Merge branch 'for-upstream' of git://git.kernel.org/pub/scm/linux/kernel/git/dvrabel...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / dcache.c
blobe7a1a99b7464ef442a8af4ed20b0d45a4db25a65
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 if (dname_external(dentry))
73 kfree(dentry->d_name.name);
74 kmem_cache_free(dentry_cache, dentry);
77 static void d_callback(struct rcu_head *head)
79 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
80 __d_free(dentry);
84 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
85 * inside dcache_lock.
87 static void d_free(struct dentry *dentry)
89 if (dentry->d_op && dentry->d_op->d_release)
90 dentry->d_op->d_release(dentry);
91 /* if dentry was never inserted into hash, immediate free is OK */
92 if (hlist_unhashed(&dentry->d_hash))
93 __d_free(dentry);
94 else
95 call_rcu(&dentry->d_u.d_rcu, d_callback);
99 * Release the dentry's inode, using the filesystem
100 * d_iput() operation if defined.
102 static void dentry_iput(struct dentry * dentry)
103 __releases(dentry->d_lock)
104 __releases(dcache_lock)
106 struct inode *inode = dentry->d_inode;
107 if (inode) {
108 dentry->d_inode = NULL;
109 list_del_init(&dentry->d_alias);
110 spin_unlock(&dentry->d_lock);
111 spin_unlock(&dcache_lock);
112 if (!inode->i_nlink)
113 fsnotify_inoderemove(inode);
114 if (dentry->d_op && dentry->d_op->d_iput)
115 dentry->d_op->d_iput(dentry, inode);
116 else
117 iput(inode);
118 } else {
119 spin_unlock(&dentry->d_lock);
120 spin_unlock(&dcache_lock);
125 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
127 static void dentry_lru_add(struct dentry *dentry)
129 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
130 dentry->d_sb->s_nr_dentry_unused++;
131 dentry_stat.nr_unused++;
134 static void dentry_lru_add_tail(struct dentry *dentry)
136 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
137 dentry->d_sb->s_nr_dentry_unused++;
138 dentry_stat.nr_unused++;
141 static void dentry_lru_del(struct dentry *dentry)
143 if (!list_empty(&dentry->d_lru)) {
144 list_del(&dentry->d_lru);
145 dentry->d_sb->s_nr_dentry_unused--;
146 dentry_stat.nr_unused--;
150 static void dentry_lru_del_init(struct dentry *dentry)
152 if (likely(!list_empty(&dentry->d_lru))) {
153 list_del_init(&dentry->d_lru);
154 dentry->d_sb->s_nr_dentry_unused--;
155 dentry_stat.nr_unused--;
160 * d_kill - kill dentry and return parent
161 * @dentry: dentry to kill
163 * The dentry must already be unhashed and removed from the LRU.
165 * If this is the root of the dentry tree, return NULL.
167 static struct dentry *d_kill(struct dentry *dentry)
168 __releases(dentry->d_lock)
169 __releases(dcache_lock)
171 struct dentry *parent;
173 list_del(&dentry->d_u.d_child);
174 dentry_stat.nr_dentry--; /* For d_free, below */
175 /*drops the locks, at that point nobody can reach this dentry */
176 dentry_iput(dentry);
177 parent = dentry->d_parent;
178 d_free(dentry);
179 return dentry == parent ? NULL : parent;
183 * This is dput
185 * This is complicated by the fact that we do not want to put
186 * dentries that are no longer on any hash chain on the unused
187 * list: we'd much rather just get rid of them immediately.
189 * However, that implies that we have to traverse the dentry
190 * tree upwards to the parents which might _also_ now be
191 * scheduled for deletion (it may have been only waiting for
192 * its last child to go away).
194 * This tail recursion is done by hand as we don't want to depend
195 * on the compiler to always get this right (gcc generally doesn't).
196 * Real recursion would eat up our stack space.
200 * dput - release a dentry
201 * @dentry: dentry to release
203 * Release a dentry. This will drop the usage count and if appropriate
204 * call the dentry unlink method as well as removing it from the queues and
205 * releasing its resources. If the parent dentries were scheduled for release
206 * they too may now get deleted.
208 * no dcache lock, please.
211 void dput(struct dentry *dentry)
213 if (!dentry)
214 return;
216 repeat:
217 if (atomic_read(&dentry->d_count) == 1)
218 might_sleep();
219 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
220 return;
222 spin_lock(&dentry->d_lock);
223 if (atomic_read(&dentry->d_count)) {
224 spin_unlock(&dentry->d_lock);
225 spin_unlock(&dcache_lock);
226 return;
230 * AV: ->d_delete() is _NOT_ allowed to block now.
232 if (dentry->d_op && dentry->d_op->d_delete) {
233 if (dentry->d_op->d_delete(dentry))
234 goto unhash_it;
236 /* Unreachable? Get rid of it */
237 if (d_unhashed(dentry))
238 goto kill_it;
239 if (list_empty(&dentry->d_lru)) {
240 dentry->d_flags |= DCACHE_REFERENCED;
241 dentry_lru_add(dentry);
243 spin_unlock(&dentry->d_lock);
244 spin_unlock(&dcache_lock);
245 return;
247 unhash_it:
248 __d_drop(dentry);
249 kill_it:
250 /* if dentry was on the d_lru list delete it from there */
251 dentry_lru_del(dentry);
252 dentry = d_kill(dentry);
253 if (dentry)
254 goto repeat;
258 * d_invalidate - invalidate a dentry
259 * @dentry: dentry to invalidate
261 * Try to invalidate the dentry if it turns out to be
262 * possible. If there are other dentries that can be
263 * reached through this one we can't delete it and we
264 * return -EBUSY. On success we return 0.
266 * no dcache lock.
269 int d_invalidate(struct dentry * dentry)
272 * If it's already been dropped, return OK.
274 spin_lock(&dcache_lock);
275 if (d_unhashed(dentry)) {
276 spin_unlock(&dcache_lock);
277 return 0;
280 * Check whether to do a partial shrink_dcache
281 * to get rid of unused child entries.
283 if (!list_empty(&dentry->d_subdirs)) {
284 spin_unlock(&dcache_lock);
285 shrink_dcache_parent(dentry);
286 spin_lock(&dcache_lock);
290 * Somebody else still using it?
292 * If it's a directory, we can't drop it
293 * for fear of somebody re-populating it
294 * with children (even though dropping it
295 * would make it unreachable from the root,
296 * we might still populate it if it was a
297 * working directory or similar).
299 spin_lock(&dentry->d_lock);
300 if (atomic_read(&dentry->d_count) > 1) {
301 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
302 spin_unlock(&dentry->d_lock);
303 spin_unlock(&dcache_lock);
304 return -EBUSY;
308 __d_drop(dentry);
309 spin_unlock(&dentry->d_lock);
310 spin_unlock(&dcache_lock);
311 return 0;
314 /* This should be called _only_ with dcache_lock held */
316 static inline struct dentry * __dget_locked(struct dentry *dentry)
318 atomic_inc(&dentry->d_count);
319 dentry_lru_del_init(dentry);
320 return dentry;
323 struct dentry * dget_locked(struct dentry *dentry)
325 return __dget_locked(dentry);
329 * d_find_alias - grab a hashed alias of inode
330 * @inode: inode in question
331 * @want_discon: flag, used by d_splice_alias, to request
332 * that only a DISCONNECTED alias be returned.
334 * If inode has a hashed alias, or is a directory and has any alias,
335 * acquire the reference to alias and return it. Otherwise return NULL.
336 * Notice that if inode is a directory there can be only one alias and
337 * it can be unhashed only if it has no children, or if it is the root
338 * of a filesystem.
340 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
341 * any other hashed alias over that one unless @want_discon is set,
342 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
345 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
347 struct list_head *head, *next, *tmp;
348 struct dentry *alias, *discon_alias=NULL;
350 head = &inode->i_dentry;
351 next = inode->i_dentry.next;
352 while (next != head) {
353 tmp = next;
354 next = tmp->next;
355 prefetch(next);
356 alias = list_entry(tmp, struct dentry, d_alias);
357 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
358 if (IS_ROOT(alias) &&
359 (alias->d_flags & DCACHE_DISCONNECTED))
360 discon_alias = alias;
361 else if (!want_discon) {
362 __dget_locked(alias);
363 return alias;
367 if (discon_alias)
368 __dget_locked(discon_alias);
369 return discon_alias;
372 struct dentry * d_find_alias(struct inode *inode)
374 struct dentry *de = NULL;
376 if (!list_empty(&inode->i_dentry)) {
377 spin_lock(&dcache_lock);
378 de = __d_find_alias(inode, 0);
379 spin_unlock(&dcache_lock);
381 return de;
385 * Try to kill dentries associated with this inode.
386 * WARNING: you must own a reference to inode.
388 void d_prune_aliases(struct inode *inode)
390 struct dentry *dentry;
391 restart:
392 spin_lock(&dcache_lock);
393 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
394 spin_lock(&dentry->d_lock);
395 if (!atomic_read(&dentry->d_count)) {
396 __dget_locked(dentry);
397 __d_drop(dentry);
398 spin_unlock(&dentry->d_lock);
399 spin_unlock(&dcache_lock);
400 dput(dentry);
401 goto restart;
403 spin_unlock(&dentry->d_lock);
405 spin_unlock(&dcache_lock);
409 * Throw away a dentry - free the inode, dput the parent. This requires that
410 * the LRU list has already been removed.
412 * Try to prune ancestors as well. This is necessary to prevent
413 * quadratic behavior of shrink_dcache_parent(), but is also expected
414 * to be beneficial in reducing dentry cache fragmentation.
416 static void prune_one_dentry(struct dentry * dentry)
417 __releases(dentry->d_lock)
418 __releases(dcache_lock)
419 __acquires(dcache_lock)
421 __d_drop(dentry);
422 dentry = d_kill(dentry);
425 * Prune ancestors. Locking is simpler than in dput(),
426 * because dcache_lock needs to be taken anyway.
428 spin_lock(&dcache_lock);
429 while (dentry) {
430 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
431 return;
433 if (dentry->d_op && dentry->d_op->d_delete)
434 dentry->d_op->d_delete(dentry);
435 dentry_lru_del_init(dentry);
436 __d_drop(dentry);
437 dentry = d_kill(dentry);
438 spin_lock(&dcache_lock);
443 * Shrink the dentry LRU on a given superblock.
444 * @sb : superblock to shrink dentry LRU.
445 * @count: If count is NULL, we prune all dentries on superblock.
446 * @flags: If flags is non-zero, we need to do special processing based on
447 * which flags are set. This means we don't need to maintain multiple
448 * similar copies of this loop.
450 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
452 LIST_HEAD(referenced);
453 LIST_HEAD(tmp);
454 struct dentry *dentry;
455 int cnt = 0;
457 BUG_ON(!sb);
458 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
459 spin_lock(&dcache_lock);
460 if (count != NULL)
461 /* called from prune_dcache() and shrink_dcache_parent() */
462 cnt = *count;
463 restart:
464 if (count == NULL)
465 list_splice_init(&sb->s_dentry_lru, &tmp);
466 else {
467 while (!list_empty(&sb->s_dentry_lru)) {
468 dentry = list_entry(sb->s_dentry_lru.prev,
469 struct dentry, d_lru);
470 BUG_ON(dentry->d_sb != sb);
472 spin_lock(&dentry->d_lock);
474 * If we are honouring the DCACHE_REFERENCED flag and
475 * the dentry has this flag set, don't free it. Clear
476 * the flag and put it back on the LRU.
478 if ((flags & DCACHE_REFERENCED)
479 && (dentry->d_flags & DCACHE_REFERENCED)) {
480 dentry->d_flags &= ~DCACHE_REFERENCED;
481 list_move_tail(&dentry->d_lru, &referenced);
482 spin_unlock(&dentry->d_lock);
483 } else {
484 list_move_tail(&dentry->d_lru, &tmp);
485 spin_unlock(&dentry->d_lock);
486 cnt--;
487 if (!cnt)
488 break;
490 cond_resched_lock(&dcache_lock);
493 while (!list_empty(&tmp)) {
494 dentry = list_entry(tmp.prev, struct dentry, d_lru);
495 dentry_lru_del_init(dentry);
496 spin_lock(&dentry->d_lock);
498 * We found an inuse dentry which was not removed from
499 * the LRU because of laziness during lookup. Do not free
500 * it - just keep it off the LRU list.
502 if (atomic_read(&dentry->d_count)) {
503 spin_unlock(&dentry->d_lock);
504 continue;
506 prune_one_dentry(dentry);
507 /* dentry->d_lock was dropped in prune_one_dentry() */
508 cond_resched_lock(&dcache_lock);
510 if (count == NULL && !list_empty(&sb->s_dentry_lru))
511 goto restart;
512 if (count != NULL)
513 *count = cnt;
514 if (!list_empty(&referenced))
515 list_splice(&referenced, &sb->s_dentry_lru);
516 spin_unlock(&dcache_lock);
520 * prune_dcache - shrink the dcache
521 * @count: number of entries to try to free
523 * Shrink the dcache. This is done when we need more memory, or simply when we
524 * need to unmount something (at which point we need to unuse all dentries).
526 * This function may fail to free any resources if all the dentries are in use.
528 static void prune_dcache(int count)
530 struct super_block *sb;
531 int w_count;
532 int unused = dentry_stat.nr_unused;
533 int prune_ratio;
534 int pruned;
536 if (unused == 0 || count == 0)
537 return;
538 spin_lock(&dcache_lock);
539 restart:
540 if (count >= unused)
541 prune_ratio = 1;
542 else
543 prune_ratio = unused / count;
544 spin_lock(&sb_lock);
545 list_for_each_entry(sb, &super_blocks, s_list) {
546 if (sb->s_nr_dentry_unused == 0)
547 continue;
548 sb->s_count++;
549 /* Now, we reclaim unused dentrins with fairness.
550 * We reclaim them same percentage from each superblock.
551 * We calculate number of dentries to scan on this sb
552 * as follows, but the implementation is arranged to avoid
553 * overflows:
554 * number of dentries to scan on this sb =
555 * count * (number of dentries on this sb /
556 * number of dentries in the machine)
558 spin_unlock(&sb_lock);
559 if (prune_ratio != 1)
560 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
561 else
562 w_count = sb->s_nr_dentry_unused;
563 pruned = w_count;
565 * We need to be sure this filesystem isn't being unmounted,
566 * otherwise we could race with generic_shutdown_super(), and
567 * end up holding a reference to an inode while the filesystem
568 * is unmounted. So we try to get s_umount, and make sure
569 * s_root isn't NULL.
571 if (down_read_trylock(&sb->s_umount)) {
572 if ((sb->s_root != NULL) &&
573 (!list_empty(&sb->s_dentry_lru))) {
574 spin_unlock(&dcache_lock);
575 __shrink_dcache_sb(sb, &w_count,
576 DCACHE_REFERENCED);
577 pruned -= w_count;
578 spin_lock(&dcache_lock);
580 up_read(&sb->s_umount);
582 spin_lock(&sb_lock);
583 count -= pruned;
585 * restart only when sb is no longer on the list and
586 * we have more work to do.
588 if (__put_super_and_need_restart(sb) && count > 0) {
589 spin_unlock(&sb_lock);
590 goto restart;
593 spin_unlock(&sb_lock);
594 spin_unlock(&dcache_lock);
598 * shrink_dcache_sb - shrink dcache for a superblock
599 * @sb: superblock
601 * Shrink the dcache for the specified super block. This
602 * is used to free the dcache before unmounting a file
603 * system
605 void shrink_dcache_sb(struct super_block * sb)
607 __shrink_dcache_sb(sb, NULL, 0);
611 * destroy a single subtree of dentries for unmount
612 * - see the comments on shrink_dcache_for_umount() for a description of the
613 * locking
615 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
617 struct dentry *parent;
618 unsigned detached = 0;
620 BUG_ON(!IS_ROOT(dentry));
622 /* detach this root from the system */
623 spin_lock(&dcache_lock);
624 dentry_lru_del_init(dentry);
625 __d_drop(dentry);
626 spin_unlock(&dcache_lock);
628 for (;;) {
629 /* descend to the first leaf in the current subtree */
630 while (!list_empty(&dentry->d_subdirs)) {
631 struct dentry *loop;
633 /* this is a branch with children - detach all of them
634 * from the system in one go */
635 spin_lock(&dcache_lock);
636 list_for_each_entry(loop, &dentry->d_subdirs,
637 d_u.d_child) {
638 dentry_lru_del_init(loop);
639 __d_drop(loop);
640 cond_resched_lock(&dcache_lock);
642 spin_unlock(&dcache_lock);
644 /* move to the first child */
645 dentry = list_entry(dentry->d_subdirs.next,
646 struct dentry, d_u.d_child);
649 /* consume the dentries from this leaf up through its parents
650 * until we find one with children or run out altogether */
651 do {
652 struct inode *inode;
654 if (atomic_read(&dentry->d_count) != 0) {
655 printk(KERN_ERR
656 "BUG: Dentry %p{i=%lx,n=%s}"
657 " still in use (%d)"
658 " [unmount of %s %s]\n",
659 dentry,
660 dentry->d_inode ?
661 dentry->d_inode->i_ino : 0UL,
662 dentry->d_name.name,
663 atomic_read(&dentry->d_count),
664 dentry->d_sb->s_type->name,
665 dentry->d_sb->s_id);
666 BUG();
669 parent = dentry->d_parent;
670 if (parent == dentry)
671 parent = NULL;
672 else
673 atomic_dec(&parent->d_count);
675 list_del(&dentry->d_u.d_child);
676 detached++;
678 inode = dentry->d_inode;
679 if (inode) {
680 dentry->d_inode = NULL;
681 list_del_init(&dentry->d_alias);
682 if (dentry->d_op && dentry->d_op->d_iput)
683 dentry->d_op->d_iput(dentry, inode);
684 else
685 iput(inode);
688 d_free(dentry);
690 /* finished when we fall off the top of the tree,
691 * otherwise we ascend to the parent and move to the
692 * next sibling if there is one */
693 if (!parent)
694 goto out;
696 dentry = parent;
698 } while (list_empty(&dentry->d_subdirs));
700 dentry = list_entry(dentry->d_subdirs.next,
701 struct dentry, d_u.d_child);
703 out:
704 /* several dentries were freed, need to correct nr_dentry */
705 spin_lock(&dcache_lock);
706 dentry_stat.nr_dentry -= detached;
707 spin_unlock(&dcache_lock);
711 * destroy the dentries attached to a superblock on unmounting
712 * - we don't need to use dentry->d_lock, and only need dcache_lock when
713 * removing the dentry from the system lists and hashes because:
714 * - the superblock is detached from all mountings and open files, so the
715 * dentry trees will not be rearranged by the VFS
716 * - s_umount is write-locked, so the memory pressure shrinker will ignore
717 * any dentries belonging to this superblock that it comes across
718 * - the filesystem itself is no longer permitted to rearrange the dentries
719 * in this superblock
721 void shrink_dcache_for_umount(struct super_block *sb)
723 struct dentry *dentry;
725 if (down_read_trylock(&sb->s_umount))
726 BUG();
728 dentry = sb->s_root;
729 sb->s_root = NULL;
730 atomic_dec(&dentry->d_count);
731 shrink_dcache_for_umount_subtree(dentry);
733 while (!hlist_empty(&sb->s_anon)) {
734 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
735 shrink_dcache_for_umount_subtree(dentry);
740 * Search for at least 1 mount point in the dentry's subdirs.
741 * We descend to the next level whenever the d_subdirs
742 * list is non-empty and continue searching.
746 * have_submounts - check for mounts over a dentry
747 * @parent: dentry to check.
749 * Return true if the parent or its subdirectories contain
750 * a mount point
753 int have_submounts(struct dentry *parent)
755 struct dentry *this_parent = parent;
756 struct list_head *next;
758 spin_lock(&dcache_lock);
759 if (d_mountpoint(parent))
760 goto positive;
761 repeat:
762 next = this_parent->d_subdirs.next;
763 resume:
764 while (next != &this_parent->d_subdirs) {
765 struct list_head *tmp = next;
766 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
767 next = tmp->next;
768 /* Have we found a mount point ? */
769 if (d_mountpoint(dentry))
770 goto positive;
771 if (!list_empty(&dentry->d_subdirs)) {
772 this_parent = dentry;
773 goto repeat;
777 * All done at this level ... ascend and resume the search.
779 if (this_parent != parent) {
780 next = this_parent->d_u.d_child.next;
781 this_parent = this_parent->d_parent;
782 goto resume;
784 spin_unlock(&dcache_lock);
785 return 0; /* No mount points found in tree */
786 positive:
787 spin_unlock(&dcache_lock);
788 return 1;
792 * Search the dentry child list for the specified parent,
793 * and move any unused dentries to the end of the unused
794 * list for prune_dcache(). We descend to the next level
795 * whenever the d_subdirs list is non-empty and continue
796 * searching.
798 * It returns zero iff there are no unused children,
799 * otherwise it returns the number of children moved to
800 * the end of the unused list. This may not be the total
801 * number of unused children, because select_parent can
802 * drop the lock and return early due to latency
803 * constraints.
805 static int select_parent(struct dentry * parent)
807 struct dentry *this_parent = parent;
808 struct list_head *next;
809 int found = 0;
811 spin_lock(&dcache_lock);
812 repeat:
813 next = this_parent->d_subdirs.next;
814 resume:
815 while (next != &this_parent->d_subdirs) {
816 struct list_head *tmp = next;
817 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
818 next = tmp->next;
820 dentry_lru_del_init(dentry);
822 * move only zero ref count dentries to the end
823 * of the unused list for prune_dcache
825 if (!atomic_read(&dentry->d_count)) {
826 dentry_lru_add_tail(dentry);
827 found++;
831 * We can return to the caller if we have found some (this
832 * ensures forward progress). We'll be coming back to find
833 * the rest.
835 if (found && need_resched())
836 goto out;
839 * Descend a level if the d_subdirs list is non-empty.
841 if (!list_empty(&dentry->d_subdirs)) {
842 this_parent = dentry;
843 goto repeat;
847 * All done at this level ... ascend and resume the search.
849 if (this_parent != parent) {
850 next = this_parent->d_u.d_child.next;
851 this_parent = this_parent->d_parent;
852 goto resume;
854 out:
855 spin_unlock(&dcache_lock);
856 return found;
860 * shrink_dcache_parent - prune dcache
861 * @parent: parent of entries to prune
863 * Prune the dcache to remove unused children of the parent dentry.
866 void shrink_dcache_parent(struct dentry * parent)
868 struct super_block *sb = parent->d_sb;
869 int found;
871 while ((found = select_parent(parent)) != 0)
872 __shrink_dcache_sb(sb, &found, 0);
876 * Scan `nr' dentries and return the number which remain.
878 * We need to avoid reentering the filesystem if the caller is performing a
879 * GFP_NOFS allocation attempt. One example deadlock is:
881 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
882 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
883 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
885 * In this case we return -1 to tell the caller that we baled.
887 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
889 if (nr) {
890 if (!(gfp_mask & __GFP_FS))
891 return -1;
892 prune_dcache(nr);
894 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
897 static struct shrinker dcache_shrinker = {
898 .shrink = shrink_dcache_memory,
899 .seeks = DEFAULT_SEEKS,
903 * d_alloc - allocate a dcache entry
904 * @parent: parent of entry to allocate
905 * @name: qstr of the name
907 * Allocates a dentry. It returns %NULL if there is insufficient memory
908 * available. On a success the dentry is returned. The name passed in is
909 * copied and the copy passed in may be reused after this call.
912 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
914 struct dentry *dentry;
915 char *dname;
917 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
918 if (!dentry)
919 return NULL;
921 if (name->len > DNAME_INLINE_LEN-1) {
922 dname = kmalloc(name->len + 1, GFP_KERNEL);
923 if (!dname) {
924 kmem_cache_free(dentry_cache, dentry);
925 return NULL;
927 } else {
928 dname = dentry->d_iname;
930 dentry->d_name.name = dname;
932 dentry->d_name.len = name->len;
933 dentry->d_name.hash = name->hash;
934 memcpy(dname, name->name, name->len);
935 dname[name->len] = 0;
937 atomic_set(&dentry->d_count, 1);
938 dentry->d_flags = DCACHE_UNHASHED;
939 spin_lock_init(&dentry->d_lock);
940 dentry->d_inode = NULL;
941 dentry->d_parent = NULL;
942 dentry->d_sb = NULL;
943 dentry->d_op = NULL;
944 dentry->d_fsdata = NULL;
945 dentry->d_mounted = 0;
946 #ifdef CONFIG_PROFILING
947 dentry->d_cookie = NULL;
948 #endif
949 INIT_HLIST_NODE(&dentry->d_hash);
950 INIT_LIST_HEAD(&dentry->d_lru);
951 INIT_LIST_HEAD(&dentry->d_subdirs);
952 INIT_LIST_HEAD(&dentry->d_alias);
954 if (parent) {
955 dentry->d_parent = dget(parent);
956 dentry->d_sb = parent->d_sb;
957 } else {
958 INIT_LIST_HEAD(&dentry->d_u.d_child);
961 spin_lock(&dcache_lock);
962 if (parent)
963 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
964 dentry_stat.nr_dentry++;
965 spin_unlock(&dcache_lock);
967 return dentry;
970 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
972 struct qstr q;
974 q.name = name;
975 q.len = strlen(name);
976 q.hash = full_name_hash(q.name, q.len);
977 return d_alloc(parent, &q);
981 * d_instantiate - fill in inode information for a dentry
982 * @entry: dentry to complete
983 * @inode: inode to attach to this dentry
985 * Fill in inode information in the entry.
987 * This turns negative dentries into productive full members
988 * of society.
990 * NOTE! This assumes that the inode count has been incremented
991 * (or otherwise set) by the caller to indicate that it is now
992 * in use by the dcache.
995 void d_instantiate(struct dentry *entry, struct inode * inode)
997 BUG_ON(!list_empty(&entry->d_alias));
998 spin_lock(&dcache_lock);
999 if (inode)
1000 list_add(&entry->d_alias, &inode->i_dentry);
1001 entry->d_inode = inode;
1002 fsnotify_d_instantiate(entry, inode);
1003 spin_unlock(&dcache_lock);
1004 security_d_instantiate(entry, inode);
1008 * d_instantiate_unique - instantiate a non-aliased dentry
1009 * @entry: dentry to instantiate
1010 * @inode: inode to attach to this dentry
1012 * Fill in inode information in the entry. On success, it returns NULL.
1013 * If an unhashed alias of "entry" already exists, then we return the
1014 * aliased dentry instead and drop one reference to inode.
1016 * Note that in order to avoid conflicts with rename() etc, the caller
1017 * had better be holding the parent directory semaphore.
1019 * This also assumes that the inode count has been incremented
1020 * (or otherwise set) by the caller to indicate that it is now
1021 * in use by the dcache.
1023 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1024 struct inode *inode)
1026 struct dentry *alias;
1027 int len = entry->d_name.len;
1028 const char *name = entry->d_name.name;
1029 unsigned int hash = entry->d_name.hash;
1031 if (!inode) {
1032 entry->d_inode = NULL;
1033 return NULL;
1036 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1037 struct qstr *qstr = &alias->d_name;
1039 if (qstr->hash != hash)
1040 continue;
1041 if (alias->d_parent != entry->d_parent)
1042 continue;
1043 if (qstr->len != len)
1044 continue;
1045 if (memcmp(qstr->name, name, len))
1046 continue;
1047 dget_locked(alias);
1048 return alias;
1051 list_add(&entry->d_alias, &inode->i_dentry);
1052 entry->d_inode = inode;
1053 fsnotify_d_instantiate(entry, inode);
1054 return NULL;
1057 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1059 struct dentry *result;
1061 BUG_ON(!list_empty(&entry->d_alias));
1063 spin_lock(&dcache_lock);
1064 result = __d_instantiate_unique(entry, inode);
1065 spin_unlock(&dcache_lock);
1067 if (!result) {
1068 security_d_instantiate(entry, inode);
1069 return NULL;
1072 BUG_ON(!d_unhashed(result));
1073 iput(inode);
1074 return result;
1077 EXPORT_SYMBOL(d_instantiate_unique);
1080 * d_alloc_root - allocate root dentry
1081 * @root_inode: inode to allocate the root for
1083 * Allocate a root ("/") dentry for the inode given. The inode is
1084 * instantiated and returned. %NULL is returned if there is insufficient
1085 * memory or the inode passed is %NULL.
1088 struct dentry * d_alloc_root(struct inode * root_inode)
1090 struct dentry *res = NULL;
1092 if (root_inode) {
1093 static const struct qstr name = { .name = "/", .len = 1 };
1095 res = d_alloc(NULL, &name);
1096 if (res) {
1097 res->d_sb = root_inode->i_sb;
1098 res->d_parent = res;
1099 d_instantiate(res, root_inode);
1102 return res;
1105 static inline struct hlist_head *d_hash(struct dentry *parent,
1106 unsigned long hash)
1108 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1109 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1110 return dentry_hashtable + (hash & D_HASHMASK);
1114 * d_alloc_anon - allocate an anonymous dentry
1115 * @inode: inode to allocate the dentry for
1117 * This is similar to d_alloc_root. It is used by filesystems when
1118 * creating a dentry for a given inode, often in the process of
1119 * mapping a filehandle to a dentry. The returned dentry may be
1120 * anonymous, or may have a full name (if the inode was already
1121 * in the cache). The file system may need to make further
1122 * efforts to connect this dentry into the dcache properly.
1124 * When called on a directory inode, we must ensure that
1125 * the inode only ever has one dentry. If a dentry is
1126 * found, that is returned instead of allocating a new one.
1128 * On successful return, the reference to the inode has been transferred
1129 * to the dentry. If %NULL is returned (indicating kmalloc failure),
1130 * the reference on the inode has not been released.
1133 struct dentry * d_alloc_anon(struct inode *inode)
1135 static const struct qstr anonstring = { .name = "" };
1136 struct dentry *tmp;
1137 struct dentry *res;
1139 if ((res = d_find_alias(inode))) {
1140 iput(inode);
1141 return res;
1144 tmp = d_alloc(NULL, &anonstring);
1145 if (!tmp)
1146 return NULL;
1148 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1150 spin_lock(&dcache_lock);
1151 res = __d_find_alias(inode, 0);
1152 if (!res) {
1153 /* attach a disconnected dentry */
1154 res = tmp;
1155 tmp = NULL;
1156 spin_lock(&res->d_lock);
1157 res->d_sb = inode->i_sb;
1158 res->d_parent = res;
1159 res->d_inode = inode;
1160 res->d_flags |= DCACHE_DISCONNECTED;
1161 res->d_flags &= ~DCACHE_UNHASHED;
1162 list_add(&res->d_alias, &inode->i_dentry);
1163 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
1164 spin_unlock(&res->d_lock);
1166 inode = NULL; /* don't drop reference */
1168 spin_unlock(&dcache_lock);
1170 if (inode)
1171 iput(inode);
1172 if (tmp)
1173 dput(tmp);
1174 return res;
1179 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1180 * @inode: the inode which may have a disconnected dentry
1181 * @dentry: a negative dentry which we want to point to the inode.
1183 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1184 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1185 * and return it, else simply d_add the inode to the dentry and return NULL.
1187 * This is needed in the lookup routine of any filesystem that is exportable
1188 * (via knfsd) so that we can build dcache paths to directories effectively.
1190 * If a dentry was found and moved, then it is returned. Otherwise NULL
1191 * is returned. This matches the expected return value of ->lookup.
1194 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1196 struct dentry *new = NULL;
1198 if (inode && S_ISDIR(inode->i_mode)) {
1199 spin_lock(&dcache_lock);
1200 new = __d_find_alias(inode, 1);
1201 if (new) {
1202 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1203 fsnotify_d_instantiate(new, inode);
1204 spin_unlock(&dcache_lock);
1205 security_d_instantiate(new, inode);
1206 d_rehash(dentry);
1207 d_move(new, dentry);
1208 iput(inode);
1209 } else {
1210 /* d_instantiate takes dcache_lock, so we do it by hand */
1211 list_add(&dentry->d_alias, &inode->i_dentry);
1212 dentry->d_inode = inode;
1213 fsnotify_d_instantiate(dentry, inode);
1214 spin_unlock(&dcache_lock);
1215 security_d_instantiate(dentry, inode);
1216 d_rehash(dentry);
1218 } else
1219 d_add(dentry, inode);
1220 return new;
1224 * d_add_ci - lookup or allocate new dentry with case-exact name
1225 * @inode: the inode case-insensitive lookup has found
1226 * @dentry: the negative dentry that was passed to the parent's lookup func
1227 * @name: the case-exact name to be associated with the returned dentry
1229 * This is to avoid filling the dcache with case-insensitive names to the
1230 * same inode, only the actual correct case is stored in the dcache for
1231 * case-insensitive filesystems.
1233 * For a case-insensitive lookup match and if the the case-exact dentry
1234 * already exists in in the dcache, use it and return it.
1236 * If no entry exists with the exact case name, allocate new dentry with
1237 * the exact case, and return the spliced entry.
1239 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1240 struct qstr *name)
1242 int error;
1243 struct dentry *found;
1244 struct dentry *new;
1246 /* Does a dentry matching the name exist already? */
1247 found = d_hash_and_lookup(dentry->d_parent, name);
1248 /* If not, create it now and return */
1249 if (!found) {
1250 new = d_alloc(dentry->d_parent, name);
1251 if (!new) {
1252 error = -ENOMEM;
1253 goto err_out;
1255 found = d_splice_alias(inode, new);
1256 if (found) {
1257 dput(new);
1258 return found;
1260 return new;
1262 /* Matching dentry exists, check if it is negative. */
1263 if (found->d_inode) {
1264 if (unlikely(found->d_inode != inode)) {
1265 /* This can't happen because bad inodes are unhashed. */
1266 BUG_ON(!is_bad_inode(inode));
1267 BUG_ON(!is_bad_inode(found->d_inode));
1270 * Already have the inode and the dentry attached, decrement
1271 * the reference count to balance the iget() done
1272 * earlier on. We found the dentry using d_lookup() so it
1273 * cannot be disconnected and thus we do not need to worry
1274 * about any NFS/disconnectedness issues here.
1276 iput(inode);
1277 return found;
1280 * Negative dentry: instantiate it unless the inode is a directory and
1281 * has a 'disconnected' dentry (i.e. IS_ROOT and DCACHE_DISCONNECTED),
1282 * in which case d_move() that in place of the found dentry.
1284 if (!S_ISDIR(inode->i_mode)) {
1285 /* Not a directory; everything is easy. */
1286 d_instantiate(found, inode);
1287 return found;
1289 spin_lock(&dcache_lock);
1290 if (list_empty(&inode->i_dentry)) {
1292 * Directory without a 'disconnected' dentry; we need to do
1293 * d_instantiate() by hand because it takes dcache_lock which
1294 * we already hold.
1296 list_add(&found->d_alias, &inode->i_dentry);
1297 found->d_inode = inode;
1298 spin_unlock(&dcache_lock);
1299 security_d_instantiate(found, inode);
1300 return found;
1303 * Directory with a 'disconnected' dentry; get a reference to the
1304 * 'disconnected' dentry.
1306 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1307 dget_locked(new);
1308 spin_unlock(&dcache_lock);
1309 /* Do security vodoo. */
1310 security_d_instantiate(found, inode);
1311 /* Move new in place of found. */
1312 d_move(new, found);
1313 /* Balance the iget() we did above. */
1314 iput(inode);
1315 /* Throw away found. */
1316 dput(found);
1317 /* Use new as the actual dentry. */
1318 return new;
1320 err_out:
1321 iput(inode);
1322 return ERR_PTR(error);
1326 * d_lookup - search for a dentry
1327 * @parent: parent dentry
1328 * @name: qstr of name we wish to find
1330 * Searches the children of the parent dentry for the name in question. If
1331 * the dentry is found its reference count is incremented and the dentry
1332 * is returned. The caller must use d_put to free the entry when it has
1333 * finished using it. %NULL is returned on failure.
1335 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1336 * Memory barriers are used while updating and doing lockless traversal.
1337 * To avoid races with d_move while rename is happening, d_lock is used.
1339 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1340 * and name pointer in one structure pointed by d_qstr.
1342 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1343 * lookup is going on.
1345 * The dentry unused LRU is not updated even if lookup finds the required dentry
1346 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1347 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1348 * acquisition.
1350 * d_lookup() is protected against the concurrent renames in some unrelated
1351 * directory using the seqlockt_t rename_lock.
1354 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1356 struct dentry * dentry = NULL;
1357 unsigned long seq;
1359 do {
1360 seq = read_seqbegin(&rename_lock);
1361 dentry = __d_lookup(parent, name);
1362 if (dentry)
1363 break;
1364 } while (read_seqretry(&rename_lock, seq));
1365 return dentry;
1368 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1370 unsigned int len = name->len;
1371 unsigned int hash = name->hash;
1372 const unsigned char *str = name->name;
1373 struct hlist_head *head = d_hash(parent,hash);
1374 struct dentry *found = NULL;
1375 struct hlist_node *node;
1376 struct dentry *dentry;
1378 rcu_read_lock();
1380 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1381 struct qstr *qstr;
1383 if (dentry->d_name.hash != hash)
1384 continue;
1385 if (dentry->d_parent != parent)
1386 continue;
1388 spin_lock(&dentry->d_lock);
1391 * Recheck the dentry after taking the lock - d_move may have
1392 * changed things. Don't bother checking the hash because we're
1393 * about to compare the whole name anyway.
1395 if (dentry->d_parent != parent)
1396 goto next;
1398 /* non-existing due to RCU? */
1399 if (d_unhashed(dentry))
1400 goto next;
1403 * It is safe to compare names since d_move() cannot
1404 * change the qstr (protected by d_lock).
1406 qstr = &dentry->d_name;
1407 if (parent->d_op && parent->d_op->d_compare) {
1408 if (parent->d_op->d_compare(parent, qstr, name))
1409 goto next;
1410 } else {
1411 if (qstr->len != len)
1412 goto next;
1413 if (memcmp(qstr->name, str, len))
1414 goto next;
1417 atomic_inc(&dentry->d_count);
1418 found = dentry;
1419 spin_unlock(&dentry->d_lock);
1420 break;
1421 next:
1422 spin_unlock(&dentry->d_lock);
1424 rcu_read_unlock();
1426 return found;
1430 * d_hash_and_lookup - hash the qstr then search for a dentry
1431 * @dir: Directory to search in
1432 * @name: qstr of name we wish to find
1434 * On hash failure or on lookup failure NULL is returned.
1436 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1438 struct dentry *dentry = NULL;
1441 * Check for a fs-specific hash function. Note that we must
1442 * calculate the standard hash first, as the d_op->d_hash()
1443 * routine may choose to leave the hash value unchanged.
1445 name->hash = full_name_hash(name->name, name->len);
1446 if (dir->d_op && dir->d_op->d_hash) {
1447 if (dir->d_op->d_hash(dir, name) < 0)
1448 goto out;
1450 dentry = d_lookup(dir, name);
1451 out:
1452 return dentry;
1456 * d_validate - verify dentry provided from insecure source
1457 * @dentry: The dentry alleged to be valid child of @dparent
1458 * @dparent: The parent dentry (known to be valid)
1459 * @hash: Hash of the dentry
1460 * @len: Length of the name
1462 * An insecure source has sent us a dentry, here we verify it and dget() it.
1463 * This is used by ncpfs in its readdir implementation.
1464 * Zero is returned in the dentry is invalid.
1467 int d_validate(struct dentry *dentry, struct dentry *dparent)
1469 struct hlist_head *base;
1470 struct hlist_node *lhp;
1472 /* Check whether the ptr might be valid at all.. */
1473 if (!kmem_ptr_validate(dentry_cache, dentry))
1474 goto out;
1476 if (dentry->d_parent != dparent)
1477 goto out;
1479 spin_lock(&dcache_lock);
1480 base = d_hash(dparent, dentry->d_name.hash);
1481 hlist_for_each(lhp,base) {
1482 /* hlist_for_each_entry_rcu() not required for d_hash list
1483 * as it is parsed under dcache_lock
1485 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1486 __dget_locked(dentry);
1487 spin_unlock(&dcache_lock);
1488 return 1;
1491 spin_unlock(&dcache_lock);
1492 out:
1493 return 0;
1497 * When a file is deleted, we have two options:
1498 * - turn this dentry into a negative dentry
1499 * - unhash this dentry and free it.
1501 * Usually, we want to just turn this into
1502 * a negative dentry, but if anybody else is
1503 * currently using the dentry or the inode
1504 * we can't do that and we fall back on removing
1505 * it from the hash queues and waiting for
1506 * it to be deleted later when it has no users
1510 * d_delete - delete a dentry
1511 * @dentry: The dentry to delete
1513 * Turn the dentry into a negative dentry if possible, otherwise
1514 * remove it from the hash queues so it can be deleted later
1517 void d_delete(struct dentry * dentry)
1519 int isdir = 0;
1521 * Are we the only user?
1523 spin_lock(&dcache_lock);
1524 spin_lock(&dentry->d_lock);
1525 isdir = S_ISDIR(dentry->d_inode->i_mode);
1526 if (atomic_read(&dentry->d_count) == 1) {
1527 dentry_iput(dentry);
1528 fsnotify_nameremove(dentry, isdir);
1529 return;
1532 if (!d_unhashed(dentry))
1533 __d_drop(dentry);
1535 spin_unlock(&dentry->d_lock);
1536 spin_unlock(&dcache_lock);
1538 fsnotify_nameremove(dentry, isdir);
1541 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1544 entry->d_flags &= ~DCACHE_UNHASHED;
1545 hlist_add_head_rcu(&entry->d_hash, list);
1548 static void _d_rehash(struct dentry * entry)
1550 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1554 * d_rehash - add an entry back to the hash
1555 * @entry: dentry to add to the hash
1557 * Adds a dentry to the hash according to its name.
1560 void d_rehash(struct dentry * entry)
1562 spin_lock(&dcache_lock);
1563 spin_lock(&entry->d_lock);
1564 _d_rehash(entry);
1565 spin_unlock(&entry->d_lock);
1566 spin_unlock(&dcache_lock);
1569 #define do_switch(x,y) do { \
1570 __typeof__ (x) __tmp = x; \
1571 x = y; y = __tmp; } while (0)
1574 * When switching names, the actual string doesn't strictly have to
1575 * be preserved in the target - because we're dropping the target
1576 * anyway. As such, we can just do a simple memcpy() to copy over
1577 * the new name before we switch.
1579 * Note that we have to be a lot more careful about getting the hash
1580 * switched - we have to switch the hash value properly even if it
1581 * then no longer matches the actual (corrupted) string of the target.
1582 * The hash value has to match the hash queue that the dentry is on..
1584 static void switch_names(struct dentry *dentry, struct dentry *target)
1586 if (dname_external(target)) {
1587 if (dname_external(dentry)) {
1589 * Both external: swap the pointers
1591 do_switch(target->d_name.name, dentry->d_name.name);
1592 } else {
1594 * dentry:internal, target:external. Steal target's
1595 * storage and make target internal.
1597 memcpy(target->d_iname, dentry->d_name.name,
1598 dentry->d_name.len + 1);
1599 dentry->d_name.name = target->d_name.name;
1600 target->d_name.name = target->d_iname;
1602 } else {
1603 if (dname_external(dentry)) {
1605 * dentry:external, target:internal. Give dentry's
1606 * storage to target and make dentry internal
1608 memcpy(dentry->d_iname, target->d_name.name,
1609 target->d_name.len + 1);
1610 target->d_name.name = dentry->d_name.name;
1611 dentry->d_name.name = dentry->d_iname;
1612 } else {
1614 * Both are internal. Just copy target to dentry
1616 memcpy(dentry->d_iname, target->d_name.name,
1617 target->d_name.len + 1);
1623 * We cannibalize "target" when moving dentry on top of it,
1624 * because it's going to be thrown away anyway. We could be more
1625 * polite about it, though.
1627 * This forceful removal will result in ugly /proc output if
1628 * somebody holds a file open that got deleted due to a rename.
1629 * We could be nicer about the deleted file, and let it show
1630 * up under the name it had before it was deleted rather than
1631 * under the original name of the file that was moved on top of it.
1635 * d_move_locked - move a dentry
1636 * @dentry: entry to move
1637 * @target: new dentry
1639 * Update the dcache to reflect the move of a file name. Negative
1640 * dcache entries should not be moved in this way.
1642 static void d_move_locked(struct dentry * dentry, struct dentry * target)
1644 struct hlist_head *list;
1646 if (!dentry->d_inode)
1647 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1649 write_seqlock(&rename_lock);
1651 * XXXX: do we really need to take target->d_lock?
1653 if (target < dentry) {
1654 spin_lock(&target->d_lock);
1655 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1656 } else {
1657 spin_lock(&dentry->d_lock);
1658 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1661 /* Move the dentry to the target hash queue, if on different bucket */
1662 if (d_unhashed(dentry))
1663 goto already_unhashed;
1665 hlist_del_rcu(&dentry->d_hash);
1667 already_unhashed:
1668 list = d_hash(target->d_parent, target->d_name.hash);
1669 __d_rehash(dentry, list);
1671 /* Unhash the target: dput() will then get rid of it */
1672 __d_drop(target);
1674 list_del(&dentry->d_u.d_child);
1675 list_del(&target->d_u.d_child);
1677 /* Switch the names.. */
1678 switch_names(dentry, target);
1679 do_switch(dentry->d_name.len, target->d_name.len);
1680 do_switch(dentry->d_name.hash, target->d_name.hash);
1682 /* ... and switch the parents */
1683 if (IS_ROOT(dentry)) {
1684 dentry->d_parent = target->d_parent;
1685 target->d_parent = target;
1686 INIT_LIST_HEAD(&target->d_u.d_child);
1687 } else {
1688 do_switch(dentry->d_parent, target->d_parent);
1690 /* And add them back to the (new) parent lists */
1691 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1694 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1695 spin_unlock(&target->d_lock);
1696 fsnotify_d_move(dentry);
1697 spin_unlock(&dentry->d_lock);
1698 write_sequnlock(&rename_lock);
1702 * d_move - move a dentry
1703 * @dentry: entry to move
1704 * @target: new dentry
1706 * Update the dcache to reflect the move of a file name. Negative
1707 * dcache entries should not be moved in this way.
1710 void d_move(struct dentry * dentry, struct dentry * target)
1712 spin_lock(&dcache_lock);
1713 d_move_locked(dentry, target);
1714 spin_unlock(&dcache_lock);
1718 * Helper that returns 1 if p1 is a parent of p2, else 0
1720 static int d_isparent(struct dentry *p1, struct dentry *p2)
1722 struct dentry *p;
1724 for (p = p2; p->d_parent != p; p = p->d_parent) {
1725 if (p->d_parent == p1)
1726 return 1;
1728 return 0;
1732 * This helper attempts to cope with remotely renamed directories
1734 * It assumes that the caller is already holding
1735 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1737 * Note: If ever the locking in lock_rename() changes, then please
1738 * remember to update this too...
1740 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
1741 __releases(dcache_lock)
1743 struct mutex *m1 = NULL, *m2 = NULL;
1744 struct dentry *ret;
1746 /* If alias and dentry share a parent, then no extra locks required */
1747 if (alias->d_parent == dentry->d_parent)
1748 goto out_unalias;
1750 /* Check for loops */
1751 ret = ERR_PTR(-ELOOP);
1752 if (d_isparent(alias, dentry))
1753 goto out_err;
1755 /* See lock_rename() */
1756 ret = ERR_PTR(-EBUSY);
1757 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1758 goto out_err;
1759 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1760 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1761 goto out_err;
1762 m2 = &alias->d_parent->d_inode->i_mutex;
1763 out_unalias:
1764 d_move_locked(alias, dentry);
1765 ret = alias;
1766 out_err:
1767 spin_unlock(&dcache_lock);
1768 if (m2)
1769 mutex_unlock(m2);
1770 if (m1)
1771 mutex_unlock(m1);
1772 return ret;
1776 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1777 * named dentry in place of the dentry to be replaced.
1779 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1781 struct dentry *dparent, *aparent;
1783 switch_names(dentry, anon);
1784 do_switch(dentry->d_name.len, anon->d_name.len);
1785 do_switch(dentry->d_name.hash, anon->d_name.hash);
1787 dparent = dentry->d_parent;
1788 aparent = anon->d_parent;
1790 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1791 list_del(&dentry->d_u.d_child);
1792 if (!IS_ROOT(dentry))
1793 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1794 else
1795 INIT_LIST_HEAD(&dentry->d_u.d_child);
1797 anon->d_parent = (dparent == dentry) ? anon : dparent;
1798 list_del(&anon->d_u.d_child);
1799 if (!IS_ROOT(anon))
1800 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1801 else
1802 INIT_LIST_HEAD(&anon->d_u.d_child);
1804 anon->d_flags &= ~DCACHE_DISCONNECTED;
1808 * d_materialise_unique - introduce an inode into the tree
1809 * @dentry: candidate dentry
1810 * @inode: inode to bind to the dentry, to which aliases may be attached
1812 * Introduces an dentry into the tree, substituting an extant disconnected
1813 * root directory alias in its place if there is one
1815 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1817 struct dentry *actual;
1819 BUG_ON(!d_unhashed(dentry));
1821 spin_lock(&dcache_lock);
1823 if (!inode) {
1824 actual = dentry;
1825 dentry->d_inode = NULL;
1826 goto found_lock;
1829 if (S_ISDIR(inode->i_mode)) {
1830 struct dentry *alias;
1832 /* Does an aliased dentry already exist? */
1833 alias = __d_find_alias(inode, 0);
1834 if (alias) {
1835 actual = alias;
1836 /* Is this an anonymous mountpoint that we could splice
1837 * into our tree? */
1838 if (IS_ROOT(alias)) {
1839 spin_lock(&alias->d_lock);
1840 __d_materialise_dentry(dentry, alias);
1841 __d_drop(alias);
1842 goto found;
1844 /* Nope, but we must(!) avoid directory aliasing */
1845 actual = __d_unalias(dentry, alias);
1846 if (IS_ERR(actual))
1847 dput(alias);
1848 goto out_nolock;
1852 /* Add a unique reference */
1853 actual = __d_instantiate_unique(dentry, inode);
1854 if (!actual)
1855 actual = dentry;
1856 else if (unlikely(!d_unhashed(actual)))
1857 goto shouldnt_be_hashed;
1859 found_lock:
1860 spin_lock(&actual->d_lock);
1861 found:
1862 _d_rehash(actual);
1863 spin_unlock(&actual->d_lock);
1864 spin_unlock(&dcache_lock);
1865 out_nolock:
1866 if (actual == dentry) {
1867 security_d_instantiate(dentry, inode);
1868 return NULL;
1871 iput(inode);
1872 return actual;
1874 shouldnt_be_hashed:
1875 spin_unlock(&dcache_lock);
1876 BUG();
1879 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
1881 *buflen -= namelen;
1882 if (*buflen < 0)
1883 return -ENAMETOOLONG;
1884 *buffer -= namelen;
1885 memcpy(*buffer, str, namelen);
1886 return 0;
1889 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1891 return prepend(buffer, buflen, name->name, name->len);
1895 * __d_path - return the path of a dentry
1896 * @path: the dentry/vfsmount to report
1897 * @root: root vfsmnt/dentry (may be modified by this function)
1898 * @buffer: buffer to return value in
1899 * @buflen: buffer length
1901 * Convert a dentry into an ASCII path name. If the entry has been deleted
1902 * the string " (deleted)" is appended. Note that this is ambiguous.
1904 * Returns the buffer or an error code if the path was too long.
1906 * "buflen" should be positive. Caller holds the dcache_lock.
1908 * If path is not reachable from the supplied root, then the value of
1909 * root is changed (without modifying refcounts).
1911 char *__d_path(const struct path *path, struct path *root,
1912 char *buffer, int buflen)
1914 struct dentry *dentry = path->dentry;
1915 struct vfsmount *vfsmnt = path->mnt;
1916 char *end = buffer + buflen;
1917 char *retval;
1919 spin_lock(&vfsmount_lock);
1920 prepend(&end, &buflen, "\0", 1);
1921 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
1922 (prepend(&end, &buflen, " (deleted)", 10) != 0))
1923 goto Elong;
1925 if (buflen < 1)
1926 goto Elong;
1927 /* Get '/' right */
1928 retval = end-1;
1929 *retval = '/';
1931 for (;;) {
1932 struct dentry * parent;
1934 if (dentry == root->dentry && vfsmnt == root->mnt)
1935 break;
1936 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1937 /* Global root? */
1938 if (vfsmnt->mnt_parent == vfsmnt) {
1939 goto global_root;
1941 dentry = vfsmnt->mnt_mountpoint;
1942 vfsmnt = vfsmnt->mnt_parent;
1943 continue;
1945 parent = dentry->d_parent;
1946 prefetch(parent);
1947 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
1948 (prepend(&end, &buflen, "/", 1) != 0))
1949 goto Elong;
1950 retval = end;
1951 dentry = parent;
1954 out:
1955 spin_unlock(&vfsmount_lock);
1956 return retval;
1958 global_root:
1959 retval += 1; /* hit the slash */
1960 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
1961 goto Elong;
1962 root->mnt = vfsmnt;
1963 root->dentry = dentry;
1964 goto out;
1966 Elong:
1967 retval = ERR_PTR(-ENAMETOOLONG);
1968 goto out;
1972 * d_path - return the path of a dentry
1973 * @path: path to report
1974 * @buf: buffer to return value in
1975 * @buflen: buffer length
1977 * Convert a dentry into an ASCII path name. If the entry has been deleted
1978 * the string " (deleted)" is appended. Note that this is ambiguous.
1980 * Returns the buffer or an error code if the path was too long.
1982 * "buflen" should be positive.
1984 char *d_path(const struct path *path, char *buf, int buflen)
1986 char *res;
1987 struct path root;
1988 struct path tmp;
1991 * We have various synthetic filesystems that never get mounted. On
1992 * these filesystems dentries are never used for lookup purposes, and
1993 * thus don't need to be hashed. They also don't need a name until a
1994 * user wants to identify the object in /proc/pid/fd/. The little hack
1995 * below allows us to generate a name for these objects on demand:
1997 if (path->dentry->d_op && path->dentry->d_op->d_dname)
1998 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2000 read_lock(&current->fs->lock);
2001 root = current->fs->root;
2002 path_get(&root);
2003 read_unlock(&current->fs->lock);
2004 spin_lock(&dcache_lock);
2005 tmp = root;
2006 res = __d_path(path, &tmp, buf, buflen);
2007 spin_unlock(&dcache_lock);
2008 path_put(&root);
2009 return res;
2013 * Helper function for dentry_operations.d_dname() members
2015 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2016 const char *fmt, ...)
2018 va_list args;
2019 char temp[64];
2020 int sz;
2022 va_start(args, fmt);
2023 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2024 va_end(args);
2026 if (sz > sizeof(temp) || sz > buflen)
2027 return ERR_PTR(-ENAMETOOLONG);
2029 buffer += buflen - sz;
2030 return memcpy(buffer, temp, sz);
2034 * Write full pathname from the root of the filesystem into the buffer.
2036 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2038 char *end = buf + buflen;
2039 char *retval;
2041 spin_lock(&dcache_lock);
2042 prepend(&end, &buflen, "\0", 1);
2043 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
2044 (prepend(&end, &buflen, "//deleted", 9) != 0))
2045 goto Elong;
2046 if (buflen < 1)
2047 goto Elong;
2048 /* Get '/' right */
2049 retval = end-1;
2050 *retval = '/';
2052 while (!IS_ROOT(dentry)) {
2053 struct dentry *parent = dentry->d_parent;
2055 prefetch(parent);
2056 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
2057 (prepend(&end, &buflen, "/", 1) != 0))
2058 goto Elong;
2060 retval = end;
2061 dentry = parent;
2063 spin_unlock(&dcache_lock);
2064 return retval;
2065 Elong:
2066 spin_unlock(&dcache_lock);
2067 return ERR_PTR(-ENAMETOOLONG);
2071 * NOTE! The user-level library version returns a
2072 * character pointer. The kernel system call just
2073 * returns the length of the buffer filled (which
2074 * includes the ending '\0' character), or a negative
2075 * error value. So libc would do something like
2077 * char *getcwd(char * buf, size_t size)
2079 * int retval;
2081 * retval = sys_getcwd(buf, size);
2082 * if (retval >= 0)
2083 * return buf;
2084 * errno = -retval;
2085 * return NULL;
2088 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
2090 int error;
2091 struct path pwd, root;
2092 char *page = (char *) __get_free_page(GFP_USER);
2094 if (!page)
2095 return -ENOMEM;
2097 read_lock(&current->fs->lock);
2098 pwd = current->fs->pwd;
2099 path_get(&pwd);
2100 root = current->fs->root;
2101 path_get(&root);
2102 read_unlock(&current->fs->lock);
2104 error = -ENOENT;
2105 /* Has the current directory has been unlinked? */
2106 spin_lock(&dcache_lock);
2107 if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) {
2108 unsigned long len;
2109 struct path tmp = root;
2110 char * cwd;
2112 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
2113 spin_unlock(&dcache_lock);
2115 error = PTR_ERR(cwd);
2116 if (IS_ERR(cwd))
2117 goto out;
2119 error = -ERANGE;
2120 len = PAGE_SIZE + page - cwd;
2121 if (len <= size) {
2122 error = len;
2123 if (copy_to_user(buf, cwd, len))
2124 error = -EFAULT;
2126 } else
2127 spin_unlock(&dcache_lock);
2129 out:
2130 path_put(&pwd);
2131 path_put(&root);
2132 free_page((unsigned long) page);
2133 return error;
2137 * Test whether new_dentry is a subdirectory of old_dentry.
2139 * Trivially implemented using the dcache structure
2143 * is_subdir - is new dentry a subdirectory of old_dentry
2144 * @new_dentry: new dentry
2145 * @old_dentry: old dentry
2147 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2148 * Returns 0 otherwise.
2149 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2152 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
2154 int result;
2155 struct dentry * saved = new_dentry;
2156 unsigned long seq;
2158 /* need rcu_readlock to protect against the d_parent trashing due to
2159 * d_move
2161 rcu_read_lock();
2162 do {
2163 /* for restarting inner loop in case of seq retry */
2164 new_dentry = saved;
2165 result = 0;
2166 seq = read_seqbegin(&rename_lock);
2167 for (;;) {
2168 if (new_dentry != old_dentry) {
2169 struct dentry * parent = new_dentry->d_parent;
2170 if (parent == new_dentry)
2171 break;
2172 new_dentry = parent;
2173 continue;
2175 result = 1;
2176 break;
2178 } while (read_seqretry(&rename_lock, seq));
2179 rcu_read_unlock();
2181 return result;
2184 void d_genocide(struct dentry *root)
2186 struct dentry *this_parent = root;
2187 struct list_head *next;
2189 spin_lock(&dcache_lock);
2190 repeat:
2191 next = this_parent->d_subdirs.next;
2192 resume:
2193 while (next != &this_parent->d_subdirs) {
2194 struct list_head *tmp = next;
2195 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2196 next = tmp->next;
2197 if (d_unhashed(dentry)||!dentry->d_inode)
2198 continue;
2199 if (!list_empty(&dentry->d_subdirs)) {
2200 this_parent = dentry;
2201 goto repeat;
2203 atomic_dec(&dentry->d_count);
2205 if (this_parent != root) {
2206 next = this_parent->d_u.d_child.next;
2207 atomic_dec(&this_parent->d_count);
2208 this_parent = this_parent->d_parent;
2209 goto resume;
2211 spin_unlock(&dcache_lock);
2215 * find_inode_number - check for dentry with name
2216 * @dir: directory to check
2217 * @name: Name to find.
2219 * Check whether a dentry already exists for the given name,
2220 * and return the inode number if it has an inode. Otherwise
2221 * 0 is returned.
2223 * This routine is used to post-process directory listings for
2224 * filesystems using synthetic inode numbers, and is necessary
2225 * to keep getcwd() working.
2228 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2230 struct dentry * dentry;
2231 ino_t ino = 0;
2233 dentry = d_hash_and_lookup(dir, name);
2234 if (dentry) {
2235 if (dentry->d_inode)
2236 ino = dentry->d_inode->i_ino;
2237 dput(dentry);
2239 return ino;
2242 static __initdata unsigned long dhash_entries;
2243 static int __init set_dhash_entries(char *str)
2245 if (!str)
2246 return 0;
2247 dhash_entries = simple_strtoul(str, &str, 0);
2248 return 1;
2250 __setup("dhash_entries=", set_dhash_entries);
2252 static void __init dcache_init_early(void)
2254 int loop;
2256 /* If hashes are distributed across NUMA nodes, defer
2257 * hash allocation until vmalloc space is available.
2259 if (hashdist)
2260 return;
2262 dentry_hashtable =
2263 alloc_large_system_hash("Dentry cache",
2264 sizeof(struct hlist_head),
2265 dhash_entries,
2267 HASH_EARLY,
2268 &d_hash_shift,
2269 &d_hash_mask,
2272 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2273 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2276 static void __init dcache_init(void)
2278 int loop;
2281 * A constructor could be added for stable state like the lists,
2282 * but it is probably not worth it because of the cache nature
2283 * of the dcache.
2285 dentry_cache = KMEM_CACHE(dentry,
2286 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2288 register_shrinker(&dcache_shrinker);
2290 /* Hash may have been set up in dcache_init_early */
2291 if (!hashdist)
2292 return;
2294 dentry_hashtable =
2295 alloc_large_system_hash("Dentry cache",
2296 sizeof(struct hlist_head),
2297 dhash_entries,
2300 &d_hash_shift,
2301 &d_hash_mask,
2304 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2305 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2308 /* SLAB cache for __getname() consumers */
2309 struct kmem_cache *names_cachep __read_mostly;
2311 /* SLAB cache for file structures */
2312 struct kmem_cache *filp_cachep __read_mostly;
2314 EXPORT_SYMBOL(d_genocide);
2316 void __init vfs_caches_init_early(void)
2318 dcache_init_early();
2319 inode_init_early();
2322 void __init vfs_caches_init(unsigned long mempages)
2324 unsigned long reserve;
2326 /* Base hash sizes on available memory, with a reserve equal to
2327 150% of current kernel size */
2329 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2330 mempages -= reserve;
2332 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
2333 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2335 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
2336 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2338 dcache_init();
2339 inode_init();
2340 files_init(mempages);
2341 mnt_init();
2342 bdev_cache_init();
2343 chrdev_init();
2346 EXPORT_SYMBOL(d_alloc);
2347 EXPORT_SYMBOL(d_alloc_anon);
2348 EXPORT_SYMBOL(d_alloc_root);
2349 EXPORT_SYMBOL(d_delete);
2350 EXPORT_SYMBOL(d_find_alias);
2351 EXPORT_SYMBOL(d_instantiate);
2352 EXPORT_SYMBOL(d_invalidate);
2353 EXPORT_SYMBOL(d_lookup);
2354 EXPORT_SYMBOL(d_move);
2355 EXPORT_SYMBOL_GPL(d_materialise_unique);
2356 EXPORT_SYMBOL(d_path);
2357 EXPORT_SYMBOL(d_prune_aliases);
2358 EXPORT_SYMBOL(d_rehash);
2359 EXPORT_SYMBOL(d_splice_alias);
2360 EXPORT_SYMBOL(d_add_ci);
2361 EXPORT_SYMBOL(d_validate);
2362 EXPORT_SYMBOL(dget_locked);
2363 EXPORT_SYMBOL(dput);
2364 EXPORT_SYMBOL(find_inode_number);
2365 EXPORT_SYMBOL(have_submounts);
2366 EXPORT_SYMBOL(names_cachep);
2367 EXPORT_SYMBOL(shrink_dcache_parent);
2368 EXPORT_SYMBOL(shrink_dcache_sb);