blob: d09d93819b4d45413bb7f1e73669fba4f8225c5c [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 Virodca33252010-07-25 02:31:46 +0400539 struct super_block *sb, *p = NULL;
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 Virodca33252010-07-25 02:31:46 +0400553 list_for_each_entry(sb, &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);
Al Virodca33252010-07-25 02:31:46 +0400593 if (p)
594 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700595 count -= pruned;
Al Virodca33252010-07-25 02:31:46 +0400596 p = sb;
Al Viro79893c12010-03-22 20:27:55 -0400597 /* more work left to do? */
598 if (count <= 0)
599 break;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700600 }
Al Virodca33252010-07-25 02:31:46 +0400601 if (p)
602 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700603 spin_unlock(&sb_lock);
604 spin_unlock(&dcache_lock);
605}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607/**
608 * shrink_dcache_sb - shrink dcache for a superblock
609 * @sb: superblock
610 *
611 * Shrink the dcache for the specified super block. This
612 * is used to free the dcache before unmounting a file
613 * system
614 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615void shrink_dcache_sb(struct super_block * sb)
616{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700617 __shrink_dcache_sb(sb, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700619EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621/*
David Howellsc636ebd2006-10-11 01:22:19 -0700622 * destroy a single subtree of dentries for unmount
623 * - see the comments on shrink_dcache_for_umount() for a description of the
624 * locking
625 */
626static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
627{
628 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700629 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700630
631 BUG_ON(!IS_ROOT(dentry));
632
633 /* detach this root from the system */
634 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700635 dentry_lru_del_init(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700636 __d_drop(dentry);
637 spin_unlock(&dcache_lock);
638
639 for (;;) {
640 /* descend to the first leaf in the current subtree */
641 while (!list_empty(&dentry->d_subdirs)) {
642 struct dentry *loop;
643
644 /* this is a branch with children - detach all of them
645 * from the system in one go */
646 spin_lock(&dcache_lock);
647 list_for_each_entry(loop, &dentry->d_subdirs,
648 d_u.d_child) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700649 dentry_lru_del_init(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700650 __d_drop(loop);
651 cond_resched_lock(&dcache_lock);
652 }
653 spin_unlock(&dcache_lock);
654
655 /* move to the first child */
656 dentry = list_entry(dentry->d_subdirs.next,
657 struct dentry, d_u.d_child);
658 }
659
660 /* consume the dentries from this leaf up through its parents
661 * until we find one with children or run out altogether */
662 do {
663 struct inode *inode;
664
665 if (atomic_read(&dentry->d_count) != 0) {
666 printk(KERN_ERR
667 "BUG: Dentry %p{i=%lx,n=%s}"
668 " still in use (%d)"
669 " [unmount of %s %s]\n",
670 dentry,
671 dentry->d_inode ?
672 dentry->d_inode->i_ino : 0UL,
673 dentry->d_name.name,
674 atomic_read(&dentry->d_count),
675 dentry->d_sb->s_type->name,
676 dentry->d_sb->s_id);
677 BUG();
678 }
679
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900680 if (IS_ROOT(dentry))
David Howellsc636ebd2006-10-11 01:22:19 -0700681 parent = NULL;
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900682 else {
683 parent = dentry->d_parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700684 atomic_dec(&parent->d_count);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900685 }
David Howellsc636ebd2006-10-11 01:22:19 -0700686
687 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700688 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700689
690 inode = dentry->d_inode;
691 if (inode) {
692 dentry->d_inode = NULL;
693 list_del_init(&dentry->d_alias);
694 if (dentry->d_op && dentry->d_op->d_iput)
695 dentry->d_op->d_iput(dentry, inode);
696 else
697 iput(inode);
698 }
699
700 d_free(dentry);
701
702 /* finished when we fall off the top of the tree,
703 * otherwise we ascend to the parent and move to the
704 * next sibling if there is one */
705 if (!parent)
David Howellsf8713572006-10-28 10:38:46 -0700706 goto out;
David Howellsc636ebd2006-10-11 01:22:19 -0700707
708 dentry = parent;
709
710 } while (list_empty(&dentry->d_subdirs));
711
712 dentry = list_entry(dentry->d_subdirs.next,
713 struct dentry, d_u.d_child);
714 }
David Howellsf8713572006-10-28 10:38:46 -0700715out:
716 /* several dentries were freed, need to correct nr_dentry */
717 spin_lock(&dcache_lock);
718 dentry_stat.nr_dentry -= detached;
719 spin_unlock(&dcache_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700720}
721
722/*
723 * destroy the dentries attached to a superblock on unmounting
724 * - we don't need to use dentry->d_lock, and only need dcache_lock when
725 * removing the dentry from the system lists and hashes because:
726 * - the superblock is detached from all mountings and open files, so the
727 * dentry trees will not be rearranged by the VFS
728 * - s_umount is write-locked, so the memory pressure shrinker will ignore
729 * any dentries belonging to this superblock that it comes across
730 * - the filesystem itself is no longer permitted to rearrange the dentries
731 * in this superblock
732 */
733void shrink_dcache_for_umount(struct super_block *sb)
734{
735 struct dentry *dentry;
736
737 if (down_read_trylock(&sb->s_umount))
738 BUG();
739
740 dentry = sb->s_root;
741 sb->s_root = NULL;
742 atomic_dec(&dentry->d_count);
743 shrink_dcache_for_umount_subtree(dentry);
744
745 while (!hlist_empty(&sb->s_anon)) {
746 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
747 shrink_dcache_for_umount_subtree(dentry);
748 }
749}
750
751/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 * Search for at least 1 mount point in the dentry's subdirs.
753 * We descend to the next level whenever the d_subdirs
754 * list is non-empty and continue searching.
755 */
756
757/**
758 * have_submounts - check for mounts over a dentry
759 * @parent: dentry to check.
760 *
761 * Return true if the parent or its subdirectories contain
762 * a mount point
763 */
764
765int have_submounts(struct dentry *parent)
766{
767 struct dentry *this_parent = parent;
768 struct list_head *next;
769
770 spin_lock(&dcache_lock);
771 if (d_mountpoint(parent))
772 goto positive;
773repeat:
774 next = this_parent->d_subdirs.next;
775resume:
776 while (next != &this_parent->d_subdirs) {
777 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800778 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 next = tmp->next;
780 /* Have we found a mount point ? */
781 if (d_mountpoint(dentry))
782 goto positive;
783 if (!list_empty(&dentry->d_subdirs)) {
784 this_parent = dentry;
785 goto repeat;
786 }
787 }
788 /*
789 * All done at this level ... ascend and resume the search.
790 */
791 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800792 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 this_parent = this_parent->d_parent;
794 goto resume;
795 }
796 spin_unlock(&dcache_lock);
797 return 0; /* No mount points found in tree */
798positive:
799 spin_unlock(&dcache_lock);
800 return 1;
801}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700802EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804/*
805 * Search the dentry child list for the specified parent,
806 * and move any unused dentries to the end of the unused
807 * list for prune_dcache(). We descend to the next level
808 * whenever the d_subdirs list is non-empty and continue
809 * searching.
810 *
811 * It returns zero iff there are no unused children,
812 * otherwise it returns the number of children moved to
813 * the end of the unused list. This may not be the total
814 * number of unused children, because select_parent can
815 * drop the lock and return early due to latency
816 * constraints.
817 */
818static int select_parent(struct dentry * parent)
819{
820 struct dentry *this_parent = parent;
821 struct list_head *next;
822 int found = 0;
823
824 spin_lock(&dcache_lock);
825repeat:
826 next = this_parent->d_subdirs.next;
827resume:
828 while (next != &this_parent->d_subdirs) {
829 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800830 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 next = tmp->next;
832
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700833 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 /*
835 * move only zero ref count dentries to the end
836 * of the unused list for prune_dcache
837 */
838 if (!atomic_read(&dentry->d_count)) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700839 dentry_lru_add_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 found++;
841 }
842
843 /*
844 * We can return to the caller if we have found some (this
845 * ensures forward progress). We'll be coming back to find
846 * the rest.
847 */
848 if (found && need_resched())
849 goto out;
850
851 /*
852 * Descend a level if the d_subdirs list is non-empty.
853 */
854 if (!list_empty(&dentry->d_subdirs)) {
855 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 goto repeat;
857 }
858 }
859 /*
860 * All done at this level ... ascend and resume the search.
861 */
862 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800863 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 goto resume;
866 }
867out:
868 spin_unlock(&dcache_lock);
869 return found;
870}
871
872/**
873 * shrink_dcache_parent - prune dcache
874 * @parent: parent of entries to prune
875 *
876 * Prune the dcache to remove unused children of the parent dentry.
877 */
878
879void shrink_dcache_parent(struct dentry * parent)
880{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700881 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 int found;
883
884 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700885 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700887EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889/*
890 * Scan `nr' dentries and return the number which remain.
891 *
892 * We need to avoid reentering the filesystem if the caller is performing a
893 * GFP_NOFS allocation attempt. One example deadlock is:
894 *
895 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
896 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
897 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
898 *
899 * In this case we return -1 to tell the caller that we baled.
900 */
Dave Chinner7f8275d2010-07-19 14:56:17 +1000901static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
903 if (nr) {
904 if (!(gfp_mask & __GFP_FS))
905 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700906 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 }
908 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
909}
910
Rusty Russell8e1f9362007-07-17 04:03:17 -0700911static struct shrinker dcache_shrinker = {
912 .shrink = shrink_dcache_memory,
913 .seeks = DEFAULT_SEEKS,
914};
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916/**
917 * d_alloc - allocate a dcache entry
918 * @parent: parent of entry to allocate
919 * @name: qstr of the name
920 *
921 * Allocates a dentry. It returns %NULL if there is insufficient memory
922 * available. On a success the dentry is returned. The name passed in is
923 * copied and the copy passed in may be reused after this call.
924 */
925
926struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
927{
928 struct dentry *dentry;
929 char *dname;
930
Mel Gormane12ba742007-10-16 01:25:52 -0700931 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 if (!dentry)
933 return NULL;
934
935 if (name->len > DNAME_INLINE_LEN-1) {
936 dname = kmalloc(name->len + 1, GFP_KERNEL);
937 if (!dname) {
938 kmem_cache_free(dentry_cache, dentry);
939 return NULL;
940 }
941 } else {
942 dname = dentry->d_iname;
943 }
944 dentry->d_name.name = dname;
945
946 dentry->d_name.len = name->len;
947 dentry->d_name.hash = name->hash;
948 memcpy(dname, name->name, name->len);
949 dname[name->len] = 0;
950
951 atomic_set(&dentry->d_count, 1);
952 dentry->d_flags = DCACHE_UNHASHED;
953 spin_lock_init(&dentry->d_lock);
954 dentry->d_inode = NULL;
955 dentry->d_parent = NULL;
956 dentry->d_sb = NULL;
957 dentry->d_op = NULL;
958 dentry->d_fsdata = NULL;
959 dentry->d_mounted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 INIT_HLIST_NODE(&dentry->d_hash);
961 INIT_LIST_HEAD(&dentry->d_lru);
962 INIT_LIST_HEAD(&dentry->d_subdirs);
963 INIT_LIST_HEAD(&dentry->d_alias);
964
965 if (parent) {
966 dentry->d_parent = dget(parent);
967 dentry->d_sb = parent->d_sb;
968 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800969 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 }
971
972 spin_lock(&dcache_lock);
973 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800974 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 dentry_stat.nr_dentry++;
976 spin_unlock(&dcache_lock);
977
978 return dentry;
979}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700980EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981
982struct dentry *d_alloc_name(struct dentry *parent, const char *name)
983{
984 struct qstr q;
985
986 q.name = name;
987 q.len = strlen(name);
988 q.hash = full_name_hash(q.name, q.len);
989 return d_alloc(parent, &q);
990}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -0400991EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
OGAWA Hirofumi360da902008-10-16 07:50:28 +0900993/* the caller must hold dcache_lock */
994static void __d_instantiate(struct dentry *dentry, struct inode *inode)
995{
996 if (inode)
997 list_add(&dentry->d_alias, &inode->i_dentry);
998 dentry->d_inode = inode;
999 fsnotify_d_instantiate(dentry, inode);
1000}
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002/**
1003 * d_instantiate - fill in inode information for a dentry
1004 * @entry: dentry to complete
1005 * @inode: inode to attach to this dentry
1006 *
1007 * Fill in inode information in the entry.
1008 *
1009 * This turns negative dentries into productive full members
1010 * of society.
1011 *
1012 * NOTE! This assumes that the inode count has been incremented
1013 * (or otherwise set) by the caller to indicate that it is now
1014 * in use by the dcache.
1015 */
1016
1017void d_instantiate(struct dentry *entry, struct inode * inode)
1018{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001019 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 spin_lock(&dcache_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001021 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 spin_unlock(&dcache_lock);
1023 security_d_instantiate(entry, inode);
1024}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001025EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027/**
1028 * d_instantiate_unique - instantiate a non-aliased dentry
1029 * @entry: dentry to instantiate
1030 * @inode: inode to attach to this dentry
1031 *
1032 * Fill in inode information in the entry. On success, it returns NULL.
1033 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001034 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 *
1036 * Note that in order to avoid conflicts with rename() etc, the caller
1037 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001038 *
1039 * This also assumes that the inode count has been incremented
1040 * (or otherwise set) by the caller to indicate that it is now
1041 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 */
David Howells770bfad2006-08-22 20:06:07 -04001043static struct dentry *__d_instantiate_unique(struct dentry *entry,
1044 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
1046 struct dentry *alias;
1047 int len = entry->d_name.len;
1048 const char *name = entry->d_name.name;
1049 unsigned int hash = entry->d_name.hash;
1050
David Howells770bfad2006-08-22 20:06:07 -04001051 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001052 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001053 return NULL;
1054 }
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1057 struct qstr *qstr = &alias->d_name;
1058
1059 if (qstr->hash != hash)
1060 continue;
1061 if (alias->d_parent != entry->d_parent)
1062 continue;
1063 if (qstr->len != len)
1064 continue;
1065 if (memcmp(qstr->name, name, len))
1066 continue;
1067 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 return alias;
1069 }
David Howells770bfad2006-08-22 20:06:07 -04001070
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001071 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return NULL;
1073}
David Howells770bfad2006-08-22 20:06:07 -04001074
1075struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1076{
1077 struct dentry *result;
1078
1079 BUG_ON(!list_empty(&entry->d_alias));
1080
1081 spin_lock(&dcache_lock);
1082 result = __d_instantiate_unique(entry, inode);
1083 spin_unlock(&dcache_lock);
1084
1085 if (!result) {
1086 security_d_instantiate(entry, inode);
1087 return NULL;
1088 }
1089
1090 BUG_ON(!d_unhashed(result));
1091 iput(inode);
1092 return result;
1093}
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095EXPORT_SYMBOL(d_instantiate_unique);
1096
1097/**
1098 * d_alloc_root - allocate root dentry
1099 * @root_inode: inode to allocate the root for
1100 *
1101 * Allocate a root ("/") dentry for the inode given. The inode is
1102 * instantiated and returned. %NULL is returned if there is insufficient
1103 * memory or the inode passed is %NULL.
1104 */
1105
1106struct dentry * d_alloc_root(struct inode * root_inode)
1107{
1108 struct dentry *res = NULL;
1109
1110 if (root_inode) {
1111 static const struct qstr name = { .name = "/", .len = 1 };
1112
1113 res = d_alloc(NULL, &name);
1114 if (res) {
1115 res->d_sb = root_inode->i_sb;
1116 res->d_parent = res;
1117 d_instantiate(res, root_inode);
1118 }
1119 }
1120 return res;
1121}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001122EXPORT_SYMBOL(d_alloc_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124static inline struct hlist_head *d_hash(struct dentry *parent,
1125 unsigned long hash)
1126{
1127 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1128 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1129 return dentry_hashtable + (hash & D_HASHMASK);
1130}
1131
1132/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001133 * d_obtain_alias - find or allocate a dentry for a given inode
1134 * @inode: inode to allocate the dentry for
1135 *
1136 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1137 * similar open by handle operations. The returned dentry may be anonymous,
1138 * or may have a full name (if the inode was already in the cache).
1139 *
1140 * When called on a directory inode, we must ensure that the inode only ever
1141 * has one dentry. If a dentry is found, that is returned instead of
1142 * allocating a new one.
1143 *
1144 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001145 * to the dentry. In case of an error the reference on the inode is released.
1146 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1147 * be passed in and will be the error will be propagate to the return value,
1148 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001149 */
1150struct dentry *d_obtain_alias(struct inode *inode)
1151{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001152 static const struct qstr anonstring = { .name = "" };
1153 struct dentry *tmp;
1154 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001155
1156 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001157 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001158 if (IS_ERR(inode))
1159 return ERR_CAST(inode);
1160
Christoph Hellwig9308a612008-08-11 15:49:12 +02001161 res = d_find_alias(inode);
1162 if (res)
1163 goto out_iput;
1164
1165 tmp = d_alloc(NULL, &anonstring);
1166 if (!tmp) {
1167 res = ERR_PTR(-ENOMEM);
1168 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001169 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001170 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1171
1172 spin_lock(&dcache_lock);
1173 res = __d_find_alias(inode, 0);
1174 if (res) {
1175 spin_unlock(&dcache_lock);
1176 dput(tmp);
1177 goto out_iput;
1178 }
1179
1180 /* attach a disconnected dentry */
1181 spin_lock(&tmp->d_lock);
1182 tmp->d_sb = inode->i_sb;
1183 tmp->d_inode = inode;
1184 tmp->d_flags |= DCACHE_DISCONNECTED;
1185 tmp->d_flags &= ~DCACHE_UNHASHED;
1186 list_add(&tmp->d_alias, &inode->i_dentry);
1187 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1188 spin_unlock(&tmp->d_lock);
1189
1190 spin_unlock(&dcache_lock);
1191 return tmp;
1192
1193 out_iput:
1194 iput(inode);
1195 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001196}
Benny Halevyadc48722009-02-27 14:02:59 -08001197EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199/**
1200 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1201 * @inode: the inode which may have a disconnected dentry
1202 * @dentry: a negative dentry which we want to point to the inode.
1203 *
1204 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1205 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1206 * and return it, else simply d_add the inode to the dentry and return NULL.
1207 *
1208 * This is needed in the lookup routine of any filesystem that is exportable
1209 * (via knfsd) so that we can build dcache paths to directories effectively.
1210 *
1211 * If a dentry was found and moved, then it is returned. Otherwise NULL
1212 * is returned. This matches the expected return value of ->lookup.
1213 *
1214 */
1215struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1216{
1217 struct dentry *new = NULL;
1218
NeilBrown21c0d8f2006-10-04 02:16:16 -07001219 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 spin_lock(&dcache_lock);
1221 new = __d_find_alias(inode, 1);
1222 if (new) {
1223 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1224 spin_unlock(&dcache_lock);
1225 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 d_move(new, dentry);
1227 iput(inode);
1228 } else {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001229 /* already taking dcache_lock, so d_add() by hand */
1230 __d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 spin_unlock(&dcache_lock);
1232 security_d_instantiate(dentry, inode);
1233 d_rehash(dentry);
1234 }
1235 } else
1236 d_add(dentry, inode);
1237 return new;
1238}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001239EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
Barry Naujok94035402008-05-21 16:50:46 +10001241/**
1242 * d_add_ci - lookup or allocate new dentry with case-exact name
1243 * @inode: the inode case-insensitive lookup has found
1244 * @dentry: the negative dentry that was passed to the parent's lookup func
1245 * @name: the case-exact name to be associated with the returned dentry
1246 *
1247 * This is to avoid filling the dcache with case-insensitive names to the
1248 * same inode, only the actual correct case is stored in the dcache for
1249 * case-insensitive filesystems.
1250 *
1251 * For a case-insensitive lookup match and if the the case-exact dentry
1252 * already exists in in the dcache, use it and return it.
1253 *
1254 * If no entry exists with the exact case name, allocate new dentry with
1255 * the exact case, and return the spliced entry.
1256 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001257struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001258 struct qstr *name)
1259{
1260 int error;
1261 struct dentry *found;
1262 struct dentry *new;
1263
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001264 /*
1265 * First check if a dentry matching the name already exists,
1266 * if not go ahead and create it now.
1267 */
Barry Naujok94035402008-05-21 16:50:46 +10001268 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001269 if (!found) {
1270 new = d_alloc(dentry->d_parent, name);
1271 if (!new) {
1272 error = -ENOMEM;
1273 goto err_out;
1274 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001275
Barry Naujok94035402008-05-21 16:50:46 +10001276 found = d_splice_alias(inode, new);
1277 if (found) {
1278 dput(new);
1279 return found;
1280 }
1281 return new;
1282 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001283
1284 /*
1285 * If a matching dentry exists, and it's not negative use it.
1286 *
1287 * Decrement the reference count to balance the iget() done
1288 * earlier on.
1289 */
Barry Naujok94035402008-05-21 16:50:46 +10001290 if (found->d_inode) {
1291 if (unlikely(found->d_inode != inode)) {
1292 /* This can't happen because bad inodes are unhashed. */
1293 BUG_ON(!is_bad_inode(inode));
1294 BUG_ON(!is_bad_inode(found->d_inode));
1295 }
Barry Naujok94035402008-05-21 16:50:46 +10001296 iput(inode);
1297 return found;
1298 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001299
Barry Naujok94035402008-05-21 16:50:46 +10001300 /*
1301 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001302 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001303 */
Barry Naujok94035402008-05-21 16:50:46 +10001304 spin_lock(&dcache_lock);
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001305 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001306 __d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001307 spin_unlock(&dcache_lock);
1308 security_d_instantiate(found, inode);
1309 return found;
1310 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001311
Barry Naujok94035402008-05-21 16:50:46 +10001312 /*
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001313 * In case a directory already has a (disconnected) entry grab a
1314 * reference to it, move it in place and use it.
Barry Naujok94035402008-05-21 16:50:46 +10001315 */
1316 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1317 dget_locked(new);
1318 spin_unlock(&dcache_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001319 security_d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001320 d_move(new, found);
Barry Naujok94035402008-05-21 16:50:46 +10001321 iput(inode);
Barry Naujok94035402008-05-21 16:50:46 +10001322 dput(found);
Barry Naujok94035402008-05-21 16:50:46 +10001323 return new;
1324
1325err_out:
1326 iput(inode);
1327 return ERR_PTR(error);
1328}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001329EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
1331/**
1332 * d_lookup - search for a dentry
1333 * @parent: parent dentry
1334 * @name: qstr of name we wish to find
1335 *
1336 * Searches the children of the parent dentry for the name in question. If
1337 * the dentry is found its reference count is incremented and the dentry
Zhaoleibe42c4c2008-12-01 14:34:58 -08001338 * is returned. The caller must use dput to free the entry when it has
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 * finished using it. %NULL is returned on failure.
1340 *
1341 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1342 * Memory barriers are used while updating and doing lockless traversal.
1343 * To avoid races with d_move while rename is happening, d_lock is used.
1344 *
1345 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1346 * and name pointer in one structure pointed by d_qstr.
1347 *
1348 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1349 * lookup is going on.
1350 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001351 * The dentry unused LRU is not updated even if lookup finds the required dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1353 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1354 * acquisition.
1355 *
1356 * d_lookup() is protected against the concurrent renames in some unrelated
1357 * directory using the seqlockt_t rename_lock.
1358 */
1359
1360struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1361{
1362 struct dentry * dentry = NULL;
1363 unsigned long seq;
1364
1365 do {
1366 seq = read_seqbegin(&rename_lock);
1367 dentry = __d_lookup(parent, name);
1368 if (dentry)
1369 break;
1370 } while (read_seqretry(&rename_lock, seq));
1371 return dentry;
1372}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001373EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
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}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001500EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
1502/*
1503 * When a file is deleted, we have two options:
1504 * - turn this dentry into a negative dentry
1505 * - unhash this dentry and free it.
1506 *
1507 * Usually, we want to just turn this into
1508 * a negative dentry, but if anybody else is
1509 * currently using the dentry or the inode
1510 * we can't do that and we fall back on removing
1511 * it from the hash queues and waiting for
1512 * it to be deleted later when it has no users
1513 */
1514
1515/**
1516 * d_delete - delete a dentry
1517 * @dentry: The dentry to delete
1518 *
1519 * Turn the dentry into a negative dentry if possible, otherwise
1520 * remove it from the hash queues so it can be deleted later
1521 */
1522
1523void d_delete(struct dentry * dentry)
1524{
John McCutchan7a91bf72005-08-08 13:52:16 -04001525 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 /*
1527 * Are we the only user?
1528 */
1529 spin_lock(&dcache_lock);
1530 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001531 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 if (atomic_read(&dentry->d_count) == 1) {
Al Viro13e3c5e2010-05-21 16:11:04 -04001533 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001535 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 return;
1537 }
1538
1539 if (!d_unhashed(dentry))
1540 __d_drop(dentry);
1541
1542 spin_unlock(&dentry->d_lock);
1543 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001544
1545 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001547EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1550{
1551
1552 entry->d_flags &= ~DCACHE_UNHASHED;
1553 hlist_add_head_rcu(&entry->d_hash, list);
1554}
1555
David Howells770bfad2006-08-22 20:06:07 -04001556static void _d_rehash(struct dentry * entry)
1557{
1558 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1559}
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561/**
1562 * d_rehash - add an entry back to the hash
1563 * @entry: dentry to add to the hash
1564 *
1565 * Adds a dentry to the hash according to its name.
1566 */
1567
1568void d_rehash(struct dentry * entry)
1569{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 spin_lock(&dcache_lock);
1571 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001572 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 spin_unlock(&entry->d_lock);
1574 spin_unlock(&dcache_lock);
1575}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001576EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578/*
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 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001596 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 } 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);
Al Virodc711ca2008-11-03 15:03:50 -05001623 dentry->d_name.len = target->d_name.len;
1624 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 }
1626 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001627 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628}
1629
1630/*
1631 * We cannibalize "target" when moving dentry on top of it,
1632 * because it's going to be thrown away anyway. We could be more
1633 * polite about it, though.
1634 *
1635 * This forceful removal will result in ugly /proc output if
1636 * somebody holds a file open that got deleted due to a rename.
1637 * We could be nicer about the deleted file, and let it show
J. Bruce Fieldsbc154b12007-10-16 23:29:42 -07001638 * up under the name it had before it was deleted rather than
1639 * under the original name of the file that was moved on top of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 */
1641
Trond Myklebust9eaef272006-10-21 10:24:20 -07001642/*
1643 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 * @dentry: entry to move
1645 * @target: new dentry
1646 *
1647 * Update the dcache to reflect the move of a file name. Negative
1648 * dcache entries should not be moved in this way.
1649 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001650static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
1652 struct hlist_head *list;
1653
1654 if (!dentry->d_inode)
1655 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 write_seqlock(&rename_lock);
1658 /*
1659 * XXXX: do we really need to take target->d_lock?
1660 */
1661 if (target < dentry) {
1662 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001663 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 } else {
1665 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001666 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 }
1668
1669 /* Move the dentry to the target hash queue, if on different bucket */
Denis Chengf77e3492007-10-16 23:30:11 -07001670 if (d_unhashed(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 goto already_unhashed;
1672
1673 hlist_del_rcu(&dentry->d_hash);
1674
1675already_unhashed:
1676 list = d_hash(target->d_parent, target->d_name.hash);
1677 __d_rehash(dentry, list);
1678
1679 /* Unhash the target: dput() will then get rid of it */
1680 __d_drop(target);
1681
Eric Dumazet5160ee62006-01-08 01:03:32 -08001682 list_del(&dentry->d_u.d_child);
1683 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
1685 /* Switch the names.. */
1686 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001687 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 /* ... and switch the parents */
1690 if (IS_ROOT(dentry)) {
1691 dentry->d_parent = target->d_parent;
1692 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001693 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001695 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
1697 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001698 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 }
1700
Eric Dumazet5160ee62006-01-08 01:03:32 -08001701 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001703 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 spin_unlock(&dentry->d_lock);
1705 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001706}
1707
1708/**
1709 * d_move - move a dentry
1710 * @dentry: entry to move
1711 * @target: new dentry
1712 *
1713 * Update the dcache to reflect the move of a file name. Negative
1714 * dcache entries should not be moved in this way.
1715 */
1716
1717void d_move(struct dentry * dentry, struct dentry * target)
1718{
1719 spin_lock(&dcache_lock);
1720 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 spin_unlock(&dcache_lock);
1722}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001723EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001725/**
1726 * d_ancestor - search for an ancestor
1727 * @p1: ancestor dentry
1728 * @p2: child dentry
1729 *
1730 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1731 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07001732 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001733struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001734{
1735 struct dentry *p;
1736
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001737 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07001738 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001739 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001740 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001741 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001742}
1743
1744/*
1745 * This helper attempts to cope with remotely renamed directories
1746 *
1747 * It assumes that the caller is already holding
1748 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1749 *
1750 * Note: If ever the locking in lock_rename() changes, then please
1751 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07001752 */
1753static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001754 __releases(dcache_lock)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001755{
1756 struct mutex *m1 = NULL, *m2 = NULL;
1757 struct dentry *ret;
1758
1759 /* If alias and dentry share a parent, then no extra locks required */
1760 if (alias->d_parent == dentry->d_parent)
1761 goto out_unalias;
1762
1763 /* Check for loops */
1764 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001765 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07001766 goto out_err;
1767
1768 /* See lock_rename() */
1769 ret = ERR_PTR(-EBUSY);
1770 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1771 goto out_err;
1772 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1773 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1774 goto out_err;
1775 m2 = &alias->d_parent->d_inode->i_mutex;
1776out_unalias:
1777 d_move_locked(alias, dentry);
1778 ret = alias;
1779out_err:
1780 spin_unlock(&dcache_lock);
1781 if (m2)
1782 mutex_unlock(m2);
1783 if (m1)
1784 mutex_unlock(m1);
1785 return ret;
1786}
1787
1788/*
David Howells770bfad2006-08-22 20:06:07 -04001789 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1790 * named dentry in place of the dentry to be replaced.
1791 */
1792static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1793{
1794 struct dentry *dparent, *aparent;
1795
1796 switch_names(dentry, anon);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001797 swap(dentry->d_name.hash, anon->d_name.hash);
David Howells770bfad2006-08-22 20:06:07 -04001798
1799 dparent = dentry->d_parent;
1800 aparent = anon->d_parent;
1801
1802 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1803 list_del(&dentry->d_u.d_child);
1804 if (!IS_ROOT(dentry))
1805 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1806 else
1807 INIT_LIST_HEAD(&dentry->d_u.d_child);
1808
1809 anon->d_parent = (dparent == dentry) ? anon : dparent;
1810 list_del(&anon->d_u.d_child);
1811 if (!IS_ROOT(anon))
1812 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1813 else
1814 INIT_LIST_HEAD(&anon->d_u.d_child);
1815
1816 anon->d_flags &= ~DCACHE_DISCONNECTED;
1817}
1818
1819/**
1820 * d_materialise_unique - introduce an inode into the tree
1821 * @dentry: candidate dentry
1822 * @inode: inode to bind to the dentry, to which aliases may be attached
1823 *
1824 * Introduces an dentry into the tree, substituting an extant disconnected
1825 * root directory alias in its place if there is one
1826 */
1827struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1828{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001829 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001830
1831 BUG_ON(!d_unhashed(dentry));
1832
1833 spin_lock(&dcache_lock);
1834
1835 if (!inode) {
1836 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001837 __d_instantiate(dentry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001838 goto found_lock;
1839 }
1840
Trond Myklebust9eaef272006-10-21 10:24:20 -07001841 if (S_ISDIR(inode->i_mode)) {
1842 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001843
Trond Myklebust9eaef272006-10-21 10:24:20 -07001844 /* Does an aliased dentry already exist? */
1845 alias = __d_find_alias(inode, 0);
1846 if (alias) {
1847 actual = alias;
1848 /* Is this an anonymous mountpoint that we could splice
1849 * into our tree? */
1850 if (IS_ROOT(alias)) {
1851 spin_lock(&alias->d_lock);
1852 __d_materialise_dentry(dentry, alias);
1853 __d_drop(alias);
1854 goto found;
1855 }
1856 /* Nope, but we must(!) avoid directory aliasing */
1857 actual = __d_unalias(dentry, alias);
1858 if (IS_ERR(actual))
1859 dput(alias);
1860 goto out_nolock;
1861 }
David Howells770bfad2006-08-22 20:06:07 -04001862 }
1863
1864 /* Add a unique reference */
1865 actual = __d_instantiate_unique(dentry, inode);
1866 if (!actual)
1867 actual = dentry;
1868 else if (unlikely(!d_unhashed(actual)))
1869 goto shouldnt_be_hashed;
1870
1871found_lock:
1872 spin_lock(&actual->d_lock);
1873found:
1874 _d_rehash(actual);
1875 spin_unlock(&actual->d_lock);
1876 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001877out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001878 if (actual == dentry) {
1879 security_d_instantiate(dentry, inode);
1880 return NULL;
1881 }
1882
1883 iput(inode);
1884 return actual;
1885
David Howells770bfad2006-08-22 20:06:07 -04001886shouldnt_be_hashed:
1887 spin_unlock(&dcache_lock);
1888 BUG();
David Howells770bfad2006-08-22 20:06:07 -04001889}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001890EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04001891
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001892static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01001893{
1894 *buflen -= namelen;
1895 if (*buflen < 0)
1896 return -ENAMETOOLONG;
1897 *buffer -= namelen;
1898 memcpy(*buffer, str, namelen);
1899 return 0;
1900}
1901
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001902static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1903{
1904 return prepend(buffer, buflen, name->name, name->len);
1905}
1906
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907/**
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001908 * Prepend path string to a buffer
1909 *
1910 * @path: the dentry/vfsmount to report
1911 * @root: root vfsmnt/dentry (may be modified by this function)
1912 * @buffer: pointer to the end of the buffer
1913 * @buflen: pointer to buffer length
1914 *
1915 * Caller holds the dcache_lock.
1916 *
1917 * If path is not reachable from the supplied root, then the value of
1918 * root is changed (without modifying refcounts).
1919 */
1920static int prepend_path(const struct path *path, struct path *root,
1921 char **buffer, int *buflen)
1922{
1923 struct dentry *dentry = path->dentry;
1924 struct vfsmount *vfsmnt = path->mnt;
1925 bool slash = false;
1926 int error = 0;
1927
1928 spin_lock(&vfsmount_lock);
1929 while (dentry != root->dentry || vfsmnt != root->mnt) {
1930 struct dentry * parent;
1931
1932 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1933 /* Global root? */
1934 if (vfsmnt->mnt_parent == vfsmnt) {
1935 goto global_root;
1936 }
1937 dentry = vfsmnt->mnt_mountpoint;
1938 vfsmnt = vfsmnt->mnt_parent;
1939 continue;
1940 }
1941 parent = dentry->d_parent;
1942 prefetch(parent);
1943 error = prepend_name(buffer, buflen, &dentry->d_name);
1944 if (!error)
1945 error = prepend(buffer, buflen, "/", 1);
1946 if (error)
1947 break;
1948
1949 slash = true;
1950 dentry = parent;
1951 }
1952
1953out:
1954 if (!error && !slash)
1955 error = prepend(buffer, buflen, "/", 1);
1956
1957 spin_unlock(&vfsmount_lock);
1958 return error;
1959
1960global_root:
1961 /*
1962 * Filesystems needing to implement special "root names"
1963 * should do so with ->d_dname()
1964 */
1965 if (IS_ROOT(dentry) &&
1966 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
1967 WARN(1, "Root dentry has weird name <%.*s>\n",
1968 (int) dentry->d_name.len, dentry->d_name.name);
1969 }
1970 root->mnt = vfsmnt;
1971 root->dentry = dentry;
1972 goto out;
1973}
1974
1975/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001976 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001977 * @path: the dentry/vfsmount to report
1978 * @root: root vfsmnt/dentry (may be modified by this function)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 * @buffer: buffer to return value in
1980 * @buflen: buffer length
1981 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001982 * Convert a dentry into an ASCII path name. If the entry has been deleted
1983 * the string " (deleted)" is appended. Note that this is ambiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08001985 * Returns a pointer into the buffer or an error code if the
1986 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08001987 *
1988 * "buflen" should be positive. Caller holds the dcache_lock.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001989 *
1990 * If path is not reachable from the supplied root, then the value of
1991 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001993char *__d_path(const struct path *path, struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001994 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001996 char *res = buf + buflen;
1997 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001999 prepend(&res, &buflen, "\0", 1);
2000 if (d_unlinked(path->dentry)) {
2001 error = prepend(&res, &buflen, " (deleted)", 10);
2002 if (error)
2003 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 }
2005
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002006 error = prepend_path(path, root, &res, &buflen);
2007 if (error)
2008 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002010 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011}
2012
Jan Bluncka03a8a702008-02-14 19:38:32 -08002013/**
2014 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002015 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002016 * @buf: buffer to return value in
2017 * @buflen: buffer length
2018 *
2019 * Convert a dentry into an ASCII path name. If the entry has been deleted
2020 * the string " (deleted)" is appended. Note that this is ambiguous.
2021 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002022 * Returns a pointer into the buffer or an error code if the path was
2023 * too long. Note: Callers should use the returned pointer, not the passed
2024 * in buffer, to use the name! The implementation often starts at an offset
2025 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002026 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002027 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002028 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002029char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030{
2031 char *res;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002032 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002033 struct path tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002035 /*
2036 * We have various synthetic filesystems that never get mounted. On
2037 * these filesystems dentries are never used for lookup purposes, and
2038 * thus don't need to be hashed. They also don't need a name until a
2039 * user wants to identify the object in /proc/pid/fd/. The little hack
2040 * below allows us to generate a name for these objects on demand:
2041 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002042 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2043 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002044
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002045 get_fs_root(current->fs, &root);
Linus Torvalds552ce542007-02-13 12:08:18 -08002046 spin_lock(&dcache_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002047 tmp = root;
2048 res = __d_path(path, &tmp, buf, buflen);
Linus Torvalds552ce542007-02-13 12:08:18 -08002049 spin_unlock(&dcache_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002050 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 return res;
2052}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002053EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
2055/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002056 * Helper function for dentry_operations.d_dname() members
2057 */
2058char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2059 const char *fmt, ...)
2060{
2061 va_list args;
2062 char temp[64];
2063 int sz;
2064
2065 va_start(args, fmt);
2066 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2067 va_end(args);
2068
2069 if (sz > sizeof(temp) || sz > buflen)
2070 return ERR_PTR(-ENAMETOOLONG);
2071
2072 buffer += buflen - sz;
2073 return memcpy(buffer, temp, sz);
2074}
2075
2076/*
Ram Pai6092d042008-03-27 13:06:20 +01002077 * Write full pathname from the root of the filesystem into the buffer.
2078 */
Al Viroc1031352010-06-06 22:31:14 -04002079char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002080{
2081 char *end = buf + buflen;
2082 char *retval;
2083
Ram Pai6092d042008-03-27 13:06:20 +01002084 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002085 if (buflen < 1)
2086 goto Elong;
2087 /* Get '/' right */
2088 retval = end-1;
2089 *retval = '/';
2090
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002091 while (!IS_ROOT(dentry)) {
2092 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01002093
Ram Pai6092d042008-03-27 13:06:20 +01002094 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002095 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01002096 (prepend(&end, &buflen, "/", 1) != 0))
2097 goto Elong;
2098
2099 retval = end;
2100 dentry = parent;
2101 }
Al Viroc1031352010-06-06 22:31:14 -04002102 return retval;
2103Elong:
2104 return ERR_PTR(-ENAMETOOLONG);
2105}
2106EXPORT_SYMBOL(__dentry_path);
2107
2108char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2109{
2110 char *p = NULL;
2111 char *retval;
2112
2113 spin_lock(&dcache_lock);
2114 if (d_unlinked(dentry)) {
2115 p = buf + buflen;
2116 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2117 goto Elong;
2118 buflen++;
2119 }
2120 retval = __dentry_path(dentry, buf, buflen);
Ram Pai6092d042008-03-27 13:06:20 +01002121 spin_unlock(&dcache_lock);
Al Viroc1031352010-06-06 22:31:14 -04002122 if (!IS_ERR(retval) && p)
2123 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002124 return retval;
2125Elong:
2126 spin_unlock(&dcache_lock);
2127 return ERR_PTR(-ENAMETOOLONG);
2128}
2129
2130/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 * NOTE! The user-level library version returns a
2132 * character pointer. The kernel system call just
2133 * returns the length of the buffer filled (which
2134 * includes the ending '\0' character), or a negative
2135 * error value. So libc would do something like
2136 *
2137 * char *getcwd(char * buf, size_t size)
2138 * {
2139 * int retval;
2140 *
2141 * retval = sys_getcwd(buf, size);
2142 * if (retval >= 0)
2143 * return buf;
2144 * errno = -retval;
2145 * return NULL;
2146 * }
2147 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002148SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149{
Linus Torvalds552ce542007-02-13 12:08:18 -08002150 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002151 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002152 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
2154 if (!page)
2155 return -ENOMEM;
2156
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002157 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
Linus Torvalds552ce542007-02-13 12:08:18 -08002159 error = -ENOENT;
Linus Torvalds552ce542007-02-13 12:08:18 -08002160 spin_lock(&dcache_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002161 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002162 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002163 struct path tmp = root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002164 char * cwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002166 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
Linus Torvalds552ce542007-02-13 12:08:18 -08002167 spin_unlock(&dcache_lock);
2168
2169 error = PTR_ERR(cwd);
2170 if (IS_ERR(cwd))
2171 goto out;
2172
2173 error = -ERANGE;
2174 len = PAGE_SIZE + page - cwd;
2175 if (len <= size) {
2176 error = len;
2177 if (copy_to_user(buf, cwd, len))
2178 error = -EFAULT;
2179 }
2180 } else
2181 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182
2183out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002184 path_put(&pwd);
2185 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 free_page((unsigned long) page);
2187 return error;
2188}
2189
2190/*
2191 * Test whether new_dentry is a subdirectory of old_dentry.
2192 *
2193 * Trivially implemented using the dcache structure
2194 */
2195
2196/**
2197 * is_subdir - is new dentry a subdirectory of old_dentry
2198 * @new_dentry: new dentry
2199 * @old_dentry: old dentry
2200 *
2201 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2202 * Returns 0 otherwise.
2203 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2204 */
2205
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002206int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
2208 int result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 unsigned long seq;
2210
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002211 if (new_dentry == old_dentry)
2212 return 1;
2213
2214 /*
2215 * Need rcu_readlock to protect against the d_parent trashing
2216 * due to d_move
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 */
2218 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002219 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 seq = read_seqbegin(&rename_lock);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002222 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002224 else
2225 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 } while (read_seqretry(&rename_lock, seq));
2227 rcu_read_unlock();
2228
2229 return result;
2230}
2231
Al Viro2096f752010-01-30 13:16:21 -05002232int path_is_under(struct path *path1, struct path *path2)
2233{
2234 struct vfsmount *mnt = path1->mnt;
2235 struct dentry *dentry = path1->dentry;
2236 int res;
2237 spin_lock(&vfsmount_lock);
2238 if (mnt != path2->mnt) {
2239 for (;;) {
2240 if (mnt->mnt_parent == mnt) {
2241 spin_unlock(&vfsmount_lock);
2242 return 0;
2243 }
2244 if (mnt->mnt_parent == path2->mnt)
2245 break;
2246 mnt = mnt->mnt_parent;
2247 }
2248 dentry = mnt->mnt_mountpoint;
2249 }
2250 res = is_subdir(dentry, path2->dentry);
2251 spin_unlock(&vfsmount_lock);
2252 return res;
2253}
2254EXPORT_SYMBOL(path_is_under);
2255
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256void d_genocide(struct dentry *root)
2257{
2258 struct dentry *this_parent = root;
2259 struct list_head *next;
2260
2261 spin_lock(&dcache_lock);
2262repeat:
2263 next = this_parent->d_subdirs.next;
2264resume:
2265 while (next != &this_parent->d_subdirs) {
2266 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002267 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 next = tmp->next;
2269 if (d_unhashed(dentry)||!dentry->d_inode)
2270 continue;
2271 if (!list_empty(&dentry->d_subdirs)) {
2272 this_parent = dentry;
2273 goto repeat;
2274 }
2275 atomic_dec(&dentry->d_count);
2276 }
2277 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002278 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 atomic_dec(&this_parent->d_count);
2280 this_parent = this_parent->d_parent;
2281 goto resume;
2282 }
2283 spin_unlock(&dcache_lock);
2284}
2285
2286/**
2287 * find_inode_number - check for dentry with name
2288 * @dir: directory to check
2289 * @name: Name to find.
2290 *
2291 * Check whether a dentry already exists for the given name,
2292 * and return the inode number if it has an inode. Otherwise
2293 * 0 is returned.
2294 *
2295 * This routine is used to post-process directory listings for
2296 * filesystems using synthetic inode numbers, and is necessary
2297 * to keep getcwd() working.
2298 */
2299
2300ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2301{
2302 struct dentry * dentry;
2303 ino_t ino = 0;
2304
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002305 dentry = d_hash_and_lookup(dir, name);
2306 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 if (dentry->d_inode)
2308 ino = dentry->d_inode->i_ino;
2309 dput(dentry);
2310 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 return ino;
2312}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002313EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
2315static __initdata unsigned long dhash_entries;
2316static int __init set_dhash_entries(char *str)
2317{
2318 if (!str)
2319 return 0;
2320 dhash_entries = simple_strtoul(str, &str, 0);
2321 return 1;
2322}
2323__setup("dhash_entries=", set_dhash_entries);
2324
2325static void __init dcache_init_early(void)
2326{
2327 int loop;
2328
2329 /* If hashes are distributed across NUMA nodes, defer
2330 * hash allocation until vmalloc space is available.
2331 */
2332 if (hashdist)
2333 return;
2334
2335 dentry_hashtable =
2336 alloc_large_system_hash("Dentry cache",
2337 sizeof(struct hlist_head),
2338 dhash_entries,
2339 13,
2340 HASH_EARLY,
2341 &d_hash_shift,
2342 &d_hash_mask,
2343 0);
2344
2345 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2346 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2347}
2348
Denis Cheng74bf17c2007-10-16 23:26:30 -07002349static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350{
2351 int loop;
2352
2353 /*
2354 * A constructor could be added for stable state like the lists,
2355 * but it is probably not worth it because of the cache nature
2356 * of the dcache.
2357 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002358 dentry_cache = KMEM_CACHE(dentry,
2359 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360
Rusty Russell8e1f9362007-07-17 04:03:17 -07002361 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362
2363 /* Hash may have been set up in dcache_init_early */
2364 if (!hashdist)
2365 return;
2366
2367 dentry_hashtable =
2368 alloc_large_system_hash("Dentry cache",
2369 sizeof(struct hlist_head),
2370 dhash_entries,
2371 13,
2372 0,
2373 &d_hash_shift,
2374 &d_hash_mask,
2375 0);
2376
2377 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2378 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2379}
2380
2381/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002382struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002383EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385EXPORT_SYMBOL(d_genocide);
2386
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387void __init vfs_caches_init_early(void)
2388{
2389 dcache_init_early();
2390 inode_init_early();
2391}
2392
2393void __init vfs_caches_init(unsigned long mempages)
2394{
2395 unsigned long reserve;
2396
2397 /* Base hash sizes on available memory, with a reserve equal to
2398 150% of current kernel size */
2399
2400 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2401 mempages -= reserve;
2402
2403 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002404 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405
Denis Cheng74bf17c2007-10-16 23:26:30 -07002406 dcache_init();
2407 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002409 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 bdev_cache_init();
2411 chrdev_init();
2412}