blob: 11dc83092d4aa48c78650f6dee4b48b834d25d7e [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{
Eric Dumazet5160ee62006-01-08 01:03:32 -080074 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
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);
Eric Dumazet5160ee62006-01-08 01:03:32 -080089 call_rcu(&dentry->d_u.d_rcu, d_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090}
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 */
Arjan van de Ven858119e2006-01-14 13:20:43 -080097static void dentry_iput(struct dentry * dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
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 }
Eric Dumazet5160ee62006-01-08 01:03:32 -0800196 list_del(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 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);
Eric Dumazet5160ee62006-01-08 01:03:32 -0800370 list_del(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 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;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800521 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 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) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800535 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 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;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800572 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 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) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800613 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 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 *
Paul E. McKenney665a7582005-11-07 00:59:17 -0800647 * parsing d_hash list does not hlist_for_each_entry_rcu() as it
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 * 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 */
Al Viro27496a82005-10-21 03:20:48 -0400692static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
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;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800746#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800748#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 INIT_HLIST_NODE(&dentry->d_hash);
750 INIT_LIST_HEAD(&dentry->d_lru);
751 INIT_LIST_HEAD(&dentry->d_subdirs);
752 INIT_LIST_HEAD(&dentry->d_alias);
753
754 if (parent) {
755 dentry->d_parent = dget(parent);
756 dentry->d_sb = parent->d_sb;
757 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800758 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
760
761 spin_lock(&dcache_lock);
762 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800763 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 dentry_stat.nr_dentry++;
765 spin_unlock(&dcache_lock);
766
767 return dentry;
768}
769
770struct dentry *d_alloc_name(struct dentry *parent, const char *name)
771{
772 struct qstr q;
773
774 q.name = name;
775 q.len = strlen(name);
776 q.hash = full_name_hash(q.name, q.len);
777 return d_alloc(parent, &q);
778}
779
780/**
781 * d_instantiate - fill in inode information for a dentry
782 * @entry: dentry to complete
783 * @inode: inode to attach to this dentry
784 *
785 * Fill in inode information in the entry.
786 *
787 * This turns negative dentries into productive full members
788 * of society.
789 *
790 * NOTE! This assumes that the inode count has been incremented
791 * (or otherwise set) by the caller to indicate that it is now
792 * in use by the dcache.
793 */
794
795void d_instantiate(struct dentry *entry, struct inode * inode)
796{
797 if (!list_empty(&entry->d_alias)) BUG();
798 spin_lock(&dcache_lock);
799 if (inode)
800 list_add(&entry->d_alias, &inode->i_dentry);
801 entry->d_inode = inode;
802 spin_unlock(&dcache_lock);
803 security_d_instantiate(entry, inode);
804}
805
806/**
807 * d_instantiate_unique - instantiate a non-aliased dentry
808 * @entry: dentry to instantiate
809 * @inode: inode to attach to this dentry
810 *
811 * Fill in inode information in the entry. On success, it returns NULL.
812 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -0800813 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 *
815 * Note that in order to avoid conflicts with rename() etc, the caller
816 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -0800817 *
818 * This also assumes that the inode count has been incremented
819 * (or otherwise set) by the caller to indicate that it is now
820 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 */
822struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
823{
824 struct dentry *alias;
825 int len = entry->d_name.len;
826 const char *name = entry->d_name.name;
827 unsigned int hash = entry->d_name.hash;
828
829 BUG_ON(!list_empty(&entry->d_alias));
830 spin_lock(&dcache_lock);
831 if (!inode)
832 goto do_negative;
833 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
834 struct qstr *qstr = &alias->d_name;
835
836 if (qstr->hash != hash)
837 continue;
838 if (alias->d_parent != entry->d_parent)
839 continue;
840 if (qstr->len != len)
841 continue;
842 if (memcmp(qstr->name, name, len))
843 continue;
844 dget_locked(alias);
845 spin_unlock(&dcache_lock);
846 BUG_ON(!d_unhashed(alias));
Oleg Drokine866cfa2006-01-09 20:52:51 -0800847 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 return alias;
849 }
850 list_add(&entry->d_alias, &inode->i_dentry);
851do_negative:
852 entry->d_inode = inode;
853 spin_unlock(&dcache_lock);
854 security_d_instantiate(entry, inode);
855 return NULL;
856}
857EXPORT_SYMBOL(d_instantiate_unique);
858
859/**
860 * d_alloc_root - allocate root dentry
861 * @root_inode: inode to allocate the root for
862 *
863 * Allocate a root ("/") dentry for the inode given. The inode is
864 * instantiated and returned. %NULL is returned if there is insufficient
865 * memory or the inode passed is %NULL.
866 */
867
868struct dentry * d_alloc_root(struct inode * root_inode)
869{
870 struct dentry *res = NULL;
871
872 if (root_inode) {
873 static const struct qstr name = { .name = "/", .len = 1 };
874
875 res = d_alloc(NULL, &name);
876 if (res) {
877 res->d_sb = root_inode->i_sb;
878 res->d_parent = res;
879 d_instantiate(res, root_inode);
880 }
881 }
882 return res;
883}
884
885static inline struct hlist_head *d_hash(struct dentry *parent,
886 unsigned long hash)
887{
888 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
889 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
890 return dentry_hashtable + (hash & D_HASHMASK);
891}
892
893/**
894 * d_alloc_anon - allocate an anonymous dentry
895 * @inode: inode to allocate the dentry for
896 *
897 * This is similar to d_alloc_root. It is used by filesystems when
898 * creating a dentry for a given inode, often in the process of
899 * mapping a filehandle to a dentry. The returned dentry may be
900 * anonymous, or may have a full name (if the inode was already
901 * in the cache). The file system may need to make further
902 * efforts to connect this dentry into the dcache properly.
903 *
904 * When called on a directory inode, we must ensure that
905 * the inode only ever has one dentry. If a dentry is
906 * found, that is returned instead of allocating a new one.
907 *
908 * On successful return, the reference to the inode has been transferred
909 * to the dentry. If %NULL is returned (indicating kmalloc failure),
910 * the reference on the inode has not been released.
911 */
912
913struct dentry * d_alloc_anon(struct inode *inode)
914{
915 static const struct qstr anonstring = { .name = "" };
916 struct dentry *tmp;
917 struct dentry *res;
918
919 if ((res = d_find_alias(inode))) {
920 iput(inode);
921 return res;
922 }
923
924 tmp = d_alloc(NULL, &anonstring);
925 if (!tmp)
926 return NULL;
927
928 tmp->d_parent = tmp; /* make sure dput doesn't croak */
929
930 spin_lock(&dcache_lock);
931 res = __d_find_alias(inode, 0);
932 if (!res) {
933 /* attach a disconnected dentry */
934 res = tmp;
935 tmp = NULL;
936 spin_lock(&res->d_lock);
937 res->d_sb = inode->i_sb;
938 res->d_parent = res;
939 res->d_inode = inode;
940 res->d_flags |= DCACHE_DISCONNECTED;
941 res->d_flags &= ~DCACHE_UNHASHED;
942 list_add(&res->d_alias, &inode->i_dentry);
943 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
944 spin_unlock(&res->d_lock);
945
946 inode = NULL; /* don't drop reference */
947 }
948 spin_unlock(&dcache_lock);
949
950 if (inode)
951 iput(inode);
952 if (tmp)
953 dput(tmp);
954 return res;
955}
956
957
958/**
959 * d_splice_alias - splice a disconnected dentry into the tree if one exists
960 * @inode: the inode which may have a disconnected dentry
961 * @dentry: a negative dentry which we want to point to the inode.
962 *
963 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
964 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
965 * and return it, else simply d_add the inode to the dentry and return NULL.
966 *
967 * This is needed in the lookup routine of any filesystem that is exportable
968 * (via knfsd) so that we can build dcache paths to directories effectively.
969 *
970 * If a dentry was found and moved, then it is returned. Otherwise NULL
971 * is returned. This matches the expected return value of ->lookup.
972 *
973 */
974struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
975{
976 struct dentry *new = NULL;
977
978 if (inode) {
979 spin_lock(&dcache_lock);
980 new = __d_find_alias(inode, 1);
981 if (new) {
982 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
983 spin_unlock(&dcache_lock);
984 security_d_instantiate(new, inode);
985 d_rehash(dentry);
986 d_move(new, dentry);
987 iput(inode);
988 } else {
989 /* d_instantiate takes dcache_lock, so we do it by hand */
990 list_add(&dentry->d_alias, &inode->i_dentry);
991 dentry->d_inode = inode;
992 spin_unlock(&dcache_lock);
993 security_d_instantiate(dentry, inode);
994 d_rehash(dentry);
995 }
996 } else
997 d_add(dentry, inode);
998 return new;
999}
1000
1001
1002/**
1003 * d_lookup - search for a dentry
1004 * @parent: parent dentry
1005 * @name: qstr of name we wish to find
1006 *
1007 * Searches the children of the parent dentry for the name in question. If
1008 * the dentry is found its reference count is incremented and the dentry
1009 * is returned. The caller must use d_put to free the entry when it has
1010 * finished using it. %NULL is returned on failure.
1011 *
1012 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1013 * Memory barriers are used while updating and doing lockless traversal.
1014 * To avoid races with d_move while rename is happening, d_lock is used.
1015 *
1016 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1017 * and name pointer in one structure pointed by d_qstr.
1018 *
1019 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1020 * lookup is going on.
1021 *
1022 * dentry_unused list is not updated even if lookup finds the required dentry
1023 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1024 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1025 * acquisition.
1026 *
1027 * d_lookup() is protected against the concurrent renames in some unrelated
1028 * directory using the seqlockt_t rename_lock.
1029 */
1030
1031struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1032{
1033 struct dentry * dentry = NULL;
1034 unsigned long seq;
1035
1036 do {
1037 seq = read_seqbegin(&rename_lock);
1038 dentry = __d_lookup(parent, name);
1039 if (dentry)
1040 break;
1041 } while (read_seqretry(&rename_lock, seq));
1042 return dentry;
1043}
1044
1045struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1046{
1047 unsigned int len = name->len;
1048 unsigned int hash = name->hash;
1049 const unsigned char *str = name->name;
1050 struct hlist_head *head = d_hash(parent,hash);
1051 struct dentry *found = NULL;
1052 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001053 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
1055 rcu_read_lock();
1056
Paul E. McKenney665a7582005-11-07 00:59:17 -08001057 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 struct qstr *qstr;
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (dentry->d_name.hash != hash)
1061 continue;
1062 if (dentry->d_parent != parent)
1063 continue;
1064
1065 spin_lock(&dentry->d_lock);
1066
1067 /*
1068 * Recheck the dentry after taking the lock - d_move may have
1069 * changed things. Don't bother checking the hash because we're
1070 * about to compare the whole name anyway.
1071 */
1072 if (dentry->d_parent != parent)
1073 goto next;
1074
1075 /*
1076 * It is safe to compare names since d_move() cannot
1077 * change the qstr (protected by d_lock).
1078 */
1079 qstr = &dentry->d_name;
1080 if (parent->d_op && parent->d_op->d_compare) {
1081 if (parent->d_op->d_compare(parent, qstr, name))
1082 goto next;
1083 } else {
1084 if (qstr->len != len)
1085 goto next;
1086 if (memcmp(qstr->name, str, len))
1087 goto next;
1088 }
1089
1090 if (!d_unhashed(dentry)) {
1091 atomic_inc(&dentry->d_count);
1092 found = dentry;
1093 }
1094 spin_unlock(&dentry->d_lock);
1095 break;
1096next:
1097 spin_unlock(&dentry->d_lock);
1098 }
1099 rcu_read_unlock();
1100
1101 return found;
1102}
1103
1104/**
1105 * d_validate - verify dentry provided from insecure source
1106 * @dentry: The dentry alleged to be valid child of @dparent
1107 * @dparent: The parent dentry (known to be valid)
1108 * @hash: Hash of the dentry
1109 * @len: Length of the name
1110 *
1111 * An insecure source has sent us a dentry, here we verify it and dget() it.
1112 * This is used by ncpfs in its readdir implementation.
1113 * Zero is returned in the dentry is invalid.
1114 */
1115
1116int d_validate(struct dentry *dentry, struct dentry *dparent)
1117{
1118 struct hlist_head *base;
1119 struct hlist_node *lhp;
1120
1121 /* Check whether the ptr might be valid at all.. */
1122 if (!kmem_ptr_validate(dentry_cache, dentry))
1123 goto out;
1124
1125 if (dentry->d_parent != dparent)
1126 goto out;
1127
1128 spin_lock(&dcache_lock);
1129 base = d_hash(dparent, dentry->d_name.hash);
1130 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001131 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 * as it is parsed under dcache_lock
1133 */
1134 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1135 __dget_locked(dentry);
1136 spin_unlock(&dcache_lock);
1137 return 1;
1138 }
1139 }
1140 spin_unlock(&dcache_lock);
1141out:
1142 return 0;
1143}
1144
1145/*
1146 * When a file is deleted, we have two options:
1147 * - turn this dentry into a negative dentry
1148 * - unhash this dentry and free it.
1149 *
1150 * Usually, we want to just turn this into
1151 * a negative dentry, but if anybody else is
1152 * currently using the dentry or the inode
1153 * we can't do that and we fall back on removing
1154 * it from the hash queues and waiting for
1155 * it to be deleted later when it has no users
1156 */
1157
1158/**
1159 * d_delete - delete a dentry
1160 * @dentry: The dentry to delete
1161 *
1162 * Turn the dentry into a negative dentry if possible, otherwise
1163 * remove it from the hash queues so it can be deleted later
1164 */
1165
1166void d_delete(struct dentry * dentry)
1167{
John McCutchan7a91bf72005-08-08 13:52:16 -04001168 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 /*
1170 * Are we the only user?
1171 */
1172 spin_lock(&dcache_lock);
1173 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001174 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (atomic_read(&dentry->d_count) == 1) {
1176 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001177 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 return;
1179 }
1180
1181 if (!d_unhashed(dentry))
1182 __d_drop(dentry);
1183
1184 spin_unlock(&dentry->d_lock);
1185 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001186
1187 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
1190static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1191{
1192
1193 entry->d_flags &= ~DCACHE_UNHASHED;
1194 hlist_add_head_rcu(&entry->d_hash, list);
1195}
1196
1197/**
1198 * d_rehash - add an entry back to the hash
1199 * @entry: dentry to add to the hash
1200 *
1201 * Adds a dentry to the hash according to its name.
1202 */
1203
1204void d_rehash(struct dentry * entry)
1205{
1206 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1207
1208 spin_lock(&dcache_lock);
1209 spin_lock(&entry->d_lock);
1210 __d_rehash(entry, list);
1211 spin_unlock(&entry->d_lock);
1212 spin_unlock(&dcache_lock);
1213}
1214
1215#define do_switch(x,y) do { \
1216 __typeof__ (x) __tmp = x; \
1217 x = y; y = __tmp; } while (0)
1218
1219/*
1220 * When switching names, the actual string doesn't strictly have to
1221 * be preserved in the target - because we're dropping the target
1222 * anyway. As such, we can just do a simple memcpy() to copy over
1223 * the new name before we switch.
1224 *
1225 * Note that we have to be a lot more careful about getting the hash
1226 * switched - we have to switch the hash value properly even if it
1227 * then no longer matches the actual (corrupted) string of the target.
1228 * The hash value has to match the hash queue that the dentry is on..
1229 */
1230static void switch_names(struct dentry *dentry, struct dentry *target)
1231{
1232 if (dname_external(target)) {
1233 if (dname_external(dentry)) {
1234 /*
1235 * Both external: swap the pointers
1236 */
1237 do_switch(target->d_name.name, dentry->d_name.name);
1238 } else {
1239 /*
1240 * dentry:internal, target:external. Steal target's
1241 * storage and make target internal.
1242 */
1243 dentry->d_name.name = target->d_name.name;
1244 target->d_name.name = target->d_iname;
1245 }
1246 } else {
1247 if (dname_external(dentry)) {
1248 /*
1249 * dentry:external, target:internal. Give dentry's
1250 * storage to target and make dentry internal
1251 */
1252 memcpy(dentry->d_iname, target->d_name.name,
1253 target->d_name.len + 1);
1254 target->d_name.name = dentry->d_name.name;
1255 dentry->d_name.name = dentry->d_iname;
1256 } else {
1257 /*
1258 * Both are internal. Just copy target to dentry
1259 */
1260 memcpy(dentry->d_iname, target->d_name.name,
1261 target->d_name.len + 1);
1262 }
1263 }
1264}
1265
1266/*
1267 * We cannibalize "target" when moving dentry on top of it,
1268 * because it's going to be thrown away anyway. We could be more
1269 * polite about it, though.
1270 *
1271 * This forceful removal will result in ugly /proc output if
1272 * somebody holds a file open that got deleted due to a rename.
1273 * We could be nicer about the deleted file, and let it show
1274 * up under the name it got deleted rather than the name that
1275 * deleted it.
1276 */
1277
1278/**
1279 * d_move - move a dentry
1280 * @dentry: entry to move
1281 * @target: new dentry
1282 *
1283 * Update the dcache to reflect the move of a file name. Negative
1284 * dcache entries should not be moved in this way.
1285 */
1286
1287void d_move(struct dentry * dentry, struct dentry * target)
1288{
1289 struct hlist_head *list;
1290
1291 if (!dentry->d_inode)
1292 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1293
1294 spin_lock(&dcache_lock);
1295 write_seqlock(&rename_lock);
1296 /*
1297 * XXXX: do we really need to take target->d_lock?
1298 */
1299 if (target < dentry) {
1300 spin_lock(&target->d_lock);
1301 spin_lock(&dentry->d_lock);
1302 } else {
1303 spin_lock(&dentry->d_lock);
1304 spin_lock(&target->d_lock);
1305 }
1306
1307 /* Move the dentry to the target hash queue, if on different bucket */
1308 if (dentry->d_flags & DCACHE_UNHASHED)
1309 goto already_unhashed;
1310
1311 hlist_del_rcu(&dentry->d_hash);
1312
1313already_unhashed:
1314 list = d_hash(target->d_parent, target->d_name.hash);
1315 __d_rehash(dentry, list);
1316
1317 /* Unhash the target: dput() will then get rid of it */
1318 __d_drop(target);
1319
Eric Dumazet5160ee62006-01-08 01:03:32 -08001320 list_del(&dentry->d_u.d_child);
1321 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323 /* Switch the names.. */
1324 switch_names(dentry, target);
1325 do_switch(dentry->d_name.len, target->d_name.len);
1326 do_switch(dentry->d_name.hash, target->d_name.hash);
1327
1328 /* ... and switch the parents */
1329 if (IS_ROOT(dentry)) {
1330 dentry->d_parent = target->d_parent;
1331 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001332 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 } else {
1334 do_switch(dentry->d_parent, target->d_parent);
1335
1336 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001337 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
1339
Eric Dumazet5160ee62006-01-08 01:03:32 -08001340 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 spin_unlock(&target->d_lock);
1342 spin_unlock(&dentry->d_lock);
1343 write_sequnlock(&rename_lock);
1344 spin_unlock(&dcache_lock);
1345}
1346
1347/**
1348 * d_path - return the path of a dentry
1349 * @dentry: dentry to report
1350 * @vfsmnt: vfsmnt to which the dentry belongs
1351 * @root: root dentry
1352 * @rootmnt: vfsmnt to which the root dentry belongs
1353 * @buffer: buffer to return value in
1354 * @buflen: buffer length
1355 *
1356 * Convert a dentry into an ASCII path name. If the entry has been deleted
1357 * the string " (deleted)" is appended. Note that this is ambiguous.
1358 *
1359 * Returns the buffer or an error code if the path was too long.
1360 *
1361 * "buflen" should be positive. Caller holds the dcache_lock.
1362 */
1363static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1364 struct dentry *root, struct vfsmount *rootmnt,
1365 char *buffer, int buflen)
1366{
1367 char * end = buffer+buflen;
1368 char * retval;
1369 int namelen;
1370
1371 *--end = '\0';
1372 buflen--;
1373 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1374 buflen -= 10;
1375 end -= 10;
1376 if (buflen < 0)
1377 goto Elong;
1378 memcpy(end, " (deleted)", 10);
1379 }
1380
1381 if (buflen < 1)
1382 goto Elong;
1383 /* Get '/' right */
1384 retval = end-1;
1385 *retval = '/';
1386
1387 for (;;) {
1388 struct dentry * parent;
1389
1390 if (dentry == root && vfsmnt == rootmnt)
1391 break;
1392 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1393 /* Global root? */
1394 spin_lock(&vfsmount_lock);
1395 if (vfsmnt->mnt_parent == vfsmnt) {
1396 spin_unlock(&vfsmount_lock);
1397 goto global_root;
1398 }
1399 dentry = vfsmnt->mnt_mountpoint;
1400 vfsmnt = vfsmnt->mnt_parent;
1401 spin_unlock(&vfsmount_lock);
1402 continue;
1403 }
1404 parent = dentry->d_parent;
1405 prefetch(parent);
1406 namelen = dentry->d_name.len;
1407 buflen -= namelen + 1;
1408 if (buflen < 0)
1409 goto Elong;
1410 end -= namelen;
1411 memcpy(end, dentry->d_name.name, namelen);
1412 *--end = '/';
1413 retval = end;
1414 dentry = parent;
1415 }
1416
1417 return retval;
1418
1419global_root:
1420 namelen = dentry->d_name.len;
1421 buflen -= namelen;
1422 if (buflen < 0)
1423 goto Elong;
1424 retval -= namelen-1; /* hit the slash */
1425 memcpy(retval, dentry->d_name.name, namelen);
1426 return retval;
1427Elong:
1428 return ERR_PTR(-ENAMETOOLONG);
1429}
1430
1431/* write full pathname into buffer and return start of pathname */
1432char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1433 char *buf, int buflen)
1434{
1435 char *res;
1436 struct vfsmount *rootmnt;
1437 struct dentry *root;
1438
1439 read_lock(&current->fs->lock);
1440 rootmnt = mntget(current->fs->rootmnt);
1441 root = dget(current->fs->root);
1442 read_unlock(&current->fs->lock);
1443 spin_lock(&dcache_lock);
1444 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1445 spin_unlock(&dcache_lock);
1446 dput(root);
1447 mntput(rootmnt);
1448 return res;
1449}
1450
1451/*
1452 * NOTE! The user-level library version returns a
1453 * character pointer. The kernel system call just
1454 * returns the length of the buffer filled (which
1455 * includes the ending '\0' character), or a negative
1456 * error value. So libc would do something like
1457 *
1458 * char *getcwd(char * buf, size_t size)
1459 * {
1460 * int retval;
1461 *
1462 * retval = sys_getcwd(buf, size);
1463 * if (retval >= 0)
1464 * return buf;
1465 * errno = -retval;
1466 * return NULL;
1467 * }
1468 */
1469asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1470{
1471 int error;
1472 struct vfsmount *pwdmnt, *rootmnt;
1473 struct dentry *pwd, *root;
1474 char *page = (char *) __get_free_page(GFP_USER);
1475
1476 if (!page)
1477 return -ENOMEM;
1478
1479 read_lock(&current->fs->lock);
1480 pwdmnt = mntget(current->fs->pwdmnt);
1481 pwd = dget(current->fs->pwd);
1482 rootmnt = mntget(current->fs->rootmnt);
1483 root = dget(current->fs->root);
1484 read_unlock(&current->fs->lock);
1485
1486 error = -ENOENT;
1487 /* Has the current directory has been unlinked? */
1488 spin_lock(&dcache_lock);
1489 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1490 unsigned long len;
1491 char * cwd;
1492
1493 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1494 spin_unlock(&dcache_lock);
1495
1496 error = PTR_ERR(cwd);
1497 if (IS_ERR(cwd))
1498 goto out;
1499
1500 error = -ERANGE;
1501 len = PAGE_SIZE + page - cwd;
1502 if (len <= size) {
1503 error = len;
1504 if (copy_to_user(buf, cwd, len))
1505 error = -EFAULT;
1506 }
1507 } else
1508 spin_unlock(&dcache_lock);
1509
1510out:
1511 dput(pwd);
1512 mntput(pwdmnt);
1513 dput(root);
1514 mntput(rootmnt);
1515 free_page((unsigned long) page);
1516 return error;
1517}
1518
1519/*
1520 * Test whether new_dentry is a subdirectory of old_dentry.
1521 *
1522 * Trivially implemented using the dcache structure
1523 */
1524
1525/**
1526 * is_subdir - is new dentry a subdirectory of old_dentry
1527 * @new_dentry: new dentry
1528 * @old_dentry: old dentry
1529 *
1530 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1531 * Returns 0 otherwise.
1532 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1533 */
1534
1535int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1536{
1537 int result;
1538 struct dentry * saved = new_dentry;
1539 unsigned long seq;
1540
1541 /* need rcu_readlock to protect against the d_parent trashing due to
1542 * d_move
1543 */
1544 rcu_read_lock();
1545 do {
1546 /* for restarting inner loop in case of seq retry */
1547 new_dentry = saved;
1548 result = 0;
1549 seq = read_seqbegin(&rename_lock);
1550 for (;;) {
1551 if (new_dentry != old_dentry) {
1552 struct dentry * parent = new_dentry->d_parent;
1553 if (parent == new_dentry)
1554 break;
1555 new_dentry = parent;
1556 continue;
1557 }
1558 result = 1;
1559 break;
1560 }
1561 } while (read_seqretry(&rename_lock, seq));
1562 rcu_read_unlock();
1563
1564 return result;
1565}
1566
1567void d_genocide(struct dentry *root)
1568{
1569 struct dentry *this_parent = root;
1570 struct list_head *next;
1571
1572 spin_lock(&dcache_lock);
1573repeat:
1574 next = this_parent->d_subdirs.next;
1575resume:
1576 while (next != &this_parent->d_subdirs) {
1577 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001578 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 next = tmp->next;
1580 if (d_unhashed(dentry)||!dentry->d_inode)
1581 continue;
1582 if (!list_empty(&dentry->d_subdirs)) {
1583 this_parent = dentry;
1584 goto repeat;
1585 }
1586 atomic_dec(&dentry->d_count);
1587 }
1588 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08001589 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 atomic_dec(&this_parent->d_count);
1591 this_parent = this_parent->d_parent;
1592 goto resume;
1593 }
1594 spin_unlock(&dcache_lock);
1595}
1596
1597/**
1598 * find_inode_number - check for dentry with name
1599 * @dir: directory to check
1600 * @name: Name to find.
1601 *
1602 * Check whether a dentry already exists for the given name,
1603 * and return the inode number if it has an inode. Otherwise
1604 * 0 is returned.
1605 *
1606 * This routine is used to post-process directory listings for
1607 * filesystems using synthetic inode numbers, and is necessary
1608 * to keep getcwd() working.
1609 */
1610
1611ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1612{
1613 struct dentry * dentry;
1614 ino_t ino = 0;
1615
1616 /*
1617 * Check for a fs-specific hash function. Note that we must
1618 * calculate the standard hash first, as the d_op->d_hash()
1619 * routine may choose to leave the hash value unchanged.
1620 */
1621 name->hash = full_name_hash(name->name, name->len);
1622 if (dir->d_op && dir->d_op->d_hash)
1623 {
1624 if (dir->d_op->d_hash(dir, name) != 0)
1625 goto out;
1626 }
1627
1628 dentry = d_lookup(dir, name);
1629 if (dentry)
1630 {
1631 if (dentry->d_inode)
1632 ino = dentry->d_inode->i_ino;
1633 dput(dentry);
1634 }
1635out:
1636 return ino;
1637}
1638
1639static __initdata unsigned long dhash_entries;
1640static int __init set_dhash_entries(char *str)
1641{
1642 if (!str)
1643 return 0;
1644 dhash_entries = simple_strtoul(str, &str, 0);
1645 return 1;
1646}
1647__setup("dhash_entries=", set_dhash_entries);
1648
1649static void __init dcache_init_early(void)
1650{
1651 int loop;
1652
1653 /* If hashes are distributed across NUMA nodes, defer
1654 * hash allocation until vmalloc space is available.
1655 */
1656 if (hashdist)
1657 return;
1658
1659 dentry_hashtable =
1660 alloc_large_system_hash("Dentry cache",
1661 sizeof(struct hlist_head),
1662 dhash_entries,
1663 13,
1664 HASH_EARLY,
1665 &d_hash_shift,
1666 &d_hash_mask,
1667 0);
1668
1669 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1670 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1671}
1672
1673static void __init dcache_init(unsigned long mempages)
1674{
1675 int loop;
1676
1677 /*
1678 * A constructor could be added for stable state like the lists,
1679 * but it is probably not worth it because of the cache nature
1680 * of the dcache.
1681 */
1682 dentry_cache = kmem_cache_create("dentry_cache",
1683 sizeof(struct dentry),
1684 0,
1685 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC,
1686 NULL, NULL);
1687
1688 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1689
1690 /* Hash may have been set up in dcache_init_early */
1691 if (!hashdist)
1692 return;
1693
1694 dentry_hashtable =
1695 alloc_large_system_hash("Dentry cache",
1696 sizeof(struct hlist_head),
1697 dhash_entries,
1698 13,
1699 0,
1700 &d_hash_shift,
1701 &d_hash_mask,
1702 0);
1703
1704 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1705 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1706}
1707
1708/* SLAB cache for __getname() consumers */
1709kmem_cache_t *names_cachep;
1710
1711/* SLAB cache for file structures */
1712kmem_cache_t *filp_cachep;
1713
1714EXPORT_SYMBOL(d_genocide);
1715
1716extern void bdev_cache_init(void);
1717extern void chrdev_init(void);
1718
1719void __init vfs_caches_init_early(void)
1720{
1721 dcache_init_early();
1722 inode_init_early();
1723}
1724
1725void __init vfs_caches_init(unsigned long mempages)
1726{
1727 unsigned long reserve;
1728
1729 /* Base hash sizes on available memory, with a reserve equal to
1730 150% of current kernel size */
1731
1732 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1733 mempages -= reserve;
1734
1735 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1736 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1737
1738 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -08001739 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
1741 dcache_init(mempages);
1742 inode_init(mempages);
1743 files_init(mempages);
1744 mnt_init(mempages);
1745 bdev_cache_init();
1746 chrdev_init();
1747}
1748
1749EXPORT_SYMBOL(d_alloc);
1750EXPORT_SYMBOL(d_alloc_anon);
1751EXPORT_SYMBOL(d_alloc_root);
1752EXPORT_SYMBOL(d_delete);
1753EXPORT_SYMBOL(d_find_alias);
1754EXPORT_SYMBOL(d_instantiate);
1755EXPORT_SYMBOL(d_invalidate);
1756EXPORT_SYMBOL(d_lookup);
1757EXPORT_SYMBOL(d_move);
1758EXPORT_SYMBOL(d_path);
1759EXPORT_SYMBOL(d_prune_aliases);
1760EXPORT_SYMBOL(d_rehash);
1761EXPORT_SYMBOL(d_splice_alias);
1762EXPORT_SYMBOL(d_validate);
1763EXPORT_SYMBOL(dget_locked);
1764EXPORT_SYMBOL(dput);
1765EXPORT_SYMBOL(find_inode_number);
1766EXPORT_SYMBOL(have_submounts);
1767EXPORT_SYMBOL(names_cachep);
1768EXPORT_SYMBOL(shrink_dcache_parent);
1769EXPORT_SYMBOL(shrink_dcache_sb);