blob: 1b4a3a34ec57f20d605386c129c9ebf315dd1765 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
24#include <linux/smp_lock.h>
25#include <linux/hash.h>
26#include <linux/cache.h>
27#include <linux/module.h>
28#include <linux/mount.h>
29#include <linux/file.h>
30#include <asm/uaccess.h>
31#include <linux/security.h>
32#include <linux/seqlock.h>
33#include <linux/swap.h>
34#include <linux/bootmem.h>
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Eric Dumazetfa3536c2006-03-26 01:37:24 -080037int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
39
40 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
Ingo Molnare4d91912006-07-03 00:24:34 -070041static __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43EXPORT_SYMBOL(dcache_lock);
44
Eric Dumazetfa3536c2006-03-26 01:37:24 -080045static kmem_cache_t *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
48
49/*
50 * This is the single most critical data structure when it comes
51 * to the dcache: the hashtable for lookups. Somebody should try
52 * to make this good - I've just made it work.
53 *
54 * This hash-function tries to avoid losing too many bits of hash
55 * information, yet avoid using a prime hash-size or similar.
56 */
57#define D_HASHBITS d_hash_shift
58#define D_HASHMASK d_hash_mask
59
Eric Dumazetfa3536c2006-03-26 01:37:24 -080060static unsigned int d_hash_mask __read_mostly;
61static unsigned int d_hash_shift __read_mostly;
62static struct hlist_head *dentry_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063static LIST_HEAD(dentry_unused);
64
65/* Statistics gathering. */
66struct dentry_stat_t dentry_stat = {
67 .age_limit = 45,
68};
69
70static void d_callback(struct rcu_head *head)
71{
Eric Dumazet5160ee62006-01-08 01:03:32 -080072 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74 if (dname_external(dentry))
75 kfree(dentry->d_name.name);
76 kmem_cache_free(dentry_cache, dentry);
77}
78
79/*
80 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
81 * inside dcache_lock.
82 */
83static void d_free(struct dentry *dentry)
84{
85 if (dentry->d_op && dentry->d_op->d_release)
86 dentry->d_op->d_release(dentry);
Eric Dumazet5160ee62006-01-08 01:03:32 -080087 call_rcu(&dentry->d_u.d_rcu, d_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088}
89
90/*
91 * Release the dentry's inode, using the filesystem
92 * d_iput() operation if defined.
93 * Called with dcache_lock and per dentry lock held, drops both.
94 */
Arjan van de Ven858119e2006-01-14 13:20:43 -080095static void dentry_iput(struct dentry * dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
97 struct inode *inode = dentry->d_inode;
98 if (inode) {
99 dentry->d_inode = NULL;
100 list_del_init(&dentry->d_alias);
101 spin_unlock(&dentry->d_lock);
102 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700103 if (!inode->i_nlink)
104 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 if (dentry->d_op && dentry->d_op->d_iput)
106 dentry->d_op->d_iput(dentry, inode);
107 else
108 iput(inode);
109 } else {
110 spin_unlock(&dentry->d_lock);
111 spin_unlock(&dcache_lock);
112 }
113}
114
115/*
116 * This is dput
117 *
118 * This is complicated by the fact that we do not want to put
119 * dentries that are no longer on any hash chain on the unused
120 * list: we'd much rather just get rid of them immediately.
121 *
122 * However, that implies that we have to traverse the dentry
123 * tree upwards to the parents which might _also_ now be
124 * scheduled for deletion (it may have been only waiting for
125 * its last child to go away).
126 *
127 * This tail recursion is done by hand as we don't want to depend
128 * on the compiler to always get this right (gcc generally doesn't).
129 * Real recursion would eat up our stack space.
130 */
131
132/*
133 * dput - release a dentry
134 * @dentry: dentry to release
135 *
136 * Release a dentry. This will drop the usage count and if appropriate
137 * call the dentry unlink method as well as removing it from the queues and
138 * releasing its resources. If the parent dentries were scheduled for release
139 * they too may now get deleted.
140 *
141 * no dcache lock, please.
142 */
143
144void dput(struct dentry *dentry)
145{
146 if (!dentry)
147 return;
148
149repeat:
150 if (atomic_read(&dentry->d_count) == 1)
151 might_sleep();
152 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
153 return;
154
155 spin_lock(&dentry->d_lock);
156 if (atomic_read(&dentry->d_count)) {
157 spin_unlock(&dentry->d_lock);
158 spin_unlock(&dcache_lock);
159 return;
160 }
161
162 /*
163 * AV: ->d_delete() is _NOT_ allowed to block now.
164 */
165 if (dentry->d_op && dentry->d_op->d_delete) {
166 if (dentry->d_op->d_delete(dentry))
167 goto unhash_it;
168 }
169 /* Unreachable? Get rid of it */
170 if (d_unhashed(dentry))
171 goto kill_it;
172 if (list_empty(&dentry->d_lru)) {
173 dentry->d_flags |= DCACHE_REFERENCED;
174 list_add(&dentry->d_lru, &dentry_unused);
175 dentry_stat.nr_unused++;
176 }
177 spin_unlock(&dentry->d_lock);
178 spin_unlock(&dcache_lock);
179 return;
180
181unhash_it:
182 __d_drop(dentry);
183
184kill_it: {
185 struct dentry *parent;
186
187 /* If dentry was on d_lru list
188 * delete it from there
189 */
190 if (!list_empty(&dentry->d_lru)) {
191 list_del(&dentry->d_lru);
192 dentry_stat.nr_unused--;
193 }
Eric Dumazet5160ee62006-01-08 01:03:32 -0800194 list_del(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 dentry_stat.nr_dentry--; /* For d_free, below */
196 /*drops the locks, at that point nobody can reach this dentry */
197 dentry_iput(dentry);
198 parent = dentry->d_parent;
199 d_free(dentry);
200 if (dentry == parent)
201 return;
202 dentry = parent;
203 goto repeat;
204 }
205}
206
207/**
208 * d_invalidate - invalidate a dentry
209 * @dentry: dentry to invalidate
210 *
211 * Try to invalidate the dentry if it turns out to be
212 * possible. If there are other dentries that can be
213 * reached through this one we can't delete it and we
214 * return -EBUSY. On success we return 0.
215 *
216 * no dcache lock.
217 */
218
219int d_invalidate(struct dentry * dentry)
220{
221 /*
222 * If it's already been dropped, return OK.
223 */
224 spin_lock(&dcache_lock);
225 if (d_unhashed(dentry)) {
226 spin_unlock(&dcache_lock);
227 return 0;
228 }
229 /*
230 * Check whether to do a partial shrink_dcache
231 * to get rid of unused child entries.
232 */
233 if (!list_empty(&dentry->d_subdirs)) {
234 spin_unlock(&dcache_lock);
235 shrink_dcache_parent(dentry);
236 spin_lock(&dcache_lock);
237 }
238
239 /*
240 * Somebody else still using it?
241 *
242 * If it's a directory, we can't drop it
243 * for fear of somebody re-populating it
244 * with children (even though dropping it
245 * would make it unreachable from the root,
246 * we might still populate it if it was a
247 * working directory or similar).
248 */
249 spin_lock(&dentry->d_lock);
250 if (atomic_read(&dentry->d_count) > 1) {
251 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
252 spin_unlock(&dentry->d_lock);
253 spin_unlock(&dcache_lock);
254 return -EBUSY;
255 }
256 }
257
258 __d_drop(dentry);
259 spin_unlock(&dentry->d_lock);
260 spin_unlock(&dcache_lock);
261 return 0;
262}
263
264/* This should be called _only_ with dcache_lock held */
265
266static inline struct dentry * __dget_locked(struct dentry *dentry)
267{
268 atomic_inc(&dentry->d_count);
269 if (!list_empty(&dentry->d_lru)) {
270 dentry_stat.nr_unused--;
271 list_del_init(&dentry->d_lru);
272 }
273 return dentry;
274}
275
276struct dentry * dget_locked(struct dentry *dentry)
277{
278 return __dget_locked(dentry);
279}
280
281/**
282 * d_find_alias - grab a hashed alias of inode
283 * @inode: inode in question
284 * @want_discon: flag, used by d_splice_alias, to request
285 * that only a DISCONNECTED alias be returned.
286 *
287 * If inode has a hashed alias, or is a directory and has any alias,
288 * acquire the reference to alias and return it. Otherwise return NULL.
289 * Notice that if inode is a directory there can be only one alias and
290 * it can be unhashed only if it has no children, or if it is the root
291 * of a filesystem.
292 *
293 * If the inode has a DCACHE_DISCONNECTED alias, then prefer
294 * any other hashed alias over that one unless @want_discon is set,
295 * in which case only return a DCACHE_DISCONNECTED alias.
296 */
297
298static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
299{
300 struct list_head *head, *next, *tmp;
301 struct dentry *alias, *discon_alias=NULL;
302
303 head = &inode->i_dentry;
304 next = inode->i_dentry.next;
305 while (next != head) {
306 tmp = next;
307 next = tmp->next;
308 prefetch(next);
309 alias = list_entry(tmp, struct dentry, d_alias);
310 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
311 if (alias->d_flags & DCACHE_DISCONNECTED)
312 discon_alias = alias;
313 else if (!want_discon) {
314 __dget_locked(alias);
315 return alias;
316 }
317 }
318 }
319 if (discon_alias)
320 __dget_locked(discon_alias);
321 return discon_alias;
322}
323
324struct dentry * d_find_alias(struct inode *inode)
325{
David Howells214fda12006-03-25 03:06:36 -0800326 struct dentry *de = NULL;
327
328 if (!list_empty(&inode->i_dentry)) {
329 spin_lock(&dcache_lock);
330 de = __d_find_alias(inode, 0);
331 spin_unlock(&dcache_lock);
332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 return de;
334}
335
336/*
337 * Try to kill dentries associated with this inode.
338 * WARNING: you must own a reference to inode.
339 */
340void d_prune_aliases(struct inode *inode)
341{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700342 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343restart:
344 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700345 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 spin_lock(&dentry->d_lock);
347 if (!atomic_read(&dentry->d_count)) {
348 __dget_locked(dentry);
349 __d_drop(dentry);
350 spin_unlock(&dentry->d_lock);
351 spin_unlock(&dcache_lock);
352 dput(dentry);
353 goto restart;
354 }
355 spin_unlock(&dentry->d_lock);
356 }
357 spin_unlock(&dcache_lock);
358}
359
360/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700361 * Throw away a dentry - free the inode, dput the parent. This requires that
362 * the LRU list has already been removed.
363 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 * Called with dcache_lock, drops it and then regains.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700365 * Called with dentry->d_lock held, drops it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 */
Andrew Mortond702ccb2006-06-22 14:47:31 -0700367static void prune_one_dentry(struct dentry * dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368{
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
NeilBrown0feae5c2006-06-22 14:47:28 -0700385 * @sb: if given, ignore dentries for other superblocks
386 * which are being unmounted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 *
388 * Shrink the dcache. This is done when we need
389 * more memory, or simply when we need to unmount
390 * something (at which point we need to unuse
391 * all dentries).
392 *
393 * This function may fail to free any resources if
394 * all the dentries are in use.
395 */
396
NeilBrown0feae5c2006-06-22 14:47:28 -0700397static void prune_dcache(int count, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
399 spin_lock(&dcache_lock);
400 for (; count ; count--) {
401 struct dentry *dentry;
402 struct list_head *tmp;
NeilBrown0feae5c2006-06-22 14:47:28 -0700403 struct rw_semaphore *s_umount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405 cond_resched_lock(&dcache_lock);
406
407 tmp = dentry_unused.prev;
Hua Zhongf58a1eb2006-06-25 05:49:32 -0700408 if (sb) {
NeilBrown0feae5c2006-06-22 14:47:28 -0700409 /* Try to find a dentry for this sb, but don't try
410 * too hard, if they aren't near the tail they will
411 * be moved down again soon
412 */
413 int skip = count;
414 while (skip && tmp != &dentry_unused &&
415 list_entry(tmp, struct dentry, d_lru)->d_sb != sb) {
416 skip--;
417 tmp = tmp->prev;
418 }
419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 if (tmp == &dentry_unused)
421 break;
422 list_del_init(tmp);
423 prefetch(dentry_unused.prev);
424 dentry_stat.nr_unused--;
425 dentry = list_entry(tmp, struct dentry, d_lru);
426
427 spin_lock(&dentry->d_lock);
428 /*
429 * We found an inuse dentry which was not removed from
430 * dentry_unused because of laziness during lookup. Do not free
431 * it - just keep it off the dentry_unused list.
432 */
433 if (atomic_read(&dentry->d_count)) {
434 spin_unlock(&dentry->d_lock);
435 continue;
436 }
437 /* If the dentry was recently referenced, don't free it. */
438 if (dentry->d_flags & DCACHE_REFERENCED) {
439 dentry->d_flags &= ~DCACHE_REFERENCED;
440 list_add(&dentry->d_lru, &dentry_unused);
441 dentry_stat.nr_unused++;
442 spin_unlock(&dentry->d_lock);
443 continue;
444 }
NeilBrown0feae5c2006-06-22 14:47:28 -0700445 /*
446 * If the dentry is not DCACHED_REFERENCED, it is time
447 * to remove it from the dcache, provided the super block is
448 * NULL (which means we are trying to reclaim memory)
449 * or this dentry belongs to the same super block that
450 * we want to shrink.
451 */
452 /*
453 * If this dentry is for "my" filesystem, then I can prune it
454 * without taking the s_umount lock (I already hold it).
455 */
456 if (sb && dentry->d_sb == sb) {
457 prune_one_dentry(dentry);
458 continue;
459 }
460 /*
461 * ...otherwise we need to be sure this filesystem isn't being
462 * unmounted, otherwise we could race with
463 * generic_shutdown_super(), and end up holding a reference to
464 * an inode while the filesystem is unmounted.
465 * So we try to get s_umount, and make sure s_root isn't NULL.
466 * (Take a local copy of s_umount to avoid a use-after-free of
467 * `dentry').
468 */
469 s_umount = &dentry->d_sb->s_umount;
470 if (down_read_trylock(s_umount)) {
471 if (dentry->d_sb->s_root != NULL) {
472 prune_one_dentry(dentry);
473 up_read(s_umount);
474 continue;
475 }
476 up_read(s_umount);
477 }
478 spin_unlock(&dentry->d_lock);
479 /* Cannot remove the first dentry, and it isn't appropriate
480 * to move it to the head of the list, so give up, and try
481 * later
482 */
483 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
485 spin_unlock(&dcache_lock);
486}
487
488/*
489 * Shrink the dcache for the specified super block.
490 * This allows us to unmount a device without disturbing
491 * the dcache for the other devices.
492 *
493 * This implementation makes just two traversals of the
494 * unused list. On the first pass we move the selected
495 * dentries to the most recent end, and on the second
496 * pass we free them. The second pass must restart after
497 * each dput(), but since the target dentries are all at
498 * the end, it's really just a single traversal.
499 */
500
501/**
502 * shrink_dcache_sb - shrink dcache for a superblock
503 * @sb: superblock
504 *
505 * Shrink the dcache for the specified super block. This
506 * is used to free the dcache before unmounting a file
507 * system
508 */
509
510void shrink_dcache_sb(struct super_block * sb)
511{
512 struct list_head *tmp, *next;
513 struct dentry *dentry;
514
515 /*
516 * Pass one ... move the dentries for the specified
517 * superblock to the most recent end of the unused list.
518 */
519 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700520 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 dentry = list_entry(tmp, struct dentry, d_lru);
522 if (dentry->d_sb != sb)
523 continue;
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700524 list_move(tmp, &dentry_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 }
526
527 /*
528 * Pass two ... free the dentries for this superblock.
529 */
530repeat:
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700531 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 dentry = list_entry(tmp, struct dentry, d_lru);
533 if (dentry->d_sb != sb)
534 continue;
535 dentry_stat.nr_unused--;
536 list_del_init(tmp);
537 spin_lock(&dentry->d_lock);
538 if (atomic_read(&dentry->d_count)) {
539 spin_unlock(&dentry->d_lock);
540 continue;
541 }
542 prune_one_dentry(dentry);
Kirill Korotaev2ab13462006-03-25 03:07:45 -0800543 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 goto repeat;
545 }
546 spin_unlock(&dcache_lock);
547}
548
549/*
550 * Search for at least 1 mount point in the dentry's subdirs.
551 * We descend to the next level whenever the d_subdirs
552 * list is non-empty and continue searching.
553 */
554
555/**
556 * have_submounts - check for mounts over a dentry
557 * @parent: dentry to check.
558 *
559 * Return true if the parent or its subdirectories contain
560 * a mount point
561 */
562
563int have_submounts(struct dentry *parent)
564{
565 struct dentry *this_parent = parent;
566 struct list_head *next;
567
568 spin_lock(&dcache_lock);
569 if (d_mountpoint(parent))
570 goto positive;
571repeat:
572 next = this_parent->d_subdirs.next;
573resume:
574 while (next != &this_parent->d_subdirs) {
575 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800576 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 next = tmp->next;
578 /* Have we found a mount point ? */
579 if (d_mountpoint(dentry))
580 goto positive;
581 if (!list_empty(&dentry->d_subdirs)) {
582 this_parent = dentry;
583 goto repeat;
584 }
585 }
586 /*
587 * All done at this level ... ascend and resume the search.
588 */
589 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800590 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 this_parent = this_parent->d_parent;
592 goto resume;
593 }
594 spin_unlock(&dcache_lock);
595 return 0; /* No mount points found in tree */
596positive:
597 spin_unlock(&dcache_lock);
598 return 1;
599}
600
601/*
602 * Search the dentry child list for the specified parent,
603 * and move any unused dentries to the end of the unused
604 * list for prune_dcache(). We descend to the next level
605 * whenever the d_subdirs list is non-empty and continue
606 * searching.
607 *
608 * It returns zero iff there are no unused children,
609 * otherwise it returns the number of children moved to
610 * the end of the unused list. This may not be the total
611 * number of unused children, because select_parent can
612 * drop the lock and return early due to latency
613 * constraints.
614 */
615static int select_parent(struct dentry * parent)
616{
617 struct dentry *this_parent = parent;
618 struct list_head *next;
619 int found = 0;
620
621 spin_lock(&dcache_lock);
622repeat:
623 next = this_parent->d_subdirs.next;
624resume:
625 while (next != &this_parent->d_subdirs) {
626 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800627 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 next = tmp->next;
629
630 if (!list_empty(&dentry->d_lru)) {
631 dentry_stat.nr_unused--;
632 list_del_init(&dentry->d_lru);
633 }
634 /*
635 * move only zero ref count dentries to the end
636 * of the unused list for prune_dcache
637 */
638 if (!atomic_read(&dentry->d_count)) {
Akinobu Mita8e130592006-06-26 00:24:37 -0700639 list_add_tail(&dentry->d_lru, &dentry_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 dentry_stat.nr_unused++;
641 found++;
642 }
643
644 /*
645 * We can return to the caller if we have found some (this
646 * ensures forward progress). We'll be coming back to find
647 * the rest.
648 */
649 if (found && need_resched())
650 goto out;
651
652 /*
653 * Descend a level if the d_subdirs list is non-empty.
654 */
655 if (!list_empty(&dentry->d_subdirs)) {
656 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 goto repeat;
658 }
659 }
660 /*
661 * All done at this level ... ascend and resume the search.
662 */
663 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800664 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 goto resume;
667 }
668out:
669 spin_unlock(&dcache_lock);
670 return found;
671}
672
673/**
674 * shrink_dcache_parent - prune dcache
675 * @parent: parent of entries to prune
676 *
677 * Prune the dcache to remove unused children of the parent dentry.
678 */
679
680void shrink_dcache_parent(struct dentry * parent)
681{
682 int found;
683
684 while ((found = select_parent(parent)) != 0)
NeilBrown0feae5c2006-06-22 14:47:28 -0700685 prune_dcache(found, parent->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686}
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688/*
689 * Scan `nr' dentries and return the number which remain.
690 *
691 * We need to avoid reentering the filesystem if the caller is performing a
692 * GFP_NOFS allocation attempt. One example deadlock is:
693 *
694 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
695 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
696 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
697 *
698 * In this case we return -1 to tell the caller that we baled.
699 */
Al Viro27496a82005-10-21 03:20:48 -0400700static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
702 if (nr) {
703 if (!(gfp_mask & __GFP_FS))
704 return -1;
NeilBrown0feae5c2006-06-22 14:47:28 -0700705 prune_dcache(nr, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
708}
709
710/**
711 * d_alloc - allocate a dcache entry
712 * @parent: parent of entry to allocate
713 * @name: qstr of the name
714 *
715 * Allocates a dentry. It returns %NULL if there is insufficient memory
716 * available. On a success the dentry is returned. The name passed in is
717 * copied and the copy passed in may be reused after this call.
718 */
719
720struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
721{
722 struct dentry *dentry;
723 char *dname;
724
725 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
726 if (!dentry)
727 return NULL;
728
729 if (name->len > DNAME_INLINE_LEN-1) {
730 dname = kmalloc(name->len + 1, GFP_KERNEL);
731 if (!dname) {
732 kmem_cache_free(dentry_cache, dentry);
733 return NULL;
734 }
735 } else {
736 dname = dentry->d_iname;
737 }
738 dentry->d_name.name = dname;
739
740 dentry->d_name.len = name->len;
741 dentry->d_name.hash = name->hash;
742 memcpy(dname, name->name, name->len);
743 dname[name->len] = 0;
744
745 atomic_set(&dentry->d_count, 1);
746 dentry->d_flags = DCACHE_UNHASHED;
747 spin_lock_init(&dentry->d_lock);
748 dentry->d_inode = NULL;
749 dentry->d_parent = NULL;
750 dentry->d_sb = NULL;
751 dentry->d_op = NULL;
752 dentry->d_fsdata = NULL;
753 dentry->d_mounted = 0;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800754#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800756#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 INIT_HLIST_NODE(&dentry->d_hash);
758 INIT_LIST_HEAD(&dentry->d_lru);
759 INIT_LIST_HEAD(&dentry->d_subdirs);
760 INIT_LIST_HEAD(&dentry->d_alias);
761
762 if (parent) {
763 dentry->d_parent = dget(parent);
764 dentry->d_sb = parent->d_sb;
765 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800766 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 }
768
769 spin_lock(&dcache_lock);
770 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800771 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 dentry_stat.nr_dentry++;
773 spin_unlock(&dcache_lock);
774
775 return dentry;
776}
777
778struct dentry *d_alloc_name(struct dentry *parent, const char *name)
779{
780 struct qstr q;
781
782 q.name = name;
783 q.len = strlen(name);
784 q.hash = full_name_hash(q.name, q.len);
785 return d_alloc(parent, &q);
786}
787
788/**
789 * d_instantiate - fill in inode information for a dentry
790 * @entry: dentry to complete
791 * @inode: inode to attach to this dentry
792 *
793 * Fill in inode information in the entry.
794 *
795 * This turns negative dentries into productive full members
796 * of society.
797 *
798 * NOTE! This assumes that the inode count has been incremented
799 * (or otherwise set) by the caller to indicate that it is now
800 * in use by the dcache.
801 */
802
803void d_instantiate(struct dentry *entry, struct inode * inode)
804{
Eric Sesterhenn28133c72006-03-26 18:25:39 +0200805 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 spin_lock(&dcache_lock);
807 if (inode)
808 list_add(&entry->d_alias, &inode->i_dentry);
809 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800810 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 spin_unlock(&dcache_lock);
812 security_d_instantiate(entry, inode);
813}
814
815/**
816 * d_instantiate_unique - instantiate a non-aliased dentry
817 * @entry: dentry to instantiate
818 * @inode: inode to attach to this dentry
819 *
820 * Fill in inode information in the entry. On success, it returns NULL.
821 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -0800822 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 *
824 * Note that in order to avoid conflicts with rename() etc, the caller
825 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -0800826 *
827 * This also assumes that the inode count has been incremented
828 * (or otherwise set) by the caller to indicate that it is now
829 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 */
831struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
832{
833 struct dentry *alias;
834 int len = entry->d_name.len;
835 const char *name = entry->d_name.name;
836 unsigned int hash = entry->d_name.hash;
837
838 BUG_ON(!list_empty(&entry->d_alias));
839 spin_lock(&dcache_lock);
840 if (!inode)
841 goto do_negative;
842 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
843 struct qstr *qstr = &alias->d_name;
844
845 if (qstr->hash != hash)
846 continue;
847 if (alias->d_parent != entry->d_parent)
848 continue;
849 if (qstr->len != len)
850 continue;
851 if (memcmp(qstr->name, name, len))
852 continue;
853 dget_locked(alias);
854 spin_unlock(&dcache_lock);
855 BUG_ON(!d_unhashed(alias));
Oleg Drokine866cfa2006-01-09 20:52:51 -0800856 iput(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 return alias;
858 }
859 list_add(&entry->d_alias, &inode->i_dentry);
860do_negative:
861 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800862 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 spin_unlock(&dcache_lock);
864 security_d_instantiate(entry, inode);
865 return NULL;
866}
867EXPORT_SYMBOL(d_instantiate_unique);
868
869/**
870 * d_alloc_root - allocate root dentry
871 * @root_inode: inode to allocate the root for
872 *
873 * Allocate a root ("/") dentry for the inode given. The inode is
874 * instantiated and returned. %NULL is returned if there is insufficient
875 * memory or the inode passed is %NULL.
876 */
877
878struct dentry * d_alloc_root(struct inode * root_inode)
879{
880 struct dentry *res = NULL;
881
882 if (root_inode) {
883 static const struct qstr name = { .name = "/", .len = 1 };
884
885 res = d_alloc(NULL, &name);
886 if (res) {
887 res->d_sb = root_inode->i_sb;
888 res->d_parent = res;
889 d_instantiate(res, root_inode);
890 }
891 }
892 return res;
893}
894
895static inline struct hlist_head *d_hash(struct dentry *parent,
896 unsigned long hash)
897{
898 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
899 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
900 return dentry_hashtable + (hash & D_HASHMASK);
901}
902
903/**
904 * d_alloc_anon - allocate an anonymous dentry
905 * @inode: inode to allocate the dentry for
906 *
907 * This is similar to d_alloc_root. It is used by filesystems when
908 * creating a dentry for a given inode, often in the process of
909 * mapping a filehandle to a dentry. The returned dentry may be
910 * anonymous, or may have a full name (if the inode was already
911 * in the cache). The file system may need to make further
912 * efforts to connect this dentry into the dcache properly.
913 *
914 * When called on a directory inode, we must ensure that
915 * the inode only ever has one dentry. If a dentry is
916 * found, that is returned instead of allocating a new one.
917 *
918 * On successful return, the reference to the inode has been transferred
919 * to the dentry. If %NULL is returned (indicating kmalloc failure),
920 * the reference on the inode has not been released.
921 */
922
923struct dentry * d_alloc_anon(struct inode *inode)
924{
925 static const struct qstr anonstring = { .name = "" };
926 struct dentry *tmp;
927 struct dentry *res;
928
929 if ((res = d_find_alias(inode))) {
930 iput(inode);
931 return res;
932 }
933
934 tmp = d_alloc(NULL, &anonstring);
935 if (!tmp)
936 return NULL;
937
938 tmp->d_parent = tmp; /* make sure dput doesn't croak */
939
940 spin_lock(&dcache_lock);
941 res = __d_find_alias(inode, 0);
942 if (!res) {
943 /* attach a disconnected dentry */
944 res = tmp;
945 tmp = NULL;
946 spin_lock(&res->d_lock);
947 res->d_sb = inode->i_sb;
948 res->d_parent = res;
949 res->d_inode = inode;
950 res->d_flags |= DCACHE_DISCONNECTED;
951 res->d_flags &= ~DCACHE_UNHASHED;
952 list_add(&res->d_alias, &inode->i_dentry);
953 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
954 spin_unlock(&res->d_lock);
955
956 inode = NULL; /* don't drop reference */
957 }
958 spin_unlock(&dcache_lock);
959
960 if (inode)
961 iput(inode);
962 if (tmp)
963 dput(tmp);
964 return res;
965}
966
967
968/**
969 * d_splice_alias - splice a disconnected dentry into the tree if one exists
970 * @inode: the inode which may have a disconnected dentry
971 * @dentry: a negative dentry which we want to point to the inode.
972 *
973 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
974 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
975 * and return it, else simply d_add the inode to the dentry and return NULL.
976 *
977 * This is needed in the lookup routine of any filesystem that is exportable
978 * (via knfsd) so that we can build dcache paths to directories effectively.
979 *
980 * If a dentry was found and moved, then it is returned. Otherwise NULL
981 * is returned. This matches the expected return value of ->lookup.
982 *
983 */
984struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
985{
986 struct dentry *new = NULL;
987
988 if (inode) {
989 spin_lock(&dcache_lock);
990 new = __d_find_alias(inode, 1);
991 if (new) {
992 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Pigginc32ccd82006-03-25 03:07:09 -0800993 fsnotify_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 spin_unlock(&dcache_lock);
995 security_d_instantiate(new, inode);
996 d_rehash(dentry);
997 d_move(new, dentry);
998 iput(inode);
999 } else {
1000 /* d_instantiate takes dcache_lock, so we do it by hand */
1001 list_add(&dentry->d_alias, &inode->i_dentry);
1002 dentry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001003 fsnotify_d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 spin_unlock(&dcache_lock);
1005 security_d_instantiate(dentry, inode);
1006 d_rehash(dentry);
1007 }
1008 } else
1009 d_add(dentry, inode);
1010 return new;
1011}
1012
1013
1014/**
1015 * d_lookup - search for a dentry
1016 * @parent: parent dentry
1017 * @name: qstr of name we wish to find
1018 *
1019 * Searches the children of the parent dentry for the name in question. If
1020 * the dentry is found its reference count is incremented and the dentry
1021 * is returned. The caller must use d_put to free the entry when it has
1022 * finished using it. %NULL is returned on failure.
1023 *
1024 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1025 * Memory barriers are used while updating and doing lockless traversal.
1026 * To avoid races with d_move while rename is happening, d_lock is used.
1027 *
1028 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1029 * and name pointer in one structure pointed by d_qstr.
1030 *
1031 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1032 * lookup is going on.
1033 *
1034 * dentry_unused list is not updated even if lookup finds the required dentry
1035 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1036 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1037 * acquisition.
1038 *
1039 * d_lookup() is protected against the concurrent renames in some unrelated
1040 * directory using the seqlockt_t rename_lock.
1041 */
1042
1043struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1044{
1045 struct dentry * dentry = NULL;
1046 unsigned long seq;
1047
1048 do {
1049 seq = read_seqbegin(&rename_lock);
1050 dentry = __d_lookup(parent, name);
1051 if (dentry)
1052 break;
1053 } while (read_seqretry(&rename_lock, seq));
1054 return dentry;
1055}
1056
1057struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1058{
1059 unsigned int len = name->len;
1060 unsigned int hash = name->hash;
1061 const unsigned char *str = name->name;
1062 struct hlist_head *head = d_hash(parent,hash);
1063 struct dentry *found = NULL;
1064 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001065 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067 rcu_read_lock();
1068
Paul E. McKenney665a7582005-11-07 00:59:17 -08001069 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 struct qstr *qstr;
1071
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (dentry->d_name.hash != hash)
1073 continue;
1074 if (dentry->d_parent != parent)
1075 continue;
1076
1077 spin_lock(&dentry->d_lock);
1078
1079 /*
1080 * Recheck the dentry after taking the lock - d_move may have
1081 * changed things. Don't bother checking the hash because we're
1082 * about to compare the whole name anyway.
1083 */
1084 if (dentry->d_parent != parent)
1085 goto next;
1086
1087 /*
1088 * It is safe to compare names since d_move() cannot
1089 * change the qstr (protected by d_lock).
1090 */
1091 qstr = &dentry->d_name;
1092 if (parent->d_op && parent->d_op->d_compare) {
1093 if (parent->d_op->d_compare(parent, qstr, name))
1094 goto next;
1095 } else {
1096 if (qstr->len != len)
1097 goto next;
1098 if (memcmp(qstr->name, str, len))
1099 goto next;
1100 }
1101
1102 if (!d_unhashed(dentry)) {
1103 atomic_inc(&dentry->d_count);
1104 found = dentry;
1105 }
1106 spin_unlock(&dentry->d_lock);
1107 break;
1108next:
1109 spin_unlock(&dentry->d_lock);
1110 }
1111 rcu_read_unlock();
1112
1113 return found;
1114}
1115
1116/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001117 * d_hash_and_lookup - hash the qstr then search for a dentry
1118 * @dir: Directory to search in
1119 * @name: qstr of name we wish to find
1120 *
1121 * On hash failure or on lookup failure NULL is returned.
1122 */
1123struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1124{
1125 struct dentry *dentry = NULL;
1126
1127 /*
1128 * Check for a fs-specific hash function. Note that we must
1129 * calculate the standard hash first, as the d_op->d_hash()
1130 * routine may choose to leave the hash value unchanged.
1131 */
1132 name->hash = full_name_hash(name->name, name->len);
1133 if (dir->d_op && dir->d_op->d_hash) {
1134 if (dir->d_op->d_hash(dir, name) < 0)
1135 goto out;
1136 }
1137 dentry = d_lookup(dir, name);
1138out:
1139 return dentry;
1140}
1141
1142/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 * d_validate - verify dentry provided from insecure source
1144 * @dentry: The dentry alleged to be valid child of @dparent
1145 * @dparent: The parent dentry (known to be valid)
1146 * @hash: Hash of the dentry
1147 * @len: Length of the name
1148 *
1149 * An insecure source has sent us a dentry, here we verify it and dget() it.
1150 * This is used by ncpfs in its readdir implementation.
1151 * Zero is returned in the dentry is invalid.
1152 */
1153
1154int d_validate(struct dentry *dentry, struct dentry *dparent)
1155{
1156 struct hlist_head *base;
1157 struct hlist_node *lhp;
1158
1159 /* Check whether the ptr might be valid at all.. */
1160 if (!kmem_ptr_validate(dentry_cache, dentry))
1161 goto out;
1162
1163 if (dentry->d_parent != dparent)
1164 goto out;
1165
1166 spin_lock(&dcache_lock);
1167 base = d_hash(dparent, dentry->d_name.hash);
1168 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001169 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 * as it is parsed under dcache_lock
1171 */
1172 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1173 __dget_locked(dentry);
1174 spin_unlock(&dcache_lock);
1175 return 1;
1176 }
1177 }
1178 spin_unlock(&dcache_lock);
1179out:
1180 return 0;
1181}
1182
1183/*
1184 * When a file is deleted, we have two options:
1185 * - turn this dentry into a negative dentry
1186 * - unhash this dentry and free it.
1187 *
1188 * Usually, we want to just turn this into
1189 * a negative dentry, but if anybody else is
1190 * currently using the dentry or the inode
1191 * we can't do that and we fall back on removing
1192 * it from the hash queues and waiting for
1193 * it to be deleted later when it has no users
1194 */
1195
1196/**
1197 * d_delete - delete a dentry
1198 * @dentry: The dentry to delete
1199 *
1200 * Turn the dentry into a negative dentry if possible, otherwise
1201 * remove it from the hash queues so it can be deleted later
1202 */
1203
1204void d_delete(struct dentry * dentry)
1205{
John McCutchan7a91bf72005-08-08 13:52:16 -04001206 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 /*
1208 * Are we the only user?
1209 */
1210 spin_lock(&dcache_lock);
1211 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001212 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 if (atomic_read(&dentry->d_count) == 1) {
1214 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001215 fsnotify_nameremove(dentry, isdir);
Amy Griffis7a2bd3f2006-03-31 02:30:54 -08001216
1217 /* remove this and other inotify debug checks after 2.6.18 */
1218 dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 return;
1220 }
1221
1222 if (!d_unhashed(dentry))
1223 __d_drop(dentry);
1224
1225 spin_unlock(&dentry->d_lock);
1226 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001227
1228 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229}
1230
1231static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1232{
1233
1234 entry->d_flags &= ~DCACHE_UNHASHED;
1235 hlist_add_head_rcu(&entry->d_hash, list);
1236}
1237
1238/**
1239 * d_rehash - add an entry back to the hash
1240 * @entry: dentry to add to the hash
1241 *
1242 * Adds a dentry to the hash according to its name.
1243 */
1244
1245void d_rehash(struct dentry * entry)
1246{
1247 struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash);
1248
1249 spin_lock(&dcache_lock);
1250 spin_lock(&entry->d_lock);
1251 __d_rehash(entry, list);
1252 spin_unlock(&entry->d_lock);
1253 spin_unlock(&dcache_lock);
1254}
1255
1256#define do_switch(x,y) do { \
1257 __typeof__ (x) __tmp = x; \
1258 x = y; y = __tmp; } while (0)
1259
1260/*
1261 * When switching names, the actual string doesn't strictly have to
1262 * be preserved in the target - because we're dropping the target
1263 * anyway. As such, we can just do a simple memcpy() to copy over
1264 * the new name before we switch.
1265 *
1266 * Note that we have to be a lot more careful about getting the hash
1267 * switched - we have to switch the hash value properly even if it
1268 * then no longer matches the actual (corrupted) string of the target.
1269 * The hash value has to match the hash queue that the dentry is on..
1270 */
1271static void switch_names(struct dentry *dentry, struct dentry *target)
1272{
1273 if (dname_external(target)) {
1274 if (dname_external(dentry)) {
1275 /*
1276 * Both external: swap the pointers
1277 */
1278 do_switch(target->d_name.name, dentry->d_name.name);
1279 } else {
1280 /*
1281 * dentry:internal, target:external. Steal target's
1282 * storage and make target internal.
1283 */
1284 dentry->d_name.name = target->d_name.name;
1285 target->d_name.name = target->d_iname;
1286 }
1287 } else {
1288 if (dname_external(dentry)) {
1289 /*
1290 * dentry:external, target:internal. Give dentry's
1291 * storage to target and make dentry internal
1292 */
1293 memcpy(dentry->d_iname, target->d_name.name,
1294 target->d_name.len + 1);
1295 target->d_name.name = dentry->d_name.name;
1296 dentry->d_name.name = dentry->d_iname;
1297 } else {
1298 /*
1299 * Both are internal. Just copy target to dentry
1300 */
1301 memcpy(dentry->d_iname, target->d_name.name,
1302 target->d_name.len + 1);
1303 }
1304 }
1305}
1306
1307/*
1308 * We cannibalize "target" when moving dentry on top of it,
1309 * because it's going to be thrown away anyway. We could be more
1310 * polite about it, though.
1311 *
1312 * This forceful removal will result in ugly /proc output if
1313 * somebody holds a file open that got deleted due to a rename.
1314 * We could be nicer about the deleted file, and let it show
1315 * up under the name it got deleted rather than the name that
1316 * deleted it.
1317 */
1318
1319/**
1320 * d_move - move a dentry
1321 * @dentry: entry to move
1322 * @target: new dentry
1323 *
1324 * Update the dcache to reflect the move of a file name. Negative
1325 * dcache entries should not be moved in this way.
1326 */
1327
1328void d_move(struct dentry * dentry, struct dentry * target)
1329{
1330 struct hlist_head *list;
1331
1332 if (!dentry->d_inode)
1333 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1334
1335 spin_lock(&dcache_lock);
1336 write_seqlock(&rename_lock);
1337 /*
1338 * XXXX: do we really need to take target->d_lock?
1339 */
1340 if (target < dentry) {
1341 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001342 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 } else {
1344 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001345 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 }
1347
1348 /* Move the dentry to the target hash queue, if on different bucket */
1349 if (dentry->d_flags & DCACHE_UNHASHED)
1350 goto already_unhashed;
1351
1352 hlist_del_rcu(&dentry->d_hash);
1353
1354already_unhashed:
1355 list = d_hash(target->d_parent, target->d_name.hash);
1356 __d_rehash(dentry, list);
1357
1358 /* Unhash the target: dput() will then get rid of it */
1359 __d_drop(target);
1360
Eric Dumazet5160ee62006-01-08 01:03:32 -08001361 list_del(&dentry->d_u.d_child);
1362 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
1364 /* Switch the names.. */
1365 switch_names(dentry, target);
1366 do_switch(dentry->d_name.len, target->d_name.len);
1367 do_switch(dentry->d_name.hash, target->d_name.hash);
1368
1369 /* ... and switch the parents */
1370 if (IS_ROOT(dentry)) {
1371 dentry->d_parent = target->d_parent;
1372 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001373 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 } else {
1375 do_switch(dentry->d_parent, target->d_parent);
1376
1377 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001378 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 }
1380
Eric Dumazet5160ee62006-01-08 01:03:32 -08001381 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001383 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 spin_unlock(&dentry->d_lock);
1385 write_sequnlock(&rename_lock);
1386 spin_unlock(&dcache_lock);
1387}
1388
1389/**
1390 * d_path - return the path of a dentry
1391 * @dentry: dentry to report
1392 * @vfsmnt: vfsmnt to which the dentry belongs
1393 * @root: root dentry
1394 * @rootmnt: vfsmnt to which the root dentry belongs
1395 * @buffer: buffer to return value in
1396 * @buflen: buffer length
1397 *
1398 * Convert a dentry into an ASCII path name. If the entry has been deleted
1399 * the string " (deleted)" is appended. Note that this is ambiguous.
1400 *
1401 * Returns the buffer or an error code if the path was too long.
1402 *
1403 * "buflen" should be positive. Caller holds the dcache_lock.
1404 */
1405static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1406 struct dentry *root, struct vfsmount *rootmnt,
1407 char *buffer, int buflen)
1408{
1409 char * end = buffer+buflen;
1410 char * retval;
1411 int namelen;
1412
1413 *--end = '\0';
1414 buflen--;
1415 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
1416 buflen -= 10;
1417 end -= 10;
1418 if (buflen < 0)
1419 goto Elong;
1420 memcpy(end, " (deleted)", 10);
1421 }
1422
1423 if (buflen < 1)
1424 goto Elong;
1425 /* Get '/' right */
1426 retval = end-1;
1427 *retval = '/';
1428
1429 for (;;) {
1430 struct dentry * parent;
1431
1432 if (dentry == root && vfsmnt == rootmnt)
1433 break;
1434 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1435 /* Global root? */
1436 spin_lock(&vfsmount_lock);
1437 if (vfsmnt->mnt_parent == vfsmnt) {
1438 spin_unlock(&vfsmount_lock);
1439 goto global_root;
1440 }
1441 dentry = vfsmnt->mnt_mountpoint;
1442 vfsmnt = vfsmnt->mnt_parent;
1443 spin_unlock(&vfsmount_lock);
1444 continue;
1445 }
1446 parent = dentry->d_parent;
1447 prefetch(parent);
1448 namelen = dentry->d_name.len;
1449 buflen -= namelen + 1;
1450 if (buflen < 0)
1451 goto Elong;
1452 end -= namelen;
1453 memcpy(end, dentry->d_name.name, namelen);
1454 *--end = '/';
1455 retval = end;
1456 dentry = parent;
1457 }
1458
1459 return retval;
1460
1461global_root:
1462 namelen = dentry->d_name.len;
1463 buflen -= namelen;
1464 if (buflen < 0)
1465 goto Elong;
1466 retval -= namelen-1; /* hit the slash */
1467 memcpy(retval, dentry->d_name.name, namelen);
1468 return retval;
1469Elong:
1470 return ERR_PTR(-ENAMETOOLONG);
1471}
1472
1473/* write full pathname into buffer and return start of pathname */
1474char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1475 char *buf, int buflen)
1476{
1477 char *res;
1478 struct vfsmount *rootmnt;
1479 struct dentry *root;
1480
1481 read_lock(&current->fs->lock);
1482 rootmnt = mntget(current->fs->rootmnt);
1483 root = dget(current->fs->root);
1484 read_unlock(&current->fs->lock);
1485 spin_lock(&dcache_lock);
1486 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1487 spin_unlock(&dcache_lock);
1488 dput(root);
1489 mntput(rootmnt);
1490 return res;
1491}
1492
1493/*
1494 * NOTE! The user-level library version returns a
1495 * character pointer. The kernel system call just
1496 * returns the length of the buffer filled (which
1497 * includes the ending '\0' character), or a negative
1498 * error value. So libc would do something like
1499 *
1500 * char *getcwd(char * buf, size_t size)
1501 * {
1502 * int retval;
1503 *
1504 * retval = sys_getcwd(buf, size);
1505 * if (retval >= 0)
1506 * return buf;
1507 * errno = -retval;
1508 * return NULL;
1509 * }
1510 */
1511asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1512{
1513 int error;
1514 struct vfsmount *pwdmnt, *rootmnt;
1515 struct dentry *pwd, *root;
1516 char *page = (char *) __get_free_page(GFP_USER);
1517
1518 if (!page)
1519 return -ENOMEM;
1520
1521 read_lock(&current->fs->lock);
1522 pwdmnt = mntget(current->fs->pwdmnt);
1523 pwd = dget(current->fs->pwd);
1524 rootmnt = mntget(current->fs->rootmnt);
1525 root = dget(current->fs->root);
1526 read_unlock(&current->fs->lock);
1527
1528 error = -ENOENT;
1529 /* Has the current directory has been unlinked? */
1530 spin_lock(&dcache_lock);
1531 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1532 unsigned long len;
1533 char * cwd;
1534
1535 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1536 spin_unlock(&dcache_lock);
1537
1538 error = PTR_ERR(cwd);
1539 if (IS_ERR(cwd))
1540 goto out;
1541
1542 error = -ERANGE;
1543 len = PAGE_SIZE + page - cwd;
1544 if (len <= size) {
1545 error = len;
1546 if (copy_to_user(buf, cwd, len))
1547 error = -EFAULT;
1548 }
1549 } else
1550 spin_unlock(&dcache_lock);
1551
1552out:
1553 dput(pwd);
1554 mntput(pwdmnt);
1555 dput(root);
1556 mntput(rootmnt);
1557 free_page((unsigned long) page);
1558 return error;
1559}
1560
1561/*
1562 * Test whether new_dentry is a subdirectory of old_dentry.
1563 *
1564 * Trivially implemented using the dcache structure
1565 */
1566
1567/**
1568 * is_subdir - is new dentry a subdirectory of old_dentry
1569 * @new_dentry: new dentry
1570 * @old_dentry: old dentry
1571 *
1572 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1573 * Returns 0 otherwise.
1574 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1575 */
1576
1577int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1578{
1579 int result;
1580 struct dentry * saved = new_dentry;
1581 unsigned long seq;
1582
1583 /* need rcu_readlock to protect against the d_parent trashing due to
1584 * d_move
1585 */
1586 rcu_read_lock();
1587 do {
1588 /* for restarting inner loop in case of seq retry */
1589 new_dentry = saved;
1590 result = 0;
1591 seq = read_seqbegin(&rename_lock);
1592 for (;;) {
1593 if (new_dentry != old_dentry) {
1594 struct dentry * parent = new_dentry->d_parent;
1595 if (parent == new_dentry)
1596 break;
1597 new_dentry = parent;
1598 continue;
1599 }
1600 result = 1;
1601 break;
1602 }
1603 } while (read_seqretry(&rename_lock, seq));
1604 rcu_read_unlock();
1605
1606 return result;
1607}
1608
1609void d_genocide(struct dentry *root)
1610{
1611 struct dentry *this_parent = root;
1612 struct list_head *next;
1613
1614 spin_lock(&dcache_lock);
1615repeat:
1616 next = this_parent->d_subdirs.next;
1617resume:
1618 while (next != &this_parent->d_subdirs) {
1619 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001620 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 next = tmp->next;
1622 if (d_unhashed(dentry)||!dentry->d_inode)
1623 continue;
1624 if (!list_empty(&dentry->d_subdirs)) {
1625 this_parent = dentry;
1626 goto repeat;
1627 }
1628 atomic_dec(&dentry->d_count);
1629 }
1630 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08001631 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 atomic_dec(&this_parent->d_count);
1633 this_parent = this_parent->d_parent;
1634 goto resume;
1635 }
1636 spin_unlock(&dcache_lock);
1637}
1638
1639/**
1640 * find_inode_number - check for dentry with name
1641 * @dir: directory to check
1642 * @name: Name to find.
1643 *
1644 * Check whether a dentry already exists for the given name,
1645 * and return the inode number if it has an inode. Otherwise
1646 * 0 is returned.
1647 *
1648 * This routine is used to post-process directory listings for
1649 * filesystems using synthetic inode numbers, and is necessary
1650 * to keep getcwd() working.
1651 */
1652
1653ino_t find_inode_number(struct dentry *dir, struct qstr *name)
1654{
1655 struct dentry * dentry;
1656 ino_t ino = 0;
1657
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001658 dentry = d_hash_and_lookup(dir, name);
1659 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 if (dentry->d_inode)
1661 ino = dentry->d_inode->i_ino;
1662 dput(dentry);
1663 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return ino;
1665}
1666
1667static __initdata unsigned long dhash_entries;
1668static int __init set_dhash_entries(char *str)
1669{
1670 if (!str)
1671 return 0;
1672 dhash_entries = simple_strtoul(str, &str, 0);
1673 return 1;
1674}
1675__setup("dhash_entries=", set_dhash_entries);
1676
1677static void __init dcache_init_early(void)
1678{
1679 int loop;
1680
1681 /* If hashes are distributed across NUMA nodes, defer
1682 * hash allocation until vmalloc space is available.
1683 */
1684 if (hashdist)
1685 return;
1686
1687 dentry_hashtable =
1688 alloc_large_system_hash("Dentry cache",
1689 sizeof(struct hlist_head),
1690 dhash_entries,
1691 13,
1692 HASH_EARLY,
1693 &d_hash_shift,
1694 &d_hash_mask,
1695 0);
1696
1697 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1698 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1699}
1700
1701static void __init dcache_init(unsigned long mempages)
1702{
1703 int loop;
1704
1705 /*
1706 * A constructor could be added for stable state like the lists,
1707 * but it is probably not worth it because of the cache nature
1708 * of the dcache.
1709 */
1710 dentry_cache = kmem_cache_create("dentry_cache",
1711 sizeof(struct dentry),
1712 0,
Paul Jacksonb0196002006-03-24 03:16:09 -08001713 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
1714 SLAB_MEM_SPREAD),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 NULL, NULL);
1716
1717 set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory);
1718
1719 /* Hash may have been set up in dcache_init_early */
1720 if (!hashdist)
1721 return;
1722
1723 dentry_hashtable =
1724 alloc_large_system_hash("Dentry cache",
1725 sizeof(struct hlist_head),
1726 dhash_entries,
1727 13,
1728 0,
1729 &d_hash_shift,
1730 &d_hash_mask,
1731 0);
1732
1733 for (loop = 0; loop < (1 << d_hash_shift); loop++)
1734 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
1735}
1736
1737/* SLAB cache for __getname() consumers */
Eric Dumazetfa3536c2006-03-26 01:37:24 -08001738kmem_cache_t *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740/* SLAB cache for file structures */
Eric Dumazetfa3536c2006-03-26 01:37:24 -08001741kmem_cache_t *filp_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
1743EXPORT_SYMBOL(d_genocide);
1744
1745extern void bdev_cache_init(void);
1746extern void chrdev_init(void);
1747
1748void __init vfs_caches_init_early(void)
1749{
1750 dcache_init_early();
1751 inode_init_early();
1752}
1753
1754void __init vfs_caches_init(unsigned long mempages)
1755{
1756 unsigned long reserve;
1757
1758 /* Base hash sizes on available memory, with a reserve equal to
1759 150% of current kernel size */
1760
1761 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
1762 mempages -= reserve;
1763
1764 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
1765 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1766
1767 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Dipankar Sarma529bf6b2006-03-07 21:55:35 -08001768 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
1770 dcache_init(mempages);
1771 inode_init(mempages);
1772 files_init(mempages);
1773 mnt_init(mempages);
1774 bdev_cache_init();
1775 chrdev_init();
1776}
1777
1778EXPORT_SYMBOL(d_alloc);
1779EXPORT_SYMBOL(d_alloc_anon);
1780EXPORT_SYMBOL(d_alloc_root);
1781EXPORT_SYMBOL(d_delete);
1782EXPORT_SYMBOL(d_find_alias);
1783EXPORT_SYMBOL(d_instantiate);
1784EXPORT_SYMBOL(d_invalidate);
1785EXPORT_SYMBOL(d_lookup);
1786EXPORT_SYMBOL(d_move);
1787EXPORT_SYMBOL(d_path);
1788EXPORT_SYMBOL(d_prune_aliases);
1789EXPORT_SYMBOL(d_rehash);
1790EXPORT_SYMBOL(d_splice_alias);
1791EXPORT_SYMBOL(d_validate);
1792EXPORT_SYMBOL(dget_locked);
1793EXPORT_SYMBOL(dput);
1794EXPORT_SYMBOL(find_inode_number);
1795EXPORT_SYMBOL(have_submounts);
1796EXPORT_SYMBOL(names_cachep);
1797EXPORT_SYMBOL(shrink_dcache_parent);
1798EXPORT_SYMBOL(shrink_dcache_sb);