IB/mthca: Query port fix
[linux-2.6/verdex.git] / fs / dcache.c
blob2355bddad8de12609bcc0b517a151883638c79eb
1 /*
2 * fs/dcache.c
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
9 /*
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
19 #include <linux/mm.h>
20 #include <linux/fs.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/smp_lock.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 static __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
44 EXPORT_SYMBOL(dcache_lock);
46 static kmem_cache_t *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;
64 static LIST_HEAD(dentry_unused);
66 /* Statistics gathering. */
67 struct dentry_stat_t dentry_stat = {
68 .age_limit = 45,
71 static void d_callback(struct rcu_head *head)
73 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
75 if (dname_external(dentry))
76 kfree(dentry->d_name.name);
77 kmem_cache_free(dentry_cache, dentry);
81 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
82 * inside dcache_lock.
84 static void d_free(struct dentry *dentry)
86 if (dentry->d_op && dentry->d_op->d_release)
87 dentry->d_op->d_release(dentry);
88 call_rcu(&dentry->d_u.d_rcu, d_callback);
92 * Release the dentry's inode, using the filesystem
93 * d_iput() operation if defined.
94 * Called with dcache_lock and per dentry lock held, drops both.
96 static void dentry_iput(struct dentry * dentry)
98 struct inode *inode = dentry->d_inode;
99 if (inode) {
100 dentry->d_inode = NULL;
101 list_del_init(&dentry->d_alias);
102 spin_unlock(&dentry->d_lock);
103 spin_unlock(&dcache_lock);
104 if (!inode->i_nlink)
105 fsnotify_inoderemove(inode);
106 if (dentry->d_op && dentry->d_op->d_iput)
107 dentry->d_op->d_iput(dentry, inode);
108 else
109 iput(inode);
110 } else {
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
117 * This is dput
119 * This is complicated by the fact that we do not want to put
120 * dentries that are no longer on any hash chain on the unused
121 * list: we'd much rather just get rid of them immediately.
123 * However, that implies that we have to traverse the dentry
124 * tree upwards to the parents which might _also_ now be
125 * scheduled for deletion (it may have been only waiting for
126 * its last child to go away).
128 * This tail recursion is done by hand as we don't want to depend
129 * on the compiler to always get this right (gcc generally doesn't).
130 * Real recursion would eat up our stack space.
134 * dput - release a dentry
135 * @dentry: dentry to release
137 * Release a dentry. This will drop the usage count and if appropriate
138 * call the dentry unlink method as well as removing it from the queues and
139 * releasing its resources. If the parent dentries were scheduled for release
140 * they too may now get deleted.
142 * no dcache lock, please.
145 void dput(struct dentry *dentry)
147 if (!dentry)
148 return;
150 repeat:
151 if (atomic_read(&dentry->d_count) == 1)
152 might_sleep();
153 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
154 return;
156 spin_lock(&dentry->d_lock);
157 if (atomic_read(&dentry->d_count)) {
158 spin_unlock(&dentry->d_lock);
159 spin_unlock(&dcache_lock);
160 return;
164 * AV: ->d_delete() is _NOT_ allowed to block now.
166 if (dentry->d_op && dentry->d_op->d_delete) {
167 if (dentry->d_op->d_delete(dentry))
168 goto unhash_it;
170 /* Unreachable? Get rid of it */
171 if (d_unhashed(dentry))
172 goto kill_it;
173 if (list_empty(&dentry->d_lru)) {
174 dentry->d_flags |= DCACHE_REFERENCED;
175 list_add(&dentry->d_lru, &dentry_unused);
176 dentry_stat.nr_unused++;
178 spin_unlock(&dentry->d_lock);
179 spin_unlock(&dcache_lock);
180 return;
182 unhash_it:
183 __d_drop(dentry);
185 kill_it: {
186 struct dentry *parent;
188 /* If dentry was on d_lru list
189 * delete it from there
191 if (!list_empty(&dentry->d_lru)) {
192 list_del(&dentry->d_lru);
193 dentry_stat.nr_unused--;
195 list_del(&dentry->d_u.d_child);
196 dentry_stat.nr_dentry--; /* For d_free, below */
197 /*drops the locks, at that point nobody can reach this dentry */
198 dentry_iput(dentry);
199 parent = dentry->d_parent;
200 d_free(dentry);
201 if (dentry == parent)
202 return;
203 dentry = parent;
204 goto repeat;
209 * d_invalidate - invalidate a dentry
210 * @dentry: dentry to invalidate
212 * Try to invalidate the dentry if it turns out to be
213 * possible. If there are other dentries that can be
214 * reached through this one we can't delete it and we
215 * return -EBUSY. On success we return 0.
217 * no dcache lock.
220 int d_invalidate(struct dentry * dentry)
223 * If it's already been dropped, return OK.
225 spin_lock(&dcache_lock);
226 if (d_unhashed(dentry)) {
227 spin_unlock(&dcache_lock);
228 return 0;
231 * Check whether to do a partial shrink_dcache
232 * to get rid of unused child entries.
234 if (!list_empty(&dentry->d_subdirs)) {
235 spin_unlock(&dcache_lock);
236 shrink_dcache_parent(dentry);
237 spin_lock(&dcache_lock);
241 * Somebody else still using it?
243 * If it's a directory, we can't drop it
244 * for fear of somebody re-populating it
245 * with children (even though dropping it
246 * would make it unreachable from the root,
247 * we might still populate it if it was a
248 * working directory or similar).
250 spin_lock(&dentry->d_lock);
251 if (atomic_read(&dentry->d_count) > 1) {
252 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
253 spin_unlock(&dentry->d_lock);
254 spin_unlock(&dcache_lock);
255 return -EBUSY;
259 __d_drop(dentry);
260 spin_unlock(&dentry->d_lock);
261 spin_unlock(&dcache_lock);
262 return 0;
265 /* This should be called _only_ with dcache_lock held */
267 static inline struct dentry * __dget_locked(struct dentry *dentry)
269 atomic_inc(&dentry->d_count);
270 if (!list_empty(&dentry->d_lru)) {
271 dentry_stat.nr_unused--;
272 list_del_init(&dentry->d_lru);
274 return dentry;
277 struct dentry * dget_locked(struct dentry *dentry)
279 return __dget_locked(dentry);
283 * d_find_alias - grab a hashed alias of inode
284 * @inode: inode in question
285 * @want_discon: flag, used by d_splice_alias, to request
286 * that only a DISCONNECTED alias be returned.
288 * If inode has a hashed alias, or is a directory and has any alias,
289 * acquire the reference to alias and return it. Otherwise return NULL.
290 * Notice that if inode is a directory there can be only one alias and
291 * it can be unhashed only if it has no children, or if it is the root
292 * of a filesystem.
294 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
295 * any other hashed alias over that one unless @want_discon is set,
296 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
299 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
301 struct list_head *head, *next, *tmp;
302 struct dentry *alias, *discon_alias=NULL;
304 head = &inode->i_dentry;
305 next = inode->i_dentry.next;
306 while (next != head) {
307 tmp = next;
308 next = tmp->next;
309 prefetch(next);
310 alias = list_entry(tmp, struct dentry, d_alias);
311 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
312 if (IS_ROOT(alias) &&
313 (alias->d_flags & DCACHE_DISCONNECTED))
314 discon_alias = alias;
315 else if (!want_discon) {
316 __dget_locked(alias);
317 return alias;
321 if (discon_alias)
322 __dget_locked(discon_alias);
323 return discon_alias;
326 struct dentry * d_find_alias(struct inode *inode)
328 struct dentry *de = NULL;
330 if (!list_empty(&inode->i_dentry)) {
331 spin_lock(&dcache_lock);
332 de = __d_find_alias(inode, 0);
333 spin_unlock(&dcache_lock);
335 return de;
339 * Try to kill dentries associated with this inode.
340 * WARNING: you must own a reference to inode.
342 void d_prune_aliases(struct inode *inode)
344 struct dentry *dentry;
345 restart:
346 spin_lock(&dcache_lock);
347 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
348 spin_lock(&dentry->d_lock);
349 if (!atomic_read(&dentry->d_count)) {
350 __dget_locked(dentry);
351 __d_drop(dentry);
352 spin_unlock(&dentry->d_lock);
353 spin_unlock(&dcache_lock);
354 dput(dentry);
355 goto restart;
357 spin_unlock(&dentry->d_lock);
359 spin_unlock(&dcache_lock);
363 * Throw away a dentry - free the inode, dput the parent. This requires that
364 * the LRU list has already been removed.
366 * Called with dcache_lock, drops it and then regains.
367 * Called with dentry->d_lock held, drops it.
369 static void prune_one_dentry(struct dentry * dentry)
371 struct dentry * parent;
373 __d_drop(dentry);
374 list_del(&dentry->d_u.d_child);
375 dentry_stat.nr_dentry--; /* For d_free, below */
376 dentry_iput(dentry);
377 parent = dentry->d_parent;
378 d_free(dentry);
379 if (parent != dentry)
380 dput(parent);
381 spin_lock(&dcache_lock);
385 * prune_dcache - shrink the dcache
386 * @count: number of entries to try and free
387 * @sb: if given, ignore dentries for other superblocks
388 * which are being unmounted.
390 * Shrink the dcache. This is done when we need
391 * more memory, or simply when we need to unmount
392 * something (at which point we need to unuse
393 * all dentries).
395 * This function may fail to free any resources if
396 * all the dentries are in use.
399 static void prune_dcache(int count, struct super_block *sb)
401 spin_lock(&dcache_lock);
402 for (; count ; count--) {
403 struct dentry *dentry;
404 struct list_head *tmp;
405 struct rw_semaphore *s_umount;
407 cond_resched_lock(&dcache_lock);
409 tmp = dentry_unused.prev;
410 if (sb) {
411 /* Try to find a dentry for this sb, but don't try
412 * too hard, if they aren't near the tail they will
413 * be moved down again soon
415 int skip = count;
416 while (skip && tmp != &dentry_unused &&
417 list_entry(tmp, struct dentry, d_lru)->d_sb != sb) {
418 skip--;
419 tmp = tmp->prev;
422 if (tmp == &dentry_unused)
423 break;
424 list_del_init(tmp);
425 prefetch(dentry_unused.prev);
426 dentry_stat.nr_unused--;
427 dentry = list_entry(tmp, struct dentry, d_lru);
429 spin_lock(&dentry->d_lock);
431 * We found an inuse dentry which was not removed from
432 * dentry_unused because of laziness during lookup. Do not free
433 * it - just keep it off the dentry_unused list.
435 if (atomic_read(&dentry->d_count)) {
436 spin_unlock(&dentry->d_lock);
437 continue;
439 /* If the dentry was recently referenced, don't free it. */
440 if (dentry->d_flags & DCACHE_REFERENCED) {
441 dentry->d_flags &= ~DCACHE_REFERENCED;
442 list_add(&dentry->d_lru, &dentry_unused);
443 dentry_stat.nr_unused++;
444 spin_unlock(&dentry->d_lock);
445 continue;
448 * If the dentry is not DCACHED_REFERENCED, it is time
449 * to remove it from the dcache, provided the super block is
450 * NULL (which means we are trying to reclaim memory)
451 * or this dentry belongs to the same super block that
452 * we want to shrink.
455 * If this dentry is for "my" filesystem, then I can prune it
456 * without taking the s_umount lock (I already hold it).
458 if (sb && dentry->d_sb == sb) {
459 prune_one_dentry(dentry);
460 continue;
463 * ...otherwise we need to be sure this filesystem isn't being
464 * unmounted, otherwise we could race with
465 * generic_shutdown_super(), and end up holding a reference to
466 * an inode while the filesystem is unmounted.
467 * So we try to get s_umount, and make sure s_root isn't NULL.
468 * (Take a local copy of s_umount to avoid a use-after-free of
469 * `dentry').
471 s_umount = &dentry->d_sb->s_umount;
472 if (down_read_trylock(s_umount)) {
473 if (dentry->d_sb->s_root != NULL) {
474 prune_one_dentry(dentry);
475 up_read(s_umount);
476 continue;
478 up_read(s_umount);
480 spin_unlock(&dentry->d_lock);
481 /* Cannot remove the first dentry, and it isn't appropriate
482 * to move it to the head of the list, so give up, and try
483 * later
485 break;
487 spin_unlock(&dcache_lock);
491 * Shrink the dcache for the specified super block.
492 * This allows us to unmount a device without disturbing
493 * the dcache for the other devices.
495 * This implementation makes just two traversals of the
496 * unused list. On the first pass we move the selected
497 * dentries to the most recent end, and on the second
498 * pass we free them. The second pass must restart after
499 * each dput(), but since the target dentries are all at
500 * the end, it's really just a single traversal.
504 * shrink_dcache_sb - shrink dcache for a superblock
505 * @sb: superblock
507 * Shrink the dcache for the specified super block. This
508 * is used to free the dcache before unmounting a file
509 * system
512 void shrink_dcache_sb(struct super_block * sb)
514 struct list_head *tmp, *next;
515 struct dentry *dentry;
518 * Pass one ... move the dentries for the specified
519 * superblock to the most recent end of the unused list.
521 spin_lock(&dcache_lock);
522 list_for_each_safe(tmp, next, &dentry_unused) {
523 dentry = list_entry(tmp, struct dentry, d_lru);
524 if (dentry->d_sb != sb)
525 continue;
526 list_move(tmp, &dentry_unused);
530 * Pass two ... free the dentries for this superblock.
532 repeat:
533 list_for_each_safe(tmp, next, &dentry_unused) {
534 dentry = list_entry(tmp, struct dentry, d_lru);
535 if (dentry->d_sb != sb)
536 continue;
537 dentry_stat.nr_unused--;
538 list_del_init(tmp);
539 spin_lock(&dentry->d_lock);
540 if (atomic_read(&dentry->d_count)) {
541 spin_unlock(&dentry->d_lock);
542 continue;
544 prune_one_dentry(dentry);
545 cond_resched_lock(&dcache_lock);
546 goto repeat;
548 spin_unlock(&dcache_lock);
552 * Search for at least 1 mount point in the dentry's subdirs.
553 * We descend to the next level whenever the d_subdirs
554 * list is non-empty and continue searching.
558 * have_submounts - check for mounts over a dentry
559 * @parent: dentry to check.
561 * Return true if the parent or its subdirectories contain
562 * a mount point
565 int have_submounts(struct dentry *parent)
567 struct dentry *this_parent = parent;
568 struct list_head *next;
570 spin_lock(&dcache_lock);
571 if (d_mountpoint(parent))
572 goto positive;
573 repeat:
574 next = this_parent->d_subdirs.next;
575 resume:
576 while (next != &this_parent->d_subdirs) {
577 struct list_head *tmp = next;
578 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
579 next = tmp->next;
580 /* Have we found a mount point ? */
581 if (d_mountpoint(dentry))
582 goto positive;
583 if (!list_empty(&dentry->d_subdirs)) {
584 this_parent = dentry;
585 goto repeat;
589 * All done at this level ... ascend and resume the search.
591 if (this_parent != parent) {
592 next = this_parent->d_u.d_child.next;
593 this_parent = this_parent->d_parent;
594 goto resume;
596 spin_unlock(&dcache_lock);
597 return 0; /* No mount points found in tree */
598 positive:
599 spin_unlock(&dcache_lock);
600 return 1;
604 * Search the dentry child list for the specified parent,
605 * and move any unused dentries to the end of the unused
606 * list for prune_dcache(). We descend to the next level
607 * whenever the d_subdirs list is non-empty and continue
608 * searching.
610 * It returns zero iff there are no unused children,
611 * otherwise it returns the number of children moved to
612 * the end of the unused list. This may not be the total
613 * number of unused children, because select_parent can
614 * drop the lock and return early due to latency
615 * constraints.
617 static int select_parent(struct dentry * parent)
619 struct dentry *this_parent = parent;
620 struct list_head *next;
621 int found = 0;
623 spin_lock(&dcache_lock);
624 repeat:
625 next = this_parent->d_subdirs.next;
626 resume:
627 while (next != &this_parent->d_subdirs) {
628 struct list_head *tmp = next;
629 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
630 next = tmp->next;
632 if (!list_empty(&dentry->d_lru)) {
633 dentry_stat.nr_unused--;
634 list_del_init(&dentry->d_lru);
637 * move only zero ref count dentries to the end
638 * of the unused list for prune_dcache
640 if (!atomic_read(&dentry->d_count)) {
641 list_add_tail(&dentry->d_lru, &dentry_unused);
642 dentry_stat.nr_unused++;
643 found++;
647 * We can return to the caller if we have found some (this
648 * ensures forward progress). We'll be coming back to find
649 * the rest.
651 if (found && need_resched())
652 goto out;
655 * Descend a level if the d_subdirs list is non-empty.
657 if (!list_empty(&dentry->d_subdirs)) {
658 this_parent = dentry;
659 goto repeat;
663 * All done at this level ... ascend and resume the search.
665 if (this_parent != parent) {
666 next = this_parent->d_u.d_child.next;
667 this_parent = this_parent->d_parent;
668 goto resume;
670 out:
671 spin_unlock(&dcache_lock);
672 return found;
676 * shrink_dcache_parent - prune dcache
677 * @parent: parent of entries to prune
679 * Prune the dcache to remove unused children of the parent dentry.
682 void shrink_dcache_parent(struct dentry * parent)
684 int found;
686 while ((found = select_parent(parent)) != 0)
687 prune_dcache(found, parent->d_sb);
691 * Scan `nr' dentries and return the number which remain.
693 * We need to avoid reentering the filesystem if the caller is performing a
694 * GFP_NOFS allocation attempt. One example deadlock is:
696 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
697 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
698 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
700 * In this case we return -1 to tell the caller that we baled.
702 static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
704 if (nr) {
705 if (!(gfp_mask & __GFP_FS))
706 return -1;
707 prune_dcache(nr, NULL);
709 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
713 * d_alloc - allocate a dcache entry
714 * @parent: parent of entry to allocate
715 * @name: qstr of the name
717 * Allocates a dentry. It returns %NULL if there is insufficient memory
718 * available. On a success the dentry is returned. The name passed in is
719 * copied and the copy passed in may be reused after this call.
722 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
724 struct dentry *dentry;
725 char *dname;
727 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
728 if (!dentry)
729 return NULL;
731 if (name->len > DNAME_INLINE_LEN-1) {
732 dname = kmalloc(name->len + 1, GFP_KERNEL);
733 if (!dname) {
734 kmem_cache_free(dentry_cache, dentry);
735 return NULL;
737 } else {
738 dname = dentry->d_iname;
740 dentry->d_name.name = dname;
742 dentry->d_name.len = name->len;
743 dentry->d_name.hash = name->hash;
744 memcpy(dname, name->name, name->len);
745 dname[name->len] = 0;
747 atomic_set(&dentry->d_count, 1);
748 dentry->d_flags = DCACHE_UNHASHED;
749 spin_lock_init(&dentry->d_lock);
750 dentry->d_inode = NULL;
751 dentry->d_parent = NULL;
752 dentry->d_sb = NULL;
753 dentry->d_op = NULL;
754 dentry->d_fsdata = NULL;
755 dentry->d_mounted = 0;
756 #ifdef CONFIG_PROFILING
757 dentry->d_cookie = NULL;
758 #endif
759 INIT_HLIST_NODE(&dentry->d_hash);
760 INIT_LIST_HEAD(&dentry->d_lru);
761 INIT_LIST_HEAD(&dentry->d_subdirs);
762 INIT_LIST_HEAD(&dentry->d_alias);
764 if (parent) {
765 dentry->d_parent = dget(parent);
766 dentry->d_sb = parent->d_sb;
767 } else {
768 INIT_LIST_HEAD(&dentry->d_u.d_child);
771 spin_lock(&dcache_lock);
772 if (parent)
773 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
774 dentry_stat.nr_dentry++;
775 spin_unlock(&dcache_lock);
777 return dentry;
780 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
782 struct qstr q;
784 q.name = name;
785 q.len = strlen(name);
786 q.hash = full_name_hash(q.name, q.len);
787 return d_alloc(parent, &q);
791 * d_instantiate - fill in inode information for a dentry
792 * @entry: dentry to complete
793 * @inode: inode to attach to this dentry
795 * Fill in inode information in the entry.
797 * This turns negative dentries into productive full members
798 * of society.
800 * NOTE! This assumes that the inode count has been incremented
801 * (or otherwise set) by the caller to indicate that it is now
802 * in use by the dcache.
805 void d_instantiate(struct dentry *entry, struct inode * inode)
807 BUG_ON(!list_empty(&entry->d_alias));
808 spin_lock(&dcache_lock);
809 if (inode)
810 list_add(&entry->d_alias, &inode->i_dentry);
811 entry->d_inode = inode;
812 fsnotify_d_instantiate(entry, inode);
813 spin_unlock(&dcache_lock);
814 security_d_instantiate(entry, inode);
818 * d_instantiate_unique - instantiate a non-aliased dentry
819 * @entry: dentry to instantiate
820 * @inode: inode to attach to this dentry
822 * Fill in inode information in the entry. On success, it returns NULL.
823 * If an unhashed alias of "entry" already exists, then we return the
824 * aliased dentry instead and drop one reference to inode.
826 * Note that in order to avoid conflicts with rename() etc, the caller
827 * had better be holding the parent directory semaphore.
829 * This also assumes that the inode count has been incremented
830 * (or otherwise set) by the caller to indicate that it is now
831 * in use by the dcache.
833 static struct dentry *__d_instantiate_unique(struct dentry *entry,
834 struct inode *inode)
836 struct dentry *alias;
837 int len = entry->d_name.len;
838 const char *name = entry->d_name.name;
839 unsigned int hash = entry->d_name.hash;
841 if (!inode) {
842 entry->d_inode = NULL;
843 return NULL;
846 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
847 struct qstr *qstr = &alias->d_name;
849 if (qstr->hash != hash)
850 continue;
851 if (alias->d_parent != entry->d_parent)
852 continue;
853 if (qstr->len != len)
854 continue;
855 if (memcmp(qstr->name, name, len))
856 continue;
857 dget_locked(alias);
858 return alias;
861 list_add(&entry->d_alias, &inode->i_dentry);
862 entry->d_inode = inode;
863 fsnotify_d_instantiate(entry, inode);
864 return NULL;
867 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
869 struct dentry *result;
871 BUG_ON(!list_empty(&entry->d_alias));
873 spin_lock(&dcache_lock);
874 result = __d_instantiate_unique(entry, inode);
875 spin_unlock(&dcache_lock);
877 if (!result) {
878 security_d_instantiate(entry, inode);
879 return NULL;
882 BUG_ON(!d_unhashed(result));
883 iput(inode);
884 return result;
887 EXPORT_SYMBOL(d_instantiate_unique);
890 * d_alloc_root - allocate root dentry
891 * @root_inode: inode to allocate the root for
893 * Allocate a root ("/") dentry for the inode given. The inode is
894 * instantiated and returned. %NULL is returned if there is insufficient
895 * memory or the inode passed is %NULL.
898 struct dentry * d_alloc_root(struct inode * root_inode)
900 struct dentry *res = NULL;
902 if (root_inode) {
903 static const struct qstr name = { .name = "/", .len = 1 };
905 res = d_alloc(NULL, &name);
906 if (res) {
907 res->d_sb = root_inode->i_sb;
908 res->d_parent = res;
909 d_instantiate(res, root_inode);
912 return res;
915 static inline struct hlist_head *d_hash(struct dentry *parent,
916 unsigned long hash)
918 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
919 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
920 return dentry_hashtable + (hash & D_HASHMASK);
924 * d_alloc_anon - allocate an anonymous dentry
925 * @inode: inode to allocate the dentry for
927 * This is similar to d_alloc_root. It is used by filesystems when
928 * creating a dentry for a given inode, often in the process of
929 * mapping a filehandle to a dentry. The returned dentry may be
930 * anonymous, or may have a full name (if the inode was already
931 * in the cache). The file system may need to make further
932 * efforts to connect this dentry into the dcache properly.
934 * When called on a directory inode, we must ensure that
935 * the inode only ever has one dentry. If a dentry is
936 * found, that is returned instead of allocating a new one.
938 * On successful return, the reference to the inode has been transferred
939 * to the dentry. If %NULL is returned (indicating kmalloc failure),
940 * the reference on the inode has not been released.
943 struct dentry * d_alloc_anon(struct inode *inode)
945 static const struct qstr anonstring = { .name = "" };
946 struct dentry *tmp;
947 struct dentry *res;
949 if ((res = d_find_alias(inode))) {
950 iput(inode);
951 return res;
954 tmp = d_alloc(NULL, &anonstring);
955 if (!tmp)
956 return NULL;
958 tmp->d_parent = tmp; /* make sure dput doesn't croak */
960 spin_lock(&dcache_lock);
961 res = __d_find_alias(inode, 0);
962 if (!res) {
963 /* attach a disconnected dentry */
964 res = tmp;
965 tmp = NULL;
966 spin_lock(&res->d_lock);
967 res->d_sb = inode->i_sb;
968 res->d_parent = res;
969 res->d_inode = inode;
970 res->d_flags |= DCACHE_DISCONNECTED;
971 res->d_flags &= ~DCACHE_UNHASHED;
972 list_add(&res->d_alias, &inode->i_dentry);
973 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
974 spin_unlock(&res->d_lock);
976 inode = NULL; /* don't drop reference */
978 spin_unlock(&dcache_lock);
980 if (inode)
981 iput(inode);
982 if (tmp)
983 dput(tmp);
984 return res;
989 * d_splice_alias - splice a disconnected dentry into the tree if one exists
990 * @inode: the inode which may have a disconnected dentry
991 * @dentry: a negative dentry which we want to point to the inode.
993 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
994 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
995 * and return it, else simply d_add the inode to the dentry and return NULL.
997 * This is needed in the lookup routine of any filesystem that is exportable
998 * (via knfsd) so that we can build dcache paths to directories effectively.
1000 * If a dentry was found and moved, then it is returned. Otherwise NULL
1001 * is returned. This matches the expected return value of ->lookup.
1004 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1006 struct dentry *new = NULL;
1008 if (inode && S_ISDIR(inode->i_mode)) {
1009 spin_lock(&dcache_lock);
1010 new = __d_find_alias(inode, 1);
1011 if (new) {
1012 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1013 fsnotify_d_instantiate(new, inode);
1014 spin_unlock(&dcache_lock);
1015 security_d_instantiate(new, inode);
1016 d_rehash(dentry);
1017 d_move(new, dentry);
1018 iput(inode);
1019 } else {
1020 /* d_instantiate takes dcache_lock, so we do it by hand */
1021 list_add(&dentry->d_alias, &inode->i_dentry);
1022 dentry->d_inode = inode;
1023 fsnotify_d_instantiate(dentry, inode);
1024 spin_unlock(&dcache_lock);
1025 security_d_instantiate(dentry, inode);
1026 d_rehash(dentry);
1028 } else
1029 d_add(dentry, inode);
1030 return new;
1035 * d_lookup - search for a dentry
1036 * @parent: parent dentry
1037 * @name: qstr of name we wish to find
1039 * Searches the children of the parent dentry for the name in question. If
1040 * the dentry is found its reference count is incremented and the dentry
1041 * is returned. The caller must use d_put to free the entry when it has
1042 * finished using it. %NULL is returned on failure.
1044 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1045 * Memory barriers are used while updating and doing lockless traversal.
1046 * To avoid races with d_move while rename is happening, d_lock is used.
1048 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1049 * and name pointer in one structure pointed by d_qstr.
1051 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1052 * lookup is going on.
1054 * dentry_unused list is not updated even if lookup finds the required dentry
1055 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1056 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1057 * acquisition.
1059 * d_lookup() is protected against the concurrent renames in some unrelated
1060 * directory using the seqlockt_t rename_lock.
1063 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1065 struct dentry * dentry = NULL;
1066 unsigned long seq;
1068 do {
1069 seq = read_seqbegin(&rename_lock);
1070 dentry = __d_lookup(parent, name);
1071 if (dentry)
1072 break;
1073 } while (read_seqretry(&rename_lock, seq));
1074 return dentry;
1077 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1079 unsigned int len = name->len;
1080 unsigned int hash = name->hash;
1081 const unsigned char *str = name->name;
1082 struct hlist_head *head = d_hash(parent,hash);
1083 struct dentry *found = NULL;
1084 struct hlist_node *node;
1085 struct dentry *dentry;
1087 rcu_read_lock();
1089 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1090 struct qstr *qstr;
1092 if (dentry->d_name.hash != hash)
1093 continue;
1094 if (dentry->d_parent != parent)
1095 continue;
1097 spin_lock(&dentry->d_lock);
1100 * Recheck the dentry after taking the lock - d_move may have
1101 * changed things. Don't bother checking the hash because we're
1102 * about to compare the whole name anyway.
1104 if (dentry->d_parent != parent)
1105 goto next;
1108 * It is safe to compare names since d_move() cannot
1109 * change the qstr (protected by d_lock).
1111 qstr = &dentry->d_name;
1112 if (parent->d_op && parent->d_op->d_compare) {
1113 if (parent->d_op->d_compare(parent, qstr, name))
1114 goto next;
1115 } else {
1116 if (qstr->len != len)
1117 goto next;
1118 if (memcmp(qstr->name, str, len))
1119 goto next;
1122 if (!d_unhashed(dentry)) {
1123 atomic_inc(&dentry->d_count);
1124 found = dentry;
1126 spin_unlock(&dentry->d_lock);
1127 break;
1128 next:
1129 spin_unlock(&dentry->d_lock);
1131 rcu_read_unlock();
1133 return found;
1137 * d_hash_and_lookup - hash the qstr then search for a dentry
1138 * @dir: Directory to search in
1139 * @name: qstr of name we wish to find
1141 * On hash failure or on lookup failure NULL is returned.
1143 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1145 struct dentry *dentry = NULL;
1148 * Check for a fs-specific hash function. Note that we must
1149 * calculate the standard hash first, as the d_op->d_hash()
1150 * routine may choose to leave the hash value unchanged.
1152 name->hash = full_name_hash(name->name, name->len);
1153 if (dir->d_op && dir->d_op->d_hash) {
1154 if (dir->d_op->d_hash(dir, name) < 0)
1155 goto out;
1157 dentry = d_lookup(dir, name);
1158 out:
1159 return dentry;
1163 * d_validate - verify dentry provided from insecure source
1164 * @dentry: The dentry alleged to be valid child of @dparent
1165 * @dparent: The parent dentry (known to be valid)
1166 * @hash: Hash of the dentry
1167 * @len: Length of the name
1169 * An insecure source has sent us a dentry, here we verify it and dget() it.
1170 * This is used by ncpfs in its readdir implementation.
1171 * Zero is returned in the dentry is invalid.
1174 int d_validate(struct dentry *dentry, struct dentry *dparent)
1176 struct hlist_head *base;
1177 struct hlist_node *lhp;
1179 /* Check whether the ptr might be valid at all.. */
1180 if (!kmem_ptr_validate(dentry_cache, dentry))
1181 goto out;
1183 if (dentry->d_parent != dparent)
1184 goto out;
1186 spin_lock(&dcache_lock);
1187 base = d_hash(dparent, dentry->d_name.hash);
1188 hlist_for_each(lhp,base) {
1189 /* hlist_for_each_entry_rcu() not required for d_hash list
1190 * as it is parsed under dcache_lock
1192 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1193 __dget_locked(dentry);
1194 spin_unlock(&dcache_lock);
1195 return 1;
1198 spin_unlock(&dcache_lock);
1199 out:
1200 return 0;
1204 * When a file is deleted, we have two options:
1205 * - turn this dentry into a negative dentry
1206 * - unhash this dentry and free it.
1208 * Usually, we want to just turn this into
1209 * a negative dentry, but if anybody else is
1210 * currently using the dentry or the inode
1211 * we can't do that and we fall back on removing
1212 * it from the hash queues and waiting for
1213 * it to be deleted later when it has no users
1217 * d_delete - delete a dentry
1218 * @dentry: The dentry to delete
1220 * Turn the dentry into a negative dentry if possible, otherwise
1221 * remove it from the hash queues so it can be deleted later
1224 void d_delete(struct dentry * dentry)
1226 int isdir = 0;
1228 * Are we the only user?
1230 spin_lock(&dcache_lock);
1231 spin_lock(&dentry->d_lock);
1232 isdir = S_ISDIR(dentry->d_inode->i_mode);
1233 if (atomic_read(&dentry->d_count) == 1) {
1234 dentry_iput(dentry);
1235 fsnotify_nameremove(dentry, isdir);
1237 /* remove this and other inotify debug checks after 2.6.18 */
1238 dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
1239 return;
1242 if (!d_unhashed(dentry))
1243 __d_drop(dentry);
1245 spin_unlock(&dentry->d_lock);
1246 spin_unlock(&dcache_lock);
1248 fsnotify_nameremove(dentry, isdir);
1251 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1254 entry->d_flags &= ~DCACHE_UNHASHED;
1255 hlist_add_head_rcu(&entry->d_hash, list);
1258 static void _d_rehash(struct dentry * entry)
1260 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1264 * d_rehash - add an entry back to the hash
1265 * @entry: dentry to add to the hash
1267 * Adds a dentry to the hash according to its name.
1270 void d_rehash(struct dentry * entry)
1272 spin_lock(&dcache_lock);
1273 spin_lock(&entry->d_lock);
1274 _d_rehash(entry);
1275 spin_unlock(&entry->d_lock);
1276 spin_unlock(&dcache_lock);
1279 #define do_switch(x,y) do { \
1280 __typeof__ (x) __tmp = x; \
1281 x = y; y = __tmp; } while (0)
1284 * When switching names, the actual string doesn't strictly have to
1285 * be preserved in the target - because we're dropping the target
1286 * anyway. As such, we can just do a simple memcpy() to copy over
1287 * the new name before we switch.
1289 * Note that we have to be a lot more careful about getting the hash
1290 * switched - we have to switch the hash value properly even if it
1291 * then no longer matches the actual (corrupted) string of the target.
1292 * The hash value has to match the hash queue that the dentry is on..
1294 static void switch_names(struct dentry *dentry, struct dentry *target)
1296 if (dname_external(target)) {
1297 if (dname_external(dentry)) {
1299 * Both external: swap the pointers
1301 do_switch(target->d_name.name, dentry->d_name.name);
1302 } else {
1304 * dentry:internal, target:external. Steal target's
1305 * storage and make target internal.
1307 dentry->d_name.name = target->d_name.name;
1308 target->d_name.name = target->d_iname;
1310 } else {
1311 if (dname_external(dentry)) {
1313 * dentry:external, target:internal. Give dentry's
1314 * storage to target and make dentry internal
1316 memcpy(dentry->d_iname, target->d_name.name,
1317 target->d_name.len + 1);
1318 target->d_name.name = dentry->d_name.name;
1319 dentry->d_name.name = dentry->d_iname;
1320 } else {
1322 * Both are internal. Just copy target to dentry
1324 memcpy(dentry->d_iname, target->d_name.name,
1325 target->d_name.len + 1);
1331 * We cannibalize "target" when moving dentry on top of it,
1332 * because it's going to be thrown away anyway. We could be more
1333 * polite about it, though.
1335 * This forceful removal will result in ugly /proc output if
1336 * somebody holds a file open that got deleted due to a rename.
1337 * We could be nicer about the deleted file, and let it show
1338 * up under the name it got deleted rather than the name that
1339 * deleted it.
1343 * d_move - move a dentry
1344 * @dentry: entry to move
1345 * @target: new dentry
1347 * Update the dcache to reflect the move of a file name. Negative
1348 * dcache entries should not be moved in this way.
1351 void d_move(struct dentry * dentry, struct dentry * target)
1353 struct hlist_head *list;
1355 if (!dentry->d_inode)
1356 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1358 spin_lock(&dcache_lock);
1359 write_seqlock(&rename_lock);
1361 * XXXX: do we really need to take target->d_lock?
1363 if (target < dentry) {
1364 spin_lock(&target->d_lock);
1365 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1366 } else {
1367 spin_lock(&dentry->d_lock);
1368 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1371 /* Move the dentry to the target hash queue, if on different bucket */
1372 if (dentry->d_flags & DCACHE_UNHASHED)
1373 goto already_unhashed;
1375 hlist_del_rcu(&dentry->d_hash);
1377 already_unhashed:
1378 list = d_hash(target->d_parent, target->d_name.hash);
1379 __d_rehash(dentry, list);
1381 /* Unhash the target: dput() will then get rid of it */
1382 __d_drop(target);
1384 list_del(&dentry->d_u.d_child);
1385 list_del(&target->d_u.d_child);
1387 /* Switch the names.. */
1388 switch_names(dentry, target);
1389 do_switch(dentry->d_name.len, target->d_name.len);
1390 do_switch(dentry->d_name.hash, target->d_name.hash);
1392 /* ... and switch the parents */
1393 if (IS_ROOT(dentry)) {
1394 dentry->d_parent = target->d_parent;
1395 target->d_parent = target;
1396 INIT_LIST_HEAD(&target->d_u.d_child);
1397 } else {
1398 do_switch(dentry->d_parent, target->d_parent);
1400 /* And add them back to the (new) parent lists */
1401 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1404 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1405 spin_unlock(&target->d_lock);
1406 fsnotify_d_move(dentry);
1407 spin_unlock(&dentry->d_lock);
1408 write_sequnlock(&rename_lock);
1409 spin_unlock(&dcache_lock);
1413 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1414 * named dentry in place of the dentry to be replaced.
1416 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1418 struct dentry *dparent, *aparent;
1420 switch_names(dentry, anon);
1421 do_switch(dentry->d_name.len, anon->d_name.len);
1422 do_switch(dentry->d_name.hash, anon->d_name.hash);
1424 dparent = dentry->d_parent;
1425 aparent = anon->d_parent;
1427 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1428 list_del(&dentry->d_u.d_child);
1429 if (!IS_ROOT(dentry))
1430 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1431 else
1432 INIT_LIST_HEAD(&dentry->d_u.d_child);
1434 anon->d_parent = (dparent == dentry) ? anon : dparent;
1435 list_del(&anon->d_u.d_child);
1436 if (!IS_ROOT(anon))
1437 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1438 else
1439 INIT_LIST_HEAD(&anon->d_u.d_child);
1441 anon->d_flags &= ~DCACHE_DISCONNECTED;
1445 * d_materialise_unique - introduce an inode into the tree
1446 * @dentry: candidate dentry
1447 * @inode: inode to bind to the dentry, to which aliases may be attached
1449 * Introduces an dentry into the tree, substituting an extant disconnected
1450 * root directory alias in its place if there is one
1452 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1454 struct dentry *alias, *actual;
1456 BUG_ON(!d_unhashed(dentry));
1458 spin_lock(&dcache_lock);
1460 if (!inode) {
1461 actual = dentry;
1462 dentry->d_inode = NULL;
1463 goto found_lock;
1466 /* See if a disconnected directory already exists as an anonymous root
1467 * that we should splice into the tree instead */
1468 if (S_ISDIR(inode->i_mode) && (alias = __d_find_alias(inode, 1))) {
1469 spin_lock(&alias->d_lock);
1471 /* Is this a mountpoint that we could splice into our tree? */
1472 if (IS_ROOT(alias))
1473 goto connect_mountpoint;
1475 if (alias->d_name.len == dentry->d_name.len &&
1476 alias->d_parent == dentry->d_parent &&
1477 memcmp(alias->d_name.name,
1478 dentry->d_name.name,
1479 dentry->d_name.len) == 0)
1480 goto replace_with_alias;
1482 spin_unlock(&alias->d_lock);
1484 /* Doh! Seem to be aliasing directories for some reason... */
1485 dput(alias);
1488 /* Add a unique reference */
1489 actual = __d_instantiate_unique(dentry, inode);
1490 if (!actual)
1491 actual = dentry;
1492 else if (unlikely(!d_unhashed(actual)))
1493 goto shouldnt_be_hashed;
1495 found_lock:
1496 spin_lock(&actual->d_lock);
1497 found:
1498 _d_rehash(actual);
1499 spin_unlock(&actual->d_lock);
1500 spin_unlock(&dcache_lock);
1502 if (actual == dentry) {
1503 security_d_instantiate(dentry, inode);
1504 return NULL;
1507 iput(inode);
1508 return actual;
1510 /* Convert the anonymous/root alias into an ordinary dentry */
1511 connect_mountpoint:
1512 __d_materialise_dentry(dentry, alias);
1514 /* Replace the candidate dentry with the alias in the tree */
1515 replace_with_alias:
1516 __d_drop(alias);
1517 actual = alias;
1518 goto found;
1520 shouldnt_be_hashed:
1521 spin_unlock(&dcache_lock);
1522 BUG();
1523 goto shouldnt_be_hashed;
1527 * d_path - return the path of a dentry
1528 * @dentry: dentry to report
1529 * @vfsmnt: vfsmnt to which the dentry belongs
1530 * @root: root dentry
1531 * @rootmnt: vfsmnt to which the root dentry belongs
1532 * @buffer: buffer to return value in
1533 * @buflen: buffer length
1535 * Convert a dentry into an ASCII path name. If the entry has been deleted
1536 * the string " (deleted)" is appended. Note that this is ambiguous.
1538 * Returns the buffer or an error code if the path was too long.
1540 * "buflen" should be positive. Caller holds the dcache_lock.
1542 static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1543 struct dentry *root, struct vfsmount *rootmnt,
1544 char *buffer, int buflen)
1546 char * end = buffer+buflen;
1547 char * retval;
1548 int namelen;
1550 *--end = '\0';
1551 buflen--;
1552 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1553 buflen -= 10;
1554 end -= 10;
1555 if (buflen < 0)
1556 goto Elong;
1557 memcpy(end, " (deleted)", 10);
1560 if (buflen < 1)
1561 goto Elong;
1562 /* Get '/' right */
1563 retval = end-1;
1564 *retval = '/';
1566 for (;;) {
1567 struct dentry * parent;
1569 if (dentry == root && vfsmnt == rootmnt)
1570 break;
1571 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1572 /* Global root? */
1573 spin_lock(&vfsmount_lock);
1574 if (vfsmnt->mnt_parent == vfsmnt) {
1575 spin_unlock(&vfsmount_lock);
1576 goto global_root;
1578 dentry = vfsmnt->mnt_mountpoint;
1579 vfsmnt = vfsmnt->mnt_parent;
1580 spin_unlock(&vfsmount_lock);
1581 continue;
1583 parent = dentry->d_parent;
1584 prefetch(parent);
1585 namelen = dentry->d_name.len;
1586 buflen -= namelen + 1;
1587 if (buflen < 0)
1588 goto Elong;
1589 end -= namelen;
1590 memcpy(end, dentry->d_name.name, namelen);
1591 *--end = '/';
1592 retval = end;
1593 dentry = parent;
1596 return retval;
1598 global_root:
1599 namelen = dentry->d_name.len;
1600 buflen -= namelen;
1601 if (buflen < 0)
1602 goto Elong;
1603 retval -= namelen-1; /* hit the slash */
1604 memcpy(retval, dentry->d_name.name, namelen);
1605 return retval;
1606 Elong:
1607 return ERR_PTR(-ENAMETOOLONG);
1610 /* write full pathname into buffer and return start of pathname */
1611 char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1612 char *buf, int buflen)
1614 char *res;
1615 struct vfsmount *rootmnt;
1616 struct dentry *root;
1618 read_lock(&current->fs->lock);
1619 rootmnt = mntget(current->fs->rootmnt);
1620 root = dget(current->fs->root);
1621 read_unlock(&current->fs->lock);
1622 spin_lock(&dcache_lock);
1623 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1624 spin_unlock(&dcache_lock);
1625 dput(root);
1626 mntput(rootmnt);
1627 return res;
1631 * NOTE! The user-level library version returns a
1632 * character pointer. The kernel system call just
1633 * returns the length of the buffer filled (which
1634 * includes the ending '\0' character), or a negative
1635 * error value. So libc would do something like
1637 * char *getcwd(char * buf, size_t size)
1639 * int retval;
1641 * retval = sys_getcwd(buf, size);
1642 * if (retval >= 0)
1643 * return buf;
1644 * errno = -retval;
1645 * return NULL;
1648 asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1650 int error;
1651 struct vfsmount *pwdmnt, *rootmnt;
1652 struct dentry *pwd, *root;
1653 char *page = (char *) __get_free_page(GFP_USER);
1655 if (!page)
1656 return -ENOMEM;
1658 read_lock(&current->fs->lock);
1659 pwdmnt = mntget(current->fs->pwdmnt);
1660 pwd = dget(current->fs->pwd);
1661 rootmnt = mntget(current->fs->rootmnt);
1662 root = dget(current->fs->root);
1663 read_unlock(&current->fs->lock);
1665 error = -ENOENT;
1666 /* Has the current directory has been unlinked? */
1667 spin_lock(&dcache_lock);
1668 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1669 unsigned long len;
1670 char * cwd;
1672 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1673 spin_unlock(&dcache_lock);
1675 error = PTR_ERR(cwd);
1676 if (IS_ERR(cwd))
1677 goto out;
1679 error = -ERANGE;
1680 len = PAGE_SIZE + page - cwd;
1681 if (len <= size) {
1682 error = len;
1683 if (copy_to_user(buf, cwd, len))
1684 error = -EFAULT;
1686 } else
1687 spin_unlock(&dcache_lock);
1689 out:
1690 dput(pwd);
1691 mntput(pwdmnt);
1692 dput(root);
1693 mntput(rootmnt);
1694 free_page((unsigned long) page);
1695 return error;
1699 * Test whether new_dentry is a subdirectory of old_dentry.
1701 * Trivially implemented using the dcache structure
1705 * is_subdir - is new dentry a subdirectory of old_dentry
1706 * @new_dentry: new dentry
1707 * @old_dentry: old dentry
1709 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1710 * Returns 0 otherwise.
1711 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1714 int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1716 int result;
1717 struct dentry * saved = new_dentry;
1718 unsigned long seq;
1720 /* need rcu_readlock to protect against the d_parent trashing due to
1721 * d_move
1723 rcu_read_lock();
1724 do {
1725 /* for restarting inner loop in case of seq retry */
1726 new_dentry = saved;
1727 result = 0;
1728 seq = read_seqbegin(&rename_lock);
1729 for (;;) {
1730 if (new_dentry != old_dentry) {
1731 struct dentry * parent = new_dentry->d_parent;
1732 if (parent == new_dentry)
1733 break;
1734 new_dentry = parent;
1735 continue;
1737 result = 1;
1738 break;
1740 } while (read_seqretry(&rename_lock, seq));
1741 rcu_read_unlock();
1743 return result;
1746 void d_genocide(struct dentry *root)
1748 struct dentry *this_parent = root;
1749 struct list_head *next;
1751 spin_lock(&dcache_lock);
1752 repeat:
1753 next = this_parent->d_subdirs.next;
1754 resume:
1755 while (next != &this_parent->d_subdirs) {
1756 struct list_head *tmp = next;
1757 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1758 next = tmp->next;
1759 if (d_unhashed(dentry)||!dentry->d_inode)
1760 continue;
1761 if (!list_empty(&dentry->d_subdirs)) {
1762 this_parent = dentry;
1763 goto repeat;
1765 atomic_dec(&dentry->d_count);
1767 if (this_parent != root) {
1768 next = this_parent->d_u.d_child.next;
1769 atomic_dec(&this_parent->d_count);
1770 this_parent = this_parent->d_parent;
1771 goto resume;
1773 spin_unlock(&dcache_lock);
1777 * find_inode_number - check for dentry with name
1778 * @dir: directory to check
1779 * @name: Name to find.
1781 * Check whether a dentry already exists for the given name,
1782 * and return the inode number if it has an inode. Otherwise
1783 * 0 is returned.
1785 * This routine is used to post-process directory listings for
1786 * filesystems using synthetic inode numbers, and is necessary
1787 * to keep getcwd() working.
1790 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1792 struct dentry * dentry;
1793 ino_t ino = 0;
1795 dentry = d_hash_and_lookup(dir, name);
1796 if (dentry) {
1797 if (dentry->d_inode)
1798 ino = dentry->d_inode->i_ino;
1799 dput(dentry);
1801 return ino;
1804 static __initdata unsigned long dhash_entries;
1805 static int __init set_dhash_entries(char *str)
1807 if (!str)
1808 return 0;
1809 dhash_entries = simple_strtoul(str, &str, 0);
1810 return 1;
1812 __setup("dhash_entries=", set_dhash_entries);
1814 static void __init dcache_init_early(void)
1816 int loop;
1818 /* If hashes are distributed across NUMA nodes, defer
1819 * hash allocation until vmalloc space is available.
1821 if (hashdist)
1822 return;
1824 dentry_hashtable =
1825 alloc_large_system_hash("Dentry cache",
1826 sizeof(struct hlist_head),
1827 dhash_entries,
1829 HASH_EARLY,
1830 &d_hash_shift,
1831 &d_hash_mask,
1834 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1835 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1838 static void __init dcache_init(unsigned long mempages)
1840 int loop;
1843 * A constructor could be added for stable state like the lists,
1844 * but it is probably not worth it because of the cache nature
1845 * of the dcache.
1847 dentry_cache = kmem_cache_create("dentry_cache",
1848 sizeof(struct dentry),
1850 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1851 SLAB_MEM_SPREAD),
1852 NULL, NULL);
1854 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1856 /* Hash may have been set up in dcache_init_early */
1857 if (!hashdist)
1858 return;
1860 dentry_hashtable =
1861 alloc_large_system_hash("Dentry cache",
1862 sizeof(struct hlist_head),
1863 dhash_entries,
1866 &d_hash_shift,
1867 &d_hash_mask,
1870 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1871 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1874 /* SLAB cache for __getname() consumers */
1875 kmem_cache_t *names_cachep __read_mostly;
1877 /* SLAB cache for file structures */
1878 kmem_cache_t *filp_cachep __read_mostly;
1880 EXPORT_SYMBOL(d_genocide);
1882 void __init vfs_caches_init_early(void)
1884 dcache_init_early();
1885 inode_init_early();
1888 void __init vfs_caches_init(unsigned long mempages)
1890 unsigned long reserve;
1892 /* Base hash sizes on available memory, with a reserve equal to
1893 150% of current kernel size */
1895 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1896 mempages -= reserve;
1898 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1899 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1901 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1902 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1904 dcache_init(mempages);
1905 inode_init(mempages);
1906 files_init(mempages);
1907 mnt_init(mempages);
1908 bdev_cache_init();
1909 chrdev_init();
1912 EXPORT_SYMBOL(d_alloc);
1913 EXPORT_SYMBOL(d_alloc_anon);
1914 EXPORT_SYMBOL(d_alloc_root);
1915 EXPORT_SYMBOL(d_delete);
1916 EXPORT_SYMBOL(d_find_alias);
1917 EXPORT_SYMBOL(d_instantiate);
1918 EXPORT_SYMBOL(d_invalidate);
1919 EXPORT_SYMBOL(d_lookup);
1920 EXPORT_SYMBOL(d_move);
1921 EXPORT_SYMBOL_GPL(d_materialise_unique);
1922 EXPORT_SYMBOL(d_path);
1923 EXPORT_SYMBOL(d_prune_aliases);
1924 EXPORT_SYMBOL(d_rehash);
1925 EXPORT_SYMBOL(d_splice_alias);
1926 EXPORT_SYMBOL(d_validate);
1927 EXPORT_SYMBOL(dget_locked);
1928 EXPORT_SYMBOL(dput);
1929 EXPORT_SYMBOL(find_inode_number);
1930 EXPORT_SYMBOL(have_submounts);
1931 EXPORT_SYMBOL(names_cachep);
1932 EXPORT_SYMBOL(shrink_dcache_parent);
1933 EXPORT_SYMBOL(shrink_dcache_sb);