ACPI: thinkpad-acpi: clean up hotkey_notify()
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / dcache.c
blobca1fd3b7ea9b6f0b1f5dfef46f186dfa4862a59f
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/fdtable.h>
21 #include <linux/fs.h>
22 #include <linux/fsnotify.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/hash.h>
26 #include <linux/cache.h>
27 #include <linux/module.h>
28 #include <linux/mount.h>
29 #include <linux/file.h>
30 #include <asm/uaccess.h>
31 #include <linux/security.h>
32 #include <linux/seqlock.h>
33 #include <linux/swap.h>
34 #include <linux/bootmem.h>
35 #include "internal.h"
38 int sysctl_vfs_cache_pressure __read_mostly = 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
42 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
44 EXPORT_SYMBOL(dcache_lock);
46 static struct kmem_cache *dentry_cache __read_mostly;
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
61 static unsigned int d_hash_mask __read_mostly;
62 static unsigned int d_hash_shift __read_mostly;
63 static struct hlist_head *dentry_hashtable __read_mostly;
65 /* Statistics gathering. */
66 struct dentry_stat_t dentry_stat = {
67 .age_limit = 45,
70 static void __d_free(struct dentry *dentry)
72 WARN_ON(!list_empty(&dentry->d_alias));
73 if (dname_external(dentry))
74 kfree(dentry->d_name.name);
75 kmem_cache_free(dentry_cache, dentry);
78 static void d_callback(struct rcu_head *head)
80 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
81 __d_free(dentry);
85 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
86 * inside dcache_lock.
88 static void d_free(struct dentry *dentry)
90 if (dentry->d_op && dentry->d_op->d_release)
91 dentry->d_op->d_release(dentry);
92 /* if dentry was never inserted into hash, immediate free is OK */
93 if (hlist_unhashed(&dentry->d_hash))
94 __d_free(dentry);
95 else
96 call_rcu(&dentry->d_u.d_rcu, d_callback);
100 * Release the dentry's inode, using the filesystem
101 * d_iput() operation if defined.
103 static void dentry_iput(struct dentry * dentry)
104 __releases(dentry->d_lock)
105 __releases(dcache_lock)
107 struct inode *inode = dentry->d_inode;
108 if (inode) {
109 dentry->d_inode = NULL;
110 list_del_init(&dentry->d_alias);
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
113 if (!inode->i_nlink)
114 fsnotify_inoderemove(inode);
115 if (dentry->d_op && dentry->d_op->d_iput)
116 dentry->d_op->d_iput(dentry, inode);
117 else
118 iput(inode);
119 } else {
120 spin_unlock(&dentry->d_lock);
121 spin_unlock(&dcache_lock);
126 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
128 static void dentry_lru_add(struct dentry *dentry)
130 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
131 dentry->d_sb->s_nr_dentry_unused++;
132 dentry_stat.nr_unused++;
135 static void dentry_lru_add_tail(struct dentry *dentry)
137 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
138 dentry->d_sb->s_nr_dentry_unused++;
139 dentry_stat.nr_unused++;
142 static void dentry_lru_del(struct dentry *dentry)
144 if (!list_empty(&dentry->d_lru)) {
145 list_del(&dentry->d_lru);
146 dentry->d_sb->s_nr_dentry_unused--;
147 dentry_stat.nr_unused--;
151 static void dentry_lru_del_init(struct dentry *dentry)
153 if (likely(!list_empty(&dentry->d_lru))) {
154 list_del_init(&dentry->d_lru);
155 dentry->d_sb->s_nr_dentry_unused--;
156 dentry_stat.nr_unused--;
161 * d_kill - kill dentry and return parent
162 * @dentry: dentry to kill
164 * The dentry must already be unhashed and removed from the LRU.
166 * If this is the root of the dentry tree, return NULL.
168 static struct dentry *d_kill(struct dentry *dentry)
169 __releases(dentry->d_lock)
170 __releases(dcache_lock)
172 struct dentry *parent;
174 list_del(&dentry->d_u.d_child);
175 dentry_stat.nr_dentry--; /* For d_free, below */
176 /*drops the locks, at that point nobody can reach this dentry */
177 dentry_iput(dentry);
178 if (IS_ROOT(dentry))
179 parent = NULL;
180 else
181 parent = dentry->d_parent;
182 d_free(dentry);
183 return parent;
187 * This is dput
189 * This is complicated by the fact that we do not want to put
190 * dentries that are no longer on any hash chain on the unused
191 * list: we'd much rather just get rid of them immediately.
193 * However, that implies that we have to traverse the dentry
194 * tree upwards to the parents which might _also_ now be
195 * scheduled for deletion (it may have been only waiting for
196 * its last child to go away).
198 * This tail recursion is done by hand as we don't want to depend
199 * on the compiler to always get this right (gcc generally doesn't).
200 * Real recursion would eat up our stack space.
204 * dput - release a dentry
205 * @dentry: dentry to release
207 * Release a dentry. This will drop the usage count and if appropriate
208 * call the dentry unlink method as well as removing it from the queues and
209 * releasing its resources. If the parent dentries were scheduled for release
210 * they too may now get deleted.
212 * no dcache lock, please.
215 void dput(struct dentry *dentry)
217 if (!dentry)
218 return;
220 repeat:
221 if (atomic_read(&dentry->d_count) == 1)
222 might_sleep();
223 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
224 return;
226 spin_lock(&dentry->d_lock);
227 if (atomic_read(&dentry->d_count)) {
228 spin_unlock(&dentry->d_lock);
229 spin_unlock(&dcache_lock);
230 return;
234 * AV: ->d_delete() is _NOT_ allowed to block now.
236 if (dentry->d_op && dentry->d_op->d_delete) {
237 if (dentry->d_op->d_delete(dentry))
238 goto unhash_it;
240 /* Unreachable? Get rid of it */
241 if (d_unhashed(dentry))
242 goto kill_it;
243 if (list_empty(&dentry->d_lru)) {
244 dentry->d_flags |= DCACHE_REFERENCED;
245 dentry_lru_add(dentry);
247 spin_unlock(&dentry->d_lock);
248 spin_unlock(&dcache_lock);
249 return;
251 unhash_it:
252 __d_drop(dentry);
253 kill_it:
254 /* if dentry was on the d_lru list delete it from there */
255 dentry_lru_del(dentry);
256 dentry = d_kill(dentry);
257 if (dentry)
258 goto repeat;
262 * d_invalidate - invalidate a dentry
263 * @dentry: dentry to invalidate
265 * Try to invalidate the dentry if it turns out to be
266 * possible. If there are other dentries that can be
267 * reached through this one we can't delete it and we
268 * return -EBUSY. On success we return 0.
270 * no dcache lock.
273 int d_invalidate(struct dentry * dentry)
276 * If it's already been dropped, return OK.
278 spin_lock(&dcache_lock);
279 if (d_unhashed(dentry)) {
280 spin_unlock(&dcache_lock);
281 return 0;
284 * Check whether to do a partial shrink_dcache
285 * to get rid of unused child entries.
287 if (!list_empty(&dentry->d_subdirs)) {
288 spin_unlock(&dcache_lock);
289 shrink_dcache_parent(dentry);
290 spin_lock(&dcache_lock);
294 * Somebody else still using it?
296 * If it's a directory, we can't drop it
297 * for fear of somebody re-populating it
298 * with children (even though dropping it
299 * would make it unreachable from the root,
300 * we might still populate it if it was a
301 * working directory or similar).
303 spin_lock(&dentry->d_lock);
304 if (atomic_read(&dentry->d_count) > 1) {
305 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
306 spin_unlock(&dentry->d_lock);
307 spin_unlock(&dcache_lock);
308 return -EBUSY;
312 __d_drop(dentry);
313 spin_unlock(&dentry->d_lock);
314 spin_unlock(&dcache_lock);
315 return 0;
318 /* This should be called _only_ with dcache_lock held */
320 static inline struct dentry * __dget_locked(struct dentry *dentry)
322 atomic_inc(&dentry->d_count);
323 dentry_lru_del_init(dentry);
324 return dentry;
327 struct dentry * dget_locked(struct dentry *dentry)
329 return __dget_locked(dentry);
333 * d_find_alias - grab a hashed alias of inode
334 * @inode: inode in question
335 * @want_discon: flag, used by d_splice_alias, to request
336 * that only a DISCONNECTED alias be returned.
338 * If inode has a hashed alias, or is a directory and has any alias,
339 * acquire the reference to alias and return it. Otherwise return NULL.
340 * Notice that if inode is a directory there can be only one alias and
341 * it can be unhashed only if it has no children, or if it is the root
342 * of a filesystem.
344 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
345 * any other hashed alias over that one unless @want_discon is set,
346 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
349 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
351 struct list_head *head, *next, *tmp;
352 struct dentry *alias, *discon_alias=NULL;
354 head = &inode->i_dentry;
355 next = inode->i_dentry.next;
356 while (next != head) {
357 tmp = next;
358 next = tmp->next;
359 prefetch(next);
360 alias = list_entry(tmp, struct dentry, d_alias);
361 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
362 if (IS_ROOT(alias) &&
363 (alias->d_flags & DCACHE_DISCONNECTED))
364 discon_alias = alias;
365 else if (!want_discon) {
366 __dget_locked(alias);
367 return alias;
371 if (discon_alias)
372 __dget_locked(discon_alias);
373 return discon_alias;
376 struct dentry * d_find_alias(struct inode *inode)
378 struct dentry *de = NULL;
380 if (!list_empty(&inode->i_dentry)) {
381 spin_lock(&dcache_lock);
382 de = __d_find_alias(inode, 0);
383 spin_unlock(&dcache_lock);
385 return de;
389 * Try to kill dentries associated with this inode.
390 * WARNING: you must own a reference to inode.
392 void d_prune_aliases(struct inode *inode)
394 struct dentry *dentry;
395 restart:
396 spin_lock(&dcache_lock);
397 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
398 spin_lock(&dentry->d_lock);
399 if (!atomic_read(&dentry->d_count)) {
400 __dget_locked(dentry);
401 __d_drop(dentry);
402 spin_unlock(&dentry->d_lock);
403 spin_unlock(&dcache_lock);
404 dput(dentry);
405 goto restart;
407 spin_unlock(&dentry->d_lock);
409 spin_unlock(&dcache_lock);
413 * Throw away a dentry - free the inode, dput the parent. This requires that
414 * the LRU list has already been removed.
416 * Try to prune ancestors as well. This is necessary to prevent
417 * quadratic behavior of shrink_dcache_parent(), but is also expected
418 * to be beneficial in reducing dentry cache fragmentation.
420 static void prune_one_dentry(struct dentry * dentry)
421 __releases(dentry->d_lock)
422 __releases(dcache_lock)
423 __acquires(dcache_lock)
425 __d_drop(dentry);
426 dentry = d_kill(dentry);
429 * Prune ancestors. Locking is simpler than in dput(),
430 * because dcache_lock needs to be taken anyway.
432 spin_lock(&dcache_lock);
433 while (dentry) {
434 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
435 return;
437 if (dentry->d_op && dentry->d_op->d_delete)
438 dentry->d_op->d_delete(dentry);
439 dentry_lru_del_init(dentry);
440 __d_drop(dentry);
441 dentry = d_kill(dentry);
442 spin_lock(&dcache_lock);
447 * Shrink the dentry LRU on a given superblock.
448 * @sb : superblock to shrink dentry LRU.
449 * @count: If count is NULL, we prune all dentries on superblock.
450 * @flags: If flags is non-zero, we need to do special processing based on
451 * which flags are set. This means we don't need to maintain multiple
452 * similar copies of this loop.
454 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
456 LIST_HEAD(referenced);
457 LIST_HEAD(tmp);
458 struct dentry *dentry;
459 int cnt = 0;
461 BUG_ON(!sb);
462 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
463 spin_lock(&dcache_lock);
464 if (count != NULL)
465 /* called from prune_dcache() and shrink_dcache_parent() */
466 cnt = *count;
467 restart:
468 if (count == NULL)
469 list_splice_init(&sb->s_dentry_lru, &tmp);
470 else {
471 while (!list_empty(&sb->s_dentry_lru)) {
472 dentry = list_entry(sb->s_dentry_lru.prev,
473 struct dentry, d_lru);
474 BUG_ON(dentry->d_sb != sb);
476 spin_lock(&dentry->d_lock);
478 * If we are honouring the DCACHE_REFERENCED flag and
479 * the dentry has this flag set, don't free it. Clear
480 * the flag and put it back on the LRU.
482 if ((flags & DCACHE_REFERENCED)
483 && (dentry->d_flags & DCACHE_REFERENCED)) {
484 dentry->d_flags &= ~DCACHE_REFERENCED;
485 list_move_tail(&dentry->d_lru, &referenced);
486 spin_unlock(&dentry->d_lock);
487 } else {
488 list_move_tail(&dentry->d_lru, &tmp);
489 spin_unlock(&dentry->d_lock);
490 cnt--;
491 if (!cnt)
492 break;
494 cond_resched_lock(&dcache_lock);
497 while (!list_empty(&tmp)) {
498 dentry = list_entry(tmp.prev, struct dentry, d_lru);
499 dentry_lru_del_init(dentry);
500 spin_lock(&dentry->d_lock);
502 * We found an inuse dentry which was not removed from
503 * the LRU because of laziness during lookup. Do not free
504 * it - just keep it off the LRU list.
506 if (atomic_read(&dentry->d_count)) {
507 spin_unlock(&dentry->d_lock);
508 continue;
510 prune_one_dentry(dentry);
511 /* dentry->d_lock was dropped in prune_one_dentry() */
512 cond_resched_lock(&dcache_lock);
514 if (count == NULL && !list_empty(&sb->s_dentry_lru))
515 goto restart;
516 if (count != NULL)
517 *count = cnt;
518 if (!list_empty(&referenced))
519 list_splice(&referenced, &sb->s_dentry_lru);
520 spin_unlock(&dcache_lock);
524 * prune_dcache - shrink the dcache
525 * @count: number of entries to try to free
527 * Shrink the dcache. This is done when we need more memory, or simply when we
528 * need to unmount something (at which point we need to unuse all dentries).
530 * This function may fail to free any resources if all the dentries are in use.
532 static void prune_dcache(int count)
534 struct super_block *sb;
535 int w_count;
536 int unused = dentry_stat.nr_unused;
537 int prune_ratio;
538 int pruned;
540 if (unused == 0 || count == 0)
541 return;
542 spin_lock(&dcache_lock);
543 restart:
544 if (count >= unused)
545 prune_ratio = 1;
546 else
547 prune_ratio = unused / count;
548 spin_lock(&sb_lock);
549 list_for_each_entry(sb, &super_blocks, s_list) {
550 if (sb->s_nr_dentry_unused == 0)
551 continue;
552 sb->s_count++;
553 /* Now, we reclaim unused dentrins with fairness.
554 * We reclaim them same percentage from each superblock.
555 * We calculate number of dentries to scan on this sb
556 * as follows, but the implementation is arranged to avoid
557 * overflows:
558 * number of dentries to scan on this sb =
559 * count * (number of dentries on this sb /
560 * number of dentries in the machine)
562 spin_unlock(&sb_lock);
563 if (prune_ratio != 1)
564 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
565 else
566 w_count = sb->s_nr_dentry_unused;
567 pruned = w_count;
569 * We need to be sure this filesystem isn't being unmounted,
570 * otherwise we could race with generic_shutdown_super(), and
571 * end up holding a reference to an inode while the filesystem
572 * is unmounted. So we try to get s_umount, and make sure
573 * s_root isn't NULL.
575 if (down_read_trylock(&sb->s_umount)) {
576 if ((sb->s_root != NULL) &&
577 (!list_empty(&sb->s_dentry_lru))) {
578 spin_unlock(&dcache_lock);
579 __shrink_dcache_sb(sb, &w_count,
580 DCACHE_REFERENCED);
581 pruned -= w_count;
582 spin_lock(&dcache_lock);
584 up_read(&sb->s_umount);
586 spin_lock(&sb_lock);
587 count -= pruned;
589 * restart only when sb is no longer on the list and
590 * we have more work to do.
592 if (__put_super_and_need_restart(sb) && count > 0) {
593 spin_unlock(&sb_lock);
594 goto restart;
597 spin_unlock(&sb_lock);
598 spin_unlock(&dcache_lock);
602 * shrink_dcache_sb - shrink dcache for a superblock
603 * @sb: superblock
605 * Shrink the dcache for the specified super block. This
606 * is used to free the dcache before unmounting a file
607 * system
609 void shrink_dcache_sb(struct super_block * sb)
611 __shrink_dcache_sb(sb, NULL, 0);
615 * destroy a single subtree of dentries for unmount
616 * - see the comments on shrink_dcache_for_umount() for a description of the
617 * locking
619 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
621 struct dentry *parent;
622 unsigned detached = 0;
624 BUG_ON(!IS_ROOT(dentry));
626 /* detach this root from the system */
627 spin_lock(&dcache_lock);
628 dentry_lru_del_init(dentry);
629 __d_drop(dentry);
630 spin_unlock(&dcache_lock);
632 for (;;) {
633 /* descend to the first leaf in the current subtree */
634 while (!list_empty(&dentry->d_subdirs)) {
635 struct dentry *loop;
637 /* this is a branch with children - detach all of them
638 * from the system in one go */
639 spin_lock(&dcache_lock);
640 list_for_each_entry(loop, &dentry->d_subdirs,
641 d_u.d_child) {
642 dentry_lru_del_init(loop);
643 __d_drop(loop);
644 cond_resched_lock(&dcache_lock);
646 spin_unlock(&dcache_lock);
648 /* move to the first child */
649 dentry = list_entry(dentry->d_subdirs.next,
650 struct dentry, d_u.d_child);
653 /* consume the dentries from this leaf up through its parents
654 * until we find one with children or run out altogether */
655 do {
656 struct inode *inode;
658 if (atomic_read(&dentry->d_count) != 0) {
659 printk(KERN_ERR
660 "BUG: Dentry %p{i=%lx,n=%s}"
661 " still in use (%d)"
662 " [unmount of %s %s]\n",
663 dentry,
664 dentry->d_inode ?
665 dentry->d_inode->i_ino : 0UL,
666 dentry->d_name.name,
667 atomic_read(&dentry->d_count),
668 dentry->d_sb->s_type->name,
669 dentry->d_sb->s_id);
670 BUG();
673 if (IS_ROOT(dentry))
674 parent = NULL;
675 else {
676 parent = dentry->d_parent;
677 atomic_dec(&parent->d_count);
680 list_del(&dentry->d_u.d_child);
681 detached++;
683 inode = dentry->d_inode;
684 if (inode) {
685 dentry->d_inode = NULL;
686 list_del_init(&dentry->d_alias);
687 if (dentry->d_op && dentry->d_op->d_iput)
688 dentry->d_op->d_iput(dentry, inode);
689 else
690 iput(inode);
693 d_free(dentry);
695 /* finished when we fall off the top of the tree,
696 * otherwise we ascend to the parent and move to the
697 * next sibling if there is one */
698 if (!parent)
699 goto out;
701 dentry = parent;
703 } while (list_empty(&dentry->d_subdirs));
705 dentry = list_entry(dentry->d_subdirs.next,
706 struct dentry, d_u.d_child);
708 out:
709 /* several dentries were freed, need to correct nr_dentry */
710 spin_lock(&dcache_lock);
711 dentry_stat.nr_dentry -= detached;
712 spin_unlock(&dcache_lock);
716 * destroy the dentries attached to a superblock on unmounting
717 * - we don't need to use dentry->d_lock, and only need dcache_lock when
718 * removing the dentry from the system lists and hashes because:
719 * - the superblock is detached from all mountings and open files, so the
720 * dentry trees will not be rearranged by the VFS
721 * - s_umount is write-locked, so the memory pressure shrinker will ignore
722 * any dentries belonging to this superblock that it comes across
723 * - the filesystem itself is no longer permitted to rearrange the dentries
724 * in this superblock
726 void shrink_dcache_for_umount(struct super_block *sb)
728 struct dentry *dentry;
730 if (down_read_trylock(&sb->s_umount))
731 BUG();
733 dentry = sb->s_root;
734 sb->s_root = NULL;
735 atomic_dec(&dentry->d_count);
736 shrink_dcache_for_umount_subtree(dentry);
738 while (!hlist_empty(&sb->s_anon)) {
739 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
740 shrink_dcache_for_umount_subtree(dentry);
745 * Search for at least 1 mount point in the dentry's subdirs.
746 * We descend to the next level whenever the d_subdirs
747 * list is non-empty and continue searching.
751 * have_submounts - check for mounts over a dentry
752 * @parent: dentry to check.
754 * Return true if the parent or its subdirectories contain
755 * a mount point
758 int have_submounts(struct dentry *parent)
760 struct dentry *this_parent = parent;
761 struct list_head *next;
763 spin_lock(&dcache_lock);
764 if (d_mountpoint(parent))
765 goto positive;
766 repeat:
767 next = this_parent->d_subdirs.next;
768 resume:
769 while (next != &this_parent->d_subdirs) {
770 struct list_head *tmp = next;
771 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
772 next = tmp->next;
773 /* Have we found a mount point ? */
774 if (d_mountpoint(dentry))
775 goto positive;
776 if (!list_empty(&dentry->d_subdirs)) {
777 this_parent = dentry;
778 goto repeat;
782 * All done at this level ... ascend and resume the search.
784 if (this_parent != parent) {
785 next = this_parent->d_u.d_child.next;
786 this_parent = this_parent->d_parent;
787 goto resume;
789 spin_unlock(&dcache_lock);
790 return 0; /* No mount points found in tree */
791 positive:
792 spin_unlock(&dcache_lock);
793 return 1;
797 * Search the dentry child list for the specified parent,
798 * and move any unused dentries to the end of the unused
799 * list for prune_dcache(). We descend to the next level
800 * whenever the d_subdirs list is non-empty and continue
801 * searching.
803 * It returns zero iff there are no unused children,
804 * otherwise it returns the number of children moved to
805 * the end of the unused list. This may not be the total
806 * number of unused children, because select_parent can
807 * drop the lock and return early due to latency
808 * constraints.
810 static int select_parent(struct dentry * parent)
812 struct dentry *this_parent = parent;
813 struct list_head *next;
814 int found = 0;
816 spin_lock(&dcache_lock);
817 repeat:
818 next = this_parent->d_subdirs.next;
819 resume:
820 while (next != &this_parent->d_subdirs) {
821 struct list_head *tmp = next;
822 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
823 next = tmp->next;
825 dentry_lru_del_init(dentry);
827 * move only zero ref count dentries to the end
828 * of the unused list for prune_dcache
830 if (!atomic_read(&dentry->d_count)) {
831 dentry_lru_add_tail(dentry);
832 found++;
836 * We can return to the caller if we have found some (this
837 * ensures forward progress). We'll be coming back to find
838 * the rest.
840 if (found && need_resched())
841 goto out;
844 * Descend a level if the d_subdirs list is non-empty.
846 if (!list_empty(&dentry->d_subdirs)) {
847 this_parent = dentry;
848 goto repeat;
852 * All done at this level ... ascend and resume the search.
854 if (this_parent != parent) {
855 next = this_parent->d_u.d_child.next;
856 this_parent = this_parent->d_parent;
857 goto resume;
859 out:
860 spin_unlock(&dcache_lock);
861 return found;
865 * shrink_dcache_parent - prune dcache
866 * @parent: parent of entries to prune
868 * Prune the dcache to remove unused children of the parent dentry.
871 void shrink_dcache_parent(struct dentry * parent)
873 struct super_block *sb = parent->d_sb;
874 int found;
876 while ((found = select_parent(parent)) != 0)
877 __shrink_dcache_sb(sb, &found, 0);
881 * Scan `nr' dentries and return the number which remain.
883 * We need to avoid reentering the filesystem if the caller is performing a
884 * GFP_NOFS allocation attempt. One example deadlock is:
886 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
887 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
888 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
890 * In this case we return -1 to tell the caller that we baled.
892 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
894 if (nr) {
895 if (!(gfp_mask & __GFP_FS))
896 return -1;
897 prune_dcache(nr);
899 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
902 static struct shrinker dcache_shrinker = {
903 .shrink = shrink_dcache_memory,
904 .seeks = DEFAULT_SEEKS,
908 * d_alloc - allocate a dcache entry
909 * @parent: parent of entry to allocate
910 * @name: qstr of the name
912 * Allocates a dentry. It returns %NULL if there is insufficient memory
913 * available. On a success the dentry is returned. The name passed in is
914 * copied and the copy passed in may be reused after this call.
917 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
919 struct dentry *dentry;
920 char *dname;
922 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
923 if (!dentry)
924 return NULL;
926 if (name->len > DNAME_INLINE_LEN-1) {
927 dname = kmalloc(name->len + 1, GFP_KERNEL);
928 if (!dname) {
929 kmem_cache_free(dentry_cache, dentry);
930 return NULL;
932 } else {
933 dname = dentry->d_iname;
935 dentry->d_name.name = dname;
937 dentry->d_name.len = name->len;
938 dentry->d_name.hash = name->hash;
939 memcpy(dname, name->name, name->len);
940 dname[name->len] = 0;
942 atomic_set(&dentry->d_count, 1);
943 dentry->d_flags = DCACHE_UNHASHED;
944 spin_lock_init(&dentry->d_lock);
945 dentry->d_inode = NULL;
946 dentry->d_parent = NULL;
947 dentry->d_sb = NULL;
948 dentry->d_op = NULL;
949 dentry->d_fsdata = NULL;
950 dentry->d_mounted = 0;
951 #ifdef CONFIG_PROFILING
952 dentry->d_cookie = NULL;
953 #endif
954 INIT_HLIST_NODE(&dentry->d_hash);
955 INIT_LIST_HEAD(&dentry->d_lru);
956 INIT_LIST_HEAD(&dentry->d_subdirs);
957 INIT_LIST_HEAD(&dentry->d_alias);
959 if (parent) {
960 dentry->d_parent = dget(parent);
961 dentry->d_sb = parent->d_sb;
962 } else {
963 INIT_LIST_HEAD(&dentry->d_u.d_child);
966 spin_lock(&dcache_lock);
967 if (parent)
968 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
969 dentry_stat.nr_dentry++;
970 spin_unlock(&dcache_lock);
972 return dentry;
975 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
977 struct qstr q;
979 q.name = name;
980 q.len = strlen(name);
981 q.hash = full_name_hash(q.name, q.len);
982 return d_alloc(parent, &q);
985 /* the caller must hold dcache_lock */
986 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
988 if (inode)
989 list_add(&dentry->d_alias, &inode->i_dentry);
990 dentry->d_inode = inode;
991 fsnotify_d_instantiate(dentry, inode);
995 * d_instantiate - fill in inode information for a dentry
996 * @entry: dentry to complete
997 * @inode: inode to attach to this dentry
999 * Fill in inode information in the entry.
1001 * This turns negative dentries into productive full members
1002 * of society.
1004 * NOTE! This assumes that the inode count has been incremented
1005 * (or otherwise set) by the caller to indicate that it is now
1006 * in use by the dcache.
1009 void d_instantiate(struct dentry *entry, struct inode * inode)
1011 BUG_ON(!list_empty(&entry->d_alias));
1012 spin_lock(&dcache_lock);
1013 __d_instantiate(entry, inode);
1014 spin_unlock(&dcache_lock);
1015 security_d_instantiate(entry, inode);
1019 * d_instantiate_unique - instantiate a non-aliased dentry
1020 * @entry: dentry to instantiate
1021 * @inode: inode to attach to this dentry
1023 * Fill in inode information in the entry. On success, it returns NULL.
1024 * If an unhashed alias of "entry" already exists, then we return the
1025 * aliased dentry instead and drop one reference to inode.
1027 * Note that in order to avoid conflicts with rename() etc, the caller
1028 * had better be holding the parent directory semaphore.
1030 * This also assumes that the inode count has been incremented
1031 * (or otherwise set) by the caller to indicate that it is now
1032 * in use by the dcache.
1034 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1035 struct inode *inode)
1037 struct dentry *alias;
1038 int len = entry->d_name.len;
1039 const char *name = entry->d_name.name;
1040 unsigned int hash = entry->d_name.hash;
1042 if (!inode) {
1043 __d_instantiate(entry, NULL);
1044 return NULL;
1047 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1048 struct qstr *qstr = &alias->d_name;
1050 if (qstr->hash != hash)
1051 continue;
1052 if (alias->d_parent != entry->d_parent)
1053 continue;
1054 if (qstr->len != len)
1055 continue;
1056 if (memcmp(qstr->name, name, len))
1057 continue;
1058 dget_locked(alias);
1059 return alias;
1062 __d_instantiate(entry, inode);
1063 return NULL;
1066 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1068 struct dentry *result;
1070 BUG_ON(!list_empty(&entry->d_alias));
1072 spin_lock(&dcache_lock);
1073 result = __d_instantiate_unique(entry, inode);
1074 spin_unlock(&dcache_lock);
1076 if (!result) {
1077 security_d_instantiate(entry, inode);
1078 return NULL;
1081 BUG_ON(!d_unhashed(result));
1082 iput(inode);
1083 return result;
1086 EXPORT_SYMBOL(d_instantiate_unique);
1089 * d_alloc_root - allocate root dentry
1090 * @root_inode: inode to allocate the root for
1092 * Allocate a root ("/") dentry for the inode given. The inode is
1093 * instantiated and returned. %NULL is returned if there is insufficient
1094 * memory or the inode passed is %NULL.
1097 struct dentry * d_alloc_root(struct inode * root_inode)
1099 struct dentry *res = NULL;
1101 if (root_inode) {
1102 static const struct qstr name = { .name = "/", .len = 1 };
1104 res = d_alloc(NULL, &name);
1105 if (res) {
1106 res->d_sb = root_inode->i_sb;
1107 res->d_parent = res;
1108 d_instantiate(res, root_inode);
1111 return res;
1114 static inline struct hlist_head *d_hash(struct dentry *parent,
1115 unsigned long hash)
1117 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1118 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1119 return dentry_hashtable + (hash & D_HASHMASK);
1123 * d_obtain_alias - find or allocate a dentry for a given inode
1124 * @inode: inode to allocate the dentry for
1126 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1127 * similar open by handle operations. The returned dentry may be anonymous,
1128 * or may have a full name (if the inode was already in the cache).
1130 * When called on a directory inode, we must ensure that the inode only ever
1131 * has one dentry. If a dentry is found, that is returned instead of
1132 * allocating a new one.
1134 * On successful return, the reference to the inode has been transferred
1135 * to the dentry. In case of an error the reference on the inode is released.
1136 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1137 * be passed in and will be the error will be propagate to the return value,
1138 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1140 struct dentry *d_obtain_alias(struct inode *inode)
1142 static const struct qstr anonstring = { .name = "" };
1143 struct dentry *tmp;
1144 struct dentry *res;
1146 if (!inode)
1147 return ERR_PTR(-ESTALE);
1148 if (IS_ERR(inode))
1149 return ERR_CAST(inode);
1151 res = d_find_alias(inode);
1152 if (res)
1153 goto out_iput;
1155 tmp = d_alloc(NULL, &anonstring);
1156 if (!tmp) {
1157 res = ERR_PTR(-ENOMEM);
1158 goto out_iput;
1160 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1162 spin_lock(&dcache_lock);
1163 res = __d_find_alias(inode, 0);
1164 if (res) {
1165 spin_unlock(&dcache_lock);
1166 dput(tmp);
1167 goto out_iput;
1170 /* attach a disconnected dentry */
1171 spin_lock(&tmp->d_lock);
1172 tmp->d_sb = inode->i_sb;
1173 tmp->d_inode = inode;
1174 tmp->d_flags |= DCACHE_DISCONNECTED;
1175 tmp->d_flags &= ~DCACHE_UNHASHED;
1176 list_add(&tmp->d_alias, &inode->i_dentry);
1177 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1178 spin_unlock(&tmp->d_lock);
1180 spin_unlock(&dcache_lock);
1181 return tmp;
1183 out_iput:
1184 iput(inode);
1185 return res;
1187 EXPORT_SYMBOL_GPL(d_obtain_alias);
1190 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1191 * @inode: the inode which may have a disconnected dentry
1192 * @dentry: a negative dentry which we want to point to the inode.
1194 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1195 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1196 * and return it, else simply d_add the inode to the dentry and return NULL.
1198 * This is needed in the lookup routine of any filesystem that is exportable
1199 * (via knfsd) so that we can build dcache paths to directories effectively.
1201 * If a dentry was found and moved, then it is returned. Otherwise NULL
1202 * is returned. This matches the expected return value of ->lookup.
1205 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1207 struct dentry *new = NULL;
1209 if (inode && S_ISDIR(inode->i_mode)) {
1210 spin_lock(&dcache_lock);
1211 new = __d_find_alias(inode, 1);
1212 if (new) {
1213 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1214 spin_unlock(&dcache_lock);
1215 security_d_instantiate(new, inode);
1216 d_rehash(dentry);
1217 d_move(new, dentry);
1218 iput(inode);
1219 } else {
1220 /* already taking dcache_lock, so d_add() by hand */
1221 __d_instantiate(dentry, inode);
1222 spin_unlock(&dcache_lock);
1223 security_d_instantiate(dentry, inode);
1224 d_rehash(dentry);
1226 } else
1227 d_add(dentry, inode);
1228 return new;
1232 * d_add_ci - lookup or allocate new dentry with case-exact name
1233 * @inode: the inode case-insensitive lookup has found
1234 * @dentry: the negative dentry that was passed to the parent's lookup func
1235 * @name: the case-exact name to be associated with the returned dentry
1237 * This is to avoid filling the dcache with case-insensitive names to the
1238 * same inode, only the actual correct case is stored in the dcache for
1239 * case-insensitive filesystems.
1241 * For a case-insensitive lookup match and if the the case-exact dentry
1242 * already exists in in the dcache, use it and return it.
1244 * If no entry exists with the exact case name, allocate new dentry with
1245 * the exact case, and return the spliced entry.
1247 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1248 struct qstr *name)
1250 int error;
1251 struct dentry *found;
1252 struct dentry *new;
1254 /* Does a dentry matching the name exist already? */
1255 found = d_hash_and_lookup(dentry->d_parent, name);
1256 /* If not, create it now and return */
1257 if (!found) {
1258 new = d_alloc(dentry->d_parent, name);
1259 if (!new) {
1260 error = -ENOMEM;
1261 goto err_out;
1263 found = d_splice_alias(inode, new);
1264 if (found) {
1265 dput(new);
1266 return found;
1268 return new;
1270 /* Matching dentry exists, check if it is negative. */
1271 if (found->d_inode) {
1272 if (unlikely(found->d_inode != inode)) {
1273 /* This can't happen because bad inodes are unhashed. */
1274 BUG_ON(!is_bad_inode(inode));
1275 BUG_ON(!is_bad_inode(found->d_inode));
1278 * Already have the inode and the dentry attached, decrement
1279 * the reference count to balance the iget() done
1280 * earlier on. We found the dentry using d_lookup() so it
1281 * cannot be disconnected and thus we do not need to worry
1282 * about any NFS/disconnectedness issues here.
1284 iput(inode);
1285 return found;
1288 * Negative dentry: instantiate it unless the inode is a directory and
1289 * has a 'disconnected' dentry (i.e. IS_ROOT and DCACHE_DISCONNECTED),
1290 * in which case d_move() that in place of the found dentry.
1292 if (!S_ISDIR(inode->i_mode)) {
1293 /* Not a directory; everything is easy. */
1294 d_instantiate(found, inode);
1295 return found;
1297 spin_lock(&dcache_lock);
1298 if (list_empty(&inode->i_dentry)) {
1300 * Directory without a 'disconnected' dentry; we need to do
1301 * d_instantiate() by hand because it takes dcache_lock which
1302 * we already hold.
1304 __d_instantiate(found, inode);
1305 spin_unlock(&dcache_lock);
1306 security_d_instantiate(found, inode);
1307 return found;
1310 * Directory with a 'disconnected' dentry; get a reference to the
1311 * 'disconnected' dentry.
1313 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1314 dget_locked(new);
1315 spin_unlock(&dcache_lock);
1316 /* Do security vodoo. */
1317 security_d_instantiate(found, inode);
1318 /* Move new in place of found. */
1319 d_move(new, found);
1320 /* Balance the iget() we did above. */
1321 iput(inode);
1322 /* Throw away found. */
1323 dput(found);
1324 /* Use new as the actual dentry. */
1325 return new;
1327 err_out:
1328 iput(inode);
1329 return ERR_PTR(error);
1333 * d_lookup - search for a dentry
1334 * @parent: parent dentry
1335 * @name: qstr of name we wish to find
1337 * Searches the children of the parent dentry for the name in question. If
1338 * the dentry is found its reference count is incremented and the dentry
1339 * is returned. The caller must use d_put to free the entry when it has
1340 * finished using it. %NULL is returned on failure.
1342 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1343 * Memory barriers are used while updating and doing lockless traversal.
1344 * To avoid races with d_move while rename is happening, d_lock is used.
1346 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1347 * and name pointer in one structure pointed by d_qstr.
1349 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1350 * lookup is going on.
1352 * The dentry unused LRU is not updated even if lookup finds the required dentry
1353 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1354 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1355 * acquisition.
1357 * d_lookup() is protected against the concurrent renames in some unrelated
1358 * directory using the seqlockt_t rename_lock.
1361 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1363 struct dentry * dentry = NULL;
1364 unsigned long seq;
1366 do {
1367 seq = read_seqbegin(&rename_lock);
1368 dentry = __d_lookup(parent, name);
1369 if (dentry)
1370 break;
1371 } while (read_seqretry(&rename_lock, seq));
1372 return dentry;
1375 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1377 unsigned int len = name->len;
1378 unsigned int hash = name->hash;
1379 const unsigned char *str = name->name;
1380 struct hlist_head *head = d_hash(parent,hash);
1381 struct dentry *found = NULL;
1382 struct hlist_node *node;
1383 struct dentry *dentry;
1385 rcu_read_lock();
1387 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1388 struct qstr *qstr;
1390 if (dentry->d_name.hash != hash)
1391 continue;
1392 if (dentry->d_parent != parent)
1393 continue;
1395 spin_lock(&dentry->d_lock);
1398 * Recheck the dentry after taking the lock - d_move may have
1399 * changed things. Don't bother checking the hash because we're
1400 * about to compare the whole name anyway.
1402 if (dentry->d_parent != parent)
1403 goto next;
1405 /* non-existing due to RCU? */
1406 if (d_unhashed(dentry))
1407 goto next;
1410 * It is safe to compare names since d_move() cannot
1411 * change the qstr (protected by d_lock).
1413 qstr = &dentry->d_name;
1414 if (parent->d_op && parent->d_op->d_compare) {
1415 if (parent->d_op->d_compare(parent, qstr, name))
1416 goto next;
1417 } else {
1418 if (qstr->len != len)
1419 goto next;
1420 if (memcmp(qstr->name, str, len))
1421 goto next;
1424 atomic_inc(&dentry->d_count);
1425 found = dentry;
1426 spin_unlock(&dentry->d_lock);
1427 break;
1428 next:
1429 spin_unlock(&dentry->d_lock);
1431 rcu_read_unlock();
1433 return found;
1437 * d_hash_and_lookup - hash the qstr then search for a dentry
1438 * @dir: Directory to search in
1439 * @name: qstr of name we wish to find
1441 * On hash failure or on lookup failure NULL is returned.
1443 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1445 struct dentry *dentry = NULL;
1448 * Check for a fs-specific hash function. Note that we must
1449 * calculate the standard hash first, as the d_op->d_hash()
1450 * routine may choose to leave the hash value unchanged.
1452 name->hash = full_name_hash(name->name, name->len);
1453 if (dir->d_op && dir->d_op->d_hash) {
1454 if (dir->d_op->d_hash(dir, name) < 0)
1455 goto out;
1457 dentry = d_lookup(dir, name);
1458 out:
1459 return dentry;
1463 * d_validate - verify dentry provided from insecure source
1464 * @dentry: The dentry alleged to be valid child of @dparent
1465 * @dparent: The parent dentry (known to be valid)
1467 * An insecure source has sent us a dentry, here we verify it and dget() it.
1468 * This is used by ncpfs in its readdir implementation.
1469 * Zero is returned in the dentry is invalid.
1472 int d_validate(struct dentry *dentry, struct dentry *dparent)
1474 struct hlist_head *base;
1475 struct hlist_node *lhp;
1477 /* Check whether the ptr might be valid at all.. */
1478 if (!kmem_ptr_validate(dentry_cache, dentry))
1479 goto out;
1481 if (dentry->d_parent != dparent)
1482 goto out;
1484 spin_lock(&dcache_lock);
1485 base = d_hash(dparent, dentry->d_name.hash);
1486 hlist_for_each(lhp,base) {
1487 /* hlist_for_each_entry_rcu() not required for d_hash list
1488 * as it is parsed under dcache_lock
1490 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1491 __dget_locked(dentry);
1492 spin_unlock(&dcache_lock);
1493 return 1;
1496 spin_unlock(&dcache_lock);
1497 out:
1498 return 0;
1502 * When a file is deleted, we have two options:
1503 * - turn this dentry into a negative dentry
1504 * - unhash this dentry and free it.
1506 * Usually, we want to just turn this into
1507 * a negative dentry, but if anybody else is
1508 * currently using the dentry or the inode
1509 * we can't do that and we fall back on removing
1510 * it from the hash queues and waiting for
1511 * it to be deleted later when it has no users
1515 * d_delete - delete a dentry
1516 * @dentry: The dentry to delete
1518 * Turn the dentry into a negative dentry if possible, otherwise
1519 * remove it from the hash queues so it can be deleted later
1522 void d_delete(struct dentry * dentry)
1524 int isdir = 0;
1526 * Are we the only user?
1528 spin_lock(&dcache_lock);
1529 spin_lock(&dentry->d_lock);
1530 isdir = S_ISDIR(dentry->d_inode->i_mode);
1531 if (atomic_read(&dentry->d_count) == 1) {
1532 dentry_iput(dentry);
1533 fsnotify_nameremove(dentry, isdir);
1534 return;
1537 if (!d_unhashed(dentry))
1538 __d_drop(dentry);
1540 spin_unlock(&dentry->d_lock);
1541 spin_unlock(&dcache_lock);
1543 fsnotify_nameremove(dentry, isdir);
1546 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1549 entry->d_flags &= ~DCACHE_UNHASHED;
1550 hlist_add_head_rcu(&entry->d_hash, list);
1553 static void _d_rehash(struct dentry * entry)
1555 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1559 * d_rehash - add an entry back to the hash
1560 * @entry: dentry to add to the hash
1562 * Adds a dentry to the hash according to its name.
1565 void d_rehash(struct dentry * entry)
1567 spin_lock(&dcache_lock);
1568 spin_lock(&entry->d_lock);
1569 _d_rehash(entry);
1570 spin_unlock(&entry->d_lock);
1571 spin_unlock(&dcache_lock);
1574 #define do_switch(x,y) do { \
1575 __typeof__ (x) __tmp = x; \
1576 x = y; y = __tmp; } while (0)
1579 * When switching names, the actual string doesn't strictly have to
1580 * be preserved in the target - because we're dropping the target
1581 * anyway. As such, we can just do a simple memcpy() to copy over
1582 * the new name before we switch.
1584 * Note that we have to be a lot more careful about getting the hash
1585 * switched - we have to switch the hash value properly even if it
1586 * then no longer matches the actual (corrupted) string of the target.
1587 * The hash value has to match the hash queue that the dentry is on..
1589 static void switch_names(struct dentry *dentry, struct dentry *target)
1591 if (dname_external(target)) {
1592 if (dname_external(dentry)) {
1594 * Both external: swap the pointers
1596 do_switch(target->d_name.name, dentry->d_name.name);
1597 } else {
1599 * dentry:internal, target:external. Steal target's
1600 * storage and make target internal.
1602 memcpy(target->d_iname, dentry->d_name.name,
1603 dentry->d_name.len + 1);
1604 dentry->d_name.name = target->d_name.name;
1605 target->d_name.name = target->d_iname;
1607 } else {
1608 if (dname_external(dentry)) {
1610 * dentry:external, target:internal. Give dentry's
1611 * storage to target and make dentry internal
1613 memcpy(dentry->d_iname, target->d_name.name,
1614 target->d_name.len + 1);
1615 target->d_name.name = dentry->d_name.name;
1616 dentry->d_name.name = dentry->d_iname;
1617 } else {
1619 * Both are internal. Just copy target to dentry
1621 memcpy(dentry->d_iname, target->d_name.name,
1622 target->d_name.len + 1);
1623 dentry->d_name.len = target->d_name.len;
1624 return;
1627 do_switch(dentry->d_name.len, target->d_name.len);
1631 * We cannibalize "target" when moving dentry on top of it,
1632 * because it's going to be thrown away anyway. We could be more
1633 * polite about it, though.
1635 * This forceful removal will result in ugly /proc output if
1636 * somebody holds a file open that got deleted due to a rename.
1637 * We could be nicer about the deleted file, and let it show
1638 * up under the name it had before it was deleted rather than
1639 * under the original name of the file that was moved on top of it.
1643 * d_move_locked - move a dentry
1644 * @dentry: entry to move
1645 * @target: new dentry
1647 * Update the dcache to reflect the move of a file name. Negative
1648 * dcache entries should not be moved in this way.
1650 static void d_move_locked(struct dentry * dentry, struct dentry * target)
1652 struct hlist_head *list;
1654 if (!dentry->d_inode)
1655 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1657 write_seqlock(&rename_lock);
1659 * XXXX: do we really need to take target->d_lock?
1661 if (target < dentry) {
1662 spin_lock(&target->d_lock);
1663 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1664 } else {
1665 spin_lock(&dentry->d_lock);
1666 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1669 /* Move the dentry to the target hash queue, if on different bucket */
1670 if (d_unhashed(dentry))
1671 goto already_unhashed;
1673 hlist_del_rcu(&dentry->d_hash);
1675 already_unhashed:
1676 list = d_hash(target->d_parent, target->d_name.hash);
1677 __d_rehash(dentry, list);
1679 /* Unhash the target: dput() will then get rid of it */
1680 __d_drop(target);
1682 list_del(&dentry->d_u.d_child);
1683 list_del(&target->d_u.d_child);
1685 /* Switch the names.. */
1686 switch_names(dentry, target);
1687 do_switch(dentry->d_name.hash, target->d_name.hash);
1689 /* ... and switch the parents */
1690 if (IS_ROOT(dentry)) {
1691 dentry->d_parent = target->d_parent;
1692 target->d_parent = target;
1693 INIT_LIST_HEAD(&target->d_u.d_child);
1694 } else {
1695 do_switch(dentry->d_parent, target->d_parent);
1697 /* And add them back to the (new) parent lists */
1698 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1701 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1702 spin_unlock(&target->d_lock);
1703 fsnotify_d_move(dentry);
1704 spin_unlock(&dentry->d_lock);
1705 write_sequnlock(&rename_lock);
1709 * d_move - move a dentry
1710 * @dentry: entry to move
1711 * @target: new dentry
1713 * Update the dcache to reflect the move of a file name. Negative
1714 * dcache entries should not be moved in this way.
1717 void d_move(struct dentry * dentry, struct dentry * target)
1719 spin_lock(&dcache_lock);
1720 d_move_locked(dentry, target);
1721 spin_unlock(&dcache_lock);
1725 * d_ancestor - search for an ancestor
1726 * @p1: ancestor dentry
1727 * @p2: child dentry
1729 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1730 * an ancestor of p2, else NULL.
1732 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
1734 struct dentry *p;
1736 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
1737 if (p->d_parent == p1)
1738 return p;
1740 return NULL;
1744 * This helper attempts to cope with remotely renamed directories
1746 * It assumes that the caller is already holding
1747 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1749 * Note: If ever the locking in lock_rename() changes, then please
1750 * remember to update this too...
1752 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
1753 __releases(dcache_lock)
1755 struct mutex *m1 = NULL, *m2 = NULL;
1756 struct dentry *ret;
1758 /* If alias and dentry share a parent, then no extra locks required */
1759 if (alias->d_parent == dentry->d_parent)
1760 goto out_unalias;
1762 /* Check for loops */
1763 ret = ERR_PTR(-ELOOP);
1764 if (d_ancestor(alias, dentry))
1765 goto out_err;
1767 /* See lock_rename() */
1768 ret = ERR_PTR(-EBUSY);
1769 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1770 goto out_err;
1771 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1772 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1773 goto out_err;
1774 m2 = &alias->d_parent->d_inode->i_mutex;
1775 out_unalias:
1776 d_move_locked(alias, dentry);
1777 ret = alias;
1778 out_err:
1779 spin_unlock(&dcache_lock);
1780 if (m2)
1781 mutex_unlock(m2);
1782 if (m1)
1783 mutex_unlock(m1);
1784 return ret;
1788 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1789 * named dentry in place of the dentry to be replaced.
1791 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1793 struct dentry *dparent, *aparent;
1795 switch_names(dentry, anon);
1796 do_switch(dentry->d_name.hash, anon->d_name.hash);
1798 dparent = dentry->d_parent;
1799 aparent = anon->d_parent;
1801 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1802 list_del(&dentry->d_u.d_child);
1803 if (!IS_ROOT(dentry))
1804 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1805 else
1806 INIT_LIST_HEAD(&dentry->d_u.d_child);
1808 anon->d_parent = (dparent == dentry) ? anon : dparent;
1809 list_del(&anon->d_u.d_child);
1810 if (!IS_ROOT(anon))
1811 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1812 else
1813 INIT_LIST_HEAD(&anon->d_u.d_child);
1815 anon->d_flags &= ~DCACHE_DISCONNECTED;
1819 * d_materialise_unique - introduce an inode into the tree
1820 * @dentry: candidate dentry
1821 * @inode: inode to bind to the dentry, to which aliases may be attached
1823 * Introduces an dentry into the tree, substituting an extant disconnected
1824 * root directory alias in its place if there is one
1826 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1828 struct dentry *actual;
1830 BUG_ON(!d_unhashed(dentry));
1832 spin_lock(&dcache_lock);
1834 if (!inode) {
1835 actual = dentry;
1836 __d_instantiate(dentry, NULL);
1837 goto found_lock;
1840 if (S_ISDIR(inode->i_mode)) {
1841 struct dentry *alias;
1843 /* Does an aliased dentry already exist? */
1844 alias = __d_find_alias(inode, 0);
1845 if (alias) {
1846 actual = alias;
1847 /* Is this an anonymous mountpoint that we could splice
1848 * into our tree? */
1849 if (IS_ROOT(alias)) {
1850 spin_lock(&alias->d_lock);
1851 __d_materialise_dentry(dentry, alias);
1852 __d_drop(alias);
1853 goto found;
1855 /* Nope, but we must(!) avoid directory aliasing */
1856 actual = __d_unalias(dentry, alias);
1857 if (IS_ERR(actual))
1858 dput(alias);
1859 goto out_nolock;
1863 /* Add a unique reference */
1864 actual = __d_instantiate_unique(dentry, inode);
1865 if (!actual)
1866 actual = dentry;
1867 else if (unlikely(!d_unhashed(actual)))
1868 goto shouldnt_be_hashed;
1870 found_lock:
1871 spin_lock(&actual->d_lock);
1872 found:
1873 _d_rehash(actual);
1874 spin_unlock(&actual->d_lock);
1875 spin_unlock(&dcache_lock);
1876 out_nolock:
1877 if (actual == dentry) {
1878 security_d_instantiate(dentry, inode);
1879 return NULL;
1882 iput(inode);
1883 return actual;
1885 shouldnt_be_hashed:
1886 spin_unlock(&dcache_lock);
1887 BUG();
1890 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
1892 *buflen -= namelen;
1893 if (*buflen < 0)
1894 return -ENAMETOOLONG;
1895 *buffer -= namelen;
1896 memcpy(*buffer, str, namelen);
1897 return 0;
1900 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1902 return prepend(buffer, buflen, name->name, name->len);
1906 * __d_path - return the path of a dentry
1907 * @path: the dentry/vfsmount to report
1908 * @root: root vfsmnt/dentry (may be modified by this function)
1909 * @buffer: buffer to return value in
1910 * @buflen: buffer length
1912 * Convert a dentry into an ASCII path name. If the entry has been deleted
1913 * the string " (deleted)" is appended. Note that this is ambiguous.
1915 * Returns the buffer or an error code if the path was too long.
1917 * "buflen" should be positive. Caller holds the dcache_lock.
1919 * If path is not reachable from the supplied root, then the value of
1920 * root is changed (without modifying refcounts).
1922 char *__d_path(const struct path *path, struct path *root,
1923 char *buffer, int buflen)
1925 struct dentry *dentry = path->dentry;
1926 struct vfsmount *vfsmnt = path->mnt;
1927 char *end = buffer + buflen;
1928 char *retval;
1930 spin_lock(&vfsmount_lock);
1931 prepend(&end, &buflen, "\0", 1);
1932 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
1933 (prepend(&end, &buflen, " (deleted)", 10) != 0))
1934 goto Elong;
1936 if (buflen < 1)
1937 goto Elong;
1938 /* Get '/' right */
1939 retval = end-1;
1940 *retval = '/';
1942 for (;;) {
1943 struct dentry * parent;
1945 if (dentry == root->dentry && vfsmnt == root->mnt)
1946 break;
1947 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1948 /* Global root? */
1949 if (vfsmnt->mnt_parent == vfsmnt) {
1950 goto global_root;
1952 dentry = vfsmnt->mnt_mountpoint;
1953 vfsmnt = vfsmnt->mnt_parent;
1954 continue;
1956 parent = dentry->d_parent;
1957 prefetch(parent);
1958 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
1959 (prepend(&end, &buflen, "/", 1) != 0))
1960 goto Elong;
1961 retval = end;
1962 dentry = parent;
1965 out:
1966 spin_unlock(&vfsmount_lock);
1967 return retval;
1969 global_root:
1970 retval += 1; /* hit the slash */
1971 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
1972 goto Elong;
1973 root->mnt = vfsmnt;
1974 root->dentry = dentry;
1975 goto out;
1977 Elong:
1978 retval = ERR_PTR(-ENAMETOOLONG);
1979 goto out;
1983 * d_path - return the path of a dentry
1984 * @path: path to report
1985 * @buf: buffer to return value in
1986 * @buflen: buffer length
1988 * Convert a dentry into an ASCII path name. If the entry has been deleted
1989 * the string " (deleted)" is appended. Note that this is ambiguous.
1991 * Returns the buffer or an error code if the path was too long.
1993 * "buflen" should be positive.
1995 char *d_path(const struct path *path, char *buf, int buflen)
1997 char *res;
1998 struct path root;
1999 struct path tmp;
2002 * We have various synthetic filesystems that never get mounted. On
2003 * these filesystems dentries are never used for lookup purposes, and
2004 * thus don't need to be hashed. They also don't need a name until a
2005 * user wants to identify the object in /proc/pid/fd/. The little hack
2006 * below allows us to generate a name for these objects on demand:
2008 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2009 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2011 read_lock(&current->fs->lock);
2012 root = current->fs->root;
2013 path_get(&root);
2014 read_unlock(&current->fs->lock);
2015 spin_lock(&dcache_lock);
2016 tmp = root;
2017 res = __d_path(path, &tmp, buf, buflen);
2018 spin_unlock(&dcache_lock);
2019 path_put(&root);
2020 return res;
2024 * Helper function for dentry_operations.d_dname() members
2026 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2027 const char *fmt, ...)
2029 va_list args;
2030 char temp[64];
2031 int sz;
2033 va_start(args, fmt);
2034 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2035 va_end(args);
2037 if (sz > sizeof(temp) || sz > buflen)
2038 return ERR_PTR(-ENAMETOOLONG);
2040 buffer += buflen - sz;
2041 return memcpy(buffer, temp, sz);
2045 * Write full pathname from the root of the filesystem into the buffer.
2047 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2049 char *end = buf + buflen;
2050 char *retval;
2052 spin_lock(&dcache_lock);
2053 prepend(&end, &buflen, "\0", 1);
2054 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
2055 (prepend(&end, &buflen, "//deleted", 9) != 0))
2056 goto Elong;
2057 if (buflen < 1)
2058 goto Elong;
2059 /* Get '/' right */
2060 retval = end-1;
2061 *retval = '/';
2063 while (!IS_ROOT(dentry)) {
2064 struct dentry *parent = dentry->d_parent;
2066 prefetch(parent);
2067 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
2068 (prepend(&end, &buflen, "/", 1) != 0))
2069 goto Elong;
2071 retval = end;
2072 dentry = parent;
2074 spin_unlock(&dcache_lock);
2075 return retval;
2076 Elong:
2077 spin_unlock(&dcache_lock);
2078 return ERR_PTR(-ENAMETOOLONG);
2082 * NOTE! The user-level library version returns a
2083 * character pointer. The kernel system call just
2084 * returns the length of the buffer filled (which
2085 * includes the ending '\0' character), or a negative
2086 * error value. So libc would do something like
2088 * char *getcwd(char * buf, size_t size)
2090 * int retval;
2092 * retval = sys_getcwd(buf, size);
2093 * if (retval >= 0)
2094 * return buf;
2095 * errno = -retval;
2096 * return NULL;
2099 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2101 int error;
2102 struct path pwd, root;
2103 char *page = (char *) __get_free_page(GFP_USER);
2105 if (!page)
2106 return -ENOMEM;
2108 read_lock(&current->fs->lock);
2109 pwd = current->fs->pwd;
2110 path_get(&pwd);
2111 root = current->fs->root;
2112 path_get(&root);
2113 read_unlock(&current->fs->lock);
2115 error = -ENOENT;
2116 /* Has the current directory has been unlinked? */
2117 spin_lock(&dcache_lock);
2118 if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) {
2119 unsigned long len;
2120 struct path tmp = root;
2121 char * cwd;
2123 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
2124 spin_unlock(&dcache_lock);
2126 error = PTR_ERR(cwd);
2127 if (IS_ERR(cwd))
2128 goto out;
2130 error = -ERANGE;
2131 len = PAGE_SIZE + page - cwd;
2132 if (len <= size) {
2133 error = len;
2134 if (copy_to_user(buf, cwd, len))
2135 error = -EFAULT;
2137 } else
2138 spin_unlock(&dcache_lock);
2140 out:
2141 path_put(&pwd);
2142 path_put(&root);
2143 free_page((unsigned long) page);
2144 return error;
2148 * Test whether new_dentry is a subdirectory of old_dentry.
2150 * Trivially implemented using the dcache structure
2154 * is_subdir - is new dentry a subdirectory of old_dentry
2155 * @new_dentry: new dentry
2156 * @old_dentry: old dentry
2158 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2159 * Returns 0 otherwise.
2160 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2163 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2165 int result;
2166 unsigned long seq;
2168 /* FIXME: This is old behavior, needed? Please check callers. */
2169 if (new_dentry == old_dentry)
2170 return 1;
2173 * Need rcu_readlock to protect against the d_parent trashing
2174 * due to d_move
2176 rcu_read_lock();
2177 do {
2178 /* for restarting inner loop in case of seq retry */
2179 seq = read_seqbegin(&rename_lock);
2180 if (d_ancestor(old_dentry, new_dentry))
2181 result = 1;
2182 else
2183 result = 0;
2184 } while (read_seqretry(&rename_lock, seq));
2185 rcu_read_unlock();
2187 return result;
2190 void d_genocide(struct dentry *root)
2192 struct dentry *this_parent = root;
2193 struct list_head *next;
2195 spin_lock(&dcache_lock);
2196 repeat:
2197 next = this_parent->d_subdirs.next;
2198 resume:
2199 while (next != &this_parent->d_subdirs) {
2200 struct list_head *tmp = next;
2201 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2202 next = tmp->next;
2203 if (d_unhashed(dentry)||!dentry->d_inode)
2204 continue;
2205 if (!list_empty(&dentry->d_subdirs)) {
2206 this_parent = dentry;
2207 goto repeat;
2209 atomic_dec(&dentry->d_count);
2211 if (this_parent != root) {
2212 next = this_parent->d_u.d_child.next;
2213 atomic_dec(&this_parent->d_count);
2214 this_parent = this_parent->d_parent;
2215 goto resume;
2217 spin_unlock(&dcache_lock);
2221 * find_inode_number - check for dentry with name
2222 * @dir: directory to check
2223 * @name: Name to find.
2225 * Check whether a dentry already exists for the given name,
2226 * and return the inode number if it has an inode. Otherwise
2227 * 0 is returned.
2229 * This routine is used to post-process directory listings for
2230 * filesystems using synthetic inode numbers, and is necessary
2231 * to keep getcwd() working.
2234 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2236 struct dentry * dentry;
2237 ino_t ino = 0;
2239 dentry = d_hash_and_lookup(dir, name);
2240 if (dentry) {
2241 if (dentry->d_inode)
2242 ino = dentry->d_inode->i_ino;
2243 dput(dentry);
2245 return ino;
2248 static __initdata unsigned long dhash_entries;
2249 static int __init set_dhash_entries(char *str)
2251 if (!str)
2252 return 0;
2253 dhash_entries = simple_strtoul(str, &str, 0);
2254 return 1;
2256 __setup("dhash_entries=", set_dhash_entries);
2258 static void __init dcache_init_early(void)
2260 int loop;
2262 /* If hashes are distributed across NUMA nodes, defer
2263 * hash allocation until vmalloc space is available.
2265 if (hashdist)
2266 return;
2268 dentry_hashtable =
2269 alloc_large_system_hash("Dentry cache",
2270 sizeof(struct hlist_head),
2271 dhash_entries,
2273 HASH_EARLY,
2274 &d_hash_shift,
2275 &d_hash_mask,
2278 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2279 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2282 static void __init dcache_init(void)
2284 int loop;
2287 * A constructor could be added for stable state like the lists,
2288 * but it is probably not worth it because of the cache nature
2289 * of the dcache.
2291 dentry_cache = KMEM_CACHE(dentry,
2292 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2294 register_shrinker(&dcache_shrinker);
2296 /* Hash may have been set up in dcache_init_early */
2297 if (!hashdist)
2298 return;
2300 dentry_hashtable =
2301 alloc_large_system_hash("Dentry cache",
2302 sizeof(struct hlist_head),
2303 dhash_entries,
2306 &d_hash_shift,
2307 &d_hash_mask,
2310 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2311 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2314 /* SLAB cache for __getname() consumers */
2315 struct kmem_cache *names_cachep __read_mostly;
2317 /* SLAB cache for file structures */
2318 struct kmem_cache *filp_cachep __read_mostly;
2320 EXPORT_SYMBOL(d_genocide);
2322 void __init vfs_caches_init_early(void)
2324 dcache_init_early();
2325 inode_init_early();
2328 void __init vfs_caches_init(unsigned long mempages)
2330 unsigned long reserve;
2332 /* Base hash sizes on available memory, with a reserve equal to
2333 150% of current kernel size */
2335 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2336 mempages -= reserve;
2338 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
2339 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2341 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
2342 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2344 dcache_init();
2345 inode_init();
2346 files_init(mempages);
2347 mnt_init();
2348 bdev_cache_init();
2349 chrdev_init();
2352 EXPORT_SYMBOL(d_alloc);
2353 EXPORT_SYMBOL(d_alloc_root);
2354 EXPORT_SYMBOL(d_delete);
2355 EXPORT_SYMBOL(d_find_alias);
2356 EXPORT_SYMBOL(d_instantiate);
2357 EXPORT_SYMBOL(d_invalidate);
2358 EXPORT_SYMBOL(d_lookup);
2359 EXPORT_SYMBOL(d_move);
2360 EXPORT_SYMBOL_GPL(d_materialise_unique);
2361 EXPORT_SYMBOL(d_path);
2362 EXPORT_SYMBOL(d_prune_aliases);
2363 EXPORT_SYMBOL(d_rehash);
2364 EXPORT_SYMBOL(d_splice_alias);
2365 EXPORT_SYMBOL(d_add_ci);
2366 EXPORT_SYMBOL(d_validate);
2367 EXPORT_SYMBOL(dget_locked);
2368 EXPORT_SYMBOL(dput);
2369 EXPORT_SYMBOL(find_inode_number);
2370 EXPORT_SYMBOL(have_submounts);
2371 EXPORT_SYMBOL(names_cachep);
2372 EXPORT_SYMBOL(shrink_dcache_parent);
2373 EXPORT_SYMBOL(shrink_dcache_sb);