blob: 19458d3995024123c226e4de7a950ff71319462b [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Eric Dumazetfa3536c2006-03-26 01:37:24 -080038int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
40
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
Adrian Bunk75c96f82005-05-05 16:16:09 -070042static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44EXPORT_SYMBOL(dcache_lock);
45
Eric Dumazetfa3536c2006-03-26 01:37:24 -080046static kmem_cache_t *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
49
50/*
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
54 *
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
57 */
58#define D_HASHBITS d_hash_shift
59#define D_HASHMASK d_hash_mask
60
Eric Dumazetfa3536c2006-03-26 01:37:24 -080061static unsigned int d_hash_mask __read_mostly;
62static unsigned int d_hash_shift __read_mostly;
63static struct hlist_head *dentry_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064static LIST_HEAD(dentry_unused);
65
66/* Statistics gathering. */
67struct dentry_stat_t dentry_stat = {
68 .age_limit = 45,
69};
70
71static void d_callback(struct rcu_head *head)
72{
Eric Dumazet5160ee62006-01-08 01:03:32 -080073 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75 if (dname_external(dentry))
76 kfree(dentry->d_name.name);
77 kmem_cache_free(dentry_cache, dentry);
78}
79
80/*
81 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
82 * inside dcache_lock.
83 */
84static void d_free(struct dentry *dentry)
85{
86 if (dentry->d_op && dentry->d_op->d_release)
87 dentry->d_op->d_release(dentry);
Eric Dumazet5160ee62006-01-08 01:03:32 -080088 call_rcu(&dentry->d_u.d_rcu, d_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089}
90
91/*
92 * Release the dentry's inode, using the filesystem
93 * d_iput() operation if defined.
94 * Called with dcache_lock and per dentry lock held, drops both.
95 */
Arjan van de Ven858119e2006-01-14 13:20:43 -080096static void dentry_iput(struct dentry * dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097{
98 struct inode *inode = dentry->d_inode;
99 if (inode) {
100 dentry->d_inode = NULL;
101 list_del_init(&dentry->d_alias);
102 spin_unlock(&dentry->d_lock);
103 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700104 if (!inode->i_nlink)
105 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 if (dentry->d_op && dentry->d_op->d_iput)
107 dentry->d_op->d_iput(dentry, inode);
108 else
109 iput(inode);
110 } else {
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
113 }
114}
115
116/*
117 * This is dput
118 *
119 * This is complicated by the fact that we do not want to put
120 * dentries that are no longer on any hash chain on the unused
121 * list: we'd much rather just get rid of them immediately.
122 *
123 * However, that implies that we have to traverse the dentry
124 * tree upwards to the parents which might _also_ now be
125 * scheduled for deletion (it may have been only waiting for
126 * its last child to go away).
127 *
128 * This tail recursion is done by hand as we don't want to depend
129 * on the compiler to always get this right (gcc generally doesn't).
130 * Real recursion would eat up our stack space.
131 */
132
133/*
134 * dput - release a dentry
135 * @dentry: dentry to release
136 *
137 * Release a dentry. This will drop the usage count and if appropriate
138 * call the dentry unlink method as well as removing it from the queues and
139 * releasing its resources. If the parent dentries were scheduled for release
140 * they too may now get deleted.
141 *
142 * no dcache lock, please.
143 */
144
145void dput(struct dentry *dentry)
146{
147 if (!dentry)
148 return;
149
150repeat:
151 if (atomic_read(&dentry->d_count) == 1)
152 might_sleep();
153 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
154 return;
155
156 spin_lock(&dentry->d_lock);
157 if (atomic_read(&dentry->d_count)) {
158 spin_unlock(&dentry->d_lock);
159 spin_unlock(&dcache_lock);
160 return;
161 }
162
163 /*
164 * AV: ->d_delete() is _NOT_ allowed to block now.
165 */
166 if (dentry->d_op && dentry->d_op->d_delete) {
167 if (dentry->d_op->d_delete(dentry))
168 goto unhash_it;
169 }
170 /* Unreachable? Get rid of it */
171 if (d_unhashed(dentry))
172 goto kill_it;
173 if (list_empty(&dentry->d_lru)) {
174 dentry->d_flags |= DCACHE_REFERENCED;
175 list_add(&dentry->d_lru, &dentry_unused);
176 dentry_stat.nr_unused++;
177 }
178 spin_unlock(&dentry->d_lock);
179 spin_unlock(&dcache_lock);
180 return;
181
182unhash_it:
183 __d_drop(dentry);
184
185kill_it: {
186 struct dentry *parent;
187
188 /* If dentry was on d_lru list
189 * delete it from there
190 */
191 if (!list_empty(&dentry->d_lru)) {
192 list_del(&dentry->d_lru);
193 dentry_stat.nr_unused--;
194 }
Eric Dumazet5160ee62006-01-08 01:03:32 -0800195 list_del(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 dentry_stat.nr_dentry--; /* For d_free, below */
197 /*drops the locks, at that point nobody can reach this dentry */
198 dentry_iput(dentry);
199 parent = dentry->d_parent;
200 d_free(dentry);
201 if (dentry == parent)
202 return;
203 dentry = parent;
204 goto repeat;
205 }
206}
207
208/**
209 * d_invalidate - invalidate a dentry
210 * @dentry: dentry to invalidate
211 *
212 * Try to invalidate the dentry if it turns out to be
213 * possible. If there are other dentries that can be
214 * reached through this one we can't delete it and we
215 * return -EBUSY. On success we return 0.
216 *
217 * no dcache lock.
218 */
219
220int d_invalidate(struct dentry * dentry)
221{
222 /*
223 * If it's already been dropped, return OK.
224 */
225 spin_lock(&dcache_lock);
226 if (d_unhashed(dentry)) {
227 spin_unlock(&dcache_lock);
228 return 0;
229 }
230 /*
231 * Check whether to do a partial shrink_dcache
232 * to get rid of unused child entries.
233 */
234 if (!list_empty(&dentry->d_subdirs)) {
235 spin_unlock(&dcache_lock);
236 shrink_dcache_parent(dentry);
237 spin_lock(&dcache_lock);
238 }
239
240 /*
241 * Somebody else still using it?
242 *
243 * If it's a directory, we can't drop it
244 * for fear of somebody re-populating it
245 * with children (even though dropping it
246 * would make it unreachable from the root,
247 * we might still populate it if it was a
248 * working directory or similar).
249 */
250 spin_lock(&dentry->d_lock);
251 if (atomic_read(&dentry->d_count) > 1) {
252 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
253 spin_unlock(&dentry->d_lock);
254 spin_unlock(&dcache_lock);
255 return -EBUSY;
256 }
257 }
258
259 __d_drop(dentry);
260 spin_unlock(&dentry->d_lock);
261 spin_unlock(&dcache_lock);
262 return 0;
263}
264
265/* This should be called _only_ with dcache_lock held */
266
267static inline struct dentry * __dget_locked(struct dentry *dentry)
268{
269 atomic_inc(&dentry->d_count);
270 if (!list_empty(&dentry->d_lru)) {
271 dentry_stat.nr_unused--;
272 list_del_init(&dentry->d_lru);
273 }
274 return dentry;
275}
276
277struct dentry * dget_locked(struct dentry *dentry)
278{
279 return __dget_locked(dentry);
280}
281
282/**
283 * d_find_alias - grab a hashed alias of inode
284 * @inode: inode in question
285 * @want_discon: flag, used by d_splice_alias, to request
286 * that only a DISCONNECTED alias be returned.
287 *
288 * If inode has a hashed alias, or is a directory and has any alias,
289 * acquire the reference to alias and return it. Otherwise return NULL.
290 * Notice that if inode is a directory there can be only one alias and
291 * it can be unhashed only if it has no children, or if it is the root
292 * of a filesystem.
293 *
294 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
295 * any other hashed alias over that one unless @want_discon is set,
296 * in which case only return a DCACHE_DISCONNECTED alias.
297 */
298
299static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
300{
301 struct list_head *head, *next, *tmp;
302 struct dentry *alias, *discon_alias=NULL;
303
304 head = &inode->i_dentry;
305 next = inode->i_dentry.next;
306 while (next != head) {
307 tmp = next;
308 next = tmp->next;
309 prefetch(next);
310 alias = list_entry(tmp, struct dentry, d_alias);
311 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
312 if (alias->d_flags & DCACHE_DISCONNECTED)
313 discon_alias = alias;
314 else if (!want_discon) {
315 __dget_locked(alias);
316 return alias;
317 }
318 }
319 }
320 if (discon_alias)
321 __dget_locked(discon_alias);
322 return discon_alias;
323}
324
325struct dentry * d_find_alias(struct inode *inode)
326{
David Howells214fda12006-03-25 03:06:36 -0800327 struct dentry *de = NULL;
328
329 if (!list_empty(&inode->i_dentry)) {
330 spin_lock(&dcache_lock);
331 de = __d_find_alias(inode, 0);
332 spin_unlock(&dcache_lock);
333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 return de;
335}
336
337/*
338 * Try to kill dentries associated with this inode.
339 * WARNING: you must own a reference to inode.
340 */
341void d_prune_aliases(struct inode *inode)
342{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700343 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344restart:
345 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700346 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 spin_lock(&dentry->d_lock);
348 if (!atomic_read(&dentry->d_count)) {
349 __dget_locked(dentry);
350 __d_drop(dentry);
351 spin_unlock(&dentry->d_lock);
352 spin_unlock(&dcache_lock);
353 dput(dentry);
354 goto restart;
355 }
356 spin_unlock(&dentry->d_lock);
357 }
358 spin_unlock(&dcache_lock);
359}
360
361/*
362 * Throw away a dentry - free the inode, dput the parent.
363 * This requires that the LRU list has already been
364 * removed.
365 * Called with dcache_lock, drops it and then regains.
366 */
367static inline void prune_one_dentry(struct dentry * dentry)
368{
369 struct dentry * parent;
370
371 __d_drop(dentry);
Eric Dumazet5160ee62006-01-08 01:03:32 -0800372 list_del(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 dentry_stat.nr_dentry--; /* For d_free, below */
374 dentry_iput(dentry);
375 parent = dentry->d_parent;
376 d_free(dentry);
377 if (parent != dentry)
378 dput(parent);
379 spin_lock(&dcache_lock);
380}
381
382/**
383 * prune_dcache - shrink the dcache
384 * @count: number of entries to try and free
385 *
386 * Shrink the dcache. This is done when we need
387 * more memory, or simply when we need to unmount
388 * something (at which point we need to unuse
389 * all dentries).
390 *
391 * This function may fail to free any resources if
392 * all the dentries are in use.
393 */
394
395static void prune_dcache(int count)
396{
397 spin_lock(&dcache_lock);
398 for (; count ; count--) {
399 struct dentry *dentry;
400 struct list_head *tmp;
401
402 cond_resched_lock(&dcache_lock);
403
404 tmp = dentry_unused.prev;
405 if (tmp == &dentry_unused)
406 break;
407 list_del_init(tmp);
408 prefetch(dentry_unused.prev);
409 dentry_stat.nr_unused--;
410 dentry = list_entry(tmp, struct dentry, d_lru);
411
412 spin_lock(&dentry->d_lock);
413 /*
414 * We found an inuse dentry which was not removed from
415 * dentry_unused because of laziness during lookup. Do not free
416 * it - just keep it off the dentry_unused list.
417 */
418 if (atomic_read(&dentry->d_count)) {
419 spin_unlock(&dentry->d_lock);
420 continue;
421 }
422 /* If the dentry was recently referenced, don't free it. */
423 if (dentry->d_flags & DCACHE_REFERENCED) {
424 dentry->d_flags &= ~DCACHE_REFERENCED;
425 list_add(&dentry->d_lru, &dentry_unused);
426 dentry_stat.nr_unused++;
427 spin_unlock(&dentry->d_lock);
428 continue;
429 }
430 prune_one_dentry(dentry);
431 }
432 spin_unlock(&dcache_lock);
433}
434
435/*
436 * Shrink the dcache for the specified super block.
437 * This allows us to unmount a device without disturbing
438 * the dcache for the other devices.
439 *
440 * This implementation makes just two traversals of the
441 * unused list. On the first pass we move the selected
442 * dentries to the most recent end, and on the second
443 * pass we free them. The second pass must restart after
444 * each dput(), but since the target dentries are all at
445 * the end, it's really just a single traversal.
446 */
447
448/**
449 * shrink_dcache_sb - shrink dcache for a superblock
450 * @sb: superblock
451 *
452 * Shrink the dcache for the specified super block. This
453 * is used to free the dcache before unmounting a file
454 * system
455 */
456
457void shrink_dcache_sb(struct super_block * sb)
458{
459 struct list_head *tmp, *next;
460 struct dentry *dentry;
461
462 /*
463 * Pass one ... move the dentries for the specified
464 * superblock to the most recent end of the unused list.
465 */
466 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700467 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 dentry = list_entry(tmp, struct dentry, d_lru);
469 if (dentry->d_sb != sb)
470 continue;
471 list_del(tmp);
472 list_add(tmp, &dentry_unused);
473 }
474
475 /*
476 * Pass two ... free the dentries for this superblock.
477 */
478repeat:
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700479 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 dentry = list_entry(tmp, struct dentry, d_lru);
481 if (dentry->d_sb != sb)
482 continue;
483 dentry_stat.nr_unused--;
484 list_del_init(tmp);
485 spin_lock(&dentry->d_lock);
486 if (atomic_read(&dentry->d_count)) {
487 spin_unlock(&dentry->d_lock);
488 continue;
489 }
490 prune_one_dentry(dentry);
Kirill Korotaev2ab13462006-03-25 03:07:45 -0800491 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 goto repeat;
493 }
494 spin_unlock(&dcache_lock);
495}
496
497/*
498 * Search for at least 1 mount point in the dentry's subdirs.
499 * We descend to the next level whenever the d_subdirs
500 * list is non-empty and continue searching.
501 */
502
503/**
504 * have_submounts - check for mounts over a dentry
505 * @parent: dentry to check.
506 *
507 * Return true if the parent or its subdirectories contain
508 * a mount point
509 */
510
511int have_submounts(struct dentry *parent)
512{
513 struct dentry *this_parent = parent;
514 struct list_head *next;
515
516 spin_lock(&dcache_lock);
517 if (d_mountpoint(parent))
518 goto positive;
519repeat:
520 next = this_parent->d_subdirs.next;
521resume:
522 while (next != &this_parent->d_subdirs) {
523 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800524 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 next = tmp->next;
526 /* Have we found a mount point ? */
527 if (d_mountpoint(dentry))
528 goto positive;
529 if (!list_empty(&dentry->d_subdirs)) {
530 this_parent = dentry;
531 goto repeat;
532 }
533 }
534 /*
535 * All done at this level ... ascend and resume the search.
536 */
537 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800538 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 this_parent = this_parent->d_parent;
540 goto resume;
541 }
542 spin_unlock(&dcache_lock);
543 return 0; /* No mount points found in tree */
544positive:
545 spin_unlock(&dcache_lock);
546 return 1;
547}
548
549/*
550 * Search the dentry child list for the specified parent,
551 * and move any unused dentries to the end of the unused
552 * list for prune_dcache(). We descend to the next level
553 * whenever the d_subdirs list is non-empty and continue
554 * searching.
555 *
556 * It returns zero iff there are no unused children,
557 * otherwise it returns the number of children moved to
558 * the end of the unused list. This may not be the total
559 * number of unused children, because select_parent can
560 * drop the lock and return early due to latency
561 * constraints.
562 */
563static int select_parent(struct dentry * parent)
564{
565 struct dentry *this_parent = parent;
566 struct list_head *next;
567 int found = 0;
568
569 spin_lock(&dcache_lock);
570repeat:
571 next = this_parent->d_subdirs.next;
572resume:
573 while (next != &this_parent->d_subdirs) {
574 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800575 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 next = tmp->next;
577
578 if (!list_empty(&dentry->d_lru)) {
579 dentry_stat.nr_unused--;
580 list_del_init(&dentry->d_lru);
581 }
582 /*
583 * move only zero ref count dentries to the end
584 * of the unused list for prune_dcache
585 */
586 if (!atomic_read(&dentry->d_count)) {
587 list_add(&dentry->d_lru, dentry_unused.prev);
588 dentry_stat.nr_unused++;
589 found++;
590 }
591
592 /*
593 * We can return to the caller if we have found some (this
594 * ensures forward progress). We'll be coming back to find
595 * the rest.
596 */
597 if (found && need_resched())
598 goto out;
599
600 /*
601 * Descend a level if the d_subdirs list is non-empty.
602 */
603 if (!list_empty(&dentry->d_subdirs)) {
604 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 goto repeat;
606 }
607 }
608 /*
609 * All done at this level ... ascend and resume the search.
610 */
611 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800612 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 goto resume;
615 }
616out:
617 spin_unlock(&dcache_lock);
618 return found;
619}
620
621/**
622 * shrink_dcache_parent - prune dcache
623 * @parent: parent of entries to prune
624 *
625 * Prune the dcache to remove unused children of the parent dentry.
626 */
627
628void shrink_dcache_parent(struct dentry * parent)
629{
630 int found;
631
632 while ((found = select_parent(parent)) != 0)
633 prune_dcache(found);
634}
635
636/**
637 * shrink_dcache_anon - further prune the cache
638 * @head: head of d_hash list of dentries to prune
639 *
640 * Prune the dentries that are anonymous
641 *
Paul E. McKenney665a7582005-11-07 00:59:17 -0800642 * parsing d_hash list does not hlist_for_each_entry_rcu() as it
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 * done under dcache_lock.
644 *
645 */
646void shrink_dcache_anon(struct hlist_head *head)
647{
648 struct hlist_node *lp;
649 int found;
650 do {
651 found = 0;
652 spin_lock(&dcache_lock);
653 hlist_for_each(lp, head) {
654 struct dentry *this = hlist_entry(lp, struct dentry, d_hash);
655 if (!list_empty(&this->d_lru)) {
656 dentry_stat.nr_unused--;
657 list_del_init(&this->d_lru);
658 }
659
660 /*
661 * move only zero ref count dentries to the end
662 * of the unused list for prune_dcache
663 */
664 if (!atomic_read(&this->d_count)) {
665 list_add_tail(&this->d_lru, &dentry_unused);
666 dentry_stat.nr_unused++;
667 found++;
668 }
669 }
670 spin_unlock(&dcache_lock);
671 prune_dcache(found);
672 } while(found);
673}
674
675/*
676 * Scan `nr' dentries and return the number which remain.
677 *
678 * We need to avoid reentering the filesystem if the caller is performing a
679 * GFP_NOFS allocation attempt. One example deadlock is:
680 *
681 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
682 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
683 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
684 *
685 * In this case we return -1 to tell the caller that we baled.
686 */
Al Viro27496a82005-10-21 03:20:48 -0400687static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
689 if (nr) {
690 if (!(gfp_mask & __GFP_FS))
691 return -1;
692 prune_dcache(nr);
693 }
694 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
695}
696
697/**
698 * d_alloc - allocate a dcache entry
699 * @parent: parent of entry to allocate
700 * @name: qstr of the name
701 *
702 * Allocates a dentry. It returns %NULL if there is insufficient memory
703 * available. On a success the dentry is returned. The name passed in is
704 * copied and the copy passed in may be reused after this call.
705 */
706
707struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
708{
709 struct dentry *dentry;
710 char *dname;
711
712 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
713 if (!dentry)
714 return NULL;
715
716 if (name->len > DNAME_INLINE_LEN-1) {
717 dname = kmalloc(name->len + 1, GFP_KERNEL);
718 if (!dname) {
719 kmem_cache_free(dentry_cache, dentry);
720 return NULL;
721 }
722 } else {
723 dname = dentry->d_iname;
724 }
725 dentry->d_name.name = dname;
726
727 dentry->d_name.len = name->len;
728 dentry->d_name.hash = name->hash;
729 memcpy(dname, name->name, name->len);
730 dname[name->len] = 0;
731
732 atomic_set(&dentry->d_count, 1);
733 dentry->d_flags = DCACHE_UNHASHED;
734 spin_lock_init(&dentry->d_lock);
735 dentry->d_inode = NULL;
736 dentry->d_parent = NULL;
737 dentry->d_sb = NULL;
738 dentry->d_op = NULL;
739 dentry->d_fsdata = NULL;
740 dentry->d_mounted = 0;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800741#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800743#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 INIT_HLIST_NODE(&dentry->d_hash);
745 INIT_LIST_HEAD(&dentry->d_lru);
746 INIT_LIST_HEAD(&dentry->d_subdirs);
747 INIT_LIST_HEAD(&dentry->d_alias);
748
749 if (parent) {
750 dentry->d_parent = dget(parent);
751 dentry->d_sb = parent->d_sb;
752 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800753 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 }
755
756 spin_lock(&dcache_lock);
757 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800758 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 dentry_stat.nr_dentry++;
760 spin_unlock(&dcache_lock);
761
762 return dentry;
763}
764
765struct dentry *d_alloc_name(struct dentry *parent, const char *name)
766{
767 struct qstr q;
768
769 q.name = name;
770 q.len = strlen(name);
771 q.hash = full_name_hash(q.name, q.len);
772 return d_alloc(parent, &q);
773}
774
775/**
776 * d_instantiate - fill in inode information for a dentry
777 * @entry: dentry to complete
778 * @inode: inode to attach to this dentry
779 *
780 * Fill in inode information in the entry.
781 *
782 * This turns negative dentries into productive full members
783 * of society.
784 *
785 * NOTE! This assumes that the inode count has been incremented
786 * (or otherwise set) by the caller to indicate that it is now
787 * in use by the dcache.
788 */
789
790void d_instantiate(struct dentry *entry, struct inode * inode)
791{
Eric Sesterhenn28133c72006-03-26 18:25:39 +0200792 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 spin_lock(&dcache_lock);
794 if (inode)
795 list_add(&entry->d_alias, &inode->i_dentry);
796 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800797 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 spin_unlock(&dcache_lock);
799 security_d_instantiate(entry, inode);
800}
801
802/**
803 * d_instantiate_unique - instantiate a non-aliased dentry
804 * @entry: dentry to instantiate
805 * @inode: inode to attach to this dentry
806 *
807 * Fill in inode information in the entry. On success, it returns NULL.
808 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -0800809 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 *
811 * Note that in order to avoid conflicts with rename() etc, the caller
812 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -0800813 *
814 * This also assumes that the inode count has been incremented
815 * (or otherwise set) by the caller to indicate that it is now
816 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 */
818struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
819{
820 struct dentry *alias;
821 int len = entry->d_name.len;
822 const char *name = entry->d_name.name;
823 unsigned int hash = entry->d_name.hash;
824
825 BUG_ON(!list_empty(&entry->d_alias));
826 spin_lock(&dcache_lock);
827 if (!inode)
828 goto do_negative;
829 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
830 struct qstr *qstr = &alias->d_name;
831
832 if (qstr->hash != hash)
833 continue;
834 if (alias->d_parent != entry->d_parent)
835 continue;
836 if (qstr->len != len)
837 continue;
838 if (memcmp(qstr->name, name, len))
839 continue;
840 dget_locked(alias);
841 spin_unlock(&dcache_lock);
842 BUG_ON(!d_unhashed(alias));
Oleg Drokine866cfa2006-01-09 20:52:51 -0800843 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 return alias;
845 }
846 list_add(&entry->d_alias, &inode->i_dentry);
847do_negative:
848 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800849 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 spin_unlock(&dcache_lock);
851 security_d_instantiate(entry, inode);
852 return NULL;
853}
854EXPORT_SYMBOL(d_instantiate_unique);
855
856/**
857 * d_alloc_root - allocate root dentry
858 * @root_inode: inode to allocate the root for
859 *
860 * Allocate a root ("/") dentry for the inode given. The inode is
861 * instantiated and returned. %NULL is returned if there is insufficient
862 * memory or the inode passed is %NULL.
863 */
864
865struct dentry * d_alloc_root(struct inode * root_inode)
866{
867 struct dentry *res = NULL;
868
869 if (root_inode) {
870 static const struct qstr name = { .name = "/", .len = 1 };
871
872 res = d_alloc(NULL, &name);
873 if (res) {
874 res->d_sb = root_inode->i_sb;
875 res->d_parent = res;
876 d_instantiate(res, root_inode);
877 }
878 }
879 return res;
880}
881
882static inline struct hlist_head *d_hash(struct dentry *parent,
883 unsigned long hash)
884{
885 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
886 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
887 return dentry_hashtable + (hash & D_HASHMASK);
888}
889
890/**
891 * d_alloc_anon - allocate an anonymous dentry
892 * @inode: inode to allocate the dentry for
893 *
894 * This is similar to d_alloc_root. It is used by filesystems when
895 * creating a dentry for a given inode, often in the process of
896 * mapping a filehandle to a dentry. The returned dentry may be
897 * anonymous, or may have a full name (if the inode was already
898 * in the cache). The file system may need to make further
899 * efforts to connect this dentry into the dcache properly.
900 *
901 * When called on a directory inode, we must ensure that
902 * the inode only ever has one dentry. If a dentry is
903 * found, that is returned instead of allocating a new one.
904 *
905 * On successful return, the reference to the inode has been transferred
906 * to the dentry. If %NULL is returned (indicating kmalloc failure),
907 * the reference on the inode has not been released.
908 */
909
910struct dentry * d_alloc_anon(struct inode *inode)
911{
912 static const struct qstr anonstring = { .name = "" };
913 struct dentry *tmp;
914 struct dentry *res;
915
916 if ((res = d_find_alias(inode))) {
917 iput(inode);
918 return res;
919 }
920
921 tmp = d_alloc(NULL, &anonstring);
922 if (!tmp)
923 return NULL;
924
925 tmp->d_parent = tmp; /* make sure dput doesn't croak */
926
927 spin_lock(&dcache_lock);
928 res = __d_find_alias(inode, 0);
929 if (!res) {
930 /* attach a disconnected dentry */
931 res = tmp;
932 tmp = NULL;
933 spin_lock(&res->d_lock);
934 res->d_sb = inode->i_sb;
935 res->d_parent = res;
936 res->d_inode = inode;
937 res->d_flags |= DCACHE_DISCONNECTED;
938 res->d_flags &= ~DCACHE_UNHASHED;
939 list_add(&res->d_alias, &inode->i_dentry);
940 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
941 spin_unlock(&res->d_lock);
942
943 inode = NULL; /* don't drop reference */
944 }
945 spin_unlock(&dcache_lock);
946
947 if (inode)
948 iput(inode);
949 if (tmp)
950 dput(tmp);
951 return res;
952}
953
954
955/**
956 * d_splice_alias - splice a disconnected dentry into the tree if one exists
957 * @inode: the inode which may have a disconnected dentry
958 * @dentry: a negative dentry which we want to point to the inode.
959 *
960 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
961 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
962 * and return it, else simply d_add the inode to the dentry and return NULL.
963 *
964 * This is needed in the lookup routine of any filesystem that is exportable
965 * (via knfsd) so that we can build dcache paths to directories effectively.
966 *
967 * If a dentry was found and moved, then it is returned. Otherwise NULL
968 * is returned. This matches the expected return value of ->lookup.
969 *
970 */
971struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
972{
973 struct dentry *new = NULL;
974
975 if (inode) {
976 spin_lock(&dcache_lock);
977 new = __d_find_alias(inode, 1);
978 if (new) {
979 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Pigginc32ccd82006-03-25 03:07:09 -0800980 fsnotify_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 spin_unlock(&dcache_lock);
982 security_d_instantiate(new, inode);
983 d_rehash(dentry);
984 d_move(new, dentry);
985 iput(inode);
986 } else {
987 /* d_instantiate takes dcache_lock, so we do it by hand */
988 list_add(&dentry->d_alias, &inode->i_dentry);
989 dentry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800990 fsnotify_d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 spin_unlock(&dcache_lock);
992 security_d_instantiate(dentry, inode);
993 d_rehash(dentry);
994 }
995 } else
996 d_add(dentry, inode);
997 return new;
998}
999
1000
1001/**
1002 * d_lookup - search for a dentry
1003 * @parent: parent dentry
1004 * @name: qstr of name we wish to find
1005 *
1006 * Searches the children of the parent dentry for the name in question. If
1007 * the dentry is found its reference count is incremented and the dentry
1008 * is returned. The caller must use d_put to free the entry when it has
1009 * finished using it. %NULL is returned on failure.
1010 *
1011 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1012 * Memory barriers are used while updating and doing lockless traversal.
1013 * To avoid races with d_move while rename is happening, d_lock is used.
1014 *
1015 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1016 * and name pointer in one structure pointed by d_qstr.
1017 *
1018 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1019 * lookup is going on.
1020 *
1021 * dentry_unused list is not updated even if lookup finds the required dentry
1022 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1023 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1024 * acquisition.
1025 *
1026 * d_lookup() is protected against the concurrent renames in some unrelated
1027 * directory using the seqlockt_t rename_lock.
1028 */
1029
1030struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1031{
1032 struct dentry * dentry = NULL;
1033 unsigned long seq;
1034
1035 do {
1036 seq = read_seqbegin(&rename_lock);
1037 dentry = __d_lookup(parent, name);
1038 if (dentry)
1039 break;
1040 } while (read_seqretry(&rename_lock, seq));
1041 return dentry;
1042}
1043
1044struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1045{
1046 unsigned int len = name->len;
1047 unsigned int hash = name->hash;
1048 const unsigned char *str = name->name;
1049 struct hlist_head *head = d_hash(parent,hash);
1050 struct dentry *found = NULL;
1051 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001052 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
1054 rcu_read_lock();
1055
Paul E. McKenney665a7582005-11-07 00:59:17 -08001056 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 struct qstr *qstr;
1058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (dentry->d_name.hash != hash)
1060 continue;
1061 if (dentry->d_parent != parent)
1062 continue;
1063
1064 spin_lock(&dentry->d_lock);
1065
1066 /*
1067 * Recheck the dentry after taking the lock - d_move may have
1068 * changed things. Don't bother checking the hash because we're
1069 * about to compare the whole name anyway.
1070 */
1071 if (dentry->d_parent != parent)
1072 goto next;
1073
1074 /*
1075 * It is safe to compare names since d_move() cannot
1076 * change the qstr (protected by d_lock).
1077 */
1078 qstr = &dentry->d_name;
1079 if (parent->d_op && parent->d_op->d_compare) {
1080 if (parent->d_op->d_compare(parent, qstr, name))
1081 goto next;
1082 } else {
1083 if (qstr->len != len)
1084 goto next;
1085 if (memcmp(qstr->name, str, len))
1086 goto next;
1087 }
1088
1089 if (!d_unhashed(dentry)) {
1090 atomic_inc(&dentry->d_count);
1091 found = dentry;
1092 }
1093 spin_unlock(&dentry->d_lock);
1094 break;
1095next:
1096 spin_unlock(&dentry->d_lock);
1097 }
1098 rcu_read_unlock();
1099
1100 return found;
1101}
1102
1103/**
1104 * d_validate - verify dentry provided from insecure source
1105 * @dentry: The dentry alleged to be valid child of @dparent
1106 * @dparent: The parent dentry (known to be valid)
1107 * @hash: Hash of the dentry
1108 * @len: Length of the name
1109 *
1110 * An insecure source has sent us a dentry, here we verify it and dget() it.
1111 * This is used by ncpfs in its readdir implementation.
1112 * Zero is returned in the dentry is invalid.
1113 */
1114
1115int d_validate(struct dentry *dentry, struct dentry *dparent)
1116{
1117 struct hlist_head *base;
1118 struct hlist_node *lhp;
1119
1120 /* Check whether the ptr might be valid at all.. */
1121 if (!kmem_ptr_validate(dentry_cache, dentry))
1122 goto out;
1123
1124 if (dentry->d_parent != dparent)
1125 goto out;
1126
1127 spin_lock(&dcache_lock);
1128 base = d_hash(dparent, dentry->d_name.hash);
1129 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001130 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 * as it is parsed under dcache_lock
1132 */
1133 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1134 __dget_locked(dentry);
1135 spin_unlock(&dcache_lock);
1136 return 1;
1137 }
1138 }
1139 spin_unlock(&dcache_lock);
1140out:
1141 return 0;
1142}
1143
1144/*
1145 * When a file is deleted, we have two options:
1146 * - turn this dentry into a negative dentry
1147 * - unhash this dentry and free it.
1148 *
1149 * Usually, we want to just turn this into
1150 * a negative dentry, but if anybody else is
1151 * currently using the dentry or the inode
1152 * we can't do that and we fall back on removing
1153 * it from the hash queues and waiting for
1154 * it to be deleted later when it has no users
1155 */
1156
1157/**
1158 * d_delete - delete a dentry
1159 * @dentry: The dentry to delete
1160 *
1161 * Turn the dentry into a negative dentry if possible, otherwise
1162 * remove it from the hash queues so it can be deleted later
1163 */
1164
1165void d_delete(struct dentry * dentry)
1166{
John McCutchan7a91bf72005-08-08 13:52:16 -04001167 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 /*
1169 * Are we the only user?
1170 */
1171 spin_lock(&dcache_lock);
1172 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001173 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 if (atomic_read(&dentry->d_count) == 1) {
Nick Pigginc32ccd82006-03-25 03:07:09 -08001175 /* remove this and other inotify debug checks after 2.6.18 */
1176 dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
1177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001179 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 return;
1181 }
1182
1183 if (!d_unhashed(dentry))
1184 __d_drop(dentry);
1185
1186 spin_unlock(&dentry->d_lock);
1187 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001188
1189 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190}
1191
1192static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1193{
1194
1195 entry->d_flags &= ~DCACHE_UNHASHED;
1196 hlist_add_head_rcu(&entry->d_hash, list);
1197}
1198
1199/**
1200 * d_rehash - add an entry back to the hash
1201 * @entry: dentry to add to the hash
1202 *
1203 * Adds a dentry to the hash according to its name.
1204 */
1205
1206void d_rehash(struct dentry * entry)
1207{
1208 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1209
1210 spin_lock(&dcache_lock);
1211 spin_lock(&entry->d_lock);
1212 __d_rehash(entry, list);
1213 spin_unlock(&entry->d_lock);
1214 spin_unlock(&dcache_lock);
1215}
1216
1217#define do_switch(x,y) do { \
1218 __typeof__ (x) __tmp = x; \
1219 x = y; y = __tmp; } while (0)
1220
1221/*
1222 * When switching names, the actual string doesn't strictly have to
1223 * be preserved in the target - because we're dropping the target
1224 * anyway. As such, we can just do a simple memcpy() to copy over
1225 * the new name before we switch.
1226 *
1227 * Note that we have to be a lot more careful about getting the hash
1228 * switched - we have to switch the hash value properly even if it
1229 * then no longer matches the actual (corrupted) string of the target.
1230 * The hash value has to match the hash queue that the dentry is on..
1231 */
1232static void switch_names(struct dentry *dentry, struct dentry *target)
1233{
1234 if (dname_external(target)) {
1235 if (dname_external(dentry)) {
1236 /*
1237 * Both external: swap the pointers
1238 */
1239 do_switch(target->d_name.name, dentry->d_name.name);
1240 } else {
1241 /*
1242 * dentry:internal, target:external. Steal target's
1243 * storage and make target internal.
1244 */
1245 dentry->d_name.name = target->d_name.name;
1246 target->d_name.name = target->d_iname;
1247 }
1248 } else {
1249 if (dname_external(dentry)) {
1250 /*
1251 * dentry:external, target:internal. Give dentry's
1252 * storage to target and make dentry internal
1253 */
1254 memcpy(dentry->d_iname, target->d_name.name,
1255 target->d_name.len + 1);
1256 target->d_name.name = dentry->d_name.name;
1257 dentry->d_name.name = dentry->d_iname;
1258 } else {
1259 /*
1260 * Both are internal. Just copy target to dentry
1261 */
1262 memcpy(dentry->d_iname, target->d_name.name,
1263 target->d_name.len + 1);
1264 }
1265 }
1266}
1267
1268/*
1269 * We cannibalize "target" when moving dentry on top of it,
1270 * because it's going to be thrown away anyway. We could be more
1271 * polite about it, though.
1272 *
1273 * This forceful removal will result in ugly /proc output if
1274 * somebody holds a file open that got deleted due to a rename.
1275 * We could be nicer about the deleted file, and let it show
1276 * up under the name it got deleted rather than the name that
1277 * deleted it.
1278 */
1279
1280/**
1281 * d_move - move a dentry
1282 * @dentry: entry to move
1283 * @target: new dentry
1284 *
1285 * Update the dcache to reflect the move of a file name. Negative
1286 * dcache entries should not be moved in this way.
1287 */
1288
1289void d_move(struct dentry * dentry, struct dentry * target)
1290{
1291 struct hlist_head *list;
1292
1293 if (!dentry->d_inode)
1294 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1295
1296 spin_lock(&dcache_lock);
1297 write_seqlock(&rename_lock);
1298 /*
1299 * XXXX: do we really need to take target->d_lock?
1300 */
1301 if (target < dentry) {
1302 spin_lock(&target->d_lock);
1303 spin_lock(&dentry->d_lock);
1304 } else {
1305 spin_lock(&dentry->d_lock);
1306 spin_lock(&target->d_lock);
1307 }
1308
1309 /* Move the dentry to the target hash queue, if on different bucket */
1310 if (dentry->d_flags & DCACHE_UNHASHED)
1311 goto already_unhashed;
1312
1313 hlist_del_rcu(&dentry->d_hash);
1314
1315already_unhashed:
1316 list = d_hash(target->d_parent, target->d_name.hash);
1317 __d_rehash(dentry, list);
1318
1319 /* Unhash the target: dput() will then get rid of it */
1320 __d_drop(target);
1321
Eric Dumazet5160ee62006-01-08 01:03:32 -08001322 list_del(&dentry->d_u.d_child);
1323 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 /* Switch the names.. */
1326 switch_names(dentry, target);
1327 do_switch(dentry->d_name.len, target->d_name.len);
1328 do_switch(dentry->d_name.hash, target->d_name.hash);
1329
1330 /* ... and switch the parents */
1331 if (IS_ROOT(dentry)) {
1332 dentry->d_parent = target->d_parent;
1333 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001334 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 } else {
1336 do_switch(dentry->d_parent, target->d_parent);
1337
1338 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001339 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 }
1341
Eric Dumazet5160ee62006-01-08 01:03:32 -08001342 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001344 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 spin_unlock(&dentry->d_lock);
1346 write_sequnlock(&rename_lock);
1347 spin_unlock(&dcache_lock);
1348}
1349
1350/**
1351 * d_path - return the path of a dentry
1352 * @dentry: dentry to report
1353 * @vfsmnt: vfsmnt to which the dentry belongs
1354 * @root: root dentry
1355 * @rootmnt: vfsmnt to which the root dentry belongs
1356 * @buffer: buffer to return value in
1357 * @buflen: buffer length
1358 *
1359 * Convert a dentry into an ASCII path name. If the entry has been deleted
1360 * the string " (deleted)" is appended. Note that this is ambiguous.
1361 *
1362 * Returns the buffer or an error code if the path was too long.
1363 *
1364 * "buflen" should be positive. Caller holds the dcache_lock.
1365 */
1366static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1367 struct dentry *root, struct vfsmount *rootmnt,
1368 char *buffer, int buflen)
1369{
1370 char * end = buffer+buflen;
1371 char * retval;
1372 int namelen;
1373
1374 *--end = '\0';
1375 buflen--;
1376 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1377 buflen -= 10;
1378 end -= 10;
1379 if (buflen < 0)
1380 goto Elong;
1381 memcpy(end, " (deleted)", 10);
1382 }
1383
1384 if (buflen < 1)
1385 goto Elong;
1386 /* Get '/' right */
1387 retval = end-1;
1388 *retval = '/';
1389
1390 for (;;) {
1391 struct dentry * parent;
1392
1393 if (dentry == root && vfsmnt == rootmnt)
1394 break;
1395 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1396 /* Global root? */
1397 spin_lock(&vfsmount_lock);
1398 if (vfsmnt->mnt_parent == vfsmnt) {
1399 spin_unlock(&vfsmount_lock);
1400 goto global_root;
1401 }
1402 dentry = vfsmnt->mnt_mountpoint;
1403 vfsmnt = vfsmnt->mnt_parent;
1404 spin_unlock(&vfsmount_lock);
1405 continue;
1406 }
1407 parent = dentry->d_parent;
1408 prefetch(parent);
1409 namelen = dentry->d_name.len;
1410 buflen -= namelen + 1;
1411 if (buflen < 0)
1412 goto Elong;
1413 end -= namelen;
1414 memcpy(end, dentry->d_name.name, namelen);
1415 *--end = '/';
1416 retval = end;
1417 dentry = parent;
1418 }
1419
1420 return retval;
1421
1422global_root:
1423 namelen = dentry->d_name.len;
1424 buflen -= namelen;
1425 if (buflen < 0)
1426 goto Elong;
1427 retval -= namelen-1; /* hit the slash */
1428 memcpy(retval, dentry->d_name.name, namelen);
1429 return retval;
1430Elong:
1431 return ERR_PTR(-ENAMETOOLONG);
1432}
1433
1434/* write full pathname into buffer and return start of pathname */
1435char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1436 char *buf, int buflen)
1437{
1438 char *res;
1439 struct vfsmount *rootmnt;
1440 struct dentry *root;
1441
1442 read_lock(&current->fs->lock);
1443 rootmnt = mntget(current->fs->rootmnt);
1444 root = dget(current->fs->root);
1445 read_unlock(&current->fs->lock);
1446 spin_lock(&dcache_lock);
1447 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1448 spin_unlock(&dcache_lock);
1449 dput(root);
1450 mntput(rootmnt);
1451 return res;
1452}
1453
1454/*
1455 * NOTE! The user-level library version returns a
1456 * character pointer. The kernel system call just
1457 * returns the length of the buffer filled (which
1458 * includes the ending '\0' character), or a negative
1459 * error value. So libc would do something like
1460 *
1461 * char *getcwd(char * buf, size_t size)
1462 * {
1463 * int retval;
1464 *
1465 * retval = sys_getcwd(buf, size);
1466 * if (retval >= 0)
1467 * return buf;
1468 * errno = -retval;
1469 * return NULL;
1470 * }
1471 */
1472asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1473{
1474 int error;
1475 struct vfsmount *pwdmnt, *rootmnt;
1476 struct dentry *pwd, *root;
1477 char *page = (char *) __get_free_page(GFP_USER);
1478
1479 if (!page)
1480 return -ENOMEM;
1481
1482 read_lock(&current->fs->lock);
1483 pwdmnt = mntget(current->fs->pwdmnt);
1484 pwd = dget(current->fs->pwd);
1485 rootmnt = mntget(current->fs->rootmnt);
1486 root = dget(current->fs->root);
1487 read_unlock(&current->fs->lock);
1488
1489 error = -ENOENT;
1490 /* Has the current directory has been unlinked? */
1491 spin_lock(&dcache_lock);
1492 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1493 unsigned long len;
1494 char * cwd;
1495
1496 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1497 spin_unlock(&dcache_lock);
1498
1499 error = PTR_ERR(cwd);
1500 if (IS_ERR(cwd))
1501 goto out;
1502
1503 error = -ERANGE;
1504 len = PAGE_SIZE + page - cwd;
1505 if (len <= size) {
1506 error = len;
1507 if (copy_to_user(buf, cwd, len))
1508 error = -EFAULT;
1509 }
1510 } else
1511 spin_unlock(&dcache_lock);
1512
1513out:
1514 dput(pwd);
1515 mntput(pwdmnt);
1516 dput(root);
1517 mntput(rootmnt);
1518 free_page((unsigned long) page);
1519 return error;
1520}
1521
1522/*
1523 * Test whether new_dentry is a subdirectory of old_dentry.
1524 *
1525 * Trivially implemented using the dcache structure
1526 */
1527
1528/**
1529 * is_subdir - is new dentry a subdirectory of old_dentry
1530 * @new_dentry: new dentry
1531 * @old_dentry: old dentry
1532 *
1533 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1534 * Returns 0 otherwise.
1535 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1536 */
1537
1538int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1539{
1540 int result;
1541 struct dentry * saved = new_dentry;
1542 unsigned long seq;
1543
1544 /* need rcu_readlock to protect against the d_parent trashing due to
1545 * d_move
1546 */
1547 rcu_read_lock();
1548 do {
1549 /* for restarting inner loop in case of seq retry */
1550 new_dentry = saved;
1551 result = 0;
1552 seq = read_seqbegin(&rename_lock);
1553 for (;;) {
1554 if (new_dentry != old_dentry) {
1555 struct dentry * parent = new_dentry->d_parent;
1556 if (parent == new_dentry)
1557 break;
1558 new_dentry = parent;
1559 continue;
1560 }
1561 result = 1;
1562 break;
1563 }
1564 } while (read_seqretry(&rename_lock, seq));
1565 rcu_read_unlock();
1566
1567 return result;
1568}
1569
1570void d_genocide(struct dentry *root)
1571{
1572 struct dentry *this_parent = root;
1573 struct list_head *next;
1574
1575 spin_lock(&dcache_lock);
1576repeat:
1577 next = this_parent->d_subdirs.next;
1578resume:
1579 while (next != &this_parent->d_subdirs) {
1580 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001581 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 next = tmp->next;
1583 if (d_unhashed(dentry)||!dentry->d_inode)
1584 continue;
1585 if (!list_empty(&dentry->d_subdirs)) {
1586 this_parent = dentry;
1587 goto repeat;
1588 }
1589 atomic_dec(&dentry->d_count);
1590 }
1591 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08001592 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 atomic_dec(&this_parent->d_count);
1594 this_parent = this_parent->d_parent;
1595 goto resume;
1596 }
1597 spin_unlock(&dcache_lock);
1598}
1599
1600/**
1601 * find_inode_number - check for dentry with name
1602 * @dir: directory to check
1603 * @name: Name to find.
1604 *
1605 * Check whether a dentry already exists for the given name,
1606 * and return the inode number if it has an inode. Otherwise
1607 * 0 is returned.
1608 *
1609 * This routine is used to post-process directory listings for
1610 * filesystems using synthetic inode numbers, and is necessary
1611 * to keep getcwd() working.
1612 */
1613
1614ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1615{
1616 struct dentry * dentry;
1617 ino_t ino = 0;
1618
1619 /*
1620 * Check for a fs-specific hash function. Note that we must
1621 * calculate the standard hash first, as the d_op->d_hash()
1622 * routine may choose to leave the hash value unchanged.
1623 */
1624 name->hash = full_name_hash(name->name, name->len);
1625 if (dir->d_op && dir->d_op->d_hash)
1626 {
1627 if (dir->d_op->d_hash(dir, name) != 0)
1628 goto out;
1629 }
1630
1631 dentry = d_lookup(dir, name);
1632 if (dentry)
1633 {
1634 if (dentry->d_inode)
1635 ino = dentry->d_inode->i_ino;
1636 dput(dentry);
1637 }
1638out:
1639 return ino;
1640}
1641
1642static __initdata unsigned long dhash_entries;
1643static int __init set_dhash_entries(char *str)
1644{
1645 if (!str)
1646 return 0;
1647 dhash_entries = simple_strtoul(str, &str, 0);
1648 return 1;
1649}
1650__setup("dhash_entries=", set_dhash_entries);
1651
1652static void __init dcache_init_early(void)
1653{
1654 int loop;
1655
1656 /* If hashes are distributed across NUMA nodes, defer
1657 * hash allocation until vmalloc space is available.
1658 */
1659 if (hashdist)
1660 return;
1661
1662 dentry_hashtable =
1663 alloc_large_system_hash("Dentry cache",
1664 sizeof(struct hlist_head),
1665 dhash_entries,
1666 13,
1667 HASH_EARLY,
1668 &d_hash_shift,
1669 &d_hash_mask,
1670 0);
1671
1672 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1673 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1674}
1675
1676static void __init dcache_init(unsigned long mempages)
1677{
1678 int loop;
1679
1680 /*
1681 * A constructor could be added for stable state like the lists,
1682 * but it is probably not worth it because of the cache nature
1683 * of the dcache.
1684 */
1685 dentry_cache = kmem_cache_create("dentry_cache",
1686 sizeof(struct dentry),
1687 0,
Paul Jacksonb0196002006-03-24 03:16:09 -08001688 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1689 SLAB_MEM_SPREAD),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 NULL, NULL);
1691
1692 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1693
1694 /* Hash may have been set up in dcache_init_early */
1695 if (!hashdist)
1696 return;
1697
1698 dentry_hashtable =
1699 alloc_large_system_hash("Dentry cache",
1700 sizeof(struct hlist_head),
1701 dhash_entries,
1702 13,
1703 0,
1704 &d_hash_shift,
1705 &d_hash_mask,
1706 0);
1707
1708 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1709 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1710}
1711
1712/* SLAB cache for __getname() consumers */
Eric Dumazetfa3536c2006-03-26 01:37:24 -08001713kmem_cache_t *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
1715/* SLAB cache for file structures */
Eric Dumazetfa3536c2006-03-26 01:37:24 -08001716kmem_cache_t *filp_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718EXPORT_SYMBOL(d_genocide);
1719
1720extern void bdev_cache_init(void);
1721extern void chrdev_init(void);
1722
1723void __init vfs_caches_init_early(void)
1724{
1725 dcache_init_early();
1726 inode_init_early();
1727}
1728
1729void __init vfs_caches_init(unsigned long mempages)
1730{
1731 unsigned long reserve;
1732
1733 /* Base hash sizes on available memory, with a reserve equal to
1734 150% of current kernel size */
1735
1736 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1737 mempages -= reserve;
1738
1739 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1740 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1741
1742 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -08001743 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744
1745 dcache_init(mempages);
1746 inode_init(mempages);
1747 files_init(mempages);
1748 mnt_init(mempages);
1749 bdev_cache_init();
1750 chrdev_init();
1751}
1752
1753EXPORT_SYMBOL(d_alloc);
1754EXPORT_SYMBOL(d_alloc_anon);
1755EXPORT_SYMBOL(d_alloc_root);
1756EXPORT_SYMBOL(d_delete);
1757EXPORT_SYMBOL(d_find_alias);
1758EXPORT_SYMBOL(d_instantiate);
1759EXPORT_SYMBOL(d_invalidate);
1760EXPORT_SYMBOL(d_lookup);
1761EXPORT_SYMBOL(d_move);
1762EXPORT_SYMBOL(d_path);
1763EXPORT_SYMBOL(d_prune_aliases);
1764EXPORT_SYMBOL(d_rehash);
1765EXPORT_SYMBOL(d_splice_alias);
1766EXPORT_SYMBOL(d_validate);
1767EXPORT_SYMBOL(dget_locked);
1768EXPORT_SYMBOL(dput);
1769EXPORT_SYMBOL(find_inode_number);
1770EXPORT_SYMBOL(have_submounts);
1771EXPORT_SYMBOL(names_cachep);
1772EXPORT_SYMBOL(shrink_dcache_parent);
1773EXPORT_SYMBOL(shrink_dcache_sb);