blob: a100fa35a48f3702fbe5522e2636501dc8eb93f3 [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}
260
261/**
262 * d_invalidate - invalidate a dentry
263 * @dentry: dentry to invalidate
264 *
265 * Try to invalidate the dentry if it turns out to be
266 * possible. If there are other dentries that can be
267 * reached through this one we can't delete it and we
268 * return -EBUSY. On success we return 0.
269 *
270 * no dcache lock.
271 */
272
273int d_invalidate(struct dentry * dentry)
274{
275 /*
276 * If it's already been dropped, return OK.
277 */
278 spin_lock(&dcache_lock);
279 if (d_unhashed(dentry)) {
280 spin_unlock(&dcache_lock);
281 return 0;
282 }
283 /*
284 * Check whether to do a partial shrink_dcache
285 * to get rid of unused child entries.
286 */
287 if (!list_empty(&dentry->d_subdirs)) {
288 spin_unlock(&dcache_lock);
289 shrink_dcache_parent(dentry);
290 spin_lock(&dcache_lock);
291 }
292
293 /*
294 * Somebody else still using it?
295 *
296 * If it's a directory, we can't drop it
297 * for fear of somebody re-populating it
298 * with children (even though dropping it
299 * would make it unreachable from the root,
300 * we might still populate it if it was a
301 * working directory or similar).
302 */
303 spin_lock(&dentry->d_lock);
304 if (atomic_read(&dentry->d_count) > 1) {
305 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
306 spin_unlock(&dentry->d_lock);
307 spin_unlock(&dcache_lock);
308 return -EBUSY;
309 }
310 }
311
312 __d_drop(dentry);
313 spin_unlock(&dentry->d_lock);
314 spin_unlock(&dcache_lock);
315 return 0;
316}
317
318/* This should be called _only_ with dcache_lock held */
319
320static inline struct dentry * __dget_locked(struct dentry *dentry)
321{
322 atomic_inc(&dentry->d_count);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700323 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 return dentry;
325}
326
327struct dentry * dget_locked(struct dentry *dentry)
328{
329 return __dget_locked(dentry);
330}
331
332/**
333 * d_find_alias - grab a hashed alias of inode
334 * @inode: inode in question
335 * @want_discon: flag, used by d_splice_alias, to request
336 * that only a DISCONNECTED alias be returned.
337 *
338 * If inode has a hashed alias, or is a directory and has any alias,
339 * acquire the reference to alias and return it. Otherwise return NULL.
340 * Notice that if inode is a directory there can be only one alias and
341 * it can be unhashed only if it has no children, or if it is the root
342 * of a filesystem.
343 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700344 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700346 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 */
348
349static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
350{
351 struct list_head *head, *next, *tmp;
352 struct dentry *alias, *discon_alias=NULL;
353
354 head = &inode->i_dentry;
355 next = inode->i_dentry.next;
356 while (next != head) {
357 tmp = next;
358 next = tmp->next;
359 prefetch(next);
360 alias = list_entry(tmp, struct dentry, d_alias);
361 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700362 if (IS_ROOT(alias) &&
363 (alias->d_flags & DCACHE_DISCONNECTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 discon_alias = alias;
365 else if (!want_discon) {
366 __dget_locked(alias);
367 return alias;
368 }
369 }
370 }
371 if (discon_alias)
372 __dget_locked(discon_alias);
373 return discon_alias;
374}
375
376struct dentry * d_find_alias(struct inode *inode)
377{
David Howells214fda12006-03-25 03:06:36 -0800378 struct dentry *de = NULL;
379
380 if (!list_empty(&inode->i_dentry)) {
381 spin_lock(&dcache_lock);
382 de = __d_find_alias(inode, 0);
383 spin_unlock(&dcache_lock);
384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 return de;
386}
387
388/*
389 * Try to kill dentries associated with this inode.
390 * WARNING: you must own a reference to inode.
391 */
392void d_prune_aliases(struct inode *inode)
393{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700394 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395restart:
396 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700397 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 spin_lock(&dentry->d_lock);
399 if (!atomic_read(&dentry->d_count)) {
400 __dget_locked(dentry);
401 __d_drop(dentry);
402 spin_unlock(&dentry->d_lock);
403 spin_unlock(&dcache_lock);
404 dput(dentry);
405 goto restart;
406 }
407 spin_unlock(&dentry->d_lock);
408 }
409 spin_unlock(&dcache_lock);
410}
411
412/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700413 * Throw away a dentry - free the inode, dput the parent. This requires that
414 * the LRU list has already been removed.
415 *
Miklos Szeredi85864e12007-10-16 23:27:09 -0700416 * Try to prune ancestors as well. This is necessary to prevent
417 * quadratic behavior of shrink_dcache_parent(), but is also expected
418 * to be beneficial in reducing dentry cache fragmentation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 */
Miklos Szeredi85864e12007-10-16 23:27:09 -0700420static void prune_one_dentry(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200421 __releases(dentry->d_lock)
422 __releases(dcache_lock)
423 __acquires(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700426 dentry = d_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700427
428 /*
429 * Prune ancestors. Locking is simpler than in dput(),
430 * because dcache_lock needs to be taken anyway.
431 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 spin_lock(&dcache_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700433 while (dentry) {
434 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
435 return;
436
437 if (dentry->d_op && dentry->d_op->d_delete)
438 dentry->d_op->d_delete(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700439 dentry_lru_del_init(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700440 __d_drop(dentry);
441 dentry = d_kill(dentry);
442 spin_lock(&dcache_lock);
443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444}
445
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700446/*
447 * Shrink the dentry LRU on a given superblock.
448 * @sb : superblock to shrink dentry LRU.
449 * @count: If count is NULL, we prune all dentries on superblock.
450 * @flags: If flags is non-zero, we need to do special processing based on
451 * which flags are set. This means we don't need to maintain multiple
452 * similar copies of this loop.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700454static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700456 LIST_HEAD(referenced);
457 LIST_HEAD(tmp);
458 struct dentry *dentry;
459 int cnt = 0;
460
461 BUG_ON(!sb);
462 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700464 if (count != NULL)
465 /* called from prune_dcache() and shrink_dcache_parent() */
466 cnt = *count;
467restart:
468 if (count == NULL)
469 list_splice_init(&sb->s_dentry_lru, &tmp);
470 else {
471 while (!list_empty(&sb->s_dentry_lru)) {
472 dentry = list_entry(sb->s_dentry_lru.prev,
473 struct dentry, d_lru);
474 BUG_ON(dentry->d_sb != sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700476 spin_lock(&dentry->d_lock);
477 /*
478 * If we are honouring the DCACHE_REFERENCED flag and
479 * the dentry has this flag set, don't free it. Clear
480 * the flag and put it back on the LRU.
NeilBrown0feae5c2006-06-22 14:47:28 -0700481 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700482 if ((flags & DCACHE_REFERENCED)
483 && (dentry->d_flags & DCACHE_REFERENCED)) {
484 dentry->d_flags &= ~DCACHE_REFERENCED;
npiggin@suse.dec490d792009-04-26 20:25:53 +1000485 list_move(&dentry->d_lru, &referenced);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700486 spin_unlock(&dentry->d_lock);
487 } else {
488 list_move_tail(&dentry->d_lru, &tmp);
489 spin_unlock(&dentry->d_lock);
490 cnt--;
491 if (!cnt)
492 break;
NeilBrown0feae5c2006-06-22 14:47:28 -0700493 }
Kentaro Makitaf3c6ba92008-07-25 19:44:40 -0700494 cond_resched_lock(&dcache_lock);
NeilBrown0feae5c2006-06-22 14:47:28 -0700495 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700496 }
497 while (!list_empty(&tmp)) {
498 dentry = list_entry(tmp.prev, struct dentry, d_lru);
499 dentry_lru_del_init(dentry);
500 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 /*
502 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700503 * the LRU because of laziness during lookup. Do not free
504 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700506 if (atomic_read(&dentry->d_count)) {
507 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 continue;
509 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700510 prune_one_dentry(dentry);
511 /* dentry->d_lock was dropped in prune_one_dentry() */
512 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700514 if (count == NULL && !list_empty(&sb->s_dentry_lru))
515 goto restart;
516 if (count != NULL)
517 *count = cnt;
518 if (!list_empty(&referenced))
519 list_splice(&referenced, &sb->s_dentry_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 spin_unlock(&dcache_lock);
521}
522
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700523/**
524 * prune_dcache - shrink the dcache
525 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700527 * Shrink the dcache. This is done when we need more memory, or simply when we
528 * need to unmount something (at which point we need to unuse all dentries).
529 *
530 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700532static void prune_dcache(int count)
533{
534 struct super_block *sb;
535 int w_count;
536 int unused = dentry_stat.nr_unused;
537 int prune_ratio;
538 int pruned;
539
540 if (unused == 0 || count == 0)
541 return;
542 spin_lock(&dcache_lock);
543restart:
544 if (count >= unused)
545 prune_ratio = 1;
546 else
547 prune_ratio = unused / count;
548 spin_lock(&sb_lock);
549 list_for_each_entry(sb, &super_blocks, s_list) {
550 if (sb->s_nr_dentry_unused == 0)
551 continue;
552 sb->s_count++;
553 /* Now, we reclaim unused dentrins with fairness.
554 * We reclaim them same percentage from each superblock.
555 * We calculate number of dentries to scan on this sb
556 * as follows, but the implementation is arranged to avoid
557 * overflows:
558 * number of dentries to scan on this sb =
559 * count * (number of dentries on this sb /
560 * number of dentries in the machine)
561 */
562 spin_unlock(&sb_lock);
563 if (prune_ratio != 1)
564 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
565 else
566 w_count = sb->s_nr_dentry_unused;
567 pruned = w_count;
568 /*
569 * We need to be sure this filesystem isn't being unmounted,
570 * otherwise we could race with generic_shutdown_super(), and
571 * end up holding a reference to an inode while the filesystem
572 * is unmounted. So we try to get s_umount, and make sure
573 * s_root isn't NULL.
574 */
575 if (down_read_trylock(&sb->s_umount)) {
576 if ((sb->s_root != NULL) &&
577 (!list_empty(&sb->s_dentry_lru))) {
578 spin_unlock(&dcache_lock);
579 __shrink_dcache_sb(sb, &w_count,
580 DCACHE_REFERENCED);
581 pruned -= w_count;
582 spin_lock(&dcache_lock);
583 }
584 up_read(&sb->s_umount);
585 }
586 spin_lock(&sb_lock);
587 count -= pruned;
588 /*
589 * restart only when sb is no longer on the list and
590 * we have more work to do.
591 */
592 if (__put_super_and_need_restart(sb) && count > 0) {
593 spin_unlock(&sb_lock);
594 goto restart;
595 }
596 }
597 spin_unlock(&sb_lock);
598 spin_unlock(&dcache_lock);
599}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601/**
602 * shrink_dcache_sb - shrink dcache for a superblock
603 * @sb: superblock
604 *
605 * Shrink the dcache for the specified super block. This
606 * is used to free the dcache before unmounting a file
607 * system
608 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609void shrink_dcache_sb(struct super_block * sb)
610{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700611 __shrink_dcache_sb(sb, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612}
613
614/*
David Howellsc636ebd2006-10-11 01:22:19 -0700615 * destroy a single subtree of dentries for unmount
616 * - see the comments on shrink_dcache_for_umount() for a description of the
617 * locking
618 */
619static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
620{
621 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700622 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700623
624 BUG_ON(!IS_ROOT(dentry));
625
626 /* detach this root from the system */
627 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700628 dentry_lru_del_init(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700629 __d_drop(dentry);
630 spin_unlock(&dcache_lock);
631
632 for (;;) {
633 /* descend to the first leaf in the current subtree */
634 while (!list_empty(&dentry->d_subdirs)) {
635 struct dentry *loop;
636
637 /* this is a branch with children - detach all of them
638 * from the system in one go */
639 spin_lock(&dcache_lock);
640 list_for_each_entry(loop, &dentry->d_subdirs,
641 d_u.d_child) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700642 dentry_lru_del_init(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700643 __d_drop(loop);
644 cond_resched_lock(&dcache_lock);
645 }
646 spin_unlock(&dcache_lock);
647
648 /* move to the first child */
649 dentry = list_entry(dentry->d_subdirs.next,
650 struct dentry, d_u.d_child);
651 }
652
653 /* consume the dentries from this leaf up through its parents
654 * until we find one with children or run out altogether */
655 do {
656 struct inode *inode;
657
658 if (atomic_read(&dentry->d_count) != 0) {
659 printk(KERN_ERR
660 "BUG: Dentry %p{i=%lx,n=%s}"
661 " still in use (%d)"
662 " [unmount of %s %s]\n",
663 dentry,
664 dentry->d_inode ?
665 dentry->d_inode->i_ino : 0UL,
666 dentry->d_name.name,
667 atomic_read(&dentry->d_count),
668 dentry->d_sb->s_type->name,
669 dentry->d_sb->s_id);
670 BUG();
671 }
672
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900673 if (IS_ROOT(dentry))
David Howellsc636ebd2006-10-11 01:22:19 -0700674 parent = NULL;
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900675 else {
676 parent = dentry->d_parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700677 atomic_dec(&parent->d_count);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900678 }
David Howellsc636ebd2006-10-11 01:22:19 -0700679
680 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700681 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700682
683 inode = dentry->d_inode;
684 if (inode) {
685 dentry->d_inode = NULL;
686 list_del_init(&dentry->d_alias);
687 if (dentry->d_op && dentry->d_op->d_iput)
688 dentry->d_op->d_iput(dentry, inode);
689 else
690 iput(inode);
691 }
692
693 d_free(dentry);
694
695 /* finished when we fall off the top of the tree,
696 * otherwise we ascend to the parent and move to the
697 * next sibling if there is one */
698 if (!parent)
David Howellsf8713572006-10-28 10:38:46 -0700699 goto out;
David Howellsc636ebd2006-10-11 01:22:19 -0700700
701 dentry = parent;
702
703 } while (list_empty(&dentry->d_subdirs));
704
705 dentry = list_entry(dentry->d_subdirs.next,
706 struct dentry, d_u.d_child);
707 }
David Howellsf8713572006-10-28 10:38:46 -0700708out:
709 /* several dentries were freed, need to correct nr_dentry */
710 spin_lock(&dcache_lock);
711 dentry_stat.nr_dentry -= detached;
712 spin_unlock(&dcache_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700713}
714
715/*
716 * destroy the dentries attached to a superblock on unmounting
717 * - we don't need to use dentry->d_lock, and only need dcache_lock when
718 * removing the dentry from the system lists and hashes because:
719 * - the superblock is detached from all mountings and open files, so the
720 * dentry trees will not be rearranged by the VFS
721 * - s_umount is write-locked, so the memory pressure shrinker will ignore
722 * any dentries belonging to this superblock that it comes across
723 * - the filesystem itself is no longer permitted to rearrange the dentries
724 * in this superblock
725 */
726void shrink_dcache_for_umount(struct super_block *sb)
727{
728 struct dentry *dentry;
729
730 if (down_read_trylock(&sb->s_umount))
731 BUG();
732
733 dentry = sb->s_root;
734 sb->s_root = NULL;
735 atomic_dec(&dentry->d_count);
736 shrink_dcache_for_umount_subtree(dentry);
737
738 while (!hlist_empty(&sb->s_anon)) {
739 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
740 shrink_dcache_for_umount_subtree(dentry);
741 }
742}
743
744/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 * Search for at least 1 mount point in the dentry's subdirs.
746 * We descend to the next level whenever the d_subdirs
747 * list is non-empty and continue searching.
748 */
749
750/**
751 * have_submounts - check for mounts over a dentry
752 * @parent: dentry to check.
753 *
754 * Return true if the parent or its subdirectories contain
755 * a mount point
756 */
757
758int have_submounts(struct dentry *parent)
759{
760 struct dentry *this_parent = parent;
761 struct list_head *next;
762
763 spin_lock(&dcache_lock);
764 if (d_mountpoint(parent))
765 goto positive;
766repeat:
767 next = this_parent->d_subdirs.next;
768resume:
769 while (next != &this_parent->d_subdirs) {
770 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800771 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 next = tmp->next;
773 /* Have we found a mount point ? */
774 if (d_mountpoint(dentry))
775 goto positive;
776 if (!list_empty(&dentry->d_subdirs)) {
777 this_parent = dentry;
778 goto repeat;
779 }
780 }
781 /*
782 * All done at this level ... ascend and resume the search.
783 */
784 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800785 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 this_parent = this_parent->d_parent;
787 goto resume;
788 }
789 spin_unlock(&dcache_lock);
790 return 0; /* No mount points found in tree */
791positive:
792 spin_unlock(&dcache_lock);
793 return 1;
794}
795
796/*
797 * Search the dentry child list for the specified parent,
798 * and move any unused dentries to the end of the unused
799 * list for prune_dcache(). We descend to the next level
800 * whenever the d_subdirs list is non-empty and continue
801 * searching.
802 *
803 * It returns zero iff there are no unused children,
804 * otherwise it returns the number of children moved to
805 * the end of the unused list. This may not be the total
806 * number of unused children, because select_parent can
807 * drop the lock and return early due to latency
808 * constraints.
809 */
810static int select_parent(struct dentry * parent)
811{
812 struct dentry *this_parent = parent;
813 struct list_head *next;
814 int found = 0;
815
816 spin_lock(&dcache_lock);
817repeat:
818 next = this_parent->d_subdirs.next;
819resume:
820 while (next != &this_parent->d_subdirs) {
821 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800822 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 next = tmp->next;
824
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700825 dentry_lru_del_init(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 /*
827 * move only zero ref count dentries to the end
828 * of the unused list for prune_dcache
829 */
830 if (!atomic_read(&dentry->d_count)) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700831 dentry_lru_add_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 found++;
833 }
834
835 /*
836 * We can return to the caller if we have found some (this
837 * ensures forward progress). We'll be coming back to find
838 * the rest.
839 */
840 if (found && need_resched())
841 goto out;
842
843 /*
844 * Descend a level if the d_subdirs list is non-empty.
845 */
846 if (!list_empty(&dentry->d_subdirs)) {
847 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 goto repeat;
849 }
850 }
851 /*
852 * All done at this level ... ascend and resume the search.
853 */
854 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800855 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 goto resume;
858 }
859out:
860 spin_unlock(&dcache_lock);
861 return found;
862}
863
864/**
865 * shrink_dcache_parent - prune dcache
866 * @parent: parent of entries to prune
867 *
868 * Prune the dcache to remove unused children of the parent dentry.
869 */
870
871void shrink_dcache_parent(struct dentry * parent)
872{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700873 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 int found;
875
876 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700877 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880/*
881 * Scan `nr' dentries and return the number which remain.
882 *
883 * We need to avoid reentering the filesystem if the caller is performing a
884 * GFP_NOFS allocation attempt. One example deadlock is:
885 *
886 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
887 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
888 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
889 *
890 * In this case we return -1 to tell the caller that we baled.
891 */
Al Viro27496a82005-10-21 03:20:48 -0400892static int shrink_dcache_memory(int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
894 if (nr) {
895 if (!(gfp_mask & __GFP_FS))
896 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700897 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 }
899 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
900}
901
Rusty Russell8e1f9362007-07-17 04:03:17 -0700902static struct shrinker dcache_shrinker = {
903 .shrink = shrink_dcache_memory,
904 .seeks = DEFAULT_SEEKS,
905};
906
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907/**
908 * d_alloc - allocate a dcache entry
909 * @parent: parent of entry to allocate
910 * @name: qstr of the name
911 *
912 * Allocates a dentry. It returns %NULL if there is insufficient memory
913 * available. On a success the dentry is returned. The name passed in is
914 * copied and the copy passed in may be reused after this call.
915 */
916
917struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
918{
919 struct dentry *dentry;
920 char *dname;
921
Mel Gormane12ba742007-10-16 01:25:52 -0700922 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 if (!dentry)
924 return NULL;
925
926 if (name->len > DNAME_INLINE_LEN-1) {
927 dname = kmalloc(name->len + 1, GFP_KERNEL);
928 if (!dname) {
929 kmem_cache_free(dentry_cache, dentry);
930 return NULL;
931 }
932 } else {
933 dname = dentry->d_iname;
934 }
935 dentry->d_name.name = dname;
936
937 dentry->d_name.len = name->len;
938 dentry->d_name.hash = name->hash;
939 memcpy(dname, name->name, name->len);
940 dname[name->len] = 0;
941
942 atomic_set(&dentry->d_count, 1);
943 dentry->d_flags = DCACHE_UNHASHED;
944 spin_lock_init(&dentry->d_lock);
945 dentry->d_inode = NULL;
946 dentry->d_parent = NULL;
947 dentry->d_sb = NULL;
948 dentry->d_op = NULL;
949 dentry->d_fsdata = NULL;
950 dentry->d_mounted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 INIT_HLIST_NODE(&dentry->d_hash);
952 INIT_LIST_HEAD(&dentry->d_lru);
953 INIT_LIST_HEAD(&dentry->d_subdirs);
954 INIT_LIST_HEAD(&dentry->d_alias);
955
956 if (parent) {
957 dentry->d_parent = dget(parent);
958 dentry->d_sb = parent->d_sb;
959 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800960 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 }
962
963 spin_lock(&dcache_lock);
964 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800965 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 dentry_stat.nr_dentry++;
967 spin_unlock(&dcache_lock);
968
969 return dentry;
970}
971
972struct dentry *d_alloc_name(struct dentry *parent, const char *name)
973{
974 struct qstr q;
975
976 q.name = name;
977 q.len = strlen(name);
978 q.hash = full_name_hash(q.name, q.len);
979 return d_alloc(parent, &q);
980}
981
OGAWA Hirofumi360da902008-10-16 07:50:28 +0900982/* the caller must hold dcache_lock */
983static void __d_instantiate(struct dentry *dentry, struct inode *inode)
984{
985 if (inode)
986 list_add(&dentry->d_alias, &inode->i_dentry);
987 dentry->d_inode = inode;
988 fsnotify_d_instantiate(dentry, inode);
989}
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991/**
992 * d_instantiate - fill in inode information for a dentry
993 * @entry: dentry to complete
994 * @inode: inode to attach to this dentry
995 *
996 * Fill in inode information in the entry.
997 *
998 * This turns negative dentries into productive full members
999 * of society.
1000 *
1001 * NOTE! This assumes that the inode count has been incremented
1002 * (or otherwise set) by the caller to indicate that it is now
1003 * in use by the dcache.
1004 */
1005
1006void d_instantiate(struct dentry *entry, struct inode * inode)
1007{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001008 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 spin_lock(&dcache_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001010 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 spin_unlock(&dcache_lock);
1012 security_d_instantiate(entry, inode);
1013}
1014
1015/**
1016 * d_instantiate_unique - instantiate a non-aliased dentry
1017 * @entry: dentry to instantiate
1018 * @inode: inode to attach to this dentry
1019 *
1020 * Fill in inode information in the entry. On success, it returns NULL.
1021 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001022 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 *
1024 * Note that in order to avoid conflicts with rename() etc, the caller
1025 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001026 *
1027 * This also assumes that the inode count has been incremented
1028 * (or otherwise set) by the caller to indicate that it is now
1029 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 */
David Howells770bfad2006-08-22 20:06:07 -04001031static struct dentry *__d_instantiate_unique(struct dentry *entry,
1032 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
1034 struct dentry *alias;
1035 int len = entry->d_name.len;
1036 const char *name = entry->d_name.name;
1037 unsigned int hash = entry->d_name.hash;
1038
David Howells770bfad2006-08-22 20:06:07 -04001039 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001040 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001041 return NULL;
1042 }
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1045 struct qstr *qstr = &alias->d_name;
1046
1047 if (qstr->hash != hash)
1048 continue;
1049 if (alias->d_parent != entry->d_parent)
1050 continue;
1051 if (qstr->len != len)
1052 continue;
1053 if (memcmp(qstr->name, name, len))
1054 continue;
1055 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 return alias;
1057 }
David Howells770bfad2006-08-22 20:06:07 -04001058
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001059 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 return NULL;
1061}
David Howells770bfad2006-08-22 20:06:07 -04001062
1063struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1064{
1065 struct dentry *result;
1066
1067 BUG_ON(!list_empty(&entry->d_alias));
1068
1069 spin_lock(&dcache_lock);
1070 result = __d_instantiate_unique(entry, inode);
1071 spin_unlock(&dcache_lock);
1072
1073 if (!result) {
1074 security_d_instantiate(entry, inode);
1075 return NULL;
1076 }
1077
1078 BUG_ON(!d_unhashed(result));
1079 iput(inode);
1080 return result;
1081}
1082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083EXPORT_SYMBOL(d_instantiate_unique);
1084
1085/**
1086 * d_alloc_root - allocate root dentry
1087 * @root_inode: inode to allocate the root for
1088 *
1089 * Allocate a root ("/") dentry for the inode given. The inode is
1090 * instantiated and returned. %NULL is returned if there is insufficient
1091 * memory or the inode passed is %NULL.
1092 */
1093
1094struct dentry * d_alloc_root(struct inode * root_inode)
1095{
1096 struct dentry *res = NULL;
1097
1098 if (root_inode) {
1099 static const struct qstr name = { .name = "/", .len = 1 };
1100
1101 res = d_alloc(NULL, &name);
1102 if (res) {
1103 res->d_sb = root_inode->i_sb;
1104 res->d_parent = res;
1105 d_instantiate(res, root_inode);
1106 }
1107 }
1108 return res;
1109}
1110
1111static inline struct hlist_head *d_hash(struct dentry *parent,
1112 unsigned long hash)
1113{
1114 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1115 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1116 return dentry_hashtable + (hash & D_HASHMASK);
1117}
1118
1119/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001120 * d_obtain_alias - find or allocate a dentry for a given inode
1121 * @inode: inode to allocate the dentry for
1122 *
1123 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1124 * similar open by handle operations. The returned dentry may be anonymous,
1125 * or may have a full name (if the inode was already in the cache).
1126 *
1127 * When called on a directory inode, we must ensure that the inode only ever
1128 * has one dentry. If a dentry is found, that is returned instead of
1129 * allocating a new one.
1130 *
1131 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001132 * to the dentry. In case of an error the reference on the inode is released.
1133 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1134 * be passed in and will be the error will be propagate to the return value,
1135 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001136 */
1137struct dentry *d_obtain_alias(struct inode *inode)
1138{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001139 static const struct qstr anonstring = { .name = "" };
1140 struct dentry *tmp;
1141 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001142
1143 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001144 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001145 if (IS_ERR(inode))
1146 return ERR_CAST(inode);
1147
Christoph Hellwig9308a612008-08-11 15:49:12 +02001148 res = d_find_alias(inode);
1149 if (res)
1150 goto out_iput;
1151
1152 tmp = d_alloc(NULL, &anonstring);
1153 if (!tmp) {
1154 res = ERR_PTR(-ENOMEM);
1155 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001156 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001157 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1158
1159 spin_lock(&dcache_lock);
1160 res = __d_find_alias(inode, 0);
1161 if (res) {
1162 spin_unlock(&dcache_lock);
1163 dput(tmp);
1164 goto out_iput;
1165 }
1166
1167 /* attach a disconnected dentry */
1168 spin_lock(&tmp->d_lock);
1169 tmp->d_sb = inode->i_sb;
1170 tmp->d_inode = inode;
1171 tmp->d_flags |= DCACHE_DISCONNECTED;
1172 tmp->d_flags &= ~DCACHE_UNHASHED;
1173 list_add(&tmp->d_alias, &inode->i_dentry);
1174 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1175 spin_unlock(&tmp->d_lock);
1176
1177 spin_unlock(&dcache_lock);
1178 return tmp;
1179
1180 out_iput:
1181 iput(inode);
1182 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001183}
Benny Halevyadc48722009-02-27 14:02:59 -08001184EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186/**
1187 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1188 * @inode: the inode which may have a disconnected dentry
1189 * @dentry: a negative dentry which we want to point to the inode.
1190 *
1191 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1192 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1193 * and return it, else simply d_add the inode to the dentry and return NULL.
1194 *
1195 * This is needed in the lookup routine of any filesystem that is exportable
1196 * (via knfsd) so that we can build dcache paths to directories effectively.
1197 *
1198 * If a dentry was found and moved, then it is returned. Otherwise NULL
1199 * is returned. This matches the expected return value of ->lookup.
1200 *
1201 */
1202struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1203{
1204 struct dentry *new = NULL;
1205
NeilBrown21c0d8f2006-10-04 02:16:16 -07001206 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 spin_lock(&dcache_lock);
1208 new = __d_find_alias(inode, 1);
1209 if (new) {
1210 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1211 spin_unlock(&dcache_lock);
1212 security_d_instantiate(new, inode);
1213 d_rehash(dentry);
1214 d_move(new, dentry);
1215 iput(inode);
1216 } else {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001217 /* already taking dcache_lock, so d_add() by hand */
1218 __d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 spin_unlock(&dcache_lock);
1220 security_d_instantiate(dentry, inode);
1221 d_rehash(dentry);
1222 }
1223 } else
1224 d_add(dentry, inode);
1225 return new;
1226}
1227
Barry Naujok94035402008-05-21 16:50:46 +10001228/**
1229 * d_add_ci - lookup or allocate new dentry with case-exact name
1230 * @inode: the inode case-insensitive lookup has found
1231 * @dentry: the negative dentry that was passed to the parent's lookup func
1232 * @name: the case-exact name to be associated with the returned dentry
1233 *
1234 * This is to avoid filling the dcache with case-insensitive names to the
1235 * same inode, only the actual correct case is stored in the dcache for
1236 * case-insensitive filesystems.
1237 *
1238 * For a case-insensitive lookup match and if the the case-exact dentry
1239 * already exists in in the dcache, use it and return it.
1240 *
1241 * If no entry exists with the exact case name, allocate new dentry with
1242 * the exact case, and return the spliced entry.
1243 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001244struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001245 struct qstr *name)
1246{
1247 int error;
1248 struct dentry *found;
1249 struct dentry *new;
1250
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001251 /*
1252 * First check if a dentry matching the name already exists,
1253 * if not go ahead and create it now.
1254 */
Barry Naujok94035402008-05-21 16:50:46 +10001255 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001256 if (!found) {
1257 new = d_alloc(dentry->d_parent, name);
1258 if (!new) {
1259 error = -ENOMEM;
1260 goto err_out;
1261 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001262
Barry Naujok94035402008-05-21 16:50:46 +10001263 found = d_splice_alias(inode, new);
1264 if (found) {
1265 dput(new);
1266 return found;
1267 }
1268 return new;
1269 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001270
1271 /*
1272 * If a matching dentry exists, and it's not negative use it.
1273 *
1274 * Decrement the reference count to balance the iget() done
1275 * earlier on.
1276 */
Barry Naujok94035402008-05-21 16:50:46 +10001277 if (found->d_inode) {
1278 if (unlikely(found->d_inode != inode)) {
1279 /* This can't happen because bad inodes are unhashed. */
1280 BUG_ON(!is_bad_inode(inode));
1281 BUG_ON(!is_bad_inode(found->d_inode));
1282 }
Barry Naujok94035402008-05-21 16:50:46 +10001283 iput(inode);
1284 return found;
1285 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001286
Barry Naujok94035402008-05-21 16:50:46 +10001287 /*
1288 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001289 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001290 */
Barry Naujok94035402008-05-21 16:50:46 +10001291 spin_lock(&dcache_lock);
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001292 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001293 __d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001294 spin_unlock(&dcache_lock);
1295 security_d_instantiate(found, inode);
1296 return found;
1297 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001298
Barry Naujok94035402008-05-21 16:50:46 +10001299 /*
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001300 * In case a directory already has a (disconnected) entry grab a
1301 * reference to it, move it in place and use it.
Barry Naujok94035402008-05-21 16:50:46 +10001302 */
1303 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1304 dget_locked(new);
1305 spin_unlock(&dcache_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001306 security_d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001307 d_move(new, found);
Barry Naujok94035402008-05-21 16:50:46 +10001308 iput(inode);
Barry Naujok94035402008-05-21 16:50:46 +10001309 dput(found);
Barry Naujok94035402008-05-21 16:50:46 +10001310 return new;
1311
1312err_out:
1313 iput(inode);
1314 return ERR_PTR(error);
1315}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
1317/**
1318 * d_lookup - search for a dentry
1319 * @parent: parent dentry
1320 * @name: qstr of name we wish to find
1321 *
1322 * Searches the children of the parent dentry for the name in question. If
1323 * the dentry is found its reference count is incremented and the dentry
Zhaoleibe42c4c2008-12-01 14:34:58 -08001324 * is returned. The caller must use dput to free the entry when it has
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 * finished using it. %NULL is returned on failure.
1326 *
1327 * __d_lookup is dcache_lock free. The hash list is protected using RCU.
1328 * Memory barriers are used while updating and doing lockless traversal.
1329 * To avoid races with d_move while rename is happening, d_lock is used.
1330 *
1331 * Overflows in memcmp(), while d_move, are avoided by keeping the length
1332 * and name pointer in one structure pointed by d_qstr.
1333 *
1334 * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while
1335 * lookup is going on.
1336 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001337 * The dentry unused LRU is not updated even if lookup finds the required dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 * in there. It is updated in places such as prune_dcache, shrink_dcache_sb,
1339 * select_parent and __dget_locked. This laziness saves lookup from dcache_lock
1340 * acquisition.
1341 *
1342 * d_lookup() is protected against the concurrent renames in some unrelated
1343 * directory using the seqlockt_t rename_lock.
1344 */
1345
1346struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1347{
1348 struct dentry * dentry = NULL;
1349 unsigned long seq;
1350
1351 do {
1352 seq = read_seqbegin(&rename_lock);
1353 dentry = __d_lookup(parent, name);
1354 if (dentry)
1355 break;
1356 } while (read_seqretry(&rename_lock, seq));
1357 return dentry;
1358}
1359
1360struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1361{
1362 unsigned int len = name->len;
1363 unsigned int hash = name->hash;
1364 const unsigned char *str = name->name;
1365 struct hlist_head *head = d_hash(parent,hash);
1366 struct dentry *found = NULL;
1367 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001368 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370 rcu_read_lock();
1371
Paul E. McKenney665a7582005-11-07 00:59:17 -08001372 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 struct qstr *qstr;
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 if (dentry->d_name.hash != hash)
1376 continue;
1377 if (dentry->d_parent != parent)
1378 continue;
1379
1380 spin_lock(&dentry->d_lock);
1381
1382 /*
1383 * Recheck the dentry after taking the lock - d_move may have
1384 * changed things. Don't bother checking the hash because we're
1385 * about to compare the whole name anyway.
1386 */
1387 if (dentry->d_parent != parent)
1388 goto next;
1389
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001390 /* non-existing due to RCU? */
1391 if (d_unhashed(dentry))
1392 goto next;
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 /*
1395 * It is safe to compare names since d_move() cannot
1396 * change the qstr (protected by d_lock).
1397 */
1398 qstr = &dentry->d_name;
1399 if (parent->d_op && parent->d_op->d_compare) {
1400 if (parent->d_op->d_compare(parent, qstr, name))
1401 goto next;
1402 } else {
1403 if (qstr->len != len)
1404 goto next;
1405 if (memcmp(qstr->name, str, len))
1406 goto next;
1407 }
1408
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001409 atomic_inc(&dentry->d_count);
1410 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 spin_unlock(&dentry->d_lock);
1412 break;
1413next:
1414 spin_unlock(&dentry->d_lock);
1415 }
1416 rcu_read_unlock();
1417
1418 return found;
1419}
1420
1421/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001422 * d_hash_and_lookup - hash the qstr then search for a dentry
1423 * @dir: Directory to search in
1424 * @name: qstr of name we wish to find
1425 *
1426 * On hash failure or on lookup failure NULL is returned.
1427 */
1428struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1429{
1430 struct dentry *dentry = NULL;
1431
1432 /*
1433 * Check for a fs-specific hash function. Note that we must
1434 * calculate the standard hash first, as the d_op->d_hash()
1435 * routine may choose to leave the hash value unchanged.
1436 */
1437 name->hash = full_name_hash(name->name, name->len);
1438 if (dir->d_op && dir->d_op->d_hash) {
1439 if (dir->d_op->d_hash(dir, name) < 0)
1440 goto out;
1441 }
1442 dentry = d_lookup(dir, name);
1443out:
1444 return dentry;
1445}
1446
1447/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 * d_validate - verify dentry provided from insecure source
1449 * @dentry: The dentry alleged to be valid child of @dparent
1450 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 *
1452 * An insecure source has sent us a dentry, here we verify it and dget() it.
1453 * This is used by ncpfs in its readdir implementation.
1454 * Zero is returned in the dentry is invalid.
1455 */
1456
1457int d_validate(struct dentry *dentry, struct dentry *dparent)
1458{
1459 struct hlist_head *base;
1460 struct hlist_node *lhp;
1461
1462 /* Check whether the ptr might be valid at all.. */
1463 if (!kmem_ptr_validate(dentry_cache, dentry))
1464 goto out;
1465
1466 if (dentry->d_parent != dparent)
1467 goto out;
1468
1469 spin_lock(&dcache_lock);
1470 base = d_hash(dparent, dentry->d_name.hash);
1471 hlist_for_each(lhp,base) {
Paul E. McKenney665a7582005-11-07 00:59:17 -08001472 /* hlist_for_each_entry_rcu() not required for d_hash list
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 * as it is parsed under dcache_lock
1474 */
1475 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1476 __dget_locked(dentry);
1477 spin_unlock(&dcache_lock);
1478 return 1;
1479 }
1480 }
1481 spin_unlock(&dcache_lock);
1482out:
1483 return 0;
1484}
1485
1486/*
1487 * When a file is deleted, we have two options:
1488 * - turn this dentry into a negative dentry
1489 * - unhash this dentry and free it.
1490 *
1491 * Usually, we want to just turn this into
1492 * a negative dentry, but if anybody else is
1493 * currently using the dentry or the inode
1494 * we can't do that and we fall back on removing
1495 * it from the hash queues and waiting for
1496 * it to be deleted later when it has no users
1497 */
1498
1499/**
1500 * d_delete - delete a dentry
1501 * @dentry: The dentry to delete
1502 *
1503 * Turn the dentry into a negative dentry if possible, otherwise
1504 * remove it from the hash queues so it can be deleted later
1505 */
1506
1507void d_delete(struct dentry * dentry)
1508{
John McCutchan7a91bf72005-08-08 13:52:16 -04001509 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 /*
1511 * Are we the only user?
1512 */
1513 spin_lock(&dcache_lock);
1514 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001515 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 if (atomic_read(&dentry->d_count) == 1) {
1517 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001518 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 return;
1520 }
1521
1522 if (!d_unhashed(dentry))
1523 __d_drop(dentry);
1524
1525 spin_unlock(&dentry->d_lock);
1526 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001527
1528 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529}
1530
1531static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1532{
1533
1534 entry->d_flags &= ~DCACHE_UNHASHED;
1535 hlist_add_head_rcu(&entry->d_hash, list);
1536}
1537
David Howells770bfad2006-08-22 20:06:07 -04001538static void _d_rehash(struct dentry * entry)
1539{
1540 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1541}
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543/**
1544 * d_rehash - add an entry back to the hash
1545 * @entry: dentry to add to the hash
1546 *
1547 * Adds a dentry to the hash according to its name.
1548 */
1549
1550void d_rehash(struct dentry * entry)
1551{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 spin_lock(&dcache_lock);
1553 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001554 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 spin_unlock(&entry->d_lock);
1556 spin_unlock(&dcache_lock);
1557}
1558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559/*
1560 * When switching names, the actual string doesn't strictly have to
1561 * be preserved in the target - because we're dropping the target
1562 * anyway. As such, we can just do a simple memcpy() to copy over
1563 * the new name before we switch.
1564 *
1565 * Note that we have to be a lot more careful about getting the hash
1566 * switched - we have to switch the hash value properly even if it
1567 * then no longer matches the actual (corrupted) string of the target.
1568 * The hash value has to match the hash queue that the dentry is on..
1569 */
1570static void switch_names(struct dentry *dentry, struct dentry *target)
1571{
1572 if (dname_external(target)) {
1573 if (dname_external(dentry)) {
1574 /*
1575 * Both external: swap the pointers
1576 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001577 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 } else {
1579 /*
1580 * dentry:internal, target:external. Steal target's
1581 * storage and make target internal.
1582 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07001583 memcpy(target->d_iname, dentry->d_name.name,
1584 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 dentry->d_name.name = target->d_name.name;
1586 target->d_name.name = target->d_iname;
1587 }
1588 } else {
1589 if (dname_external(dentry)) {
1590 /*
1591 * dentry:external, target:internal. Give dentry's
1592 * storage to target and make dentry internal
1593 */
1594 memcpy(dentry->d_iname, target->d_name.name,
1595 target->d_name.len + 1);
1596 target->d_name.name = dentry->d_name.name;
1597 dentry->d_name.name = dentry->d_iname;
1598 } else {
1599 /*
1600 * Both are internal. Just copy target to dentry
1601 */
1602 memcpy(dentry->d_iname, target->d_name.name,
1603 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05001604 dentry->d_name.len = target->d_name.len;
1605 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 }
1607 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001608 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609}
1610
1611/*
1612 * We cannibalize "target" when moving dentry on top of it,
1613 * because it's going to be thrown away anyway. We could be more
1614 * polite about it, though.
1615 *
1616 * This forceful removal will result in ugly /proc output if
1617 * somebody holds a file open that got deleted due to a rename.
1618 * We could be nicer about the deleted file, and let it show
J. Bruce Fieldsbc154b12007-10-16 23:29:42 -07001619 * up under the name it had before it was deleted rather than
1620 * under the original name of the file that was moved on top of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 */
1622
Trond Myklebust9eaef272006-10-21 10:24:20 -07001623/*
1624 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 * @dentry: entry to move
1626 * @target: new dentry
1627 *
1628 * Update the dcache to reflect the move of a file name. Negative
1629 * dcache entries should not be moved in this way.
1630 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001631static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632{
1633 struct hlist_head *list;
1634
1635 if (!dentry->d_inode)
1636 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1637
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 write_seqlock(&rename_lock);
1639 /*
1640 * XXXX: do we really need to take target->d_lock?
1641 */
1642 if (target < dentry) {
1643 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001644 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 } else {
1646 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001647 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
1649
1650 /* Move the dentry to the target hash queue, if on different bucket */
Denis Chengf77e3492007-10-16 23:30:11 -07001651 if (d_unhashed(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 goto already_unhashed;
1653
1654 hlist_del_rcu(&dentry->d_hash);
1655
1656already_unhashed:
1657 list = d_hash(target->d_parent, target->d_name.hash);
1658 __d_rehash(dentry, list);
1659
1660 /* Unhash the target: dput() will then get rid of it */
1661 __d_drop(target);
1662
Eric Dumazet5160ee62006-01-08 01:03:32 -08001663 list_del(&dentry->d_u.d_child);
1664 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666 /* Switch the names.. */
1667 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001668 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
1670 /* ... and switch the parents */
1671 if (IS_ROOT(dentry)) {
1672 dentry->d_parent = target->d_parent;
1673 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001674 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001676 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
1678 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001679 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
1681
Eric Dumazet5160ee62006-01-08 01:03:32 -08001682 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001684 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 spin_unlock(&dentry->d_lock);
1686 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001687}
1688
1689/**
1690 * d_move - move a dentry
1691 * @dentry: entry to move
1692 * @target: new dentry
1693 *
1694 * Update the dcache to reflect the move of a file name. Negative
1695 * dcache entries should not be moved in this way.
1696 */
1697
1698void d_move(struct dentry * dentry, struct dentry * target)
1699{
1700 spin_lock(&dcache_lock);
1701 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 spin_unlock(&dcache_lock);
1703}
1704
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001705/**
1706 * d_ancestor - search for an ancestor
1707 * @p1: ancestor dentry
1708 * @p2: child dentry
1709 *
1710 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1711 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07001712 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001713struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001714{
1715 struct dentry *p;
1716
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001717 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07001718 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001719 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001720 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001721 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001722}
1723
1724/*
1725 * This helper attempts to cope with remotely renamed directories
1726 *
1727 * It assumes that the caller is already holding
1728 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1729 *
1730 * Note: If ever the locking in lock_rename() changes, then please
1731 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07001732 */
1733static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001734 __releases(dcache_lock)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001735{
1736 struct mutex *m1 = NULL, *m2 = NULL;
1737 struct dentry *ret;
1738
1739 /* If alias and dentry share a parent, then no extra locks required */
1740 if (alias->d_parent == dentry->d_parent)
1741 goto out_unalias;
1742
1743 /* Check for loops */
1744 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001745 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07001746 goto out_err;
1747
1748 /* See lock_rename() */
1749 ret = ERR_PTR(-EBUSY);
1750 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1751 goto out_err;
1752 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1753 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1754 goto out_err;
1755 m2 = &alias->d_parent->d_inode->i_mutex;
1756out_unalias:
1757 d_move_locked(alias, dentry);
1758 ret = alias;
1759out_err:
1760 spin_unlock(&dcache_lock);
1761 if (m2)
1762 mutex_unlock(m2);
1763 if (m1)
1764 mutex_unlock(m1);
1765 return ret;
1766}
1767
1768/*
David Howells770bfad2006-08-22 20:06:07 -04001769 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1770 * named dentry in place of the dentry to be replaced.
1771 */
1772static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1773{
1774 struct dentry *dparent, *aparent;
1775
1776 switch_names(dentry, anon);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001777 swap(dentry->d_name.hash, anon->d_name.hash);
David Howells770bfad2006-08-22 20:06:07 -04001778
1779 dparent = dentry->d_parent;
1780 aparent = anon->d_parent;
1781
1782 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1783 list_del(&dentry->d_u.d_child);
1784 if (!IS_ROOT(dentry))
1785 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1786 else
1787 INIT_LIST_HEAD(&dentry->d_u.d_child);
1788
1789 anon->d_parent = (dparent == dentry) ? anon : dparent;
1790 list_del(&anon->d_u.d_child);
1791 if (!IS_ROOT(anon))
1792 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1793 else
1794 INIT_LIST_HEAD(&anon->d_u.d_child);
1795
1796 anon->d_flags &= ~DCACHE_DISCONNECTED;
1797}
1798
1799/**
1800 * d_materialise_unique - introduce an inode into the tree
1801 * @dentry: candidate dentry
1802 * @inode: inode to bind to the dentry, to which aliases may be attached
1803 *
1804 * Introduces an dentry into the tree, substituting an extant disconnected
1805 * root directory alias in its place if there is one
1806 */
1807struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1808{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001809 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001810
1811 BUG_ON(!d_unhashed(dentry));
1812
1813 spin_lock(&dcache_lock);
1814
1815 if (!inode) {
1816 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001817 __d_instantiate(dentry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001818 goto found_lock;
1819 }
1820
Trond Myklebust9eaef272006-10-21 10:24:20 -07001821 if (S_ISDIR(inode->i_mode)) {
1822 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001823
Trond Myklebust9eaef272006-10-21 10:24:20 -07001824 /* Does an aliased dentry already exist? */
1825 alias = __d_find_alias(inode, 0);
1826 if (alias) {
1827 actual = alias;
1828 /* Is this an anonymous mountpoint that we could splice
1829 * into our tree? */
1830 if (IS_ROOT(alias)) {
1831 spin_lock(&alias->d_lock);
1832 __d_materialise_dentry(dentry, alias);
1833 __d_drop(alias);
1834 goto found;
1835 }
1836 /* Nope, but we must(!) avoid directory aliasing */
1837 actual = __d_unalias(dentry, alias);
1838 if (IS_ERR(actual))
1839 dput(alias);
1840 goto out_nolock;
1841 }
David Howells770bfad2006-08-22 20:06:07 -04001842 }
1843
1844 /* Add a unique reference */
1845 actual = __d_instantiate_unique(dentry, inode);
1846 if (!actual)
1847 actual = dentry;
1848 else if (unlikely(!d_unhashed(actual)))
1849 goto shouldnt_be_hashed;
1850
1851found_lock:
1852 spin_lock(&actual->d_lock);
1853found:
1854 _d_rehash(actual);
1855 spin_unlock(&actual->d_lock);
1856 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001857out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001858 if (actual == dentry) {
1859 security_d_instantiate(dentry, inode);
1860 return NULL;
1861 }
1862
1863 iput(inode);
1864 return actual;
1865
David Howells770bfad2006-08-22 20:06:07 -04001866shouldnt_be_hashed:
1867 spin_unlock(&dcache_lock);
1868 BUG();
David Howells770bfad2006-08-22 20:06:07 -04001869}
1870
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001871static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01001872{
1873 *buflen -= namelen;
1874 if (*buflen < 0)
1875 return -ENAMETOOLONG;
1876 *buffer -= namelen;
1877 memcpy(*buffer, str, namelen);
1878 return 0;
1879}
1880
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001881static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1882{
1883 return prepend(buffer, buflen, name->name, name->len);
1884}
1885
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001887 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001888 * @path: the dentry/vfsmount to report
1889 * @root: root vfsmnt/dentry (may be modified by this function)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 * @buffer: buffer to return value in
1891 * @buflen: buffer length
1892 *
Linus Torvalds552ce542007-02-13 12:08:18 -08001893 * Convert a dentry into an ASCII path name. If the entry has been deleted
1894 * the string " (deleted)" is appended. Note that this is ambiguous.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08001896 * Returns a pointer into the buffer or an error code if the
1897 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08001898 *
1899 * "buflen" should be positive. Caller holds the dcache_lock.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001900 *
1901 * If path is not reachable from the supplied root, then the value of
1902 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001904char *__d_path(const struct path *path, struct path *root,
1905 char *buffer, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906{
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001907 struct dentry *dentry = path->dentry;
1908 struct vfsmount *vfsmnt = path->mnt;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001909 char *end = buffer + buflen;
1910 char *retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001912 spin_lock(&vfsmount_lock);
Ram Pai6092d042008-03-27 13:06:20 +01001913 prepend(&end, &buflen, "\0", 1);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001914 if (d_unlinked(dentry) &&
Ram Pai6092d042008-03-27 13:06:20 +01001915 (prepend(&end, &buflen, " (deleted)", 10) != 0))
Linus Torvalds552ce542007-02-13 12:08:18 -08001916 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001917
1918 if (buflen < 1)
1919 goto Elong;
1920 /* Get '/' right */
1921 retval = end-1;
1922 *retval = '/';
1923
1924 for (;;) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 struct dentry * parent;
1926
Jan Blunck329c97f2008-02-14 19:38:31 -08001927 if (dentry == root->dentry && vfsmnt == root->mnt)
Linus Torvalds552ce542007-02-13 12:08:18 -08001928 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08001930 /* Global root? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 if (vfsmnt->mnt_parent == vfsmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 goto global_root;
1933 }
1934 dentry = vfsmnt->mnt_mountpoint;
1935 vfsmnt = vfsmnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 continue;
1937 }
1938 parent = dentry->d_parent;
1939 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001940 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01001941 (prepend(&end, &buflen, "/", 1) != 0))
Linus Torvalds552ce542007-02-13 12:08:18 -08001942 goto Elong;
Linus Torvalds552ce542007-02-13 12:08:18 -08001943 retval = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 dentry = parent;
1945 }
1946
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001947out:
1948 spin_unlock(&vfsmount_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08001949 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951global_root:
Ram Pai6092d042008-03-27 13:06:20 +01001952 retval += 1; /* hit the slash */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001953 if (prepend_name(&retval, &buflen, &dentry->d_name) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 goto Elong;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001955 root->mnt = vfsmnt;
1956 root->dentry = dentry;
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001957 goto out;
1958
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959Elong:
Andreas Gruenbacherbe285c72008-06-16 13:28:07 +02001960 retval = ERR_PTR(-ENAMETOOLONG);
1961 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962}
1963
Jan Bluncka03a8a702008-02-14 19:38:32 -08001964/**
1965 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08001966 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08001967 * @buf: buffer to return value in
1968 * @buflen: buffer length
1969 *
1970 * Convert a dentry into an ASCII path name. If the entry has been deleted
1971 * the string " (deleted)" is appended. Note that this is ambiguous.
1972 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08001973 * Returns a pointer into the buffer or an error code if the path was
1974 * too long. Note: Callers should use the returned pointer, not the passed
1975 * in buffer, to use the name! The implementation often starts at an offset
1976 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08001977 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001978 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08001979 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07001980char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981{
1982 char *res;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001983 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01001984 struct path tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001986 /*
1987 * We have various synthetic filesystems that never get mounted. On
1988 * these filesystems dentries are never used for lookup purposes, and
1989 * thus don't need to be hashed. They also don't need a name until a
1990 * user wants to identify the object in /proc/pid/fd/. The little hack
1991 * below allows us to generate a name for these objects on demand:
1992 */
Jan Blunckcf28b482008-02-14 19:38:44 -08001993 if (path->dentry->d_op && path->dentry->d_op->d_dname)
1994 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07001995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001997 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01001998 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 read_unlock(&current->fs->lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002000 spin_lock(&dcache_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002001 tmp = root;
2002 res = __d_path(path, &tmp, buf, buflen);
Linus Torvalds552ce542007-02-13 12:08:18 -08002003 spin_unlock(&dcache_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002004 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 return res;
2006}
2007
2008/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002009 * Helper function for dentry_operations.d_dname() members
2010 */
2011char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2012 const char *fmt, ...)
2013{
2014 va_list args;
2015 char temp[64];
2016 int sz;
2017
2018 va_start(args, fmt);
2019 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2020 va_end(args);
2021
2022 if (sz > sizeof(temp) || sz > buflen)
2023 return ERR_PTR(-ENAMETOOLONG);
2024
2025 buffer += buflen - sz;
2026 return memcpy(buffer, temp, sz);
2027}
2028
2029/*
Ram Pai6092d042008-03-27 13:06:20 +01002030 * Write full pathname from the root of the filesystem into the buffer.
2031 */
2032char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2033{
2034 char *end = buf + buflen;
2035 char *retval;
2036
2037 spin_lock(&dcache_lock);
2038 prepend(&end, &buflen, "\0", 1);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002039 if (d_unlinked(dentry) &&
Ram Pai6092d042008-03-27 13:06:20 +01002040 (prepend(&end, &buflen, "//deleted", 9) != 0))
2041 goto Elong;
2042 if (buflen < 1)
2043 goto Elong;
2044 /* Get '/' right */
2045 retval = end-1;
2046 *retval = '/';
2047
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002048 while (!IS_ROOT(dentry)) {
2049 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01002050
Ram Pai6092d042008-03-27 13:06:20 +01002051 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002052 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01002053 (prepend(&end, &buflen, "/", 1) != 0))
2054 goto Elong;
2055
2056 retval = end;
2057 dentry = parent;
2058 }
2059 spin_unlock(&dcache_lock);
2060 return retval;
2061Elong:
2062 spin_unlock(&dcache_lock);
2063 return ERR_PTR(-ENAMETOOLONG);
2064}
2065
2066/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 * NOTE! The user-level library version returns a
2068 * character pointer. The kernel system call just
2069 * returns the length of the buffer filled (which
2070 * includes the ending '\0' character), or a negative
2071 * error value. So libc would do something like
2072 *
2073 * char *getcwd(char * buf, size_t size)
2074 * {
2075 * int retval;
2076 *
2077 * retval = sys_getcwd(buf, size);
2078 * if (retval >= 0)
2079 * return buf;
2080 * errno = -retval;
2081 * return NULL;
2082 * }
2083 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002084SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085{
Linus Torvalds552ce542007-02-13 12:08:18 -08002086 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002087 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002088 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
2090 if (!page)
2091 return -ENOMEM;
2092
2093 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002094 pwd = current->fs->pwd;
Ram Pai6092d042008-03-27 13:06:20 +01002095 path_get(&pwd);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002096 root = current->fs->root;
Ram Pai6092d042008-03-27 13:06:20 +01002097 path_get(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 read_unlock(&current->fs->lock);
2099
Linus Torvalds552ce542007-02-13 12:08:18 -08002100 error = -ENOENT;
Linus Torvalds552ce542007-02-13 12:08:18 -08002101 spin_lock(&dcache_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002102 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002103 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002104 struct path tmp = root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002105 char * cwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002107 cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE);
Linus Torvalds552ce542007-02-13 12:08:18 -08002108 spin_unlock(&dcache_lock);
2109
2110 error = PTR_ERR(cwd);
2111 if (IS_ERR(cwd))
2112 goto out;
2113
2114 error = -ERANGE;
2115 len = PAGE_SIZE + page - cwd;
2116 if (len <= size) {
2117 error = len;
2118 if (copy_to_user(buf, cwd, len))
2119 error = -EFAULT;
2120 }
2121 } else
2122 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
2124out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002125 path_put(&pwd);
2126 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 free_page((unsigned long) page);
2128 return error;
2129}
2130
2131/*
2132 * Test whether new_dentry is a subdirectory of old_dentry.
2133 *
2134 * Trivially implemented using the dcache structure
2135 */
2136
2137/**
2138 * is_subdir - is new dentry a subdirectory of old_dentry
2139 * @new_dentry: new dentry
2140 * @old_dentry: old dentry
2141 *
2142 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2143 * Returns 0 otherwise.
2144 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2145 */
2146
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002147int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148{
2149 int result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 unsigned long seq;
2151
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002152 if (new_dentry == old_dentry)
2153 return 1;
2154
2155 /*
2156 * Need rcu_readlock to protect against the d_parent trashing
2157 * due to d_move
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 */
2159 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002160 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 seq = read_seqbegin(&rename_lock);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002163 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002165 else
2166 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 } while (read_seqretry(&rename_lock, seq));
2168 rcu_read_unlock();
2169
2170 return result;
2171}
2172
2173void d_genocide(struct dentry *root)
2174{
2175 struct dentry *this_parent = root;
2176 struct list_head *next;
2177
2178 spin_lock(&dcache_lock);
2179repeat:
2180 next = this_parent->d_subdirs.next;
2181resume:
2182 while (next != &this_parent->d_subdirs) {
2183 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002184 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 next = tmp->next;
2186 if (d_unhashed(dentry)||!dentry->d_inode)
2187 continue;
2188 if (!list_empty(&dentry->d_subdirs)) {
2189 this_parent = dentry;
2190 goto repeat;
2191 }
2192 atomic_dec(&dentry->d_count);
2193 }
2194 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002195 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 atomic_dec(&this_parent->d_count);
2197 this_parent = this_parent->d_parent;
2198 goto resume;
2199 }
2200 spin_unlock(&dcache_lock);
2201}
2202
2203/**
2204 * find_inode_number - check for dentry with name
2205 * @dir: directory to check
2206 * @name: Name to find.
2207 *
2208 * Check whether a dentry already exists for the given name,
2209 * and return the inode number if it has an inode. Otherwise
2210 * 0 is returned.
2211 *
2212 * This routine is used to post-process directory listings for
2213 * filesystems using synthetic inode numbers, and is necessary
2214 * to keep getcwd() working.
2215 */
2216
2217ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2218{
2219 struct dentry * dentry;
2220 ino_t ino = 0;
2221
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002222 dentry = d_hash_and_lookup(dir, name);
2223 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 if (dentry->d_inode)
2225 ino = dentry->d_inode->i_ino;
2226 dput(dentry);
2227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 return ino;
2229}
2230
2231static __initdata unsigned long dhash_entries;
2232static int __init set_dhash_entries(char *str)
2233{
2234 if (!str)
2235 return 0;
2236 dhash_entries = simple_strtoul(str, &str, 0);
2237 return 1;
2238}
2239__setup("dhash_entries=", set_dhash_entries);
2240
2241static void __init dcache_init_early(void)
2242{
2243 int loop;
2244
2245 /* If hashes are distributed across NUMA nodes, defer
2246 * hash allocation until vmalloc space is available.
2247 */
2248 if (hashdist)
2249 return;
2250
2251 dentry_hashtable =
2252 alloc_large_system_hash("Dentry cache",
2253 sizeof(struct hlist_head),
2254 dhash_entries,
2255 13,
2256 HASH_EARLY,
2257 &d_hash_shift,
2258 &d_hash_mask,
2259 0);
2260
2261 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2262 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2263}
2264
Denis Cheng74bf17c2007-10-16 23:26:30 -07002265static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266{
2267 int loop;
2268
2269 /*
2270 * A constructor could be added for stable state like the lists,
2271 * but it is probably not worth it because of the cache nature
2272 * of the dcache.
2273 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002274 dentry_cache = KMEM_CACHE(dentry,
2275 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Rusty Russell8e1f9362007-07-17 04:03:17 -07002277 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278
2279 /* Hash may have been set up in dcache_init_early */
2280 if (!hashdist)
2281 return;
2282
2283 dentry_hashtable =
2284 alloc_large_system_hash("Dentry cache",
2285 sizeof(struct hlist_head),
2286 dhash_entries,
2287 13,
2288 0,
2289 &d_hash_shift,
2290 &d_hash_mask,
2291 0);
2292
2293 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2294 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2295}
2296
2297/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002298struct kmem_cache *names_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300EXPORT_SYMBOL(d_genocide);
2301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302void __init vfs_caches_init_early(void)
2303{
2304 dcache_init_early();
2305 inode_init_early();
2306}
2307
2308void __init vfs_caches_init(unsigned long mempages)
2309{
2310 unsigned long reserve;
2311
2312 /* Base hash sizes on available memory, with a reserve equal to
2313 150% of current kernel size */
2314
2315 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2316 mempages -= reserve;
2317
2318 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002319 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320
Denis Cheng74bf17c2007-10-16 23:26:30 -07002321 dcache_init();
2322 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002324 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 bdev_cache_init();
2326 chrdev_init();
2327}
2328
2329EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330EXPORT_SYMBOL(d_alloc_root);
2331EXPORT_SYMBOL(d_delete);
2332EXPORT_SYMBOL(d_find_alias);
2333EXPORT_SYMBOL(d_instantiate);
2334EXPORT_SYMBOL(d_invalidate);
2335EXPORT_SYMBOL(d_lookup);
2336EXPORT_SYMBOL(d_move);
David Howells770bfad2006-08-22 20:06:07 -04002337EXPORT_SYMBOL_GPL(d_materialise_unique);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338EXPORT_SYMBOL(d_path);
2339EXPORT_SYMBOL(d_prune_aliases);
2340EXPORT_SYMBOL(d_rehash);
2341EXPORT_SYMBOL(d_splice_alias);
Barry Naujok94035402008-05-21 16:50:46 +10002342EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343EXPORT_SYMBOL(d_validate);
2344EXPORT_SYMBOL(dget_locked);
2345EXPORT_SYMBOL(dput);
2346EXPORT_SYMBOL(find_inode_number);
2347EXPORT_SYMBOL(have_submounts);
2348EXPORT_SYMBOL(names_cachep);
2349EXPORT_SYMBOL(shrink_dcache_parent);
2350EXPORT_SYMBOL(shrink_dcache_sb);