Merge branch 'firewire-kernel-streaming' of git://git.alsa-project.org/alsa-kprivate
[firewire-audio.git] / fs / dcache.c
blob611ffe928c03c7f90aa1c9e0774fa60b86dd1982
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 "internal.h"
41 * Usage:
42 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects:
45 * - the dcache hash table
46 * s_anon bl list spinlock protects:
47 * - the s_anon list (see __d_drop)
48 * dcache_lru_lock protects:
49 * - the dcache lru lists and counters
50 * d_lock protects:
51 * - d_flags
52 * - d_name
53 * - d_lru
54 * - d_count
55 * - d_unhashed()
56 * - d_parent and d_subdirs
57 * - childrens' d_child and d_parent
58 * - d_alias, d_inode
60 * Ordering:
61 * dentry->d_inode->i_lock
62 * dentry->d_lock
63 * dcache_lru_lock
64 * dcache_hash_bucket lock
65 * s_anon lock
67 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock
69 * ...
70 * dentry->d_parent->d_lock
71 * dentry->d_lock
73 * If no ancestor relationship:
74 * if (dentry1 < dentry2)
75 * dentry1->d_lock
76 * dentry2->d_lock
78 int sysctl_vfs_cache_pressure __read_mostly = 100;
79 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
81 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
84 EXPORT_SYMBOL(rename_lock);
86 static struct kmem_cache *dentry_cache __read_mostly;
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
96 #define D_HASHBITS d_hash_shift
97 #define D_HASHMASK d_hash_mask
99 static unsigned int d_hash_mask __read_mostly;
100 static unsigned int d_hash_shift __read_mostly;
102 struct dcache_hash_bucket {
103 struct hlist_bl_head head;
105 static struct dcache_hash_bucket *dentry_hashtable __read_mostly;
107 static inline struct dcache_hash_bucket *d_hash(struct dentry *parent,
108 unsigned long hash)
110 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
111 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
112 return dentry_hashtable + (hash & D_HASHMASK);
115 static inline void spin_lock_bucket(struct dcache_hash_bucket *b)
117 bit_spin_lock(0, (unsigned long *)&b->head.first);
120 static inline void spin_unlock_bucket(struct dcache_hash_bucket *b)
122 __bit_spin_unlock(0, (unsigned long *)&b->head.first);
125 /* Statistics gathering. */
126 struct dentry_stat_t dentry_stat = {
127 .age_limit = 45,
130 static DEFINE_PER_CPU(unsigned int, nr_dentry);
132 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
133 static int get_nr_dentry(void)
135 int i;
136 int sum = 0;
137 for_each_possible_cpu(i)
138 sum += per_cpu(nr_dentry, i);
139 return sum < 0 ? 0 : sum;
142 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
143 size_t *lenp, loff_t *ppos)
145 dentry_stat.nr_dentry = get_nr_dentry();
146 return proc_dointvec(table, write, buffer, lenp, ppos);
148 #endif
150 static void __d_free(struct rcu_head *head)
152 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
154 WARN_ON(!list_empty(&dentry->d_alias));
155 if (dname_external(dentry))
156 kfree(dentry->d_name.name);
157 kmem_cache_free(dentry_cache, dentry);
161 * no locks, please.
163 static void d_free(struct dentry *dentry)
165 BUG_ON(dentry->d_count);
166 this_cpu_dec(nr_dentry);
167 if (dentry->d_op && dentry->d_op->d_release)
168 dentry->d_op->d_release(dentry);
170 /* if dentry was never inserted into hash, immediate free is OK */
171 if (hlist_bl_unhashed(&dentry->d_hash))
172 __d_free(&dentry->d_u.d_rcu);
173 else
174 call_rcu(&dentry->d_u.d_rcu, __d_free);
178 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
179 * @dentry: the target dentry
180 * After this call, in-progress rcu-walk path lookup will fail. This
181 * should be called after unhashing, and after changing d_inode (if
182 * the dentry has not already been unhashed).
184 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
186 assert_spin_locked(&dentry->d_lock);
187 /* Go through a barrier */
188 write_seqcount_barrier(&dentry->d_seq);
192 * Release the dentry's inode, using the filesystem
193 * d_iput() operation if defined. Dentry has no refcount
194 * and is unhashed.
196 static void dentry_iput(struct dentry * dentry)
197 __releases(dentry->d_lock)
198 __releases(dentry->d_inode->i_lock)
200 struct inode *inode = dentry->d_inode;
201 if (inode) {
202 dentry->d_inode = NULL;
203 list_del_init(&dentry->d_alias);
204 spin_unlock(&dentry->d_lock);
205 spin_unlock(&inode->i_lock);
206 if (!inode->i_nlink)
207 fsnotify_inoderemove(inode);
208 if (dentry->d_op && dentry->d_op->d_iput)
209 dentry->d_op->d_iput(dentry, inode);
210 else
211 iput(inode);
212 } else {
213 spin_unlock(&dentry->d_lock);
218 * Release the dentry's inode, using the filesystem
219 * d_iput() operation if defined. dentry remains in-use.
221 static void dentry_unlink_inode(struct dentry * dentry)
222 __releases(dentry->d_lock)
223 __releases(dentry->d_inode->i_lock)
225 struct inode *inode = dentry->d_inode;
226 dentry->d_inode = NULL;
227 list_del_init(&dentry->d_alias);
228 dentry_rcuwalk_barrier(dentry);
229 spin_unlock(&dentry->d_lock);
230 spin_unlock(&inode->i_lock);
231 if (!inode->i_nlink)
232 fsnotify_inoderemove(inode);
233 if (dentry->d_op && dentry->d_op->d_iput)
234 dentry->d_op->d_iput(dentry, inode);
235 else
236 iput(inode);
240 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
242 static void dentry_lru_add(struct dentry *dentry)
244 if (list_empty(&dentry->d_lru)) {
245 spin_lock(&dcache_lru_lock);
246 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
247 dentry->d_sb->s_nr_dentry_unused++;
248 dentry_stat.nr_unused++;
249 spin_unlock(&dcache_lru_lock);
253 static void __dentry_lru_del(struct dentry *dentry)
255 list_del_init(&dentry->d_lru);
256 dentry->d_sb->s_nr_dentry_unused--;
257 dentry_stat.nr_unused--;
260 static void dentry_lru_del(struct dentry *dentry)
262 if (!list_empty(&dentry->d_lru)) {
263 spin_lock(&dcache_lru_lock);
264 __dentry_lru_del(dentry);
265 spin_unlock(&dcache_lru_lock);
269 static void dentry_lru_move_tail(struct dentry *dentry)
271 spin_lock(&dcache_lru_lock);
272 if (list_empty(&dentry->d_lru)) {
273 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
274 dentry->d_sb->s_nr_dentry_unused++;
275 dentry_stat.nr_unused++;
276 } else {
277 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
279 spin_unlock(&dcache_lru_lock);
283 * d_kill - kill dentry and return parent
284 * @dentry: dentry to kill
285 * @parent: parent dentry
287 * The dentry must already be unhashed and removed from the LRU.
289 * If this is the root of the dentry tree, return NULL.
291 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
292 * d_kill.
294 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
295 __releases(dentry->d_lock)
296 __releases(parent->d_lock)
297 __releases(dentry->d_inode->i_lock)
299 dentry->d_parent = NULL;
300 list_del(&dentry->d_u.d_child);
301 if (parent)
302 spin_unlock(&parent->d_lock);
303 dentry_iput(dentry);
305 * dentry_iput drops the locks, at which point nobody (except
306 * transient RCU lookups) can reach this dentry.
308 d_free(dentry);
309 return parent;
313 * d_drop - drop a dentry
314 * @dentry: dentry to drop
316 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
317 * be found through a VFS lookup any more. Note that this is different from
318 * deleting the dentry - d_delete will try to mark the dentry negative if
319 * possible, giving a successful _negative_ lookup, while d_drop will
320 * just make the cache lookup fail.
322 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
323 * reason (NFS timeouts or autofs deletes).
325 * __d_drop requires dentry->d_lock.
327 void __d_drop(struct dentry *dentry)
329 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
330 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) {
331 bit_spin_lock(0,
332 (unsigned long *)&dentry->d_sb->s_anon.first);
333 dentry->d_flags |= DCACHE_UNHASHED;
334 hlist_bl_del_init(&dentry->d_hash);
335 __bit_spin_unlock(0,
336 (unsigned long *)&dentry->d_sb->s_anon.first);
337 } else {
338 struct dcache_hash_bucket *b;
339 b = d_hash(dentry->d_parent, dentry->d_name.hash);
340 spin_lock_bucket(b);
342 * We may not actually need to put DCACHE_UNHASHED
343 * manipulations under the hash lock, but follow
344 * the principle of least surprise.
346 dentry->d_flags |= DCACHE_UNHASHED;
347 hlist_bl_del_rcu(&dentry->d_hash);
348 spin_unlock_bucket(b);
349 dentry_rcuwalk_barrier(dentry);
353 EXPORT_SYMBOL(__d_drop);
355 void d_drop(struct dentry *dentry)
357 spin_lock(&dentry->d_lock);
358 __d_drop(dentry);
359 spin_unlock(&dentry->d_lock);
361 EXPORT_SYMBOL(d_drop);
364 * Finish off a dentry we've decided to kill.
365 * dentry->d_lock must be held, returns with it unlocked.
366 * If ref is non-zero, then decrement the refcount too.
367 * Returns dentry requiring refcount drop, or NULL if we're done.
369 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
370 __releases(dentry->d_lock)
372 struct inode *inode;
373 struct dentry *parent;
375 inode = dentry->d_inode;
376 if (inode && !spin_trylock(&inode->i_lock)) {
377 relock:
378 spin_unlock(&dentry->d_lock);
379 cpu_relax();
380 return dentry; /* try again with same dentry */
382 if (IS_ROOT(dentry))
383 parent = NULL;
384 else
385 parent = dentry->d_parent;
386 if (parent && !spin_trylock(&parent->d_lock)) {
387 if (inode)
388 spin_unlock(&inode->i_lock);
389 goto relock;
392 if (ref)
393 dentry->d_count--;
394 /* if dentry was on the d_lru list delete it from there */
395 dentry_lru_del(dentry);
396 /* if it was on the hash then remove it */
397 __d_drop(dentry);
398 return d_kill(dentry, parent);
402 * This is dput
404 * This is complicated by the fact that we do not want to put
405 * dentries that are no longer on any hash chain on the unused
406 * list: we'd much rather just get rid of them immediately.
408 * However, that implies that we have to traverse the dentry
409 * tree upwards to the parents which might _also_ now be
410 * scheduled for deletion (it may have been only waiting for
411 * its last child to go away).
413 * This tail recursion is done by hand as we don't want to depend
414 * on the compiler to always get this right (gcc generally doesn't).
415 * Real recursion would eat up our stack space.
419 * dput - release a dentry
420 * @dentry: dentry to release
422 * Release a dentry. This will drop the usage count and if appropriate
423 * call the dentry unlink method as well as removing it from the queues and
424 * releasing its resources. If the parent dentries were scheduled for release
425 * they too may now get deleted.
427 void dput(struct dentry *dentry)
429 if (!dentry)
430 return;
432 repeat:
433 if (dentry->d_count == 1)
434 might_sleep();
435 spin_lock(&dentry->d_lock);
436 BUG_ON(!dentry->d_count);
437 if (dentry->d_count > 1) {
438 dentry->d_count--;
439 spin_unlock(&dentry->d_lock);
440 return;
443 if (dentry->d_flags & DCACHE_OP_DELETE) {
444 if (dentry->d_op->d_delete(dentry))
445 goto kill_it;
448 /* Unreachable? Get rid of it */
449 if (d_unhashed(dentry))
450 goto kill_it;
452 /* Otherwise leave it cached and ensure it's on the LRU */
453 dentry->d_flags |= DCACHE_REFERENCED;
454 dentry_lru_add(dentry);
456 dentry->d_count--;
457 spin_unlock(&dentry->d_lock);
458 return;
460 kill_it:
461 dentry = dentry_kill(dentry, 1);
462 if (dentry)
463 goto repeat;
465 EXPORT_SYMBOL(dput);
468 * d_invalidate - invalidate a dentry
469 * @dentry: dentry to invalidate
471 * Try to invalidate the dentry if it turns out to be
472 * possible. If there are other dentries that can be
473 * reached through this one we can't delete it and we
474 * return -EBUSY. On success we return 0.
476 * no dcache lock.
479 int d_invalidate(struct dentry * dentry)
482 * If it's already been dropped, return OK.
484 spin_lock(&dentry->d_lock);
485 if (d_unhashed(dentry)) {
486 spin_unlock(&dentry->d_lock);
487 return 0;
490 * Check whether to do a partial shrink_dcache
491 * to get rid of unused child entries.
493 if (!list_empty(&dentry->d_subdirs)) {
494 spin_unlock(&dentry->d_lock);
495 shrink_dcache_parent(dentry);
496 spin_lock(&dentry->d_lock);
500 * Somebody else still using it?
502 * If it's a directory, we can't drop it
503 * for fear of somebody re-populating it
504 * with children (even though dropping it
505 * would make it unreachable from the root,
506 * we might still populate it if it was a
507 * working directory or similar).
509 if (dentry->d_count > 1) {
510 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
511 spin_unlock(&dentry->d_lock);
512 return -EBUSY;
516 __d_drop(dentry);
517 spin_unlock(&dentry->d_lock);
518 return 0;
520 EXPORT_SYMBOL(d_invalidate);
522 /* This must be called with d_lock held */
523 static inline void __dget_dlock(struct dentry *dentry)
525 dentry->d_count++;
528 static inline void __dget(struct dentry *dentry)
530 spin_lock(&dentry->d_lock);
531 __dget_dlock(dentry);
532 spin_unlock(&dentry->d_lock);
535 struct dentry *dget_parent(struct dentry *dentry)
537 struct dentry *ret;
539 repeat:
541 * Don't need rcu_dereference because we re-check it was correct under
542 * the lock.
544 rcu_read_lock();
545 ret = dentry->d_parent;
546 if (!ret) {
547 rcu_read_unlock();
548 goto out;
550 spin_lock(&ret->d_lock);
551 if (unlikely(ret != dentry->d_parent)) {
552 spin_unlock(&ret->d_lock);
553 rcu_read_unlock();
554 goto repeat;
556 rcu_read_unlock();
557 BUG_ON(!ret->d_count);
558 ret->d_count++;
559 spin_unlock(&ret->d_lock);
560 out:
561 return ret;
563 EXPORT_SYMBOL(dget_parent);
566 * d_find_alias - grab a hashed alias of inode
567 * @inode: inode in question
568 * @want_discon: flag, used by d_splice_alias, to request
569 * that only a DISCONNECTED alias be returned.
571 * If inode has a hashed alias, or is a directory and has any alias,
572 * acquire the reference to alias and return it. Otherwise return NULL.
573 * Notice that if inode is a directory there can be only one alias and
574 * it can be unhashed only if it has no children, or if it is the root
575 * of a filesystem.
577 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
578 * any other hashed alias over that one unless @want_discon is set,
579 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
581 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
583 struct dentry *alias, *discon_alias;
585 again:
586 discon_alias = NULL;
587 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
588 spin_lock(&alias->d_lock);
589 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
590 if (IS_ROOT(alias) &&
591 (alias->d_flags & DCACHE_DISCONNECTED)) {
592 discon_alias = alias;
593 } else if (!want_discon) {
594 __dget_dlock(alias);
595 spin_unlock(&alias->d_lock);
596 return alias;
599 spin_unlock(&alias->d_lock);
601 if (discon_alias) {
602 alias = discon_alias;
603 spin_lock(&alias->d_lock);
604 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
605 if (IS_ROOT(alias) &&
606 (alias->d_flags & DCACHE_DISCONNECTED)) {
607 __dget_dlock(alias);
608 spin_unlock(&alias->d_lock);
609 return alias;
612 spin_unlock(&alias->d_lock);
613 goto again;
615 return NULL;
618 struct dentry *d_find_alias(struct inode *inode)
620 struct dentry *de = NULL;
622 if (!list_empty(&inode->i_dentry)) {
623 spin_lock(&inode->i_lock);
624 de = __d_find_alias(inode, 0);
625 spin_unlock(&inode->i_lock);
627 return de;
629 EXPORT_SYMBOL(d_find_alias);
632 * Try to kill dentries associated with this inode.
633 * WARNING: you must own a reference to inode.
635 void d_prune_aliases(struct inode *inode)
637 struct dentry *dentry;
638 restart:
639 spin_lock(&inode->i_lock);
640 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
641 spin_lock(&dentry->d_lock);
642 if (!dentry->d_count) {
643 __dget_dlock(dentry);
644 __d_drop(dentry);
645 spin_unlock(&dentry->d_lock);
646 spin_unlock(&inode->i_lock);
647 dput(dentry);
648 goto restart;
650 spin_unlock(&dentry->d_lock);
652 spin_unlock(&inode->i_lock);
654 EXPORT_SYMBOL(d_prune_aliases);
657 * Try to throw away a dentry - free the inode, dput the parent.
658 * Requires dentry->d_lock is held, and dentry->d_count == 0.
659 * Releases dentry->d_lock.
661 * This may fail if locks cannot be acquired no problem, just try again.
663 static void try_prune_one_dentry(struct dentry *dentry)
664 __releases(dentry->d_lock)
666 struct dentry *parent;
668 parent = dentry_kill(dentry, 0);
670 * If dentry_kill returns NULL, we have nothing more to do.
671 * if it returns the same dentry, trylocks failed. In either
672 * case, just loop again.
674 * Otherwise, we need to prune ancestors too. This is necessary
675 * to prevent quadratic behavior of shrink_dcache_parent(), but
676 * is also expected to be beneficial in reducing dentry cache
677 * fragmentation.
679 if (!parent)
680 return;
681 if (parent == dentry)
682 return;
684 /* Prune ancestors. */
685 dentry = parent;
686 while (dentry) {
687 spin_lock(&dentry->d_lock);
688 if (dentry->d_count > 1) {
689 dentry->d_count--;
690 spin_unlock(&dentry->d_lock);
691 return;
693 dentry = dentry_kill(dentry, 1);
697 static void shrink_dentry_list(struct list_head *list)
699 struct dentry *dentry;
701 rcu_read_lock();
702 for (;;) {
703 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
704 if (&dentry->d_lru == list)
705 break; /* empty */
706 spin_lock(&dentry->d_lock);
707 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
708 spin_unlock(&dentry->d_lock);
709 continue;
713 * We found an inuse dentry which was not removed from
714 * the LRU because of laziness during lookup. Do not free
715 * it - just keep it off the LRU list.
717 if (dentry->d_count) {
718 dentry_lru_del(dentry);
719 spin_unlock(&dentry->d_lock);
720 continue;
723 rcu_read_unlock();
725 try_prune_one_dentry(dentry);
727 rcu_read_lock();
729 rcu_read_unlock();
733 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
734 * @sb: superblock to shrink dentry LRU.
735 * @count: number of entries to prune
736 * @flags: flags to control the dentry processing
738 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
740 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
742 /* called from prune_dcache() and shrink_dcache_parent() */
743 struct dentry *dentry;
744 LIST_HEAD(referenced);
745 LIST_HEAD(tmp);
746 int cnt = *count;
748 relock:
749 spin_lock(&dcache_lru_lock);
750 while (!list_empty(&sb->s_dentry_lru)) {
751 dentry = list_entry(sb->s_dentry_lru.prev,
752 struct dentry, d_lru);
753 BUG_ON(dentry->d_sb != sb);
755 if (!spin_trylock(&dentry->d_lock)) {
756 spin_unlock(&dcache_lru_lock);
757 cpu_relax();
758 goto relock;
762 * If we are honouring the DCACHE_REFERENCED flag and the
763 * dentry has this flag set, don't free it. Clear the flag
764 * and put it back on the LRU.
766 if (flags & DCACHE_REFERENCED &&
767 dentry->d_flags & DCACHE_REFERENCED) {
768 dentry->d_flags &= ~DCACHE_REFERENCED;
769 list_move(&dentry->d_lru, &referenced);
770 spin_unlock(&dentry->d_lock);
771 } else {
772 list_move_tail(&dentry->d_lru, &tmp);
773 spin_unlock(&dentry->d_lock);
774 if (!--cnt)
775 break;
777 cond_resched_lock(&dcache_lru_lock);
779 if (!list_empty(&referenced))
780 list_splice(&referenced, &sb->s_dentry_lru);
781 spin_unlock(&dcache_lru_lock);
783 shrink_dentry_list(&tmp);
785 *count = cnt;
789 * prune_dcache - shrink the dcache
790 * @count: number of entries to try to free
792 * Shrink the dcache. This is done when we need more memory, or simply when we
793 * need to unmount something (at which point we need to unuse all dentries).
795 * This function may fail to free any resources if all the dentries are in use.
797 static void prune_dcache(int count)
799 struct super_block *sb, *p = NULL;
800 int w_count;
801 int unused = dentry_stat.nr_unused;
802 int prune_ratio;
803 int pruned;
805 if (unused == 0 || count == 0)
806 return;
807 if (count >= unused)
808 prune_ratio = 1;
809 else
810 prune_ratio = unused / count;
811 spin_lock(&sb_lock);
812 list_for_each_entry(sb, &super_blocks, s_list) {
813 if (list_empty(&sb->s_instances))
814 continue;
815 if (sb->s_nr_dentry_unused == 0)
816 continue;
817 sb->s_count++;
818 /* Now, we reclaim unused dentrins with fairness.
819 * We reclaim them same percentage from each superblock.
820 * We calculate number of dentries to scan on this sb
821 * as follows, but the implementation is arranged to avoid
822 * overflows:
823 * number of dentries to scan on this sb =
824 * count * (number of dentries on this sb /
825 * number of dentries in the machine)
827 spin_unlock(&sb_lock);
828 if (prune_ratio != 1)
829 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
830 else
831 w_count = sb->s_nr_dentry_unused;
832 pruned = w_count;
834 * We need to be sure this filesystem isn't being unmounted,
835 * otherwise we could race with generic_shutdown_super(), and
836 * end up holding a reference to an inode while the filesystem
837 * is unmounted. So we try to get s_umount, and make sure
838 * s_root isn't NULL.
840 if (down_read_trylock(&sb->s_umount)) {
841 if ((sb->s_root != NULL) &&
842 (!list_empty(&sb->s_dentry_lru))) {
843 __shrink_dcache_sb(sb, &w_count,
844 DCACHE_REFERENCED);
845 pruned -= w_count;
847 up_read(&sb->s_umount);
849 spin_lock(&sb_lock);
850 if (p)
851 __put_super(p);
852 count -= pruned;
853 p = sb;
854 /* more work left to do? */
855 if (count <= 0)
856 break;
858 if (p)
859 __put_super(p);
860 spin_unlock(&sb_lock);
864 * shrink_dcache_sb - shrink dcache for a superblock
865 * @sb: superblock
867 * Shrink the dcache for the specified super block. This is used to free
868 * the dcache before unmounting a file system.
870 void shrink_dcache_sb(struct super_block *sb)
872 LIST_HEAD(tmp);
874 spin_lock(&dcache_lru_lock);
875 while (!list_empty(&sb->s_dentry_lru)) {
876 list_splice_init(&sb->s_dentry_lru, &tmp);
877 spin_unlock(&dcache_lru_lock);
878 shrink_dentry_list(&tmp);
879 spin_lock(&dcache_lru_lock);
881 spin_unlock(&dcache_lru_lock);
883 EXPORT_SYMBOL(shrink_dcache_sb);
886 * destroy a single subtree of dentries for unmount
887 * - see the comments on shrink_dcache_for_umount() for a description of the
888 * locking
890 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
892 struct dentry *parent;
893 unsigned detached = 0;
895 BUG_ON(!IS_ROOT(dentry));
897 /* detach this root from the system */
898 spin_lock(&dentry->d_lock);
899 dentry_lru_del(dentry);
900 __d_drop(dentry);
901 spin_unlock(&dentry->d_lock);
903 for (;;) {
904 /* descend to the first leaf in the current subtree */
905 while (!list_empty(&dentry->d_subdirs)) {
906 struct dentry *loop;
908 /* this is a branch with children - detach all of them
909 * from the system in one go */
910 spin_lock(&dentry->d_lock);
911 list_for_each_entry(loop, &dentry->d_subdirs,
912 d_u.d_child) {
913 spin_lock_nested(&loop->d_lock,
914 DENTRY_D_LOCK_NESTED);
915 dentry_lru_del(loop);
916 __d_drop(loop);
917 spin_unlock(&loop->d_lock);
919 spin_unlock(&dentry->d_lock);
921 /* move to the first child */
922 dentry = list_entry(dentry->d_subdirs.next,
923 struct dentry, d_u.d_child);
926 /* consume the dentries from this leaf up through its parents
927 * until we find one with children or run out altogether */
928 do {
929 struct inode *inode;
931 if (dentry->d_count != 0) {
932 printk(KERN_ERR
933 "BUG: Dentry %p{i=%lx,n=%s}"
934 " still in use (%d)"
935 " [unmount of %s %s]\n",
936 dentry,
937 dentry->d_inode ?
938 dentry->d_inode->i_ino : 0UL,
939 dentry->d_name.name,
940 dentry->d_count,
941 dentry->d_sb->s_type->name,
942 dentry->d_sb->s_id);
943 BUG();
946 if (IS_ROOT(dentry)) {
947 parent = NULL;
948 list_del(&dentry->d_u.d_child);
949 } else {
950 parent = dentry->d_parent;
951 spin_lock(&parent->d_lock);
952 parent->d_count--;
953 list_del(&dentry->d_u.d_child);
954 spin_unlock(&parent->d_lock);
957 detached++;
959 inode = dentry->d_inode;
960 if (inode) {
961 dentry->d_inode = NULL;
962 list_del_init(&dentry->d_alias);
963 if (dentry->d_op && dentry->d_op->d_iput)
964 dentry->d_op->d_iput(dentry, inode);
965 else
966 iput(inode);
969 d_free(dentry);
971 /* finished when we fall off the top of the tree,
972 * otherwise we ascend to the parent and move to the
973 * next sibling if there is one */
974 if (!parent)
975 return;
976 dentry = parent;
977 } while (list_empty(&dentry->d_subdirs));
979 dentry = list_entry(dentry->d_subdirs.next,
980 struct dentry, d_u.d_child);
985 * destroy the dentries attached to a superblock on unmounting
986 * - we don't need to use dentry->d_lock because:
987 * - the superblock is detached from all mountings and open files, so the
988 * dentry trees will not be rearranged by the VFS
989 * - s_umount is write-locked, so the memory pressure shrinker will ignore
990 * any dentries belonging to this superblock that it comes across
991 * - the filesystem itself is no longer permitted to rearrange the dentries
992 * in this superblock
994 void shrink_dcache_for_umount(struct super_block *sb)
996 struct dentry *dentry;
998 if (down_read_trylock(&sb->s_umount))
999 BUG();
1001 dentry = sb->s_root;
1002 sb->s_root = NULL;
1003 spin_lock(&dentry->d_lock);
1004 dentry->d_count--;
1005 spin_unlock(&dentry->d_lock);
1006 shrink_dcache_for_umount_subtree(dentry);
1008 while (!hlist_bl_empty(&sb->s_anon)) {
1009 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1010 shrink_dcache_for_umount_subtree(dentry);
1015 * Search for at least 1 mount point in the dentry's subdirs.
1016 * We descend to the next level whenever the d_subdirs
1017 * list is non-empty and continue searching.
1021 * have_submounts - check for mounts over a dentry
1022 * @parent: dentry to check.
1024 * Return true if the parent or its subdirectories contain
1025 * a mount point
1027 int have_submounts(struct dentry *parent)
1029 struct dentry *this_parent;
1030 struct list_head *next;
1031 unsigned seq;
1032 int locked = 0;
1034 seq = read_seqbegin(&rename_lock);
1035 again:
1036 this_parent = parent;
1038 if (d_mountpoint(parent))
1039 goto positive;
1040 spin_lock(&this_parent->d_lock);
1041 repeat:
1042 next = this_parent->d_subdirs.next;
1043 resume:
1044 while (next != &this_parent->d_subdirs) {
1045 struct list_head *tmp = next;
1046 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1047 next = tmp->next;
1049 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1050 /* Have we found a mount point ? */
1051 if (d_mountpoint(dentry)) {
1052 spin_unlock(&dentry->d_lock);
1053 spin_unlock(&this_parent->d_lock);
1054 goto positive;
1056 if (!list_empty(&dentry->d_subdirs)) {
1057 spin_unlock(&this_parent->d_lock);
1058 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1059 this_parent = dentry;
1060 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1061 goto repeat;
1063 spin_unlock(&dentry->d_lock);
1066 * All done at this level ... ascend and resume the search.
1068 if (this_parent != parent) {
1069 struct dentry *tmp;
1070 struct dentry *child;
1072 tmp = this_parent->d_parent;
1073 rcu_read_lock();
1074 spin_unlock(&this_parent->d_lock);
1075 child = this_parent;
1076 this_parent = tmp;
1077 spin_lock(&this_parent->d_lock);
1078 /* might go back up the wrong parent if we have had a rename
1079 * or deletion */
1080 if (this_parent != child->d_parent ||
1081 (!locked && read_seqretry(&rename_lock, seq))) {
1082 spin_unlock(&this_parent->d_lock);
1083 rcu_read_unlock();
1084 goto rename_retry;
1086 rcu_read_unlock();
1087 next = child->d_u.d_child.next;
1088 goto resume;
1090 spin_unlock(&this_parent->d_lock);
1091 if (!locked && read_seqretry(&rename_lock, seq))
1092 goto rename_retry;
1093 if (locked)
1094 write_sequnlock(&rename_lock);
1095 return 0; /* No mount points found in tree */
1096 positive:
1097 if (!locked && read_seqretry(&rename_lock, seq))
1098 goto rename_retry;
1099 if (locked)
1100 write_sequnlock(&rename_lock);
1101 return 1;
1103 rename_retry:
1104 locked = 1;
1105 write_seqlock(&rename_lock);
1106 goto again;
1108 EXPORT_SYMBOL(have_submounts);
1111 * Search the dentry child list for the specified parent,
1112 * and move any unused dentries to the end of the unused
1113 * list for prune_dcache(). We descend to the next level
1114 * whenever the d_subdirs list is non-empty and continue
1115 * searching.
1117 * It returns zero iff there are no unused children,
1118 * otherwise it returns the number of children moved to
1119 * the end of the unused list. This may not be the total
1120 * number of unused children, because select_parent can
1121 * drop the lock and return early due to latency
1122 * constraints.
1124 static int select_parent(struct dentry * parent)
1126 struct dentry *this_parent;
1127 struct list_head *next;
1128 unsigned seq;
1129 int found = 0;
1130 int locked = 0;
1132 seq = read_seqbegin(&rename_lock);
1133 again:
1134 this_parent = parent;
1135 spin_lock(&this_parent->d_lock);
1136 repeat:
1137 next = this_parent->d_subdirs.next;
1138 resume:
1139 while (next != &this_parent->d_subdirs) {
1140 struct list_head *tmp = next;
1141 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1142 next = tmp->next;
1144 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1147 * move only zero ref count dentries to the end
1148 * of the unused list for prune_dcache
1150 if (!dentry->d_count) {
1151 dentry_lru_move_tail(dentry);
1152 found++;
1153 } else {
1154 dentry_lru_del(dentry);
1158 * We can return to the caller if we have found some (this
1159 * ensures forward progress). We'll be coming back to find
1160 * the rest.
1162 if (found && need_resched()) {
1163 spin_unlock(&dentry->d_lock);
1164 goto out;
1168 * Descend a level if the d_subdirs list is non-empty.
1170 if (!list_empty(&dentry->d_subdirs)) {
1171 spin_unlock(&this_parent->d_lock);
1172 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1173 this_parent = dentry;
1174 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1175 goto repeat;
1178 spin_unlock(&dentry->d_lock);
1181 * All done at this level ... ascend and resume the search.
1183 if (this_parent != parent) {
1184 struct dentry *tmp;
1185 struct dentry *child;
1187 tmp = this_parent->d_parent;
1188 rcu_read_lock();
1189 spin_unlock(&this_parent->d_lock);
1190 child = this_parent;
1191 this_parent = tmp;
1192 spin_lock(&this_parent->d_lock);
1193 /* might go back up the wrong parent if we have had a rename
1194 * or deletion */
1195 if (this_parent != child->d_parent ||
1196 (!locked && read_seqretry(&rename_lock, seq))) {
1197 spin_unlock(&this_parent->d_lock);
1198 rcu_read_unlock();
1199 goto rename_retry;
1201 rcu_read_unlock();
1202 next = child->d_u.d_child.next;
1203 goto resume;
1205 out:
1206 spin_unlock(&this_parent->d_lock);
1207 if (!locked && read_seqretry(&rename_lock, seq))
1208 goto rename_retry;
1209 if (locked)
1210 write_sequnlock(&rename_lock);
1211 return found;
1213 rename_retry:
1214 if (found)
1215 return found;
1216 locked = 1;
1217 write_seqlock(&rename_lock);
1218 goto again;
1222 * shrink_dcache_parent - prune dcache
1223 * @parent: parent of entries to prune
1225 * Prune the dcache to remove unused children of the parent dentry.
1228 void shrink_dcache_parent(struct dentry * parent)
1230 struct super_block *sb = parent->d_sb;
1231 int found;
1233 while ((found = select_parent(parent)) != 0)
1234 __shrink_dcache_sb(sb, &found, 0);
1236 EXPORT_SYMBOL(shrink_dcache_parent);
1239 * Scan `nr' dentries and return the number which remain.
1241 * We need to avoid reentering the filesystem if the caller is performing a
1242 * GFP_NOFS allocation attempt. One example deadlock is:
1244 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1245 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1246 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1248 * In this case we return -1 to tell the caller that we baled.
1250 static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1252 if (nr) {
1253 if (!(gfp_mask & __GFP_FS))
1254 return -1;
1255 prune_dcache(nr);
1258 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
1261 static struct shrinker dcache_shrinker = {
1262 .shrink = shrink_dcache_memory,
1263 .seeks = DEFAULT_SEEKS,
1267 * d_alloc - allocate a dcache entry
1268 * @parent: parent of entry to allocate
1269 * @name: qstr of the name
1271 * Allocates a dentry. It returns %NULL if there is insufficient memory
1272 * available. On a success the dentry is returned. The name passed in is
1273 * copied and the copy passed in may be reused after this call.
1276 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1278 struct dentry *dentry;
1279 char *dname;
1281 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1282 if (!dentry)
1283 return NULL;
1285 if (name->len > DNAME_INLINE_LEN-1) {
1286 dname = kmalloc(name->len + 1, GFP_KERNEL);
1287 if (!dname) {
1288 kmem_cache_free(dentry_cache, dentry);
1289 return NULL;
1291 } else {
1292 dname = dentry->d_iname;
1294 dentry->d_name.name = dname;
1296 dentry->d_name.len = name->len;
1297 dentry->d_name.hash = name->hash;
1298 memcpy(dname, name->name, name->len);
1299 dname[name->len] = 0;
1301 dentry->d_count = 1;
1302 dentry->d_flags = DCACHE_UNHASHED;
1303 spin_lock_init(&dentry->d_lock);
1304 seqcount_init(&dentry->d_seq);
1305 dentry->d_inode = NULL;
1306 dentry->d_parent = NULL;
1307 dentry->d_sb = NULL;
1308 dentry->d_op = NULL;
1309 dentry->d_fsdata = NULL;
1310 INIT_HLIST_BL_NODE(&dentry->d_hash);
1311 INIT_LIST_HEAD(&dentry->d_lru);
1312 INIT_LIST_HEAD(&dentry->d_subdirs);
1313 INIT_LIST_HEAD(&dentry->d_alias);
1314 INIT_LIST_HEAD(&dentry->d_u.d_child);
1316 if (parent) {
1317 spin_lock(&parent->d_lock);
1319 * don't need child lock because it is not subject
1320 * to concurrency here
1322 __dget_dlock(parent);
1323 dentry->d_parent = parent;
1324 dentry->d_sb = parent->d_sb;
1325 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1326 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1327 spin_unlock(&parent->d_lock);
1330 this_cpu_inc(nr_dentry);
1332 return dentry;
1334 EXPORT_SYMBOL(d_alloc);
1336 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1338 struct dentry *dentry = d_alloc(NULL, name);
1339 if (dentry) {
1340 dentry->d_sb = sb;
1341 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1342 dentry->d_parent = dentry;
1343 dentry->d_flags |= DCACHE_DISCONNECTED;
1345 return dentry;
1347 EXPORT_SYMBOL(d_alloc_pseudo);
1349 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1351 struct qstr q;
1353 q.name = 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 |
1364 DCACHE_OP_COMPARE |
1365 DCACHE_OP_REVALIDATE |
1366 DCACHE_OP_DELETE ));
1367 dentry->d_op = op;
1368 if (!op)
1369 return;
1370 if (op->d_hash)
1371 dentry->d_flags |= DCACHE_OP_HASH;
1372 if (op->d_compare)
1373 dentry->d_flags |= DCACHE_OP_COMPARE;
1374 if (op->d_revalidate)
1375 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1376 if (op->d_delete)
1377 dentry->d_flags |= DCACHE_OP_DELETE;
1380 EXPORT_SYMBOL(d_set_d_op);
1382 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1384 spin_lock(&dentry->d_lock);
1385 if (inode) {
1386 if (unlikely(IS_AUTOMOUNT(inode)))
1387 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1388 list_add(&dentry->d_alias, &inode->i_dentry);
1390 dentry->d_inode = inode;
1391 dentry_rcuwalk_barrier(dentry);
1392 spin_unlock(&dentry->d_lock);
1393 fsnotify_d_instantiate(dentry, inode);
1397 * d_instantiate - fill in inode information for a dentry
1398 * @entry: dentry to complete
1399 * @inode: inode to attach to this dentry
1401 * Fill in inode information in the entry.
1403 * This turns negative dentries into productive full members
1404 * of society.
1406 * NOTE! This assumes that the inode count has been incremented
1407 * (or otherwise set) by the caller to indicate that it is now
1408 * in use by the dcache.
1411 void d_instantiate(struct dentry *entry, struct inode * inode)
1413 BUG_ON(!list_empty(&entry->d_alias));
1414 if (inode)
1415 spin_lock(&inode->i_lock);
1416 __d_instantiate(entry, inode);
1417 if (inode)
1418 spin_unlock(&inode->i_lock);
1419 security_d_instantiate(entry, inode);
1421 EXPORT_SYMBOL(d_instantiate);
1424 * d_instantiate_unique - instantiate a non-aliased dentry
1425 * @entry: dentry to instantiate
1426 * @inode: inode to attach to this dentry
1428 * Fill in inode information in the entry. On success, it returns NULL.
1429 * If an unhashed alias of "entry" already exists, then we return the
1430 * aliased dentry instead and drop one reference to inode.
1432 * Note that in order to avoid conflicts with rename() etc, the caller
1433 * had better be holding the parent directory semaphore.
1435 * This also assumes that the inode count has been incremented
1436 * (or otherwise set) by the caller to indicate that it is now
1437 * in use by the dcache.
1439 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1440 struct inode *inode)
1442 struct dentry *alias;
1443 int len = entry->d_name.len;
1444 const char *name = entry->d_name.name;
1445 unsigned int hash = entry->d_name.hash;
1447 if (!inode) {
1448 __d_instantiate(entry, NULL);
1449 return NULL;
1452 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1453 struct qstr *qstr = &alias->d_name;
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 (qstr->hash != hash)
1461 continue;
1462 if (alias->d_parent != entry->d_parent)
1463 continue;
1464 if (dentry_cmp(qstr->name, qstr->len, name, len))
1465 continue;
1466 __dget(alias);
1467 return alias;
1470 __d_instantiate(entry, inode);
1471 return NULL;
1474 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1476 struct dentry *result;
1478 BUG_ON(!list_empty(&entry->d_alias));
1480 if (inode)
1481 spin_lock(&inode->i_lock);
1482 result = __d_instantiate_unique(entry, inode);
1483 if (inode)
1484 spin_unlock(&inode->i_lock);
1486 if (!result) {
1487 security_d_instantiate(entry, inode);
1488 return NULL;
1491 BUG_ON(!d_unhashed(result));
1492 iput(inode);
1493 return result;
1496 EXPORT_SYMBOL(d_instantiate_unique);
1499 * d_alloc_root - allocate root dentry
1500 * @root_inode: inode to allocate the root for
1502 * Allocate a root ("/") dentry for the inode given. The inode is
1503 * instantiated and returned. %NULL is returned if there is insufficient
1504 * memory or the inode passed is %NULL.
1507 struct dentry * d_alloc_root(struct inode * root_inode)
1509 struct dentry *res = NULL;
1511 if (root_inode) {
1512 static const struct qstr name = { .name = "/", .len = 1 };
1514 res = d_alloc(NULL, &name);
1515 if (res) {
1516 res->d_sb = root_inode->i_sb;
1517 d_set_d_op(res, res->d_sb->s_d_op);
1518 res->d_parent = res;
1519 d_instantiate(res, root_inode);
1522 return res;
1524 EXPORT_SYMBOL(d_alloc_root);
1526 static struct dentry * __d_find_any_alias(struct inode *inode)
1528 struct dentry *alias;
1530 if (list_empty(&inode->i_dentry))
1531 return NULL;
1532 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1533 __dget(alias);
1534 return alias;
1537 static struct dentry * d_find_any_alias(struct inode *inode)
1539 struct dentry *de;
1541 spin_lock(&inode->i_lock);
1542 de = __d_find_any_alias(inode);
1543 spin_unlock(&inode->i_lock);
1544 return de;
1549 * d_obtain_alias - find or allocate a dentry for a given inode
1550 * @inode: inode to allocate the dentry for
1552 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1553 * similar open by handle operations. The returned dentry may be anonymous,
1554 * or may have a full name (if the inode was already in the cache).
1556 * When called on a directory inode, we must ensure that the inode only ever
1557 * has one dentry. If a dentry is found, that is returned instead of
1558 * allocating a new one.
1560 * On successful return, the reference to the inode has been transferred
1561 * to the dentry. In case of an error the reference on the inode is released.
1562 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1563 * be passed in and will be the error will be propagate to the return value,
1564 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1566 struct dentry *d_obtain_alias(struct inode *inode)
1568 static const struct qstr anonstring = { .name = "" };
1569 struct dentry *tmp;
1570 struct dentry *res;
1572 if (!inode)
1573 return ERR_PTR(-ESTALE);
1574 if (IS_ERR(inode))
1575 return ERR_CAST(inode);
1577 res = d_find_any_alias(inode);
1578 if (res)
1579 goto out_iput;
1581 tmp = d_alloc(NULL, &anonstring);
1582 if (!tmp) {
1583 res = ERR_PTR(-ENOMEM);
1584 goto out_iput;
1586 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1589 spin_lock(&inode->i_lock);
1590 res = __d_find_any_alias(inode);
1591 if (res) {
1592 spin_unlock(&inode->i_lock);
1593 dput(tmp);
1594 goto out_iput;
1597 /* attach a disconnected dentry */
1598 spin_lock(&tmp->d_lock);
1599 tmp->d_sb = inode->i_sb;
1600 d_set_d_op(tmp, tmp->d_sb->s_d_op);
1601 tmp->d_inode = inode;
1602 tmp->d_flags |= DCACHE_DISCONNECTED;
1603 list_add(&tmp->d_alias, &inode->i_dentry);
1604 bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1605 tmp->d_flags &= ~DCACHE_UNHASHED;
1606 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1607 __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1608 spin_unlock(&tmp->d_lock);
1609 spin_unlock(&inode->i_lock);
1611 return tmp;
1613 out_iput:
1614 iput(inode);
1615 return res;
1617 EXPORT_SYMBOL(d_obtain_alias);
1620 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1621 * @inode: the inode which may have a disconnected dentry
1622 * @dentry: a negative dentry which we want to point to the inode.
1624 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1625 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1626 * and return it, else simply d_add the inode to the dentry and return NULL.
1628 * This is needed in the lookup routine of any filesystem that is exportable
1629 * (via knfsd) so that we can build dcache paths to directories effectively.
1631 * If a dentry was found and moved, then it is returned. Otherwise NULL
1632 * is returned. This matches the expected return value of ->lookup.
1635 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1637 struct dentry *new = NULL;
1639 if (inode && S_ISDIR(inode->i_mode)) {
1640 spin_lock(&inode->i_lock);
1641 new = __d_find_alias(inode, 1);
1642 if (new) {
1643 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1644 spin_unlock(&inode->i_lock);
1645 security_d_instantiate(new, inode);
1646 d_move(new, dentry);
1647 iput(inode);
1648 } else {
1649 /* already taking inode->i_lock, so d_add() by hand */
1650 __d_instantiate(dentry, inode);
1651 spin_unlock(&inode->i_lock);
1652 security_d_instantiate(dentry, inode);
1653 d_rehash(dentry);
1655 } else
1656 d_add(dentry, inode);
1657 return new;
1659 EXPORT_SYMBOL(d_splice_alias);
1662 * d_add_ci - lookup or allocate new dentry with case-exact name
1663 * @inode: the inode case-insensitive lookup has found
1664 * @dentry: the negative dentry that was passed to the parent's lookup func
1665 * @name: the case-exact name to be associated with the returned dentry
1667 * This is to avoid filling the dcache with case-insensitive names to the
1668 * same inode, only the actual correct case is stored in the dcache for
1669 * case-insensitive filesystems.
1671 * For a case-insensitive lookup match and if the the case-exact dentry
1672 * already exists in in the dcache, use it and return it.
1674 * If no entry exists with the exact case name, allocate new dentry with
1675 * the exact case, and return the spliced entry.
1677 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1678 struct qstr *name)
1680 int error;
1681 struct dentry *found;
1682 struct dentry *new;
1685 * First check if a dentry matching the name already exists,
1686 * if not go ahead and create it now.
1688 found = d_hash_and_lookup(dentry->d_parent, name);
1689 if (!found) {
1690 new = d_alloc(dentry->d_parent, name);
1691 if (!new) {
1692 error = -ENOMEM;
1693 goto err_out;
1696 found = d_splice_alias(inode, new);
1697 if (found) {
1698 dput(new);
1699 return found;
1701 return new;
1705 * If a matching dentry exists, and it's not negative use it.
1707 * Decrement the reference count to balance the iget() done
1708 * earlier on.
1710 if (found->d_inode) {
1711 if (unlikely(found->d_inode != inode)) {
1712 /* This can't happen because bad inodes are unhashed. */
1713 BUG_ON(!is_bad_inode(inode));
1714 BUG_ON(!is_bad_inode(found->d_inode));
1716 iput(inode);
1717 return found;
1721 * Negative dentry: instantiate it unless the inode is a directory and
1722 * already has a dentry.
1724 spin_lock(&inode->i_lock);
1725 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1726 __d_instantiate(found, inode);
1727 spin_unlock(&inode->i_lock);
1728 security_d_instantiate(found, inode);
1729 return found;
1733 * In case a directory already has a (disconnected) entry grab a
1734 * reference to it, move it in place and use it.
1736 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1737 __dget(new);
1738 spin_unlock(&inode->i_lock);
1739 security_d_instantiate(found, inode);
1740 d_move(new, found);
1741 iput(inode);
1742 dput(found);
1743 return new;
1745 err_out:
1746 iput(inode);
1747 return ERR_PTR(error);
1749 EXPORT_SYMBOL(d_add_ci);
1752 * __d_lookup_rcu - search for a dentry (racy, store-free)
1753 * @parent: parent dentry
1754 * @name: qstr of name we wish to find
1755 * @seq: returns d_seq value at the point where the dentry was found
1756 * @inode: returns dentry->d_inode when the inode was found valid.
1757 * Returns: dentry, or NULL
1759 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1760 * resolution (store-free path walking) design described in
1761 * Documentation/filesystems/path-lookup.txt.
1763 * This is not to be used outside core vfs.
1765 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1766 * held, and rcu_read_lock held. The returned dentry must not be stored into
1767 * without taking d_lock and checking d_seq sequence count against @seq
1768 * returned here.
1770 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1771 * function.
1773 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1774 * the returned dentry, so long as its parent's seqlock is checked after the
1775 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1776 * is formed, giving integrity down the path walk.
1778 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1779 unsigned *seq, struct inode **inode)
1781 unsigned int len = name->len;
1782 unsigned int hash = name->hash;
1783 const unsigned char *str = name->name;
1784 struct dcache_hash_bucket *b = d_hash(parent, hash);
1785 struct hlist_bl_node *node;
1786 struct dentry *dentry;
1789 * Note: There is significant duplication with __d_lookup_rcu which is
1790 * required to prevent single threaded performance regressions
1791 * especially on architectures where smp_rmb (in seqcounts) are costly.
1792 * Keep the two functions in sync.
1796 * The hash list is protected using RCU.
1798 * Carefully use d_seq when comparing a candidate dentry, to avoid
1799 * races with d_move().
1801 * It is possible that concurrent renames can mess up our list
1802 * walk here and result in missing our dentry, resulting in the
1803 * false-negative result. d_lookup() protects against concurrent
1804 * renames using rename_lock seqlock.
1806 * See Documentation/vfs/dcache-locking.txt for more details.
1808 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1809 struct inode *i;
1810 const char *tname;
1811 int tlen;
1813 if (dentry->d_name.hash != hash)
1814 continue;
1816 seqretry:
1817 *seq = read_seqcount_begin(&dentry->d_seq);
1818 if (dentry->d_parent != parent)
1819 continue;
1820 if (d_unhashed(dentry))
1821 continue;
1822 tlen = dentry->d_name.len;
1823 tname = dentry->d_name.name;
1824 i = dentry->d_inode;
1825 prefetch(tname);
1826 if (i)
1827 prefetch(i);
1829 * This seqcount check is required to ensure name and
1830 * len are loaded atomically, so as not to walk off the
1831 * edge of memory when walking. If we could load this
1832 * atomically some other way, we could drop this check.
1834 if (read_seqcount_retry(&dentry->d_seq, *seq))
1835 goto seqretry;
1836 if (parent->d_flags & DCACHE_OP_COMPARE) {
1837 if (parent->d_op->d_compare(parent, *inode,
1838 dentry, i,
1839 tlen, tname, name))
1840 continue;
1841 } else {
1842 if (dentry_cmp(tname, tlen, str, len))
1843 continue;
1846 * No extra seqcount check is required after the name
1847 * compare. The caller must perform a seqcount check in
1848 * order to do anything useful with the returned dentry
1849 * anyway.
1851 *inode = i;
1852 return dentry;
1854 return NULL;
1858 * d_lookup - search for a dentry
1859 * @parent: parent dentry
1860 * @name: qstr of name we wish to find
1861 * Returns: dentry, or NULL
1863 * d_lookup searches the children of the parent dentry for the name in
1864 * question. If the dentry is found its reference count is incremented and the
1865 * dentry is returned. The caller must use dput to free the entry when it has
1866 * finished using it. %NULL is returned if the dentry does not exist.
1868 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1870 struct dentry *dentry;
1871 unsigned seq;
1873 do {
1874 seq = read_seqbegin(&rename_lock);
1875 dentry = __d_lookup(parent, name);
1876 if (dentry)
1877 break;
1878 } while (read_seqretry(&rename_lock, seq));
1879 return dentry;
1881 EXPORT_SYMBOL(d_lookup);
1884 * __d_lookup - search for a dentry (racy)
1885 * @parent: parent dentry
1886 * @name: qstr of name we wish to find
1887 * Returns: dentry, or NULL
1889 * __d_lookup is like d_lookup, however it may (rarely) return a
1890 * false-negative result due to unrelated rename activity.
1892 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1893 * however it must be used carefully, eg. with a following d_lookup in
1894 * the case of failure.
1896 * __d_lookup callers must be commented.
1898 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1900 unsigned int len = name->len;
1901 unsigned int hash = name->hash;
1902 const unsigned char *str = name->name;
1903 struct dcache_hash_bucket *b = d_hash(parent, hash);
1904 struct hlist_bl_node *node;
1905 struct dentry *found = NULL;
1906 struct dentry *dentry;
1909 * Note: There is significant duplication with __d_lookup_rcu which is
1910 * required to prevent single threaded performance regressions
1911 * especially on architectures where smp_rmb (in seqcounts) are costly.
1912 * Keep the two functions in sync.
1916 * The hash list is protected using RCU.
1918 * Take d_lock when comparing a candidate dentry, to avoid races
1919 * with d_move().
1921 * It is possible that concurrent renames can mess up our list
1922 * walk here and result in missing our dentry, resulting in the
1923 * false-negative result. d_lookup() protects against concurrent
1924 * renames using rename_lock seqlock.
1926 * See Documentation/vfs/dcache-locking.txt for more details.
1928 rcu_read_lock();
1930 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
1931 const char *tname;
1932 int tlen;
1934 if (dentry->d_name.hash != hash)
1935 continue;
1937 spin_lock(&dentry->d_lock);
1938 if (dentry->d_parent != parent)
1939 goto next;
1940 if (d_unhashed(dentry))
1941 goto next;
1944 * It is safe to compare names since d_move() cannot
1945 * change the qstr (protected by d_lock).
1947 tlen = dentry->d_name.len;
1948 tname = dentry->d_name.name;
1949 if (parent->d_flags & DCACHE_OP_COMPARE) {
1950 if (parent->d_op->d_compare(parent, parent->d_inode,
1951 dentry, dentry->d_inode,
1952 tlen, tname, name))
1953 goto next;
1954 } else {
1955 if (dentry_cmp(tname, tlen, str, len))
1956 goto next;
1959 dentry->d_count++;
1960 found = dentry;
1961 spin_unlock(&dentry->d_lock);
1962 break;
1963 next:
1964 spin_unlock(&dentry->d_lock);
1966 rcu_read_unlock();
1968 return found;
1972 * d_hash_and_lookup - hash the qstr then search for a dentry
1973 * @dir: Directory to search in
1974 * @name: qstr of name we wish to find
1976 * On hash failure or on lookup failure NULL is returned.
1978 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1980 struct dentry *dentry = NULL;
1983 * Check for a fs-specific hash function. Note that we must
1984 * calculate the standard hash first, as the d_op->d_hash()
1985 * routine may choose to leave the hash value unchanged.
1987 name->hash = full_name_hash(name->name, name->len);
1988 if (dir->d_flags & DCACHE_OP_HASH) {
1989 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1990 goto out;
1992 dentry = d_lookup(dir, name);
1993 out:
1994 return dentry;
1998 * d_validate - verify dentry provided from insecure source (deprecated)
1999 * @dentry: The dentry alleged to be valid child of @dparent
2000 * @dparent: The parent dentry (known to be valid)
2002 * An insecure source has sent us a dentry, here we verify it and dget() it.
2003 * This is used by ncpfs in its readdir implementation.
2004 * Zero is returned in the dentry is invalid.
2006 * This function is slow for big directories, and deprecated, do not use it.
2008 int d_validate(struct dentry *dentry, struct dentry *dparent)
2010 struct dentry *child;
2012 spin_lock(&dparent->d_lock);
2013 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2014 if (dentry == child) {
2015 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2016 __dget_dlock(dentry);
2017 spin_unlock(&dentry->d_lock);
2018 spin_unlock(&dparent->d_lock);
2019 return 1;
2022 spin_unlock(&dparent->d_lock);
2024 return 0;
2026 EXPORT_SYMBOL(d_validate);
2029 * When a file is deleted, we have two options:
2030 * - turn this dentry into a negative dentry
2031 * - unhash this dentry and free it.
2033 * Usually, we want to just turn this into
2034 * a negative dentry, but if anybody else is
2035 * currently using the dentry or the inode
2036 * we can't do that and we fall back on removing
2037 * it from the hash queues and waiting for
2038 * it to be deleted later when it has no users
2042 * d_delete - delete a dentry
2043 * @dentry: The dentry to delete
2045 * Turn the dentry into a negative dentry if possible, otherwise
2046 * remove it from the hash queues so it can be deleted later
2049 void d_delete(struct dentry * dentry)
2051 struct inode *inode;
2052 int isdir = 0;
2054 * Are we the only user?
2056 again:
2057 spin_lock(&dentry->d_lock);
2058 inode = dentry->d_inode;
2059 isdir = S_ISDIR(inode->i_mode);
2060 if (dentry->d_count == 1) {
2061 if (inode && !spin_trylock(&inode->i_lock)) {
2062 spin_unlock(&dentry->d_lock);
2063 cpu_relax();
2064 goto again;
2066 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2067 dentry_unlink_inode(dentry);
2068 fsnotify_nameremove(dentry, isdir);
2069 return;
2072 if (!d_unhashed(dentry))
2073 __d_drop(dentry);
2075 spin_unlock(&dentry->d_lock);
2077 fsnotify_nameremove(dentry, isdir);
2079 EXPORT_SYMBOL(d_delete);
2081 static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b)
2083 BUG_ON(!d_unhashed(entry));
2084 spin_lock_bucket(b);
2085 entry->d_flags &= ~DCACHE_UNHASHED;
2086 hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
2087 spin_unlock_bucket(b);
2090 static void _d_rehash(struct dentry * entry)
2092 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2096 * d_rehash - add an entry back to the hash
2097 * @entry: dentry to add to the hash
2099 * Adds a dentry to the hash according to its name.
2102 void d_rehash(struct dentry * entry)
2104 spin_lock(&entry->d_lock);
2105 _d_rehash(entry);
2106 spin_unlock(&entry->d_lock);
2108 EXPORT_SYMBOL(d_rehash);
2111 * dentry_update_name_case - update case insensitive dentry with a new name
2112 * @dentry: dentry to be updated
2113 * @name: new name
2115 * Update a case insensitive dentry with new case of name.
2117 * dentry must have been returned by d_lookup with name @name. Old and new
2118 * name lengths must match (ie. no d_compare which allows mismatched name
2119 * lengths).
2121 * Parent inode i_mutex must be held over d_lookup and into this call (to
2122 * keep renames and concurrent inserts, and readdir(2) away).
2124 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2126 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2127 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2129 spin_lock(&dentry->d_lock);
2130 write_seqcount_begin(&dentry->d_seq);
2131 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2132 write_seqcount_end(&dentry->d_seq);
2133 spin_unlock(&dentry->d_lock);
2135 EXPORT_SYMBOL(dentry_update_name_case);
2137 static void switch_names(struct dentry *dentry, struct dentry *target)
2139 if (dname_external(target)) {
2140 if (dname_external(dentry)) {
2142 * Both external: swap the pointers
2144 swap(target->d_name.name, dentry->d_name.name);
2145 } else {
2147 * dentry:internal, target:external. Steal target's
2148 * storage and make target internal.
2150 memcpy(target->d_iname, dentry->d_name.name,
2151 dentry->d_name.len + 1);
2152 dentry->d_name.name = target->d_name.name;
2153 target->d_name.name = target->d_iname;
2155 } else {
2156 if (dname_external(dentry)) {
2158 * dentry:external, target:internal. Give dentry's
2159 * storage to target and make dentry internal
2161 memcpy(dentry->d_iname, target->d_name.name,
2162 target->d_name.len + 1);
2163 target->d_name.name = dentry->d_name.name;
2164 dentry->d_name.name = dentry->d_iname;
2165 } else {
2167 * Both are internal. Just copy target to dentry
2169 memcpy(dentry->d_iname, target->d_name.name,
2170 target->d_name.len + 1);
2171 dentry->d_name.len = target->d_name.len;
2172 return;
2175 swap(dentry->d_name.len, target->d_name.len);
2178 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2181 * XXXX: do we really need to take target->d_lock?
2183 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2184 spin_lock(&target->d_parent->d_lock);
2185 else {
2186 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2187 spin_lock(&dentry->d_parent->d_lock);
2188 spin_lock_nested(&target->d_parent->d_lock,
2189 DENTRY_D_LOCK_NESTED);
2190 } else {
2191 spin_lock(&target->d_parent->d_lock);
2192 spin_lock_nested(&dentry->d_parent->d_lock,
2193 DENTRY_D_LOCK_NESTED);
2196 if (target < dentry) {
2197 spin_lock_nested(&target->d_lock, 2);
2198 spin_lock_nested(&dentry->d_lock, 3);
2199 } else {
2200 spin_lock_nested(&dentry->d_lock, 2);
2201 spin_lock_nested(&target->d_lock, 3);
2205 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2206 struct dentry *target)
2208 if (target->d_parent != dentry->d_parent)
2209 spin_unlock(&dentry->d_parent->d_lock);
2210 if (target->d_parent != target)
2211 spin_unlock(&target->d_parent->d_lock);
2215 * When switching names, the actual string doesn't strictly have to
2216 * be preserved in the target - because we're dropping the target
2217 * anyway. As such, we can just do a simple memcpy() to copy over
2218 * the new name before we switch.
2220 * Note that we have to be a lot more careful about getting the hash
2221 * switched - we have to switch the hash value properly even if it
2222 * then no longer matches the actual (corrupted) string of the target.
2223 * The hash value has to match the hash queue that the dentry is on..
2226 * d_move - move a dentry
2227 * @dentry: entry to move
2228 * @target: new dentry
2230 * Update the dcache to reflect the move of a file name. Negative
2231 * dcache entries should not be moved in this way.
2233 void d_move(struct dentry * dentry, struct dentry * target)
2235 if (!dentry->d_inode)
2236 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2238 BUG_ON(d_ancestor(dentry, target));
2239 BUG_ON(d_ancestor(target, dentry));
2241 write_seqlock(&rename_lock);
2243 dentry_lock_for_move(dentry, target);
2245 write_seqcount_begin(&dentry->d_seq);
2246 write_seqcount_begin(&target->d_seq);
2248 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2251 * Move the dentry to the target hash queue. Don't bother checking
2252 * for the same hash queue because of how unlikely it is.
2254 __d_drop(dentry);
2255 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2257 /* Unhash the target: dput() will then get rid of it */
2258 __d_drop(target);
2260 list_del(&dentry->d_u.d_child);
2261 list_del(&target->d_u.d_child);
2263 /* Switch the names.. */
2264 switch_names(dentry, target);
2265 swap(dentry->d_name.hash, target->d_name.hash);
2267 /* ... and switch the parents */
2268 if (IS_ROOT(dentry)) {
2269 dentry->d_parent = target->d_parent;
2270 target->d_parent = target;
2271 INIT_LIST_HEAD(&target->d_u.d_child);
2272 } else {
2273 swap(dentry->d_parent, target->d_parent);
2275 /* And add them back to the (new) parent lists */
2276 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2279 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2281 write_seqcount_end(&target->d_seq);
2282 write_seqcount_end(&dentry->d_seq);
2284 dentry_unlock_parents_for_move(dentry, target);
2285 spin_unlock(&target->d_lock);
2286 fsnotify_d_move(dentry);
2287 spin_unlock(&dentry->d_lock);
2288 write_sequnlock(&rename_lock);
2290 EXPORT_SYMBOL(d_move);
2293 * d_ancestor - search for an ancestor
2294 * @p1: ancestor dentry
2295 * @p2: child dentry
2297 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2298 * an ancestor of p2, else NULL.
2300 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2302 struct dentry *p;
2304 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2305 if (p->d_parent == p1)
2306 return p;
2308 return NULL;
2312 * This helper attempts to cope with remotely renamed directories
2314 * It assumes that the caller is already holding
2315 * dentry->d_parent->d_inode->i_mutex and the inode->i_lock
2317 * Note: If ever the locking in lock_rename() changes, then please
2318 * remember to update this too...
2320 static struct dentry *__d_unalias(struct inode *inode,
2321 struct dentry *dentry, struct dentry *alias)
2323 struct mutex *m1 = NULL, *m2 = NULL;
2324 struct dentry *ret;
2326 /* If alias and dentry share a parent, then no extra locks required */
2327 if (alias->d_parent == dentry->d_parent)
2328 goto out_unalias;
2330 /* Check for loops */
2331 ret = ERR_PTR(-ELOOP);
2332 if (d_ancestor(alias, dentry))
2333 goto out_err;
2335 /* See lock_rename() */
2336 ret = ERR_PTR(-EBUSY);
2337 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2338 goto out_err;
2339 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2340 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2341 goto out_err;
2342 m2 = &alias->d_parent->d_inode->i_mutex;
2343 out_unalias:
2344 d_move(alias, dentry);
2345 ret = alias;
2346 out_err:
2347 spin_unlock(&inode->i_lock);
2348 if (m2)
2349 mutex_unlock(m2);
2350 if (m1)
2351 mutex_unlock(m1);
2352 return ret;
2356 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2357 * named dentry in place of the dentry to be replaced.
2358 * returns with anon->d_lock held!
2360 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2362 struct dentry *dparent, *aparent;
2364 dentry_lock_for_move(anon, dentry);
2366 write_seqcount_begin(&dentry->d_seq);
2367 write_seqcount_begin(&anon->d_seq);
2369 dparent = dentry->d_parent;
2370 aparent = anon->d_parent;
2372 switch_names(dentry, anon);
2373 swap(dentry->d_name.hash, anon->d_name.hash);
2375 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2376 list_del(&dentry->d_u.d_child);
2377 if (!IS_ROOT(dentry))
2378 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2379 else
2380 INIT_LIST_HEAD(&dentry->d_u.d_child);
2382 anon->d_parent = (dparent == dentry) ? anon : dparent;
2383 list_del(&anon->d_u.d_child);
2384 if (!IS_ROOT(anon))
2385 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2386 else
2387 INIT_LIST_HEAD(&anon->d_u.d_child);
2389 write_seqcount_end(&dentry->d_seq);
2390 write_seqcount_end(&anon->d_seq);
2392 dentry_unlock_parents_for_move(anon, dentry);
2393 spin_unlock(&dentry->d_lock);
2395 /* anon->d_lock still locked, returns locked */
2396 anon->d_flags &= ~DCACHE_DISCONNECTED;
2400 * d_materialise_unique - introduce an inode into the tree
2401 * @dentry: candidate dentry
2402 * @inode: inode to bind to the dentry, to which aliases may be attached
2404 * Introduces an dentry into the tree, substituting an extant disconnected
2405 * root directory alias in its place if there is one
2407 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2409 struct dentry *actual;
2411 BUG_ON(!d_unhashed(dentry));
2413 if (!inode) {
2414 actual = dentry;
2415 __d_instantiate(dentry, NULL);
2416 d_rehash(actual);
2417 goto out_nolock;
2420 spin_lock(&inode->i_lock);
2422 if (S_ISDIR(inode->i_mode)) {
2423 struct dentry *alias;
2425 /* Does an aliased dentry already exist? */
2426 alias = __d_find_alias(inode, 0);
2427 if (alias) {
2428 actual = alias;
2429 /* Is this an anonymous mountpoint that we could splice
2430 * into our tree? */
2431 if (IS_ROOT(alias)) {
2432 __d_materialise_dentry(dentry, alias);
2433 __d_drop(alias);
2434 goto found;
2436 /* Nope, but we must(!) avoid directory aliasing */
2437 actual = __d_unalias(inode, dentry, alias);
2438 if (IS_ERR(actual))
2439 dput(alias);
2440 goto out_nolock;
2444 /* Add a unique reference */
2445 actual = __d_instantiate_unique(dentry, inode);
2446 if (!actual)
2447 actual = dentry;
2448 else
2449 BUG_ON(!d_unhashed(actual));
2451 spin_lock(&actual->d_lock);
2452 found:
2453 _d_rehash(actual);
2454 spin_unlock(&actual->d_lock);
2455 spin_unlock(&inode->i_lock);
2456 out_nolock:
2457 if (actual == dentry) {
2458 security_d_instantiate(dentry, inode);
2459 return NULL;
2462 iput(inode);
2463 return actual;
2465 EXPORT_SYMBOL_GPL(d_materialise_unique);
2467 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2469 *buflen -= namelen;
2470 if (*buflen < 0)
2471 return -ENAMETOOLONG;
2472 *buffer -= namelen;
2473 memcpy(*buffer, str, namelen);
2474 return 0;
2477 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2479 return prepend(buffer, buflen, name->name, name->len);
2483 * prepend_path - Prepend path string to a buffer
2484 * @path: the dentry/vfsmount to report
2485 * @root: root vfsmnt/dentry (may be modified by this function)
2486 * @buffer: pointer to the end of the buffer
2487 * @buflen: pointer to buffer length
2489 * Caller holds the rename_lock.
2491 * If path is not reachable from the supplied root, then the value of
2492 * root is changed (without modifying refcounts).
2494 static int prepend_path(const struct path *path, struct path *root,
2495 char **buffer, int *buflen)
2497 struct dentry *dentry = path->dentry;
2498 struct vfsmount *vfsmnt = path->mnt;
2499 bool slash = false;
2500 int error = 0;
2502 br_read_lock(vfsmount_lock);
2503 while (dentry != root->dentry || vfsmnt != root->mnt) {
2504 struct dentry * parent;
2506 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2507 /* Global root? */
2508 if (vfsmnt->mnt_parent == vfsmnt) {
2509 goto global_root;
2511 dentry = vfsmnt->mnt_mountpoint;
2512 vfsmnt = vfsmnt->mnt_parent;
2513 continue;
2515 parent = dentry->d_parent;
2516 prefetch(parent);
2517 spin_lock(&dentry->d_lock);
2518 error = prepend_name(buffer, buflen, &dentry->d_name);
2519 spin_unlock(&dentry->d_lock);
2520 if (!error)
2521 error = prepend(buffer, buflen, "/", 1);
2522 if (error)
2523 break;
2525 slash = true;
2526 dentry = parent;
2529 out:
2530 if (!error && !slash)
2531 error = prepend(buffer, buflen, "/", 1);
2533 br_read_unlock(vfsmount_lock);
2534 return error;
2536 global_root:
2538 * Filesystems needing to implement special "root names"
2539 * should do so with ->d_dname()
2541 if (IS_ROOT(dentry) &&
2542 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2543 WARN(1, "Root dentry has weird name <%.*s>\n",
2544 (int) dentry->d_name.len, dentry->d_name.name);
2546 root->mnt = vfsmnt;
2547 root->dentry = dentry;
2548 goto out;
2552 * __d_path - return the path of a dentry
2553 * @path: the dentry/vfsmount to report
2554 * @root: root vfsmnt/dentry (may be modified by this function)
2555 * @buf: buffer to return value in
2556 * @buflen: buffer length
2558 * Convert a dentry into an ASCII path name.
2560 * Returns a pointer into the buffer or an error code if the
2561 * path was too long.
2563 * "buflen" should be positive.
2565 * If path is not reachable from the supplied root, then the value of
2566 * root is changed (without modifying refcounts).
2568 char *__d_path(const struct path *path, struct path *root,
2569 char *buf, int buflen)
2571 char *res = buf + buflen;
2572 int error;
2574 prepend(&res, &buflen, "\0", 1);
2575 write_seqlock(&rename_lock);
2576 error = prepend_path(path, root, &res, &buflen);
2577 write_sequnlock(&rename_lock);
2579 if (error)
2580 return ERR_PTR(error);
2581 return res;
2585 * same as __d_path but appends "(deleted)" for unlinked files.
2587 static int path_with_deleted(const struct path *path, struct path *root,
2588 char **buf, int *buflen)
2590 prepend(buf, buflen, "\0", 1);
2591 if (d_unlinked(path->dentry)) {
2592 int error = prepend(buf, buflen, " (deleted)", 10);
2593 if (error)
2594 return error;
2597 return prepend_path(path, root, buf, buflen);
2600 static int prepend_unreachable(char **buffer, int *buflen)
2602 return prepend(buffer, buflen, "(unreachable)", 13);
2606 * d_path - return the path of a dentry
2607 * @path: path to report
2608 * @buf: buffer to return value in
2609 * @buflen: buffer length
2611 * Convert a dentry into an ASCII path name. If the entry has been deleted
2612 * the string " (deleted)" is appended. Note that this is ambiguous.
2614 * Returns a pointer into the buffer or an error code if the path was
2615 * too long. Note: Callers should use the returned pointer, not the passed
2616 * in buffer, to use the name! The implementation often starts at an offset
2617 * into the buffer, and may leave 0 bytes at the start.
2619 * "buflen" should be positive.
2621 char *d_path(const struct path *path, char *buf, int buflen)
2623 char *res = buf + buflen;
2624 struct path root;
2625 struct path tmp;
2626 int error;
2629 * We have various synthetic filesystems that never get mounted. On
2630 * these filesystems dentries are never used for lookup purposes, and
2631 * thus don't need to be hashed. They also don't need a name until a
2632 * user wants to identify the object in /proc/pid/fd/. The little hack
2633 * below allows us to generate a name for these objects on demand:
2635 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2636 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2638 get_fs_root(current->fs, &root);
2639 write_seqlock(&rename_lock);
2640 tmp = root;
2641 error = path_with_deleted(path, &tmp, &res, &buflen);
2642 if (error)
2643 res = ERR_PTR(error);
2644 write_sequnlock(&rename_lock);
2645 path_put(&root);
2646 return res;
2648 EXPORT_SYMBOL(d_path);
2651 * d_path_with_unreachable - return the path of a dentry
2652 * @path: path to report
2653 * @buf: buffer to return value in
2654 * @buflen: buffer length
2656 * The difference from d_path() is that this prepends "(unreachable)"
2657 * to paths which are unreachable from the current process' root.
2659 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2661 char *res = buf + buflen;
2662 struct path root;
2663 struct path tmp;
2664 int error;
2666 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2667 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2669 get_fs_root(current->fs, &root);
2670 write_seqlock(&rename_lock);
2671 tmp = root;
2672 error = path_with_deleted(path, &tmp, &res, &buflen);
2673 if (!error && !path_equal(&tmp, &root))
2674 error = prepend_unreachable(&res, &buflen);
2675 write_sequnlock(&rename_lock);
2676 path_put(&root);
2677 if (error)
2678 res = ERR_PTR(error);
2680 return res;
2684 * Helper function for dentry_operations.d_dname() members
2686 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2687 const char *fmt, ...)
2689 va_list args;
2690 char temp[64];
2691 int sz;
2693 va_start(args, fmt);
2694 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2695 va_end(args);
2697 if (sz > sizeof(temp) || sz > buflen)
2698 return ERR_PTR(-ENAMETOOLONG);
2700 buffer += buflen - sz;
2701 return memcpy(buffer, temp, sz);
2705 * Write full pathname from the root of the filesystem into the buffer.
2707 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2709 char *end = buf + buflen;
2710 char *retval;
2712 prepend(&end, &buflen, "\0", 1);
2713 if (buflen < 1)
2714 goto Elong;
2715 /* Get '/' right */
2716 retval = end-1;
2717 *retval = '/';
2719 while (!IS_ROOT(dentry)) {
2720 struct dentry *parent = dentry->d_parent;
2721 int error;
2723 prefetch(parent);
2724 spin_lock(&dentry->d_lock);
2725 error = prepend_name(&end, &buflen, &dentry->d_name);
2726 spin_unlock(&dentry->d_lock);
2727 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2728 goto Elong;
2730 retval = end;
2731 dentry = parent;
2733 return retval;
2734 Elong:
2735 return ERR_PTR(-ENAMETOOLONG);
2738 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2740 char *retval;
2742 write_seqlock(&rename_lock);
2743 retval = __dentry_path(dentry, buf, buflen);
2744 write_sequnlock(&rename_lock);
2746 return retval;
2748 EXPORT_SYMBOL(dentry_path_raw);
2750 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2752 char *p = NULL;
2753 char *retval;
2755 write_seqlock(&rename_lock);
2756 if (d_unlinked(dentry)) {
2757 p = buf + buflen;
2758 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2759 goto Elong;
2760 buflen++;
2762 retval = __dentry_path(dentry, buf, buflen);
2763 write_sequnlock(&rename_lock);
2764 if (!IS_ERR(retval) && p)
2765 *p = '/'; /* restore '/' overriden with '\0' */
2766 return retval;
2767 Elong:
2768 return ERR_PTR(-ENAMETOOLONG);
2772 * NOTE! The user-level library version returns a
2773 * character pointer. The kernel system call just
2774 * returns the length of the buffer filled (which
2775 * includes the ending '\0' character), or a negative
2776 * error value. So libc would do something like
2778 * char *getcwd(char * buf, size_t size)
2780 * int retval;
2782 * retval = sys_getcwd(buf, size);
2783 * if (retval >= 0)
2784 * return buf;
2785 * errno = -retval;
2786 * return NULL;
2789 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2791 int error;
2792 struct path pwd, root;
2793 char *page = (char *) __get_free_page(GFP_USER);
2795 if (!page)
2796 return -ENOMEM;
2798 get_fs_root_and_pwd(current->fs, &root, &pwd);
2800 error = -ENOENT;
2801 write_seqlock(&rename_lock);
2802 if (!d_unlinked(pwd.dentry)) {
2803 unsigned long len;
2804 struct path tmp = root;
2805 char *cwd = page + PAGE_SIZE;
2806 int buflen = PAGE_SIZE;
2808 prepend(&cwd, &buflen, "\0", 1);
2809 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2810 write_sequnlock(&rename_lock);
2812 if (error)
2813 goto out;
2815 /* Unreachable from current root */
2816 if (!path_equal(&tmp, &root)) {
2817 error = prepend_unreachable(&cwd, &buflen);
2818 if (error)
2819 goto out;
2822 error = -ERANGE;
2823 len = PAGE_SIZE + page - cwd;
2824 if (len <= size) {
2825 error = len;
2826 if (copy_to_user(buf, cwd, len))
2827 error = -EFAULT;
2829 } else {
2830 write_sequnlock(&rename_lock);
2833 out:
2834 path_put(&pwd);
2835 path_put(&root);
2836 free_page((unsigned long) page);
2837 return error;
2841 * Test whether new_dentry is a subdirectory of old_dentry.
2843 * Trivially implemented using the dcache structure
2847 * is_subdir - is new dentry a subdirectory of old_dentry
2848 * @new_dentry: new dentry
2849 * @old_dentry: old dentry
2851 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2852 * Returns 0 otherwise.
2853 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2856 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2858 int result;
2859 unsigned seq;
2861 if (new_dentry == old_dentry)
2862 return 1;
2864 do {
2865 /* for restarting inner loop in case of seq retry */
2866 seq = read_seqbegin(&rename_lock);
2868 * Need rcu_readlock to protect against the d_parent trashing
2869 * due to d_move
2871 rcu_read_lock();
2872 if (d_ancestor(old_dentry, new_dentry))
2873 result = 1;
2874 else
2875 result = 0;
2876 rcu_read_unlock();
2877 } while (read_seqretry(&rename_lock, seq));
2879 return result;
2882 int path_is_under(struct path *path1, struct path *path2)
2884 struct vfsmount *mnt = path1->mnt;
2885 struct dentry *dentry = path1->dentry;
2886 int res;
2888 br_read_lock(vfsmount_lock);
2889 if (mnt != path2->mnt) {
2890 for (;;) {
2891 if (mnt->mnt_parent == mnt) {
2892 br_read_unlock(vfsmount_lock);
2893 return 0;
2895 if (mnt->mnt_parent == path2->mnt)
2896 break;
2897 mnt = mnt->mnt_parent;
2899 dentry = mnt->mnt_mountpoint;
2901 res = is_subdir(dentry, path2->dentry);
2902 br_read_unlock(vfsmount_lock);
2903 return res;
2905 EXPORT_SYMBOL(path_is_under);
2907 void d_genocide(struct dentry *root)
2909 struct dentry *this_parent;
2910 struct list_head *next;
2911 unsigned seq;
2912 int locked = 0;
2914 seq = read_seqbegin(&rename_lock);
2915 again:
2916 this_parent = root;
2917 spin_lock(&this_parent->d_lock);
2918 repeat:
2919 next = this_parent->d_subdirs.next;
2920 resume:
2921 while (next != &this_parent->d_subdirs) {
2922 struct list_head *tmp = next;
2923 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2924 next = tmp->next;
2926 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2927 if (d_unhashed(dentry) || !dentry->d_inode) {
2928 spin_unlock(&dentry->d_lock);
2929 continue;
2931 if (!list_empty(&dentry->d_subdirs)) {
2932 spin_unlock(&this_parent->d_lock);
2933 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2934 this_parent = dentry;
2935 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2936 goto repeat;
2938 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2939 dentry->d_flags |= DCACHE_GENOCIDE;
2940 dentry->d_count--;
2942 spin_unlock(&dentry->d_lock);
2944 if (this_parent != root) {
2945 struct dentry *tmp;
2946 struct dentry *child;
2948 tmp = this_parent->d_parent;
2949 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2950 this_parent->d_flags |= DCACHE_GENOCIDE;
2951 this_parent->d_count--;
2953 rcu_read_lock();
2954 spin_unlock(&this_parent->d_lock);
2955 child = this_parent;
2956 this_parent = tmp;
2957 spin_lock(&this_parent->d_lock);
2958 /* might go back up the wrong parent if we have had a rename
2959 * or deletion */
2960 if (this_parent != child->d_parent ||
2961 (!locked && read_seqretry(&rename_lock, seq))) {
2962 spin_unlock(&this_parent->d_lock);
2963 rcu_read_unlock();
2964 goto rename_retry;
2966 rcu_read_unlock();
2967 next = child->d_u.d_child.next;
2968 goto resume;
2970 spin_unlock(&this_parent->d_lock);
2971 if (!locked && read_seqretry(&rename_lock, seq))
2972 goto rename_retry;
2973 if (locked)
2974 write_sequnlock(&rename_lock);
2975 return;
2977 rename_retry:
2978 locked = 1;
2979 write_seqlock(&rename_lock);
2980 goto again;
2984 * find_inode_number - check for dentry with name
2985 * @dir: directory to check
2986 * @name: Name to find.
2988 * Check whether a dentry already exists for the given name,
2989 * and return the inode number if it has an inode. Otherwise
2990 * 0 is returned.
2992 * This routine is used to post-process directory listings for
2993 * filesystems using synthetic inode numbers, and is necessary
2994 * to keep getcwd() working.
2997 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2999 struct dentry * dentry;
3000 ino_t ino = 0;
3002 dentry = d_hash_and_lookup(dir, name);
3003 if (dentry) {
3004 if (dentry->d_inode)
3005 ino = dentry->d_inode->i_ino;
3006 dput(dentry);
3008 return ino;
3010 EXPORT_SYMBOL(find_inode_number);
3012 static __initdata unsigned long dhash_entries;
3013 static int __init set_dhash_entries(char *str)
3015 if (!str)
3016 return 0;
3017 dhash_entries = simple_strtoul(str, &str, 0);
3018 return 1;
3020 __setup("dhash_entries=", set_dhash_entries);
3022 static void __init dcache_init_early(void)
3024 int loop;
3026 /* If hashes are distributed across NUMA nodes, defer
3027 * hash allocation until vmalloc space is available.
3029 if (hashdist)
3030 return;
3032 dentry_hashtable =
3033 alloc_large_system_hash("Dentry cache",
3034 sizeof(struct dcache_hash_bucket),
3035 dhash_entries,
3037 HASH_EARLY,
3038 &d_hash_shift,
3039 &d_hash_mask,
3042 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3043 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3046 static void __init dcache_init(void)
3048 int loop;
3051 * A constructor could be added for stable state like the lists,
3052 * but it is probably not worth it because of the cache nature
3053 * of the dcache.
3055 dentry_cache = KMEM_CACHE(dentry,
3056 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3058 register_shrinker(&dcache_shrinker);
3060 /* Hash may have been set up in dcache_init_early */
3061 if (!hashdist)
3062 return;
3064 dentry_hashtable =
3065 alloc_large_system_hash("Dentry cache",
3066 sizeof(struct dcache_hash_bucket),
3067 dhash_entries,
3070 &d_hash_shift,
3071 &d_hash_mask,
3074 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3075 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
3078 /* SLAB cache for __getname() consumers */
3079 struct kmem_cache *names_cachep __read_mostly;
3080 EXPORT_SYMBOL(names_cachep);
3082 EXPORT_SYMBOL(d_genocide);
3084 void __init vfs_caches_init_early(void)
3086 dcache_init_early();
3087 inode_init_early();
3090 void __init vfs_caches_init(unsigned long mempages)
3092 unsigned long reserve;
3094 /* Base hash sizes on available memory, with a reserve equal to
3095 150% of current kernel size */
3097 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3098 mempages -= reserve;
3100 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3101 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3103 dcache_init();
3104 inode_init();
3105 files_init(mempages);
3106 mnt_init();
3107 bdev_cache_init();
3108 chrdev_init();