blob: 48b44a714b3558c5f73099315094799b680b82ff [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/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700362 * Throw away a dentry - free the inode, dput the parent. This requires that
363 * the LRU list has already been removed.
364 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 * Called with dcache_lock, drops it and then regains.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700366 * Called with dentry->d_lock held, drops it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 */
Andrew Mortond702ccb2006-06-22 14:47:31 -0700368static void prune_one_dentry(struct dentry * dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
370 struct dentry * parent;
371
372 __d_drop(dentry);
Eric Dumazet5160ee62006-01-08 01:03:32 -0800373 list_del(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 dentry_stat.nr_dentry--; /* For d_free, below */
375 dentry_iput(dentry);
376 parent = dentry->d_parent;
377 d_free(dentry);
378 if (parent != dentry)
379 dput(parent);
380 spin_lock(&dcache_lock);
381}
382
383/**
384 * prune_dcache - shrink the dcache
385 * @count: number of entries to try and free
NeilBrown0feae5c2006-06-22 14:47:28 -0700386 * @sb: if given, ignore dentries for other superblocks
387 * which are being unmounted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 *
389 * Shrink the dcache. This is done when we need
390 * more memory, or simply when we need to unmount
391 * something (at which point we need to unuse
392 * all dentries).
393 *
394 * This function may fail to free any resources if
395 * all the dentries are in use.
396 */
397
NeilBrown0feae5c2006-06-22 14:47:28 -0700398static void prune_dcache(int count, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
400 spin_lock(&dcache_lock);
401 for (; count ; count--) {
402 struct dentry *dentry;
403 struct list_head *tmp;
NeilBrown0feae5c2006-06-22 14:47:28 -0700404 struct rw_semaphore *s_umount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
406 cond_resched_lock(&dcache_lock);
407
408 tmp = dentry_unused.prev;
Hua Zhongf58a1eb2006-06-25 05:49:32 -0700409 if (sb) {
NeilBrown0feae5c2006-06-22 14:47:28 -0700410 /* Try to find a dentry for this sb, but don't try
411 * too hard, if they aren't near the tail they will
412 * be moved down again soon
413 */
414 int skip = count;
415 while (skip && tmp != &dentry_unused &&
416 list_entry(tmp, struct dentry, d_lru)->d_sb != sb) {
417 skip--;
418 tmp = tmp->prev;
419 }
420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if (tmp == &dentry_unused)
422 break;
423 list_del_init(tmp);
424 prefetch(dentry_unused.prev);
425 dentry_stat.nr_unused--;
426 dentry = list_entry(tmp, struct dentry, d_lru);
427
428 spin_lock(&dentry->d_lock);
429 /*
430 * We found an inuse dentry which was not removed from
431 * dentry_unused because of laziness during lookup. Do not free
432 * it - just keep it off the dentry_unused list.
433 */
434 if (atomic_read(&dentry->d_count)) {
435 spin_unlock(&dentry->d_lock);
436 continue;
437 }
438 /* If the dentry was recently referenced, don't free it. */
439 if (dentry->d_flags & DCACHE_REFERENCED) {
440 dentry->d_flags &= ~DCACHE_REFERENCED;
441 list_add(&dentry->d_lru, &dentry_unused);
442 dentry_stat.nr_unused++;
443 spin_unlock(&dentry->d_lock);
444 continue;
445 }
NeilBrown0feae5c2006-06-22 14:47:28 -0700446 /*
447 * If the dentry is not DCACHED_REFERENCED, it is time
448 * to remove it from the dcache, provided the super block is
449 * NULL (which means we are trying to reclaim memory)
450 * or this dentry belongs to the same super block that
451 * we want to shrink.
452 */
453 /*
454 * If this dentry is for "my" filesystem, then I can prune it
455 * without taking the s_umount lock (I already hold it).
456 */
457 if (sb && dentry->d_sb == sb) {
458 prune_one_dentry(dentry);
459 continue;
460 }
461 /*
462 * ...otherwise we need to be sure this filesystem isn't being
463 * unmounted, otherwise we could race with
464 * generic_shutdown_super(), and end up holding a reference to
465 * an inode while the filesystem is unmounted.
466 * So we try to get s_umount, and make sure s_root isn't NULL.
467 * (Take a local copy of s_umount to avoid a use-after-free of
468 * `dentry').
469 */
470 s_umount = &dentry->d_sb->s_umount;
471 if (down_read_trylock(s_umount)) {
472 if (dentry->d_sb->s_root != NULL) {
473 prune_one_dentry(dentry);
474 up_read(s_umount);
475 continue;
476 }
477 up_read(s_umount);
478 }
479 spin_unlock(&dentry->d_lock);
480 /* Cannot remove the first dentry, and it isn't appropriate
481 * to move it to the head of the list, so give up, and try
482 * later
483 */
484 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 }
486 spin_unlock(&dcache_lock);
487}
488
489/*
490 * Shrink the dcache for the specified super block.
491 * This allows us to unmount a device without disturbing
492 * the dcache for the other devices.
493 *
494 * This implementation makes just two traversals of the
495 * unused list. On the first pass we move the selected
496 * dentries to the most recent end, and on the second
497 * pass we free them. The second pass must restart after
498 * each dput(), but since the target dentries are all at
499 * the end, it's really just a single traversal.
500 */
501
502/**
503 * shrink_dcache_sb - shrink dcache for a superblock
504 * @sb: superblock
505 *
506 * Shrink the dcache for the specified super block. This
507 * is used to free the dcache before unmounting a file
508 * system
509 */
510
511void shrink_dcache_sb(struct super_block * sb)
512{
513 struct list_head *tmp, *next;
514 struct dentry *dentry;
515
516 /*
517 * Pass one ... move the dentries for the specified
518 * superblock to the most recent end of the unused list.
519 */
520 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700521 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 dentry = list_entry(tmp, struct dentry, d_lru);
523 if (dentry->d_sb != sb)
524 continue;
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700525 list_move(tmp, &dentry_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 }
527
528 /*
529 * Pass two ... free the dentries for this superblock.
530 */
531repeat:
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700532 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 dentry = list_entry(tmp, struct dentry, d_lru);
534 if (dentry->d_sb != sb)
535 continue;
536 dentry_stat.nr_unused--;
537 list_del_init(tmp);
538 spin_lock(&dentry->d_lock);
539 if (atomic_read(&dentry->d_count)) {
540 spin_unlock(&dentry->d_lock);
541 continue;
542 }
543 prune_one_dentry(dentry);
Kirill Korotaev2ab13462006-03-25 03:07:45 -0800544 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 goto repeat;
546 }
547 spin_unlock(&dcache_lock);
548}
549
550/*
551 * Search for at least 1 mount point in the dentry's subdirs.
552 * We descend to the next level whenever the d_subdirs
553 * list is non-empty and continue searching.
554 */
555
556/**
557 * have_submounts - check for mounts over a dentry
558 * @parent: dentry to check.
559 *
560 * Return true if the parent or its subdirectories contain
561 * a mount point
562 */
563
564int have_submounts(struct dentry *parent)
565{
566 struct dentry *this_parent = parent;
567 struct list_head *next;
568
569 spin_lock(&dcache_lock);
570 if (d_mountpoint(parent))
571 goto positive;
572repeat:
573 next = this_parent->d_subdirs.next;
574resume:
575 while (next != &this_parent->d_subdirs) {
576 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800577 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 next = tmp->next;
579 /* Have we found a mount point ? */
580 if (d_mountpoint(dentry))
581 goto positive;
582 if (!list_empty(&dentry->d_subdirs)) {
583 this_parent = dentry;
584 goto repeat;
585 }
586 }
587 /*
588 * All done at this level ... ascend and resume the search.
589 */
590 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800591 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 this_parent = this_parent->d_parent;
593 goto resume;
594 }
595 spin_unlock(&dcache_lock);
596 return 0; /* No mount points found in tree */
597positive:
598 spin_unlock(&dcache_lock);
599 return 1;
600}
601
602/*
603 * Search the dentry child list for the specified parent,
604 * and move any unused dentries to the end of the unused
605 * list for prune_dcache(). We descend to the next level
606 * whenever the d_subdirs list is non-empty and continue
607 * searching.
608 *
609 * It returns zero iff there are no unused children,
610 * otherwise it returns the number of children moved to
611 * the end of the unused list. This may not be the total
612 * number of unused children, because select_parent can
613 * drop the lock and return early due to latency
614 * constraints.
615 */
616static int select_parent(struct dentry * parent)
617{
618 struct dentry *this_parent = parent;
619 struct list_head *next;
620 int found = 0;
621
622 spin_lock(&dcache_lock);
623repeat:
624 next = this_parent->d_subdirs.next;
625resume:
626 while (next != &this_parent->d_subdirs) {
627 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800628 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 next = tmp->next;
630
631 if (!list_empty(&dentry->d_lru)) {
632 dentry_stat.nr_unused--;
633 list_del_init(&dentry->d_lru);
634 }
635 /*
636 * move only zero ref count dentries to the end
637 * of the unused list for prune_dcache
638 */
639 if (!atomic_read(&dentry->d_count)) {
Akinobu Mita8e130592006-06-26 00:24:37 -0700640 list_add_tail(&dentry->d_lru, &dentry_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 dentry_stat.nr_unused++;
642 found++;
643 }
644
645 /*
646 * We can return to the caller if we have found some (this
647 * ensures forward progress). We'll be coming back to find
648 * the rest.
649 */
650 if (found && need_resched())
651 goto out;
652
653 /*
654 * Descend a level if the d_subdirs list is non-empty.
655 */
656 if (!list_empty(&dentry->d_subdirs)) {
657 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 goto repeat;
659 }
660 }
661 /*
662 * All done at this level ... ascend and resume the search.
663 */
664 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800665 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 goto resume;
668 }
669out:
670 spin_unlock(&dcache_lock);
671 return found;
672}
673
674/**
675 * shrink_dcache_parent - prune dcache
676 * @parent: parent of entries to prune
677 *
678 * Prune the dcache to remove unused children of the parent dentry.
679 */
680
681void shrink_dcache_parent(struct dentry * parent)
682{
683 int found;
684
685 while ((found = select_parent(parent)) != 0)
NeilBrown0feae5c2006-06-22 14:47:28 -0700686 prune_dcache(found, parent->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689/*
690 * Scan `nr' dentries and return the number which remain.
691 *
692 * We need to avoid reentering the filesystem if the caller is performing a
693 * GFP_NOFS allocation attempt. One example deadlock is:
694 *
695 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
696 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
697 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
698 *
699 * In this case we return -1 to tell the caller that we baled.
700 */
Al Viro27496a82005-10-21 03:20:48 -0400701static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 if (nr) {
704 if (!(gfp_mask & __GFP_FS))
705 return -1;
NeilBrown0feae5c2006-06-22 14:47:28 -0700706 prune_dcache(nr, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 }
708 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
709}
710
711/**
712 * d_alloc - allocate a dcache entry
713 * @parent: parent of entry to allocate
714 * @name: qstr of the name
715 *
716 * Allocates a dentry. It returns %NULL if there is insufficient memory
717 * available. On a success the dentry is returned. The name passed in is
718 * copied and the copy passed in may be reused after this call.
719 */
720
721struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
722{
723 struct dentry *dentry;
724 char *dname;
725
726 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
727 if (!dentry)
728 return NULL;
729
730 if (name->len > DNAME_INLINE_LEN-1) {
731 dname = kmalloc(name->len + 1, GFP_KERNEL);
732 if (!dname) {
733 kmem_cache_free(dentry_cache, dentry);
734 return NULL;
735 }
736 } else {
737 dname = dentry->d_iname;
738 }
739 dentry->d_name.name = dname;
740
741 dentry->d_name.len = name->len;
742 dentry->d_name.hash = name->hash;
743 memcpy(dname, name->name, name->len);
744 dname[name->len] = 0;
745
746 atomic_set(&dentry->d_count, 1);
747 dentry->d_flags = DCACHE_UNHASHED;
748 spin_lock_init(&dentry->d_lock);
749 dentry->d_inode = NULL;
750 dentry->d_parent = NULL;
751 dentry->d_sb = NULL;
752 dentry->d_op = NULL;
753 dentry->d_fsdata = NULL;
754 dentry->d_mounted = 0;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800755#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800757#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 INIT_HLIST_NODE(&dentry->d_hash);
759 INIT_LIST_HEAD(&dentry->d_lru);
760 INIT_LIST_HEAD(&dentry->d_subdirs);
761 INIT_LIST_HEAD(&dentry->d_alias);
762
763 if (parent) {
764 dentry->d_parent = dget(parent);
765 dentry->d_sb = parent->d_sb;
766 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800767 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
769
770 spin_lock(&dcache_lock);
771 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800772 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 dentry_stat.nr_dentry++;
774 spin_unlock(&dcache_lock);
775
776 return dentry;
777}
778
779struct dentry *d_alloc_name(struct dentry *parent, const char *name)
780{
781 struct qstr q;
782
783 q.name = name;
784 q.len = strlen(name);
785 q.hash = full_name_hash(q.name, q.len);
786 return d_alloc(parent, &q);
787}
788
789/**
790 * d_instantiate - fill in inode information for a dentry
791 * @entry: dentry to complete
792 * @inode: inode to attach to this dentry
793 *
794 * Fill in inode information in the entry.
795 *
796 * This turns negative dentries into productive full members
797 * of society.
798 *
799 * NOTE! This assumes that the inode count has been incremented
800 * (or otherwise set) by the caller to indicate that it is now
801 * in use by the dcache.
802 */
803
804void d_instantiate(struct dentry *entry, struct inode * inode)
805{
Eric Sesterhenn28133c72006-03-26 18:25:39 +0200806 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 spin_lock(&dcache_lock);
808 if (inode)
809 list_add(&entry->d_alias, &inode->i_dentry);
810 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800811 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 spin_unlock(&dcache_lock);
813 security_d_instantiate(entry, inode);
814}
815
816/**
817 * d_instantiate_unique - instantiate a non-aliased dentry
818 * @entry: dentry to instantiate
819 * @inode: inode to attach to this dentry
820 *
821 * Fill in inode information in the entry. On success, it returns NULL.
822 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -0800823 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 *
825 * Note that in order to avoid conflicts with rename() etc, the caller
826 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -0800827 *
828 * This also assumes that the inode count has been incremented
829 * (or otherwise set) by the caller to indicate that it is now
830 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 */
832struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
833{
834 struct dentry *alias;
835 int len = entry->d_name.len;
836 const char *name = entry->d_name.name;
837 unsigned int hash = entry->d_name.hash;
838
839 BUG_ON(!list_empty(&entry->d_alias));
840 spin_lock(&dcache_lock);
841 if (!inode)
842 goto do_negative;
843 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
844 struct qstr *qstr = &alias->d_name;
845
846 if (qstr->hash != hash)
847 continue;
848 if (alias->d_parent != entry->d_parent)
849 continue;
850 if (qstr->len != len)
851 continue;
852 if (memcmp(qstr->name, name, len))
853 continue;
854 dget_locked(alias);
855 spin_unlock(&dcache_lock);
856 BUG_ON(!d_unhashed(alias));
Oleg Drokine866cfa2006-01-09 20:52:51 -0800857 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 return alias;
859 }
860 list_add(&entry->d_alias, &inode->i_dentry);
861do_negative:
862 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800863 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 spin_unlock(&dcache_lock);
865 security_d_instantiate(entry, inode);
866 return NULL;
867}
868EXPORT_SYMBOL(d_instantiate_unique);
869
870/**
871 * d_alloc_root - allocate root dentry
872 * @root_inode: inode to allocate the root for
873 *
874 * Allocate a root ("/") dentry for the inode given. The inode is
875 * instantiated and returned. %NULL is returned if there is insufficient
876 * memory or the inode passed is %NULL.
877 */
878
879struct dentry * d_alloc_root(struct inode * root_inode)
880{
881 struct dentry *res = NULL;
882
883 if (root_inode) {
884 static const struct qstr name = { .name = "/", .len = 1 };
885
886 res = d_alloc(NULL, &name);
887 if (res) {
888 res->d_sb = root_inode->i_sb;
889 res->d_parent = res;
890 d_instantiate(res, root_inode);
891 }
892 }
893 return res;
894}
895
896static inline struct hlist_head *d_hash(struct dentry *parent,
897 unsigned long hash)
898{
899 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
900 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
901 return dentry_hashtable + (hash & D_HASHMASK);
902}
903
904/**
905 * d_alloc_anon - allocate an anonymous dentry
906 * @inode: inode to allocate the dentry for
907 *
908 * This is similar to d_alloc_root. It is used by filesystems when
909 * creating a dentry for a given inode, often in the process of
910 * mapping a filehandle to a dentry. The returned dentry may be
911 * anonymous, or may have a full name (if the inode was already
912 * in the cache). The file system may need to make further
913 * efforts to connect this dentry into the dcache properly.
914 *
915 * When called on a directory inode, we must ensure that
916 * the inode only ever has one dentry. If a dentry is
917 * found, that is returned instead of allocating a new one.
918 *
919 * On successful return, the reference to the inode has been transferred
920 * to the dentry. If %NULL is returned (indicating kmalloc failure),
921 * the reference on the inode has not been released.
922 */
923
924struct dentry * d_alloc_anon(struct inode *inode)
925{
926 static const struct qstr anonstring = { .name = "" };
927 struct dentry *tmp;
928 struct dentry *res;
929
930 if ((res = d_find_alias(inode))) {
931 iput(inode);
932 return res;
933 }
934
935 tmp = d_alloc(NULL, &anonstring);
936 if (!tmp)
937 return NULL;
938
939 tmp->d_parent = tmp; /* make sure dput doesn't croak */
940
941 spin_lock(&dcache_lock);
942 res = __d_find_alias(inode, 0);
943 if (!res) {
944 /* attach a disconnected dentry */
945 res = tmp;
946 tmp = NULL;
947 spin_lock(&res->d_lock);
948 res->d_sb = inode->i_sb;
949 res->d_parent = res;
950 res->d_inode = inode;
951 res->d_flags |= DCACHE_DISCONNECTED;
952 res->d_flags &= ~DCACHE_UNHASHED;
953 list_add(&res->d_alias, &inode->i_dentry);
954 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
955 spin_unlock(&res->d_lock);
956
957 inode = NULL; /* don't drop reference */
958 }
959 spin_unlock(&dcache_lock);
960
961 if (inode)
962 iput(inode);
963 if (tmp)
964 dput(tmp);
965 return res;
966}
967
968
969/**
970 * d_splice_alias - splice a disconnected dentry into the tree if one exists
971 * @inode: the inode which may have a disconnected dentry
972 * @dentry: a negative dentry which we want to point to the inode.
973 *
974 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
975 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
976 * and return it, else simply d_add the inode to the dentry and return NULL.
977 *
978 * This is needed in the lookup routine of any filesystem that is exportable
979 * (via knfsd) so that we can build dcache paths to directories effectively.
980 *
981 * If a dentry was found and moved, then it is returned. Otherwise NULL
982 * is returned. This matches the expected return value of ->lookup.
983 *
984 */
985struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
986{
987 struct dentry *new = NULL;
988
989 if (inode) {
990 spin_lock(&dcache_lock);
991 new = __d_find_alias(inode, 1);
992 if (new) {
993 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Pigginc32ccd82006-03-25 03:07:09 -0800994 fsnotify_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 spin_unlock(&dcache_lock);
996 security_d_instantiate(new, inode);
997 d_rehash(dentry);
998 d_move(new, dentry);
999 iput(inode);
1000 } else {
1001 /* d_instantiate takes dcache_lock, so we do it by hand */
1002 list_add(&dentry->d_alias, &inode->i_dentry);
1003 dentry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001004 fsnotify_d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 spin_unlock(&dcache_lock);
1006 security_d_instantiate(dentry, inode);
1007 d_rehash(dentry);
1008 }
1009 } else
1010 d_add(dentry, inode);
1011 return new;
1012}
1013
1014
1015/**
1016 * d_lookup - search for a dentry
1017 * @parent: parent dentry
1018 * @name: qstr of name we wish to find
1019 *
1020 * Searches the children of the parent dentry for the name in question. If
1021 * the dentry is found its reference count is incremented and the dentry
1022 * is returned. The caller must use d_put to free the entry when it has
1023 * finished using it. %NULL is returned on failure.
1024 *
1025 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1026 * Memory barriers are used while updating and doing lockless traversal.
1027 * To avoid races with d_move while rename is happening, d_lock is used.
1028 *
1029 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1030 * and name pointer in one structure pointed by d_qstr.
1031 *
1032 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1033 * lookup is going on.
1034 *
1035 * dentry_unused list is not updated even if lookup finds the required dentry
1036 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1037 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1038 * acquisition.
1039 *
1040 * d_lookup() is protected against the concurrent renames in some unrelated
1041 * directory using the seqlockt_t rename_lock.
1042 */
1043
1044struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1045{
1046 struct dentry * dentry = NULL;
1047 unsigned long seq;
1048
1049 do {
1050 seq = read_seqbegin(&rename_lock);
1051 dentry = __d_lookup(parent, name);
1052 if (dentry)
1053 break;
1054 } while (read_seqretry(&rename_lock, seq));
1055 return dentry;
1056}
1057
1058struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1059{
1060 unsigned int len = name->len;
1061 unsigned int hash = name->hash;
1062 const unsigned char *str = name->name;
1063 struct hlist_head *head = d_hash(parent,hash);
1064 struct dentry *found = NULL;
1065 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001066 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 rcu_read_lock();
1069
Paul E. McKenney665a7582005-11-07 00:59:17 -08001070 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 struct qstr *qstr;
1072
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (dentry->d_name.hash != hash)
1074 continue;
1075 if (dentry->d_parent != parent)
1076 continue;
1077
1078 spin_lock(&dentry->d_lock);
1079
1080 /*
1081 * Recheck the dentry after taking the lock - d_move may have
1082 * changed things. Don't bother checking the hash because we're
1083 * about to compare the whole name anyway.
1084 */
1085 if (dentry->d_parent != parent)
1086 goto next;
1087
1088 /*
1089 * It is safe to compare names since d_move() cannot
1090 * change the qstr (protected by d_lock).
1091 */
1092 qstr = &dentry->d_name;
1093 if (parent->d_op && parent->d_op->d_compare) {
1094 if (parent->d_op->d_compare(parent, qstr, name))
1095 goto next;
1096 } else {
1097 if (qstr->len != len)
1098 goto next;
1099 if (memcmp(qstr->name, str, len))
1100 goto next;
1101 }
1102
1103 if (!d_unhashed(dentry)) {
1104 atomic_inc(&dentry->d_count);
1105 found = dentry;
1106 }
1107 spin_unlock(&dentry->d_lock);
1108 break;
1109next:
1110 spin_unlock(&dentry->d_lock);
1111 }
1112 rcu_read_unlock();
1113
1114 return found;
1115}
1116
1117/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001118 * d_hash_and_lookup - hash the qstr then search for a dentry
1119 * @dir: Directory to search in
1120 * @name: qstr of name we wish to find
1121 *
1122 * On hash failure or on lookup failure NULL is returned.
1123 */
1124struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1125{
1126 struct dentry *dentry = NULL;
1127
1128 /*
1129 * Check for a fs-specific hash function. Note that we must
1130 * calculate the standard hash first, as the d_op->d_hash()
1131 * routine may choose to leave the hash value unchanged.
1132 */
1133 name->hash = full_name_hash(name->name, name->len);
1134 if (dir->d_op && dir->d_op->d_hash) {
1135 if (dir->d_op->d_hash(dir, name) < 0)
1136 goto out;
1137 }
1138 dentry = d_lookup(dir, name);
1139out:
1140 return dentry;
1141}
1142
1143/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 * d_validate - verify dentry provided from insecure source
1145 * @dentry: The dentry alleged to be valid child of @dparent
1146 * @dparent: The parent dentry (known to be valid)
1147 * @hash: Hash of the dentry
1148 * @len: Length of the name
1149 *
1150 * An insecure source has sent us a dentry, here we verify it and dget() it.
1151 * This is used by ncpfs in its readdir implementation.
1152 * Zero is returned in the dentry is invalid.
1153 */
1154
1155int d_validate(struct dentry *dentry, struct dentry *dparent)
1156{
1157 struct hlist_head *base;
1158 struct hlist_node *lhp;
1159
1160 /* Check whether the ptr might be valid at all.. */
1161 if (!kmem_ptr_validate(dentry_cache, dentry))
1162 goto out;
1163
1164 if (dentry->d_parent != dparent)
1165 goto out;
1166
1167 spin_lock(&dcache_lock);
1168 base = d_hash(dparent, dentry->d_name.hash);
1169 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001170 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 * as it is parsed under dcache_lock
1172 */
1173 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1174 __dget_locked(dentry);
1175 spin_unlock(&dcache_lock);
1176 return 1;
1177 }
1178 }
1179 spin_unlock(&dcache_lock);
1180out:
1181 return 0;
1182}
1183
1184/*
1185 * When a file is deleted, we have two options:
1186 * - turn this dentry into a negative dentry
1187 * - unhash this dentry and free it.
1188 *
1189 * Usually, we want to just turn this into
1190 * a negative dentry, but if anybody else is
1191 * currently using the dentry or the inode
1192 * we can't do that and we fall back on removing
1193 * it from the hash queues and waiting for
1194 * it to be deleted later when it has no users
1195 */
1196
1197/**
1198 * d_delete - delete a dentry
1199 * @dentry: The dentry to delete
1200 *
1201 * Turn the dentry into a negative dentry if possible, otherwise
1202 * remove it from the hash queues so it can be deleted later
1203 */
1204
1205void d_delete(struct dentry * dentry)
1206{
John McCutchan7a91bf72005-08-08 13:52:16 -04001207 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 /*
1209 * Are we the only user?
1210 */
1211 spin_lock(&dcache_lock);
1212 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001213 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 if (atomic_read(&dentry->d_count) == 1) {
1215 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001216 fsnotify_nameremove(dentry, isdir);
Amy Griffis7a2bd3f2006-03-31 02:30:54 -08001217
1218 /* remove this and other inotify debug checks after 2.6.18 */
1219 dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 return;
1221 }
1222
1223 if (!d_unhashed(dentry))
1224 __d_drop(dentry);
1225
1226 spin_unlock(&dentry->d_lock);
1227 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001228
1229 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230}
1231
1232static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1233{
1234
1235 entry->d_flags &= ~DCACHE_UNHASHED;
1236 hlist_add_head_rcu(&entry->d_hash, list);
1237}
1238
1239/**
1240 * d_rehash - add an entry back to the hash
1241 * @entry: dentry to add to the hash
1242 *
1243 * Adds a dentry to the hash according to its name.
1244 */
1245
1246void d_rehash(struct dentry * entry)
1247{
1248 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1249
1250 spin_lock(&dcache_lock);
1251 spin_lock(&entry->d_lock);
1252 __d_rehash(entry, list);
1253 spin_unlock(&entry->d_lock);
1254 spin_unlock(&dcache_lock);
1255}
1256
1257#define do_switch(x,y) do { \
1258 __typeof__ (x) __tmp = x; \
1259 x = y; y = __tmp; } while (0)
1260
1261/*
1262 * When switching names, the actual string doesn't strictly have to
1263 * be preserved in the target - because we're dropping the target
1264 * anyway. As such, we can just do a simple memcpy() to copy over
1265 * the new name before we switch.
1266 *
1267 * Note that we have to be a lot more careful about getting the hash
1268 * switched - we have to switch the hash value properly even if it
1269 * then no longer matches the actual (corrupted) string of the target.
1270 * The hash value has to match the hash queue that the dentry is on..
1271 */
1272static void switch_names(struct dentry *dentry, struct dentry *target)
1273{
1274 if (dname_external(target)) {
1275 if (dname_external(dentry)) {
1276 /*
1277 * Both external: swap the pointers
1278 */
1279 do_switch(target->d_name.name, dentry->d_name.name);
1280 } else {
1281 /*
1282 * dentry:internal, target:external. Steal target's
1283 * storage and make target internal.
1284 */
1285 dentry->d_name.name = target->d_name.name;
1286 target->d_name.name = target->d_iname;
1287 }
1288 } else {
1289 if (dname_external(dentry)) {
1290 /*
1291 * dentry:external, target:internal. Give dentry's
1292 * storage to target and make dentry internal
1293 */
1294 memcpy(dentry->d_iname, target->d_name.name,
1295 target->d_name.len + 1);
1296 target->d_name.name = dentry->d_name.name;
1297 dentry->d_name.name = dentry->d_iname;
1298 } else {
1299 /*
1300 * Both are internal. Just copy target to dentry
1301 */
1302 memcpy(dentry->d_iname, target->d_name.name,
1303 target->d_name.len + 1);
1304 }
1305 }
1306}
1307
1308/*
1309 * We cannibalize "target" when moving dentry on top of it,
1310 * because it's going to be thrown away anyway. We could be more
1311 * polite about it, though.
1312 *
1313 * This forceful removal will result in ugly /proc output if
1314 * somebody holds a file open that got deleted due to a rename.
1315 * We could be nicer about the deleted file, and let it show
1316 * up under the name it got deleted rather than the name that
1317 * deleted it.
1318 */
1319
1320/**
1321 * d_move - move a dentry
1322 * @dentry: entry to move
1323 * @target: new dentry
1324 *
1325 * Update the dcache to reflect the move of a file name. Negative
1326 * dcache entries should not be moved in this way.
1327 */
1328
1329void d_move(struct dentry * dentry, struct dentry * target)
1330{
1331 struct hlist_head *list;
1332
1333 if (!dentry->d_inode)
1334 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1335
1336 spin_lock(&dcache_lock);
1337 write_seqlock(&rename_lock);
1338 /*
1339 * XXXX: do we really need to take target->d_lock?
1340 */
1341 if (target < dentry) {
1342 spin_lock(&target->d_lock);
1343 spin_lock(&dentry->d_lock);
1344 } else {
1345 spin_lock(&dentry->d_lock);
1346 spin_lock(&target->d_lock);
1347 }
1348
1349 /* Move the dentry to the target hash queue, if on different bucket */
1350 if (dentry->d_flags & DCACHE_UNHASHED)
1351 goto already_unhashed;
1352
1353 hlist_del_rcu(&dentry->d_hash);
1354
1355already_unhashed:
1356 list = d_hash(target->d_parent, target->d_name.hash);
1357 __d_rehash(dentry, list);
1358
1359 /* Unhash the target: dput() will then get rid of it */
1360 __d_drop(target);
1361
Eric Dumazet5160ee62006-01-08 01:03:32 -08001362 list_del(&dentry->d_u.d_child);
1363 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 /* Switch the names.. */
1366 switch_names(dentry, target);
1367 do_switch(dentry->d_name.len, target->d_name.len);
1368 do_switch(dentry->d_name.hash, target->d_name.hash);
1369
1370 /* ... and switch the parents */
1371 if (IS_ROOT(dentry)) {
1372 dentry->d_parent = target->d_parent;
1373 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001374 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 } else {
1376 do_switch(dentry->d_parent, target->d_parent);
1377
1378 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001379 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
1381
Eric Dumazet5160ee62006-01-08 01:03:32 -08001382 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001384 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 spin_unlock(&dentry->d_lock);
1386 write_sequnlock(&rename_lock);
1387 spin_unlock(&dcache_lock);
1388}
1389
1390/**
1391 * d_path - return the path of a dentry
1392 * @dentry: dentry to report
1393 * @vfsmnt: vfsmnt to which the dentry belongs
1394 * @root: root dentry
1395 * @rootmnt: vfsmnt to which the root dentry belongs
1396 * @buffer: buffer to return value in
1397 * @buflen: buffer length
1398 *
1399 * Convert a dentry into an ASCII path name. If the entry has been deleted
1400 * the string " (deleted)" is appended. Note that this is ambiguous.
1401 *
1402 * Returns the buffer or an error code if the path was too long.
1403 *
1404 * "buflen" should be positive. Caller holds the dcache_lock.
1405 */
1406static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1407 struct dentry *root, struct vfsmount *rootmnt,
1408 char *buffer, int buflen)
1409{
1410 char * end = buffer+buflen;
1411 char * retval;
1412 int namelen;
1413
1414 *--end = '\0';
1415 buflen--;
1416 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1417 buflen -= 10;
1418 end -= 10;
1419 if (buflen < 0)
1420 goto Elong;
1421 memcpy(end, " (deleted)", 10);
1422 }
1423
1424 if (buflen < 1)
1425 goto Elong;
1426 /* Get '/' right */
1427 retval = end-1;
1428 *retval = '/';
1429
1430 for (;;) {
1431 struct dentry * parent;
1432
1433 if (dentry == root && vfsmnt == rootmnt)
1434 break;
1435 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1436 /* Global root? */
1437 spin_lock(&vfsmount_lock);
1438 if (vfsmnt->mnt_parent == vfsmnt) {
1439 spin_unlock(&vfsmount_lock);
1440 goto global_root;
1441 }
1442 dentry = vfsmnt->mnt_mountpoint;
1443 vfsmnt = vfsmnt->mnt_parent;
1444 spin_unlock(&vfsmount_lock);
1445 continue;
1446 }
1447 parent = dentry->d_parent;
1448 prefetch(parent);
1449 namelen = dentry->d_name.len;
1450 buflen -= namelen + 1;
1451 if (buflen < 0)
1452 goto Elong;
1453 end -= namelen;
1454 memcpy(end, dentry->d_name.name, namelen);
1455 *--end = '/';
1456 retval = end;
1457 dentry = parent;
1458 }
1459
1460 return retval;
1461
1462global_root:
1463 namelen = dentry->d_name.len;
1464 buflen -= namelen;
1465 if (buflen < 0)
1466 goto Elong;
1467 retval -= namelen-1; /* hit the slash */
1468 memcpy(retval, dentry->d_name.name, namelen);
1469 return retval;
1470Elong:
1471 return ERR_PTR(-ENAMETOOLONG);
1472}
1473
1474/* write full pathname into buffer and return start of pathname */
1475char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1476 char *buf, int buflen)
1477{
1478 char *res;
1479 struct vfsmount *rootmnt;
1480 struct dentry *root;
1481
1482 read_lock(&current->fs->lock);
1483 rootmnt = mntget(current->fs->rootmnt);
1484 root = dget(current->fs->root);
1485 read_unlock(&current->fs->lock);
1486 spin_lock(&dcache_lock);
1487 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1488 spin_unlock(&dcache_lock);
1489 dput(root);
1490 mntput(rootmnt);
1491 return res;
1492}
1493
1494/*
1495 * NOTE! The user-level library version returns a
1496 * character pointer. The kernel system call just
1497 * returns the length of the buffer filled (which
1498 * includes the ending '\0' character), or a negative
1499 * error value. So libc would do something like
1500 *
1501 * char *getcwd(char * buf, size_t size)
1502 * {
1503 * int retval;
1504 *
1505 * retval = sys_getcwd(buf, size);
1506 * if (retval >= 0)
1507 * return buf;
1508 * errno = -retval;
1509 * return NULL;
1510 * }
1511 */
1512asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1513{
1514 int error;
1515 struct vfsmount *pwdmnt, *rootmnt;
1516 struct dentry *pwd, *root;
1517 char *page = (char *) __get_free_page(GFP_USER);
1518
1519 if (!page)
1520 return -ENOMEM;
1521
1522 read_lock(&current->fs->lock);
1523 pwdmnt = mntget(current->fs->pwdmnt);
1524 pwd = dget(current->fs->pwd);
1525 rootmnt = mntget(current->fs->rootmnt);
1526 root = dget(current->fs->root);
1527 read_unlock(&current->fs->lock);
1528
1529 error = -ENOENT;
1530 /* Has the current directory has been unlinked? */
1531 spin_lock(&dcache_lock);
1532 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1533 unsigned long len;
1534 char * cwd;
1535
1536 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1537 spin_unlock(&dcache_lock);
1538
1539 error = PTR_ERR(cwd);
1540 if (IS_ERR(cwd))
1541 goto out;
1542
1543 error = -ERANGE;
1544 len = PAGE_SIZE + page - cwd;
1545 if (len <= size) {
1546 error = len;
1547 if (copy_to_user(buf, cwd, len))
1548 error = -EFAULT;
1549 }
1550 } else
1551 spin_unlock(&dcache_lock);
1552
1553out:
1554 dput(pwd);
1555 mntput(pwdmnt);
1556 dput(root);
1557 mntput(rootmnt);
1558 free_page((unsigned long) page);
1559 return error;
1560}
1561
1562/*
1563 * Test whether new_dentry is a subdirectory of old_dentry.
1564 *
1565 * Trivially implemented using the dcache structure
1566 */
1567
1568/**
1569 * is_subdir - is new dentry a subdirectory of old_dentry
1570 * @new_dentry: new dentry
1571 * @old_dentry: old dentry
1572 *
1573 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1574 * Returns 0 otherwise.
1575 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1576 */
1577
1578int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1579{
1580 int result;
1581 struct dentry * saved = new_dentry;
1582 unsigned long seq;
1583
1584 /* need rcu_readlock to protect against the d_parent trashing due to
1585 * d_move
1586 */
1587 rcu_read_lock();
1588 do {
1589 /* for restarting inner loop in case of seq retry */
1590 new_dentry = saved;
1591 result = 0;
1592 seq = read_seqbegin(&rename_lock);
1593 for (;;) {
1594 if (new_dentry != old_dentry) {
1595 struct dentry * parent = new_dentry->d_parent;
1596 if (parent == new_dentry)
1597 break;
1598 new_dentry = parent;
1599 continue;
1600 }
1601 result = 1;
1602 break;
1603 }
1604 } while (read_seqretry(&rename_lock, seq));
1605 rcu_read_unlock();
1606
1607 return result;
1608}
1609
1610void d_genocide(struct dentry *root)
1611{
1612 struct dentry *this_parent = root;
1613 struct list_head *next;
1614
1615 spin_lock(&dcache_lock);
1616repeat:
1617 next = this_parent->d_subdirs.next;
1618resume:
1619 while (next != &this_parent->d_subdirs) {
1620 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001621 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 next = tmp->next;
1623 if (d_unhashed(dentry)||!dentry->d_inode)
1624 continue;
1625 if (!list_empty(&dentry->d_subdirs)) {
1626 this_parent = dentry;
1627 goto repeat;
1628 }
1629 atomic_dec(&dentry->d_count);
1630 }
1631 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08001632 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 atomic_dec(&this_parent->d_count);
1634 this_parent = this_parent->d_parent;
1635 goto resume;
1636 }
1637 spin_unlock(&dcache_lock);
1638}
1639
1640/**
1641 * find_inode_number - check for dentry with name
1642 * @dir: directory to check
1643 * @name: Name to find.
1644 *
1645 * Check whether a dentry already exists for the given name,
1646 * and return the inode number if it has an inode. Otherwise
1647 * 0 is returned.
1648 *
1649 * This routine is used to post-process directory listings for
1650 * filesystems using synthetic inode numbers, and is necessary
1651 * to keep getcwd() working.
1652 */
1653
1654ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1655{
1656 struct dentry * dentry;
1657 ino_t ino = 0;
1658
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001659 dentry = d_hash_and_lookup(dir, name);
1660 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (dentry->d_inode)
1662 ino = dentry->d_inode->i_ino;
1663 dput(dentry);
1664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 return ino;
1666}
1667
1668static __initdata unsigned long dhash_entries;
1669static int __init set_dhash_entries(char *str)
1670{
1671 if (!str)
1672 return 0;
1673 dhash_entries = simple_strtoul(str, &str, 0);
1674 return 1;
1675}
1676__setup("dhash_entries=", set_dhash_entries);
1677
1678static void __init dcache_init_early(void)
1679{
1680 int loop;
1681
1682 /* If hashes are distributed across NUMA nodes, defer
1683 * hash allocation until vmalloc space is available.
1684 */
1685 if (hashdist)
1686 return;
1687
1688 dentry_hashtable =
1689 alloc_large_system_hash("Dentry cache",
1690 sizeof(struct hlist_head),
1691 dhash_entries,
1692 13,
1693 HASH_EARLY,
1694 &d_hash_shift,
1695 &d_hash_mask,
1696 0);
1697
1698 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1699 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1700}
1701
1702static void __init dcache_init(unsigned long mempages)
1703{
1704 int loop;
1705
1706 /*
1707 * A constructor could be added for stable state like the lists,
1708 * but it is probably not worth it because of the cache nature
1709 * of the dcache.
1710 */
1711 dentry_cache = kmem_cache_create("dentry_cache",
1712 sizeof(struct dentry),
1713 0,
Paul Jacksonb0196002006-03-24 03:16:09 -08001714 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1715 SLAB_MEM_SPREAD),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 NULL, NULL);
1717
1718 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1719
1720 /* Hash may have been set up in dcache_init_early */
1721 if (!hashdist)
1722 return;
1723
1724 dentry_hashtable =
1725 alloc_large_system_hash("Dentry cache",
1726 sizeof(struct hlist_head),
1727 dhash_entries,
1728 13,
1729 0,
1730 &d_hash_shift,
1731 &d_hash_mask,
1732 0);
1733
1734 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1735 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1736}
1737
1738/* SLAB cache for __getname() consumers */
Eric Dumazetfa3536c2006-03-26 01:37:24 -08001739kmem_cache_t *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
1741/* SLAB cache for file structures */
Eric Dumazetfa3536c2006-03-26 01:37:24 -08001742kmem_cache_t *filp_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
1744EXPORT_SYMBOL(d_genocide);
1745
1746extern void bdev_cache_init(void);
1747extern void chrdev_init(void);
1748
1749void __init vfs_caches_init_early(void)
1750{
1751 dcache_init_early();
1752 inode_init_early();
1753}
1754
1755void __init vfs_caches_init(unsigned long mempages)
1756{
1757 unsigned long reserve;
1758
1759 /* Base hash sizes on available memory, with a reserve equal to
1760 150% of current kernel size */
1761
1762 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1763 mempages -= reserve;
1764
1765 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1766 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1767
1768 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -08001769 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771 dcache_init(mempages);
1772 inode_init(mempages);
1773 files_init(mempages);
1774 mnt_init(mempages);
1775 bdev_cache_init();
1776 chrdev_init();
1777}
1778
1779EXPORT_SYMBOL(d_alloc);
1780EXPORT_SYMBOL(d_alloc_anon);
1781EXPORT_SYMBOL(d_alloc_root);
1782EXPORT_SYMBOL(d_delete);
1783EXPORT_SYMBOL(d_find_alias);
1784EXPORT_SYMBOL(d_instantiate);
1785EXPORT_SYMBOL(d_invalidate);
1786EXPORT_SYMBOL(d_lookup);
1787EXPORT_SYMBOL(d_move);
1788EXPORT_SYMBOL(d_path);
1789EXPORT_SYMBOL(d_prune_aliases);
1790EXPORT_SYMBOL(d_rehash);
1791EXPORT_SYMBOL(d_splice_alias);
1792EXPORT_SYMBOL(d_validate);
1793EXPORT_SYMBOL(dget_locked);
1794EXPORT_SYMBOL(dput);
1795EXPORT_SYMBOL(find_inode_number);
1796EXPORT_SYMBOL(have_submounts);
1797EXPORT_SYMBOL(names_cachep);
1798EXPORT_SYMBOL(shrink_dcache_parent);
1799EXPORT_SYMBOL(shrink_dcache_sb);