blob: a1d86c7f3e6644c8914edeae67a0ba02b3df6ada [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 Howells07f3f05c2006-09-30 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{
Arjan van de Venfd217f42008-10-21 06:47:33 -070072 WARN_ON(!list_empty(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 if (dname_external(dentry))
74 kfree(dentry->d_name.name);
75 kmem_cache_free(dentry_cache, dentry);
76}
77
Eric Dumazetb3423412006-12-06 20:38:48 -080078static void d_callback(struct rcu_head *head)
79{
80 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
81 __d_free(dentry);
82}
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
85 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
86 * inside dcache_lock.
87 */
88static void d_free(struct dentry *dentry)
89{
90 if (dentry->d_op && dentry->d_op->d_release)
91 dentry->d_op->d_release(dentry);
Eric Dumazetb3423412006-12-06 20:38:48 -080092 /* if dentry was never inserted into hash, immediate free is OK */
Akinobu Mitae8462ca2008-02-06 01:37:07 -080093 if (hlist_unhashed(&dentry->d_hash))
Eric Dumazetb3423412006-12-06 20:38:48 -080094 __d_free(dentry);
95 else
96 call_rcu(&dentry->d_u.d_rcu, d_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097}
98
99/*
100 * Release the dentry's inode, using the filesystem
101 * d_iput() operation if defined.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800103static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200104 __releases(dentry->d_lock)
105 __releases(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
107 struct inode *inode = dentry->d_inode;
108 if (inode) {
109 dentry->d_inode = NULL;
110 list_del_init(&dentry->d_alias);
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700113 if (!inode->i_nlink)
114 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 if (dentry->d_op && dentry->d_op->d_iput)
116 dentry->d_op->d_iput(dentry, inode);
117 else
118 iput(inode);
119 } else {
120 spin_unlock(&dentry->d_lock);
121 spin_unlock(&dcache_lock);
122 }
123}
124
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700125/*
126 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
127 */
128static void dentry_lru_add(struct dentry *dentry)
129{
130 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
131 dentry->d_sb->s_nr_dentry_unused++;
132 dentry_stat.nr_unused++;
133}
134
135static void dentry_lru_add_tail(struct dentry *dentry)
136{
137 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
138 dentry->d_sb->s_nr_dentry_unused++;
139 dentry_stat.nr_unused++;
140}
141
142static void dentry_lru_del(struct dentry *dentry)
143{
144 if (!list_empty(&dentry->d_lru)) {
145 list_del(&dentry->d_lru);
146 dentry->d_sb->s_nr_dentry_unused--;
147 dentry_stat.nr_unused--;
148 }
149}
150
151static void dentry_lru_del_init(struct dentry *dentry)
152{
153 if (likely(!list_empty(&dentry->d_lru))) {
154 list_del_init(&dentry->d_lru);
155 dentry->d_sb->s_nr_dentry_unused--;
156 dentry_stat.nr_unused--;
157 }
158}
159
Miklos Szeredid52b9082007-05-08 00:23:46 -0700160/**
161 * d_kill - kill dentry and return parent
162 * @dentry: dentry to kill
163 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200164 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700165 *
166 * If this is the root of the dentry tree, return NULL.
167 */
168static struct dentry *d_kill(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200169 __releases(dentry->d_lock)
170 __releases(dcache_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700171{
172 struct dentry *parent;
173
174 list_del(&dentry->d_u.d_child);
175 dentry_stat.nr_dentry--; /* For d_free, below */
176 /*drops the locks, at that point nobody can reach this dentry */
177 dentry_iput(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900178 if (IS_ROOT(dentry))
179 parent = NULL;
180 else
181 parent = dentry->d_parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700182 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900183 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700184}
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/*
187 * This is dput
188 *
189 * This is complicated by the fact that we do not want to put
190 * dentries that are no longer on any hash chain on the unused
191 * list: we'd much rather just get rid of them immediately.
192 *
193 * However, that implies that we have to traverse the dentry
194 * tree upwards to the parents which might _also_ now be
195 * scheduled for deletion (it may have been only waiting for
196 * its last child to go away).
197 *
198 * This tail recursion is done by hand as we don't want to depend
199 * on the compiler to always get this right (gcc generally doesn't).
200 * Real recursion would eat up our stack space.
201 */
202
203/*
204 * dput - release a dentry
205 * @dentry: dentry to release
206 *
207 * Release a dentry. This will drop the usage count and if appropriate
208 * call the dentry unlink method as well as removing it from the queues and
209 * releasing its resources. If the parent dentries were scheduled for release
210 * they too may now get deleted.
211 *
212 * no dcache lock, please.
213 */
214
215void dput(struct dentry *dentry)
216{
217 if (!dentry)
218 return;
219
220repeat:
221 if (atomic_read(&dentry->d_count) == 1)
222 might_sleep();
223 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
224 return;
225
226 spin_lock(&dentry->d_lock);
227 if (atomic_read(&dentry->d_count)) {
228 spin_unlock(&dentry->d_lock);
229 spin_unlock(&dcache_lock);
230 return;
231 }
232
233 /*
234 * AV: ->d_delete() is _NOT_ allowed to block now.
235 */
236 if (dentry->d_op && dentry->d_op->d_delete) {
237 if (dentry->d_op->d_delete(dentry))
238 goto unhash_it;
239 }
240 /* Unreachable? Get rid of it */
241 if (d_unhashed(dentry))
242 goto kill_it;
243 if (list_empty(&dentry->d_lru)) {
244 dentry->d_flags |= DCACHE_REFERENCED;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700245 dentry_lru_add(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247 spin_unlock(&dentry->d_lock);
248 spin_unlock(&dcache_lock);
249 return;
250
251unhash_it:
252 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700253kill_it:
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700254 /* if dentry was on the d_lru list delete it from there */
255 dentry_lru_del(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700256 dentry = d_kill(dentry);
257 if (dentry)
258 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
261/**
262 * d_invalidate - invalidate a dentry
263 * @dentry: dentry to invalidate
264 *
265 * Try to invalidate the dentry if it turns out to be
266 * possible. If there are other dentries that can be
267 * reached through this one we can't delete it and we
268 * return -EBUSY. On success we return 0.
269 *
270 * no dcache lock.
271 */
272
273int d_invalidate(struct dentry * dentry)
274{
275 /*
276 * If it's already been dropped, return OK.
277 */
278 spin_lock(&dcache_lock);
279 if (d_unhashed(dentry)) {
280 spin_unlock(&dcache_lock);
281 return 0;
282 }
283 /*
284 * Check whether to do a partial shrink_dcache
285 * to get rid of unused child entries.
286 */
287 if (!list_empty(&dentry->d_subdirs)) {
288 spin_unlock(&dcache_lock);
289 shrink_dcache_parent(dentry);
290 spin_lock(&dcache_lock);
291 }
292
293 /*
294 * Somebody else still using it?
295 *
296 * If it's a directory, we can't drop it
297 * for fear of somebody re-populating it
298 * with children (even though dropping it
299 * would make it unreachable from the root,
300 * we might still populate it if it was a
301 * working directory or similar).
302 */
303 spin_lock(&dentry->d_lock);
304 if (atomic_read(&dentry->d_count) > 1) {
305 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
306 spin_unlock(&dentry->d_lock);
307 spin_unlock(&dcache_lock);
308 return -EBUSY;
309 }
310 }
311
312 __d_drop(dentry);
313 spin_unlock(&dentry->d_lock);
314 spin_unlock(&dcache_lock);
315 return 0;
316}
317
318/* This should be called _only_ with dcache_lock held */
319
320static inline struct dentry * __dget_locked(struct dentry *dentry)
321{
322 atomic_inc(&dentry->d_count);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700323 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 return dentry;
325}
326
327struct dentry * dget_locked(struct dentry *dentry)
328{
329 return __dget_locked(dentry);
330}
331
332/**
333 * d_find_alias - grab a hashed alias of inode
334 * @inode: inode in question
335 * @want_discon: flag, used by d_splice_alias, to request
336 * that only a DISCONNECTED alias be returned.
337 *
338 * If inode has a hashed alias, or is a directory and has any alias,
339 * acquire the reference to alias and return it. Otherwise return NULL.
340 * Notice that if inode is a directory there can be only one alias and
341 * it can be unhashed only if it has no children, or if it is the root
342 * of a filesystem.
343 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700344 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700346 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 */
348
349static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
350{
351 struct list_head *head, *next, *tmp;
352 struct dentry *alias, *discon_alias=NULL;
353
354 head = &inode->i_dentry;
355 next = inode->i_dentry.next;
356 while (next != head) {
357 tmp = next;
358 next = tmp->next;
359 prefetch(next);
360 alias = list_entry(tmp, struct dentry, d_alias);
361 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700362 if (IS_ROOT(alias) &&
363 (alias->d_flags & DCACHE_DISCONNECTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 discon_alias = alias;
365 else if (!want_discon) {
366 __dget_locked(alias);
367 return alias;
368 }
369 }
370 }
371 if (discon_alias)
372 __dget_locked(discon_alias);
373 return discon_alias;
374}
375
376struct dentry * d_find_alias(struct inode *inode)
377{
David Howells214fda12006-03-25 03:06:36 -0800378 struct dentry *de = NULL;
379
380 if (!list_empty(&inode->i_dentry)) {
381 spin_lock(&dcache_lock);
382 de = __d_find_alias(inode, 0);
383 spin_unlock(&dcache_lock);
384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 return de;
386}
387
388/*
389 * Try to kill dentries associated with this inode.
390 * WARNING: you must own a reference to inode.
391 */
392void d_prune_aliases(struct inode *inode)
393{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700394 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395restart:
396 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700397 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 spin_lock(&dentry->d_lock);
399 if (!atomic_read(&dentry->d_count)) {
400 __dget_locked(dentry);
401 __d_drop(dentry);
402 spin_unlock(&dentry->d_lock);
403 spin_unlock(&dcache_lock);
404 dput(dentry);
405 goto restart;
406 }
407 spin_unlock(&dentry->d_lock);
408 }
409 spin_unlock(&dcache_lock);
410}
411
412/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700413 * Throw away a dentry - free the inode, dput the parent. This requires that
414 * the LRU list has already been removed.
415 *
Miklos Szeredi85864e12007-10-16 23:27:09 -0700416 * Try to prune ancestors as well. This is necessary to prevent
417 * quadratic behavior of shrink_dcache_parent(), but is also expected
418 * to be beneficial in reducing dentry cache fragmentation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 */
Miklos Szeredi85864e12007-10-16 23:27:09 -0700420static void prune_one_dentry(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200421 __releases(dentry->d_lock)
422 __releases(dcache_lock)
423 __acquires(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700426 dentry = d_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700427
428 /*
429 * Prune ancestors. Locking is simpler than in dput(),
430 * because dcache_lock needs to be taken anyway.
431 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 spin_lock(&dcache_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700433 while (dentry) {
434 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
435 return;
436
437 if (dentry->d_op && dentry->d_op->d_delete)
438 dentry->d_op->d_delete(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700439 dentry_lru_del_init(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700440 __d_drop(dentry);
441 dentry = d_kill(dentry);
442 spin_lock(&dcache_lock);
443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444}
445
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700446/*
447 * Shrink the dentry LRU on a given superblock.
448 * @sb : superblock to shrink dentry LRU.
449 * @count: If count is NULL, we prune all dentries on superblock.
450 * @flags: If flags is non-zero, we need to do special processing based on
451 * which flags are set. This means we don't need to maintain multiple
452 * similar copies of this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700454static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700456 LIST_HEAD(referenced);
457 LIST_HEAD(tmp);
458 struct dentry *dentry;
459 int cnt = 0;
460
461 BUG_ON(!sb);
462 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700464 if (count != NULL)
465 /* called from prune_dcache() and shrink_dcache_parent() */
466 cnt = *count;
467restart:
468 if (count == NULL)
469 list_splice_init(&sb->s_dentry_lru, &tmp);
470 else {
471 while (!list_empty(&sb->s_dentry_lru)) {
472 dentry = list_entry(sb->s_dentry_lru.prev,
473 struct dentry, d_lru);
474 BUG_ON(dentry->d_sb != sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700476 spin_lock(&dentry->d_lock);
477 /*
478 * If we are honouring the DCACHE_REFERENCED flag and
479 * the dentry has this flag set, don't free it. Clear
480 * the flag and put it back on the LRU.
NeilBrown0feae5c2006-06-22 14:47:28 -0700481 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700482 if ((flags & DCACHE_REFERENCED)
483 && (dentry->d_flags & DCACHE_REFERENCED)) {
484 dentry->d_flags &= ~DCACHE_REFERENCED;
485 list_move_tail(&dentry->d_lru, &referenced);
486 spin_unlock(&dentry->d_lock);
487 } else {
488 list_move_tail(&dentry->d_lru, &tmp);
489 spin_unlock(&dentry->d_lock);
490 cnt--;
491 if (!cnt)
492 break;
NeilBrown0feae5c2006-06-22 14:47:28 -0700493 }
Kentaro Makitaf3c6ba92008-07-25 19:44:40 -0700494 cond_resched_lock(&dcache_lock);
NeilBrown0feae5c2006-06-22 14:47:28 -0700495 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700496 }
497 while (!list_empty(&tmp)) {
498 dentry = list_entry(tmp.prev, struct dentry, d_lru);
499 dentry_lru_del_init(dentry);
500 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 /*
502 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700503 * the LRU because of laziness during lookup. Do not free
504 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700506 if (atomic_read(&dentry->d_count)) {
507 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 continue;
509 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700510 prune_one_dentry(dentry);
511 /* dentry->d_lock was dropped in prune_one_dentry() */
512 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700514 if (count == NULL && !list_empty(&sb->s_dentry_lru))
515 goto restart;
516 if (count != NULL)
517 *count = cnt;
518 if (!list_empty(&referenced))
519 list_splice(&referenced, &sb->s_dentry_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 spin_unlock(&dcache_lock);
521}
522
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700523/**
524 * prune_dcache - shrink the dcache
525 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700527 * Shrink the dcache. This is done when we need more memory, or simply when we
528 * need to unmount something (at which point we need to unuse all dentries).
529 *
530 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700532static void prune_dcache(int count)
533{
534 struct super_block *sb;
535 int w_count;
536 int unused = dentry_stat.nr_unused;
537 int prune_ratio;
538 int pruned;
539
540 if (unused == 0 || count == 0)
541 return;
542 spin_lock(&dcache_lock);
543restart:
544 if (count >= unused)
545 prune_ratio = 1;
546 else
547 prune_ratio = unused / count;
548 spin_lock(&sb_lock);
549 list_for_each_entry(sb, &super_blocks, s_list) {
550 if (sb->s_nr_dentry_unused == 0)
551 continue;
552 sb->s_count++;
553 /* Now, we reclaim unused dentrins with fairness.
554 * We reclaim them same percentage from each superblock.
555 * We calculate number of dentries to scan on this sb
556 * as follows, but the implementation is arranged to avoid
557 * overflows:
558 * number of dentries to scan on this sb =
559 * count * (number of dentries on this sb /
560 * number of dentries in the machine)
561 */
562 spin_unlock(&sb_lock);
563 if (prune_ratio != 1)
564 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
565 else
566 w_count = sb->s_nr_dentry_unused;
567 pruned = w_count;
568 /*
569 * We need to be sure this filesystem isn't being unmounted,
570 * otherwise we could race with generic_shutdown_super(), and
571 * end up holding a reference to an inode while the filesystem
572 * is unmounted. So we try to get s_umount, and make sure
573 * s_root isn't NULL.
574 */
575 if (down_read_trylock(&sb->s_umount)) {
576 if ((sb->s_root != NULL) &&
577 (!list_empty(&sb->s_dentry_lru))) {
578 spin_unlock(&dcache_lock);
579 __shrink_dcache_sb(sb, &w_count,
580 DCACHE_REFERENCED);
581 pruned -= w_count;
582 spin_lock(&dcache_lock);
583 }
584 up_read(&sb->s_umount);
585 }
586 spin_lock(&sb_lock);
587 count -= pruned;
588 /*
589 * restart only when sb is no longer on the list and
590 * we have more work to do.
591 */
592 if (__put_super_and_need_restart(sb) && count > 0) {
593 spin_unlock(&sb_lock);
594 goto restart;
595 }
596 }
597 spin_unlock(&sb_lock);
598 spin_unlock(&dcache_lock);
599}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601/**
602 * shrink_dcache_sb - shrink dcache for a superblock
603 * @sb: superblock
604 *
605 * Shrink the dcache for the specified super block. This
606 * is used to free the dcache before unmounting a file
607 * system
608 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609void shrink_dcache_sb(struct super_block * sb)
610{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700611 __shrink_dcache_sb(sb, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
614/*
David Howellsc636ebd2006-10-11 01:22:19 -0700615 * destroy a single subtree of dentries for unmount
616 * - see the comments on shrink_dcache_for_umount() for a description of the
617 * locking
618 */
619static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
620{
621 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700622 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700623
624 BUG_ON(!IS_ROOT(dentry));
625
626 /* detach this root from the system */
627 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700628 dentry_lru_del_init(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700629 __d_drop(dentry);
630 spin_unlock(&dcache_lock);
631
632 for (;;) {
633 /* descend to the first leaf in the current subtree */
634 while (!list_empty(&dentry->d_subdirs)) {
635 struct dentry *loop;
636
637 /* this is a branch with children - detach all of them
638 * from the system in one go */
639 spin_lock(&dcache_lock);
640 list_for_each_entry(loop, &dentry->d_subdirs,
641 d_u.d_child) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700642 dentry_lru_del_init(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700643 __d_drop(loop);
644 cond_resched_lock(&dcache_lock);
645 }
646 spin_unlock(&dcache_lock);
647
648 /* move to the first child */
649 dentry = list_entry(dentry->d_subdirs.next,
650 struct dentry, d_u.d_child);
651 }
652
653 /* consume the dentries from this leaf up through its parents
654 * until we find one with children or run out altogether */
655 do {
656 struct inode *inode;
657
658 if (atomic_read(&dentry->d_count) != 0) {
659 printk(KERN_ERR
660 "BUG: Dentry %p{i=%lx,n=%s}"
661 " still in use (%d)"
662 " [unmount of %s %s]\n",
663 dentry,
664 dentry->d_inode ?
665 dentry->d_inode->i_ino : 0UL,
666 dentry->d_name.name,
667 atomic_read(&dentry->d_count),
668 dentry->d_sb->s_type->name,
669 dentry->d_sb->s_id);
670 BUG();
671 }
672
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900673 if (IS_ROOT(dentry))
David Howellsc636ebd2006-10-11 01:22:19 -0700674 parent = NULL;
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900675 else {
676 parent = dentry->d_parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700677 atomic_dec(&parent->d_count);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900678 }
David Howellsc636ebd2006-10-11 01:22:19 -0700679
680 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700681 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700682
683 inode = dentry->d_inode;
684 if (inode) {
685 dentry->d_inode = NULL;
686 list_del_init(&dentry->d_alias);
687 if (dentry->d_op && dentry->d_op->d_iput)
688 dentry->d_op->d_iput(dentry, inode);
689 else
690 iput(inode);
691 }
692
693 d_free(dentry);
694
695 /* finished when we fall off the top of the tree,
696 * otherwise we ascend to the parent and move to the
697 * next sibling if there is one */
698 if (!parent)
David Howellsf8713572006-10-28 10:38:46 -0700699 goto out;
David Howellsc636ebd2006-10-11 01:22:19 -0700700
701 dentry = parent;
702
703 } while (list_empty(&dentry->d_subdirs));
704
705 dentry = list_entry(dentry->d_subdirs.next,
706 struct dentry, d_u.d_child);
707 }
David Howellsf8713572006-10-28 10:38:46 -0700708out:
709 /* several dentries were freed, need to correct nr_dentry */
710 spin_lock(&dcache_lock);
711 dentry_stat.nr_dentry -= detached;
712 spin_unlock(&dcache_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700713}
714
715/*
716 * destroy the dentries attached to a superblock on unmounting
717 * - we don't need to use dentry->d_lock, and only need dcache_lock when
718 * removing the dentry from the system lists and hashes because:
719 * - the superblock is detached from all mountings and open files, so the
720 * dentry trees will not be rearranged by the VFS
721 * - s_umount is write-locked, so the memory pressure shrinker will ignore
722 * any dentries belonging to this superblock that it comes across
723 * - the filesystem itself is no longer permitted to rearrange the dentries
724 * in this superblock
725 */
726void shrink_dcache_for_umount(struct super_block *sb)
727{
728 struct dentry *dentry;
729
730 if (down_read_trylock(&sb->s_umount))
731 BUG();
732
733 dentry = sb->s_root;
734 sb->s_root = NULL;
735 atomic_dec(&dentry->d_count);
736 shrink_dcache_for_umount_subtree(dentry);
737
738 while (!hlist_empty(&sb->s_anon)) {
739 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
740 shrink_dcache_for_umount_subtree(dentry);
741 }
742}
743
744/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 * Search for at least 1 mount point in the dentry's subdirs.
746 * We descend to the next level whenever the d_subdirs
747 * list is non-empty and continue searching.
748 */
749
750/**
751 * have_submounts - check for mounts over a dentry
752 * @parent: dentry to check.
753 *
754 * Return true if the parent or its subdirectories contain
755 * a mount point
756 */
757
758int have_submounts(struct dentry *parent)
759{
760 struct dentry *this_parent = parent;
761 struct list_head *next;
762
763 spin_lock(&dcache_lock);
764 if (d_mountpoint(parent))
765 goto positive;
766repeat:
767 next = this_parent->d_subdirs.next;
768resume:
769 while (next != &this_parent->d_subdirs) {
770 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800771 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 next = tmp->next;
773 /* Have we found a mount point ? */
774 if (d_mountpoint(dentry))
775 goto positive;
776 if (!list_empty(&dentry->d_subdirs)) {
777 this_parent = dentry;
778 goto repeat;
779 }
780 }
781 /*
782 * All done at this level ... ascend and resume the search.
783 */
784 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800785 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 this_parent = this_parent->d_parent;
787 goto resume;
788 }
789 spin_unlock(&dcache_lock);
790 return 0; /* No mount points found in tree */
791positive:
792 spin_unlock(&dcache_lock);
793 return 1;
794}
795
796/*
797 * Search the dentry child list for the specified parent,
798 * and move any unused dentries to the end of the unused
799 * list for prune_dcache(). We descend to the next level
800 * whenever the d_subdirs list is non-empty and continue
801 * searching.
802 *
803 * It returns zero iff there are no unused children,
804 * otherwise it returns the number of children moved to
805 * the end of the unused list. This may not be the total
806 * number of unused children, because select_parent can
807 * drop the lock and return early due to latency
808 * constraints.
809 */
810static int select_parent(struct dentry * parent)
811{
812 struct dentry *this_parent = parent;
813 struct list_head *next;
814 int found = 0;
815
816 spin_lock(&dcache_lock);
817repeat:
818 next = this_parent->d_subdirs.next;
819resume:
820 while (next != &this_parent->d_subdirs) {
821 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800822 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 next = tmp->next;
824
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700825 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 /*
827 * move only zero ref count dentries to the end
828 * of the unused list for prune_dcache
829 */
830 if (!atomic_read(&dentry->d_count)) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700831 dentry_lru_add_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 found++;
833 }
834
835 /*
836 * We can return to the caller if we have found some (this
837 * ensures forward progress). We'll be coming back to find
838 * the rest.
839 */
840 if (found && need_resched())
841 goto out;
842
843 /*
844 * Descend a level if the d_subdirs list is non-empty.
845 */
846 if (!list_empty(&dentry->d_subdirs)) {
847 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 goto repeat;
849 }
850 }
851 /*
852 * All done at this level ... ascend and resume the search.
853 */
854 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800855 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 goto resume;
858 }
859out:
860 spin_unlock(&dcache_lock);
861 return found;
862}
863
864/**
865 * shrink_dcache_parent - prune dcache
866 * @parent: parent of entries to prune
867 *
868 * Prune the dcache to remove unused children of the parent dentry.
869 */
870
871void shrink_dcache_parent(struct dentry * parent)
872{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700873 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 int found;
875
876 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700877 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880/*
881 * Scan `nr' dentries and return the number which remain.
882 *
883 * We need to avoid reentering the filesystem if the caller is performing a
884 * GFP_NOFS allocation attempt. One example deadlock is:
885 *
886 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
887 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
888 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
889 *
890 * In this case we return -1 to tell the caller that we baled.
891 */
Al Viro27496a82005-10-21 03:20:48 -0400892static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
894 if (nr) {
895 if (!(gfp_mask & __GFP_FS))
896 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700897 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
899 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
900}
901
Rusty Russell8e1f9362007-07-17 04:03:17 -0700902static struct shrinker dcache_shrinker = {
903 .shrink = shrink_dcache_memory,
904 .seeks = DEFAULT_SEEKS,
905};
906
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907/**
908 * d_alloc - allocate a dcache entry
909 * @parent: parent of entry to allocate
910 * @name: qstr of the name
911 *
912 * Allocates a dentry. It returns %NULL if there is insufficient memory
913 * available. On a success the dentry is returned. The name passed in is
914 * copied and the copy passed in may be reused after this call.
915 */
916
917struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
918{
919 struct dentry *dentry;
920 char *dname;
921
Mel Gormane12ba742007-10-16 01:25:52 -0700922 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (!dentry)
924 return NULL;
925
926 if (name->len > DNAME_INLINE_LEN-1) {
927 dname = kmalloc(name->len + 1, GFP_KERNEL);
928 if (!dname) {
929 kmem_cache_free(dentry_cache, dentry);
930 return NULL;
931 }
932 } else {
933 dname = dentry->d_iname;
934 }
935 dentry->d_name.name = dname;
936
937 dentry->d_name.len = name->len;
938 dentry->d_name.hash = name->hash;
939 memcpy(dname, name->name, name->len);
940 dname[name->len] = 0;
941
942 atomic_set(&dentry->d_count, 1);
943 dentry->d_flags = DCACHE_UNHASHED;
944 spin_lock_init(&dentry->d_lock);
945 dentry->d_inode = NULL;
946 dentry->d_parent = NULL;
947 dentry->d_sb = NULL;
948 dentry->d_op = NULL;
949 dentry->d_fsdata = NULL;
950 dentry->d_mounted = 0;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800951#ifdef CONFIG_PROFILING
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 dentry->d_cookie = NULL;
Marcelo Tosatti47ba87e2006-02-03 03:04:06 -0800953#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 INIT_HLIST_NODE(&dentry->d_hash);
955 INIT_LIST_HEAD(&dentry->d_lru);
956 INIT_LIST_HEAD(&dentry->d_subdirs);
957 INIT_LIST_HEAD(&dentry->d_alias);
958
959 if (parent) {
960 dentry->d_parent = dget(parent);
961 dentry->d_sb = parent->d_sb;
962 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800963 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 }
965
966 spin_lock(&dcache_lock);
967 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800968 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 dentry_stat.nr_dentry++;
970 spin_unlock(&dcache_lock);
971
972 return dentry;
973}
974
975struct dentry *d_alloc_name(struct dentry *parent, const char *name)
976{
977 struct qstr q;
978
979 q.name = name;
980 q.len = strlen(name);
981 q.hash = full_name_hash(q.name, q.len);
982 return d_alloc(parent, &q);
983}
984
OGAWA Hirofumi360da902008-10-16 07:50:28 +0900985/* the caller must hold dcache_lock */
986static void __d_instantiate(struct dentry *dentry, struct inode *inode)
987{
988 if (inode)
989 list_add(&dentry->d_alias, &inode->i_dentry);
990 dentry->d_inode = inode;
991 fsnotify_d_instantiate(dentry, inode);
992}
993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994/**
995 * d_instantiate - fill in inode information for a dentry
996 * @entry: dentry to complete
997 * @inode: inode to attach to this dentry
998 *
999 * Fill in inode information in the entry.
1000 *
1001 * This turns negative dentries into productive full members
1002 * of society.
1003 *
1004 * NOTE! This assumes that the inode count has been incremented
1005 * (or otherwise set) by the caller to indicate that it is now
1006 * in use by the dcache.
1007 */
1008
1009void d_instantiate(struct dentry *entry, struct inode * inode)
1010{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001011 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 spin_lock(&dcache_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001013 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 spin_unlock(&dcache_lock);
1015 security_d_instantiate(entry, inode);
1016}
1017
1018/**
1019 * d_instantiate_unique - instantiate a non-aliased dentry
1020 * @entry: dentry to instantiate
1021 * @inode: inode to attach to this dentry
1022 *
1023 * Fill in inode information in the entry. On success, it returns NULL.
1024 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001025 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 *
1027 * Note that in order to avoid conflicts with rename() etc, the caller
1028 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001029 *
1030 * This also assumes that the inode count has been incremented
1031 * (or otherwise set) by the caller to indicate that it is now
1032 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 */
David Howells770bfad2006-08-22 20:06:07 -04001034static struct dentry *__d_instantiate_unique(struct dentry *entry,
1035 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
1037 struct dentry *alias;
1038 int len = entry->d_name.len;
1039 const char *name = entry->d_name.name;
1040 unsigned int hash = entry->d_name.hash;
1041
David Howells770bfad2006-08-22 20:06:07 -04001042 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001043 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001044 return NULL;
1045 }
1046
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1048 struct qstr *qstr = &alias->d_name;
1049
1050 if (qstr->hash != hash)
1051 continue;
1052 if (alias->d_parent != entry->d_parent)
1053 continue;
1054 if (qstr->len != len)
1055 continue;
1056 if (memcmp(qstr->name, name, len))
1057 continue;
1058 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return alias;
1060 }
David Howells770bfad2006-08-22 20:06:07 -04001061
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001062 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 return NULL;
1064}
David Howells770bfad2006-08-22 20:06:07 -04001065
1066struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1067{
1068 struct dentry *result;
1069
1070 BUG_ON(!list_empty(&entry->d_alias));
1071
1072 spin_lock(&dcache_lock);
1073 result = __d_instantiate_unique(entry, inode);
1074 spin_unlock(&dcache_lock);
1075
1076 if (!result) {
1077 security_d_instantiate(entry, inode);
1078 return NULL;
1079 }
1080
1081 BUG_ON(!d_unhashed(result));
1082 iput(inode);
1083 return result;
1084}
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086EXPORT_SYMBOL(d_instantiate_unique);
1087
1088/**
1089 * d_alloc_root - allocate root dentry
1090 * @root_inode: inode to allocate the root for
1091 *
1092 * Allocate a root ("/") dentry for the inode given. The inode is
1093 * instantiated and returned. %NULL is returned if there is insufficient
1094 * memory or the inode passed is %NULL.
1095 */
1096
1097struct dentry * d_alloc_root(struct inode * root_inode)
1098{
1099 struct dentry *res = NULL;
1100
1101 if (root_inode) {
1102 static const struct qstr name = { .name = "/", .len = 1 };
1103
1104 res = d_alloc(NULL, &name);
1105 if (res) {
1106 res->d_sb = root_inode->i_sb;
1107 res->d_parent = res;
1108 d_instantiate(res, root_inode);
1109 }
1110 }
1111 return res;
1112}
1113
1114static inline struct hlist_head *d_hash(struct dentry *parent,
1115 unsigned long hash)
1116{
1117 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1118 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1119 return dentry_hashtable + (hash & D_HASHMASK);
1120}
1121
1122/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001123 * d_obtain_alias - find or allocate a dentry for a given inode
1124 * @inode: inode to allocate the dentry for
1125 *
1126 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1127 * similar open by handle operations. The returned dentry may be anonymous,
1128 * or may have a full name (if the inode was already in the cache).
1129 *
1130 * When called on a directory inode, we must ensure that the inode only ever
1131 * has one dentry. If a dentry is found, that is returned instead of
1132 * allocating a new one.
1133 *
1134 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001135 * to the dentry. In case of an error the reference on the inode is released.
1136 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1137 * be passed in and will be the error will be propagate to the return value,
1138 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001139 */
1140struct dentry *d_obtain_alias(struct inode *inode)
1141{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001142 static const struct qstr anonstring = { .name = "" };
1143 struct dentry *tmp;
1144 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001145
1146 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001147 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001148 if (IS_ERR(inode))
1149 return ERR_CAST(inode);
1150
Christoph Hellwig9308a612008-08-11 15:49:12 +02001151 res = d_find_alias(inode);
1152 if (res)
1153 goto out_iput;
1154
1155 tmp = d_alloc(NULL, &anonstring);
1156 if (!tmp) {
1157 res = ERR_PTR(-ENOMEM);
1158 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001159 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001160 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1161
1162 spin_lock(&dcache_lock);
1163 res = __d_find_alias(inode, 0);
1164 if (res) {
1165 spin_unlock(&dcache_lock);
1166 dput(tmp);
1167 goto out_iput;
1168 }
1169
1170 /* attach a disconnected dentry */
1171 spin_lock(&tmp->d_lock);
1172 tmp->d_sb = inode->i_sb;
1173 tmp->d_inode = inode;
1174 tmp->d_flags |= DCACHE_DISCONNECTED;
1175 tmp->d_flags &= ~DCACHE_UNHASHED;
1176 list_add(&tmp->d_alias, &inode->i_dentry);
1177 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1178 spin_unlock(&tmp->d_lock);
1179
1180 spin_unlock(&dcache_lock);
1181 return tmp;
1182
1183 out_iput:
1184 iput(inode);
1185 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001186}
1187EXPORT_SYMBOL_GPL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189/**
1190 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1191 * @inode: the inode which may have a disconnected dentry
1192 * @dentry: a negative dentry which we want to point to the inode.
1193 *
1194 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1195 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1196 * and return it, else simply d_add the inode to the dentry and return NULL.
1197 *
1198 * This is needed in the lookup routine of any filesystem that is exportable
1199 * (via knfsd) so that we can build dcache paths to directories effectively.
1200 *
1201 * If a dentry was found and moved, then it is returned. Otherwise NULL
1202 * is returned. This matches the expected return value of ->lookup.
1203 *
1204 */
1205struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1206{
1207 struct dentry *new = NULL;
1208
NeilBrown21c0d8f2006-10-04 02:16:16 -07001209 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 spin_lock(&dcache_lock);
1211 new = __d_find_alias(inode, 1);
1212 if (new) {
1213 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1214 spin_unlock(&dcache_lock);
1215 security_d_instantiate(new, inode);
1216 d_rehash(dentry);
1217 d_move(new, dentry);
1218 iput(inode);
1219 } else {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001220 /* already taking dcache_lock, so d_add() by hand */
1221 __d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 spin_unlock(&dcache_lock);
1223 security_d_instantiate(dentry, inode);
1224 d_rehash(dentry);
1225 }
1226 } else
1227 d_add(dentry, inode);
1228 return new;
1229}
1230
Barry Naujok94035402008-05-21 16:50:46 +10001231/**
1232 * d_add_ci - lookup or allocate new dentry with case-exact name
1233 * @inode: the inode case-insensitive lookup has found
1234 * @dentry: the negative dentry that was passed to the parent's lookup func
1235 * @name: the case-exact name to be associated with the returned dentry
1236 *
1237 * This is to avoid filling the dcache with case-insensitive names to the
1238 * same inode, only the actual correct case is stored in the dcache for
1239 * case-insensitive filesystems.
1240 *
1241 * For a case-insensitive lookup match and if the the case-exact dentry
1242 * already exists in in the dcache, use it and return it.
1243 *
1244 * If no entry exists with the exact case name, allocate new dentry with
1245 * the exact case, and return the spliced entry.
1246 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001247struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001248 struct qstr *name)
1249{
1250 int error;
1251 struct dentry *found;
1252 struct dentry *new;
1253
1254 /* Does a dentry matching the name exist already? */
1255 found = d_hash_and_lookup(dentry->d_parent, name);
1256 /* If not, create it now and return */
1257 if (!found) {
1258 new = d_alloc(dentry->d_parent, name);
1259 if (!new) {
1260 error = -ENOMEM;
1261 goto err_out;
1262 }
1263 found = d_splice_alias(inode, new);
1264 if (found) {
1265 dput(new);
1266 return found;
1267 }
1268 return new;
1269 }
1270 /* Matching dentry exists, check if it is negative. */
1271 if (found->d_inode) {
1272 if (unlikely(found->d_inode != inode)) {
1273 /* This can't happen because bad inodes are unhashed. */
1274 BUG_ON(!is_bad_inode(inode));
1275 BUG_ON(!is_bad_inode(found->d_inode));
1276 }
1277 /*
1278 * Already have the inode and the dentry attached, decrement
1279 * the reference count to balance the iget() done
1280 * earlier on. We found the dentry using d_lookup() so it
1281 * cannot be disconnected and thus we do not need to worry
1282 * about any NFS/disconnectedness issues here.
1283 */
1284 iput(inode);
1285 return found;
1286 }
1287 /*
1288 * Negative dentry: instantiate it unless the inode is a directory and
1289 * has a 'disconnected' dentry (i.e. IS_ROOT and DCACHE_DISCONNECTED),
1290 * in which case d_move() that in place of the found dentry.
1291 */
1292 if (!S_ISDIR(inode->i_mode)) {
1293 /* Not a directory; everything is easy. */
1294 d_instantiate(found, inode);
1295 return found;
1296 }
1297 spin_lock(&dcache_lock);
1298 if (list_empty(&inode->i_dentry)) {
1299 /*
1300 * Directory without a 'disconnected' dentry; we need to do
1301 * d_instantiate() by hand because it takes dcache_lock which
1302 * we already hold.
1303 */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001304 __d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001305 spin_unlock(&dcache_lock);
1306 security_d_instantiate(found, inode);
1307 return found;
1308 }
1309 /*
1310 * Directory with a 'disconnected' dentry; get a reference to the
1311 * 'disconnected' dentry.
1312 */
1313 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1314 dget_locked(new);
1315 spin_unlock(&dcache_lock);
1316 /* Do security vodoo. */
1317 security_d_instantiate(found, inode);
1318 /* Move new in place of found. */
1319 d_move(new, found);
1320 /* Balance the iget() we did above. */
1321 iput(inode);
1322 /* Throw away found. */
1323 dput(found);
1324 /* Use new as the actual dentry. */
1325 return new;
1326
1327err_out:
1328 iput(inode);
1329 return ERR_PTR(error);
1330}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332/**
1333 * d_lookup - search for a dentry
1334 * @parent: parent dentry
1335 * @name: qstr of name we wish to find
1336 *
1337 * Searches the children of the parent dentry for the name in question. If
1338 * the dentry is found its reference count is incremented and the dentry
1339 * is returned. The caller must use d_put to free the entry when it has
1340 * finished using it. %NULL is returned on failure.
1341 *
1342 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1343 * Memory barriers are used while updating and doing lockless traversal.
1344 * To avoid races with d_move while rename is happening, d_lock is used.
1345 *
1346 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1347 * and name pointer in one structure pointed by d_qstr.
1348 *
1349 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1350 * lookup is going on.
1351 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001352 * The dentry unused LRU is not updated even if lookup finds the required dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1354 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1355 * acquisition.
1356 *
1357 * d_lookup() is protected against the concurrent renames in some unrelated
1358 * directory using the seqlockt_t rename_lock.
1359 */
1360
1361struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1362{
1363 struct dentry * dentry = NULL;
1364 unsigned long seq;
1365
1366 do {
1367 seq = read_seqbegin(&rename_lock);
1368 dentry = __d_lookup(parent, name);
1369 if (dentry)
1370 break;
1371 } while (read_seqretry(&rename_lock, seq));
1372 return dentry;
1373}
1374
1375struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1376{
1377 unsigned int len = name->len;
1378 unsigned int hash = name->hash;
1379 const unsigned char *str = name->name;
1380 struct hlist_head *head = d_hash(parent,hash);
1381 struct dentry *found = NULL;
1382 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001383 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
1385 rcu_read_lock();
1386
Paul E. McKenney665a7582005-11-07 00:59:17 -08001387 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 struct qstr *qstr;
1389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if (dentry->d_name.hash != hash)
1391 continue;
1392 if (dentry->d_parent != parent)
1393 continue;
1394
1395 spin_lock(&dentry->d_lock);
1396
1397 /*
1398 * Recheck the dentry after taking the lock - d_move may have
1399 * changed things. Don't bother checking the hash because we're
1400 * about to compare the whole name anyway.
1401 */
1402 if (dentry->d_parent != parent)
1403 goto next;
1404
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001405 /* non-existing due to RCU? */
1406 if (d_unhashed(dentry))
1407 goto next;
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 /*
1410 * It is safe to compare names since d_move() cannot
1411 * change the qstr (protected by d_lock).
1412 */
1413 qstr = &dentry->d_name;
1414 if (parent->d_op && parent->d_op->d_compare) {
1415 if (parent->d_op->d_compare(parent, qstr, name))
1416 goto next;
1417 } else {
1418 if (qstr->len != len)
1419 goto next;
1420 if (memcmp(qstr->name, str, len))
1421 goto next;
1422 }
1423
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001424 atomic_inc(&dentry->d_count);
1425 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 spin_unlock(&dentry->d_lock);
1427 break;
1428next:
1429 spin_unlock(&dentry->d_lock);
1430 }
1431 rcu_read_unlock();
1432
1433 return found;
1434}
1435
1436/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001437 * d_hash_and_lookup - hash the qstr then search for a dentry
1438 * @dir: Directory to search in
1439 * @name: qstr of name we wish to find
1440 *
1441 * On hash failure or on lookup failure NULL is returned.
1442 */
1443struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1444{
1445 struct dentry *dentry = NULL;
1446
1447 /*
1448 * Check for a fs-specific hash function. Note that we must
1449 * calculate the standard hash first, as the d_op->d_hash()
1450 * routine may choose to leave the hash value unchanged.
1451 */
1452 name->hash = full_name_hash(name->name, name->len);
1453 if (dir->d_op && dir->d_op->d_hash) {
1454 if (dir->d_op->d_hash(dir, name) < 0)
1455 goto out;
1456 }
1457 dentry = d_lookup(dir, name);
1458out:
1459 return dentry;
1460}
1461
1462/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 * d_validate - verify dentry provided from insecure source
1464 * @dentry: The dentry alleged to be valid child of @dparent
1465 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 *
1467 * An insecure source has sent us a dentry, here we verify it and dget() it.
1468 * This is used by ncpfs in its readdir implementation.
1469 * Zero is returned in the dentry is invalid.
1470 */
1471
1472int d_validate(struct dentry *dentry, struct dentry *dparent)
1473{
1474 struct hlist_head *base;
1475 struct hlist_node *lhp;
1476
1477 /* Check whether the ptr might be valid at all.. */
1478 if (!kmem_ptr_validate(dentry_cache, dentry))
1479 goto out;
1480
1481 if (dentry->d_parent != dparent)
1482 goto out;
1483
1484 spin_lock(&dcache_lock);
1485 base = d_hash(dparent, dentry->d_name.hash);
1486 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001487 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 * as it is parsed under dcache_lock
1489 */
1490 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1491 __dget_locked(dentry);
1492 spin_unlock(&dcache_lock);
1493 return 1;
1494 }
1495 }
1496 spin_unlock(&dcache_lock);
1497out:
1498 return 0;
1499}
1500
1501/*
1502 * When a file is deleted, we have two options:
1503 * - turn this dentry into a negative dentry
1504 * - unhash this dentry and free it.
1505 *
1506 * Usually, we want to just turn this into
1507 * a negative dentry, but if anybody else is
1508 * currently using the dentry or the inode
1509 * we can't do that and we fall back on removing
1510 * it from the hash queues and waiting for
1511 * it to be deleted later when it has no users
1512 */
1513
1514/**
1515 * d_delete - delete a dentry
1516 * @dentry: The dentry to delete
1517 *
1518 * Turn the dentry into a negative dentry if possible, otherwise
1519 * remove it from the hash queues so it can be deleted later
1520 */
1521
1522void d_delete(struct dentry * dentry)
1523{
John McCutchan7a91bf72005-08-08 13:52:16 -04001524 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 /*
1526 * Are we the only user?
1527 */
1528 spin_lock(&dcache_lock);
1529 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001530 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 if (atomic_read(&dentry->d_count) == 1) {
1532 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001533 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 return;
1535 }
1536
1537 if (!d_unhashed(dentry))
1538 __d_drop(dentry);
1539
1540 spin_unlock(&dentry->d_lock);
1541 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001542
1543 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544}
1545
1546static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1547{
1548
1549 entry->d_flags &= ~DCACHE_UNHASHED;
1550 hlist_add_head_rcu(&entry->d_hash, list);
1551}
1552
David Howells770bfad2006-08-22 20:06:07 -04001553static void _d_rehash(struct dentry * entry)
1554{
1555 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1556}
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558/**
1559 * d_rehash - add an entry back to the hash
1560 * @entry: dentry to add to the hash
1561 *
1562 * Adds a dentry to the hash according to its name.
1563 */
1564
1565void d_rehash(struct dentry * entry)
1566{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 spin_lock(&dcache_lock);
1568 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001569 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 spin_unlock(&entry->d_lock);
1571 spin_unlock(&dcache_lock);
1572}
1573
1574#define do_switch(x,y) do { \
1575 __typeof__ (x) __tmp = x; \
1576 x = y; y = __tmp; } while (0)
1577
1578/*
1579 * When switching names, the actual string doesn't strictly have to
1580 * be preserved in the target - because we're dropping the target
1581 * anyway. As such, we can just do a simple memcpy() to copy over
1582 * the new name before we switch.
1583 *
1584 * Note that we have to be a lot more careful about getting the hash
1585 * switched - we have to switch the hash value properly even if it
1586 * then no longer matches the actual (corrupted) string of the target.
1587 * The hash value has to match the hash queue that the dentry is on..
1588 */
1589static void switch_names(struct dentry *dentry, struct dentry *target)
1590{
1591 if (dname_external(target)) {
1592 if (dname_external(dentry)) {
1593 /*
1594 * Both external: swap the pointers
1595 */
1596 do_switch(target->d_name.name, dentry->d_name.name);
1597 } else {
1598 /*
1599 * dentry:internal, target:external. Steal target's
1600 * storage and make target internal.
1601 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07001602 memcpy(target->d_iname, dentry->d_name.name,
1603 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 dentry->d_name.name = target->d_name.name;
1605 target->d_name.name = target->d_iname;
1606 }
1607 } else {
1608 if (dname_external(dentry)) {
1609 /*
1610 * dentry:external, target:internal. Give dentry's
1611 * storage to target and make dentry internal
1612 */
1613 memcpy(dentry->d_iname, target->d_name.name,
1614 target->d_name.len + 1);
1615 target->d_name.name = dentry->d_name.name;
1616 dentry->d_name.name = dentry->d_iname;
1617 } else {
1618 /*
1619 * Both are internal. Just copy target to dentry
1620 */
1621 memcpy(dentry->d_iname, target->d_name.name,
1622 target->d_name.len + 1);
1623 }
1624 }
1625}
1626
1627/*
1628 * We cannibalize "target" when moving dentry on top of it,
1629 * because it's going to be thrown away anyway. We could be more
1630 * polite about it, though.
1631 *
1632 * This forceful removal will result in ugly /proc output if
1633 * somebody holds a file open that got deleted due to a rename.
1634 * We could be nicer about the deleted file, and let it show
J. Bruce Fieldsbc154b12007-10-16 23:29:42 -07001635 * up under the name it had before it was deleted rather than
1636 * under the original name of the file that was moved on top of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 */
1638
Trond Myklebust9eaef272006-10-21 10:24:20 -07001639/*
1640 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 * @dentry: entry to move
1642 * @target: new dentry
1643 *
1644 * Update the dcache to reflect the move of a file name. Negative
1645 * dcache entries should not be moved in this way.
1646 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001647static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
1649 struct hlist_head *list;
1650
1651 if (!dentry->d_inode)
1652 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 write_seqlock(&rename_lock);
1655 /*
1656 * XXXX: do we really need to take target->d_lock?
1657 */
1658 if (target < dentry) {
1659 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001660 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 } else {
1662 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001663 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 }
1665
1666 /* Move the dentry to the target hash queue, if on different bucket */
Denis Chengf77e3492007-10-16 23:30:11 -07001667 if (d_unhashed(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 goto already_unhashed;
1669
1670 hlist_del_rcu(&dentry->d_hash);
1671
1672already_unhashed:
1673 list = d_hash(target->d_parent, target->d_name.hash);
1674 __d_rehash(dentry, list);
1675
1676 /* Unhash the target: dput() will then get rid of it */
1677 __d_drop(target);
1678
Eric Dumazet5160ee62006-01-08 01:03:32 -08001679 list_del(&dentry->d_u.d_child);
1680 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
1682 /* Switch the names.. */
1683 switch_names(dentry, target);
1684 do_switch(dentry->d_name.len, target->d_name.len);
1685 do_switch(dentry->d_name.hash, target->d_name.hash);
1686
1687 /* ... and switch the parents */
1688 if (IS_ROOT(dentry)) {
1689 dentry->d_parent = target->d_parent;
1690 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001691 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 } else {
1693 do_switch(dentry->d_parent, target->d_parent);
1694
1695 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001696 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698
Eric Dumazet5160ee62006-01-08 01:03:32 -08001699 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001701 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 spin_unlock(&dentry->d_lock);
1703 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001704}
1705
1706/**
1707 * d_move - move a dentry
1708 * @dentry: entry to move
1709 * @target: new dentry
1710 *
1711 * Update the dcache to reflect the move of a file name. Negative
1712 * dcache entries should not be moved in this way.
1713 */
1714
1715void d_move(struct dentry * dentry, struct dentry * target)
1716{
1717 spin_lock(&dcache_lock);
1718 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 spin_unlock(&dcache_lock);
1720}
1721
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001722/**
1723 * d_ancestor - search for an ancestor
1724 * @p1: ancestor dentry
1725 * @p2: child dentry
1726 *
1727 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1728 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07001729 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001730struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001731{
1732 struct dentry *p;
1733
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001734 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07001735 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001736 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001737 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001738 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001739}
1740
1741/*
1742 * This helper attempts to cope with remotely renamed directories
1743 *
1744 * It assumes that the caller is already holding
1745 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1746 *
1747 * Note: If ever the locking in lock_rename() changes, then please
1748 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07001749 */
1750static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001751 __releases(dcache_lock)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001752{
1753 struct mutex *m1 = NULL, *m2 = NULL;
1754 struct dentry *ret;
1755
1756 /* If alias and dentry share a parent, then no extra locks required */
1757 if (alias->d_parent == dentry->d_parent)
1758 goto out_unalias;
1759
1760 /* Check for loops */
1761 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001762 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07001763 goto out_err;
1764
1765 /* See lock_rename() */
1766 ret = ERR_PTR(-EBUSY);
1767 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1768 goto out_err;
1769 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1770 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1771 goto out_err;
1772 m2 = &alias->d_parent->d_inode->i_mutex;
1773out_unalias:
1774 d_move_locked(alias, dentry);
1775 ret = alias;
1776out_err:
1777 spin_unlock(&dcache_lock);
1778 if (m2)
1779 mutex_unlock(m2);
1780 if (m1)
1781 mutex_unlock(m1);
1782 return ret;
1783}
1784
1785/*
David Howells770bfad2006-08-22 20:06:07 -04001786 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1787 * named dentry in place of the dentry to be replaced.
1788 */
1789static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1790{
1791 struct dentry *dparent, *aparent;
1792
1793 switch_names(dentry, anon);
1794 do_switch(dentry->d_name.len, anon->d_name.len);
1795 do_switch(dentry->d_name.hash, anon->d_name.hash);
1796
1797 dparent = dentry->d_parent;
1798 aparent = anon->d_parent;
1799
1800 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1801 list_del(&dentry->d_u.d_child);
1802 if (!IS_ROOT(dentry))
1803 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1804 else
1805 INIT_LIST_HEAD(&dentry->d_u.d_child);
1806
1807 anon->d_parent = (dparent == dentry) ? anon : dparent;
1808 list_del(&anon->d_u.d_child);
1809 if (!IS_ROOT(anon))
1810 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1811 else
1812 INIT_LIST_HEAD(&anon->d_u.d_child);
1813
1814 anon->d_flags &= ~DCACHE_DISCONNECTED;
1815}
1816
1817/**
1818 * d_materialise_unique - introduce an inode into the tree
1819 * @dentry: candidate dentry
1820 * @inode: inode to bind to the dentry, to which aliases may be attached
1821 *
1822 * Introduces an dentry into the tree, substituting an extant disconnected
1823 * root directory alias in its place if there is one
1824 */
1825struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1826{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001827 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001828
1829 BUG_ON(!d_unhashed(dentry));
1830
1831 spin_lock(&dcache_lock);
1832
1833 if (!inode) {
1834 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001835 __d_instantiate(dentry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001836 goto found_lock;
1837 }
1838
Trond Myklebust9eaef272006-10-21 10:24:20 -07001839 if (S_ISDIR(inode->i_mode)) {
1840 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001841
Trond Myklebust9eaef272006-10-21 10:24:20 -07001842 /* Does an aliased dentry already exist? */
1843 alias = __d_find_alias(inode, 0);
1844 if (alias) {
1845 actual = alias;
1846 /* Is this an anonymous mountpoint that we could splice
1847 * into our tree? */
1848 if (IS_ROOT(alias)) {
1849 spin_lock(&alias->d_lock);
1850 __d_materialise_dentry(dentry, alias);
1851 __d_drop(alias);
1852 goto found;
1853 }
1854 /* Nope, but we must(!) avoid directory aliasing */
1855 actual = __d_unalias(dentry, alias);
1856 if (IS_ERR(actual))
1857 dput(alias);
1858 goto out_nolock;
1859 }
David Howells770bfad2006-08-22 20:06:07 -04001860 }
1861
1862 /* Add a unique reference */
1863 actual = __d_instantiate_unique(dentry, inode);
1864 if (!actual)
1865 actual = dentry;
1866 else if (unlikely(!d_unhashed(actual)))
1867 goto shouldnt_be_hashed;
1868
1869found_lock:
1870 spin_lock(&actual->d_lock);
1871found:
1872 _d_rehash(actual);
1873 spin_unlock(&actual->d_lock);
1874 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001875out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001876 if (actual == dentry) {
1877 security_d_instantiate(dentry, inode);
1878 return NULL;
1879 }
1880
1881 iput(inode);
1882 return actual;
1883
David Howells770bfad2006-08-22 20:06:07 -04001884shouldnt_be_hashed:
1885 spin_unlock(&dcache_lock);
1886 BUG();
David Howells770bfad2006-08-22 20:06:07 -04001887}
1888
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001889static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01001890{
1891 *buflen -= namelen;
1892 if (*buflen < 0)
1893 return -ENAMETOOLONG;
1894 *buffer -= namelen;
1895 memcpy(*buffer, str, namelen);
1896 return 0;
1897}
1898
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001899static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1900{
1901 return prepend(buffer, buflen, name->name, name->len);
1902}
1903
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001905 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001906 * @path: the dentry/vfsmount to report
1907 * @root: root vfsmnt/dentry (may be modified by this function)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 * @buffer: buffer to return value in
1909 * @buflen: buffer length
1910 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001911 * Convert a dentry into an ASCII path name. If the entry has been deleted
1912 * the string " (deleted)" is appended. Note that this is ambiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001914 * Returns the buffer or an error code if the path was too long.
1915 *
1916 * "buflen" should be positive. Caller holds the dcache_lock.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001917 *
1918 * If path is not reachable from the supplied root, then the value of
1919 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001921char *__d_path(const struct path *path, struct path *root,
1922 char *buffer, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923{
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001924 struct dentry *dentry = path->dentry;
1925 struct vfsmount *vfsmnt = path->mnt;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001926 char *end = buffer + buflen;
1927 char *retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001929 spin_lock(&vfsmount_lock);
Ram Pai6092d042008-03-27 13:06:20 +01001930 prepend(&end, &buflen, "\0", 1);
1931 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
1932 (prepend(&end, &buflen, " (deleted)", 10) != 0))
Linus Torvalds552ce542007-02-13 12:08:18 -08001933 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001934
1935 if (buflen < 1)
1936 goto Elong;
1937 /* Get '/' right */
1938 retval = end-1;
1939 *retval = '/';
1940
1941 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 struct dentry * parent;
1943
Jan Blunck329c97f2008-02-14 19:38:31 -08001944 if (dentry == root->dentry && vfsmnt == root->mnt)
Linus Torvalds552ce542007-02-13 12:08:18 -08001945 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08001947 /* Global root? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 if (vfsmnt->mnt_parent == vfsmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 goto global_root;
1950 }
1951 dentry = vfsmnt->mnt_mountpoint;
1952 vfsmnt = vfsmnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 continue;
1954 }
1955 parent = dentry->d_parent;
1956 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001957 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01001958 (prepend(&end, &buflen, "/", 1) != 0))
Linus Torvalds552ce542007-02-13 12:08:18 -08001959 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001960 retval = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 dentry = parent;
1962 }
1963
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001964out:
1965 spin_unlock(&vfsmount_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08001966 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968global_root:
Ram Pai6092d042008-03-27 13:06:20 +01001969 retval += 1; /* hit the slash */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001970 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 goto Elong;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001972 root->mnt = vfsmnt;
1973 root->dentry = dentry;
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001974 goto out;
1975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976Elong:
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001977 retval = ERR_PTR(-ENAMETOOLONG);
1978 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979}
1980
Jan Bluncka03a8a702008-02-14 19:38:32 -08001981/**
1982 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08001983 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08001984 * @buf: buffer to return value in
1985 * @buflen: buffer length
1986 *
1987 * Convert a dentry into an ASCII path name. If the entry has been deleted
1988 * the string " (deleted)" is appended. Note that this is ambiguous.
1989 *
1990 * Returns the buffer or an error code if the path was too long.
1991 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001992 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08001993 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07001994char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995{
1996 char *res;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001997 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001998 struct path tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002000 /*
2001 * We have various synthetic filesystems that never get mounted. On
2002 * these filesystems dentries are never used for lookup purposes, and
2003 * thus don't need to be hashed. They also don't need a name until a
2004 * user wants to identify the object in /proc/pid/fd/. The little hack
2005 * below allows us to generate a name for these objects on demand:
2006 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002007 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2008 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002009
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002011 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002012 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 read_unlock(&current->fs->lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002014 spin_lock(&dcache_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002015 tmp = root;
2016 res = __d_path(path, &tmp, buf, buflen);
Linus Torvalds552ce542007-02-13 12:08:18 -08002017 spin_unlock(&dcache_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002018 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 return res;
2020}
2021
2022/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002023 * Helper function for dentry_operations.d_dname() members
2024 */
2025char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2026 const char *fmt, ...)
2027{
2028 va_list args;
2029 char temp[64];
2030 int sz;
2031
2032 va_start(args, fmt);
2033 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2034 va_end(args);
2035
2036 if (sz > sizeof(temp) || sz > buflen)
2037 return ERR_PTR(-ENAMETOOLONG);
2038
2039 buffer += buflen - sz;
2040 return memcpy(buffer, temp, sz);
2041}
2042
2043/*
Ram Pai6092d042008-03-27 13:06:20 +01002044 * Write full pathname from the root of the filesystem into the buffer.
2045 */
2046char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2047{
2048 char *end = buf + buflen;
2049 char *retval;
2050
2051 spin_lock(&dcache_lock);
2052 prepend(&end, &buflen, "\0", 1);
2053 if (!IS_ROOT(dentry) && d_unhashed(dentry) &&
2054 (prepend(&end, &buflen, "//deleted", 9) != 0))
2055 goto Elong;
2056 if (buflen < 1)
2057 goto Elong;
2058 /* Get '/' right */
2059 retval = end-1;
2060 *retval = '/';
2061
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002062 while (!IS_ROOT(dentry)) {
2063 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01002064
Ram Pai6092d042008-03-27 13:06:20 +01002065 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002066 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01002067 (prepend(&end, &buflen, "/", 1) != 0))
2068 goto Elong;
2069
2070 retval = end;
2071 dentry = parent;
2072 }
2073 spin_unlock(&dcache_lock);
2074 return retval;
2075Elong:
2076 spin_unlock(&dcache_lock);
2077 return ERR_PTR(-ENAMETOOLONG);
2078}
2079
2080/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 * NOTE! The user-level library version returns a
2082 * character pointer. The kernel system call just
2083 * returns the length of the buffer filled (which
2084 * includes the ending '\0' character), or a negative
2085 * error value. So libc would do something like
2086 *
2087 * char *getcwd(char * buf, size_t size)
2088 * {
2089 * int retval;
2090 *
2091 * retval = sys_getcwd(buf, size);
2092 * if (retval >= 0)
2093 * return buf;
2094 * errno = -retval;
2095 * return NULL;
2096 * }
2097 */
2098asmlinkage long sys_getcwd(char __user *buf, unsigned long size)
2099{
Linus Torvalds552ce542007-02-13 12:08:18 -08002100 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002101 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002102 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
2104 if (!page)
2105 return -ENOMEM;
2106
2107 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002108 pwd = current->fs->pwd;
Ram Pai6092d042008-03-27 13:06:20 +01002109 path_get(&pwd);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002110 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002111 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 read_unlock(&current->fs->lock);
2113
Linus Torvalds552ce542007-02-13 12:08:18 -08002114 error = -ENOENT;
2115 /* Has the current directory has been unlinked? */
2116 spin_lock(&dcache_lock);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002117 if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002118 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002119 struct path tmp = root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002120 char * cwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002122 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
Linus Torvalds552ce542007-02-13 12:08:18 -08002123 spin_unlock(&dcache_lock);
2124
2125 error = PTR_ERR(cwd);
2126 if (IS_ERR(cwd))
2127 goto out;
2128
2129 error = -ERANGE;
2130 len = PAGE_SIZE + page - cwd;
2131 if (len <= size) {
2132 error = len;
2133 if (copy_to_user(buf, cwd, len))
2134 error = -EFAULT;
2135 }
2136 } else
2137 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
2139out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002140 path_put(&pwd);
2141 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 free_page((unsigned long) page);
2143 return error;
2144}
2145
2146/*
2147 * Test whether new_dentry is a subdirectory of old_dentry.
2148 *
2149 * Trivially implemented using the dcache structure
2150 */
2151
2152/**
2153 * is_subdir - is new dentry a subdirectory of old_dentry
2154 * @new_dentry: new dentry
2155 * @old_dentry: old dentry
2156 *
2157 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2158 * Returns 0 otherwise.
2159 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2160 */
2161
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002162int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163{
2164 int result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 unsigned long seq;
2166
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002167 /* FIXME: This is old behavior, needed? Please check callers. */
2168 if (new_dentry == old_dentry)
2169 return 1;
2170
2171 /*
2172 * Need rcu_readlock to protect against the d_parent trashing
2173 * due to d_move
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 */
2175 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002176 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 seq = read_seqbegin(&rename_lock);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002179 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002181 else
2182 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 } while (read_seqretry(&rename_lock, seq));
2184 rcu_read_unlock();
2185
2186 return result;
2187}
2188
2189void d_genocide(struct dentry *root)
2190{
2191 struct dentry *this_parent = root;
2192 struct list_head *next;
2193
2194 spin_lock(&dcache_lock);
2195repeat:
2196 next = this_parent->d_subdirs.next;
2197resume:
2198 while (next != &this_parent->d_subdirs) {
2199 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002200 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 next = tmp->next;
2202 if (d_unhashed(dentry)||!dentry->d_inode)
2203 continue;
2204 if (!list_empty(&dentry->d_subdirs)) {
2205 this_parent = dentry;
2206 goto repeat;
2207 }
2208 atomic_dec(&dentry->d_count);
2209 }
2210 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002211 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 atomic_dec(&this_parent->d_count);
2213 this_parent = this_parent->d_parent;
2214 goto resume;
2215 }
2216 spin_unlock(&dcache_lock);
2217}
2218
2219/**
2220 * find_inode_number - check for dentry with name
2221 * @dir: directory to check
2222 * @name: Name to find.
2223 *
2224 * Check whether a dentry already exists for the given name,
2225 * and return the inode number if it has an inode. Otherwise
2226 * 0 is returned.
2227 *
2228 * This routine is used to post-process directory listings for
2229 * filesystems using synthetic inode numbers, and is necessary
2230 * to keep getcwd() working.
2231 */
2232
2233ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2234{
2235 struct dentry * dentry;
2236 ino_t ino = 0;
2237
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002238 dentry = d_hash_and_lookup(dir, name);
2239 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 if (dentry->d_inode)
2241 ino = dentry->d_inode->i_ino;
2242 dput(dentry);
2243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 return ino;
2245}
2246
2247static __initdata unsigned long dhash_entries;
2248static int __init set_dhash_entries(char *str)
2249{
2250 if (!str)
2251 return 0;
2252 dhash_entries = simple_strtoul(str, &str, 0);
2253 return 1;
2254}
2255__setup("dhash_entries=", set_dhash_entries);
2256
2257static void __init dcache_init_early(void)
2258{
2259 int loop;
2260
2261 /* If hashes are distributed across NUMA nodes, defer
2262 * hash allocation until vmalloc space is available.
2263 */
2264 if (hashdist)
2265 return;
2266
2267 dentry_hashtable =
2268 alloc_large_system_hash("Dentry cache",
2269 sizeof(struct hlist_head),
2270 dhash_entries,
2271 13,
2272 HASH_EARLY,
2273 &d_hash_shift,
2274 &d_hash_mask,
2275 0);
2276
2277 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2278 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2279}
2280
Denis Cheng74bf17c2007-10-16 23:26:30 -07002281static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282{
2283 int loop;
2284
2285 /*
2286 * A constructor could be added for stable state like the lists,
2287 * but it is probably not worth it because of the cache nature
2288 * of the dcache.
2289 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002290 dentry_cache = KMEM_CACHE(dentry,
2291 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292
Rusty Russell8e1f9362007-07-17 04:03:17 -07002293 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294
2295 /* Hash may have been set up in dcache_init_early */
2296 if (!hashdist)
2297 return;
2298
2299 dentry_hashtable =
2300 alloc_large_system_hash("Dentry cache",
2301 sizeof(struct hlist_head),
2302 dhash_entries,
2303 13,
2304 0,
2305 &d_hash_shift,
2306 &d_hash_mask,
2307 0);
2308
2309 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2310 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2311}
2312
2313/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002314struct kmem_cache *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315
2316/* SLAB cache for file structures */
Christoph Lametere18b8902006-12-06 20:33:20 -08002317struct kmem_cache *filp_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
2319EXPORT_SYMBOL(d_genocide);
2320
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321void __init vfs_caches_init_early(void)
2322{
2323 dcache_init_early();
2324 inode_init_early();
2325}
2326
2327void __init vfs_caches_init(unsigned long mempages)
2328{
2329 unsigned long reserve;
2330
2331 /* Base hash sizes on available memory, with a reserve equal to
2332 150% of current kernel size */
2333
2334 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2335 mempages -= reserve;
2336
2337 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002338 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
2340 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002341 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342
Denis Cheng74bf17c2007-10-16 23:26:30 -07002343 dcache_init();
2344 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002346 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 bdev_cache_init();
2348 chrdev_init();
2349}
2350
2351EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352EXPORT_SYMBOL(d_alloc_root);
2353EXPORT_SYMBOL(d_delete);
2354EXPORT_SYMBOL(d_find_alias);
2355EXPORT_SYMBOL(d_instantiate);
2356EXPORT_SYMBOL(d_invalidate);
2357EXPORT_SYMBOL(d_lookup);
2358EXPORT_SYMBOL(d_move);
David Howells770bfad2006-08-22 20:06:07 -04002359EXPORT_SYMBOL_GPL(d_materialise_unique);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360EXPORT_SYMBOL(d_path);
2361EXPORT_SYMBOL(d_prune_aliases);
2362EXPORT_SYMBOL(d_rehash);
2363EXPORT_SYMBOL(d_splice_alias);
Barry Naujok94035402008-05-21 16:50:46 +10002364EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365EXPORT_SYMBOL(d_validate);
2366EXPORT_SYMBOL(dget_locked);
2367EXPORT_SYMBOL(dput);
2368EXPORT_SYMBOL(find_inode_number);
2369EXPORT_SYMBOL(have_submounts);
2370EXPORT_SYMBOL(names_cachep);
2371EXPORT_SYMBOL(shrink_dcache_parent);
2372EXPORT_SYMBOL(shrink_dcache_sb);