blob: fb10386c59bed6b34a51af4fb930eb79c86ac586 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
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.
15 */
16
17#include <linux/config.h>
18#include <linux/syscalls.h>
19#include <linux/string.h>
20#include <linux/mm.h>
21#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040022#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/slab.h>
24#include <linux/init.h>
25#include <linux/smp_lock.h>
26#include <linux/hash.h>
27#include <linux/cache.h>
28#include <linux/module.h>
29#include <linux/mount.h>
30#include <linux/file.h>
31#include <asm/uaccess.h>
32#include <linux/security.h>
33#include <linux/seqlock.h>
34#include <linux/swap.h>
35#include <linux/bootmem.h>
36
37/* #define DCACHE_DEBUG 1 */
38
39int sysctl_vfs_cache_pressure = 100;
40EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
41
42 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
Adrian Bunk75c96f82005-05-05 16:16:09 -070043static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45EXPORT_SYMBOL(dcache_lock);
46
47static kmem_cache_t *dentry_cache;
48
49#define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
50
51/*
52 * This is the single most critical data structure when it comes
53 * to the dcache: the hashtable for lookups. Somebody should try
54 * to make this good - I've just made it work.
55 *
56 * This hash-function tries to avoid losing too many bits of hash
57 * information, yet avoid using a prime hash-size or similar.
58 */
59#define D_HASHBITS d_hash_shift
60#define D_HASHMASK d_hash_mask
61
62static unsigned int d_hash_mask;
63static unsigned int d_hash_shift;
64static struct hlist_head *dentry_hashtable;
65static LIST_HEAD(dentry_unused);
66
67/* Statistics gathering. */
68struct dentry_stat_t dentry_stat = {
69 .age_limit = 45,
70};
71
72static void d_callback(struct rcu_head *head)
73{
74 struct dentry * dentry = container_of(head, struct dentry, d_rcu);
75
76 if (dname_external(dentry))
77 kfree(dentry->d_name.name);
78 kmem_cache_free(dentry_cache, dentry);
79}
80
81/*
82 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
83 * inside dcache_lock.
84 */
85static void d_free(struct dentry *dentry)
86{
87 if (dentry->d_op && dentry->d_op->d_release)
88 dentry->d_op->d_release(dentry);
89 call_rcu(&dentry->d_rcu, d_callback);
90}
91
92/*
93 * Release the dentry's inode, using the filesystem
94 * d_iput() operation if defined.
95 * Called with dcache_lock and per dentry lock held, drops both.
96 */
97static inline void dentry_iput(struct dentry * dentry)
98{
99 struct inode *inode = dentry->d_inode;
100 if (inode) {
101 dentry->d_inode = NULL;
102 list_del_init(&dentry->d_alias);
103 spin_unlock(&dentry->d_lock);
104 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700105 if (!inode->i_nlink)
106 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 if (dentry->d_op && dentry->d_op->d_iput)
108 dentry->d_op->d_iput(dentry, inode);
109 else
110 iput(inode);
111 } else {
112 spin_unlock(&dentry->d_lock);
113 spin_unlock(&dcache_lock);
114 }
115}
116
117/*
118 * This is dput
119 *
120 * This is complicated by the fact that we do not want to put
121 * dentries that are no longer on any hash chain on the unused
122 * list: we'd much rather just get rid of them immediately.
123 *
124 * However, that implies that we have to traverse the dentry
125 * tree upwards to the parents which might _also_ now be
126 * scheduled for deletion (it may have been only waiting for
127 * its last child to go away).
128 *
129 * This tail recursion is done by hand as we don't want to depend
130 * on the compiler to always get this right (gcc generally doesn't).
131 * Real recursion would eat up our stack space.
132 */
133
134/*
135 * dput - release a dentry
136 * @dentry: dentry to release
137 *
138 * Release a dentry. This will drop the usage count and if appropriate
139 * call the dentry unlink method as well as removing it from the queues and
140 * releasing its resources. If the parent dentries were scheduled for release
141 * they too may now get deleted.
142 *
143 * no dcache lock, please.
144 */
145
146void dput(struct dentry *dentry)
147{
148 if (!dentry)
149 return;
150
151repeat:
152 if (atomic_read(&dentry->d_count) == 1)
153 might_sleep();
154 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
155 return;
156
157 spin_lock(&dentry->d_lock);
158 if (atomic_read(&dentry->d_count)) {
159 spin_unlock(&dentry->d_lock);
160 spin_unlock(&dcache_lock);
161 return;
162 }
163
164 /*
165 * AV: ->d_delete() is _NOT_ allowed to block now.
166 */
167 if (dentry->d_op && dentry->d_op->d_delete) {
168 if (dentry->d_op->d_delete(dentry))
169 goto unhash_it;
170 }
171 /* Unreachable? Get rid of it */
172 if (d_unhashed(dentry))
173 goto kill_it;
174 if (list_empty(&dentry->d_lru)) {
175 dentry->d_flags |= DCACHE_REFERENCED;
176 list_add(&dentry->d_lru, &dentry_unused);
177 dentry_stat.nr_unused++;
178 }
179 spin_unlock(&dentry->d_lock);
180 spin_unlock(&dcache_lock);
181 return;
182
183unhash_it:
184 __d_drop(dentry);
185
186kill_it: {
187 struct dentry *parent;
188
189 /* If dentry was on d_lru list
190 * delete it from there
191 */
192 if (!list_empty(&dentry->d_lru)) {
193 list_del(&dentry->d_lru);
194 dentry_stat.nr_unused--;
195 }
196 list_del(&dentry->d_child);
197 dentry_stat.nr_dentry--; /* For d_free, below */
198 /*drops the locks, at that point nobody can reach this dentry */
199 dentry_iput(dentry);
200 parent = dentry->d_parent;
201 d_free(dentry);
202 if (dentry == parent)
203 return;
204 dentry = parent;
205 goto repeat;
206 }
207}
208
209/**
210 * d_invalidate - invalidate a dentry
211 * @dentry: dentry to invalidate
212 *
213 * Try to invalidate the dentry if it turns out to be
214 * possible. If there are other dentries that can be
215 * reached through this one we can't delete it and we
216 * return -EBUSY. On success we return 0.
217 *
218 * no dcache lock.
219 */
220
221int d_invalidate(struct dentry * dentry)
222{
223 /*
224 * If it's already been dropped, return OK.
225 */
226 spin_lock(&dcache_lock);
227 if (d_unhashed(dentry)) {
228 spin_unlock(&dcache_lock);
229 return 0;
230 }
231 /*
232 * Check whether to do a partial shrink_dcache
233 * to get rid of unused child entries.
234 */
235 if (!list_empty(&dentry->d_subdirs)) {
236 spin_unlock(&dcache_lock);
237 shrink_dcache_parent(dentry);
238 spin_lock(&dcache_lock);
239 }
240
241 /*
242 * Somebody else still using it?
243 *
244 * If it's a directory, we can't drop it
245 * for fear of somebody re-populating it
246 * with children (even though dropping it
247 * would make it unreachable from the root,
248 * we might still populate it if it was a
249 * working directory or similar).
250 */
251 spin_lock(&dentry->d_lock);
252 if (atomic_read(&dentry->d_count) > 1) {
253 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
254 spin_unlock(&dentry->d_lock);
255 spin_unlock(&dcache_lock);
256 return -EBUSY;
257 }
258 }
259
260 __d_drop(dentry);
261 spin_unlock(&dentry->d_lock);
262 spin_unlock(&dcache_lock);
263 return 0;
264}
265
266/* This should be called _only_ with dcache_lock held */
267
268static inline struct dentry * __dget_locked(struct dentry *dentry)
269{
270 atomic_inc(&dentry->d_count);
271 if (!list_empty(&dentry->d_lru)) {
272 dentry_stat.nr_unused--;
273 list_del_init(&dentry->d_lru);
274 }
275 return dentry;
276}
277
278struct dentry * dget_locked(struct dentry *dentry)
279{
280 return __dget_locked(dentry);
281}
282
283/**
284 * d_find_alias - grab a hashed alias of inode
285 * @inode: inode in question
286 * @want_discon: flag, used by d_splice_alias, to request
287 * that only a DISCONNECTED alias be returned.
288 *
289 * If inode has a hashed alias, or is a directory and has any alias,
290 * acquire the reference to alias and return it. Otherwise return NULL.
291 * Notice that if inode is a directory there can be only one alias and
292 * it can be unhashed only if it has no children, or if it is the root
293 * of a filesystem.
294 *
295 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
296 * any other hashed alias over that one unless @want_discon is set,
297 * in which case only return a DCACHE_DISCONNECTED alias.
298 */
299
300static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
301{
302 struct list_head *head, *next, *tmp;
303 struct dentry *alias, *discon_alias=NULL;
304
305 head = &inode->i_dentry;
306 next = inode->i_dentry.next;
307 while (next != head) {
308 tmp = next;
309 next = tmp->next;
310 prefetch(next);
311 alias = list_entry(tmp, struct dentry, d_alias);
312 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
313 if (alias->d_flags & DCACHE_DISCONNECTED)
314 discon_alias = alias;
315 else if (!want_discon) {
316 __dget_locked(alias);
317 return alias;
318 }
319 }
320 }
321 if (discon_alias)
322 __dget_locked(discon_alias);
323 return discon_alias;
324}
325
326struct dentry * d_find_alias(struct inode *inode)
327{
328 struct dentry *de;
329 spin_lock(&dcache_lock);
330 de = __d_find_alias(inode, 0);
331 spin_unlock(&dcache_lock);
332 return de;
333}
334
335/*
336 * Try to kill dentries associated with this inode.
337 * WARNING: you must own a reference to inode.
338 */
339void d_prune_aliases(struct inode *inode)
340{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700341 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342restart:
343 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700344 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 spin_lock(&dentry->d_lock);
346 if (!atomic_read(&dentry->d_count)) {
347 __dget_locked(dentry);
348 __d_drop(dentry);
349 spin_unlock(&dentry->d_lock);
350 spin_unlock(&dcache_lock);
351 dput(dentry);
352 goto restart;
353 }
354 spin_unlock(&dentry->d_lock);
355 }
356 spin_unlock(&dcache_lock);
357}
358
359/*
360 * Throw away a dentry - free the inode, dput the parent.
361 * This requires that the LRU list has already been
362 * removed.
363 * Called with dcache_lock, drops it and then regains.
364 */
365static inline void prune_one_dentry(struct dentry * dentry)
366{
367 struct dentry * parent;
368
369 __d_drop(dentry);
370 list_del(&dentry->d_child);
371 dentry_stat.nr_dentry--; /* For d_free, below */
372 dentry_iput(dentry);
373 parent = dentry->d_parent;
374 d_free(dentry);
375 if (parent != dentry)
376 dput(parent);
377 spin_lock(&dcache_lock);
378}
379
380/**
381 * prune_dcache - shrink the dcache
382 * @count: number of entries to try and free
383 *
384 * Shrink the dcache. This is done when we need
385 * more memory, or simply when we need to unmount
386 * something (at which point we need to unuse
387 * all dentries).
388 *
389 * This function may fail to free any resources if
390 * all the dentries are in use.
391 */
392
393static void prune_dcache(int count)
394{
395 spin_lock(&dcache_lock);
396 for (; count ; count--) {
397 struct dentry *dentry;
398 struct list_head *tmp;
399
400 cond_resched_lock(&dcache_lock);
401
402 tmp = dentry_unused.prev;
403 if (tmp == &dentry_unused)
404 break;
405 list_del_init(tmp);
406 prefetch(dentry_unused.prev);
407 dentry_stat.nr_unused--;
408 dentry = list_entry(tmp, struct dentry, d_lru);
409
410 spin_lock(&dentry->d_lock);
411 /*
412 * We found an inuse dentry which was not removed from
413 * dentry_unused because of laziness during lookup. Do not free
414 * it - just keep it off the dentry_unused list.
415 */
416 if (atomic_read(&dentry->d_count)) {
417 spin_unlock(&dentry->d_lock);
418 continue;
419 }
420 /* If the dentry was recently referenced, don't free it. */
421 if (dentry->d_flags & DCACHE_REFERENCED) {
422 dentry->d_flags &= ~DCACHE_REFERENCED;
423 list_add(&dentry->d_lru, &dentry_unused);
424 dentry_stat.nr_unused++;
425 spin_unlock(&dentry->d_lock);
426 continue;
427 }
428 prune_one_dentry(dentry);
429 }
430 spin_unlock(&dcache_lock);
431}
432
433/*
434 * Shrink the dcache for the specified super block.
435 * This allows us to unmount a device without disturbing
436 * the dcache for the other devices.
437 *
438 * This implementation makes just two traversals of the
439 * unused list. On the first pass we move the selected
440 * dentries to the most recent end, and on the second
441 * pass we free them. The second pass must restart after
442 * each dput(), but since the target dentries are all at
443 * the end, it's really just a single traversal.
444 */
445
446/**
447 * shrink_dcache_sb - shrink dcache for a superblock
448 * @sb: superblock
449 *
450 * Shrink the dcache for the specified super block. This
451 * is used to free the dcache before unmounting a file
452 * system
453 */
454
455void shrink_dcache_sb(struct super_block * sb)
456{
457 struct list_head *tmp, *next;
458 struct dentry *dentry;
459
460 /*
461 * Pass one ... move the dentries for the specified
462 * superblock to the most recent end of the unused list.
463 */
464 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700465 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 dentry = list_entry(tmp, struct dentry, d_lru);
467 if (dentry->d_sb != sb)
468 continue;
469 list_del(tmp);
470 list_add(tmp, &dentry_unused);
471 }
472
473 /*
474 * Pass two ... free the dentries for this superblock.
475 */
476repeat:
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700477 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 dentry = list_entry(tmp, struct dentry, d_lru);
479 if (dentry->d_sb != sb)
480 continue;
481 dentry_stat.nr_unused--;
482 list_del_init(tmp);
483 spin_lock(&dentry->d_lock);
484 if (atomic_read(&dentry->d_count)) {
485 spin_unlock(&dentry->d_lock);
486 continue;
487 }
488 prune_one_dentry(dentry);
489 goto repeat;
490 }
491 spin_unlock(&dcache_lock);
492}
493
494/*
495 * Search for at least 1 mount point in the dentry's subdirs.
496 * We descend to the next level whenever the d_subdirs
497 * list is non-empty and continue searching.
498 */
499
500/**
501 * have_submounts - check for mounts over a dentry
502 * @parent: dentry to check.
503 *
504 * Return true if the parent or its subdirectories contain
505 * a mount point
506 */
507
508int have_submounts(struct dentry *parent)
509{
510 struct dentry *this_parent = parent;
511 struct list_head *next;
512
513 spin_lock(&dcache_lock);
514 if (d_mountpoint(parent))
515 goto positive;
516repeat:
517 next = this_parent->d_subdirs.next;
518resume:
519 while (next != &this_parent->d_subdirs) {
520 struct list_head *tmp = next;
521 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
522 next = tmp->next;
523 /* Have we found a mount point ? */
524 if (d_mountpoint(dentry))
525 goto positive;
526 if (!list_empty(&dentry->d_subdirs)) {
527 this_parent = dentry;
528 goto repeat;
529 }
530 }
531 /*
532 * All done at this level ... ascend and resume the search.
533 */
534 if (this_parent != parent) {
535 next = this_parent->d_child.next;
536 this_parent = this_parent->d_parent;
537 goto resume;
538 }
539 spin_unlock(&dcache_lock);
540 return 0; /* No mount points found in tree */
541positive:
542 spin_unlock(&dcache_lock);
543 return 1;
544}
545
546/*
547 * Search the dentry child list for the specified parent,
548 * and move any unused dentries to the end of the unused
549 * list for prune_dcache(). We descend to the next level
550 * whenever the d_subdirs list is non-empty and continue
551 * searching.
552 *
553 * It returns zero iff there are no unused children,
554 * otherwise it returns the number of children moved to
555 * the end of the unused list. This may not be the total
556 * number of unused children, because select_parent can
557 * drop the lock and return early due to latency
558 * constraints.
559 */
560static int select_parent(struct dentry * parent)
561{
562 struct dentry *this_parent = parent;
563 struct list_head *next;
564 int found = 0;
565
566 spin_lock(&dcache_lock);
567repeat:
568 next = this_parent->d_subdirs.next;
569resume:
570 while (next != &this_parent->d_subdirs) {
571 struct list_head *tmp = next;
572 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
573 next = tmp->next;
574
575 if (!list_empty(&dentry->d_lru)) {
576 dentry_stat.nr_unused--;
577 list_del_init(&dentry->d_lru);
578 }
579 /*
580 * move only zero ref count dentries to the end
581 * of the unused list for prune_dcache
582 */
583 if (!atomic_read(&dentry->d_count)) {
584 list_add(&dentry->d_lru, dentry_unused.prev);
585 dentry_stat.nr_unused++;
586 found++;
587 }
588
589 /*
590 * We can return to the caller if we have found some (this
591 * ensures forward progress). We'll be coming back to find
592 * the rest.
593 */
594 if (found && need_resched())
595 goto out;
596
597 /*
598 * Descend a level if the d_subdirs list is non-empty.
599 */
600 if (!list_empty(&dentry->d_subdirs)) {
601 this_parent = dentry;
602#ifdef DCACHE_DEBUG
603printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n",
604dentry->d_parent->d_name.name, dentry->d_name.name, found);
605#endif
606 goto repeat;
607 }
608 }
609 /*
610 * All done at this level ... ascend and resume the search.
611 */
612 if (this_parent != parent) {
613 next = this_parent->d_child.next;
614 this_parent = this_parent->d_parent;
615#ifdef DCACHE_DEBUG
616printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n",
617this_parent->d_parent->d_name.name, this_parent->d_name.name, found);
618#endif
619 goto resume;
620 }
621out:
622 spin_unlock(&dcache_lock);
623 return found;
624}
625
626/**
627 * shrink_dcache_parent - prune dcache
628 * @parent: parent of entries to prune
629 *
630 * Prune the dcache to remove unused children of the parent dentry.
631 */
632
633void shrink_dcache_parent(struct dentry * parent)
634{
635 int found;
636
637 while ((found = select_parent(parent)) != 0)
638 prune_dcache(found);
639}
640
641/**
642 * shrink_dcache_anon - further prune the cache
643 * @head: head of d_hash list of dentries to prune
644 *
645 * Prune the dentries that are anonymous
646 *
647 * parsing d_hash list does not hlist_for_each_rcu() as it
648 * done under dcache_lock.
649 *
650 */
651void shrink_dcache_anon(struct hlist_head *head)
652{
653 struct hlist_node *lp;
654 int found;
655 do {
656 found = 0;
657 spin_lock(&dcache_lock);
658 hlist_for_each(lp, head) {
659 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
660 if (!list_empty(&this->d_lru)) {
661 dentry_stat.nr_unused--;
662 list_del_init(&this->d_lru);
663 }
664
665 /*
666 * move only zero ref count dentries to the end
667 * of the unused list for prune_dcache
668 */
669 if (!atomic_read(&this->d_count)) {
670 list_add_tail(&this->d_lru, &dentry_unused);
671 dentry_stat.nr_unused++;
672 found++;
673 }
674 }
675 spin_unlock(&dcache_lock);
676 prune_dcache(found);
677 } while(found);
678}
679
680/*
681 * Scan `nr' dentries and return the number which remain.
682 *
683 * We need to avoid reentering the filesystem if the caller is performing a
684 * GFP_NOFS allocation attempt. One example deadlock is:
685 *
686 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
687 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
688 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
689 *
690 * In this case we return -1 to tell the caller that we baled.
691 */
692static int shrink_dcache_memory(int nr, unsigned int gfp_mask)
693{
694 if (nr) {
695 if (!(gfp_mask & __GFP_FS))
696 return -1;
697 prune_dcache(nr);
698 }
699 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
700}
701
702/**
703 * d_alloc - allocate a dcache entry
704 * @parent: parent of entry to allocate
705 * @name: qstr of the name
706 *
707 * Allocates a dentry. It returns %NULL if there is insufficient memory
708 * available. On a success the dentry is returned. The name passed in is
709 * copied and the copy passed in may be reused after this call.
710 */
711
712struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
713{
714 struct dentry *dentry;
715 char *dname;
716
717 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
718 if (!dentry)
719 return NULL;
720
721 if (name->len > DNAME_INLINE_LEN-1) {
722 dname = kmalloc(name->len + 1, GFP_KERNEL);
723 if (!dname) {
724 kmem_cache_free(dentry_cache, dentry);
725 return NULL;
726 }
727 } else {
728 dname = dentry->d_iname;
729 }
730 dentry->d_name.name = dname;
731
732 dentry->d_name.len = name->len;
733 dentry->d_name.hash = name->hash;
734 memcpy(dname, name->name, name->len);
735 dname[name->len] = 0;
736
737 atomic_set(&dentry->d_count, 1);
738 dentry->d_flags = DCACHE_UNHASHED;
739 spin_lock_init(&dentry->d_lock);
740 dentry->d_inode = NULL;
741 dentry->d_parent = NULL;
742 dentry->d_sb = NULL;
743 dentry->d_op = NULL;
744 dentry->d_fsdata = NULL;
745 dentry->d_mounted = 0;
746 dentry->d_cookie = NULL;
747 INIT_HLIST_NODE(&dentry->d_hash);
748 INIT_LIST_HEAD(&dentry->d_lru);
749 INIT_LIST_HEAD(&dentry->d_subdirs);
750 INIT_LIST_HEAD(&dentry->d_alias);
751
752 if (parent) {
753 dentry->d_parent = dget(parent);
754 dentry->d_sb = parent->d_sb;
755 } else {
756 INIT_LIST_HEAD(&dentry->d_child);
757 }
758
759 spin_lock(&dcache_lock);
760 if (parent)
761 list_add(&dentry->d_child, &parent->d_subdirs);
762 dentry_stat.nr_dentry++;
763 spin_unlock(&dcache_lock);
764
765 return dentry;
766}
767
768struct dentry *d_alloc_name(struct dentry *parent, const char *name)
769{
770 struct qstr q;
771
772 q.name = name;
773 q.len = strlen(name);
774 q.hash = full_name_hash(q.name, q.len);
775 return d_alloc(parent, &q);
776}
777
778/**
779 * d_instantiate - fill in inode information for a dentry
780 * @entry: dentry to complete
781 * @inode: inode to attach to this dentry
782 *
783 * Fill in inode information in the entry.
784 *
785 * This turns negative dentries into productive full members
786 * of society.
787 *
788 * NOTE! This assumes that the inode count has been incremented
789 * (or otherwise set) by the caller to indicate that it is now
790 * in use by the dcache.
791 */
792
793void d_instantiate(struct dentry *entry, struct inode * inode)
794{
795 if (!list_empty(&entry->d_alias)) BUG();
796 spin_lock(&dcache_lock);
797 if (inode)
798 list_add(&entry->d_alias, &inode->i_dentry);
799 entry->d_inode = inode;
800 spin_unlock(&dcache_lock);
801 security_d_instantiate(entry, inode);
802}
803
804/**
805 * d_instantiate_unique - instantiate a non-aliased dentry
806 * @entry: dentry to instantiate
807 * @inode: inode to attach to this dentry
808 *
809 * Fill in inode information in the entry. On success, it returns NULL.
810 * If an unhashed alias of "entry" already exists, then we return the
811 * aliased dentry instead.
812 *
813 * Note that in order to avoid conflicts with rename() etc, the caller
814 * had better be holding the parent directory semaphore.
815 */
816struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
817{
818 struct dentry *alias;
819 int len = entry->d_name.len;
820 const char *name = entry->d_name.name;
821 unsigned int hash = entry->d_name.hash;
822
823 BUG_ON(!list_empty(&entry->d_alias));
824 spin_lock(&dcache_lock);
825 if (!inode)
826 goto do_negative;
827 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
828 struct qstr *qstr = &alias->d_name;
829
830 if (qstr->hash != hash)
831 continue;
832 if (alias->d_parent != entry->d_parent)
833 continue;
834 if (qstr->len != len)
835 continue;
836 if (memcmp(qstr->name, name, len))
837 continue;
838 dget_locked(alias);
839 spin_unlock(&dcache_lock);
840 BUG_ON(!d_unhashed(alias));
841 return alias;
842 }
843 list_add(&entry->d_alias, &inode->i_dentry);
844do_negative:
845 entry->d_inode = inode;
846 spin_unlock(&dcache_lock);
847 security_d_instantiate(entry, inode);
848 return NULL;
849}
850EXPORT_SYMBOL(d_instantiate_unique);
851
852/**
853 * d_alloc_root - allocate root dentry
854 * @root_inode: inode to allocate the root for
855 *
856 * Allocate a root ("/") dentry for the inode given. The inode is
857 * instantiated and returned. %NULL is returned if there is insufficient
858 * memory or the inode passed is %NULL.
859 */
860
861struct dentry * d_alloc_root(struct inode * root_inode)
862{
863 struct dentry *res = NULL;
864
865 if (root_inode) {
866 static const struct qstr name = { .name = "/", .len = 1 };
867
868 res = d_alloc(NULL, &name);
869 if (res) {
870 res->d_sb = root_inode->i_sb;
871 res->d_parent = res;
872 d_instantiate(res, root_inode);
873 }
874 }
875 return res;
876}
877
878static inline struct hlist_head *d_hash(struct dentry *parent,
879 unsigned long hash)
880{
881 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
882 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
883 return dentry_hashtable + (hash & D_HASHMASK);
884}
885
886/**
887 * d_alloc_anon - allocate an anonymous dentry
888 * @inode: inode to allocate the dentry for
889 *
890 * This is similar to d_alloc_root. It is used by filesystems when
891 * creating a dentry for a given inode, often in the process of
892 * mapping a filehandle to a dentry. The returned dentry may be
893 * anonymous, or may have a full name (if the inode was already
894 * in the cache). The file system may need to make further
895 * efforts to connect this dentry into the dcache properly.
896 *
897 * When called on a directory inode, we must ensure that
898 * the inode only ever has one dentry. If a dentry is
899 * found, that is returned instead of allocating a new one.
900 *
901 * On successful return, the reference to the inode has been transferred
902 * to the dentry. If %NULL is returned (indicating kmalloc failure),
903 * the reference on the inode has not been released.
904 */
905
906struct dentry * d_alloc_anon(struct inode *inode)
907{
908 static const struct qstr anonstring = { .name = "" };
909 struct dentry *tmp;
910 struct dentry *res;
911
912 if ((res = d_find_alias(inode))) {
913 iput(inode);
914 return res;
915 }
916
917 tmp = d_alloc(NULL, &anonstring);
918 if (!tmp)
919 return NULL;
920
921 tmp->d_parent = tmp; /* make sure dput doesn't croak */
922
923 spin_lock(&dcache_lock);
924 res = __d_find_alias(inode, 0);
925 if (!res) {
926 /* attach a disconnected dentry */
927 res = tmp;
928 tmp = NULL;
929 spin_lock(&res->d_lock);
930 res->d_sb = inode->i_sb;
931 res->d_parent = res;
932 res->d_inode = inode;
933 res->d_flags |= DCACHE_DISCONNECTED;
934 res->d_flags &= ~DCACHE_UNHASHED;
935 list_add(&res->d_alias, &inode->i_dentry);
936 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
937 spin_unlock(&res->d_lock);
938
939 inode = NULL; /* don't drop reference */
940 }
941 spin_unlock(&dcache_lock);
942
943 if (inode)
944 iput(inode);
945 if (tmp)
946 dput(tmp);
947 return res;
948}
949
950
951/**
952 * d_splice_alias - splice a disconnected dentry into the tree if one exists
953 * @inode: the inode which may have a disconnected dentry
954 * @dentry: a negative dentry which we want to point to the inode.
955 *
956 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
957 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
958 * and return it, else simply d_add the inode to the dentry and return NULL.
959 *
960 * This is needed in the lookup routine of any filesystem that is exportable
961 * (via knfsd) so that we can build dcache paths to directories effectively.
962 *
963 * If a dentry was found and moved, then it is returned. Otherwise NULL
964 * is returned. This matches the expected return value of ->lookup.
965 *
966 */
967struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
968{
969 struct dentry *new = NULL;
970
971 if (inode) {
972 spin_lock(&dcache_lock);
973 new = __d_find_alias(inode, 1);
974 if (new) {
975 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
976 spin_unlock(&dcache_lock);
977 security_d_instantiate(new, inode);
978 d_rehash(dentry);
979 d_move(new, dentry);
980 iput(inode);
981 } else {
982 /* d_instantiate takes dcache_lock, so we do it by hand */
983 list_add(&dentry->d_alias, &inode->i_dentry);
984 dentry->d_inode = inode;
985 spin_unlock(&dcache_lock);
986 security_d_instantiate(dentry, inode);
987 d_rehash(dentry);
988 }
989 } else
990 d_add(dentry, inode);
991 return new;
992}
993
994
995/**
996 * d_lookup - search for a dentry
997 * @parent: parent dentry
998 * @name: qstr of name we wish to find
999 *
1000 * Searches the children of the parent dentry for the name in question. If
1001 * the dentry is found its reference count is incremented and the dentry
1002 * is returned. The caller must use d_put to free the entry when it has
1003 * finished using it. %NULL is returned on failure.
1004 *
1005 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1006 * Memory barriers are used while updating and doing lockless traversal.
1007 * To avoid races with d_move while rename is happening, d_lock is used.
1008 *
1009 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1010 * and name pointer in one structure pointed by d_qstr.
1011 *
1012 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1013 * lookup is going on.
1014 *
1015 * dentry_unused list is not updated even if lookup finds the required dentry
1016 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1017 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1018 * acquisition.
1019 *
1020 * d_lookup() is protected against the concurrent renames in some unrelated
1021 * directory using the seqlockt_t rename_lock.
1022 */
1023
1024struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1025{
1026 struct dentry * dentry = NULL;
1027 unsigned long seq;
1028
1029 do {
1030 seq = read_seqbegin(&rename_lock);
1031 dentry = __d_lookup(parent, name);
1032 if (dentry)
1033 break;
1034 } while (read_seqretry(&rename_lock, seq));
1035 return dentry;
1036}
1037
1038struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1039{
1040 unsigned int len = name->len;
1041 unsigned int hash = name->hash;
1042 const unsigned char *str = name->name;
1043 struct hlist_head *head = d_hash(parent,hash);
1044 struct dentry *found = NULL;
1045 struct hlist_node *node;
1046
1047 rcu_read_lock();
1048
1049 hlist_for_each_rcu(node, head) {
1050 struct dentry *dentry;
1051 struct qstr *qstr;
1052
1053 dentry = hlist_entry(node, struct dentry, d_hash);
1054
1055 if (dentry->d_name.hash != hash)
1056 continue;
1057 if (dentry->d_parent != parent)
1058 continue;
1059
1060 spin_lock(&dentry->d_lock);
1061
1062 /*
1063 * Recheck the dentry after taking the lock - d_move may have
1064 * changed things. Don't bother checking the hash because we're
1065 * about to compare the whole name anyway.
1066 */
1067 if (dentry->d_parent != parent)
1068 goto next;
1069
1070 /*
1071 * It is safe to compare names since d_move() cannot
1072 * change the qstr (protected by d_lock).
1073 */
1074 qstr = &dentry->d_name;
1075 if (parent->d_op && parent->d_op->d_compare) {
1076 if (parent->d_op->d_compare(parent, qstr, name))
1077 goto next;
1078 } else {
1079 if (qstr->len != len)
1080 goto next;
1081 if (memcmp(qstr->name, str, len))
1082 goto next;
1083 }
1084
1085 if (!d_unhashed(dentry)) {
1086 atomic_inc(&dentry->d_count);
1087 found = dentry;
1088 }
1089 spin_unlock(&dentry->d_lock);
1090 break;
1091next:
1092 spin_unlock(&dentry->d_lock);
1093 }
1094 rcu_read_unlock();
1095
1096 return found;
1097}
1098
1099/**
1100 * d_validate - verify dentry provided from insecure source
1101 * @dentry: The dentry alleged to be valid child of @dparent
1102 * @dparent: The parent dentry (known to be valid)
1103 * @hash: Hash of the dentry
1104 * @len: Length of the name
1105 *
1106 * An insecure source has sent us a dentry, here we verify it and dget() it.
1107 * This is used by ncpfs in its readdir implementation.
1108 * Zero is returned in the dentry is invalid.
1109 */
1110
1111int d_validate(struct dentry *dentry, struct dentry *dparent)
1112{
1113 struct hlist_head *base;
1114 struct hlist_node *lhp;
1115
1116 /* Check whether the ptr might be valid at all.. */
1117 if (!kmem_ptr_validate(dentry_cache, dentry))
1118 goto out;
1119
1120 if (dentry->d_parent != dparent)
1121 goto out;
1122
1123 spin_lock(&dcache_lock);
1124 base = d_hash(dparent, dentry->d_name.hash);
1125 hlist_for_each(lhp,base) {
1126 /* hlist_for_each_rcu() not required for d_hash list
1127 * as it is parsed under dcache_lock
1128 */
1129 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1130 __dget_locked(dentry);
1131 spin_unlock(&dcache_lock);
1132 return 1;
1133 }
1134 }
1135 spin_unlock(&dcache_lock);
1136out:
1137 return 0;
1138}
1139
1140/*
1141 * When a file is deleted, we have two options:
1142 * - turn this dentry into a negative dentry
1143 * - unhash this dentry and free it.
1144 *
1145 * Usually, we want to just turn this into
1146 * a negative dentry, but if anybody else is
1147 * currently using the dentry or the inode
1148 * we can't do that and we fall back on removing
1149 * it from the hash queues and waiting for
1150 * it to be deleted later when it has no users
1151 */
1152
1153/**
1154 * d_delete - delete a dentry
1155 * @dentry: The dentry to delete
1156 *
1157 * Turn the dentry into a negative dentry if possible, otherwise
1158 * remove it from the hash queues so it can be deleted later
1159 */
1160
1161void d_delete(struct dentry * dentry)
1162{
John McCutchan7a91bf72005-08-08 13:52:16 -04001163 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 /*
1165 * Are we the only user?
1166 */
1167 spin_lock(&dcache_lock);
1168 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001169 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (atomic_read(&dentry->d_count) == 1) {
1171 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001172 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 return;
1174 }
1175
1176 if (!d_unhashed(dentry))
1177 __d_drop(dentry);
1178
1179 spin_unlock(&dentry->d_lock);
1180 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001181
1182 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183}
1184
1185static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1186{
1187
1188 entry->d_flags &= ~DCACHE_UNHASHED;
1189 hlist_add_head_rcu(&entry->d_hash, list);
1190}
1191
1192/**
1193 * d_rehash - add an entry back to the hash
1194 * @entry: dentry to add to the hash
1195 *
1196 * Adds a dentry to the hash according to its name.
1197 */
1198
1199void d_rehash(struct dentry * entry)
1200{
1201 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1202
1203 spin_lock(&dcache_lock);
1204 spin_lock(&entry->d_lock);
1205 __d_rehash(entry, list);
1206 spin_unlock(&entry->d_lock);
1207 spin_unlock(&dcache_lock);
1208}
1209
1210#define do_switch(x,y) do { \
1211 __typeof__ (x) __tmp = x; \
1212 x = y; y = __tmp; } while (0)
1213
1214/*
1215 * When switching names, the actual string doesn't strictly have to
1216 * be preserved in the target - because we're dropping the target
1217 * anyway. As such, we can just do a simple memcpy() to copy over
1218 * the new name before we switch.
1219 *
1220 * Note that we have to be a lot more careful about getting the hash
1221 * switched - we have to switch the hash value properly even if it
1222 * then no longer matches the actual (corrupted) string of the target.
1223 * The hash value has to match the hash queue that the dentry is on..
1224 */
1225static void switch_names(struct dentry *dentry, struct dentry *target)
1226{
1227 if (dname_external(target)) {
1228 if (dname_external(dentry)) {
1229 /*
1230 * Both external: swap the pointers
1231 */
1232 do_switch(target->d_name.name, dentry->d_name.name);
1233 } else {
1234 /*
1235 * dentry:internal, target:external. Steal target's
1236 * storage and make target internal.
1237 */
1238 dentry->d_name.name = target->d_name.name;
1239 target->d_name.name = target->d_iname;
1240 }
1241 } else {
1242 if (dname_external(dentry)) {
1243 /*
1244 * dentry:external, target:internal. Give dentry's
1245 * storage to target and make dentry internal
1246 */
1247 memcpy(dentry->d_iname, target->d_name.name,
1248 target->d_name.len + 1);
1249 target->d_name.name = dentry->d_name.name;
1250 dentry->d_name.name = dentry->d_iname;
1251 } else {
1252 /*
1253 * Both are internal. Just copy target to dentry
1254 */
1255 memcpy(dentry->d_iname, target->d_name.name,
1256 target->d_name.len + 1);
1257 }
1258 }
1259}
1260
1261/*
1262 * We cannibalize "target" when moving dentry on top of it,
1263 * because it's going to be thrown away anyway. We could be more
1264 * polite about it, though.
1265 *
1266 * This forceful removal will result in ugly /proc output if
1267 * somebody holds a file open that got deleted due to a rename.
1268 * We could be nicer about the deleted file, and let it show
1269 * up under the name it got deleted rather than the name that
1270 * deleted it.
1271 */
1272
1273/**
1274 * d_move - move a dentry
1275 * @dentry: entry to move
1276 * @target: new dentry
1277 *
1278 * Update the dcache to reflect the move of a file name. Negative
1279 * dcache entries should not be moved in this way.
1280 */
1281
1282void d_move(struct dentry * dentry, struct dentry * target)
1283{
1284 struct hlist_head *list;
1285
1286 if (!dentry->d_inode)
1287 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1288
1289 spin_lock(&dcache_lock);
1290 write_seqlock(&rename_lock);
1291 /*
1292 * XXXX: do we really need to take target->d_lock?
1293 */
1294 if (target < dentry) {
1295 spin_lock(&target->d_lock);
1296 spin_lock(&dentry->d_lock);
1297 } else {
1298 spin_lock(&dentry->d_lock);
1299 spin_lock(&target->d_lock);
1300 }
1301
1302 /* Move the dentry to the target hash queue, if on different bucket */
1303 if (dentry->d_flags & DCACHE_UNHASHED)
1304 goto already_unhashed;
1305
1306 hlist_del_rcu(&dentry->d_hash);
1307
1308already_unhashed:
1309 list = d_hash(target->d_parent, target->d_name.hash);
1310 __d_rehash(dentry, list);
1311
1312 /* Unhash the target: dput() will then get rid of it */
1313 __d_drop(target);
1314
1315 list_del(&dentry->d_child);
1316 list_del(&target->d_child);
1317
1318 /* Switch the names.. */
1319 switch_names(dentry, target);
1320 do_switch(dentry->d_name.len, target->d_name.len);
1321 do_switch(dentry->d_name.hash, target->d_name.hash);
1322
1323 /* ... and switch the parents */
1324 if (IS_ROOT(dentry)) {
1325 dentry->d_parent = target->d_parent;
1326 target->d_parent = target;
1327 INIT_LIST_HEAD(&target->d_child);
1328 } else {
1329 do_switch(dentry->d_parent, target->d_parent);
1330
1331 /* And add them back to the (new) parent lists */
1332 list_add(&target->d_child, &target->d_parent->d_subdirs);
1333 }
1334
1335 list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
1336 spin_unlock(&target->d_lock);
1337 spin_unlock(&dentry->d_lock);
1338 write_sequnlock(&rename_lock);
1339 spin_unlock(&dcache_lock);
1340}
1341
1342/**
1343 * d_path - return the path of a dentry
1344 * @dentry: dentry to report
1345 * @vfsmnt: vfsmnt to which the dentry belongs
1346 * @root: root dentry
1347 * @rootmnt: vfsmnt to which the root dentry belongs
1348 * @buffer: buffer to return value in
1349 * @buflen: buffer length
1350 *
1351 * Convert a dentry into an ASCII path name. If the entry has been deleted
1352 * the string " (deleted)" is appended. Note that this is ambiguous.
1353 *
1354 * Returns the buffer or an error code if the path was too long.
1355 *
1356 * "buflen" should be positive. Caller holds the dcache_lock.
1357 */
1358static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1359 struct dentry *root, struct vfsmount *rootmnt,
1360 char *buffer, int buflen)
1361{
1362 char * end = buffer+buflen;
1363 char * retval;
1364 int namelen;
1365
1366 *--end = '\0';
1367 buflen--;
1368 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1369 buflen -= 10;
1370 end -= 10;
1371 if (buflen < 0)
1372 goto Elong;
1373 memcpy(end, " (deleted)", 10);
1374 }
1375
1376 if (buflen < 1)
1377 goto Elong;
1378 /* Get '/' right */
1379 retval = end-1;
1380 *retval = '/';
1381
1382 for (;;) {
1383 struct dentry * parent;
1384
1385 if (dentry == root && vfsmnt == rootmnt)
1386 break;
1387 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1388 /* Global root? */
1389 spin_lock(&vfsmount_lock);
1390 if (vfsmnt->mnt_parent == vfsmnt) {
1391 spin_unlock(&vfsmount_lock);
1392 goto global_root;
1393 }
1394 dentry = vfsmnt->mnt_mountpoint;
1395 vfsmnt = vfsmnt->mnt_parent;
1396 spin_unlock(&vfsmount_lock);
1397 continue;
1398 }
1399 parent = dentry->d_parent;
1400 prefetch(parent);
1401 namelen = dentry->d_name.len;
1402 buflen -= namelen + 1;
1403 if (buflen < 0)
1404 goto Elong;
1405 end -= namelen;
1406 memcpy(end, dentry->d_name.name, namelen);
1407 *--end = '/';
1408 retval = end;
1409 dentry = parent;
1410 }
1411
1412 return retval;
1413
1414global_root:
1415 namelen = dentry->d_name.len;
1416 buflen -= namelen;
1417 if (buflen < 0)
1418 goto Elong;
1419 retval -= namelen-1; /* hit the slash */
1420 memcpy(retval, dentry->d_name.name, namelen);
1421 return retval;
1422Elong:
1423 return ERR_PTR(-ENAMETOOLONG);
1424}
1425
1426/* write full pathname into buffer and return start of pathname */
1427char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1428 char *buf, int buflen)
1429{
1430 char *res;
1431 struct vfsmount *rootmnt;
1432 struct dentry *root;
1433
1434 read_lock(&current->fs->lock);
1435 rootmnt = mntget(current->fs->rootmnt);
1436 root = dget(current->fs->root);
1437 read_unlock(&current->fs->lock);
1438 spin_lock(&dcache_lock);
1439 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1440 spin_unlock(&dcache_lock);
1441 dput(root);
1442 mntput(rootmnt);
1443 return res;
1444}
1445
1446/*
1447 * NOTE! The user-level library version returns a
1448 * character pointer. The kernel system call just
1449 * returns the length of the buffer filled (which
1450 * includes the ending '\0' character), or a negative
1451 * error value. So libc would do something like
1452 *
1453 * char *getcwd(char * buf, size_t size)
1454 * {
1455 * int retval;
1456 *
1457 * retval = sys_getcwd(buf, size);
1458 * if (retval >= 0)
1459 * return buf;
1460 * errno = -retval;
1461 * return NULL;
1462 * }
1463 */
1464asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1465{
1466 int error;
1467 struct vfsmount *pwdmnt, *rootmnt;
1468 struct dentry *pwd, *root;
1469 char *page = (char *) __get_free_page(GFP_USER);
1470
1471 if (!page)
1472 return -ENOMEM;
1473
1474 read_lock(&current->fs->lock);
1475 pwdmnt = mntget(current->fs->pwdmnt);
1476 pwd = dget(current->fs->pwd);
1477 rootmnt = mntget(current->fs->rootmnt);
1478 root = dget(current->fs->root);
1479 read_unlock(&current->fs->lock);
1480
1481 error = -ENOENT;
1482 /* Has the current directory has been unlinked? */
1483 spin_lock(&dcache_lock);
1484 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1485 unsigned long len;
1486 char * cwd;
1487
1488 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1489 spin_unlock(&dcache_lock);
1490
1491 error = PTR_ERR(cwd);
1492 if (IS_ERR(cwd))
1493 goto out;
1494
1495 error = -ERANGE;
1496 len = PAGE_SIZE + page - cwd;
1497 if (len <= size) {
1498 error = len;
1499 if (copy_to_user(buf, cwd, len))
1500 error = -EFAULT;
1501 }
1502 } else
1503 spin_unlock(&dcache_lock);
1504
1505out:
1506 dput(pwd);
1507 mntput(pwdmnt);
1508 dput(root);
1509 mntput(rootmnt);
1510 free_page((unsigned long) page);
1511 return error;
1512}
1513
1514/*
1515 * Test whether new_dentry is a subdirectory of old_dentry.
1516 *
1517 * Trivially implemented using the dcache structure
1518 */
1519
1520/**
1521 * is_subdir - is new dentry a subdirectory of old_dentry
1522 * @new_dentry: new dentry
1523 * @old_dentry: old dentry
1524 *
1525 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1526 * Returns 0 otherwise.
1527 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1528 */
1529
1530int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1531{
1532 int result;
1533 struct dentry * saved = new_dentry;
1534 unsigned long seq;
1535
1536 /* need rcu_readlock to protect against the d_parent trashing due to
1537 * d_move
1538 */
1539 rcu_read_lock();
1540 do {
1541 /* for restarting inner loop in case of seq retry */
1542 new_dentry = saved;
1543 result = 0;
1544 seq = read_seqbegin(&rename_lock);
1545 for (;;) {
1546 if (new_dentry != old_dentry) {
1547 struct dentry * parent = new_dentry->d_parent;
1548 if (parent == new_dentry)
1549 break;
1550 new_dentry = parent;
1551 continue;
1552 }
1553 result = 1;
1554 break;
1555 }
1556 } while (read_seqretry(&rename_lock, seq));
1557 rcu_read_unlock();
1558
1559 return result;
1560}
1561
1562void d_genocide(struct dentry *root)
1563{
1564 struct dentry *this_parent = root;
1565 struct list_head *next;
1566
1567 spin_lock(&dcache_lock);
1568repeat:
1569 next = this_parent->d_subdirs.next;
1570resume:
1571 while (next != &this_parent->d_subdirs) {
1572 struct list_head *tmp = next;
1573 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1574 next = tmp->next;
1575 if (d_unhashed(dentry)||!dentry->d_inode)
1576 continue;
1577 if (!list_empty(&dentry->d_subdirs)) {
1578 this_parent = dentry;
1579 goto repeat;
1580 }
1581 atomic_dec(&dentry->d_count);
1582 }
1583 if (this_parent != root) {
1584 next = this_parent->d_child.next;
1585 atomic_dec(&this_parent->d_count);
1586 this_parent = this_parent->d_parent;
1587 goto resume;
1588 }
1589 spin_unlock(&dcache_lock);
1590}
1591
1592/**
1593 * find_inode_number - check for dentry with name
1594 * @dir: directory to check
1595 * @name: Name to find.
1596 *
1597 * Check whether a dentry already exists for the given name,
1598 * and return the inode number if it has an inode. Otherwise
1599 * 0 is returned.
1600 *
1601 * This routine is used to post-process directory listings for
1602 * filesystems using synthetic inode numbers, and is necessary
1603 * to keep getcwd() working.
1604 */
1605
1606ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1607{
1608 struct dentry * dentry;
1609 ino_t ino = 0;
1610
1611 /*
1612 * Check for a fs-specific hash function. Note that we must
1613 * calculate the standard hash first, as the d_op->d_hash()
1614 * routine may choose to leave the hash value unchanged.
1615 */
1616 name->hash = full_name_hash(name->name, name->len);
1617 if (dir->d_op && dir->d_op->d_hash)
1618 {
1619 if (dir->d_op->d_hash(dir, name) != 0)
1620 goto out;
1621 }
1622
1623 dentry = d_lookup(dir, name);
1624 if (dentry)
1625 {
1626 if (dentry->d_inode)
1627 ino = dentry->d_inode->i_ino;
1628 dput(dentry);
1629 }
1630out:
1631 return ino;
1632}
1633
1634static __initdata unsigned long dhash_entries;
1635static int __init set_dhash_entries(char *str)
1636{
1637 if (!str)
1638 return 0;
1639 dhash_entries = simple_strtoul(str, &str, 0);
1640 return 1;
1641}
1642__setup("dhash_entries=", set_dhash_entries);
1643
1644static void __init dcache_init_early(void)
1645{
1646 int loop;
1647
1648 /* If hashes are distributed across NUMA nodes, defer
1649 * hash allocation until vmalloc space is available.
1650 */
1651 if (hashdist)
1652 return;
1653
1654 dentry_hashtable =
1655 alloc_large_system_hash("Dentry cache",
1656 sizeof(struct hlist_head),
1657 dhash_entries,
1658 13,
1659 HASH_EARLY,
1660 &d_hash_shift,
1661 &d_hash_mask,
1662 0);
1663
1664 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1665 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1666}
1667
1668static void __init dcache_init(unsigned long mempages)
1669{
1670 int loop;
1671
1672 /*
1673 * A constructor could be added for stable state like the lists,
1674 * but it is probably not worth it because of the cache nature
1675 * of the dcache.
1676 */
1677 dentry_cache = kmem_cache_create("dentry_cache",
1678 sizeof(struct dentry),
1679 0,
1680 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1681 NULL, NULL);
1682
1683 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1684
1685 /* Hash may have been set up in dcache_init_early */
1686 if (!hashdist)
1687 return;
1688
1689 dentry_hashtable =
1690 alloc_large_system_hash("Dentry cache",
1691 sizeof(struct hlist_head),
1692 dhash_entries,
1693 13,
1694 0,
1695 &d_hash_shift,
1696 &d_hash_mask,
1697 0);
1698
1699 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1700 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1701}
1702
1703/* SLAB cache for __getname() consumers */
1704kmem_cache_t *names_cachep;
1705
1706/* SLAB cache for file structures */
1707kmem_cache_t *filp_cachep;
1708
1709EXPORT_SYMBOL(d_genocide);
1710
1711extern void bdev_cache_init(void);
1712extern void chrdev_init(void);
1713
1714void __init vfs_caches_init_early(void)
1715{
1716 dcache_init_early();
1717 inode_init_early();
1718}
1719
1720void __init vfs_caches_init(unsigned long mempages)
1721{
1722 unsigned long reserve;
1723
1724 /* Base hash sizes on available memory, with a reserve equal to
1725 150% of current kernel size */
1726
1727 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1728 mempages -= reserve;
1729
1730 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1731 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1732
1733 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
1734 SLAB_HWCACHE_ALIGN|SLAB_PANIC, filp_ctor, filp_dtor);
1735
1736 dcache_init(mempages);
1737 inode_init(mempages);
1738 files_init(mempages);
1739 mnt_init(mempages);
1740 bdev_cache_init();
1741 chrdev_init();
1742}
1743
1744EXPORT_SYMBOL(d_alloc);
1745EXPORT_SYMBOL(d_alloc_anon);
1746EXPORT_SYMBOL(d_alloc_root);
1747EXPORT_SYMBOL(d_delete);
1748EXPORT_SYMBOL(d_find_alias);
1749EXPORT_SYMBOL(d_instantiate);
1750EXPORT_SYMBOL(d_invalidate);
1751EXPORT_SYMBOL(d_lookup);
1752EXPORT_SYMBOL(d_move);
1753EXPORT_SYMBOL(d_path);
1754EXPORT_SYMBOL(d_prune_aliases);
1755EXPORT_SYMBOL(d_rehash);
1756EXPORT_SYMBOL(d_splice_alias);
1757EXPORT_SYMBOL(d_validate);
1758EXPORT_SYMBOL(dget_locked);
1759EXPORT_SYMBOL(dput);
1760EXPORT_SYMBOL(find_inode_number);
1761EXPORT_SYMBOL(have_submounts);
1762EXPORT_SYMBOL(names_cachep);
1763EXPORT_SYMBOL(shrink_dcache_parent);
1764EXPORT_SYMBOL(shrink_dcache_sb);