blob: 920c87720385d34ea914f21c404c5955db08b38d [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
26#include <linux/module.h>
27#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
David Howells07f3f052006-09-30 20:52:18 +020034#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
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
Christoph Lametere18b8902006-12-06 20:33:20 -080045static struct kmem_cache *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
Eric Dumazetb3423412006-12-06 20:38:48 -080070static void __d_free(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 if (dname_external(dentry))
73 kfree(dentry->d_name.name);
74 kmem_cache_free(dentry_cache, dentry);
75}
76
Eric Dumazetb3423412006-12-06 20:38:48 -080077static void d_callback(struct rcu_head *head)
78{
79 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
80 __d_free(dentry);
81}
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083/*
84 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
85 * inside dcache_lock.
86 */
87static void d_free(struct dentry *dentry)
88{
89 if (dentry->d_op && dentry->d_op->d_release)
90 dentry->d_op->d_release(dentry);
Eric Dumazetb3423412006-12-06 20:38:48 -080091 /* if dentry was never inserted into hash, immediate free is OK */
92 if (dentry->d_hash.pprev == NULL)
93 __d_free(dentry);
94 else
95 call_rcu(&dentry->d_u.d_rcu, d_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
97
98/*
99 * Release the dentry's inode, using the filesystem
100 * d_iput() operation if defined.
101 * Called with dcache_lock and per dentry lock held, drops both.
102 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800103static void dentry_iput(struct dentry * dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
105 struct inode *inode = dentry->d_inode;
106 if (inode) {
107 dentry->d_inode = NULL;
108 list_del_init(&dentry->d_alias);
109 spin_unlock(&dentry->d_lock);
110 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700111 if (!inode->i_nlink)
112 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 if (dentry->d_op && dentry->d_op->d_iput)
114 dentry->d_op->d_iput(dentry, inode);
115 else
116 iput(inode);
117 } else {
118 spin_unlock(&dentry->d_lock);
119 spin_unlock(&dcache_lock);
120 }
121}
122
Miklos Szeredid52b9082007-05-08 00:23:46 -0700123/**
124 * d_kill - kill dentry and return parent
125 * @dentry: dentry to kill
126 *
127 * Called with dcache_lock and d_lock, releases both. The dentry must
128 * already be unhashed and removed from the LRU.
129 *
130 * If this is the root of the dentry tree, return NULL.
131 */
132static struct dentry *d_kill(struct dentry *dentry)
133{
134 struct dentry *parent;
135
136 list_del(&dentry->d_u.d_child);
137 dentry_stat.nr_dentry--; /* For d_free, below */
138 /*drops the locks, at that point nobody can reach this dentry */
139 dentry_iput(dentry);
140 parent = dentry->d_parent;
141 d_free(dentry);
142 return dentry == parent ? NULL : parent;
143}
144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145/*
146 * This is dput
147 *
148 * This is complicated by the fact that we do not want to put
149 * dentries that are no longer on any hash chain on the unused
150 * list: we'd much rather just get rid of them immediately.
151 *
152 * However, that implies that we have to traverse the dentry
153 * tree upwards to the parents which might _also_ now be
154 * scheduled for deletion (it may have been only waiting for
155 * its last child to go away).
156 *
157 * This tail recursion is done by hand as we don't want to depend
158 * on the compiler to always get this right (gcc generally doesn't).
159 * Real recursion would eat up our stack space.
160 */
161
162/*
163 * dput - release a dentry
164 * @dentry: dentry to release
165 *
166 * Release a dentry. This will drop the usage count and if appropriate
167 * call the dentry unlink method as well as removing it from the queues and
168 * releasing its resources. If the parent dentries were scheduled for release
169 * they too may now get deleted.
170 *
171 * no dcache lock, please.
172 */
173
174void dput(struct dentry *dentry)
175{
176 if (!dentry)
177 return;
178
179repeat:
180 if (atomic_read(&dentry->d_count) == 1)
181 might_sleep();
182 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
183 return;
184
185 spin_lock(&dentry->d_lock);
186 if (atomic_read(&dentry->d_count)) {
187 spin_unlock(&dentry->d_lock);
188 spin_unlock(&dcache_lock);
189 return;
190 }
191
192 /*
193 * AV: ->d_delete() is _NOT_ allowed to block now.
194 */
195 if (dentry->d_op && dentry->d_op->d_delete) {
196 if (dentry->d_op->d_delete(dentry))
197 goto unhash_it;
198 }
199 /* Unreachable? Get rid of it */
200 if (d_unhashed(dentry))
201 goto kill_it;
202 if (list_empty(&dentry->d_lru)) {
203 dentry->d_flags |= DCACHE_REFERENCED;
204 list_add(&dentry->d_lru, &dentry_unused);
205 dentry_stat.nr_unused++;
206 }
207 spin_unlock(&dentry->d_lock);
208 spin_unlock(&dcache_lock);
209 return;
210
211unhash_it:
212 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700213kill_it:
214 /* If dentry was on d_lru list
215 * delete it from there
216 */
217 if (!list_empty(&dentry->d_lru)) {
218 list_del(&dentry->d_lru);
219 dentry_stat.nr_unused--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 }
Miklos Szeredid52b9082007-05-08 00:23:46 -0700221 dentry = d_kill(dentry);
222 if (dentry)
223 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226/**
227 * d_invalidate - invalidate a dentry
228 * @dentry: dentry to invalidate
229 *
230 * Try to invalidate the dentry if it turns out to be
231 * possible. If there are other dentries that can be
232 * reached through this one we can't delete it and we
233 * return -EBUSY. On success we return 0.
234 *
235 * no dcache lock.
236 */
237
238int d_invalidate(struct dentry * dentry)
239{
240 /*
241 * If it's already been dropped, return OK.
242 */
243 spin_lock(&dcache_lock);
244 if (d_unhashed(dentry)) {
245 spin_unlock(&dcache_lock);
246 return 0;
247 }
248 /*
249 * Check whether to do a partial shrink_dcache
250 * to get rid of unused child entries.
251 */
252 if (!list_empty(&dentry->d_subdirs)) {
253 spin_unlock(&dcache_lock);
254 shrink_dcache_parent(dentry);
255 spin_lock(&dcache_lock);
256 }
257
258 /*
259 * Somebody else still using it?
260 *
261 * If it's a directory, we can't drop it
262 * for fear of somebody re-populating it
263 * with children (even though dropping it
264 * would make it unreachable from the root,
265 * we might still populate it if it was a
266 * working directory or similar).
267 */
268 spin_lock(&dentry->d_lock);
269 if (atomic_read(&dentry->d_count) > 1) {
270 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
271 spin_unlock(&dentry->d_lock);
272 spin_unlock(&dcache_lock);
273 return -EBUSY;
274 }
275 }
276
277 __d_drop(dentry);
278 spin_unlock(&dentry->d_lock);
279 spin_unlock(&dcache_lock);
280 return 0;
281}
282
283/* This should be called _only_ with dcache_lock held */
284
285static inline struct dentry * __dget_locked(struct dentry *dentry)
286{
287 atomic_inc(&dentry->d_count);
288 if (!list_empty(&dentry->d_lru)) {
289 dentry_stat.nr_unused--;
290 list_del_init(&dentry->d_lru);
291 }
292 return dentry;
293}
294
295struct dentry * dget_locked(struct dentry *dentry)
296{
297 return __dget_locked(dentry);
298}
299
300/**
301 * d_find_alias - grab a hashed alias of inode
302 * @inode: inode in question
303 * @want_discon: flag, used by d_splice_alias, to request
304 * that only a DISCONNECTED alias be returned.
305 *
306 * If inode has a hashed alias, or is a directory and has any alias,
307 * acquire the reference to alias and return it. Otherwise return NULL.
308 * Notice that if inode is a directory there can be only one alias and
309 * it can be unhashed only if it has no children, or if it is the root
310 * of a filesystem.
311 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700312 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700314 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 */
316
317static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
318{
319 struct list_head *head, *next, *tmp;
320 struct dentry *alias, *discon_alias=NULL;
321
322 head = &inode->i_dentry;
323 next = inode->i_dentry.next;
324 while (next != head) {
325 tmp = next;
326 next = tmp->next;
327 prefetch(next);
328 alias = list_entry(tmp, struct dentry, d_alias);
329 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700330 if (IS_ROOT(alias) &&
331 (alias->d_flags & DCACHE_DISCONNECTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 discon_alias = alias;
333 else if (!want_discon) {
334 __dget_locked(alias);
335 return alias;
336 }
337 }
338 }
339 if (discon_alias)
340 __dget_locked(discon_alias);
341 return discon_alias;
342}
343
344struct dentry * d_find_alias(struct inode *inode)
345{
David Howells214fda12006-03-25 03:06:36 -0800346 struct dentry *de = NULL;
347
348 if (!list_empty(&inode->i_dentry)) {
349 spin_lock(&dcache_lock);
350 de = __d_find_alias(inode, 0);
351 spin_unlock(&dcache_lock);
352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return de;
354}
355
356/*
357 * Try to kill dentries associated with this inode.
358 * WARNING: you must own a reference to inode.
359 */
360void d_prune_aliases(struct inode *inode)
361{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700362 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363restart:
364 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700365 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 spin_lock(&dentry->d_lock);
367 if (!atomic_read(&dentry->d_count)) {
368 __dget_locked(dentry);
369 __d_drop(dentry);
370 spin_unlock(&dentry->d_lock);
371 spin_unlock(&dcache_lock);
372 dput(dentry);
373 goto restart;
374 }
375 spin_unlock(&dentry->d_lock);
376 }
377 spin_unlock(&dcache_lock);
378}
379
380/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700381 * Throw away a dentry - free the inode, dput the parent. This requires that
382 * the LRU list has already been removed.
383 *
Miklos Szeredid52b9082007-05-08 00:23:46 -0700384 * If prune_parents is true, try to prune ancestors as well.
385 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 * Called with dcache_lock, drops it and then regains.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700387 * Called with dentry->d_lock held, drops it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700389static void prune_one_dentry(struct dentry * dentry, int prune_parents)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700392 dentry = d_kill(dentry);
393 if (!prune_parents) {
394 dput(dentry);
395 spin_lock(&dcache_lock);
396 return;
397 }
398
399 /*
400 * Prune ancestors. Locking is simpler than in dput(),
401 * because dcache_lock needs to be taken anyway.
402 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 spin_lock(&dcache_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700404 while (dentry) {
405 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
406 return;
407
408 if (dentry->d_op && dentry->d_op->d_delete)
409 dentry->d_op->d_delete(dentry);
410 if (!list_empty(&dentry->d_lru)) {
411 list_del(&dentry->d_lru);
412 dentry_stat.nr_unused--;
413 }
414 __d_drop(dentry);
415 dentry = d_kill(dentry);
416 spin_lock(&dcache_lock);
417 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
420/**
421 * prune_dcache - shrink the dcache
422 * @count: number of entries to try and free
NeilBrown0feae5c2006-06-22 14:47:28 -0700423 * @sb: if given, ignore dentries for other superblocks
424 * which are being unmounted.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700425 * @prune_parents: if true, try to prune ancestors as well in one go
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 *
427 * Shrink the dcache. This is done when we need
428 * more memory, or simply when we need to unmount
429 * something (at which point we need to unuse
430 * all dentries).
431 *
432 * This function may fail to free any resources if
433 * all the dentries are in use.
434 */
435
Miklos Szeredid52b9082007-05-08 00:23:46 -0700436static void prune_dcache(int count, struct super_block *sb, int prune_parents)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 spin_lock(&dcache_lock);
439 for (; count ; count--) {
440 struct dentry *dentry;
441 struct list_head *tmp;
NeilBrown0feae5c2006-06-22 14:47:28 -0700442 struct rw_semaphore *s_umount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
444 cond_resched_lock(&dcache_lock);
445
446 tmp = dentry_unused.prev;
Hua Zhongf58a1eb2006-06-25 05:49:32 -0700447 if (sb) {
NeilBrown0feae5c2006-06-22 14:47:28 -0700448 /* Try to find a dentry for this sb, but don't try
449 * too hard, if they aren't near the tail they will
450 * be moved down again soon
451 */
452 int skip = count;
453 while (skip && tmp != &dentry_unused &&
454 list_entry(tmp, struct dentry, d_lru)->d_sb != sb) {
455 skip--;
456 tmp = tmp->prev;
457 }
458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 if (tmp == &dentry_unused)
460 break;
461 list_del_init(tmp);
462 prefetch(dentry_unused.prev);
463 dentry_stat.nr_unused--;
464 dentry = list_entry(tmp, struct dentry, d_lru);
465
466 spin_lock(&dentry->d_lock);
467 /*
468 * We found an inuse dentry which was not removed from
469 * dentry_unused because of laziness during lookup. Do not free
470 * it - just keep it off the dentry_unused list.
471 */
472 if (atomic_read(&dentry->d_count)) {
473 spin_unlock(&dentry->d_lock);
474 continue;
475 }
476 /* If the dentry was recently referenced, don't free it. */
477 if (dentry->d_flags & DCACHE_REFERENCED) {
478 dentry->d_flags &= ~DCACHE_REFERENCED;
479 list_add(&dentry->d_lru, &dentry_unused);
480 dentry_stat.nr_unused++;
481 spin_unlock(&dentry->d_lock);
482 continue;
483 }
NeilBrown0feae5c2006-06-22 14:47:28 -0700484 /*
485 * If the dentry is not DCACHED_REFERENCED, it is time
486 * to remove it from the dcache, provided the super block is
487 * NULL (which means we are trying to reclaim memory)
488 * or this dentry belongs to the same super block that
489 * we want to shrink.
490 */
491 /*
492 * If this dentry is for "my" filesystem, then I can prune it
493 * without taking the s_umount lock (I already hold it).
494 */
495 if (sb && dentry->d_sb == sb) {
Miklos Szeredid52b9082007-05-08 00:23:46 -0700496 prune_one_dentry(dentry, prune_parents);
NeilBrown0feae5c2006-06-22 14:47:28 -0700497 continue;
498 }
499 /*
500 * ...otherwise we need to be sure this filesystem isn't being
501 * unmounted, otherwise we could race with
502 * generic_shutdown_super(), and end up holding a reference to
503 * an inode while the filesystem is unmounted.
504 * So we try to get s_umount, and make sure s_root isn't NULL.
505 * (Take a local copy of s_umount to avoid a use-after-free of
506 * `dentry').
507 */
508 s_umount = &dentry->d_sb->s_umount;
509 if (down_read_trylock(s_umount)) {
510 if (dentry->d_sb->s_root != NULL) {
Miklos Szeredid52b9082007-05-08 00:23:46 -0700511 prune_one_dentry(dentry, prune_parents);
NeilBrown0feae5c2006-06-22 14:47:28 -0700512 up_read(s_umount);
513 continue;
514 }
515 up_read(s_umount);
516 }
517 spin_unlock(&dentry->d_lock);
Vasily Averin6eac3f92006-10-28 10:38:44 -0700518 /*
519 * Insert dentry at the head of the list as inserting at the
520 * tail leads to a cycle.
NeilBrown0feae5c2006-06-22 14:47:28 -0700521 */
Vasily Averin6eac3f92006-10-28 10:38:44 -0700522 list_add(&dentry->d_lru, &dentry_unused);
523 dentry_stat.nr_unused++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
525 spin_unlock(&dcache_lock);
526}
527
528/*
529 * Shrink the dcache for the specified super block.
530 * This allows us to unmount a device without disturbing
531 * the dcache for the other devices.
532 *
533 * This implementation makes just two traversals of the
534 * unused list. On the first pass we move the selected
535 * dentries to the most recent end, and on the second
536 * pass we free them. The second pass must restart after
537 * each dput(), but since the target dentries are all at
538 * the end, it's really just a single traversal.
539 */
540
541/**
542 * shrink_dcache_sb - shrink dcache for a superblock
543 * @sb: superblock
544 *
545 * Shrink the dcache for the specified super block. This
546 * is used to free the dcache before unmounting a file
547 * system
548 */
549
550void shrink_dcache_sb(struct super_block * sb)
551{
552 struct list_head *tmp, *next;
553 struct dentry *dentry;
554
555 /*
556 * Pass one ... move the dentries for the specified
557 * superblock to the most recent end of the unused list.
558 */
559 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700560 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 dentry = list_entry(tmp, struct dentry, d_lru);
562 if (dentry->d_sb != sb)
563 continue;
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700564 list_move(tmp, &dentry_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566
567 /*
568 * Pass two ... free the dentries for this superblock.
569 */
570repeat:
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700571 list_for_each_safe(tmp, next, &dentry_unused) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 dentry = list_entry(tmp, struct dentry, d_lru);
573 if (dentry->d_sb != sb)
574 continue;
575 dentry_stat.nr_unused--;
576 list_del_init(tmp);
577 spin_lock(&dentry->d_lock);
578 if (atomic_read(&dentry->d_count)) {
579 spin_unlock(&dentry->d_lock);
580 continue;
581 }
Miklos Szeredid52b9082007-05-08 00:23:46 -0700582 prune_one_dentry(dentry, 1);
Kirill Korotaev2ab13462006-03-25 03:07:45 -0800583 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 goto repeat;
585 }
586 spin_unlock(&dcache_lock);
587}
588
589/*
David Howellsc636ebd2006-10-11 01:22:19 -0700590 * destroy a single subtree of dentries for unmount
591 * - see the comments on shrink_dcache_for_umount() for a description of the
592 * locking
593 */
594static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
595{
596 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700597 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700598
599 BUG_ON(!IS_ROOT(dentry));
600
601 /* detach this root from the system */
602 spin_lock(&dcache_lock);
603 if (!list_empty(&dentry->d_lru)) {
604 dentry_stat.nr_unused--;
605 list_del_init(&dentry->d_lru);
606 }
607 __d_drop(dentry);
608 spin_unlock(&dcache_lock);
609
610 for (;;) {
611 /* descend to the first leaf in the current subtree */
612 while (!list_empty(&dentry->d_subdirs)) {
613 struct dentry *loop;
614
615 /* this is a branch with children - detach all of them
616 * from the system in one go */
617 spin_lock(&dcache_lock);
618 list_for_each_entry(loop, &dentry->d_subdirs,
619 d_u.d_child) {
620 if (!list_empty(&loop->d_lru)) {
621 dentry_stat.nr_unused--;
622 list_del_init(&loop->d_lru);
623 }
624
625 __d_drop(loop);
626 cond_resched_lock(&dcache_lock);
627 }
628 spin_unlock(&dcache_lock);
629
630 /* move to the first child */
631 dentry = list_entry(dentry->d_subdirs.next,
632 struct dentry, d_u.d_child);
633 }
634
635 /* consume the dentries from this leaf up through its parents
636 * until we find one with children or run out altogether */
637 do {
638 struct inode *inode;
639
640 if (atomic_read(&dentry->d_count) != 0) {
641 printk(KERN_ERR
642 "BUG: Dentry %p{i=%lx,n=%s}"
643 " still in use (%d)"
644 " [unmount of %s %s]\n",
645 dentry,
646 dentry->d_inode ?
647 dentry->d_inode->i_ino : 0UL,
648 dentry->d_name.name,
649 atomic_read(&dentry->d_count),
650 dentry->d_sb->s_type->name,
651 dentry->d_sb->s_id);
652 BUG();
653 }
654
655 parent = dentry->d_parent;
656 if (parent == dentry)
657 parent = NULL;
658 else
659 atomic_dec(&parent->d_count);
660
661 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700662 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700663
664 inode = dentry->d_inode;
665 if (inode) {
666 dentry->d_inode = NULL;
667 list_del_init(&dentry->d_alias);
668 if (dentry->d_op && dentry->d_op->d_iput)
669 dentry->d_op->d_iput(dentry, inode);
670 else
671 iput(inode);
672 }
673
674 d_free(dentry);
675
676 /* finished when we fall off the top of the tree,
677 * otherwise we ascend to the parent and move to the
678 * next sibling if there is one */
679 if (!parent)
David Howellsf8713572006-10-28 10:38:46 -0700680 goto out;
David Howellsc636ebd2006-10-11 01:22:19 -0700681
682 dentry = parent;
683
684 } while (list_empty(&dentry->d_subdirs));
685
686 dentry = list_entry(dentry->d_subdirs.next,
687 struct dentry, d_u.d_child);
688 }
David Howellsf8713572006-10-28 10:38:46 -0700689out:
690 /* several dentries were freed, need to correct nr_dentry */
691 spin_lock(&dcache_lock);
692 dentry_stat.nr_dentry -= detached;
693 spin_unlock(&dcache_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700694}
695
696/*
697 * destroy the dentries attached to a superblock on unmounting
698 * - we don't need to use dentry->d_lock, and only need dcache_lock when
699 * removing the dentry from the system lists and hashes because:
700 * - the superblock is detached from all mountings and open files, so the
701 * dentry trees will not be rearranged by the VFS
702 * - s_umount is write-locked, so the memory pressure shrinker will ignore
703 * any dentries belonging to this superblock that it comes across
704 * - the filesystem itself is no longer permitted to rearrange the dentries
705 * in this superblock
706 */
707void shrink_dcache_for_umount(struct super_block *sb)
708{
709 struct dentry *dentry;
710
711 if (down_read_trylock(&sb->s_umount))
712 BUG();
713
714 dentry = sb->s_root;
715 sb->s_root = NULL;
716 atomic_dec(&dentry->d_count);
717 shrink_dcache_for_umount_subtree(dentry);
718
719 while (!hlist_empty(&sb->s_anon)) {
720 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
721 shrink_dcache_for_umount_subtree(dentry);
722 }
723}
724
725/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 * Search for at least 1 mount point in the dentry's subdirs.
727 * We descend to the next level whenever the d_subdirs
728 * list is non-empty and continue searching.
729 */
730
731/**
732 * have_submounts - check for mounts over a dentry
733 * @parent: dentry to check.
734 *
735 * Return true if the parent or its subdirectories contain
736 * a mount point
737 */
738
739int have_submounts(struct dentry *parent)
740{
741 struct dentry *this_parent = parent;
742 struct list_head *next;
743
744 spin_lock(&dcache_lock);
745 if (d_mountpoint(parent))
746 goto positive;
747repeat:
748 next = this_parent->d_subdirs.next;
749resume:
750 while (next != &this_parent->d_subdirs) {
751 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800752 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 next = tmp->next;
754 /* Have we found a mount point ? */
755 if (d_mountpoint(dentry))
756 goto positive;
757 if (!list_empty(&dentry->d_subdirs)) {
758 this_parent = dentry;
759 goto repeat;
760 }
761 }
762 /*
763 * All done at this level ... ascend and resume the search.
764 */
765 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800766 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 this_parent = this_parent->d_parent;
768 goto resume;
769 }
770 spin_unlock(&dcache_lock);
771 return 0; /* No mount points found in tree */
772positive:
773 spin_unlock(&dcache_lock);
774 return 1;
775}
776
777/*
778 * Search the dentry child list for the specified parent,
779 * and move any unused dentries to the end of the unused
780 * list for prune_dcache(). We descend to the next level
781 * whenever the d_subdirs list is non-empty and continue
782 * searching.
783 *
784 * It returns zero iff there are no unused children,
785 * otherwise it returns the number of children moved to
786 * the end of the unused list. This may not be the total
787 * number of unused children, because select_parent can
788 * drop the lock and return early due to latency
789 * constraints.
790 */
791static int select_parent(struct dentry * parent)
792{
793 struct dentry *this_parent = parent;
794 struct list_head *next;
795 int found = 0;
796
797 spin_lock(&dcache_lock);
798repeat:
799 next = this_parent->d_subdirs.next;
800resume:
801 while (next != &this_parent->d_subdirs) {
802 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800803 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 next = tmp->next;
805
806 if (!list_empty(&dentry->d_lru)) {
807 dentry_stat.nr_unused--;
808 list_del_init(&dentry->d_lru);
809 }
810 /*
811 * move only zero ref count dentries to the end
812 * of the unused list for prune_dcache
813 */
814 if (!atomic_read(&dentry->d_count)) {
Akinobu Mita8e130592006-06-26 00:24:37 -0700815 list_add_tail(&dentry->d_lru, &dentry_unused);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 dentry_stat.nr_unused++;
817 found++;
818 }
819
820 /*
821 * We can return to the caller if we have found some (this
822 * ensures forward progress). We'll be coming back to find
823 * the rest.
824 */
825 if (found && need_resched())
826 goto out;
827
828 /*
829 * Descend a level if the d_subdirs list is non-empty.
830 */
831 if (!list_empty(&dentry->d_subdirs)) {
832 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 goto repeat;
834 }
835 }
836 /*
837 * All done at this level ... ascend and resume the search.
838 */
839 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800840 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 goto resume;
843 }
844out:
845 spin_unlock(&dcache_lock);
846 return found;
847}
848
849/**
850 * shrink_dcache_parent - prune dcache
851 * @parent: parent of entries to prune
852 *
853 * Prune the dcache to remove unused children of the parent dentry.
854 */
855
856void shrink_dcache_parent(struct dentry * parent)
857{
858 int found;
859
860 while ((found = select_parent(parent)) != 0)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700861 prune_dcache(found, parent->d_sb, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862}
863
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864/*
865 * Scan `nr' dentries and return the number which remain.
866 *
867 * We need to avoid reentering the filesystem if the caller is performing a
868 * GFP_NOFS allocation attempt. One example deadlock is:
869 *
870 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
871 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
872 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
873 *
874 * In this case we return -1 to tell the caller that we baled.
875 */
Al Viro27496a82005-10-21 03:20:48 -0400876static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877{
878 if (nr) {
879 if (!(gfp_mask & __GFP_FS))
880 return -1;
Andrew Morton24c32d72007-05-08 00:23:49 -0700881 prune_dcache(nr, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 }
883 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
884}
885
Rusty Russell8e1f9362007-07-17 04:03:17 -0700886static struct shrinker dcache_shrinker = {
887 .shrink = shrink_dcache_memory,
888 .seeks = DEFAULT_SEEKS,
889};
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891/**
892 * d_alloc - allocate a dcache entry
893 * @parent: parent of entry to allocate
894 * @name: qstr of the name
895 *
896 * Allocates a dentry. It returns %NULL if there is insufficient memory
897 * available. On a success the dentry is returned. The name passed in is
898 * copied and the copy passed in may be reused after this call.
899 */
900
901struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
902{
903 struct dentry *dentry;
904 char *dname;
905
Mel Gormane12ba742007-10-16 01:25:52 -0700906 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (!dentry)
908 return NULL;
909
910 if (name->len > DNAME_INLINE_LEN-1) {
911 dname = kmalloc(name->len + 1, GFP_KERNEL);
912 if (!dname) {
913 kmem_cache_free(dentry_cache, dentry);
914 return NULL;
915 }
916 } else {
917 dname = dentry->d_iname;
918 }
919 dentry->d_name.name = dname;
920
921 dentry->d_name.len = name->len;
922 dentry->d_name.hash = name->hash;
923 memcpy(dname, name->name, name->len);
924 dname[name->len] = 0;
925
926 atomic_set(&dentry->d_count, 1);
927 dentry->d_flags = DCACHE_UNHASHED;
928 spin_lock_init(&dentry->d_lock);
929 dentry->d_inode = NULL;
930 dentry->d_parent = NULL;
931 dentry->d_sb = NULL;
932 dentry->d_op = NULL;
933 dentry->d_fsdata = NULL;
934 dentry->d_mounted = 0;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800935#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800937#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 INIT_HLIST_NODE(&dentry->d_hash);
939 INIT_LIST_HEAD(&dentry->d_lru);
940 INIT_LIST_HEAD(&dentry->d_subdirs);
941 INIT_LIST_HEAD(&dentry->d_alias);
942
943 if (parent) {
944 dentry->d_parent = dget(parent);
945 dentry->d_sb = parent->d_sb;
946 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800947 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 }
949
950 spin_lock(&dcache_lock);
951 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800952 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 dentry_stat.nr_dentry++;
954 spin_unlock(&dcache_lock);
955
956 return dentry;
957}
958
959struct dentry *d_alloc_name(struct dentry *parent, const char *name)
960{
961 struct qstr q;
962
963 q.name = name;
964 q.len = strlen(name);
965 q.hash = full_name_hash(q.name, q.len);
966 return d_alloc(parent, &q);
967}
968
969/**
970 * d_instantiate - fill in inode information for a dentry
971 * @entry: dentry to complete
972 * @inode: inode to attach to this dentry
973 *
974 * Fill in inode information in the entry.
975 *
976 * This turns negative dentries into productive full members
977 * of society.
978 *
979 * NOTE! This assumes that the inode count has been incremented
980 * (or otherwise set) by the caller to indicate that it is now
981 * in use by the dcache.
982 */
983
984void d_instantiate(struct dentry *entry, struct inode * inode)
985{
Eric Sesterhenn28133c72006-03-26 18:25:39 +0200986 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 spin_lock(&dcache_lock);
988 if (inode)
989 list_add(&entry->d_alias, &inode->i_dentry);
990 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -0800991 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 spin_unlock(&dcache_lock);
993 security_d_instantiate(entry, inode);
994}
995
996/**
997 * d_instantiate_unique - instantiate a non-aliased dentry
998 * @entry: dentry to instantiate
999 * @inode: inode to attach to this dentry
1000 *
1001 * Fill in inode information in the entry. On success, it returns NULL.
1002 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001003 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 *
1005 * Note that in order to avoid conflicts with rename() etc, the caller
1006 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001007 *
1008 * This also assumes that the inode count has been incremented
1009 * (or otherwise set) by the caller to indicate that it is now
1010 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 */
David Howells770bfad2006-08-22 20:06:07 -04001012static struct dentry *__d_instantiate_unique(struct dentry *entry,
1013 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014{
1015 struct dentry *alias;
1016 int len = entry->d_name.len;
1017 const char *name = entry->d_name.name;
1018 unsigned int hash = entry->d_name.hash;
1019
David Howells770bfad2006-08-22 20:06:07 -04001020 if (!inode) {
1021 entry->d_inode = NULL;
1022 return NULL;
1023 }
1024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1026 struct qstr *qstr = &alias->d_name;
1027
1028 if (qstr->hash != hash)
1029 continue;
1030 if (alias->d_parent != entry->d_parent)
1031 continue;
1032 if (qstr->len != len)
1033 continue;
1034 if (memcmp(qstr->name, name, len))
1035 continue;
1036 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 return alias;
1038 }
David Howells770bfad2006-08-22 20:06:07 -04001039
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 list_add(&entry->d_alias, &inode->i_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001042 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 return NULL;
1044}
David Howells770bfad2006-08-22 20:06:07 -04001045
1046struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1047{
1048 struct dentry *result;
1049
1050 BUG_ON(!list_empty(&entry->d_alias));
1051
1052 spin_lock(&dcache_lock);
1053 result = __d_instantiate_unique(entry, inode);
1054 spin_unlock(&dcache_lock);
1055
1056 if (!result) {
1057 security_d_instantiate(entry, inode);
1058 return NULL;
1059 }
1060
1061 BUG_ON(!d_unhashed(result));
1062 iput(inode);
1063 return result;
1064}
1065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066EXPORT_SYMBOL(d_instantiate_unique);
1067
1068/**
1069 * d_alloc_root - allocate root dentry
1070 * @root_inode: inode to allocate the root for
1071 *
1072 * Allocate a root ("/") dentry for the inode given. The inode is
1073 * instantiated and returned. %NULL is returned if there is insufficient
1074 * memory or the inode passed is %NULL.
1075 */
1076
1077struct dentry * d_alloc_root(struct inode * root_inode)
1078{
1079 struct dentry *res = NULL;
1080
1081 if (root_inode) {
1082 static const struct qstr name = { .name = "/", .len = 1 };
1083
1084 res = d_alloc(NULL, &name);
1085 if (res) {
1086 res->d_sb = root_inode->i_sb;
1087 res->d_parent = res;
1088 d_instantiate(res, root_inode);
1089 }
1090 }
1091 return res;
1092}
1093
1094static inline struct hlist_head *d_hash(struct dentry *parent,
1095 unsigned long hash)
1096{
1097 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1098 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1099 return dentry_hashtable + (hash & D_HASHMASK);
1100}
1101
1102/**
1103 * d_alloc_anon - allocate an anonymous dentry
1104 * @inode: inode to allocate the dentry for
1105 *
1106 * This is similar to d_alloc_root. It is used by filesystems when
1107 * creating a dentry for a given inode, often in the process of
1108 * mapping a filehandle to a dentry. The returned dentry may be
1109 * anonymous, or may have a full name (if the inode was already
1110 * in the cache). The file system may need to make further
1111 * efforts to connect this dentry into the dcache properly.
1112 *
1113 * When called on a directory inode, we must ensure that
1114 * the inode only ever has one dentry. If a dentry is
1115 * found, that is returned instead of allocating a new one.
1116 *
1117 * On successful return, the reference to the inode has been transferred
1118 * to the dentry. If %NULL is returned (indicating kmalloc failure),
1119 * the reference on the inode has not been released.
1120 */
1121
1122struct dentry * d_alloc_anon(struct inode *inode)
1123{
1124 static const struct qstr anonstring = { .name = "" };
1125 struct dentry *tmp;
1126 struct dentry *res;
1127
1128 if ((res = d_find_alias(inode))) {
1129 iput(inode);
1130 return res;
1131 }
1132
1133 tmp = d_alloc(NULL, &anonstring);
1134 if (!tmp)
1135 return NULL;
1136
1137 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1138
1139 spin_lock(&dcache_lock);
1140 res = __d_find_alias(inode, 0);
1141 if (!res) {
1142 /* attach a disconnected dentry */
1143 res = tmp;
1144 tmp = NULL;
1145 spin_lock(&res->d_lock);
1146 res->d_sb = inode->i_sb;
1147 res->d_parent = res;
1148 res->d_inode = inode;
1149 res->d_flags |= DCACHE_DISCONNECTED;
1150 res->d_flags &= ~DCACHE_UNHASHED;
1151 list_add(&res->d_alias, &inode->i_dentry);
1152 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
1153 spin_unlock(&res->d_lock);
1154
1155 inode = NULL; /* don't drop reference */
1156 }
1157 spin_unlock(&dcache_lock);
1158
1159 if (inode)
1160 iput(inode);
1161 if (tmp)
1162 dput(tmp);
1163 return res;
1164}
1165
1166
1167/**
1168 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1169 * @inode: the inode which may have a disconnected dentry
1170 * @dentry: a negative dentry which we want to point to the inode.
1171 *
1172 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1173 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1174 * and return it, else simply d_add the inode to the dentry and return NULL.
1175 *
1176 * This is needed in the lookup routine of any filesystem that is exportable
1177 * (via knfsd) so that we can build dcache paths to directories effectively.
1178 *
1179 * If a dentry was found and moved, then it is returned. Otherwise NULL
1180 * is returned. This matches the expected return value of ->lookup.
1181 *
1182 */
1183struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1184{
1185 struct dentry *new = NULL;
1186
NeilBrown21c0d8f2006-10-04 02:16:16 -07001187 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 spin_lock(&dcache_lock);
1189 new = __d_find_alias(inode, 1);
1190 if (new) {
1191 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Pigginc32ccd82006-03-25 03:07:09 -08001192 fsnotify_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 spin_unlock(&dcache_lock);
1194 security_d_instantiate(new, inode);
1195 d_rehash(dentry);
1196 d_move(new, dentry);
1197 iput(inode);
1198 } else {
1199 /* d_instantiate takes dcache_lock, so we do it by hand */
1200 list_add(&dentry->d_alias, &inode->i_dentry);
1201 dentry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001202 fsnotify_d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 spin_unlock(&dcache_lock);
1204 security_d_instantiate(dentry, inode);
1205 d_rehash(dentry);
1206 }
1207 } else
1208 d_add(dentry, inode);
1209 return new;
1210}
1211
1212
1213/**
1214 * d_lookup - search for a dentry
1215 * @parent: parent dentry
1216 * @name: qstr of name we wish to find
1217 *
1218 * Searches the children of the parent dentry for the name in question. If
1219 * the dentry is found its reference count is incremented and the dentry
1220 * is returned. The caller must use d_put to free the entry when it has
1221 * finished using it. %NULL is returned on failure.
1222 *
1223 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1224 * Memory barriers are used while updating and doing lockless traversal.
1225 * To avoid races with d_move while rename is happening, d_lock is used.
1226 *
1227 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1228 * and name pointer in one structure pointed by d_qstr.
1229 *
1230 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1231 * lookup is going on.
1232 *
1233 * dentry_unused list is not updated even if lookup finds the required dentry
1234 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1235 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1236 * acquisition.
1237 *
1238 * d_lookup() is protected against the concurrent renames in some unrelated
1239 * directory using the seqlockt_t rename_lock.
1240 */
1241
1242struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1243{
1244 struct dentry * dentry = NULL;
1245 unsigned long seq;
1246
1247 do {
1248 seq = read_seqbegin(&rename_lock);
1249 dentry = __d_lookup(parent, name);
1250 if (dentry)
1251 break;
1252 } while (read_seqretry(&rename_lock, seq));
1253 return dentry;
1254}
1255
1256struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1257{
1258 unsigned int len = name->len;
1259 unsigned int hash = name->hash;
1260 const unsigned char *str = name->name;
1261 struct hlist_head *head = d_hash(parent,hash);
1262 struct dentry *found = NULL;
1263 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001264 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 rcu_read_lock();
1267
Paul E. McKenney665a7582005-11-07 00:59:17 -08001268 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 struct qstr *qstr;
1270
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (dentry->d_name.hash != hash)
1272 continue;
1273 if (dentry->d_parent != parent)
1274 continue;
1275
1276 spin_lock(&dentry->d_lock);
1277
1278 /*
1279 * Recheck the dentry after taking the lock - d_move may have
1280 * changed things. Don't bother checking the hash because we're
1281 * about to compare the whole name anyway.
1282 */
1283 if (dentry->d_parent != parent)
1284 goto next;
1285
1286 /*
1287 * It is safe to compare names since d_move() cannot
1288 * change the qstr (protected by d_lock).
1289 */
1290 qstr = &dentry->d_name;
1291 if (parent->d_op && parent->d_op->d_compare) {
1292 if (parent->d_op->d_compare(parent, qstr, name))
1293 goto next;
1294 } else {
1295 if (qstr->len != len)
1296 goto next;
1297 if (memcmp(qstr->name, str, len))
1298 goto next;
1299 }
1300
1301 if (!d_unhashed(dentry)) {
1302 atomic_inc(&dentry->d_count);
1303 found = dentry;
1304 }
1305 spin_unlock(&dentry->d_lock);
1306 break;
1307next:
1308 spin_unlock(&dentry->d_lock);
1309 }
1310 rcu_read_unlock();
1311
1312 return found;
1313}
1314
1315/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001316 * d_hash_and_lookup - hash the qstr then search for a dentry
1317 * @dir: Directory to search in
1318 * @name: qstr of name we wish to find
1319 *
1320 * On hash failure or on lookup failure NULL is returned.
1321 */
1322struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1323{
1324 struct dentry *dentry = NULL;
1325
1326 /*
1327 * Check for a fs-specific hash function. Note that we must
1328 * calculate the standard hash first, as the d_op->d_hash()
1329 * routine may choose to leave the hash value unchanged.
1330 */
1331 name->hash = full_name_hash(name->name, name->len);
1332 if (dir->d_op && dir->d_op->d_hash) {
1333 if (dir->d_op->d_hash(dir, name) < 0)
1334 goto out;
1335 }
1336 dentry = d_lookup(dir, name);
1337out:
1338 return dentry;
1339}
1340
1341/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 * d_validate - verify dentry provided from insecure source
1343 * @dentry: The dentry alleged to be valid child of @dparent
1344 * @dparent: The parent dentry (known to be valid)
1345 * @hash: Hash of the dentry
1346 * @len: Length of the name
1347 *
1348 * An insecure source has sent us a dentry, here we verify it and dget() it.
1349 * This is used by ncpfs in its readdir implementation.
1350 * Zero is returned in the dentry is invalid.
1351 */
1352
1353int d_validate(struct dentry *dentry, struct dentry *dparent)
1354{
1355 struct hlist_head *base;
1356 struct hlist_node *lhp;
1357
1358 /* Check whether the ptr might be valid at all.. */
1359 if (!kmem_ptr_validate(dentry_cache, dentry))
1360 goto out;
1361
1362 if (dentry->d_parent != dparent)
1363 goto out;
1364
1365 spin_lock(&dcache_lock);
1366 base = d_hash(dparent, dentry->d_name.hash);
1367 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001368 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * as it is parsed under dcache_lock
1370 */
1371 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1372 __dget_locked(dentry);
1373 spin_unlock(&dcache_lock);
1374 return 1;
1375 }
1376 }
1377 spin_unlock(&dcache_lock);
1378out:
1379 return 0;
1380}
1381
1382/*
1383 * When a file is deleted, we have two options:
1384 * - turn this dentry into a negative dentry
1385 * - unhash this dentry and free it.
1386 *
1387 * Usually, we want to just turn this into
1388 * a negative dentry, but if anybody else is
1389 * currently using the dentry or the inode
1390 * we can't do that and we fall back on removing
1391 * it from the hash queues and waiting for
1392 * it to be deleted later when it has no users
1393 */
1394
1395/**
1396 * d_delete - delete a dentry
1397 * @dentry: The dentry to delete
1398 *
1399 * Turn the dentry into a negative dentry if possible, otherwise
1400 * remove it from the hash queues so it can be deleted later
1401 */
1402
1403void d_delete(struct dentry * dentry)
1404{
John McCutchan7a91bf72005-08-08 13:52:16 -04001405 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 /*
1407 * Are we the only user?
1408 */
1409 spin_lock(&dcache_lock);
1410 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001411 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if (atomic_read(&dentry->d_count) == 1) {
1413 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001414 fsnotify_nameremove(dentry, isdir);
Amy Griffis7a2bd3f2006-03-31 02:30:54 -08001415
1416 /* remove this and other inotify debug checks after 2.6.18 */
1417 dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 return;
1419 }
1420
1421 if (!d_unhashed(dentry))
1422 __d_drop(dentry);
1423
1424 spin_unlock(&dentry->d_lock);
1425 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001426
1427 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
1430static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1431{
1432
1433 entry->d_flags &= ~DCACHE_UNHASHED;
1434 hlist_add_head_rcu(&entry->d_hash, list);
1435}
1436
David Howells770bfad2006-08-22 20:06:07 -04001437static void _d_rehash(struct dentry * entry)
1438{
1439 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1440}
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442/**
1443 * d_rehash - add an entry back to the hash
1444 * @entry: dentry to add to the hash
1445 *
1446 * Adds a dentry to the hash according to its name.
1447 */
1448
1449void d_rehash(struct dentry * entry)
1450{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 spin_lock(&dcache_lock);
1452 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001453 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 spin_unlock(&entry->d_lock);
1455 spin_unlock(&dcache_lock);
1456}
1457
1458#define do_switch(x,y) do { \
1459 __typeof__ (x) __tmp = x; \
1460 x = y; y = __tmp; } while (0)
1461
1462/*
1463 * When switching names, the actual string doesn't strictly have to
1464 * be preserved in the target - because we're dropping the target
1465 * anyway. As such, we can just do a simple memcpy() to copy over
1466 * the new name before we switch.
1467 *
1468 * Note that we have to be a lot more careful about getting the hash
1469 * switched - we have to switch the hash value properly even if it
1470 * then no longer matches the actual (corrupted) string of the target.
1471 * The hash value has to match the hash queue that the dentry is on..
1472 */
1473static void switch_names(struct dentry *dentry, struct dentry *target)
1474{
1475 if (dname_external(target)) {
1476 if (dname_external(dentry)) {
1477 /*
1478 * Both external: swap the pointers
1479 */
1480 do_switch(target->d_name.name, dentry->d_name.name);
1481 } else {
1482 /*
1483 * dentry:internal, target:external. Steal target's
1484 * storage and make target internal.
1485 */
1486 dentry->d_name.name = target->d_name.name;
1487 target->d_name.name = target->d_iname;
1488 }
1489 } else {
1490 if (dname_external(dentry)) {
1491 /*
1492 * dentry:external, target:internal. Give dentry's
1493 * storage to target and make dentry internal
1494 */
1495 memcpy(dentry->d_iname, target->d_name.name,
1496 target->d_name.len + 1);
1497 target->d_name.name = dentry->d_name.name;
1498 dentry->d_name.name = dentry->d_iname;
1499 } else {
1500 /*
1501 * Both are internal. Just copy target to dentry
1502 */
1503 memcpy(dentry->d_iname, target->d_name.name,
1504 target->d_name.len + 1);
1505 }
1506 }
1507}
1508
1509/*
1510 * We cannibalize "target" when moving dentry on top of it,
1511 * because it's going to be thrown away anyway. We could be more
1512 * polite about it, though.
1513 *
1514 * This forceful removal will result in ugly /proc output if
1515 * somebody holds a file open that got deleted due to a rename.
1516 * We could be nicer about the deleted file, and let it show
1517 * up under the name it got deleted rather than the name that
1518 * deleted it.
1519 */
1520
Trond Myklebust9eaef272006-10-21 10:24:20 -07001521/*
1522 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 * @dentry: entry to move
1524 * @target: new dentry
1525 *
1526 * Update the dcache to reflect the move of a file name. Negative
1527 * dcache entries should not be moved in this way.
1528 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001529static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
1531 struct hlist_head *list;
1532
1533 if (!dentry->d_inode)
1534 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1535
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 write_seqlock(&rename_lock);
1537 /*
1538 * XXXX: do we really need to take target->d_lock?
1539 */
1540 if (target < dentry) {
1541 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001542 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 } else {
1544 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001545 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 }
1547
1548 /* Move the dentry to the target hash queue, if on different bucket */
1549 if (dentry->d_flags & DCACHE_UNHASHED)
1550 goto already_unhashed;
1551
1552 hlist_del_rcu(&dentry->d_hash);
1553
1554already_unhashed:
1555 list = d_hash(target->d_parent, target->d_name.hash);
1556 __d_rehash(dentry, list);
1557
1558 /* Unhash the target: dput() will then get rid of it */
1559 __d_drop(target);
1560
Eric Dumazet5160ee62006-01-08 01:03:32 -08001561 list_del(&dentry->d_u.d_child);
1562 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563
1564 /* Switch the names.. */
1565 switch_names(dentry, target);
1566 do_switch(dentry->d_name.len, target->d_name.len);
1567 do_switch(dentry->d_name.hash, target->d_name.hash);
1568
1569 /* ... and switch the parents */
1570 if (IS_ROOT(dentry)) {
1571 dentry->d_parent = target->d_parent;
1572 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001573 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 } else {
1575 do_switch(dentry->d_parent, target->d_parent);
1576
1577 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001578 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 }
1580
Eric Dumazet5160ee62006-01-08 01:03:32 -08001581 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001583 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 spin_unlock(&dentry->d_lock);
1585 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001586}
1587
1588/**
1589 * d_move - move a dentry
1590 * @dentry: entry to move
1591 * @target: new dentry
1592 *
1593 * Update the dcache to reflect the move of a file name. Negative
1594 * dcache entries should not be moved in this way.
1595 */
1596
1597void d_move(struct dentry * dentry, struct dentry * target)
1598{
1599 spin_lock(&dcache_lock);
1600 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 spin_unlock(&dcache_lock);
1602}
1603
David Howells770bfad2006-08-22 20:06:07 -04001604/*
Trond Myklebust9eaef272006-10-21 10:24:20 -07001605 * Helper that returns 1 if p1 is a parent of p2, else 0
1606 */
1607static int d_isparent(struct dentry *p1, struct dentry *p2)
1608{
1609 struct dentry *p;
1610
1611 for (p = p2; p->d_parent != p; p = p->d_parent) {
1612 if (p->d_parent == p1)
1613 return 1;
1614 }
1615 return 0;
1616}
1617
1618/*
1619 * This helper attempts to cope with remotely renamed directories
1620 *
1621 * It assumes that the caller is already holding
1622 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1623 *
1624 * Note: If ever the locking in lock_rename() changes, then please
1625 * remember to update this too...
1626 *
1627 * On return, dcache_lock will have been unlocked.
1628 */
1629static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
1630{
1631 struct mutex *m1 = NULL, *m2 = NULL;
1632 struct dentry *ret;
1633
1634 /* If alias and dentry share a parent, then no extra locks required */
1635 if (alias->d_parent == dentry->d_parent)
1636 goto out_unalias;
1637
1638 /* Check for loops */
1639 ret = ERR_PTR(-ELOOP);
1640 if (d_isparent(alias, dentry))
1641 goto out_err;
1642
1643 /* See lock_rename() */
1644 ret = ERR_PTR(-EBUSY);
1645 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1646 goto out_err;
1647 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1648 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1649 goto out_err;
1650 m2 = &alias->d_parent->d_inode->i_mutex;
1651out_unalias:
1652 d_move_locked(alias, dentry);
1653 ret = alias;
1654out_err:
1655 spin_unlock(&dcache_lock);
1656 if (m2)
1657 mutex_unlock(m2);
1658 if (m1)
1659 mutex_unlock(m1);
1660 return ret;
1661}
1662
1663/*
David Howells770bfad2006-08-22 20:06:07 -04001664 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1665 * named dentry in place of the dentry to be replaced.
1666 */
1667static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1668{
1669 struct dentry *dparent, *aparent;
1670
1671 switch_names(dentry, anon);
1672 do_switch(dentry->d_name.len, anon->d_name.len);
1673 do_switch(dentry->d_name.hash, anon->d_name.hash);
1674
1675 dparent = dentry->d_parent;
1676 aparent = anon->d_parent;
1677
1678 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1679 list_del(&dentry->d_u.d_child);
1680 if (!IS_ROOT(dentry))
1681 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1682 else
1683 INIT_LIST_HEAD(&dentry->d_u.d_child);
1684
1685 anon->d_parent = (dparent == dentry) ? anon : dparent;
1686 list_del(&anon->d_u.d_child);
1687 if (!IS_ROOT(anon))
1688 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1689 else
1690 INIT_LIST_HEAD(&anon->d_u.d_child);
1691
1692 anon->d_flags &= ~DCACHE_DISCONNECTED;
1693}
1694
1695/**
1696 * d_materialise_unique - introduce an inode into the tree
1697 * @dentry: candidate dentry
1698 * @inode: inode to bind to the dentry, to which aliases may be attached
1699 *
1700 * Introduces an dentry into the tree, substituting an extant disconnected
1701 * root directory alias in its place if there is one
1702 */
1703struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1704{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001705 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001706
1707 BUG_ON(!d_unhashed(dentry));
1708
1709 spin_lock(&dcache_lock);
1710
1711 if (!inode) {
1712 actual = dentry;
1713 dentry->d_inode = NULL;
1714 goto found_lock;
1715 }
1716
Trond Myklebust9eaef272006-10-21 10:24:20 -07001717 if (S_ISDIR(inode->i_mode)) {
1718 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001719
Trond Myklebust9eaef272006-10-21 10:24:20 -07001720 /* Does an aliased dentry already exist? */
1721 alias = __d_find_alias(inode, 0);
1722 if (alias) {
1723 actual = alias;
1724 /* Is this an anonymous mountpoint that we could splice
1725 * into our tree? */
1726 if (IS_ROOT(alias)) {
1727 spin_lock(&alias->d_lock);
1728 __d_materialise_dentry(dentry, alias);
1729 __d_drop(alias);
1730 goto found;
1731 }
1732 /* Nope, but we must(!) avoid directory aliasing */
1733 actual = __d_unalias(dentry, alias);
1734 if (IS_ERR(actual))
1735 dput(alias);
1736 goto out_nolock;
1737 }
David Howells770bfad2006-08-22 20:06:07 -04001738 }
1739
1740 /* Add a unique reference */
1741 actual = __d_instantiate_unique(dentry, inode);
1742 if (!actual)
1743 actual = dentry;
1744 else if (unlikely(!d_unhashed(actual)))
1745 goto shouldnt_be_hashed;
1746
1747found_lock:
1748 spin_lock(&actual->d_lock);
1749found:
1750 _d_rehash(actual);
1751 spin_unlock(&actual->d_lock);
1752 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001753out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001754 if (actual == dentry) {
1755 security_d_instantiate(dentry, inode);
1756 return NULL;
1757 }
1758
1759 iput(inode);
1760 return actual;
1761
David Howells770bfad2006-08-22 20:06:07 -04001762shouldnt_be_hashed:
1763 spin_unlock(&dcache_lock);
1764 BUG();
1765 goto shouldnt_be_hashed;
1766}
1767
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768/**
1769 * d_path - return the path of a dentry
1770 * @dentry: dentry to report
1771 * @vfsmnt: vfsmnt to which the dentry belongs
1772 * @root: root dentry
1773 * @rootmnt: vfsmnt to which the root dentry belongs
1774 * @buffer: buffer to return value in
1775 * @buflen: buffer length
1776 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001777 * Convert a dentry into an ASCII path name. If the entry has been deleted
1778 * the string " (deleted)" is appended. Note that this is ambiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001780 * Returns the buffer or an error code if the path was too long.
1781 *
1782 * "buflen" should be positive. Caller holds the dcache_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 */
Linus Torvalds552ce542007-02-13 12:08:18 -08001784static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt,
1785 struct dentry *root, struct vfsmount *rootmnt,
1786 char *buffer, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
Linus Torvalds552ce542007-02-13 12:08:18 -08001788 char * end = buffer+buflen;
1789 char * retval;
1790 int namelen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Linus Torvalds552ce542007-02-13 12:08:18 -08001792 *--end = '\0';
1793 buflen--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 if (!IS_ROOT(dentry) && d_unhashed(dentry)) {
Andreas Gruenbachereb3dfb02007-02-12 00:51:47 -08001795 buflen -= 10;
Linus Torvalds552ce542007-02-13 12:08:18 -08001796 end -= 10;
1797 if (buflen < 0)
1798 goto Elong;
1799 memcpy(end, " (deleted)", 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
Linus Torvalds552ce542007-02-13 12:08:18 -08001801
1802 if (buflen < 1)
1803 goto Elong;
1804 /* Get '/' right */
1805 retval = end-1;
1806 *retval = '/';
1807
1808 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 struct dentry * parent;
1810
Linus Torvalds552ce542007-02-13 12:08:18 -08001811 if (dentry == root && vfsmnt == rootmnt)
1812 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08001814 /* Global root? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 spin_lock(&vfsmount_lock);
1816 if (vfsmnt->mnt_parent == vfsmnt) {
1817 spin_unlock(&vfsmount_lock);
1818 goto global_root;
1819 }
1820 dentry = vfsmnt->mnt_mountpoint;
1821 vfsmnt = vfsmnt->mnt_parent;
1822 spin_unlock(&vfsmount_lock);
1823 continue;
1824 }
1825 parent = dentry->d_parent;
1826 prefetch(parent);
1827 namelen = dentry->d_name.len;
Andreas Gruenbachereb3dfb02007-02-12 00:51:47 -08001828 buflen -= namelen + 1;
Linus Torvalds552ce542007-02-13 12:08:18 -08001829 if (buflen < 0)
1830 goto Elong;
1831 end -= namelen;
1832 memcpy(end, dentry->d_name.name, namelen);
1833 *--end = '/';
1834 retval = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 dentry = parent;
1836 }
1837
Linus Torvalds552ce542007-02-13 12:08:18 -08001838 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
1840global_root:
1841 namelen = dentry->d_name.len;
Linus Torvalds552ce542007-02-13 12:08:18 -08001842 buflen -= namelen;
1843 if (buflen < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001845 retval -= namelen-1; /* hit the slash */
1846 memcpy(retval, dentry->d_name.name, namelen);
1847 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848Elong:
Linus Torvalds552ce542007-02-13 12:08:18 -08001849 return ERR_PTR(-ENAMETOOLONG);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850}
1851
1852/* write full pathname into buffer and return start of pathname */
Linus Torvalds552ce542007-02-13 12:08:18 -08001853char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt,
1854 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855{
1856 char *res;
1857 struct vfsmount *rootmnt;
1858 struct dentry *root;
1859
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001860 /*
1861 * We have various synthetic filesystems that never get mounted. On
1862 * these filesystems dentries are never used for lookup purposes, and
1863 * thus don't need to be hashed. They also don't need a name until a
1864 * user wants to identify the object in /proc/pid/fd/. The little hack
1865 * below allows us to generate a name for these objects on demand:
1866 */
1867 if (dentry->d_op && dentry->d_op->d_dname)
1868 return dentry->d_op->d_dname(dentry, buf, buflen);
1869
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 read_lock(&current->fs->lock);
1871 rootmnt = mntget(current->fs->rootmnt);
1872 root = dget(current->fs->root);
1873 read_unlock(&current->fs->lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08001874 spin_lock(&dcache_lock);
1875 res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen);
1876 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 dput(root);
1878 mntput(rootmnt);
1879 return res;
1880}
1881
1882/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001883 * Helper function for dentry_operations.d_dname() members
1884 */
1885char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
1886 const char *fmt, ...)
1887{
1888 va_list args;
1889 char temp[64];
1890 int sz;
1891
1892 va_start(args, fmt);
1893 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
1894 va_end(args);
1895
1896 if (sz > sizeof(temp) || sz > buflen)
1897 return ERR_PTR(-ENAMETOOLONG);
1898
1899 buffer += buflen - sz;
1900 return memcpy(buffer, temp, sz);
1901}
1902
1903/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 * NOTE! The user-level library version returns a
1905 * character pointer. The kernel system call just
1906 * returns the length of the buffer filled (which
1907 * includes the ending '\0' character), or a negative
1908 * error value. So libc would do something like
1909 *
1910 * char *getcwd(char * buf, size_t size)
1911 * {
1912 * int retval;
1913 *
1914 * retval = sys_getcwd(buf, size);
1915 * if (retval >= 0)
1916 * return buf;
1917 * errno = -retval;
1918 * return NULL;
1919 * }
1920 */
1921asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
1922{
Linus Torvalds552ce542007-02-13 12:08:18 -08001923 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 struct vfsmount *pwdmnt, *rootmnt;
1925 struct dentry *pwd, *root;
Linus Torvalds552ce542007-02-13 12:08:18 -08001926 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
1928 if (!page)
1929 return -ENOMEM;
1930
1931 read_lock(&current->fs->lock);
1932 pwdmnt = mntget(current->fs->pwdmnt);
1933 pwd = dget(current->fs->pwd);
1934 rootmnt = mntget(current->fs->rootmnt);
1935 root = dget(current->fs->root);
1936 read_unlock(&current->fs->lock);
1937
Linus Torvalds552ce542007-02-13 12:08:18 -08001938 error = -ENOENT;
1939 /* Has the current directory has been unlinked? */
1940 spin_lock(&dcache_lock);
1941 if (pwd->d_parent == pwd || !d_unhashed(pwd)) {
1942 unsigned long len;
1943 char * cwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944
Linus Torvalds552ce542007-02-13 12:08:18 -08001945 cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE);
1946 spin_unlock(&dcache_lock);
1947
1948 error = PTR_ERR(cwd);
1949 if (IS_ERR(cwd))
1950 goto out;
1951
1952 error = -ERANGE;
1953 len = PAGE_SIZE + page - cwd;
1954 if (len <= size) {
1955 error = len;
1956 if (copy_to_user(buf, cwd, len))
1957 error = -EFAULT;
1958 }
1959 } else
1960 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961
1962out:
1963 dput(pwd);
1964 mntput(pwdmnt);
1965 dput(root);
1966 mntput(rootmnt);
1967 free_page((unsigned long) page);
1968 return error;
1969}
1970
1971/*
1972 * Test whether new_dentry is a subdirectory of old_dentry.
1973 *
1974 * Trivially implemented using the dcache structure
1975 */
1976
1977/**
1978 * is_subdir - is new dentry a subdirectory of old_dentry
1979 * @new_dentry: new dentry
1980 * @old_dentry: old dentry
1981 *
1982 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
1983 * Returns 0 otherwise.
1984 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
1985 */
1986
1987int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
1988{
1989 int result;
1990 struct dentry * saved = new_dentry;
1991 unsigned long seq;
1992
1993 /* need rcu_readlock to protect against the d_parent trashing due to
1994 * d_move
1995 */
1996 rcu_read_lock();
1997 do {
1998 /* for restarting inner loop in case of seq retry */
1999 new_dentry = saved;
2000 result = 0;
2001 seq = read_seqbegin(&rename_lock);
2002 for (;;) {
2003 if (new_dentry != old_dentry) {
2004 struct dentry * parent = new_dentry->d_parent;
2005 if (parent == new_dentry)
2006 break;
2007 new_dentry = parent;
2008 continue;
2009 }
2010 result = 1;
2011 break;
2012 }
2013 } while (read_seqretry(&rename_lock, seq));
2014 rcu_read_unlock();
2015
2016 return result;
2017}
2018
2019void d_genocide(struct dentry *root)
2020{
2021 struct dentry *this_parent = root;
2022 struct list_head *next;
2023
2024 spin_lock(&dcache_lock);
2025repeat:
2026 next = this_parent->d_subdirs.next;
2027resume:
2028 while (next != &this_parent->d_subdirs) {
2029 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002030 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 next = tmp->next;
2032 if (d_unhashed(dentry)||!dentry->d_inode)
2033 continue;
2034 if (!list_empty(&dentry->d_subdirs)) {
2035 this_parent = dentry;
2036 goto repeat;
2037 }
2038 atomic_dec(&dentry->d_count);
2039 }
2040 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002041 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 atomic_dec(&this_parent->d_count);
2043 this_parent = this_parent->d_parent;
2044 goto resume;
2045 }
2046 spin_unlock(&dcache_lock);
2047}
2048
2049/**
2050 * find_inode_number - check for dentry with name
2051 * @dir: directory to check
2052 * @name: Name to find.
2053 *
2054 * Check whether a dentry already exists for the given name,
2055 * and return the inode number if it has an inode. Otherwise
2056 * 0 is returned.
2057 *
2058 * This routine is used to post-process directory listings for
2059 * filesystems using synthetic inode numbers, and is necessary
2060 * to keep getcwd() working.
2061 */
2062
2063ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2064{
2065 struct dentry * dentry;
2066 ino_t ino = 0;
2067
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002068 dentry = d_hash_and_lookup(dir, name);
2069 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 if (dentry->d_inode)
2071 ino = dentry->d_inode->i_ino;
2072 dput(dentry);
2073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 return ino;
2075}
2076
2077static __initdata unsigned long dhash_entries;
2078static int __init set_dhash_entries(char *str)
2079{
2080 if (!str)
2081 return 0;
2082 dhash_entries = simple_strtoul(str, &str, 0);
2083 return 1;
2084}
2085__setup("dhash_entries=", set_dhash_entries);
2086
2087static void __init dcache_init_early(void)
2088{
2089 int loop;
2090
2091 /* If hashes are distributed across NUMA nodes, defer
2092 * hash allocation until vmalloc space is available.
2093 */
2094 if (hashdist)
2095 return;
2096
2097 dentry_hashtable =
2098 alloc_large_system_hash("Dentry cache",
2099 sizeof(struct hlist_head),
2100 dhash_entries,
2101 13,
2102 HASH_EARLY,
2103 &d_hash_shift,
2104 &d_hash_mask,
2105 0);
2106
2107 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2108 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2109}
2110
Denis Cheng74bf17c2007-10-16 23:26:30 -07002111static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112{
2113 int loop;
2114
2115 /*
2116 * A constructor could be added for stable state like the lists,
2117 * but it is probably not worth it because of the cache nature
2118 * of the dcache.
2119 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002120 dentry_cache = KMEM_CACHE(dentry,
2121 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122
Rusty Russell8e1f9362007-07-17 04:03:17 -07002123 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125 /* Hash may have been set up in dcache_init_early */
2126 if (!hashdist)
2127 return;
2128
2129 dentry_hashtable =
2130 alloc_large_system_hash("Dentry cache",
2131 sizeof(struct hlist_head),
2132 dhash_entries,
2133 13,
2134 0,
2135 &d_hash_shift,
2136 &d_hash_mask,
2137 0);
2138
2139 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2140 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2141}
2142
2143/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002144struct kmem_cache *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
2146/* SLAB cache for file structures */
Christoph Lametere18b8902006-12-06 20:33:20 -08002147struct kmem_cache *filp_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
2149EXPORT_SYMBOL(d_genocide);
2150
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151void __init vfs_caches_init_early(void)
2152{
2153 dcache_init_early();
2154 inode_init_early();
2155}
2156
2157void __init vfs_caches_init(unsigned long mempages)
2158{
2159 unsigned long reserve;
2160
2161 /* Base hash sizes on available memory, with a reserve equal to
2162 150% of current kernel size */
2163
2164 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2165 mempages -= reserve;
2166
2167 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002168 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
2170 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002171 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Denis Cheng74bf17c2007-10-16 23:26:30 -07002173 dcache_init();
2174 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002176 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 bdev_cache_init();
2178 chrdev_init();
2179}
2180
2181EXPORT_SYMBOL(d_alloc);
2182EXPORT_SYMBOL(d_alloc_anon);
2183EXPORT_SYMBOL(d_alloc_root);
2184EXPORT_SYMBOL(d_delete);
2185EXPORT_SYMBOL(d_find_alias);
2186EXPORT_SYMBOL(d_instantiate);
2187EXPORT_SYMBOL(d_invalidate);
2188EXPORT_SYMBOL(d_lookup);
2189EXPORT_SYMBOL(d_move);
David Howells770bfad2006-08-22 20:06:07 -04002190EXPORT_SYMBOL_GPL(d_materialise_unique);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191EXPORT_SYMBOL(d_path);
2192EXPORT_SYMBOL(d_prune_aliases);
2193EXPORT_SYMBOL(d_rehash);
2194EXPORT_SYMBOL(d_splice_alias);
2195EXPORT_SYMBOL(d_validate);
2196EXPORT_SYMBOL(dget_locked);
2197EXPORT_SYMBOL(dput);
2198EXPORT_SYMBOL(find_inode_number);
2199EXPORT_SYMBOL(have_submounts);
2200EXPORT_SYMBOL(names_cachep);
2201EXPORT_SYMBOL(shrink_dcache_parent);
2202EXPORT_SYMBOL(shrink_dcache_sb);