blob: d96047b4a633a86cb7ae711cb29362ac6cda061a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
26#include <linux/module.h>
27#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
David Howells07f3f052006-09-30 20:52:18 +020036#include "internal.h"
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}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700260EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262/**
263 * d_invalidate - invalidate a dentry
264 * @dentry: dentry to invalidate
265 *
266 * Try to invalidate the dentry if it turns out to be
267 * possible. If there are other dentries that can be
268 * reached through this one we can't delete it and we
269 * return -EBUSY. On success we return 0.
270 *
271 * no dcache lock.
272 */
273
274int d_invalidate(struct dentry * dentry)
275{
276 /*
277 * If it's already been dropped, return OK.
278 */
279 spin_lock(&dcache_lock);
280 if (d_unhashed(dentry)) {
281 spin_unlock(&dcache_lock);
282 return 0;
283 }
284 /*
285 * Check whether to do a partial shrink_dcache
286 * to get rid of unused child entries.
287 */
288 if (!list_empty(&dentry->d_subdirs)) {
289 spin_unlock(&dcache_lock);
290 shrink_dcache_parent(dentry);
291 spin_lock(&dcache_lock);
292 }
293
294 /*
295 * Somebody else still using it?
296 *
297 * If it's a directory, we can't drop it
298 * for fear of somebody re-populating it
299 * with children (even though dropping it
300 * would make it unreachable from the root,
301 * we might still populate it if it was a
302 * working directory or similar).
303 */
304 spin_lock(&dentry->d_lock);
305 if (atomic_read(&dentry->d_count) > 1) {
306 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
307 spin_unlock(&dentry->d_lock);
308 spin_unlock(&dcache_lock);
309 return -EBUSY;
310 }
311 }
312
313 __d_drop(dentry);
314 spin_unlock(&dentry->d_lock);
315 spin_unlock(&dcache_lock);
316 return 0;
317}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700318EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320/* This should be called _only_ with dcache_lock held */
321
322static inline struct dentry * __dget_locked(struct dentry *dentry)
323{
324 atomic_inc(&dentry->d_count);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700325 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 return dentry;
327}
328
329struct dentry * dget_locked(struct dentry *dentry)
330{
331 return __dget_locked(dentry);
332}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700333EXPORT_SYMBOL(dget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335/**
336 * d_find_alias - grab a hashed alias of inode
337 * @inode: inode in question
338 * @want_discon: flag, used by d_splice_alias, to request
339 * that only a DISCONNECTED alias be returned.
340 *
341 * If inode has a hashed alias, or is a directory and has any alias,
342 * acquire the reference to alias and return it. Otherwise return NULL.
343 * Notice that if inode is a directory there can be only one alias and
344 * it can be unhashed only if it has no children, or if it is the root
345 * of a filesystem.
346 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700347 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700349 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 */
351
352static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
353{
354 struct list_head *head, *next, *tmp;
355 struct dentry *alias, *discon_alias=NULL;
356
357 head = &inode->i_dentry;
358 next = inode->i_dentry.next;
359 while (next != head) {
360 tmp = next;
361 next = tmp->next;
362 prefetch(next);
363 alias = list_entry(tmp, struct dentry, d_alias);
364 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700365 if (IS_ROOT(alias) &&
366 (alias->d_flags & DCACHE_DISCONNECTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 discon_alias = alias;
368 else if (!want_discon) {
369 __dget_locked(alias);
370 return alias;
371 }
372 }
373 }
374 if (discon_alias)
375 __dget_locked(discon_alias);
376 return discon_alias;
377}
378
379struct dentry * d_find_alias(struct inode *inode)
380{
David Howells214fda12006-03-25 03:06:36 -0800381 struct dentry *de = NULL;
382
383 if (!list_empty(&inode->i_dentry)) {
384 spin_lock(&dcache_lock);
385 de = __d_find_alias(inode, 0);
386 spin_unlock(&dcache_lock);
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 return de;
389}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700390EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392/*
393 * Try to kill dentries associated with this inode.
394 * WARNING: you must own a reference to inode.
395 */
396void d_prune_aliases(struct inode *inode)
397{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700398 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399restart:
400 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700401 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 spin_lock(&dentry->d_lock);
403 if (!atomic_read(&dentry->d_count)) {
404 __dget_locked(dentry);
405 __d_drop(dentry);
406 spin_unlock(&dentry->d_lock);
407 spin_unlock(&dcache_lock);
408 dput(dentry);
409 goto restart;
410 }
411 spin_unlock(&dentry->d_lock);
412 }
413 spin_unlock(&dcache_lock);
414}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700415EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700418 * Throw away a dentry - free the inode, dput the parent. This requires that
419 * the LRU list has already been removed.
420 *
Miklos Szeredi85864e12007-10-16 23:27:09 -0700421 * Try to prune ancestors as well. This is necessary to prevent
422 * quadratic behavior of shrink_dcache_parent(), but is also expected
423 * to be beneficial in reducing dentry cache fragmentation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 */
Miklos Szeredi85864e12007-10-16 23:27:09 -0700425static void prune_one_dentry(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200426 __releases(dentry->d_lock)
427 __releases(dcache_lock)
428 __acquires(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700431 dentry = d_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700432
433 /*
434 * Prune ancestors. Locking is simpler than in dput(),
435 * because dcache_lock needs to be taken anyway.
436 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 spin_lock(&dcache_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700438 while (dentry) {
439 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
440 return;
441
442 if (dentry->d_op && dentry->d_op->d_delete)
443 dentry->d_op->d_delete(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700444 dentry_lru_del_init(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700445 __d_drop(dentry);
446 dentry = d_kill(dentry);
447 spin_lock(&dcache_lock);
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449}
450
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700451/*
452 * Shrink the dentry LRU on a given superblock.
453 * @sb : superblock to shrink dentry LRU.
454 * @count: If count is NULL, we prune all dentries on superblock.
455 * @flags: If flags is non-zero, we need to do special processing based on
456 * which flags are set. This means we don't need to maintain multiple
457 * similar copies of this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700459static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700461 LIST_HEAD(referenced);
462 LIST_HEAD(tmp);
463 struct dentry *dentry;
464 int cnt = 0;
465
466 BUG_ON(!sb);
467 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700469 if (count != NULL)
470 /* called from prune_dcache() and shrink_dcache_parent() */
471 cnt = *count;
472restart:
473 if (count == NULL)
474 list_splice_init(&sb->s_dentry_lru, &tmp);
475 else {
476 while (!list_empty(&sb->s_dentry_lru)) {
477 dentry = list_entry(sb->s_dentry_lru.prev,
478 struct dentry, d_lru);
479 BUG_ON(dentry->d_sb != sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700481 spin_lock(&dentry->d_lock);
482 /*
483 * If we are honouring the DCACHE_REFERENCED flag and
484 * the dentry has this flag set, don't free it. Clear
485 * the flag and put it back on the LRU.
NeilBrown0feae5c2006-06-22 14:47:28 -0700486 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700487 if ((flags & DCACHE_REFERENCED)
488 && (dentry->d_flags & DCACHE_REFERENCED)) {
489 dentry->d_flags &= ~DCACHE_REFERENCED;
npiggin@suse.dec490d792009-04-26 20:25:53 +1000490 list_move(&dentry->d_lru, &referenced);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700491 spin_unlock(&dentry->d_lock);
492 } else {
493 list_move_tail(&dentry->d_lru, &tmp);
494 spin_unlock(&dentry->d_lock);
495 cnt--;
496 if (!cnt)
497 break;
NeilBrown0feae5c2006-06-22 14:47:28 -0700498 }
Kentaro Makitaf3c6ba92008-07-25 19:44:40 -0700499 cond_resched_lock(&dcache_lock);
NeilBrown0feae5c2006-06-22 14:47:28 -0700500 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700501 }
502 while (!list_empty(&tmp)) {
503 dentry = list_entry(tmp.prev, struct dentry, d_lru);
504 dentry_lru_del_init(dentry);
505 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 /*
507 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700508 * the LRU because of laziness during lookup. Do not free
509 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700511 if (atomic_read(&dentry->d_count)) {
512 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 continue;
514 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700515 prune_one_dentry(dentry);
516 /* dentry->d_lock was dropped in prune_one_dentry() */
517 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700519 if (count == NULL && !list_empty(&sb->s_dentry_lru))
520 goto restart;
521 if (count != NULL)
522 *count = cnt;
523 if (!list_empty(&referenced))
524 list_splice(&referenced, &sb->s_dentry_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 spin_unlock(&dcache_lock);
526}
527
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700528/**
529 * prune_dcache - shrink the dcache
530 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700532 * Shrink the dcache. This is done when we need more memory, or simply when we
533 * need to unmount something (at which point we need to unuse all dentries).
534 *
535 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700537static void prune_dcache(int count)
538{
Al Viro79893c12010-03-22 20:27:55 -0400539 struct super_block *sb, *n;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700540 int w_count;
541 int unused = dentry_stat.nr_unused;
542 int prune_ratio;
543 int pruned;
544
545 if (unused == 0 || count == 0)
546 return;
547 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700548 if (count >= unused)
549 prune_ratio = 1;
550 else
551 prune_ratio = unused / count;
552 spin_lock(&sb_lock);
Al Viro79893c12010-03-22 20:27:55 -0400553 list_for_each_entry_safe(sb, n, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400554 if (list_empty(&sb->s_instances))
555 continue;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700556 if (sb->s_nr_dentry_unused == 0)
557 continue;
558 sb->s_count++;
559 /* Now, we reclaim unused dentrins with fairness.
560 * We reclaim them same percentage from each superblock.
561 * We calculate number of dentries to scan on this sb
562 * as follows, but the implementation is arranged to avoid
563 * overflows:
564 * number of dentries to scan on this sb =
565 * count * (number of dentries on this sb /
566 * number of dentries in the machine)
567 */
568 spin_unlock(&sb_lock);
569 if (prune_ratio != 1)
570 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
571 else
572 w_count = sb->s_nr_dentry_unused;
573 pruned = w_count;
574 /*
575 * We need to be sure this filesystem isn't being unmounted,
576 * otherwise we could race with generic_shutdown_super(), and
577 * end up holding a reference to an inode while the filesystem
578 * is unmounted. So we try to get s_umount, and make sure
579 * s_root isn't NULL.
580 */
581 if (down_read_trylock(&sb->s_umount)) {
582 if ((sb->s_root != NULL) &&
583 (!list_empty(&sb->s_dentry_lru))) {
584 spin_unlock(&dcache_lock);
585 __shrink_dcache_sb(sb, &w_count,
586 DCACHE_REFERENCED);
587 pruned -= w_count;
588 spin_lock(&dcache_lock);
589 }
590 up_read(&sb->s_umount);
591 }
592 spin_lock(&sb_lock);
593 count -= pruned;
Al Viro79893c12010-03-22 20:27:55 -0400594 __put_super(sb);
595 /* more work left to do? */
596 if (count <= 0)
597 break;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700598 }
599 spin_unlock(&sb_lock);
600 spin_unlock(&dcache_lock);
601}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603/**
604 * shrink_dcache_sb - shrink dcache for a superblock
605 * @sb: superblock
606 *
607 * Shrink the dcache for the specified super block. This
608 * is used to free the dcache before unmounting a file
609 * system
610 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611void shrink_dcache_sb(struct super_block * sb)
612{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700613 __shrink_dcache_sb(sb, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700615EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617/*
David Howellsc636ebd2006-10-11 01:22:19 -0700618 * destroy a single subtree of dentries for unmount
619 * - see the comments on shrink_dcache_for_umount() for a description of the
620 * locking
621 */
622static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
623{
624 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700625 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700626
627 BUG_ON(!IS_ROOT(dentry));
628
629 /* detach this root from the system */
630 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700631 dentry_lru_del_init(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700632 __d_drop(dentry);
633 spin_unlock(&dcache_lock);
634
635 for (;;) {
636 /* descend to the first leaf in the current subtree */
637 while (!list_empty(&dentry->d_subdirs)) {
638 struct dentry *loop;
639
640 /* this is a branch with children - detach all of them
641 * from the system in one go */
642 spin_lock(&dcache_lock);
643 list_for_each_entry(loop, &dentry->d_subdirs,
644 d_u.d_child) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700645 dentry_lru_del_init(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700646 __d_drop(loop);
647 cond_resched_lock(&dcache_lock);
648 }
649 spin_unlock(&dcache_lock);
650
651 /* move to the first child */
652 dentry = list_entry(dentry->d_subdirs.next,
653 struct dentry, d_u.d_child);
654 }
655
656 /* consume the dentries from this leaf up through its parents
657 * until we find one with children or run out altogether */
658 do {
659 struct inode *inode;
660
661 if (atomic_read(&dentry->d_count) != 0) {
662 printk(KERN_ERR
663 "BUG: Dentry %p{i=%lx,n=%s}"
664 " still in use (%d)"
665 " [unmount of %s %s]\n",
666 dentry,
667 dentry->d_inode ?
668 dentry->d_inode->i_ino : 0UL,
669 dentry->d_name.name,
670 atomic_read(&dentry->d_count),
671 dentry->d_sb->s_type->name,
672 dentry->d_sb->s_id);
673 BUG();
674 }
675
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900676 if (IS_ROOT(dentry))
David Howellsc636ebd2006-10-11 01:22:19 -0700677 parent = NULL;
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900678 else {
679 parent = dentry->d_parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700680 atomic_dec(&parent->d_count);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900681 }
David Howellsc636ebd2006-10-11 01:22:19 -0700682
683 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700684 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700685
686 inode = dentry->d_inode;
687 if (inode) {
688 dentry->d_inode = NULL;
689 list_del_init(&dentry->d_alias);
690 if (dentry->d_op && dentry->d_op->d_iput)
691 dentry->d_op->d_iput(dentry, inode);
692 else
693 iput(inode);
694 }
695
696 d_free(dentry);
697
698 /* finished when we fall off the top of the tree,
699 * otherwise we ascend to the parent and move to the
700 * next sibling if there is one */
701 if (!parent)
David Howellsf8713572006-10-28 10:38:46 -0700702 goto out;
David Howellsc636ebd2006-10-11 01:22:19 -0700703
704 dentry = parent;
705
706 } while (list_empty(&dentry->d_subdirs));
707
708 dentry = list_entry(dentry->d_subdirs.next,
709 struct dentry, d_u.d_child);
710 }
David Howellsf8713572006-10-28 10:38:46 -0700711out:
712 /* several dentries were freed, need to correct nr_dentry */
713 spin_lock(&dcache_lock);
714 dentry_stat.nr_dentry -= detached;
715 spin_unlock(&dcache_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700716}
717
718/*
719 * destroy the dentries attached to a superblock on unmounting
720 * - we don't need to use dentry->d_lock, and only need dcache_lock when
721 * removing the dentry from the system lists and hashes because:
722 * - the superblock is detached from all mountings and open files, so the
723 * dentry trees will not be rearranged by the VFS
724 * - s_umount is write-locked, so the memory pressure shrinker will ignore
725 * any dentries belonging to this superblock that it comes across
726 * - the filesystem itself is no longer permitted to rearrange the dentries
727 * in this superblock
728 */
729void shrink_dcache_for_umount(struct super_block *sb)
730{
731 struct dentry *dentry;
732
733 if (down_read_trylock(&sb->s_umount))
734 BUG();
735
736 dentry = sb->s_root;
737 sb->s_root = NULL;
738 atomic_dec(&dentry->d_count);
739 shrink_dcache_for_umount_subtree(dentry);
740
741 while (!hlist_empty(&sb->s_anon)) {
742 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
743 shrink_dcache_for_umount_subtree(dentry);
744 }
745}
746
747/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 * Search for at least 1 mount point in the dentry's subdirs.
749 * We descend to the next level whenever the d_subdirs
750 * list is non-empty and continue searching.
751 */
752
753/**
754 * have_submounts - check for mounts over a dentry
755 * @parent: dentry to check.
756 *
757 * Return true if the parent or its subdirectories contain
758 * a mount point
759 */
760
761int have_submounts(struct dentry *parent)
762{
763 struct dentry *this_parent = parent;
764 struct list_head *next;
765
766 spin_lock(&dcache_lock);
767 if (d_mountpoint(parent))
768 goto positive;
769repeat:
770 next = this_parent->d_subdirs.next;
771resume:
772 while (next != &this_parent->d_subdirs) {
773 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800774 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 next = tmp->next;
776 /* Have we found a mount point ? */
777 if (d_mountpoint(dentry))
778 goto positive;
779 if (!list_empty(&dentry->d_subdirs)) {
780 this_parent = dentry;
781 goto repeat;
782 }
783 }
784 /*
785 * All done at this level ... ascend and resume the search.
786 */
787 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800788 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 this_parent = this_parent->d_parent;
790 goto resume;
791 }
792 spin_unlock(&dcache_lock);
793 return 0; /* No mount points found in tree */
794positive:
795 spin_unlock(&dcache_lock);
796 return 1;
797}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700798EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800/*
801 * Search the dentry child list for the specified parent,
802 * and move any unused dentries to the end of the unused
803 * list for prune_dcache(). We descend to the next level
804 * whenever the d_subdirs list is non-empty and continue
805 * searching.
806 *
807 * It returns zero iff there are no unused children,
808 * otherwise it returns the number of children moved to
809 * the end of the unused list. This may not be the total
810 * number of unused children, because select_parent can
811 * drop the lock and return early due to latency
812 * constraints.
813 */
814static int select_parent(struct dentry * parent)
815{
816 struct dentry *this_parent = parent;
817 struct list_head *next;
818 int found = 0;
819
820 spin_lock(&dcache_lock);
821repeat:
822 next = this_parent->d_subdirs.next;
823resume:
824 while (next != &this_parent->d_subdirs) {
825 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800826 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 next = tmp->next;
828
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700829 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 /*
831 * move only zero ref count dentries to the end
832 * of the unused list for prune_dcache
833 */
834 if (!atomic_read(&dentry->d_count)) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700835 dentry_lru_add_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 found++;
837 }
838
839 /*
840 * We can return to the caller if we have found some (this
841 * ensures forward progress). We'll be coming back to find
842 * the rest.
843 */
844 if (found && need_resched())
845 goto out;
846
847 /*
848 * Descend a level if the d_subdirs list is non-empty.
849 */
850 if (!list_empty(&dentry->d_subdirs)) {
851 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 goto repeat;
853 }
854 }
855 /*
856 * All done at this level ... ascend and resume the search.
857 */
858 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800859 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 goto resume;
862 }
863out:
864 spin_unlock(&dcache_lock);
865 return found;
866}
867
868/**
869 * shrink_dcache_parent - prune dcache
870 * @parent: parent of entries to prune
871 *
872 * Prune the dcache to remove unused children of the parent dentry.
873 */
874
875void shrink_dcache_parent(struct dentry * parent)
876{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700877 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 int found;
879
880 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700881 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700883EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/*
886 * Scan `nr' dentries and return the number which remain.
887 *
888 * We need to avoid reentering the filesystem if the caller is performing a
889 * GFP_NOFS allocation attempt. One example deadlock is:
890 *
891 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
892 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
893 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
894 *
895 * In this case we return -1 to tell the caller that we baled.
896 */
Al Viro27496a82005-10-21 03:20:48 -0400897static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898{
899 if (nr) {
900 if (!(gfp_mask & __GFP_FS))
901 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700902 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 }
904 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
905}
906
Rusty Russell8e1f9362007-07-17 04:03:17 -0700907static struct shrinker dcache_shrinker = {
908 .shrink = shrink_dcache_memory,
909 .seeks = DEFAULT_SEEKS,
910};
911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912/**
913 * d_alloc - allocate a dcache entry
914 * @parent: parent of entry to allocate
915 * @name: qstr of the name
916 *
917 * Allocates a dentry. It returns %NULL if there is insufficient memory
918 * available. On a success the dentry is returned. The name passed in is
919 * copied and the copy passed in may be reused after this call.
920 */
921
922struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
923{
924 struct dentry *dentry;
925 char *dname;
926
Mel Gormane12ba742007-10-16 01:25:52 -0700927 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (!dentry)
929 return NULL;
930
931 if (name->len > DNAME_INLINE_LEN-1) {
932 dname = kmalloc(name->len + 1, GFP_KERNEL);
933 if (!dname) {
934 kmem_cache_free(dentry_cache, dentry);
935 return NULL;
936 }
937 } else {
938 dname = dentry->d_iname;
939 }
940 dentry->d_name.name = dname;
941
942 dentry->d_name.len = name->len;
943 dentry->d_name.hash = name->hash;
944 memcpy(dname, name->name, name->len);
945 dname[name->len] = 0;
946
947 atomic_set(&dentry->d_count, 1);
948 dentry->d_flags = DCACHE_UNHASHED;
949 spin_lock_init(&dentry->d_lock);
950 dentry->d_inode = NULL;
951 dentry->d_parent = NULL;
952 dentry->d_sb = NULL;
953 dentry->d_op = NULL;
954 dentry->d_fsdata = NULL;
955 dentry->d_mounted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 INIT_HLIST_NODE(&dentry->d_hash);
957 INIT_LIST_HEAD(&dentry->d_lru);
958 INIT_LIST_HEAD(&dentry->d_subdirs);
959 INIT_LIST_HEAD(&dentry->d_alias);
960
961 if (parent) {
962 dentry->d_parent = dget(parent);
963 dentry->d_sb = parent->d_sb;
964 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800965 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 }
967
968 spin_lock(&dcache_lock);
969 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800970 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 dentry_stat.nr_dentry++;
972 spin_unlock(&dcache_lock);
973
974 return dentry;
975}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700976EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
978struct dentry *d_alloc_name(struct dentry *parent, const char *name)
979{
980 struct qstr q;
981
982 q.name = name;
983 q.len = strlen(name);
984 q.hash = full_name_hash(q.name, q.len);
985 return d_alloc(parent, &q);
986}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -0400987EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
OGAWA Hirofumi360da902008-10-16 07:50:28 +0900989/* the caller must hold dcache_lock */
990static void __d_instantiate(struct dentry *dentry, struct inode *inode)
991{
992 if (inode)
993 list_add(&dentry->d_alias, &inode->i_dentry);
994 dentry->d_inode = inode;
995 fsnotify_d_instantiate(dentry, inode);
996}
997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998/**
999 * d_instantiate - fill in inode information for a dentry
1000 * @entry: dentry to complete
1001 * @inode: inode to attach to this dentry
1002 *
1003 * Fill in inode information in the entry.
1004 *
1005 * This turns negative dentries into productive full members
1006 * of society.
1007 *
1008 * NOTE! This assumes that the inode count has been incremented
1009 * (or otherwise set) by the caller to indicate that it is now
1010 * in use by the dcache.
1011 */
1012
1013void d_instantiate(struct dentry *entry, struct inode * inode)
1014{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001015 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 spin_lock(&dcache_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001017 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 spin_unlock(&dcache_lock);
1019 security_d_instantiate(entry, inode);
1020}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001021EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023/**
1024 * d_instantiate_unique - instantiate a non-aliased dentry
1025 * @entry: dentry to instantiate
1026 * @inode: inode to attach to this dentry
1027 *
1028 * Fill in inode information in the entry. On success, it returns NULL.
1029 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001030 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 *
1032 * Note that in order to avoid conflicts with rename() etc, the caller
1033 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001034 *
1035 * This also assumes that the inode count has been incremented
1036 * (or otherwise set) by the caller to indicate that it is now
1037 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 */
David Howells770bfad2006-08-22 20:06:07 -04001039static struct dentry *__d_instantiate_unique(struct dentry *entry,
1040 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
1042 struct dentry *alias;
1043 int len = entry->d_name.len;
1044 const char *name = entry->d_name.name;
1045 unsigned int hash = entry->d_name.hash;
1046
David Howells770bfad2006-08-22 20:06:07 -04001047 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001048 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001049 return NULL;
1050 }
1051
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1053 struct qstr *qstr = &alias->d_name;
1054
1055 if (qstr->hash != hash)
1056 continue;
1057 if (alias->d_parent != entry->d_parent)
1058 continue;
1059 if (qstr->len != len)
1060 continue;
1061 if (memcmp(qstr->name, name, len))
1062 continue;
1063 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 return alias;
1065 }
David Howells770bfad2006-08-22 20:06:07 -04001066
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001067 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 return NULL;
1069}
David Howells770bfad2006-08-22 20:06:07 -04001070
1071struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1072{
1073 struct dentry *result;
1074
1075 BUG_ON(!list_empty(&entry->d_alias));
1076
1077 spin_lock(&dcache_lock);
1078 result = __d_instantiate_unique(entry, inode);
1079 spin_unlock(&dcache_lock);
1080
1081 if (!result) {
1082 security_d_instantiate(entry, inode);
1083 return NULL;
1084 }
1085
1086 BUG_ON(!d_unhashed(result));
1087 iput(inode);
1088 return result;
1089}
1090
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091EXPORT_SYMBOL(d_instantiate_unique);
1092
1093/**
1094 * d_alloc_root - allocate root dentry
1095 * @root_inode: inode to allocate the root for
1096 *
1097 * Allocate a root ("/") dentry for the inode given. The inode is
1098 * instantiated and returned. %NULL is returned if there is insufficient
1099 * memory or the inode passed is %NULL.
1100 */
1101
1102struct dentry * d_alloc_root(struct inode * root_inode)
1103{
1104 struct dentry *res = NULL;
1105
1106 if (root_inode) {
1107 static const struct qstr name = { .name = "/", .len = 1 };
1108
1109 res = d_alloc(NULL, &name);
1110 if (res) {
1111 res->d_sb = root_inode->i_sb;
1112 res->d_parent = res;
1113 d_instantiate(res, root_inode);
1114 }
1115 }
1116 return res;
1117}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001118EXPORT_SYMBOL(d_alloc_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
1120static inline struct hlist_head *d_hash(struct dentry *parent,
1121 unsigned long hash)
1122{
1123 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1124 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1125 return dentry_hashtable + (hash & D_HASHMASK);
1126}
1127
1128/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001129 * d_obtain_alias - find or allocate a dentry for a given inode
1130 * @inode: inode to allocate the dentry for
1131 *
1132 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1133 * similar open by handle operations. The returned dentry may be anonymous,
1134 * or may have a full name (if the inode was already in the cache).
1135 *
1136 * When called on a directory inode, we must ensure that the inode only ever
1137 * has one dentry. If a dentry is found, that is returned instead of
1138 * allocating a new one.
1139 *
1140 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001141 * to the dentry. In case of an error the reference on the inode is released.
1142 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1143 * be passed in and will be the error will be propagate to the return value,
1144 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001145 */
1146struct dentry *d_obtain_alias(struct inode *inode)
1147{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001148 static const struct qstr anonstring = { .name = "" };
1149 struct dentry *tmp;
1150 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001151
1152 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001153 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001154 if (IS_ERR(inode))
1155 return ERR_CAST(inode);
1156
Christoph Hellwig9308a612008-08-11 15:49:12 +02001157 res = d_find_alias(inode);
1158 if (res)
1159 goto out_iput;
1160
1161 tmp = d_alloc(NULL, &anonstring);
1162 if (!tmp) {
1163 res = ERR_PTR(-ENOMEM);
1164 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001165 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001166 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1167
1168 spin_lock(&dcache_lock);
1169 res = __d_find_alias(inode, 0);
1170 if (res) {
1171 spin_unlock(&dcache_lock);
1172 dput(tmp);
1173 goto out_iput;
1174 }
1175
1176 /* attach a disconnected dentry */
1177 spin_lock(&tmp->d_lock);
1178 tmp->d_sb = inode->i_sb;
1179 tmp->d_inode = inode;
1180 tmp->d_flags |= DCACHE_DISCONNECTED;
1181 tmp->d_flags &= ~DCACHE_UNHASHED;
1182 list_add(&tmp->d_alias, &inode->i_dentry);
1183 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1184 spin_unlock(&tmp->d_lock);
1185
1186 spin_unlock(&dcache_lock);
1187 return tmp;
1188
1189 out_iput:
1190 iput(inode);
1191 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001192}
Benny Halevyadc48722009-02-27 14:02:59 -08001193EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
1195/**
1196 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1197 * @inode: the inode which may have a disconnected dentry
1198 * @dentry: a negative dentry which we want to point to the inode.
1199 *
1200 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1201 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1202 * and return it, else simply d_add the inode to the dentry and return NULL.
1203 *
1204 * This is needed in the lookup routine of any filesystem that is exportable
1205 * (via knfsd) so that we can build dcache paths to directories effectively.
1206 *
1207 * If a dentry was found and moved, then it is returned. Otherwise NULL
1208 * is returned. This matches the expected return value of ->lookup.
1209 *
1210 */
1211struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1212{
1213 struct dentry *new = NULL;
1214
NeilBrown21c0d8f2006-10-04 02:16:16 -07001215 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 spin_lock(&dcache_lock);
1217 new = __d_find_alias(inode, 1);
1218 if (new) {
1219 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1220 spin_unlock(&dcache_lock);
1221 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 d_move(new, dentry);
1223 iput(inode);
1224 } else {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001225 /* already taking dcache_lock, so d_add() by hand */
1226 __d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 spin_unlock(&dcache_lock);
1228 security_d_instantiate(dentry, inode);
1229 d_rehash(dentry);
1230 }
1231 } else
1232 d_add(dentry, inode);
1233 return new;
1234}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001235EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
Barry Naujok94035402008-05-21 16:50:46 +10001237/**
1238 * d_add_ci - lookup or allocate new dentry with case-exact name
1239 * @inode: the inode case-insensitive lookup has found
1240 * @dentry: the negative dentry that was passed to the parent's lookup func
1241 * @name: the case-exact name to be associated with the returned dentry
1242 *
1243 * This is to avoid filling the dcache with case-insensitive names to the
1244 * same inode, only the actual correct case is stored in the dcache for
1245 * case-insensitive filesystems.
1246 *
1247 * For a case-insensitive lookup match and if the the case-exact dentry
1248 * already exists in in the dcache, use it and return it.
1249 *
1250 * If no entry exists with the exact case name, allocate new dentry with
1251 * the exact case, and return the spliced entry.
1252 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001253struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001254 struct qstr *name)
1255{
1256 int error;
1257 struct dentry *found;
1258 struct dentry *new;
1259
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001260 /*
1261 * First check if a dentry matching the name already exists,
1262 * if not go ahead and create it now.
1263 */
Barry Naujok94035402008-05-21 16:50:46 +10001264 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001265 if (!found) {
1266 new = d_alloc(dentry->d_parent, name);
1267 if (!new) {
1268 error = -ENOMEM;
1269 goto err_out;
1270 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001271
Barry Naujok94035402008-05-21 16:50:46 +10001272 found = d_splice_alias(inode, new);
1273 if (found) {
1274 dput(new);
1275 return found;
1276 }
1277 return new;
1278 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001279
1280 /*
1281 * If a matching dentry exists, and it's not negative use it.
1282 *
1283 * Decrement the reference count to balance the iget() done
1284 * earlier on.
1285 */
Barry Naujok94035402008-05-21 16:50:46 +10001286 if (found->d_inode) {
1287 if (unlikely(found->d_inode != inode)) {
1288 /* This can't happen because bad inodes are unhashed. */
1289 BUG_ON(!is_bad_inode(inode));
1290 BUG_ON(!is_bad_inode(found->d_inode));
1291 }
Barry Naujok94035402008-05-21 16:50:46 +10001292 iput(inode);
1293 return found;
1294 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001295
Barry Naujok94035402008-05-21 16:50:46 +10001296 /*
1297 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001298 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001299 */
Barry Naujok94035402008-05-21 16:50:46 +10001300 spin_lock(&dcache_lock);
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001301 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001302 __d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001303 spin_unlock(&dcache_lock);
1304 security_d_instantiate(found, inode);
1305 return found;
1306 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001307
Barry Naujok94035402008-05-21 16:50:46 +10001308 /*
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001309 * In case a directory already has a (disconnected) entry grab a
1310 * reference to it, move it in place and use it.
Barry Naujok94035402008-05-21 16:50:46 +10001311 */
1312 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1313 dget_locked(new);
1314 spin_unlock(&dcache_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001315 security_d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001316 d_move(new, found);
Barry Naujok94035402008-05-21 16:50:46 +10001317 iput(inode);
Barry Naujok94035402008-05-21 16:50:46 +10001318 dput(found);
Barry Naujok94035402008-05-21 16:50:46 +10001319 return new;
1320
1321err_out:
1322 iput(inode);
1323 return ERR_PTR(error);
1324}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001325EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
1327/**
1328 * d_lookup - search for a dentry
1329 * @parent: parent dentry
1330 * @name: qstr of name we wish to find
1331 *
1332 * Searches the children of the parent dentry for the name in question. If
1333 * the dentry is found its reference count is incremented and the dentry
Zhaoleibe42c4c2008-12-01 14:34:58 -08001334 * is returned. The caller must use dput to free the entry when it has
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 * finished using it. %NULL is returned on failure.
1336 *
1337 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1338 * Memory barriers are used while updating and doing lockless traversal.
1339 * To avoid races with d_move while rename is happening, d_lock is used.
1340 *
1341 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1342 * and name pointer in one structure pointed by d_qstr.
1343 *
1344 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1345 * lookup is going on.
1346 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001347 * The dentry unused LRU is not updated even if lookup finds the required dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1349 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1350 * acquisition.
1351 *
1352 * d_lookup() is protected against the concurrent renames in some unrelated
1353 * directory using the seqlockt_t rename_lock.
1354 */
1355
1356struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1357{
1358 struct dentry * dentry = NULL;
1359 unsigned long seq;
1360
1361 do {
1362 seq = read_seqbegin(&rename_lock);
1363 dentry = __d_lookup(parent, name);
1364 if (dentry)
1365 break;
1366 } while (read_seqretry(&rename_lock, seq));
1367 return dentry;
1368}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001369EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
1371struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1372{
1373 unsigned int len = name->len;
1374 unsigned int hash = name->hash;
1375 const unsigned char *str = name->name;
1376 struct hlist_head *head = d_hash(parent,hash);
1377 struct dentry *found = NULL;
1378 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001379 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 rcu_read_lock();
1382
Paul E. McKenney665a7582005-11-07 00:59:17 -08001383 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 struct qstr *qstr;
1385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 if (dentry->d_name.hash != hash)
1387 continue;
1388 if (dentry->d_parent != parent)
1389 continue;
1390
1391 spin_lock(&dentry->d_lock);
1392
1393 /*
1394 * Recheck the dentry after taking the lock - d_move may have
1395 * changed things. Don't bother checking the hash because we're
1396 * about to compare the whole name anyway.
1397 */
1398 if (dentry->d_parent != parent)
1399 goto next;
1400
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001401 /* non-existing due to RCU? */
1402 if (d_unhashed(dentry))
1403 goto next;
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 /*
1406 * It is safe to compare names since d_move() cannot
1407 * change the qstr (protected by d_lock).
1408 */
1409 qstr = &dentry->d_name;
1410 if (parent->d_op && parent->d_op->d_compare) {
1411 if (parent->d_op->d_compare(parent, qstr, name))
1412 goto next;
1413 } else {
1414 if (qstr->len != len)
1415 goto next;
1416 if (memcmp(qstr->name, str, len))
1417 goto next;
1418 }
1419
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001420 atomic_inc(&dentry->d_count);
1421 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 spin_unlock(&dentry->d_lock);
1423 break;
1424next:
1425 spin_unlock(&dentry->d_lock);
1426 }
1427 rcu_read_unlock();
1428
1429 return found;
1430}
1431
1432/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001433 * d_hash_and_lookup - hash the qstr then search for a dentry
1434 * @dir: Directory to search in
1435 * @name: qstr of name we wish to find
1436 *
1437 * On hash failure or on lookup failure NULL is returned.
1438 */
1439struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1440{
1441 struct dentry *dentry = NULL;
1442
1443 /*
1444 * Check for a fs-specific hash function. Note that we must
1445 * calculate the standard hash first, as the d_op->d_hash()
1446 * routine may choose to leave the hash value unchanged.
1447 */
1448 name->hash = full_name_hash(name->name, name->len);
1449 if (dir->d_op && dir->d_op->d_hash) {
1450 if (dir->d_op->d_hash(dir, name) < 0)
1451 goto out;
1452 }
1453 dentry = d_lookup(dir, name);
1454out:
1455 return dentry;
1456}
1457
1458/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 * d_validate - verify dentry provided from insecure source
1460 * @dentry: The dentry alleged to be valid child of @dparent
1461 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 *
1463 * An insecure source has sent us a dentry, here we verify it and dget() it.
1464 * This is used by ncpfs in its readdir implementation.
1465 * Zero is returned in the dentry is invalid.
1466 */
1467
1468int d_validate(struct dentry *dentry, struct dentry *dparent)
1469{
1470 struct hlist_head *base;
1471 struct hlist_node *lhp;
1472
1473 /* Check whether the ptr might be valid at all.. */
1474 if (!kmem_ptr_validate(dentry_cache, dentry))
1475 goto out;
1476
1477 if (dentry->d_parent != dparent)
1478 goto out;
1479
1480 spin_lock(&dcache_lock);
1481 base = d_hash(dparent, dentry->d_name.hash);
1482 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001483 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 * as it is parsed under dcache_lock
1485 */
1486 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1487 __dget_locked(dentry);
1488 spin_unlock(&dcache_lock);
1489 return 1;
1490 }
1491 }
1492 spin_unlock(&dcache_lock);
1493out:
1494 return 0;
1495}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001496EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
1498/*
1499 * When a file is deleted, we have two options:
1500 * - turn this dentry into a negative dentry
1501 * - unhash this dentry and free it.
1502 *
1503 * Usually, we want to just turn this into
1504 * a negative dentry, but if anybody else is
1505 * currently using the dentry or the inode
1506 * we can't do that and we fall back on removing
1507 * it from the hash queues and waiting for
1508 * it to be deleted later when it has no users
1509 */
1510
1511/**
1512 * d_delete - delete a dentry
1513 * @dentry: The dentry to delete
1514 *
1515 * Turn the dentry into a negative dentry if possible, otherwise
1516 * remove it from the hash queues so it can be deleted later
1517 */
1518
1519void d_delete(struct dentry * dentry)
1520{
John McCutchan7a91bf72005-08-08 13:52:16 -04001521 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 /*
1523 * Are we the only user?
1524 */
1525 spin_lock(&dcache_lock);
1526 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001527 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 if (atomic_read(&dentry->d_count) == 1) {
Al Viro13e3c5e2010-05-21 16:11:04 -04001529 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001531 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 return;
1533 }
1534
1535 if (!d_unhashed(dentry))
1536 __d_drop(dentry);
1537
1538 spin_unlock(&dentry->d_lock);
1539 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001540
1541 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001543EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1546{
1547
1548 entry->d_flags &= ~DCACHE_UNHASHED;
1549 hlist_add_head_rcu(&entry->d_hash, list);
1550}
1551
David Howells770bfad2006-08-22 20:06:07 -04001552static void _d_rehash(struct dentry * entry)
1553{
1554 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1555}
1556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557/**
1558 * d_rehash - add an entry back to the hash
1559 * @entry: dentry to add to the hash
1560 *
1561 * Adds a dentry to the hash according to its name.
1562 */
1563
1564void d_rehash(struct dentry * entry)
1565{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 spin_lock(&dcache_lock);
1567 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001568 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 spin_unlock(&entry->d_lock);
1570 spin_unlock(&dcache_lock);
1571}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001572EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574/*
1575 * When switching names, the actual string doesn't strictly have to
1576 * be preserved in the target - because we're dropping the target
1577 * anyway. As such, we can just do a simple memcpy() to copy over
1578 * the new name before we switch.
1579 *
1580 * Note that we have to be a lot more careful about getting the hash
1581 * switched - we have to switch the hash value properly even if it
1582 * then no longer matches the actual (corrupted) string of the target.
1583 * The hash value has to match the hash queue that the dentry is on..
1584 */
1585static void switch_names(struct dentry *dentry, struct dentry *target)
1586{
1587 if (dname_external(target)) {
1588 if (dname_external(dentry)) {
1589 /*
1590 * Both external: swap the pointers
1591 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001592 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 } else {
1594 /*
1595 * dentry:internal, target:external. Steal target's
1596 * storage and make target internal.
1597 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07001598 memcpy(target->d_iname, dentry->d_name.name,
1599 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 dentry->d_name.name = target->d_name.name;
1601 target->d_name.name = target->d_iname;
1602 }
1603 } else {
1604 if (dname_external(dentry)) {
1605 /*
1606 * dentry:external, target:internal. Give dentry's
1607 * storage to target and make dentry internal
1608 */
1609 memcpy(dentry->d_iname, target->d_name.name,
1610 target->d_name.len + 1);
1611 target->d_name.name = dentry->d_name.name;
1612 dentry->d_name.name = dentry->d_iname;
1613 } else {
1614 /*
1615 * Both are internal. Just copy target to dentry
1616 */
1617 memcpy(dentry->d_iname, target->d_name.name,
1618 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05001619 dentry->d_name.len = target->d_name.len;
1620 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 }
1622 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001623 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624}
1625
1626/*
1627 * We cannibalize "target" when moving dentry on top of it,
1628 * because it's going to be thrown away anyway. We could be more
1629 * polite about it, though.
1630 *
1631 * This forceful removal will result in ugly /proc output if
1632 * somebody holds a file open that got deleted due to a rename.
1633 * We could be nicer about the deleted file, and let it show
J. Bruce Fieldsbc154b12007-10-16 23:29:42 -07001634 * up under the name it had before it was deleted rather than
1635 * under the original name of the file that was moved on top of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 */
1637
Trond Myklebust9eaef272006-10-21 10:24:20 -07001638/*
1639 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 * @dentry: entry to move
1641 * @target: new dentry
1642 *
1643 * Update the dcache to reflect the move of a file name. Negative
1644 * dcache entries should not be moved in this way.
1645 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001646static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647{
1648 struct hlist_head *list;
1649
1650 if (!dentry->d_inode)
1651 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1652
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 write_seqlock(&rename_lock);
1654 /*
1655 * XXXX: do we really need to take target->d_lock?
1656 */
1657 if (target < dentry) {
1658 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001659 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 } else {
1661 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001662 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 }
1664
1665 /* Move the dentry to the target hash queue, if on different bucket */
Denis Chengf77e3492007-10-16 23:30:11 -07001666 if (d_unhashed(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 goto already_unhashed;
1668
1669 hlist_del_rcu(&dentry->d_hash);
1670
1671already_unhashed:
1672 list = d_hash(target->d_parent, target->d_name.hash);
1673 __d_rehash(dentry, list);
1674
1675 /* Unhash the target: dput() will then get rid of it */
1676 __d_drop(target);
1677
Eric Dumazet5160ee62006-01-08 01:03:32 -08001678 list_del(&dentry->d_u.d_child);
1679 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
1681 /* Switch the names.. */
1682 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001683 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
1685 /* ... and switch the parents */
1686 if (IS_ROOT(dentry)) {
1687 dentry->d_parent = target->d_parent;
1688 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001689 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001691 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
1693 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001694 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 }
1696
Eric Dumazet5160ee62006-01-08 01:03:32 -08001697 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001699 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 spin_unlock(&dentry->d_lock);
1701 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001702}
1703
1704/**
1705 * d_move - move a dentry
1706 * @dentry: entry to move
1707 * @target: new dentry
1708 *
1709 * Update the dcache to reflect the move of a file name. Negative
1710 * dcache entries should not be moved in this way.
1711 */
1712
1713void d_move(struct dentry * dentry, struct dentry * target)
1714{
1715 spin_lock(&dcache_lock);
1716 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 spin_unlock(&dcache_lock);
1718}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001719EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001721/**
1722 * d_ancestor - search for an ancestor
1723 * @p1: ancestor dentry
1724 * @p2: child dentry
1725 *
1726 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1727 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07001728 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001729struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001730{
1731 struct dentry *p;
1732
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001733 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07001734 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001735 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001736 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001737 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001738}
1739
1740/*
1741 * This helper attempts to cope with remotely renamed directories
1742 *
1743 * It assumes that the caller is already holding
1744 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1745 *
1746 * Note: If ever the locking in lock_rename() changes, then please
1747 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07001748 */
1749static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001750 __releases(dcache_lock)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001751{
1752 struct mutex *m1 = NULL, *m2 = NULL;
1753 struct dentry *ret;
1754
1755 /* If alias and dentry share a parent, then no extra locks required */
1756 if (alias->d_parent == dentry->d_parent)
1757 goto out_unalias;
1758
1759 /* Check for loops */
1760 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001761 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07001762 goto out_err;
1763
1764 /* See lock_rename() */
1765 ret = ERR_PTR(-EBUSY);
1766 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1767 goto out_err;
1768 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1769 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1770 goto out_err;
1771 m2 = &alias->d_parent->d_inode->i_mutex;
1772out_unalias:
1773 d_move_locked(alias, dentry);
1774 ret = alias;
1775out_err:
1776 spin_unlock(&dcache_lock);
1777 if (m2)
1778 mutex_unlock(m2);
1779 if (m1)
1780 mutex_unlock(m1);
1781 return ret;
1782}
1783
1784/*
David Howells770bfad2006-08-22 20:06:07 -04001785 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1786 * named dentry in place of the dentry to be replaced.
1787 */
1788static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1789{
1790 struct dentry *dparent, *aparent;
1791
1792 switch_names(dentry, anon);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001793 swap(dentry->d_name.hash, anon->d_name.hash);
David Howells770bfad2006-08-22 20:06:07 -04001794
1795 dparent = dentry->d_parent;
1796 aparent = anon->d_parent;
1797
1798 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1799 list_del(&dentry->d_u.d_child);
1800 if (!IS_ROOT(dentry))
1801 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1802 else
1803 INIT_LIST_HEAD(&dentry->d_u.d_child);
1804
1805 anon->d_parent = (dparent == dentry) ? anon : dparent;
1806 list_del(&anon->d_u.d_child);
1807 if (!IS_ROOT(anon))
1808 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1809 else
1810 INIT_LIST_HEAD(&anon->d_u.d_child);
1811
1812 anon->d_flags &= ~DCACHE_DISCONNECTED;
1813}
1814
1815/**
1816 * d_materialise_unique - introduce an inode into the tree
1817 * @dentry: candidate dentry
1818 * @inode: inode to bind to the dentry, to which aliases may be attached
1819 *
1820 * Introduces an dentry into the tree, substituting an extant disconnected
1821 * root directory alias in its place if there is one
1822 */
1823struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1824{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001825 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001826
1827 BUG_ON(!d_unhashed(dentry));
1828
1829 spin_lock(&dcache_lock);
1830
1831 if (!inode) {
1832 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001833 __d_instantiate(dentry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001834 goto found_lock;
1835 }
1836
Trond Myklebust9eaef272006-10-21 10:24:20 -07001837 if (S_ISDIR(inode->i_mode)) {
1838 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001839
Trond Myklebust9eaef272006-10-21 10:24:20 -07001840 /* Does an aliased dentry already exist? */
1841 alias = __d_find_alias(inode, 0);
1842 if (alias) {
1843 actual = alias;
1844 /* Is this an anonymous mountpoint that we could splice
1845 * into our tree? */
1846 if (IS_ROOT(alias)) {
1847 spin_lock(&alias->d_lock);
1848 __d_materialise_dentry(dentry, alias);
1849 __d_drop(alias);
1850 goto found;
1851 }
1852 /* Nope, but we must(!) avoid directory aliasing */
1853 actual = __d_unalias(dentry, alias);
1854 if (IS_ERR(actual))
1855 dput(alias);
1856 goto out_nolock;
1857 }
David Howells770bfad2006-08-22 20:06:07 -04001858 }
1859
1860 /* Add a unique reference */
1861 actual = __d_instantiate_unique(dentry, inode);
1862 if (!actual)
1863 actual = dentry;
1864 else if (unlikely(!d_unhashed(actual)))
1865 goto shouldnt_be_hashed;
1866
1867found_lock:
1868 spin_lock(&actual->d_lock);
1869found:
1870 _d_rehash(actual);
1871 spin_unlock(&actual->d_lock);
1872 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001873out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001874 if (actual == dentry) {
1875 security_d_instantiate(dentry, inode);
1876 return NULL;
1877 }
1878
1879 iput(inode);
1880 return actual;
1881
David Howells770bfad2006-08-22 20:06:07 -04001882shouldnt_be_hashed:
1883 spin_unlock(&dcache_lock);
1884 BUG();
David Howells770bfad2006-08-22 20:06:07 -04001885}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001886EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04001887
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001888static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01001889{
1890 *buflen -= namelen;
1891 if (*buflen < 0)
1892 return -ENAMETOOLONG;
1893 *buffer -= namelen;
1894 memcpy(*buffer, str, namelen);
1895 return 0;
1896}
1897
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001898static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1899{
1900 return prepend(buffer, buflen, name->name, name->len);
1901}
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001904 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001905 * @path: the dentry/vfsmount to report
1906 * @root: root vfsmnt/dentry (may be modified by this function)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 * @buffer: buffer to return value in
1908 * @buflen: buffer length
1909 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001910 * Convert a dentry into an ASCII path name. If the entry has been deleted
1911 * the string " (deleted)" is appended. Note that this is ambiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08001913 * Returns a pointer into the buffer or an error code if the
1914 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08001915 *
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);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001931 if (d_unlinked(dentry) &&
Ram Pai6092d042008-03-27 13:06:20 +01001932 (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 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08001990 * Returns a pointer into the buffer or an error code if the path was
1991 * too long. Note: Callers should use the returned pointer, not the passed
1992 * in buffer, to use the name! The implementation often starts at an offset
1993 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08001994 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001995 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08001996 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07001997char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
1999 char *res;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002000 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002001 struct path tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002003 /*
2004 * We have various synthetic filesystems that never get mounted. On
2005 * these filesystems dentries are never used for lookup purposes, and
2006 * thus don't need to be hashed. They also don't need a name until a
2007 * user wants to identify the object in /proc/pid/fd/. The little hack
2008 * below allows us to generate a name for these objects on demand:
2009 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002010 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2011 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002014 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002015 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 read_unlock(&current->fs->lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002017 spin_lock(&dcache_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002018 tmp = root;
2019 res = __d_path(path, &tmp, buf, buflen);
Linus Torvalds552ce542007-02-13 12:08:18 -08002020 spin_unlock(&dcache_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002021 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 return res;
2023}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002024EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
2026/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002027 * Helper function for dentry_operations.d_dname() members
2028 */
2029char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2030 const char *fmt, ...)
2031{
2032 va_list args;
2033 char temp[64];
2034 int sz;
2035
2036 va_start(args, fmt);
2037 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2038 va_end(args);
2039
2040 if (sz > sizeof(temp) || sz > buflen)
2041 return ERR_PTR(-ENAMETOOLONG);
2042
2043 buffer += buflen - sz;
2044 return memcpy(buffer, temp, sz);
2045}
2046
2047/*
Ram Pai6092d042008-03-27 13:06:20 +01002048 * Write full pathname from the root of the filesystem into the buffer.
2049 */
2050char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2051{
2052 char *end = buf + buflen;
2053 char *retval;
2054
2055 spin_lock(&dcache_lock);
2056 prepend(&end, &buflen, "\0", 1);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002057 if (d_unlinked(dentry) &&
Ram Pai6092d042008-03-27 13:06:20 +01002058 (prepend(&end, &buflen, "//deleted", 9) != 0))
2059 goto Elong;
2060 if (buflen < 1)
2061 goto Elong;
2062 /* Get '/' right */
2063 retval = end-1;
2064 *retval = '/';
2065
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002066 while (!IS_ROOT(dentry)) {
2067 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01002068
Ram Pai6092d042008-03-27 13:06:20 +01002069 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002070 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01002071 (prepend(&end, &buflen, "/", 1) != 0))
2072 goto Elong;
2073
2074 retval = end;
2075 dentry = parent;
2076 }
2077 spin_unlock(&dcache_lock);
2078 return retval;
2079Elong:
2080 spin_unlock(&dcache_lock);
2081 return ERR_PTR(-ENAMETOOLONG);
2082}
2083
2084/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 * NOTE! The user-level library version returns a
2086 * character pointer. The kernel system call just
2087 * returns the length of the buffer filled (which
2088 * includes the ending '\0' character), or a negative
2089 * error value. So libc would do something like
2090 *
2091 * char *getcwd(char * buf, size_t size)
2092 * {
2093 * int retval;
2094 *
2095 * retval = sys_getcwd(buf, size);
2096 * if (retval >= 0)
2097 * return buf;
2098 * errno = -retval;
2099 * return NULL;
2100 * }
2101 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002102SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
Linus Torvalds552ce542007-02-13 12:08:18 -08002104 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002105 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002106 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107
2108 if (!page)
2109 return -ENOMEM;
2110
2111 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002112 pwd = current->fs->pwd;
Ram Pai6092d042008-03-27 13:06:20 +01002113 path_get(&pwd);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002114 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002115 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 read_unlock(&current->fs->lock);
2117
Linus Torvalds552ce542007-02-13 12:08:18 -08002118 error = -ENOENT;
Linus Torvalds552ce542007-02-13 12:08:18 -08002119 spin_lock(&dcache_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002120 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002121 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002122 struct path tmp = root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002123 char * cwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002125 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
Linus Torvalds552ce542007-02-13 12:08:18 -08002126 spin_unlock(&dcache_lock);
2127
2128 error = PTR_ERR(cwd);
2129 if (IS_ERR(cwd))
2130 goto out;
2131
2132 error = -ERANGE;
2133 len = PAGE_SIZE + page - cwd;
2134 if (len <= size) {
2135 error = len;
2136 if (copy_to_user(buf, cwd, len))
2137 error = -EFAULT;
2138 }
2139 } else
2140 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141
2142out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002143 path_put(&pwd);
2144 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 free_page((unsigned long) page);
2146 return error;
2147}
2148
2149/*
2150 * Test whether new_dentry is a subdirectory of old_dentry.
2151 *
2152 * Trivially implemented using the dcache structure
2153 */
2154
2155/**
2156 * is_subdir - is new dentry a subdirectory of old_dentry
2157 * @new_dentry: new dentry
2158 * @old_dentry: old dentry
2159 *
2160 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2161 * Returns 0 otherwise.
2162 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2163 */
2164
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002165int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166{
2167 int result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 unsigned long seq;
2169
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002170 if (new_dentry == old_dentry)
2171 return 1;
2172
2173 /*
2174 * Need rcu_readlock to protect against the d_parent trashing
2175 * due to d_move
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 */
2177 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002178 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 seq = read_seqbegin(&rename_lock);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002181 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002183 else
2184 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 } while (read_seqretry(&rename_lock, seq));
2186 rcu_read_unlock();
2187
2188 return result;
2189}
2190
Al Viro2096f752010-01-30 13:16:21 -05002191int path_is_under(struct path *path1, struct path *path2)
2192{
2193 struct vfsmount *mnt = path1->mnt;
2194 struct dentry *dentry = path1->dentry;
2195 int res;
2196 spin_lock(&vfsmount_lock);
2197 if (mnt != path2->mnt) {
2198 for (;;) {
2199 if (mnt->mnt_parent == mnt) {
2200 spin_unlock(&vfsmount_lock);
2201 return 0;
2202 }
2203 if (mnt->mnt_parent == path2->mnt)
2204 break;
2205 mnt = mnt->mnt_parent;
2206 }
2207 dentry = mnt->mnt_mountpoint;
2208 }
2209 res = is_subdir(dentry, path2->dentry);
2210 spin_unlock(&vfsmount_lock);
2211 return res;
2212}
2213EXPORT_SYMBOL(path_is_under);
2214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215void d_genocide(struct dentry *root)
2216{
2217 struct dentry *this_parent = root;
2218 struct list_head *next;
2219
2220 spin_lock(&dcache_lock);
2221repeat:
2222 next = this_parent->d_subdirs.next;
2223resume:
2224 while (next != &this_parent->d_subdirs) {
2225 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002226 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 next = tmp->next;
2228 if (d_unhashed(dentry)||!dentry->d_inode)
2229 continue;
2230 if (!list_empty(&dentry->d_subdirs)) {
2231 this_parent = dentry;
2232 goto repeat;
2233 }
2234 atomic_dec(&dentry->d_count);
2235 }
2236 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002237 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 atomic_dec(&this_parent->d_count);
2239 this_parent = this_parent->d_parent;
2240 goto resume;
2241 }
2242 spin_unlock(&dcache_lock);
2243}
2244
2245/**
2246 * find_inode_number - check for dentry with name
2247 * @dir: directory to check
2248 * @name: Name to find.
2249 *
2250 * Check whether a dentry already exists for the given name,
2251 * and return the inode number if it has an inode. Otherwise
2252 * 0 is returned.
2253 *
2254 * This routine is used to post-process directory listings for
2255 * filesystems using synthetic inode numbers, and is necessary
2256 * to keep getcwd() working.
2257 */
2258
2259ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2260{
2261 struct dentry * dentry;
2262 ino_t ino = 0;
2263
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002264 dentry = d_hash_and_lookup(dir, name);
2265 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 if (dentry->d_inode)
2267 ino = dentry->d_inode->i_ino;
2268 dput(dentry);
2269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 return ino;
2271}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002272EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
2274static __initdata unsigned long dhash_entries;
2275static int __init set_dhash_entries(char *str)
2276{
2277 if (!str)
2278 return 0;
2279 dhash_entries = simple_strtoul(str, &str, 0);
2280 return 1;
2281}
2282__setup("dhash_entries=", set_dhash_entries);
2283
2284static void __init dcache_init_early(void)
2285{
2286 int loop;
2287
2288 /* If hashes are distributed across NUMA nodes, defer
2289 * hash allocation until vmalloc space is available.
2290 */
2291 if (hashdist)
2292 return;
2293
2294 dentry_hashtable =
2295 alloc_large_system_hash("Dentry cache",
2296 sizeof(struct hlist_head),
2297 dhash_entries,
2298 13,
2299 HASH_EARLY,
2300 &d_hash_shift,
2301 &d_hash_mask,
2302 0);
2303
2304 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2305 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2306}
2307
Denis Cheng74bf17c2007-10-16 23:26:30 -07002308static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309{
2310 int loop;
2311
2312 /*
2313 * A constructor could be added for stable state like the lists,
2314 * but it is probably not worth it because of the cache nature
2315 * of the dcache.
2316 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002317 dentry_cache = KMEM_CACHE(dentry,
2318 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
Rusty Russell8e1f9362007-07-17 04:03:17 -07002320 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
2322 /* Hash may have been set up in dcache_init_early */
2323 if (!hashdist)
2324 return;
2325
2326 dentry_hashtable =
2327 alloc_large_system_hash("Dentry cache",
2328 sizeof(struct hlist_head),
2329 dhash_entries,
2330 13,
2331 0,
2332 &d_hash_shift,
2333 &d_hash_mask,
2334 0);
2335
2336 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2337 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2338}
2339
2340/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002341struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002342EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344EXPORT_SYMBOL(d_genocide);
2345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346void __init vfs_caches_init_early(void)
2347{
2348 dcache_init_early();
2349 inode_init_early();
2350}
2351
2352void __init vfs_caches_init(unsigned long mempages)
2353{
2354 unsigned long reserve;
2355
2356 /* Base hash sizes on available memory, with a reserve equal to
2357 150% of current kernel size */
2358
2359 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2360 mempages -= reserve;
2361
2362 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002363 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364
Denis Cheng74bf17c2007-10-16 23:26:30 -07002365 dcache_init();
2366 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002368 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 bdev_cache_init();
2370 chrdev_init();
2371}