry->d_lock);
2434 error = prepend_name(buffer, buflen, &dentry->d_name);
2435 spin_unlock(&dentry->d_lock);
2436 if (!error)
2437 error = prepend(buffer, buflen, "/", 1);
2438 if (error)
2439 break;
2441 slash = true;
2442 dentry = parent;
2445 out:
2446 if (!error && !slash)
2447 error = prepend(buffer, buflen, "/", 1);
2449 br_read_unlock(vfsmount_lock);
2450 return error;
2452 global_root:
2454 * Filesystems needing to implement special "root names"
2455 * should do so with ->d_dname()
2457 if (IS_ROOT(dentry) &&
2458 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2459 WARN(1, "Root dentry has weird name <%.*s>\n",
2460 (int) dentry->d_name.len, dentry->d_name.name);
2462 root->mnt = vfsmnt;
2463 root->dentry = dentry;
2464 goto out;
2468 * __d_path - return the path of a dentry
2469 * @path: the dentry/vfsmount to report
2470 * @root: root vfsmnt/dentry (may be modified by this function)
2471 * @buf: buffer to return value in
2472 * @buflen: buffer length
2474 * Convert a dentry into an ASCII path name.
2476 * Returns a pointer into the buffer or an error code if the
2477 * path was too long.
2479 * "buflen" should be positive.
2481 * If path is not reachable from the supplied root, then the value of
2482 * root is changed (without modifying refcounts).
2484 char *__d_path(const struct path *path, struct path *root,
2485 char *buf, int buflen)
2487 char *res = buf + buflen;
2488 int error;
2490 prepend(&res, &buflen, "\0", 1);
2491 write_seqlock(&rename_lock);
2492 error = prepend_path(path, root, &res, &buflen);
2493 write_sequnlock(&rename_lock);
2495 if (error)
2496 return ERR_PTR(error);
2497 return res;
2501 * same as __d_path but appends "(deleted)" for unlinked files.
2503 static int path_with_deleted(const struct path *path, struct path *root,
2504 char **buf, int *buflen)
2506 prepend(buf, buflen, "\0", 1);
2507 if (d_unlinked(path->dentry)) {
2508 int error = prepend(buf, buflen, " (deleted)", 10);
2509 if (error)
2510 return error;
2513 return prepend_path(path, root, buf, buflen);
2516 static int prepend_unreachable(char **buffer, int *buflen)
2518 return prepend(buffer, buflen, "(unreachable)", 13);
2522 * d_path - return the path of a dentry
2523 * @path: path to report
2524 * @buf: buffer to return value in
2525 * @buflen: buffer length
2527 * Convert a dentry into an ASCII path name. If the entry has been deleted
2528 * the string " (deleted)" is appended. Note that this is ambiguous.
2530 * Returns a pointer into the buffer or an error code if the path was
2531 * too long. Note: Callers should use the returned pointer, not the passed
2532 * in buffer, to use the name! The implementation often starts at an offset
2533 * into the buffer, and may leave 0 bytes at the start.
2535 * "buflen" should be positive.
2537 char *d_path(const struct path *path, char *buf, int buflen)
2539 char *res = buf + buflen;
2540 struct path root;
2541 struct path tmp;
2542 int error;
2545 * We have various synthetic filesystems that never get mounted. On
2546 * these filesystems dentries are never used for lookup purposes, and
2547 * thus don't need to be hashed. They also don't need a name until a
2548 * user wants to identify the object in /proc/pid/fd/. The little hack
2549 * below allows us to generate a name for these objects on demand:
2551 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2552 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2554 get_fs_root(current->fs, &root);
2555 write_seqlock(&rename_lock);
2556 tmp = root;
2557 error = path_with_deleted(path, &tmp, &res, &buflen);
2558 if (error)
2559 res = ERR_PTR(error);
2560 write_sequnlock(&rename_lock);
2561 path_put(&root);
2562 return res;
2564 EXPORT_SYMBOL(d_path);
2567 * d_path_with_unreachable - return the path of a dentry
2568 * @path: path to report
2569 * @buf: buffer to return value in
2570 * @buflen: buffer length
2572 * The difference from d_path() is that this prepends "(unreachable)"
2573 * to paths which are unreachable from the current process' root.
2575 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2577 char *res = buf + buflen;
2578 struct path root;
2579 struct path tmp;
2580 int error;
2582 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2583 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2585 get_fs_root(current->fs, &root);
2586 write_seqlock(&rename_lock);
2587 tmp = root;
2588 error = path_with_deleted(path, &tmp, &res, &buflen);
2589 if (!error && !path_equal(&tmp, &root))
2590 error = prepend_unreachable(&res, &buflen);
2591 write_sequnlock(&rename_lock);
2592 path_put(&root);
2593 if (error)
2594 res = ERR_PTR(error);
2596 return res;
2600 * Helper function for dentry_operations.d_dname() members
2602 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2603 const char *fmt, ...)
2605 va_list args;
2606 char temp[64];
2607 int sz;
2609 va_start(args, fmt);
2610 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2611 va_end(args);
2613 if (sz > sizeof(temp) || sz > buflen)
2614 return ERR_PTR(-ENAMETOOLONG);
2616 buffer += buflen - sz;
2617 return memcpy(buffer, temp, sz);
2621 * Write full pathname from the root of the filesystem into the buffer.
2623 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2625 char *end = buf + buflen;
2626 char *retval;
2628 prepend(&end, &buflen, "\0", 1);
2629 if (buflen < 1)
2630 goto Elong;
2631 /* Get '/' right */
2632 retval = end-1;
2633 *retval = '/';
2635 while (!IS_ROOT(dentry)) {
2636 struct dentry *parent = dentry->d_parent;
2637 int error;
2639 prefetch(parent);
2640 spin_lock(&dentry->d_lock);
2641 error = prepend_name(&end, &buflen, &dentry->d_name);
2642 spin_unlock(&dentry->d_lock);
2643 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2644 goto Elong;
2646 retval = end;
2647 dentry = parent;
2649 return retval;
2650 Elong:
2651 return ERR_PTR(-ENAMETOOLONG);
2654 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2656 char *retval;
2658 write_seqlock(&rename_lock);
2659 retval = __dentry_path(dentry, buf, buflen);
2660 write_sequnlock(&rename_lock);
2662 return retval;
2664 EXPORT_SYMBOL(dentry_path_raw);
2666 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2668 char *p = NULL;
2669 char *retval;
2671 write_seqlock(&rename_lock);
2672 if (d_unlinked(dentry)) {
2673 p = buf + buflen;
2674 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2675 goto Elong;
2676 buflen++;
2678 retval = __dentry_path(dentry, buf, buflen);
2679 write_sequnlock(&rename_lock);
2680 if (!IS_ERR(retval) && p)
2681 *p = '/'; /* restore '/' overriden with '\0' */
2682 return retval;
2683 Elong:
2684 return ERR_PTR(-ENAMETOOLONG);
2688 * NOTE! The user-level library version returns a
2689 * character pointer. The kernel system call just
2690 * returns the length of the buffer filled (which
2691 * includes the ending '\0' character), or a negative
2692 * error value. So libc would do something like
2694 * char *getcwd(char * buf, size_t size)
2696 * int retval;
2698 * retval = sys_getcwd(buf, size);
2699 * if (retval >= 0)
2700 * return buf;
2701 * errno = -retval;
2702 * return NULL;
2705 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2707 int error;
2708 struct path pwd, root;
2709 char *page = (char *) __get_free_page(GFP_USER);
2711 if (!page)
2712 return -ENOMEM;
2714 get_fs_root_and_pwd(current->fs, &root, &pwd);
2716 error = -ENOENT;
2717 write_seqlock(&rename_lock);
2718 if (!d_unlinked(pwd.dentry)) {
2719 unsigned long len;
2720 struct path tmp = root;
2721 char *cwd = page + PAGE_SIZE;
2722 int buflen = PAGE_SIZE;
2724 prepend(&cwd, &buflen, "\0", 1);
2725 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2726 write_sequnlock(&rename_lock);
2728 if (error)
2729 goto out;
2731 /* Unreachable from current root */
2732 if (!path_equal(&tmp, &root)) {
2733 error = prepend_unreachable(&cwd, &buflen);
2734 if (error)
2735 goto out;
2738 error = -ERANGE;
2739 len = PAGE_SIZE + page - cwd;
2740 if (len <= size) {
2741 error = len;
2742 if (copy_to_user(buf, cwd, len))
2743 error = -EFAULT;
2745 } else {
2746 write_sequnlock(&rename_lock);
2749 out:
2750 path_put(&pwd);
2751 path_put(&root);
2752 free_page((unsigned long) page);
2753 return error;
2757 * Test whether new_dentry is a subdirectory of old_dentry.
2759 * Trivially implemented using the dcache structure
2763 * is_subdir - is new dentry a subdirectory of old_dentry
2764 * @new_dentry: new dentry
2765 * @old_dentry: old dentry
2767 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2768 * Returns 0 otherwise.
2769 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2772 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2774 int result;
2775 unsigned seq;
2777 if (new_dentry == old_dentry)
2778 return 1;
2780 do {
2781 /* for restarting inner loop in case of seq retry */
2782 seq = read_seqbegin(&rename_lock);
2784 * Need rcu_readlock to protect against the d_parent trashing
2785 * due to d_move
2787 rcu_read_lock();
2788 if (d_ancestor(old_dentry, new_dentry))
2789 result = 1;
2790 else
2791 result = 0;
2792 rcu_read_unlock();
2793 } while (read_seqretry(&rename_lock, seq));
2795 return result;
2798 int path_is_under(struct path *path1, struct path *path2)
2800 struct vfsmount *mnt = path1->mnt;
2801 struct dentry *dentry = path1->dentry;
2802 int res;
2804 br_read_lock(vfsmount_lock);
2805 if (mnt != path2->mnt) {
2806 for (;;) {
2807 if (mnt->mnt_parent == mnt) {
2808 br_read_unlock(vfsmount_lock);
2809 return 0;
2811 if (mnt->mnt_parent == path2->mnt)
2812 break;
2813 mnt = mnt->mnt_parent;
2815 dentry = mnt->mnt_mountpoint;
2817 res = is_subdir(dentry, path2->dentry);
2818 br_read_unlock(vfsmount_lock);
2819 return res;
2821 EXPORT_SYMBOL(path_is_under);
2823 void d_genocide(struct dentry *root)
2825 struct dentry *this_parent;
2826 struct list_head *next;
2827 unsigned seq;
2828 int locked = 0;
2830 seq = read_seqbegin(&rename_lock);
2831 again:
2832 this_parent = root;
2833 spin_lock(&this_parent->d_lock);
2834 repeat:
2835 next = this_parent->d_subdirs.next;
2836 resume:
2837 while (next != &this_parent->d_subdirs) {
2838 struct list_head *tmp = next;
2839 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2840 next = tmp->next;
2842 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2843 if (d_unhashed(dentry) || !dentry->d_inode) {
2844 spin_unlock(&dentry->d_lock);
2845 continue;
2847 if (!list_empty(&dentry->d_subdirs)) {
2848 spin_unlock(&this_parent->d_lock);
2849 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2850 this_parent = dentry;
2851 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2852 goto repeat;
2854 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2855 dentry->d_flags |= DCACHE_GENOCIDE;
2856 dentry->d_count--;
2858 spin_unlock(&dentry->d_lock);
2860 if (this_parent != root) {
2861 struct dentry *child = this_parent;
2862 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2863 this_parent->d_flags |= DCACHE_GENOCIDE;
2864 this_parent->d_count--;
2866 this_parent = try_to_ascend(this_parent, locked, seq);
2867 if (!this_parent)
2868 goto rename_retry;
2869 next = child->d_u.d_child.next;
2870 goto resume;
2872 spin_unlock(&this_parent->d_lock);
2873 if (!locked && read_seqretry(&rename_lock, seq))
2874 goto rename_retry;
2875 if (locked)
2876 write_sequnlock(&rename_lock);
2877 return;
2879 rename_retry:
2880 locked = 1;
2881 write_seqlock(&rename_lock);
2882 goto again;
2886 * find_inode_number - check for dentry with name
2887 * @dir: directory to check
2888 * @name: Name to find.
2890 * Check whether a dentry already exists for the given name,
2891 * and return the inode number if it has an inode. Otherwise
2892 * 0 is returned.
2894 * This routine is used to post-process directory listings for
2895 * filesystems using synthetic inode numbers, and is necessary
2896 * to keep getcwd() working.
2899 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2901 struct dentry * dentry;
2902 ino_t ino = 0;
2904 dentry = d_hash_and_lookup(dir, name);
2905 if (dentry) {
2906 if (dentry->d_inode)
2907 ino = dentry->d_inode->i_ino;
2908 dput(dentry);
2910 return ino;
2912 EXPORT_SYMBOL(find_inode_number);
2914 static __initdata unsigned long dhash_entries;
2915 static int __init set_dhash_entries(char *str)
2917 if (!str)
2918 return 0;
2919 dhash_entries = simple_strtoul(str, &str, 0);
2920 return 1;
2922 __setup("dhash_entries=", set_dhash_entries);
2924 static void __init dcache_init_early(void)
2926 int loop;
2928 /* If hashes are distributed across NUMA nodes, defer
2929 * hash allocation until vmalloc space is available.
2931 if (hashdist)
2932 return;
2934 dentry_hashtable =
2935 alloc_large_system_hash("Dentry cache",
2936 sizeof(struct hlist_bl_head),
2937 dhash_entries,
2939 HASH_EARLY,
2940 &d_hash_shift,
2941 &d_hash_mask,
2944 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2945 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
2948 static void __init dcache_init(void)
2950 int loop;
2953 * A constructor could be added for stable state like the lists,
2954 * but it is probably not worth it because of the cache nature
2955 * of the dcache.
2957 dentry_cache = KMEM_CACHE(dentry,
2958 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2960 /* Hash may have been set up in dcache_init_early */
2961 if (!hashdist)
2962 return;
2964 dentry_hashtable =
2965 alloc_large_system_hash("Dentry cache",
2966 sizeof(struct hlist_bl_head),
2967 dhash_entries,
2970 &d_hash_shift,
2971 &d_hash_mask,
2974 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2975 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
2978 /* SLAB cache for __getname() consumers */
2979 struct kmem_cache *names_cachep __read_mostly;
2980 EXPORT_SYMBOL(names_cachep);
2982 EXPORT_SYMBOL(d_genocide);
2984 void __init vfs_caches_init_early(void)
2986 dcache_init_early();
2987 inode_init_early();
2990 void __init vfs_caches_init(unsigned long mempages)
2992 unsigned long reserve;
2994 /* Base hash sizes on available memory, with a reserve equal to
2995 150% of current kernel size */
2997 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2998 mempages -= reserve;
3000 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3001 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3003 dcache_init();
3004 inode_init();
3005 files_init(mempages);
3006 mnt_init();
3007 bdev_cache_init();
3008 chrdev_init();