blob: 101663d15e9f07dabc6d5976e69cc5de369f7d37 [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>
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +020020#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040022#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/slab.h>
24#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#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>
David Howells07f3f052006-09-30 20:52:18 +020035#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Eric Dumazetfa3536c2006-03-26 01:37:24 -080038int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
40
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
Al Viro74c3cbe2007-07-22 08:04:18 -040042__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44EXPORT_SYMBOL(dcache_lock);
45
Christoph Lametere18b8902006-12-06 20:33:20 -080046static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48#define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
49
50/*
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
54 *
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
57 */
58#define D_HASHBITS d_hash_shift
59#define D_HASHMASK d_hash_mask
60
Eric Dumazetfa3536c2006-03-26 01:37:24 -080061static unsigned int d_hash_mask __read_mostly;
62static unsigned int d_hash_shift __read_mostly;
63static struct hlist_head *dentry_hashtable __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
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 */
Akinobu Mitae8462ca2008-02-06 01:37:07 -080092 if (hlist_unhashed(&dentry->d_hash))
Eric Dumazetb3423412006-12-06 20:38:48 -080093 __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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800102static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200103 __releases(dentry->d_lock)
104 __releases(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105{
106 struct inode *inode = dentry->d_inode;
107 if (inode) {
108 dentry->d_inode = NULL;
109 list_del_init(&dentry->d_alias);
110 spin_unlock(&dentry->d_lock);
111 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700112 if (!inode->i_nlink)
113 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 if (dentry->d_op && dentry->d_op->d_iput)
115 dentry->d_op->d_iput(dentry, inode);
116 else
117 iput(inode);
118 } else {
119 spin_unlock(&dentry->d_lock);
120 spin_unlock(&dcache_lock);
121 }
122}
123
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700124/*
125 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
126 */
127static void dentry_lru_add(struct dentry *dentry)
128{
129 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
130 dentry->d_sb->s_nr_dentry_unused++;
131 dentry_stat.nr_unused++;
132}
133
134static void dentry_lru_add_tail(struct dentry *dentry)
135{
136 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
137 dentry->d_sb->s_nr_dentry_unused++;
138 dentry_stat.nr_unused++;
139}
140
141static void dentry_lru_del(struct dentry *dentry)
142{
143 if (!list_empty(&dentry->d_lru)) {
144 list_del(&dentry->d_lru);
145 dentry->d_sb->s_nr_dentry_unused--;
146 dentry_stat.nr_unused--;
147 }
148}
149
150static void dentry_lru_del_init(struct dentry *dentry)
151{
152 if (likely(!list_empty(&dentry->d_lru))) {
153 list_del_init(&dentry->d_lru);
154 dentry->d_sb->s_nr_dentry_unused--;
155 dentry_stat.nr_unused--;
156 }
157}
158
Miklos Szeredid52b9082007-05-08 00:23:46 -0700159/**
160 * d_kill - kill dentry and return parent
161 * @dentry: dentry to kill
162 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200163 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700164 *
165 * If this is the root of the dentry tree, return NULL.
166 */
167static struct dentry *d_kill(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200168 __releases(dentry->d_lock)
169 __releases(dcache_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700170{
171 struct dentry *parent;
172
173 list_del(&dentry->d_u.d_child);
174 dentry_stat.nr_dentry--; /* For d_free, below */
175 /*drops the locks, at that point nobody can reach this dentry */
176 dentry_iput(dentry);
177 parent = dentry->d_parent;
178 d_free(dentry);
179 return dentry == parent ? NULL : parent;
180}
181
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182/*
183 * This is dput
184 *
185 * This is complicated by the fact that we do not want to put
186 * dentries that are no longer on any hash chain on the unused
187 * list: we'd much rather just get rid of them immediately.
188 *
189 * However, that implies that we have to traverse the dentry
190 * tree upwards to the parents which might _also_ now be
191 * scheduled for deletion (it may have been only waiting for
192 * its last child to go away).
193 *
194 * This tail recursion is done by hand as we don't want to depend
195 * on the compiler to always get this right (gcc generally doesn't).
196 * Real recursion would eat up our stack space.
197 */
198
199/*
200 * dput - release a dentry
201 * @dentry: dentry to release
202 *
203 * Release a dentry. This will drop the usage count and if appropriate
204 * call the dentry unlink method as well as removing it from the queues and
205 * releasing its resources. If the parent dentries were scheduled for release
206 * they too may now get deleted.
207 *
208 * no dcache lock, please.
209 */
210
211void dput(struct dentry *dentry)
212{
213 if (!dentry)
214 return;
215
216repeat:
217 if (atomic_read(&dentry->d_count) == 1)
218 might_sleep();
219 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
220 return;
221
222 spin_lock(&dentry->d_lock);
223 if (atomic_read(&dentry->d_count)) {
224 spin_unlock(&dentry->d_lock);
225 spin_unlock(&dcache_lock);
226 return;
227 }
228
229 /*
230 * AV: ->d_delete() is _NOT_ allowed to block now.
231 */
232 if (dentry->d_op && dentry->d_op->d_delete) {
233 if (dentry->d_op->d_delete(dentry))
234 goto unhash_it;
235 }
236 /* Unreachable? Get rid of it */
237 if (d_unhashed(dentry))
238 goto kill_it;
239 if (list_empty(&dentry->d_lru)) {
240 dentry->d_flags |= DCACHE_REFERENCED;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700241 dentry_lru_add(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 }
243 spin_unlock(&dentry->d_lock);
244 spin_unlock(&dcache_lock);
245 return;
246
247unhash_it:
248 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700249kill_it:
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700250 /* if dentry was on the d_lru list delete it from there */
251 dentry_lru_del(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700252 dentry = d_kill(dentry);
253 if (dentry)
254 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
257/**
258 * d_invalidate - invalidate a dentry
259 * @dentry: dentry to invalidate
260 *
261 * Try to invalidate the dentry if it turns out to be
262 * possible. If there are other dentries that can be
263 * reached through this one we can't delete it and we
264 * return -EBUSY. On success we return 0.
265 *
266 * no dcache lock.
267 */
268
269int d_invalidate(struct dentry * dentry)
270{
271 /*
272 * If it's already been dropped, return OK.
273 */
274 spin_lock(&dcache_lock);
275 if (d_unhashed(dentry)) {
276 spin_unlock(&dcache_lock);
277 return 0;
278 }
279 /*
280 * Check whether to do a partial shrink_dcache
281 * to get rid of unused child entries.
282 */
283 if (!list_empty(&dentry->d_subdirs)) {
284 spin_unlock(&dcache_lock);
285 shrink_dcache_parent(dentry);
286 spin_lock(&dcache_lock);
287 }
288
289 /*
290 * Somebody else still using it?
291 *
292 * If it's a directory, we can't drop it
293 * for fear of somebody re-populating it
294 * with children (even though dropping it
295 * would make it unreachable from the root,
296 * we might still populate it if it was a
297 * working directory or similar).
298 */
299 spin_lock(&dentry->d_lock);
300 if (atomic_read(&dentry->d_count) > 1) {
301 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
302 spin_unlock(&dentry->d_lock);
303 spin_unlock(&dcache_lock);
304 return -EBUSY;
305 }
306 }
307
308 __d_drop(dentry);
309 spin_unlock(&dentry->d_lock);
310 spin_unlock(&dcache_lock);
311 return 0;
312}
313
314/* This should be called _only_ with dcache_lock held */
315
316static inline struct dentry * __dget_locked(struct dentry *dentry)
317{
318 atomic_inc(&dentry->d_count);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700319 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 return dentry;
321}
322
323struct dentry * dget_locked(struct dentry *dentry)
324{
325 return __dget_locked(dentry);
326}
327
328/**
329 * d_find_alias - grab a hashed alias of inode
330 * @inode: inode in question
331 * @want_discon: flag, used by d_splice_alias, to request
332 * that only a DISCONNECTED alias be returned.
333 *
334 * If inode has a hashed alias, or is a directory and has any alias,
335 * acquire the reference to alias and return it. Otherwise return NULL.
336 * Notice that if inode is a directory there can be only one alias and
337 * it can be unhashed only if it has no children, or if it is the root
338 * of a filesystem.
339 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700340 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700342 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 */
344
345static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
346{
347 struct list_head *head, *next, *tmp;
348 struct dentry *alias, *discon_alias=NULL;
349
350 head = &inode->i_dentry;
351 next = inode->i_dentry.next;
352 while (next != head) {
353 tmp = next;
354 next = tmp->next;
355 prefetch(next);
356 alias = list_entry(tmp, struct dentry, d_alias);
357 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700358 if (IS_ROOT(alias) &&
359 (alias->d_flags & DCACHE_DISCONNECTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 discon_alias = alias;
361 else if (!want_discon) {
362 __dget_locked(alias);
363 return alias;
364 }
365 }
366 }
367 if (discon_alias)
368 __dget_locked(discon_alias);
369 return discon_alias;
370}
371
372struct dentry * d_find_alias(struct inode *inode)
373{
David Howells214fda12006-03-25 03:06:36 -0800374 struct dentry *de = NULL;
375
376 if (!list_empty(&inode->i_dentry)) {
377 spin_lock(&dcache_lock);
378 de = __d_find_alias(inode, 0);
379 spin_unlock(&dcache_lock);
380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return de;
382}
383
384/*
385 * Try to kill dentries associated with this inode.
386 * WARNING: you must own a reference to inode.
387 */
388void d_prune_aliases(struct inode *inode)
389{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700390 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391restart:
392 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700393 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 spin_lock(&dentry->d_lock);
395 if (!atomic_read(&dentry->d_count)) {
396 __dget_locked(dentry);
397 __d_drop(dentry);
398 spin_unlock(&dentry->d_lock);
399 spin_unlock(&dcache_lock);
400 dput(dentry);
401 goto restart;
402 }
403 spin_unlock(&dentry->d_lock);
404 }
405 spin_unlock(&dcache_lock);
406}
407
408/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700409 * Throw away a dentry - free the inode, dput the parent. This requires that
410 * the LRU list has already been removed.
411 *
Miklos Szeredi85864e12007-10-16 23:27:09 -0700412 * Try to prune ancestors as well. This is necessary to prevent
413 * quadratic behavior of shrink_dcache_parent(), but is also expected
414 * to be beneficial in reducing dentry cache fragmentation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 */
Miklos Szeredi85864e12007-10-16 23:27:09 -0700416static void prune_one_dentry(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200417 __releases(dentry->d_lock)
418 __releases(dcache_lock)
419 __acquires(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700422 dentry = d_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700423
424 /*
425 * Prune ancestors. Locking is simpler than in dput(),
426 * because dcache_lock needs to be taken anyway.
427 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 spin_lock(&dcache_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700429 while (dentry) {
430 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
431 return;
432
433 if (dentry->d_op && dentry->d_op->d_delete)
434 dentry->d_op->d_delete(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700435 dentry_lru_del_init(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700436 __d_drop(dentry);
437 dentry = d_kill(dentry);
438 spin_lock(&dcache_lock);
439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
441
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700442/*
443 * Shrink the dentry LRU on a given superblock.
444 * @sb : superblock to shrink dentry LRU.
445 * @count: If count is NULL, we prune all dentries on superblock.
446 * @flags: If flags is non-zero, we need to do special processing based on
447 * which flags are set. This means we don't need to maintain multiple
448 * similar copies of this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700450static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700452 LIST_HEAD(referenced);
453 LIST_HEAD(tmp);
454 struct dentry *dentry;
455 int cnt = 0;
456
457 BUG_ON(!sb);
458 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700460 if (count != NULL)
461 /* called from prune_dcache() and shrink_dcache_parent() */
462 cnt = *count;
463restart:
464 if (count == NULL)
465 list_splice_init(&sb->s_dentry_lru, &tmp);
466 else {
467 while (!list_empty(&sb->s_dentry_lru)) {
468 dentry = list_entry(sb->s_dentry_lru.prev,
469 struct dentry, d_lru);
470 BUG_ON(dentry->d_sb != sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700472 spin_lock(&dentry->d_lock);
473 /*
474 * If we are honouring the DCACHE_REFERENCED flag and
475 * the dentry has this flag set, don't free it. Clear
476 * the flag and put it back on the LRU.
NeilBrown0feae5c2006-06-22 14:47:28 -0700477 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700478 if ((flags & DCACHE_REFERENCED)
479 && (dentry->d_flags & DCACHE_REFERENCED)) {
480 dentry->d_flags &= ~DCACHE_REFERENCED;
481 list_move_tail(&dentry->d_lru, &referenced);
482 spin_unlock(&dentry->d_lock);
483 } else {
484 list_move_tail(&dentry->d_lru, &tmp);
485 spin_unlock(&dentry->d_lock);
486 cnt--;
487 if (!cnt)
488 break;
NeilBrown0feae5c2006-06-22 14:47:28 -0700489 }
Kentaro Makitaf3c6ba92008-07-25 19:44:40 -0700490 cond_resched_lock(&dcache_lock);
NeilBrown0feae5c2006-06-22 14:47:28 -0700491 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700492 }
493 while (!list_empty(&tmp)) {
494 dentry = list_entry(tmp.prev, struct dentry, d_lru);
495 dentry_lru_del_init(dentry);
496 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 /*
498 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700499 * the LRU because of laziness during lookup. Do not free
500 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700502 if (atomic_read(&dentry->d_count)) {
503 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 continue;
505 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700506 prune_one_dentry(dentry);
507 /* dentry->d_lock was dropped in prune_one_dentry() */
508 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700510 if (count == NULL && !list_empty(&sb->s_dentry_lru))
511 goto restart;
512 if (count != NULL)
513 *count = cnt;
514 if (!list_empty(&referenced))
515 list_splice(&referenced, &sb->s_dentry_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 spin_unlock(&dcache_lock);
517}
518
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700519/**
520 * prune_dcache - shrink the dcache
521 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700523 * Shrink the dcache. This is done when we need more memory, or simply when we
524 * need to unmount something (at which point we need to unuse all dentries).
525 *
526 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700528static void prune_dcache(int count)
529{
530 struct super_block *sb;
531 int w_count;
532 int unused = dentry_stat.nr_unused;
533 int prune_ratio;
534 int pruned;
535
536 if (unused == 0 || count == 0)
537 return;
538 spin_lock(&dcache_lock);
539restart:
540 if (count >= unused)
541 prune_ratio = 1;
542 else
543 prune_ratio = unused / count;
544 spin_lock(&sb_lock);
545 list_for_each_entry(sb, &super_blocks, s_list) {
546 if (sb->s_nr_dentry_unused == 0)
547 continue;
548 sb->s_count++;
549 /* Now, we reclaim unused dentrins with fairness.
550 * We reclaim them same percentage from each superblock.
551 * We calculate number of dentries to scan on this sb
552 * as follows, but the implementation is arranged to avoid
553 * overflows:
554 * number of dentries to scan on this sb =
555 * count * (number of dentries on this sb /
556 * number of dentries in the machine)
557 */
558 spin_unlock(&sb_lock);
559 if (prune_ratio != 1)
560 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
561 else
562 w_count = sb->s_nr_dentry_unused;
563 pruned = w_count;
564 /*
565 * We need to be sure this filesystem isn't being unmounted,
566 * otherwise we could race with generic_shutdown_super(), and
567 * end up holding a reference to an inode while the filesystem
568 * is unmounted. So we try to get s_umount, and make sure
569 * s_root isn't NULL.
570 */
571 if (down_read_trylock(&sb->s_umount)) {
572 if ((sb->s_root != NULL) &&
573 (!list_empty(&sb->s_dentry_lru))) {
574 spin_unlock(&dcache_lock);
575 __shrink_dcache_sb(sb, &w_count,
576 DCACHE_REFERENCED);
577 pruned -= w_count;
578 spin_lock(&dcache_lock);
579 }
580 up_read(&sb->s_umount);
581 }
582 spin_lock(&sb_lock);
583 count -= pruned;
584 /*
585 * restart only when sb is no longer on the list and
586 * we have more work to do.
587 */
588 if (__put_super_and_need_restart(sb) && count > 0) {
589 spin_unlock(&sb_lock);
590 goto restart;
591 }
592 }
593 spin_unlock(&sb_lock);
594 spin_unlock(&dcache_lock);
595}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597/**
598 * shrink_dcache_sb - shrink dcache for a superblock
599 * @sb: superblock
600 *
601 * Shrink the dcache for the specified super block. This
602 * is used to free the dcache before unmounting a file
603 * system
604 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605void shrink_dcache_sb(struct super_block * sb)
606{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700607 __shrink_dcache_sb(sb, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
609
610/*
David Howellsc636ebd2006-10-11 01:22:19 -0700611 * destroy a single subtree of dentries for unmount
612 * - see the comments on shrink_dcache_for_umount() for a description of the
613 * locking
614 */
615static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
616{
617 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700618 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700619
620 BUG_ON(!IS_ROOT(dentry));
621
622 /* detach this root from the system */
623 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700624 dentry_lru_del_init(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700625 __d_drop(dentry);
626 spin_unlock(&dcache_lock);
627
628 for (;;) {
629 /* descend to the first leaf in the current subtree */
630 while (!list_empty(&dentry->d_subdirs)) {
631 struct dentry *loop;
632
633 /* this is a branch with children - detach all of them
634 * from the system in one go */
635 spin_lock(&dcache_lock);
636 list_for_each_entry(loop, &dentry->d_subdirs,
637 d_u.d_child) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700638 dentry_lru_del_init(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700639 __d_drop(loop);
640 cond_resched_lock(&dcache_lock);
641 }
642 spin_unlock(&dcache_lock);
643
644 /* move to the first child */
645 dentry = list_entry(dentry->d_subdirs.next,
646 struct dentry, d_u.d_child);
647 }
648
649 /* consume the dentries from this leaf up through its parents
650 * until we find one with children or run out altogether */
651 do {
652 struct inode *inode;
653
654 if (atomic_read(&dentry->d_count) != 0) {
655 printk(KERN_ERR
656 "BUG: Dentry %p{i=%lx,n=%s}"
657 " still in use (%d)"
658 " [unmount of %s %s]\n",
659 dentry,
660 dentry->d_inode ?
661 dentry->d_inode->i_ino : 0UL,
662 dentry->d_name.name,
663 atomic_read(&dentry->d_count),
664 dentry->d_sb->s_type->name,
665 dentry->d_sb->s_id);
666 BUG();
667 }
668
669 parent = dentry->d_parent;
670 if (parent == dentry)
671 parent = NULL;
672 else
673 atomic_dec(&parent->d_count);
674
675 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700676 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700677
678 inode = dentry->d_inode;
679 if (inode) {
680 dentry->d_inode = NULL;
681 list_del_init(&dentry->d_alias);
682 if (dentry->d_op && dentry->d_op->d_iput)
683 dentry->d_op->d_iput(dentry, inode);
684 else
685 iput(inode);
686 }
687
688 d_free(dentry);
689
690 /* finished when we fall off the top of the tree,
691 * otherwise we ascend to the parent and move to the
692 * next sibling if there is one */
693 if (!parent)
David Howellsf8713572006-10-28 10:38:46 -0700694 goto out;
David Howellsc636ebd2006-10-11 01:22:19 -0700695
696 dentry = parent;
697
698 } while (list_empty(&dentry->d_subdirs));
699
700 dentry = list_entry(dentry->d_subdirs.next,
701 struct dentry, d_u.d_child);
702 }
David Howellsf8713572006-10-28 10:38:46 -0700703out:
704 /* several dentries were freed, need to correct nr_dentry */
705 spin_lock(&dcache_lock);
706 dentry_stat.nr_dentry -= detached;
707 spin_unlock(&dcache_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700708}
709
710/*
711 * destroy the dentries attached to a superblock on unmounting
712 * - we don't need to use dentry->d_lock, and only need dcache_lock when
713 * removing the dentry from the system lists and hashes because:
714 * - the superblock is detached from all mountings and open files, so the
715 * dentry trees will not be rearranged by the VFS
716 * - s_umount is write-locked, so the memory pressure shrinker will ignore
717 * any dentries belonging to this superblock that it comes across
718 * - the filesystem itself is no longer permitted to rearrange the dentries
719 * in this superblock
720 */
721void shrink_dcache_for_umount(struct super_block *sb)
722{
723 struct dentry *dentry;
724
725 if (down_read_trylock(&sb->s_umount))
726 BUG();
727
728 dentry = sb->s_root;
729 sb->s_root = NULL;
730 atomic_dec(&dentry->d_count);
731 shrink_dcache_for_umount_subtree(dentry);
732
733 while (!hlist_empty(&sb->s_anon)) {
734 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
735 shrink_dcache_for_umount_subtree(dentry);
736 }
737}
738
739/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 * Search for at least 1 mount point in the dentry's subdirs.
741 * We descend to the next level whenever the d_subdirs
742 * list is non-empty and continue searching.
743 */
744
745/**
746 * have_submounts - check for mounts over a dentry
747 * @parent: dentry to check.
748 *
749 * Return true if the parent or its subdirectories contain
750 * a mount point
751 */
752
753int have_submounts(struct dentry *parent)
754{
755 struct dentry *this_parent = parent;
756 struct list_head *next;
757
758 spin_lock(&dcache_lock);
759 if (d_mountpoint(parent))
760 goto positive;
761repeat:
762 next = this_parent->d_subdirs.next;
763resume:
764 while (next != &this_parent->d_subdirs) {
765 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800766 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 next = tmp->next;
768 /* Have we found a mount point ? */
769 if (d_mountpoint(dentry))
770 goto positive;
771 if (!list_empty(&dentry->d_subdirs)) {
772 this_parent = dentry;
773 goto repeat;
774 }
775 }
776 /*
777 * All done at this level ... ascend and resume the search.
778 */
779 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800780 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 this_parent = this_parent->d_parent;
782 goto resume;
783 }
784 spin_unlock(&dcache_lock);
785 return 0; /* No mount points found in tree */
786positive:
787 spin_unlock(&dcache_lock);
788 return 1;
789}
790
791/*
792 * Search the dentry child list for the specified parent,
793 * and move any unused dentries to the end of the unused
794 * list for prune_dcache(). We descend to the next level
795 * whenever the d_subdirs list is non-empty and continue
796 * searching.
797 *
798 * It returns zero iff there are no unused children,
799 * otherwise it returns the number of children moved to
800 * the end of the unused list. This may not be the total
801 * number of unused children, because select_parent can
802 * drop the lock and return early due to latency
803 * constraints.
804 */
805static int select_parent(struct dentry * parent)
806{
807 struct dentry *this_parent = parent;
808 struct list_head *next;
809 int found = 0;
810
811 spin_lock(&dcache_lock);
812repeat:
813 next = this_parent->d_subdirs.next;
814resume:
815 while (next != &this_parent->d_subdirs) {
816 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800817 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 next = tmp->next;
819
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700820 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 /*
822 * move only zero ref count dentries to the end
823 * of the unused list for prune_dcache
824 */
825 if (!atomic_read(&dentry->d_count)) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700826 dentry_lru_add_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 found++;
828 }
829
830 /*
831 * We can return to the caller if we have found some (this
832 * ensures forward progress). We'll be coming back to find
833 * the rest.
834 */
835 if (found && need_resched())
836 goto out;
837
838 /*
839 * Descend a level if the d_subdirs list is non-empty.
840 */
841 if (!list_empty(&dentry->d_subdirs)) {
842 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 goto repeat;
844 }
845 }
846 /*
847 * All done at this level ... ascend and resume the search.
848 */
849 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800850 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 goto resume;
853 }
854out:
855 spin_unlock(&dcache_lock);
856 return found;
857}
858
859/**
860 * shrink_dcache_parent - prune dcache
861 * @parent: parent of entries to prune
862 *
863 * Prune the dcache to remove unused children of the parent dentry.
864 */
865
866void shrink_dcache_parent(struct dentry * parent)
867{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700868 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 int found;
870
871 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700872 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873}
874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/*
876 * Scan `nr' dentries and return the number which remain.
877 *
878 * We need to avoid reentering the filesystem if the caller is performing a
879 * GFP_NOFS allocation attempt. One example deadlock is:
880 *
881 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
882 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
883 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
884 *
885 * In this case we return -1 to tell the caller that we baled.
886 */
Al Viro27496a82005-10-21 03:20:48 -0400887static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
889 if (nr) {
890 if (!(gfp_mask & __GFP_FS))
891 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700892 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 }
894 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
895}
896
Rusty Russell8e1f9362007-07-17 04:03:17 -0700897static struct shrinker dcache_shrinker = {
898 .shrink = shrink_dcache_memory,
899 .seeks = DEFAULT_SEEKS,
900};
901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902/**
903 * d_alloc - allocate a dcache entry
904 * @parent: parent of entry to allocate
905 * @name: qstr of the name
906 *
907 * Allocates a dentry. It returns %NULL if there is insufficient memory
908 * available. On a success the dentry is returned. The name passed in is
909 * copied and the copy passed in may be reused after this call.
910 */
911
912struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
913{
914 struct dentry *dentry;
915 char *dname;
916
Mel Gormane12ba742007-10-16 01:25:52 -0700917 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 if (!dentry)
919 return NULL;
920
921 if (name->len > DNAME_INLINE_LEN-1) {
922 dname = kmalloc(name->len + 1, GFP_KERNEL);
923 if (!dname) {
924 kmem_cache_free(dentry_cache, dentry);
925 return NULL;
926 }
927 } else {
928 dname = dentry->d_iname;
929 }
930 dentry->d_name.name = dname;
931
932 dentry->d_name.len = name->len;
933 dentry->d_name.hash = name->hash;
934 memcpy(dname, name->name, name->len);
935 dname[name->len] = 0;
936
937 atomic_set(&dentry->d_count, 1);
938 dentry->d_flags = DCACHE_UNHASHED;
939 spin_lock_init(&dentry->d_lock);
940 dentry->d_inode = NULL;
941 dentry->d_parent = NULL;
942 dentry->d_sb = NULL;
943 dentry->d_op = NULL;
944 dentry->d_fsdata = NULL;
945 dentry->d_mounted = 0;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800946#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800948#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 INIT_HLIST_NODE(&dentry->d_hash);
950 INIT_LIST_HEAD(&dentry->d_lru);
951 INIT_LIST_HEAD(&dentry->d_subdirs);
952 INIT_LIST_HEAD(&dentry->d_alias);
953
954 if (parent) {
955 dentry->d_parent = dget(parent);
956 dentry->d_sb = parent->d_sb;
957 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800958 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
960
961 spin_lock(&dcache_lock);
962 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800963 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 dentry_stat.nr_dentry++;
965 spin_unlock(&dcache_lock);
966
967 return dentry;
968}
969
970struct dentry *d_alloc_name(struct dentry *parent, const char *name)
971{
972 struct qstr q;
973
974 q.name = name;
975 q.len = strlen(name);
976 q.hash = full_name_hash(q.name, q.len);
977 return d_alloc(parent, &q);
978}
979
980/**
981 * d_instantiate - fill in inode information for a dentry
982 * @entry: dentry to complete
983 * @inode: inode to attach to this dentry
984 *
985 * Fill in inode information in the entry.
986 *
987 * This turns negative dentries into productive full members
988 * of society.
989 *
990 * NOTE! This assumes that the inode count has been incremented
991 * (or otherwise set) by the caller to indicate that it is now
992 * in use by the dcache.
993 */
994
995void d_instantiate(struct dentry *entry, struct inode * inode)
996{
Eric Sesterhenn28133c72006-03-26 18:25:39 +0200997 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 spin_lock(&dcache_lock);
999 if (inode)
1000 list_add(&entry->d_alias, &inode->i_dentry);
1001 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001002 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 spin_unlock(&dcache_lock);
1004 security_d_instantiate(entry, inode);
1005}
1006
1007/**
1008 * d_instantiate_unique - instantiate a non-aliased dentry
1009 * @entry: dentry to instantiate
1010 * @inode: inode to attach to this dentry
1011 *
1012 * Fill in inode information in the entry. On success, it returns NULL.
1013 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001014 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 *
1016 * Note that in order to avoid conflicts with rename() etc, the caller
1017 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001018 *
1019 * This also assumes that the inode count has been incremented
1020 * (or otherwise set) by the caller to indicate that it is now
1021 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 */
David Howells770bfad2006-08-22 20:06:07 -04001023static struct dentry *__d_instantiate_unique(struct dentry *entry,
1024 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 struct dentry *alias;
1027 int len = entry->d_name.len;
1028 const char *name = entry->d_name.name;
1029 unsigned int hash = entry->d_name.hash;
1030
David Howells770bfad2006-08-22 20:06:07 -04001031 if (!inode) {
1032 entry->d_inode = NULL;
1033 return NULL;
1034 }
1035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1037 struct qstr *qstr = &alias->d_name;
1038
1039 if (qstr->hash != hash)
1040 continue;
1041 if (alias->d_parent != entry->d_parent)
1042 continue;
1043 if (qstr->len != len)
1044 continue;
1045 if (memcmp(qstr->name, name, len))
1046 continue;
1047 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 return alias;
1049 }
David Howells770bfad2006-08-22 20:06:07 -04001050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 list_add(&entry->d_alias, &inode->i_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 entry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001053 fsnotify_d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 return NULL;
1055}
David Howells770bfad2006-08-22 20:06:07 -04001056
1057struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1058{
1059 struct dentry *result;
1060
1061 BUG_ON(!list_empty(&entry->d_alias));
1062
1063 spin_lock(&dcache_lock);
1064 result = __d_instantiate_unique(entry, inode);
1065 spin_unlock(&dcache_lock);
1066
1067 if (!result) {
1068 security_d_instantiate(entry, inode);
1069 return NULL;
1070 }
1071
1072 BUG_ON(!d_unhashed(result));
1073 iput(inode);
1074 return result;
1075}
1076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077EXPORT_SYMBOL(d_instantiate_unique);
1078
1079/**
1080 * d_alloc_root - allocate root dentry
1081 * @root_inode: inode to allocate the root for
1082 *
1083 * Allocate a root ("/") dentry for the inode given. The inode is
1084 * instantiated and returned. %NULL is returned if there is insufficient
1085 * memory or the inode passed is %NULL.
1086 */
1087
1088struct dentry * d_alloc_root(struct inode * root_inode)
1089{
1090 struct dentry *res = NULL;
1091
1092 if (root_inode) {
1093 static const struct qstr name = { .name = "/", .len = 1 };
1094
1095 res = d_alloc(NULL, &name);
1096 if (res) {
1097 res->d_sb = root_inode->i_sb;
1098 res->d_parent = res;
1099 d_instantiate(res, root_inode);
1100 }
1101 }
1102 return res;
1103}
1104
1105static inline struct hlist_head *d_hash(struct dentry *parent,
1106 unsigned long hash)
1107{
1108 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1109 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1110 return dentry_hashtable + (hash & D_HASHMASK);
1111}
1112
1113/**
1114 * d_alloc_anon - allocate an anonymous dentry
1115 * @inode: inode to allocate the dentry for
1116 *
1117 * This is similar to d_alloc_root. It is used by filesystems when
1118 * creating a dentry for a given inode, often in the process of
1119 * mapping a filehandle to a dentry. The returned dentry may be
1120 * anonymous, or may have a full name (if the inode was already
1121 * in the cache). The file system may need to make further
1122 * efforts to connect this dentry into the dcache properly.
1123 *
1124 * When called on a directory inode, we must ensure that
1125 * the inode only ever has one dentry. If a dentry is
1126 * found, that is returned instead of allocating a new one.
1127 *
1128 * On successful return, the reference to the inode has been transferred
1129 * to the dentry. If %NULL is returned (indicating kmalloc failure),
1130 * the reference on the inode has not been released.
1131 */
1132
1133struct dentry * d_alloc_anon(struct inode *inode)
1134{
1135 static const struct qstr anonstring = { .name = "" };
1136 struct dentry *tmp;
1137 struct dentry *res;
1138
1139 if ((res = d_find_alias(inode))) {
1140 iput(inode);
1141 return res;
1142 }
1143
1144 tmp = d_alloc(NULL, &anonstring);
1145 if (!tmp)
1146 return NULL;
1147
1148 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1149
1150 spin_lock(&dcache_lock);
1151 res = __d_find_alias(inode, 0);
1152 if (!res) {
1153 /* attach a disconnected dentry */
1154 res = tmp;
1155 tmp = NULL;
1156 spin_lock(&res->d_lock);
1157 res->d_sb = inode->i_sb;
1158 res->d_parent = res;
1159 res->d_inode = inode;
1160 res->d_flags |= DCACHE_DISCONNECTED;
1161 res->d_flags &= ~DCACHE_UNHASHED;
1162 list_add(&res->d_alias, &inode->i_dentry);
1163 hlist_add_head(&res->d_hash, &inode->i_sb->s_anon);
1164 spin_unlock(&res->d_lock);
1165
1166 inode = NULL; /* don't drop reference */
1167 }
1168 spin_unlock(&dcache_lock);
1169
1170 if (inode)
1171 iput(inode);
1172 if (tmp)
1173 dput(tmp);
1174 return res;
1175}
1176
1177
1178/**
1179 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1180 * @inode: the inode which may have a disconnected dentry
1181 * @dentry: a negative dentry which we want to point to the inode.
1182 *
1183 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1184 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1185 * and return it, else simply d_add the inode to the dentry and return NULL.
1186 *
1187 * This is needed in the lookup routine of any filesystem that is exportable
1188 * (via knfsd) so that we can build dcache paths to directories effectively.
1189 *
1190 * If a dentry was found and moved, then it is returned. Otherwise NULL
1191 * is returned. This matches the expected return value of ->lookup.
1192 *
1193 */
1194struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1195{
1196 struct dentry *new = NULL;
1197
NeilBrown21c0d8f2006-10-04 02:16:16 -07001198 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 spin_lock(&dcache_lock);
1200 new = __d_find_alias(inode, 1);
1201 if (new) {
1202 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Pigginc32ccd82006-03-25 03:07:09 -08001203 fsnotify_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 spin_unlock(&dcache_lock);
1205 security_d_instantiate(new, inode);
1206 d_rehash(dentry);
1207 d_move(new, dentry);
1208 iput(inode);
1209 } else {
1210 /* d_instantiate takes dcache_lock, so we do it by hand */
1211 list_add(&dentry->d_alias, &inode->i_dentry);
1212 dentry->d_inode = inode;
Nick Pigginc32ccd82006-03-25 03:07:09 -08001213 fsnotify_d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 spin_unlock(&dcache_lock);
1215 security_d_instantiate(dentry, inode);
1216 d_rehash(dentry);
1217 }
1218 } else
1219 d_add(dentry, inode);
1220 return new;
1221}
1222
Barry Naujok94035402008-05-21 16:50:46 +10001223/**
1224 * d_add_ci - lookup or allocate new dentry with case-exact name
1225 * @inode: the inode case-insensitive lookup has found
1226 * @dentry: the negative dentry that was passed to the parent's lookup func
1227 * @name: the case-exact name to be associated with the returned dentry
1228 *
1229 * This is to avoid filling the dcache with case-insensitive names to the
1230 * same inode, only the actual correct case is stored in the dcache for
1231 * case-insensitive filesystems.
1232 *
1233 * For a case-insensitive lookup match and if the the case-exact dentry
1234 * already exists in in the dcache, use it and return it.
1235 *
1236 * If no entry exists with the exact case name, allocate new dentry with
1237 * the exact case, and return the spliced entry.
1238 */
1239struct dentry *d_add_ci(struct inode *inode, struct dentry *dentry,
1240 struct qstr *name)
1241{
1242 int error;
1243 struct dentry *found;
1244 struct dentry *new;
1245
1246 /* Does a dentry matching the name exist already? */
1247 found = d_hash_and_lookup(dentry->d_parent, name);
1248 /* If not, create it now and return */
1249 if (!found) {
1250 new = d_alloc(dentry->d_parent, name);
1251 if (!new) {
1252 error = -ENOMEM;
1253 goto err_out;
1254 }
1255 found = d_splice_alias(inode, new);
1256 if (found) {
1257 dput(new);
1258 return found;
1259 }
1260 return new;
1261 }
1262 /* Matching dentry exists, check if it is negative. */
1263 if (found->d_inode) {
1264 if (unlikely(found->d_inode != inode)) {
1265 /* This can't happen because bad inodes are unhashed. */
1266 BUG_ON(!is_bad_inode(inode));
1267 BUG_ON(!is_bad_inode(found->d_inode));
1268 }
1269 /*
1270 * Already have the inode and the dentry attached, decrement
1271 * the reference count to balance the iget() done
1272 * earlier on. We found the dentry using d_lookup() so it
1273 * cannot be disconnected and thus we do not need to worry
1274 * about any NFS/disconnectedness issues here.
1275 */
1276 iput(inode);
1277 return found;
1278 }
1279 /*
1280 * Negative dentry: instantiate it unless the inode is a directory and
1281 * has a 'disconnected' dentry (i.e. IS_ROOT and DCACHE_DISCONNECTED),
1282 * in which case d_move() that in place of the found dentry.
1283 */
1284 if (!S_ISDIR(inode->i_mode)) {
1285 /* Not a directory; everything is easy. */
1286 d_instantiate(found, inode);
1287 return found;
1288 }
1289 spin_lock(&dcache_lock);
1290 if (list_empty(&inode->i_dentry)) {
1291 /*
1292 * Directory without a 'disconnected' dentry; we need to do
1293 * d_instantiate() by hand because it takes dcache_lock which
1294 * we already hold.
1295 */
1296 list_add(&found->d_alias, &inode->i_dentry);
1297 found->d_inode = inode;
1298 spin_unlock(&dcache_lock);
1299 security_d_instantiate(found, inode);
1300 return found;
1301 }
1302 /*
1303 * Directory with a 'disconnected' dentry; get a reference to the
1304 * 'disconnected' dentry.
1305 */
1306 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1307 dget_locked(new);
1308 spin_unlock(&dcache_lock);
1309 /* Do security vodoo. */
1310 security_d_instantiate(found, inode);
1311 /* Move new in place of found. */
1312 d_move(new, found);
1313 /* Balance the iget() we did above. */
1314 iput(inode);
1315 /* Throw away found. */
1316 dput(found);
1317 /* Use new as the actual dentry. */
1318 return new;
1319
1320err_out:
1321 iput(inode);
1322 return ERR_PTR(error);
1323}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325/**
1326 * d_lookup - search for a dentry
1327 * @parent: parent dentry
1328 * @name: qstr of name we wish to find
1329 *
1330 * Searches the children of the parent dentry for the name in question. If
1331 * the dentry is found its reference count is incremented and the dentry
1332 * is returned. The caller must use d_put to free the entry when it has
1333 * finished using it. %NULL is returned on failure.
1334 *
1335 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1336 * Memory barriers are used while updating and doing lockless traversal.
1337 * To avoid races with d_move while rename is happening, d_lock is used.
1338 *
1339 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1340 * and name pointer in one structure pointed by d_qstr.
1341 *
1342 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1343 * lookup is going on.
1344 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001345 * The dentry unused LRU is not updated even if lookup finds the required dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1347 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1348 * acquisition.
1349 *
1350 * d_lookup() is protected against the concurrent renames in some unrelated
1351 * directory using the seqlockt_t rename_lock.
1352 */
1353
1354struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1355{
1356 struct dentry * dentry = NULL;
1357 unsigned long seq;
1358
1359 do {
1360 seq = read_seqbegin(&rename_lock);
1361 dentry = __d_lookup(parent, name);
1362 if (dentry)
1363 break;
1364 } while (read_seqretry(&rename_lock, seq));
1365 return dentry;
1366}
1367
1368struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1369{
1370 unsigned int len = name->len;
1371 unsigned int hash = name->hash;
1372 const unsigned char *str = name->name;
1373 struct hlist_head *head = d_hash(parent,hash);
1374 struct dentry *found = NULL;
1375 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001376 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
1378 rcu_read_lock();
1379
Paul E. McKenney665a7582005-11-07 00:59:17 -08001380 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 struct qstr *qstr;
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (dentry->d_name.hash != hash)
1384 continue;
1385 if (dentry->d_parent != parent)
1386 continue;
1387
1388 spin_lock(&dentry->d_lock);
1389
1390 /*
1391 * Recheck the dentry after taking the lock - d_move may have
1392 * changed things. Don't bother checking the hash because we're
1393 * about to compare the whole name anyway.
1394 */
1395 if (dentry->d_parent != parent)
1396 goto next;
1397
1398 /*
1399 * It is safe to compare names since d_move() cannot
1400 * change the qstr (protected by d_lock).
1401 */
1402 qstr = &dentry->d_name;
1403 if (parent->d_op && parent->d_op->d_compare) {
1404 if (parent->d_op->d_compare(parent, qstr, name))
1405 goto next;
1406 } else {
1407 if (qstr->len != len)
1408 goto next;
1409 if (memcmp(qstr->name, str, len))
1410 goto next;
1411 }
1412
1413 if (!d_unhashed(dentry)) {
1414 atomic_inc(&dentry->d_count);
1415 found = dentry;
1416 }
1417 spin_unlock(&dentry->d_lock);
1418 break;
1419next:
1420 spin_unlock(&dentry->d_lock);
1421 }
1422 rcu_read_unlock();
1423
1424 return found;
1425}
1426
1427/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001428 * d_hash_and_lookup - hash the qstr then search for a dentry
1429 * @dir: Directory to search in
1430 * @name: qstr of name we wish to find
1431 *
1432 * On hash failure or on lookup failure NULL is returned.
1433 */
1434struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1435{
1436 struct dentry *dentry = NULL;
1437
1438 /*
1439 * Check for a fs-specific hash function. Note that we must
1440 * calculate the standard hash first, as the d_op->d_hash()
1441 * routine may choose to leave the hash value unchanged.
1442 */
1443 name->hash = full_name_hash(name->name, name->len);
1444 if (dir->d_op && dir->d_op->d_hash) {
1445 if (dir->d_op->d_hash(dir, name) < 0)
1446 goto out;
1447 }
1448 dentry = d_lookup(dir, name);
1449out:
1450 return dentry;
1451}
1452
1453/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 * d_validate - verify dentry provided from insecure source
1455 * @dentry: The dentry alleged to be valid child of @dparent
1456 * @dparent: The parent dentry (known to be valid)
1457 * @hash: Hash of the dentry
1458 * @len: Length of the name
1459 *
1460 * An insecure source has sent us a dentry, here we verify it and dget() it.
1461 * This is used by ncpfs in its readdir implementation.
1462 * Zero is returned in the dentry is invalid.
1463 */
1464
1465int d_validate(struct dentry *dentry, struct dentry *dparent)
1466{
1467 struct hlist_head *base;
1468 struct hlist_node *lhp;
1469
1470 /* Check whether the ptr might be valid at all.. */
1471 if (!kmem_ptr_validate(dentry_cache, dentry))
1472 goto out;
1473
1474 if (dentry->d_parent != dparent)
1475 goto out;
1476
1477 spin_lock(&dcache_lock);
1478 base = d_hash(dparent, dentry->d_name.hash);
1479 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001480 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 * as it is parsed under dcache_lock
1482 */
1483 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1484 __dget_locked(dentry);
1485 spin_unlock(&dcache_lock);
1486 return 1;
1487 }
1488 }
1489 spin_unlock(&dcache_lock);
1490out:
1491 return 0;
1492}
1493
1494/*
1495 * When a file is deleted, we have two options:
1496 * - turn this dentry into a negative dentry
1497 * - unhash this dentry and free it.
1498 *
1499 * Usually, we want to just turn this into
1500 * a negative dentry, but if anybody else is
1501 * currently using the dentry or the inode
1502 * we can't do that and we fall back on removing
1503 * it from the hash queues and waiting for
1504 * it to be deleted later when it has no users
1505 */
1506
1507/**
1508 * d_delete - delete a dentry
1509 * @dentry: The dentry to delete
1510 *
1511 * Turn the dentry into a negative dentry if possible, otherwise
1512 * remove it from the hash queues so it can be deleted later
1513 */
1514
1515void d_delete(struct dentry * dentry)
1516{
John McCutchan7a91bf72005-08-08 13:52:16 -04001517 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 /*
1519 * Are we the only user?
1520 */
1521 spin_lock(&dcache_lock);
1522 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001523 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 if (atomic_read(&dentry->d_count) == 1) {
1525 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001526 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 return;
1528 }
1529
1530 if (!d_unhashed(dentry))
1531 __d_drop(dentry);
1532
1533 spin_unlock(&dentry->d_lock);
1534 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001535
1536 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537}
1538
1539static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1540{
1541
1542 entry->d_flags &= ~DCACHE_UNHASHED;
1543 hlist_add_head_rcu(&entry->d_hash, list);
1544}
1545
David Howells770bfad2006-08-22 20:06:07 -04001546static void _d_rehash(struct dentry * entry)
1547{
1548 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1549}
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551/**
1552 * d_rehash - add an entry back to the hash
1553 * @entry: dentry to add to the hash
1554 *
1555 * Adds a dentry to the hash according to its name.
1556 */
1557
1558void d_rehash(struct dentry * entry)
1559{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 spin_lock(&dcache_lock);
1561 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001562 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 spin_unlock(&entry->d_lock);
1564 spin_unlock(&dcache_lock);
1565}
1566
1567#define do_switch(x,y) do { \
1568 __typeof__ (x) __tmp = x; \
1569 x = y; y = __tmp; } while (0)
1570
1571/*
1572 * When switching names, the actual string doesn't strictly have to
1573 * be preserved in the target - because we're dropping the target
1574 * anyway. As such, we can just do a simple memcpy() to copy over
1575 * the new name before we switch.
1576 *
1577 * Note that we have to be a lot more careful about getting the hash
1578 * switched - we have to switch the hash value properly even if it
1579 * then no longer matches the actual (corrupted) string of the target.
1580 * The hash value has to match the hash queue that the dentry is on..
1581 */
1582static void switch_names(struct dentry *dentry, struct dentry *target)
1583{
1584 if (dname_external(target)) {
1585 if (dname_external(dentry)) {
1586 /*
1587 * Both external: swap the pointers
1588 */
1589 do_switch(target->d_name.name, dentry->d_name.name);
1590 } else {
1591 /*
1592 * dentry:internal, target:external. Steal target's
1593 * storage and make target internal.
1594 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07001595 memcpy(target->d_iname, dentry->d_name.name,
1596 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 dentry->d_name.name = target->d_name.name;
1598 target->d_name.name = target->d_iname;
1599 }
1600 } else {
1601 if (dname_external(dentry)) {
1602 /*
1603 * dentry:external, target:internal. Give dentry's
1604 * storage to target and make dentry internal
1605 */
1606 memcpy(dentry->d_iname, target->d_name.name,
1607 target->d_name.len + 1);
1608 target->d_name.name = dentry->d_name.name;
1609 dentry->d_name.name = dentry->d_iname;
1610 } else {
1611 /*
1612 * Both are internal. Just copy target to dentry
1613 */
1614 memcpy(dentry->d_iname, target->d_name.name,
1615 target->d_name.len + 1);
1616 }
1617 }
1618}
1619
1620/*
1621 * We cannibalize "target" when moving dentry on top of it,
1622 * because it's going to be thrown away anyway. We could be more
1623 * polite about it, though.
1624 *
1625 * This forceful removal will result in ugly /proc output if
1626 * somebody holds a file open that got deleted due to a rename.
1627 * We could be nicer about the deleted file, and let it show
J. Bruce Fieldsbc154b12007-10-16 23:29:42 -07001628 * up under the name it had before it was deleted rather than
1629 * under the original name of the file that was moved on top of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 */
1631
Trond Myklebust9eaef272006-10-21 10:24:20 -07001632/*
1633 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 * @dentry: entry to move
1635 * @target: new dentry
1636 *
1637 * Update the dcache to reflect the move of a file name. Negative
1638 * dcache entries should not be moved in this way.
1639 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001640static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641{
1642 struct hlist_head *list;
1643
1644 if (!dentry->d_inode)
1645 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1646
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 write_seqlock(&rename_lock);
1648 /*
1649 * XXXX: do we really need to take target->d_lock?
1650 */
1651 if (target < dentry) {
1652 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001653 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 } else {
1655 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001656 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 }
1658
1659 /* Move the dentry to the target hash queue, if on different bucket */
Denis Chengf77e3492007-10-16 23:30:11 -07001660 if (d_unhashed(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 goto already_unhashed;
1662
1663 hlist_del_rcu(&dentry->d_hash);
1664
1665already_unhashed:
1666 list = d_hash(target->d_parent, target->d_name.hash);
1667 __d_rehash(dentry, list);
1668
1669 /* Unhash the target: dput() will then get rid of it */
1670 __d_drop(target);
1671
Eric Dumazet5160ee62006-01-08 01:03:32 -08001672 list_del(&dentry->d_u.d_child);
1673 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
1675 /* Switch the names.. */
1676 switch_names(dentry, target);
1677 do_switch(dentry->d_name.len, target->d_name.len);
1678 do_switch(dentry->d_name.hash, target->d_name.hash);
1679
1680 /* ... and switch the parents */
1681 if (IS_ROOT(dentry)) {
1682 dentry->d_parent = target->d_parent;
1683 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001684 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 } else {
1686 do_switch(dentry->d_parent, target->d_parent);
1687
1688 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001689 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 }
1691
Eric Dumazet5160ee62006-01-08 01:03:32 -08001692 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001694 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 spin_unlock(&dentry->d_lock);
1696 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001697}
1698
1699/**
1700 * d_move - move a dentry
1701 * @dentry: entry to move
1702 * @target: new dentry
1703 *
1704 * Update the dcache to reflect the move of a file name. Negative
1705 * dcache entries should not be moved in this way.
1706 */
1707
1708void d_move(struct dentry * dentry, struct dentry * target)
1709{
1710 spin_lock(&dcache_lock);
1711 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 spin_unlock(&dcache_lock);
1713}
1714
David Howells770bfad2006-08-22 20:06:07 -04001715/*
Trond Myklebust9eaef272006-10-21 10:24:20 -07001716 * Helper that returns 1 if p1 is a parent of p2, else 0
1717 */
1718static int d_isparent(struct dentry *p1, struct dentry *p2)
1719{
1720 struct dentry *p;
1721
1722 for (p = p2; p->d_parent != p; p = p->d_parent) {
1723 if (p->d_parent == p1)
1724 return 1;
1725 }
1726 return 0;
1727}
1728
1729/*
1730 * This helper attempts to cope with remotely renamed directories
1731 *
1732 * It assumes that the caller is already holding
1733 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1734 *
1735 * Note: If ever the locking in lock_rename() changes, then please
1736 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07001737 */
1738static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001739 __releases(dcache_lock)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001740{
1741 struct mutex *m1 = NULL, *m2 = NULL;
1742 struct dentry *ret;
1743
1744 /* If alias and dentry share a parent, then no extra locks required */
1745 if (alias->d_parent == dentry->d_parent)
1746 goto out_unalias;
1747
1748 /* Check for loops */
1749 ret = ERR_PTR(-ELOOP);
1750 if (d_isparent(alias, dentry))
1751 goto out_err;
1752
1753 /* See lock_rename() */
1754 ret = ERR_PTR(-EBUSY);
1755 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1756 goto out_err;
1757 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1758 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1759 goto out_err;
1760 m2 = &alias->d_parent->d_inode->i_mutex;
1761out_unalias:
1762 d_move_locked(alias, dentry);
1763 ret = alias;
1764out_err:
1765 spin_unlock(&dcache_lock);
1766 if (m2)
1767 mutex_unlock(m2);
1768 if (m1)
1769 mutex_unlock(m1);
1770 return ret;
1771}
1772
1773/*
David Howells770bfad2006-08-22 20:06:07 -04001774 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1775 * named dentry in place of the dentry to be replaced.
1776 */
1777static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1778{
1779 struct dentry *dparent, *aparent;
1780
1781 switch_names(dentry, anon);
1782 do_switch(dentry->d_name.len, anon->d_name.len);
1783 do_switch(dentry->d_name.hash, anon->d_name.hash);
1784
1785 dparent = dentry->d_parent;
1786 aparent = anon->d_parent;
1787
1788 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1789 list_del(&dentry->d_u.d_child);
1790 if (!IS_ROOT(dentry))
1791 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1792 else
1793 INIT_LIST_HEAD(&dentry->d_u.d_child);
1794
1795 anon->d_parent = (dparent == dentry) ? anon : dparent;
1796 list_del(&anon->d_u.d_child);
1797 if (!IS_ROOT(anon))
1798 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1799 else
1800 INIT_LIST_HEAD(&anon->d_u.d_child);
1801
1802 anon->d_flags &= ~DCACHE_DISCONNECTED;
1803}
1804
1805/**
1806 * d_materialise_unique - introduce an inode into the tree
1807 * @dentry: candidate dentry
1808 * @inode: inode to bind to the dentry, to which aliases may be attached
1809 *
1810 * Introduces an dentry into the tree, substituting an extant disconnected
1811 * root directory alias in its place if there is one
1812 */
1813struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1814{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001815 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001816
1817 BUG_ON(!d_unhashed(dentry));
1818
1819 spin_lock(&dcache_lock);
1820
1821 if (!inode) {
1822 actual = dentry;
1823 dentry->d_inode = NULL;
1824 goto found_lock;
1825 }
1826
Trond Myklebust9eaef272006-10-21 10:24:20 -07001827 if (S_ISDIR(inode->i_mode)) {
1828 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001829
Trond Myklebust9eaef272006-10-21 10:24:20 -07001830 /* Does an aliased dentry already exist? */
1831 alias = __d_find_alias(inode, 0);
1832 if (alias) {
1833 actual = alias;
1834 /* Is this an anonymous mountpoint that we could splice
1835 * into our tree? */
1836 if (IS_ROOT(alias)) {
1837 spin_lock(&alias->d_lock);
1838 __d_materialise_dentry(dentry, alias);
1839 __d_drop(alias);
1840 goto found;
1841 }
1842 /* Nope, but we must(!) avoid directory aliasing */
1843 actual = __d_unalias(dentry, alias);
1844 if (IS_ERR(actual))
1845 dput(alias);
1846 goto out_nolock;
1847 }
David Howells770bfad2006-08-22 20:06:07 -04001848 }
1849
1850 /* Add a unique reference */
1851 actual = __d_instantiate_unique(dentry, inode);
1852 if (!actual)
1853 actual = dentry;
1854 else if (unlikely(!d_unhashed(actual)))
1855 goto shouldnt_be_hashed;
1856
1857found_lock:
1858 spin_lock(&actual->d_lock);
1859found:
1860 _d_rehash(actual);
1861 spin_unlock(&actual->d_lock);
1862 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001863out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001864 if (actual == dentry) {
1865 security_d_instantiate(dentry, inode);
1866 return NULL;
1867 }
1868
1869 iput(inode);
1870 return actual;
1871
David Howells770bfad2006-08-22 20:06:07 -04001872shouldnt_be_hashed:
1873 spin_unlock(&dcache_lock);
1874 BUG();
David Howells770bfad2006-08-22 20:06:07 -04001875}
1876
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001877static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01001878{
1879 *buflen -= namelen;
1880 if (*buflen < 0)
1881 return -ENAMETOOLONG;
1882 *buffer -= namelen;
1883 memcpy(*buffer, str, namelen);
1884 return 0;
1885}
1886
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001887static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1888{
1889 return prepend(buffer, buflen, name->name, name->len);
1890}
1891
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001893 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001894 * @path: the dentry/vfsmount to report
1895 * @root: root vfsmnt/dentry (may be modified by this function)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 * @buffer: buffer to return value in
1897 * @buflen: buffer length
1898 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001899 * Convert a dentry into an ASCII path name. If the entry has been deleted
1900 * the string " (deleted)" is appended. Note that this is ambiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001902 * Returns the buffer or an error code if the path was too long.
1903 *
1904 * "buflen" should be positive. Caller holds the dcache_lock.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001905 *
1906 * If path is not reachable from the supplied root, then the value of
1907 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001909char *__d_path(const struct path *path, struct path *root,
1910 char *buffer, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911{
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001912 struct dentry *dentry = path->dentry;
1913 struct vfsmount *vfsmnt = path->mnt;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001914 char *end = buffer + buflen;
1915 char *retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001917 spin_lock(&vfsmount_lock);
Ram Pai6092d042008-03-27 13:06:20 +01001918 prepend(&end, &buflen, "\0", 1);
1919 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
1920 (prepend(&end, &buflen, " (deleted)", 10) != 0))
Linus Torvalds552ce542007-02-13 12:08:18 -08001921 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001922
1923 if (buflen < 1)
1924 goto Elong;
1925 /* Get '/' right */
1926 retval = end-1;
1927 *retval = '/';
1928
1929 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 struct dentry * parent;
1931
Jan Blunck329c97f2008-02-14 19:38:31 -08001932 if (dentry == root->dentry && vfsmnt == root->mnt)
Linus Torvalds552ce542007-02-13 12:08:18 -08001933 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08001935 /* Global root? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 if (vfsmnt->mnt_parent == vfsmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 goto global_root;
1938 }
1939 dentry = vfsmnt->mnt_mountpoint;
1940 vfsmnt = vfsmnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 continue;
1942 }
1943 parent = dentry->d_parent;
1944 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001945 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01001946 (prepend(&end, &buflen, "/", 1) != 0))
Linus Torvalds552ce542007-02-13 12:08:18 -08001947 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001948 retval = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 dentry = parent;
1950 }
1951
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001952out:
1953 spin_unlock(&vfsmount_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08001954 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956global_root:
Ram Pai6092d042008-03-27 13:06:20 +01001957 retval += 1; /* hit the slash */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001958 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 goto Elong;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001960 root->mnt = vfsmnt;
1961 root->dentry = dentry;
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001962 goto out;
1963
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964Elong:
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001965 retval = ERR_PTR(-ENAMETOOLONG);
1966 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967}
1968
Jan Bluncka03a8a702008-02-14 19:38:32 -08001969/**
1970 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08001971 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08001972 * @buf: buffer to return value in
1973 * @buflen: buffer length
1974 *
1975 * Convert a dentry into an ASCII path name. If the entry has been deleted
1976 * the string " (deleted)" is appended. Note that this is ambiguous.
1977 *
1978 * Returns the buffer or an error code if the path was too long.
1979 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001980 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08001981 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07001982char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983{
1984 char *res;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001985 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001986 struct path tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001988 /*
1989 * We have various synthetic filesystems that never get mounted. On
1990 * these filesystems dentries are never used for lookup purposes, and
1991 * thus don't need to be hashed. They also don't need a name until a
1992 * user wants to identify the object in /proc/pid/fd/. The little hack
1993 * below allows us to generate a name for these objects on demand:
1994 */
Jan Blunckcf28b482008-02-14 19:38:44 -08001995 if (path->dentry->d_op && path->dentry->d_op->d_dname)
1996 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001997
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001999 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002000 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 read_unlock(&current->fs->lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002002 spin_lock(&dcache_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002003 tmp = root;
2004 res = __d_path(path, &tmp, buf, buflen);
Linus Torvalds552ce542007-02-13 12:08:18 -08002005 spin_unlock(&dcache_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002006 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 return res;
2008}
2009
2010/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002011 * Helper function for dentry_operations.d_dname() members
2012 */
2013char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2014 const char *fmt, ...)
2015{
2016 va_list args;
2017 char temp[64];
2018 int sz;
2019
2020 va_start(args, fmt);
2021 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2022 va_end(args);
2023
2024 if (sz > sizeof(temp) || sz > buflen)
2025 return ERR_PTR(-ENAMETOOLONG);
2026
2027 buffer += buflen - sz;
2028 return memcpy(buffer, temp, sz);
2029}
2030
2031/*
Ram Pai6092d042008-03-27 13:06:20 +01002032 * Write full pathname from the root of the filesystem into the buffer.
2033 */
2034char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2035{
2036 char *end = buf + buflen;
2037 char *retval;
2038
2039 spin_lock(&dcache_lock);
2040 prepend(&end, &buflen, "\0", 1);
2041 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
2042 (prepend(&end, &buflen, "//deleted", 9) != 0))
2043 goto Elong;
2044 if (buflen < 1)
2045 goto Elong;
2046 /* Get '/' right */
2047 retval = end-1;
2048 *retval = '/';
2049
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002050 while (!IS_ROOT(dentry)) {
2051 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01002052
Ram Pai6092d042008-03-27 13:06:20 +01002053 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002054 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01002055 (prepend(&end, &buflen, "/", 1) != 0))
2056 goto Elong;
2057
2058 retval = end;
2059 dentry = parent;
2060 }
2061 spin_unlock(&dcache_lock);
2062 return retval;
2063Elong:
2064 spin_unlock(&dcache_lock);
2065 return ERR_PTR(-ENAMETOOLONG);
2066}
2067
2068/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 * NOTE! The user-level library version returns a
2070 * character pointer. The kernel system call just
2071 * returns the length of the buffer filled (which
2072 * includes the ending '\0' character), or a negative
2073 * error value. So libc would do something like
2074 *
2075 * char *getcwd(char * buf, size_t size)
2076 * {
2077 * int retval;
2078 *
2079 * retval = sys_getcwd(buf, size);
2080 * if (retval >= 0)
2081 * return buf;
2082 * errno = -retval;
2083 * return NULL;
2084 * }
2085 */
2086asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
2087{
Linus Torvalds552ce542007-02-13 12:08:18 -08002088 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002089 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002090 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
2092 if (!page)
2093 return -ENOMEM;
2094
2095 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002096 pwd = current->fs->pwd;
Ram Pai6092d042008-03-27 13:06:20 +01002097 path_get(&pwd);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002098 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002099 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 read_unlock(&current->fs->lock);
2101
Linus Torvalds552ce542007-02-13 12:08:18 -08002102 error = -ENOENT;
2103 /* Has the current directory has been unlinked? */
2104 spin_lock(&dcache_lock);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002105 if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002106 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002107 struct path tmp = root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002108 char * cwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002110 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
Linus Torvalds552ce542007-02-13 12:08:18 -08002111 spin_unlock(&dcache_lock);
2112
2113 error = PTR_ERR(cwd);
2114 if (IS_ERR(cwd))
2115 goto out;
2116
2117 error = -ERANGE;
2118 len = PAGE_SIZE + page - cwd;
2119 if (len <= size) {
2120 error = len;
2121 if (copy_to_user(buf, cwd, len))
2122 error = -EFAULT;
2123 }
2124 } else
2125 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
2127out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002128 path_put(&pwd);
2129 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 free_page((unsigned long) page);
2131 return error;
2132}
2133
2134/*
2135 * Test whether new_dentry is a subdirectory of old_dentry.
2136 *
2137 * Trivially implemented using the dcache structure
2138 */
2139
2140/**
2141 * is_subdir - is new dentry a subdirectory of old_dentry
2142 * @new_dentry: new dentry
2143 * @old_dentry: old dentry
2144 *
2145 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2146 * Returns 0 otherwise.
2147 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2148 */
2149
2150int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry)
2151{
2152 int result;
2153 struct dentry * saved = new_dentry;
2154 unsigned long seq;
2155
2156 /* need rcu_readlock to protect against the d_parent trashing due to
2157 * d_move
2158 */
2159 rcu_read_lock();
2160 do {
2161 /* for restarting inner loop in case of seq retry */
2162 new_dentry = saved;
2163 result = 0;
2164 seq = read_seqbegin(&rename_lock);
2165 for (;;) {
2166 if (new_dentry != old_dentry) {
2167 struct dentry * parent = new_dentry->d_parent;
2168 if (parent == new_dentry)
2169 break;
2170 new_dentry = parent;
2171 continue;
2172 }
2173 result = 1;
2174 break;
2175 }
2176 } while (read_seqretry(&rename_lock, seq));
2177 rcu_read_unlock();
2178
2179 return result;
2180}
2181
2182void d_genocide(struct dentry *root)
2183{
2184 struct dentry *this_parent = root;
2185 struct list_head *next;
2186
2187 spin_lock(&dcache_lock);
2188repeat:
2189 next = this_parent->d_subdirs.next;
2190resume:
2191 while (next != &this_parent->d_subdirs) {
2192 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002193 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 next = tmp->next;
2195 if (d_unhashed(dentry)||!dentry->d_inode)
2196 continue;
2197 if (!list_empty(&dentry->d_subdirs)) {
2198 this_parent = dentry;
2199 goto repeat;
2200 }
2201 atomic_dec(&dentry->d_count);
2202 }
2203 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002204 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 atomic_dec(&this_parent->d_count);
2206 this_parent = this_parent->d_parent;
2207 goto resume;
2208 }
2209 spin_unlock(&dcache_lock);
2210}
2211
2212/**
2213 * find_inode_number - check for dentry with name
2214 * @dir: directory to check
2215 * @name: Name to find.
2216 *
2217 * Check whether a dentry already exists for the given name,
2218 * and return the inode number if it has an inode. Otherwise
2219 * 0 is returned.
2220 *
2221 * This routine is used to post-process directory listings for
2222 * filesystems using synthetic inode numbers, and is necessary
2223 * to keep getcwd() working.
2224 */
2225
2226ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2227{
2228 struct dentry * dentry;
2229 ino_t ino = 0;
2230
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002231 dentry = d_hash_and_lookup(dir, name);
2232 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 if (dentry->d_inode)
2234 ino = dentry->d_inode->i_ino;
2235 dput(dentry);
2236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 return ino;
2238}
2239
2240static __initdata unsigned long dhash_entries;
2241static int __init set_dhash_entries(char *str)
2242{
2243 if (!str)
2244 return 0;
2245 dhash_entries = simple_strtoul(str, &str, 0);
2246 return 1;
2247}
2248__setup("dhash_entries=", set_dhash_entries);
2249
2250static void __init dcache_init_early(void)
2251{
2252 int loop;
2253
2254 /* If hashes are distributed across NUMA nodes, defer
2255 * hash allocation until vmalloc space is available.
2256 */
2257 if (hashdist)
2258 return;
2259
2260 dentry_hashtable =
2261 alloc_large_system_hash("Dentry cache",
2262 sizeof(struct hlist_head),
2263 dhash_entries,
2264 13,
2265 HASH_EARLY,
2266 &d_hash_shift,
2267 &d_hash_mask,
2268 0);
2269
2270 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2271 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2272}
2273
Denis Cheng74bf17c2007-10-16 23:26:30 -07002274static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
2276 int loop;
2277
2278 /*
2279 * A constructor could be added for stable state like the lists,
2280 * but it is probably not worth it because of the cache nature
2281 * of the dcache.
2282 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002283 dentry_cache = KMEM_CACHE(dentry,
2284 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Rusty Russell8e1f9362007-07-17 04:03:17 -07002286 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
2288 /* Hash may have been set up in dcache_init_early */
2289 if (!hashdist)
2290 return;
2291
2292 dentry_hashtable =
2293 alloc_large_system_hash("Dentry cache",
2294 sizeof(struct hlist_head),
2295 dhash_entries,
2296 13,
2297 0,
2298 &d_hash_shift,
2299 &d_hash_mask,
2300 0);
2301
2302 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2303 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2304}
2305
2306/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002307struct kmem_cache *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308
2309/* SLAB cache for file structures */
Christoph Lametere18b8902006-12-06 20:33:20 -08002310struct kmem_cache *filp_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
2312EXPORT_SYMBOL(d_genocide);
2313
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314void __init vfs_caches_init_early(void)
2315{
2316 dcache_init_early();
2317 inode_init_early();
2318}
2319
2320void __init vfs_caches_init(unsigned long mempages)
2321{
2322 unsigned long reserve;
2323
2324 /* Base hash sizes on available memory, with a reserve equal to
2325 150% of current kernel size */
2326
2327 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2328 mempages -= reserve;
2329
2330 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002331 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
2333 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002334 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335
Denis Cheng74bf17c2007-10-16 23:26:30 -07002336 dcache_init();
2337 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002339 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 bdev_cache_init();
2341 chrdev_init();
2342}
2343
2344EXPORT_SYMBOL(d_alloc);
2345EXPORT_SYMBOL(d_alloc_anon);
2346EXPORT_SYMBOL(d_alloc_root);
2347EXPORT_SYMBOL(d_delete);
2348EXPORT_SYMBOL(d_find_alias);
2349EXPORT_SYMBOL(d_instantiate);
2350EXPORT_SYMBOL(d_invalidate);
2351EXPORT_SYMBOL(d_lookup);
2352EXPORT_SYMBOL(d_move);
David Howells770bfad2006-08-22 20:06:07 -04002353EXPORT_SYMBOL_GPL(d_materialise_unique);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354EXPORT_SYMBOL(d_path);
2355EXPORT_SYMBOL(d_prune_aliases);
2356EXPORT_SYMBOL(d_rehash);
2357EXPORT_SYMBOL(d_splice_alias);
Barry Naujok94035402008-05-21 16:50:46 +10002358EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359EXPORT_SYMBOL(d_validate);
2360EXPORT_SYMBOL(dget_locked);
2361EXPORT_SYMBOL(dput);
2362EXPORT_SYMBOL(find_inode_number);
2363EXPORT_SYMBOL(have_submounts);
2364EXPORT_SYMBOL(names_cachep);
2365EXPORT_SYMBOL(shrink_dcache_parent);
2366EXPORT_SYMBOL(shrink_dcache_sb);