blob: cc2b938021796c0a8d4492030043309f087f790e [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
Christoph Hellwig312d3ca2010-10-10 05:36:23 -040070static struct percpu_counter nr_dentry __cacheline_aligned_in_smp;
71static struct percpu_counter nr_dentry_unused __cacheline_aligned_in_smp;
72
73#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
74int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
75 size_t *lenp, loff_t *ppos)
76{
77 dentry_stat.nr_dentry = percpu_counter_sum_positive(&nr_dentry);
78 dentry_stat.nr_unused = percpu_counter_sum_positive(&nr_dentry_unused);
79 return proc_dointvec(table, write, buffer, lenp, ppos);
80}
81#endif
82
Christoph Hellwig9c82ab92010-10-10 05:36:22 -040083static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -040085 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
86
Arjan van de Venfd217f42008-10-21 06:47:33 -070087 WARN_ON(!list_empty(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 if (dname_external(dentry))
89 kfree(dentry->d_name.name);
90 kmem_cache_free(dentry_cache, dentry);
91}
92
93/*
Christoph Hellwig312d3ca2010-10-10 05:36:23 -040094 * no dcache_lock, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 */
96static void d_free(struct dentry *dentry)
97{
Christoph Hellwig312d3ca2010-10-10 05:36:23 -040098 percpu_counter_dec(&nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 if (dentry->d_op && dentry->d_op->d_release)
100 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400101
Eric Dumazetb3423412006-12-06 20:38:48 -0800102 /* if dentry was never inserted into hash, immediate free is OK */
Akinobu Mitae8462ca2008-02-06 01:37:07 -0800103 if (hlist_unhashed(&dentry->d_hash))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400104 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800105 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400106 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107}
108
109/*
110 * Release the dentry's inode, using the filesystem
111 * d_iput() operation if defined.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800113static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200114 __releases(dentry->d_lock)
115 __releases(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
117 struct inode *inode = dentry->d_inode;
118 if (inode) {
119 dentry->d_inode = NULL;
120 list_del_init(&dentry->d_alias);
121 spin_unlock(&dentry->d_lock);
122 spin_unlock(&dcache_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700123 if (!inode->i_nlink)
124 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (dentry->d_op && dentry->d_op->d_iput)
126 dentry->d_op->d_iput(dentry, inode);
127 else
128 iput(inode);
129 } else {
130 spin_unlock(&dentry->d_lock);
131 spin_unlock(&dcache_lock);
132 }
133}
134
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700135/*
Christoph Hellwiga4633352010-10-10 05:36:26 -0400136 * dentry_lru_(add|del|move_tail) must be called with dcache_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700137 */
138static void dentry_lru_add(struct dentry *dentry)
139{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400140 if (list_empty(&dentry->d_lru)) {
141 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
142 dentry->d_sb->s_nr_dentry_unused++;
143 percpu_counter_inc(&nr_dentry_unused);
144 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700145}
146
147static void dentry_lru_del(struct dentry *dentry)
148{
149 if (!list_empty(&dentry->d_lru)) {
Christoph Hellwiga4633352010-10-10 05:36:26 -0400150 list_del_init(&dentry->d_lru);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700151 dentry->d_sb->s_nr_dentry_unused--;
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400152 percpu_counter_dec(&nr_dentry_unused);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700153 }
154}
155
Christoph Hellwiga4633352010-10-10 05:36:26 -0400156static void dentry_lru_move_tail(struct dentry *dentry)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700157{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400158 if (list_empty(&dentry->d_lru)) {
159 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
160 dentry->d_sb->s_nr_dentry_unused++;
161 percpu_counter_inc(&nr_dentry_unused);
162 } else {
163 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700164 }
165}
166
Miklos Szeredid52b9082007-05-08 00:23:46 -0700167/**
168 * d_kill - kill dentry and return parent
169 * @dentry: dentry to kill
170 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200171 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700172 *
173 * If this is the root of the dentry tree, return NULL.
174 */
175static struct dentry *d_kill(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200176 __releases(dentry->d_lock)
177 __releases(dcache_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700178{
179 struct dentry *parent;
180
181 list_del(&dentry->d_u.d_child);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700182 /*drops the locks, at that point nobody can reach this dentry */
183 dentry_iput(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900184 if (IS_ROOT(dentry))
185 parent = NULL;
186 else
187 parent = dentry->d_parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700188 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900189 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/*
193 * This is dput
194 *
195 * This is complicated by the fact that we do not want to put
196 * dentries that are no longer on any hash chain on the unused
197 * list: we'd much rather just get rid of them immediately.
198 *
199 * However, that implies that we have to traverse the dentry
200 * tree upwards to the parents which might _also_ now be
201 * scheduled for deletion (it may have been only waiting for
202 * its last child to go away).
203 *
204 * This tail recursion is done by hand as we don't want to depend
205 * on the compiler to always get this right (gcc generally doesn't).
206 * Real recursion would eat up our stack space.
207 */
208
209/*
210 * dput - release a dentry
211 * @dentry: dentry to release
212 *
213 * Release a dentry. This will drop the usage count and if appropriate
214 * call the dentry unlink method as well as removing it from the queues and
215 * releasing its resources. If the parent dentries were scheduled for release
216 * they too may now get deleted.
217 *
218 * no dcache lock, please.
219 */
220
221void dput(struct dentry *dentry)
222{
223 if (!dentry)
224 return;
225
226repeat:
227 if (atomic_read(&dentry->d_count) == 1)
228 might_sleep();
229 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
230 return;
231
232 spin_lock(&dentry->d_lock);
233 if (atomic_read(&dentry->d_count)) {
234 spin_unlock(&dentry->d_lock);
235 spin_unlock(&dcache_lock);
236 return;
237 }
238
239 /*
240 * AV: ->d_delete() is _NOT_ allowed to block now.
241 */
242 if (dentry->d_op && dentry->d_op->d_delete) {
243 if (dentry->d_op->d_delete(dentry))
244 goto unhash_it;
245 }
Nick Piggin265ac902010-10-10 05:36:24 -0400246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 /* Unreachable? Get rid of it */
248 if (d_unhashed(dentry))
249 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400250
251 /* Otherwise leave it cached and ensure it's on the LRU */
252 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400253 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 spin_unlock(&dentry->d_lock);
256 spin_unlock(&dcache_lock);
257 return;
258
259unhash_it:
260 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700261kill_it:
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700262 /* if dentry was on the d_lru list delete it from there */
263 dentry_lru_del(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700264 dentry = d_kill(dentry);
265 if (dentry)
266 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700268EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270/**
271 * d_invalidate - invalidate a dentry
272 * @dentry: dentry to invalidate
273 *
274 * Try to invalidate the dentry if it turns out to be
275 * possible. If there are other dentries that can be
276 * reached through this one we can't delete it and we
277 * return -EBUSY. On success we return 0.
278 *
279 * no dcache lock.
280 */
281
282int d_invalidate(struct dentry * dentry)
283{
284 /*
285 * If it's already been dropped, return OK.
286 */
287 spin_lock(&dcache_lock);
288 if (d_unhashed(dentry)) {
289 spin_unlock(&dcache_lock);
290 return 0;
291 }
292 /*
293 * Check whether to do a partial shrink_dcache
294 * to get rid of unused child entries.
295 */
296 if (!list_empty(&dentry->d_subdirs)) {
297 spin_unlock(&dcache_lock);
298 shrink_dcache_parent(dentry);
299 spin_lock(&dcache_lock);
300 }
301
302 /*
303 * Somebody else still using it?
304 *
305 * If it's a directory, we can't drop it
306 * for fear of somebody re-populating it
307 * with children (even though dropping it
308 * would make it unreachable from the root,
309 * we might still populate it if it was a
310 * working directory or similar).
311 */
312 spin_lock(&dentry->d_lock);
313 if (atomic_read(&dentry->d_count) > 1) {
314 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
315 spin_unlock(&dentry->d_lock);
316 spin_unlock(&dcache_lock);
317 return -EBUSY;
318 }
319 }
320
321 __d_drop(dentry);
322 spin_unlock(&dentry->d_lock);
323 spin_unlock(&dcache_lock);
324 return 0;
325}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700326EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328/* This should be called _only_ with dcache_lock held */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329static inline struct dentry * __dget_locked(struct dentry *dentry)
330{
331 atomic_inc(&dentry->d_count);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400332 dentry_lru_del(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 return dentry;
334}
335
336struct dentry * dget_locked(struct dentry *dentry)
337{
338 return __dget_locked(dentry);
339}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700340EXPORT_SYMBOL(dget_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
342/**
343 * d_find_alias - grab a hashed alias of inode
344 * @inode: inode in question
345 * @want_discon: flag, used by d_splice_alias, to request
346 * that only a DISCONNECTED alias be returned.
347 *
348 * If inode has a hashed alias, or is a directory and has any alias,
349 * acquire the reference to alias and return it. Otherwise return NULL.
350 * Notice that if inode is a directory there can be only one alias and
351 * it can be unhashed only if it has no children, or if it is the root
352 * of a filesystem.
353 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700354 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700356 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 */
358
359static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
360{
361 struct list_head *head, *next, *tmp;
362 struct dentry *alias, *discon_alias=NULL;
363
364 head = &inode->i_dentry;
365 next = inode->i_dentry.next;
366 while (next != head) {
367 tmp = next;
368 next = tmp->next;
369 prefetch(next);
370 alias = list_entry(tmp, struct dentry, d_alias);
371 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700372 if (IS_ROOT(alias) &&
373 (alias->d_flags & DCACHE_DISCONNECTED))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 discon_alias = alias;
375 else if (!want_discon) {
376 __dget_locked(alias);
377 return alias;
378 }
379 }
380 }
381 if (discon_alias)
382 __dget_locked(discon_alias);
383 return discon_alias;
384}
385
386struct dentry * d_find_alias(struct inode *inode)
387{
David Howells214fda12006-03-25 03:06:36 -0800388 struct dentry *de = NULL;
389
390 if (!list_empty(&inode->i_dentry)) {
391 spin_lock(&dcache_lock);
392 de = __d_find_alias(inode, 0);
393 spin_unlock(&dcache_lock);
394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 return de;
396}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700397EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399/*
400 * Try to kill dentries associated with this inode.
401 * WARNING: you must own a reference to inode.
402 */
403void d_prune_aliases(struct inode *inode)
404{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700405 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406restart:
407 spin_lock(&dcache_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700408 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 spin_lock(&dentry->d_lock);
410 if (!atomic_read(&dentry->d_count)) {
411 __dget_locked(dentry);
412 __d_drop(dentry);
413 spin_unlock(&dentry->d_lock);
414 spin_unlock(&dcache_lock);
415 dput(dentry);
416 goto restart;
417 }
418 spin_unlock(&dentry->d_lock);
419 }
420 spin_unlock(&dcache_lock);
421}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700422EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
424/*
Andrew Mortond702ccb2006-06-22 14:47:31 -0700425 * Throw away a dentry - free the inode, dput the parent. This requires that
426 * the LRU list has already been removed.
427 *
Miklos Szeredi85864e12007-10-16 23:27:09 -0700428 * Try to prune ancestors as well. This is necessary to prevent
429 * quadratic behavior of shrink_dcache_parent(), but is also expected
430 * to be beneficial in reducing dentry cache fragmentation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 */
Miklos Szeredi85864e12007-10-16 23:27:09 -0700432static void prune_one_dentry(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200433 __releases(dentry->d_lock)
434 __releases(dcache_lock)
435 __acquires(dcache_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 __d_drop(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700438 dentry = d_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700439
440 /*
441 * Prune ancestors. Locking is simpler than in dput(),
442 * because dcache_lock needs to be taken anyway.
443 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 spin_lock(&dcache_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700445 while (dentry) {
446 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
447 return;
448
449 if (dentry->d_op && dentry->d_op->d_delete)
450 dentry->d_op->d_delete(dentry);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400451 dentry_lru_del(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700452 __d_drop(dentry);
453 dentry = d_kill(dentry);
454 spin_lock(&dcache_lock);
455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400458static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700460 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700461
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400462 while (!list_empty(list)) {
463 dentry = list_entry(list->prev, struct dentry, d_lru);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400464 dentry_lru_del(dentry);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400465
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 /*
467 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700468 * the LRU because of laziness during lookup. Do not free
469 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400471 spin_lock(&dentry->d_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700472 if (atomic_read(&dentry->d_count)) {
473 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 continue;
475 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700476 prune_one_dentry(dentry);
477 /* dentry->d_lock was dropped in prune_one_dentry() */
478 cond_resched_lock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400480}
481
482/**
483 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
484 * @sb: superblock to shrink dentry LRU.
485 * @count: number of entries to prune
486 * @flags: flags to control the dentry processing
487 *
488 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
489 */
490static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
491{
492 /* called from prune_dcache() and shrink_dcache_parent() */
493 struct dentry *dentry;
494 LIST_HEAD(referenced);
495 LIST_HEAD(tmp);
496 int cnt = *count;
497
498 spin_lock(&dcache_lock);
499 while (!list_empty(&sb->s_dentry_lru)) {
500 dentry = list_entry(sb->s_dentry_lru.prev,
501 struct dentry, d_lru);
502 BUG_ON(dentry->d_sb != sb);
503
504 /*
505 * If we are honouring the DCACHE_REFERENCED flag and the
506 * dentry has this flag set, don't free it. Clear the flag
507 * and put it back on the LRU.
508 */
509 if (flags & DCACHE_REFERENCED) {
510 spin_lock(&dentry->d_lock);
511 if (dentry->d_flags & DCACHE_REFERENCED) {
512 dentry->d_flags &= ~DCACHE_REFERENCED;
513 list_move(&dentry->d_lru, &referenced);
514 spin_unlock(&dentry->d_lock);
515 cond_resched_lock(&dcache_lock);
516 continue;
517 }
518 spin_unlock(&dentry->d_lock);
519 }
520
521 list_move_tail(&dentry->d_lru, &tmp);
522 if (!--cnt)
523 break;
524 cond_resched_lock(&dcache_lock);
525 }
526
527 *count = cnt;
528 shrink_dentry_list(&tmp);
529
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700530 if (!list_empty(&referenced))
531 list_splice(&referenced, &sb->s_dentry_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 spin_unlock(&dcache_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400533
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534}
535
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700536/**
537 * prune_dcache - shrink the dcache
538 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700540 * Shrink the dcache. This is done when we need more memory, or simply when we
541 * need to unmount something (at which point we need to unuse all dentries).
542 *
543 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700545static void prune_dcache(int count)
546{
Al Virodca33252010-07-25 02:31:46 +0400547 struct super_block *sb, *p = NULL;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700548 int w_count;
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400549 int unused = percpu_counter_sum_positive(&nr_dentry_unused);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700550 int prune_ratio;
551 int pruned;
552
553 if (unused == 0 || count == 0)
554 return;
555 spin_lock(&dcache_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700556 if (count >= unused)
557 prune_ratio = 1;
558 else
559 prune_ratio = unused / count;
560 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400561 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400562 if (list_empty(&sb->s_instances))
563 continue;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700564 if (sb->s_nr_dentry_unused == 0)
565 continue;
566 sb->s_count++;
567 /* Now, we reclaim unused dentrins with fairness.
568 * We reclaim them same percentage from each superblock.
569 * We calculate number of dentries to scan on this sb
570 * as follows, but the implementation is arranged to avoid
571 * overflows:
572 * number of dentries to scan on this sb =
573 * count * (number of dentries on this sb /
574 * number of dentries in the machine)
575 */
576 spin_unlock(&sb_lock);
577 if (prune_ratio != 1)
578 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
579 else
580 w_count = sb->s_nr_dentry_unused;
581 pruned = w_count;
582 /*
583 * We need to be sure this filesystem isn't being unmounted,
584 * otherwise we could race with generic_shutdown_super(), and
585 * end up holding a reference to an inode while the filesystem
586 * is unmounted. So we try to get s_umount, and make sure
587 * s_root isn't NULL.
588 */
589 if (down_read_trylock(&sb->s_umount)) {
590 if ((sb->s_root != NULL) &&
591 (!list_empty(&sb->s_dentry_lru))) {
592 spin_unlock(&dcache_lock);
593 __shrink_dcache_sb(sb, &w_count,
594 DCACHE_REFERENCED);
595 pruned -= w_count;
596 spin_lock(&dcache_lock);
597 }
598 up_read(&sb->s_umount);
599 }
600 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400601 if (p)
602 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700603 count -= pruned;
Al Virodca33252010-07-25 02:31:46 +0400604 p = sb;
Al Viro79893c12010-03-22 20:27:55 -0400605 /* more work left to do? */
606 if (count <= 0)
607 break;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700608 }
Al Virodca33252010-07-25 02:31:46 +0400609 if (p)
610 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700611 spin_unlock(&sb_lock);
612 spin_unlock(&dcache_lock);
613}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615/**
616 * shrink_dcache_sb - shrink dcache for a superblock
617 * @sb: superblock
618 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400619 * Shrink the dcache for the specified super block. This is used to free
620 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400622void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400624 LIST_HEAD(tmp);
625
626 spin_lock(&dcache_lock);
627 while (!list_empty(&sb->s_dentry_lru)) {
628 list_splice_init(&sb->s_dentry_lru, &tmp);
629 shrink_dentry_list(&tmp);
630 }
631 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700633EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635/*
David Howellsc636ebd2006-10-11 01:22:19 -0700636 * destroy a single subtree of dentries for unmount
637 * - see the comments on shrink_dcache_for_umount() for a description of the
638 * locking
639 */
640static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
641{
642 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700643 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700644
645 BUG_ON(!IS_ROOT(dentry));
646
647 /* detach this root from the system */
648 spin_lock(&dcache_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400649 dentry_lru_del(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700650 __d_drop(dentry);
651 spin_unlock(&dcache_lock);
652
653 for (;;) {
654 /* descend to the first leaf in the current subtree */
655 while (!list_empty(&dentry->d_subdirs)) {
656 struct dentry *loop;
657
658 /* this is a branch with children - detach all of them
659 * from the system in one go */
660 spin_lock(&dcache_lock);
661 list_for_each_entry(loop, &dentry->d_subdirs,
662 d_u.d_child) {
Christoph Hellwiga4633352010-10-10 05:36:26 -0400663 dentry_lru_del(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700664 __d_drop(loop);
665 cond_resched_lock(&dcache_lock);
666 }
667 spin_unlock(&dcache_lock);
668
669 /* move to the first child */
670 dentry = list_entry(dentry->d_subdirs.next,
671 struct dentry, d_u.d_child);
672 }
673
674 /* consume the dentries from this leaf up through its parents
675 * until we find one with children or run out altogether */
676 do {
677 struct inode *inode;
678
679 if (atomic_read(&dentry->d_count) != 0) {
680 printk(KERN_ERR
681 "BUG: Dentry %p{i=%lx,n=%s}"
682 " still in use (%d)"
683 " [unmount of %s %s]\n",
684 dentry,
685 dentry->d_inode ?
686 dentry->d_inode->i_ino : 0UL,
687 dentry->d_name.name,
688 atomic_read(&dentry->d_count),
689 dentry->d_sb->s_type->name,
690 dentry->d_sb->s_id);
691 BUG();
692 }
693
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900694 if (IS_ROOT(dentry))
David Howellsc636ebd2006-10-11 01:22:19 -0700695 parent = NULL;
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900696 else {
697 parent = dentry->d_parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700698 atomic_dec(&parent->d_count);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900699 }
David Howellsc636ebd2006-10-11 01:22:19 -0700700
701 list_del(&dentry->d_u.d_child);
David Howellsf8713572006-10-28 10:38:46 -0700702 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700703
704 inode = dentry->d_inode;
705 if (inode) {
706 dentry->d_inode = NULL;
707 list_del_init(&dentry->d_alias);
708 if (dentry->d_op && dentry->d_op->d_iput)
709 dentry->d_op->d_iput(dentry, inode);
710 else
711 iput(inode);
712 }
713
714 d_free(dentry);
715
716 /* finished when we fall off the top of the tree,
717 * otherwise we ascend to the parent and move to the
718 * next sibling if there is one */
719 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400720 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700721 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700722 } while (list_empty(&dentry->d_subdirs));
723
724 dentry = list_entry(dentry->d_subdirs.next,
725 struct dentry, d_u.d_child);
726 }
727}
728
729/*
730 * destroy the dentries attached to a superblock on unmounting
731 * - we don't need to use dentry->d_lock, and only need dcache_lock when
732 * removing the dentry from the system lists and hashes because:
733 * - the superblock is detached from all mountings and open files, so the
734 * dentry trees will not be rearranged by the VFS
735 * - s_umount is write-locked, so the memory pressure shrinker will ignore
736 * any dentries belonging to this superblock that it comes across
737 * - the filesystem itself is no longer permitted to rearrange the dentries
738 * in this superblock
739 */
740void shrink_dcache_for_umount(struct super_block *sb)
741{
742 struct dentry *dentry;
743
744 if (down_read_trylock(&sb->s_umount))
745 BUG();
746
747 dentry = sb->s_root;
748 sb->s_root = NULL;
749 atomic_dec(&dentry->d_count);
750 shrink_dcache_for_umount_subtree(dentry);
751
752 while (!hlist_empty(&sb->s_anon)) {
753 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
754 shrink_dcache_for_umount_subtree(dentry);
755 }
756}
757
758/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 * Search for at least 1 mount point in the dentry's subdirs.
760 * We descend to the next level whenever the d_subdirs
761 * list is non-empty and continue searching.
762 */
763
764/**
765 * have_submounts - check for mounts over a dentry
766 * @parent: dentry to check.
767 *
768 * Return true if the parent or its subdirectories contain
769 * a mount point
770 */
771
772int have_submounts(struct dentry *parent)
773{
774 struct dentry *this_parent = parent;
775 struct list_head *next;
776
777 spin_lock(&dcache_lock);
778 if (d_mountpoint(parent))
779 goto positive;
780repeat:
781 next = this_parent->d_subdirs.next;
782resume:
783 while (next != &this_parent->d_subdirs) {
784 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800785 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 next = tmp->next;
787 /* Have we found a mount point ? */
788 if (d_mountpoint(dentry))
789 goto positive;
790 if (!list_empty(&dentry->d_subdirs)) {
791 this_parent = dentry;
792 goto repeat;
793 }
794 }
795 /*
796 * All done at this level ... ascend and resume the search.
797 */
798 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800799 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 this_parent = this_parent->d_parent;
801 goto resume;
802 }
803 spin_unlock(&dcache_lock);
804 return 0; /* No mount points found in tree */
805positive:
806 spin_unlock(&dcache_lock);
807 return 1;
808}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700809EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810
811/*
812 * Search the dentry child list for the specified parent,
813 * and move any unused dentries to the end of the unused
814 * list for prune_dcache(). We descend to the next level
815 * whenever the d_subdirs list is non-empty and continue
816 * searching.
817 *
818 * It returns zero iff there are no unused children,
819 * otherwise it returns the number of children moved to
820 * the end of the unused list. This may not be the total
821 * number of unused children, because select_parent can
822 * drop the lock and return early due to latency
823 * constraints.
824 */
825static int select_parent(struct dentry * parent)
826{
827 struct dentry *this_parent = parent;
828 struct list_head *next;
829 int found = 0;
830
831 spin_lock(&dcache_lock);
832repeat:
833 next = this_parent->d_subdirs.next;
834resume:
835 while (next != &this_parent->d_subdirs) {
836 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -0800837 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 next = tmp->next;
839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 /*
841 * move only zero ref count dentries to the end
842 * of the unused list for prune_dcache
843 */
844 if (!atomic_read(&dentry->d_count)) {
Christoph Hellwiga4633352010-10-10 05:36:26 -0400845 dentry_lru_move_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 found++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400847 } else {
848 dentry_lru_del(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
850
851 /*
852 * We can return to the caller if we have found some (this
853 * ensures forward progress). We'll be coming back to find
854 * the rest.
855 */
856 if (found && need_resched())
857 goto out;
858
859 /*
860 * Descend a level if the d_subdirs list is non-empty.
861 */
862 if (!list_empty(&dentry->d_subdirs)) {
863 this_parent = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 goto repeat;
865 }
866 }
867 /*
868 * All done at this level ... ascend and resume the search.
869 */
870 if (this_parent != parent) {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800871 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 this_parent = this_parent->d_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 goto resume;
874 }
875out:
876 spin_unlock(&dcache_lock);
877 return found;
878}
879
880/**
881 * shrink_dcache_parent - prune dcache
882 * @parent: parent of entries to prune
883 *
884 * Prune the dcache to remove unused children of the parent dentry.
885 */
886
887void shrink_dcache_parent(struct dentry * parent)
888{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700889 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 int found;
891
892 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700893 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700895EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897/*
898 * Scan `nr' dentries and return the number which remain.
899 *
900 * We need to avoid reentering the filesystem if the caller is performing a
901 * GFP_NOFS allocation attempt. One example deadlock is:
902 *
903 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
904 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
905 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
906 *
907 * In this case we return -1 to tell the caller that we baled.
908 */
Dave Chinner7f8275d2010-07-19 14:56:17 +1000909static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400911 int nr_unused;
912
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 if (nr) {
914 if (!(gfp_mask & __GFP_FS))
915 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700916 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 }
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400918
919 nr_unused = percpu_counter_sum_positive(&nr_dentry_unused);
920 return (nr_unused / 100) * sysctl_vfs_cache_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921}
922
Rusty Russell8e1f9362007-07-17 04:03:17 -0700923static struct shrinker dcache_shrinker = {
924 .shrink = shrink_dcache_memory,
925 .seeks = DEFAULT_SEEKS,
926};
927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928/**
929 * d_alloc - allocate a dcache entry
930 * @parent: parent of entry to allocate
931 * @name: qstr of the name
932 *
933 * Allocates a dentry. It returns %NULL if there is insufficient memory
934 * available. On a success the dentry is returned. The name passed in is
935 * copied and the copy passed in may be reused after this call.
936 */
937
938struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
939{
940 struct dentry *dentry;
941 char *dname;
942
Mel Gormane12ba742007-10-16 01:25:52 -0700943 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 if (!dentry)
945 return NULL;
946
947 if (name->len > DNAME_INLINE_LEN-1) {
948 dname = kmalloc(name->len + 1, GFP_KERNEL);
949 if (!dname) {
950 kmem_cache_free(dentry_cache, dentry);
951 return NULL;
952 }
953 } else {
954 dname = dentry->d_iname;
955 }
956 dentry->d_name.name = dname;
957
958 dentry->d_name.len = name->len;
959 dentry->d_name.hash = name->hash;
960 memcpy(dname, name->name, name->len);
961 dname[name->len] = 0;
962
963 atomic_set(&dentry->d_count, 1);
964 dentry->d_flags = DCACHE_UNHASHED;
965 spin_lock_init(&dentry->d_lock);
966 dentry->d_inode = NULL;
967 dentry->d_parent = NULL;
968 dentry->d_sb = NULL;
969 dentry->d_op = NULL;
970 dentry->d_fsdata = NULL;
971 dentry->d_mounted = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 INIT_HLIST_NODE(&dentry->d_hash);
973 INIT_LIST_HEAD(&dentry->d_lru);
974 INIT_LIST_HEAD(&dentry->d_subdirs);
975 INIT_LIST_HEAD(&dentry->d_alias);
976
977 if (parent) {
978 dentry->d_parent = dget(parent);
979 dentry->d_sb = parent->d_sb;
980 } else {
Eric Dumazet5160ee62006-01-08 01:03:32 -0800981 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 }
983
984 spin_lock(&dcache_lock);
985 if (parent)
Eric Dumazet5160ee62006-01-08 01:03:32 -0800986 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 spin_unlock(&dcache_lock);
988
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400989 percpu_counter_inc(&nr_dentry);
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 return dentry;
992}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700993EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995struct dentry *d_alloc_name(struct dentry *parent, const char *name)
996{
997 struct qstr q;
998
999 q.name = name;
1000 q.len = strlen(name);
1001 q.hash = full_name_hash(q.name, q.len);
1002 return d_alloc(parent, &q);
1003}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001004EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001006/* the caller must hold dcache_lock */
1007static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1008{
1009 if (inode)
1010 list_add(&dentry->d_alias, &inode->i_dentry);
1011 dentry->d_inode = inode;
1012 fsnotify_d_instantiate(dentry, inode);
1013}
1014
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015/**
1016 * d_instantiate - fill in inode information for a dentry
1017 * @entry: dentry to complete
1018 * @inode: inode to attach to this dentry
1019 *
1020 * Fill in inode information in the entry.
1021 *
1022 * This turns negative dentries into productive full members
1023 * of society.
1024 *
1025 * NOTE! This assumes that the inode count has been incremented
1026 * (or otherwise set) by the caller to indicate that it is now
1027 * in use by the dcache.
1028 */
1029
1030void d_instantiate(struct dentry *entry, struct inode * inode)
1031{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001032 BUG_ON(!list_empty(&entry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 spin_lock(&dcache_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001034 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 spin_unlock(&dcache_lock);
1036 security_d_instantiate(entry, inode);
1037}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001038EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040/**
1041 * d_instantiate_unique - instantiate a non-aliased dentry
1042 * @entry: dentry to instantiate
1043 * @inode: inode to attach to this dentry
1044 *
1045 * Fill in inode information in the entry. On success, it returns NULL.
1046 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001047 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 *
1049 * Note that in order to avoid conflicts with rename() etc, the caller
1050 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001051 *
1052 * This also assumes that the inode count has been incremented
1053 * (or otherwise set) by the caller to indicate that it is now
1054 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 */
David Howells770bfad2006-08-22 20:06:07 -04001056static struct dentry *__d_instantiate_unique(struct dentry *entry,
1057 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058{
1059 struct dentry *alias;
1060 int len = entry->d_name.len;
1061 const char *name = entry->d_name.name;
1062 unsigned int hash = entry->d_name.hash;
1063
David Howells770bfad2006-08-22 20:06:07 -04001064 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001065 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001066 return NULL;
1067 }
1068
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1070 struct qstr *qstr = &alias->d_name;
1071
1072 if (qstr->hash != hash)
1073 continue;
1074 if (alias->d_parent != entry->d_parent)
1075 continue;
1076 if (qstr->len != len)
1077 continue;
1078 if (memcmp(qstr->name, name, len))
1079 continue;
1080 dget_locked(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 return alias;
1082 }
David Howells770bfad2006-08-22 20:06:07 -04001083
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001084 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 return NULL;
1086}
David Howells770bfad2006-08-22 20:06:07 -04001087
1088struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1089{
1090 struct dentry *result;
1091
1092 BUG_ON(!list_empty(&entry->d_alias));
1093
1094 spin_lock(&dcache_lock);
1095 result = __d_instantiate_unique(entry, inode);
1096 spin_unlock(&dcache_lock);
1097
1098 if (!result) {
1099 security_d_instantiate(entry, inode);
1100 return NULL;
1101 }
1102
1103 BUG_ON(!d_unhashed(result));
1104 iput(inode);
1105 return result;
1106}
1107
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108EXPORT_SYMBOL(d_instantiate_unique);
1109
1110/**
1111 * d_alloc_root - allocate root dentry
1112 * @root_inode: inode to allocate the root for
1113 *
1114 * Allocate a root ("/") dentry for the inode given. The inode is
1115 * instantiated and returned. %NULL is returned if there is insufficient
1116 * memory or the inode passed is %NULL.
1117 */
1118
1119struct dentry * d_alloc_root(struct inode * root_inode)
1120{
1121 struct dentry *res = NULL;
1122
1123 if (root_inode) {
1124 static const struct qstr name = { .name = "/", .len = 1 };
1125
1126 res = d_alloc(NULL, &name);
1127 if (res) {
1128 res->d_sb = root_inode->i_sb;
1129 res->d_parent = res;
1130 d_instantiate(res, root_inode);
1131 }
1132 }
1133 return res;
1134}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001135EXPORT_SYMBOL(d_alloc_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137static inline struct hlist_head *d_hash(struct dentry *parent,
1138 unsigned long hash)
1139{
1140 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1141 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1142 return dentry_hashtable + (hash & D_HASHMASK);
1143}
1144
1145/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001146 * d_obtain_alias - find or allocate a dentry for a given inode
1147 * @inode: inode to allocate the dentry for
1148 *
1149 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1150 * similar open by handle operations. The returned dentry may be anonymous,
1151 * or may have a full name (if the inode was already in the cache).
1152 *
1153 * When called on a directory inode, we must ensure that the inode only ever
1154 * has one dentry. If a dentry is found, that is returned instead of
1155 * allocating a new one.
1156 *
1157 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001158 * to the dentry. In case of an error the reference on the inode is released.
1159 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1160 * be passed in and will be the error will be propagate to the return value,
1161 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001162 */
1163struct dentry *d_obtain_alias(struct inode *inode)
1164{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001165 static const struct qstr anonstring = { .name = "" };
1166 struct dentry *tmp;
1167 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001168
1169 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001170 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001171 if (IS_ERR(inode))
1172 return ERR_CAST(inode);
1173
Christoph Hellwig9308a612008-08-11 15:49:12 +02001174 res = d_find_alias(inode);
1175 if (res)
1176 goto out_iput;
1177
1178 tmp = d_alloc(NULL, &anonstring);
1179 if (!tmp) {
1180 res = ERR_PTR(-ENOMEM);
1181 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001182 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001183 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1184
1185 spin_lock(&dcache_lock);
1186 res = __d_find_alias(inode, 0);
1187 if (res) {
1188 spin_unlock(&dcache_lock);
1189 dput(tmp);
1190 goto out_iput;
1191 }
1192
1193 /* attach a disconnected dentry */
1194 spin_lock(&tmp->d_lock);
1195 tmp->d_sb = inode->i_sb;
1196 tmp->d_inode = inode;
1197 tmp->d_flags |= DCACHE_DISCONNECTED;
1198 tmp->d_flags &= ~DCACHE_UNHASHED;
1199 list_add(&tmp->d_alias, &inode->i_dentry);
1200 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1201 spin_unlock(&tmp->d_lock);
1202
1203 spin_unlock(&dcache_lock);
1204 return tmp;
1205
1206 out_iput:
1207 iput(inode);
1208 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001209}
Benny Halevyadc48722009-02-27 14:02:59 -08001210EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
1212/**
1213 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1214 * @inode: the inode which may have a disconnected dentry
1215 * @dentry: a negative dentry which we want to point to the inode.
1216 *
1217 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1218 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1219 * and return it, else simply d_add the inode to the dentry and return NULL.
1220 *
1221 * This is needed in the lookup routine of any filesystem that is exportable
1222 * (via knfsd) so that we can build dcache paths to directories effectively.
1223 *
1224 * If a dentry was found and moved, then it is returned. Otherwise NULL
1225 * is returned. This matches the expected return value of ->lookup.
1226 *
1227 */
1228struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1229{
1230 struct dentry *new = NULL;
1231
NeilBrown21c0d8f2006-10-04 02:16:16 -07001232 if (inode && S_ISDIR(inode->i_mode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 spin_lock(&dcache_lock);
1234 new = __d_find_alias(inode, 1);
1235 if (new) {
1236 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1237 spin_unlock(&dcache_lock);
1238 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 d_move(new, dentry);
1240 iput(inode);
1241 } else {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001242 /* already taking dcache_lock, so d_add() by hand */
1243 __d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 spin_unlock(&dcache_lock);
1245 security_d_instantiate(dentry, inode);
1246 d_rehash(dentry);
1247 }
1248 } else
1249 d_add(dentry, inode);
1250 return new;
1251}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001252EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Barry Naujok94035402008-05-21 16:50:46 +10001254/**
1255 * d_add_ci - lookup or allocate new dentry with case-exact name
1256 * @inode: the inode case-insensitive lookup has found
1257 * @dentry: the negative dentry that was passed to the parent's lookup func
1258 * @name: the case-exact name to be associated with the returned dentry
1259 *
1260 * This is to avoid filling the dcache with case-insensitive names to the
1261 * same inode, only the actual correct case is stored in the dcache for
1262 * case-insensitive filesystems.
1263 *
1264 * For a case-insensitive lookup match and if the the case-exact dentry
1265 * already exists in in the dcache, use it and return it.
1266 *
1267 * If no entry exists with the exact case name, allocate new dentry with
1268 * the exact case, and return the spliced entry.
1269 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001270struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001271 struct qstr *name)
1272{
1273 int error;
1274 struct dentry *found;
1275 struct dentry *new;
1276
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001277 /*
1278 * First check if a dentry matching the name already exists,
1279 * if not go ahead and create it now.
1280 */
Barry Naujok94035402008-05-21 16:50:46 +10001281 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001282 if (!found) {
1283 new = d_alloc(dentry->d_parent, name);
1284 if (!new) {
1285 error = -ENOMEM;
1286 goto err_out;
1287 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001288
Barry Naujok94035402008-05-21 16:50:46 +10001289 found = d_splice_alias(inode, new);
1290 if (found) {
1291 dput(new);
1292 return found;
1293 }
1294 return new;
1295 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001296
1297 /*
1298 * If a matching dentry exists, and it's not negative use it.
1299 *
1300 * Decrement the reference count to balance the iget() done
1301 * earlier on.
1302 */
Barry Naujok94035402008-05-21 16:50:46 +10001303 if (found->d_inode) {
1304 if (unlikely(found->d_inode != inode)) {
1305 /* This can't happen because bad inodes are unhashed. */
1306 BUG_ON(!is_bad_inode(inode));
1307 BUG_ON(!is_bad_inode(found->d_inode));
1308 }
Barry Naujok94035402008-05-21 16:50:46 +10001309 iput(inode);
1310 return found;
1311 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001312
Barry Naujok94035402008-05-21 16:50:46 +10001313 /*
1314 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001315 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001316 */
Barry Naujok94035402008-05-21 16:50:46 +10001317 spin_lock(&dcache_lock);
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001318 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001319 __d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001320 spin_unlock(&dcache_lock);
1321 security_d_instantiate(found, inode);
1322 return found;
1323 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001324
Barry Naujok94035402008-05-21 16:50:46 +10001325 /*
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001326 * In case a directory already has a (disconnected) entry grab a
1327 * reference to it, move it in place and use it.
Barry Naujok94035402008-05-21 16:50:46 +10001328 */
1329 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1330 dget_locked(new);
1331 spin_unlock(&dcache_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001332 security_d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001333 d_move(new, found);
Barry Naujok94035402008-05-21 16:50:46 +10001334 iput(inode);
Barry Naujok94035402008-05-21 16:50:46 +10001335 dput(found);
Barry Naujok94035402008-05-21 16:50:46 +10001336 return new;
1337
1338err_out:
1339 iput(inode);
1340 return ERR_PTR(error);
1341}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001342EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
1344/**
1345 * d_lookup - search for a dentry
1346 * @parent: parent dentry
1347 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001348 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001350 * d_lookup searches the children of the parent dentry for the name in
1351 * question. If the dentry is found its reference count is incremented and the
1352 * dentry is returned. The caller must use dput to free the entry when it has
1353 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1356{
1357 struct dentry * dentry = NULL;
1358 unsigned long seq;
1359
1360 do {
1361 seq = read_seqbegin(&rename_lock);
1362 dentry = __d_lookup(parent, name);
1363 if (dentry)
1364 break;
1365 } while (read_seqretry(&rename_lock, seq));
1366 return dentry;
1367}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001368EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
Nick Pigginb04f7842010-08-18 04:37:34 +10001370/*
1371 * __d_lookup - search for a dentry (racy)
1372 * @parent: parent dentry
1373 * @name: qstr of name we wish to find
1374 * Returns: dentry, or NULL
1375 *
1376 * __d_lookup is like d_lookup, however it may (rarely) return a
1377 * false-negative result due to unrelated rename activity.
1378 *
1379 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1380 * however it must be used carefully, eg. with a following d_lookup in
1381 * the case of failure.
1382 *
1383 * __d_lookup callers must be commented.
1384 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1386{
1387 unsigned int len = name->len;
1388 unsigned int hash = name->hash;
1389 const unsigned char *str = name->name;
1390 struct hlist_head *head = d_hash(parent,hash);
1391 struct dentry *found = NULL;
1392 struct hlist_node *node;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001393 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
Nick Pigginb04f7842010-08-18 04:37:34 +10001395 /*
1396 * The hash list is protected using RCU.
1397 *
1398 * Take d_lock when comparing a candidate dentry, to avoid races
1399 * with d_move().
1400 *
1401 * It is possible that concurrent renames can mess up our list
1402 * walk here and result in missing our dentry, resulting in the
1403 * false-negative result. d_lookup() protects against concurrent
1404 * renames using rename_lock seqlock.
1405 *
1406 * See Documentation/vfs/dcache-locking.txt for more details.
1407 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 rcu_read_lock();
1409
Paul E. McKenney665a7582005-11-07 00:59:17 -08001410 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 struct qstr *qstr;
1412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 if (dentry->d_name.hash != hash)
1414 continue;
1415 if (dentry->d_parent != parent)
1416 continue;
1417
1418 spin_lock(&dentry->d_lock);
1419
1420 /*
1421 * Recheck the dentry after taking the lock - d_move may have
Nick Pigginb04f7842010-08-18 04:37:34 +10001422 * changed things. Don't bother checking the hash because
1423 * we're about to compare the whole name anyway.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 */
1425 if (dentry->d_parent != parent)
1426 goto next;
1427
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001428 /* non-existing due to RCU? */
1429 if (d_unhashed(dentry))
1430 goto next;
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 /*
1433 * It is safe to compare names since d_move() cannot
1434 * change the qstr (protected by d_lock).
1435 */
1436 qstr = &dentry->d_name;
1437 if (parent->d_op && parent->d_op->d_compare) {
1438 if (parent->d_op->d_compare(parent, qstr, name))
1439 goto next;
1440 } else {
1441 if (qstr->len != len)
1442 goto next;
1443 if (memcmp(qstr->name, str, len))
1444 goto next;
1445 }
1446
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001447 atomic_inc(&dentry->d_count);
1448 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 spin_unlock(&dentry->d_lock);
1450 break;
1451next:
1452 spin_unlock(&dentry->d_lock);
1453 }
1454 rcu_read_unlock();
1455
1456 return found;
1457}
1458
1459/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001460 * d_hash_and_lookup - hash the qstr then search for a dentry
1461 * @dir: Directory to search in
1462 * @name: qstr of name we wish to find
1463 *
1464 * On hash failure or on lookup failure NULL is returned.
1465 */
1466struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1467{
1468 struct dentry *dentry = NULL;
1469
1470 /*
1471 * Check for a fs-specific hash function. Note that we must
1472 * calculate the standard hash first, as the d_op->d_hash()
1473 * routine may choose to leave the hash value unchanged.
1474 */
1475 name->hash = full_name_hash(name->name, name->len);
1476 if (dir->d_op && dir->d_op->d_hash) {
1477 if (dir->d_op->d_hash(dir, name) < 0)
1478 goto out;
1479 }
1480 dentry = d_lookup(dir, name);
1481out:
1482 return dentry;
1483}
1484
1485/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 * d_validate - verify dentry provided from insecure source
1487 * @dentry: The dentry alleged to be valid child of @dparent
1488 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 *
1490 * An insecure source has sent us a dentry, here we verify it and dget() it.
1491 * This is used by ncpfs in its readdir implementation.
1492 * Zero is returned in the dentry is invalid.
1493 */
Nick Piggind3a23e12011-01-05 20:01:21 +11001494
1495int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496{
Nick Piggind3a23e12011-01-05 20:01:21 +11001497 struct hlist_head *base;
1498 struct hlist_node *lhp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500 /* Check whether the ptr might be valid at all.. */
1501 if (!kmem_ptr_validate(dentry_cache, dentry))
Nick Piggind3a23e12011-01-05 20:01:21 +11001502 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Nick Piggind3a23e12011-01-05 20:01:21 +11001504 if (dentry->d_parent != dparent)
1505 goto out;
1506
1507 spin_lock(&dcache_lock);
1508 base = d_hash(dparent, dentry->d_name.hash);
1509 hlist_for_each(lhp,base) {
1510 /* hlist_for_each_entry_rcu() not required for d_hash list
1511 * as it is parsed under dcache_lock
1512 */
1513 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1514 __dget_locked(dentry);
1515 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return 1;
1517 }
1518 }
Nick Piggind3a23e12011-01-05 20:01:21 +11001519 spin_unlock(&dcache_lock);
1520out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 return 0;
1522}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001523EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
1525/*
1526 * When a file is deleted, we have two options:
1527 * - turn this dentry into a negative dentry
1528 * - unhash this dentry and free it.
1529 *
1530 * Usually, we want to just turn this into
1531 * a negative dentry, but if anybody else is
1532 * currently using the dentry or the inode
1533 * we can't do that and we fall back on removing
1534 * it from the hash queues and waiting for
1535 * it to be deleted later when it has no users
1536 */
1537
1538/**
1539 * d_delete - delete a dentry
1540 * @dentry: The dentry to delete
1541 *
1542 * Turn the dentry into a negative dentry if possible, otherwise
1543 * remove it from the hash queues so it can be deleted later
1544 */
1545
1546void d_delete(struct dentry * dentry)
1547{
John McCutchan7a91bf72005-08-08 13:52:16 -04001548 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 /*
1550 * Are we the only user?
1551 */
1552 spin_lock(&dcache_lock);
1553 spin_lock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001554 isdir = S_ISDIR(dentry->d_inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 if (atomic_read(&dentry->d_count) == 1) {
Al Viro13e3c5e2010-05-21 16:11:04 -04001556 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 dentry_iput(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001558 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 return;
1560 }
1561
1562 if (!d_unhashed(dentry))
1563 __d_drop(dentry);
1564
1565 spin_unlock(&dentry->d_lock);
1566 spin_unlock(&dcache_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04001567
1568 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001570EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1573{
1574
1575 entry->d_flags &= ~DCACHE_UNHASHED;
1576 hlist_add_head_rcu(&entry->d_hash, list);
1577}
1578
David Howells770bfad2006-08-22 20:06:07 -04001579static void _d_rehash(struct dentry * entry)
1580{
1581 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1582}
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584/**
1585 * d_rehash - add an entry back to the hash
1586 * @entry: dentry to add to the hash
1587 *
1588 * Adds a dentry to the hash according to its name.
1589 */
1590
1591void d_rehash(struct dentry * entry)
1592{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 spin_lock(&dcache_lock);
1594 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04001595 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 spin_unlock(&entry->d_lock);
1597 spin_unlock(&dcache_lock);
1598}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001599EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601/*
1602 * When switching names, the actual string doesn't strictly have to
1603 * be preserved in the target - because we're dropping the target
1604 * anyway. As such, we can just do a simple memcpy() to copy over
1605 * the new name before we switch.
1606 *
1607 * Note that we have to be a lot more careful about getting the hash
1608 * switched - we have to switch the hash value properly even if it
1609 * then no longer matches the actual (corrupted) string of the target.
1610 * The hash value has to match the hash queue that the dentry is on..
1611 */
1612static void switch_names(struct dentry *dentry, struct dentry *target)
1613{
1614 if (dname_external(target)) {
1615 if (dname_external(dentry)) {
1616 /*
1617 * Both external: swap the pointers
1618 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001619 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 } else {
1621 /*
1622 * dentry:internal, target:external. Steal target's
1623 * storage and make target internal.
1624 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07001625 memcpy(target->d_iname, dentry->d_name.name,
1626 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 dentry->d_name.name = target->d_name.name;
1628 target->d_name.name = target->d_iname;
1629 }
1630 } else {
1631 if (dname_external(dentry)) {
1632 /*
1633 * dentry:external, target:internal. Give dentry's
1634 * storage to target and make dentry internal
1635 */
1636 memcpy(dentry->d_iname, target->d_name.name,
1637 target->d_name.len + 1);
1638 target->d_name.name = dentry->d_name.name;
1639 dentry->d_name.name = dentry->d_iname;
1640 } else {
1641 /*
1642 * Both are internal. Just copy target to dentry
1643 */
1644 memcpy(dentry->d_iname, target->d_name.name,
1645 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05001646 dentry->d_name.len = target->d_name.len;
1647 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
1649 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001650 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651}
1652
1653/*
1654 * We cannibalize "target" when moving dentry on top of it,
1655 * because it's going to be thrown away anyway. We could be more
1656 * polite about it, though.
1657 *
1658 * This forceful removal will result in ugly /proc output if
1659 * somebody holds a file open that got deleted due to a rename.
1660 * We could be nicer about the deleted file, and let it show
J. Bruce Fieldsbc154b12007-10-16 23:29:42 -07001661 * up under the name it had before it was deleted rather than
1662 * under the original name of the file that was moved on top of it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 */
1664
Trond Myklebust9eaef272006-10-21 10:24:20 -07001665/*
1666 * d_move_locked - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 * @dentry: entry to move
1668 * @target: new dentry
1669 *
1670 * Update the dcache to reflect the move of a file name. Negative
1671 * dcache entries should not be moved in this way.
1672 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07001673static void d_move_locked(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674{
1675 struct hlist_head *list;
1676
1677 if (!dentry->d_inode)
1678 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1679
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 write_seqlock(&rename_lock);
1681 /*
1682 * XXXX: do we really need to take target->d_lock?
1683 */
1684 if (target < dentry) {
1685 spin_lock(&target->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001686 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 } else {
1688 spin_lock(&dentry->d_lock);
Ingo Molnara90b9c02006-07-03 00:25:04 -07001689 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 }
1691
1692 /* Move the dentry to the target hash queue, if on different bucket */
Denis Chengf77e3492007-10-16 23:30:11 -07001693 if (d_unhashed(dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 goto already_unhashed;
1695
1696 hlist_del_rcu(&dentry->d_hash);
1697
1698already_unhashed:
1699 list = d_hash(target->d_parent, target->d_name.hash);
1700 __d_rehash(dentry, list);
1701
1702 /* Unhash the target: dput() will then get rid of it */
1703 __d_drop(target);
1704
Eric Dumazet5160ee62006-01-08 01:03:32 -08001705 list_del(&dentry->d_u.d_child);
1706 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
1708 /* Switch the names.. */
1709 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001710 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712 /* ... and switch the parents */
1713 if (IS_ROOT(dentry)) {
1714 dentry->d_parent = target->d_parent;
1715 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001716 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001718 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
1720 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08001721 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 }
1723
Eric Dumazet5160ee62006-01-08 01:03:32 -08001724 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08001726 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 spin_unlock(&dentry->d_lock);
1728 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001729}
1730
1731/**
1732 * d_move - move a dentry
1733 * @dentry: entry to move
1734 * @target: new dentry
1735 *
1736 * Update the dcache to reflect the move of a file name. Negative
1737 * dcache entries should not be moved in this way.
1738 */
1739
1740void d_move(struct dentry * dentry, struct dentry * target)
1741{
1742 spin_lock(&dcache_lock);
1743 d_move_locked(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 spin_unlock(&dcache_lock);
1745}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001746EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001748/**
1749 * d_ancestor - search for an ancestor
1750 * @p1: ancestor dentry
1751 * @p2: child dentry
1752 *
1753 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1754 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07001755 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001756struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001757{
1758 struct dentry *p;
1759
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001760 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07001761 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001762 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001763 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001764 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001765}
1766
1767/*
1768 * This helper attempts to cope with remotely renamed directories
1769 *
1770 * It assumes that the caller is already holding
1771 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1772 *
1773 * Note: If ever the locking in lock_rename() changes, then please
1774 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07001775 */
1776static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001777 __releases(dcache_lock)
Trond Myklebust9eaef272006-10-21 10:24:20 -07001778{
1779 struct mutex *m1 = NULL, *m2 = NULL;
1780 struct dentry *ret;
1781
1782 /* If alias and dentry share a parent, then no extra locks required */
1783 if (alias->d_parent == dentry->d_parent)
1784 goto out_unalias;
1785
1786 /* Check for loops */
1787 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09001788 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07001789 goto out_err;
1790
1791 /* See lock_rename() */
1792 ret = ERR_PTR(-EBUSY);
1793 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1794 goto out_err;
1795 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1796 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1797 goto out_err;
1798 m2 = &alias->d_parent->d_inode->i_mutex;
1799out_unalias:
1800 d_move_locked(alias, dentry);
1801 ret = alias;
1802out_err:
1803 spin_unlock(&dcache_lock);
1804 if (m2)
1805 mutex_unlock(m2);
1806 if (m1)
1807 mutex_unlock(m1);
1808 return ret;
1809}
1810
1811/*
David Howells770bfad2006-08-22 20:06:07 -04001812 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1813 * named dentry in place of the dentry to be replaced.
1814 */
1815static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1816{
1817 struct dentry *dparent, *aparent;
1818
1819 switch_names(dentry, anon);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08001820 swap(dentry->d_name.hash, anon->d_name.hash);
David Howells770bfad2006-08-22 20:06:07 -04001821
1822 dparent = dentry->d_parent;
1823 aparent = anon->d_parent;
1824
1825 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1826 list_del(&dentry->d_u.d_child);
1827 if (!IS_ROOT(dentry))
1828 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1829 else
1830 INIT_LIST_HEAD(&dentry->d_u.d_child);
1831
1832 anon->d_parent = (dparent == dentry) ? anon : dparent;
1833 list_del(&anon->d_u.d_child);
1834 if (!IS_ROOT(anon))
1835 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1836 else
1837 INIT_LIST_HEAD(&anon->d_u.d_child);
1838
1839 anon->d_flags &= ~DCACHE_DISCONNECTED;
1840}
1841
1842/**
1843 * d_materialise_unique - introduce an inode into the tree
1844 * @dentry: candidate dentry
1845 * @inode: inode to bind to the dentry, to which aliases may be attached
1846 *
1847 * Introduces an dentry into the tree, substituting an extant disconnected
1848 * root directory alias in its place if there is one
1849 */
1850struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1851{
Trond Myklebust9eaef272006-10-21 10:24:20 -07001852 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04001853
1854 BUG_ON(!d_unhashed(dentry));
1855
1856 spin_lock(&dcache_lock);
1857
1858 if (!inode) {
1859 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001860 __d_instantiate(dentry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001861 goto found_lock;
1862 }
1863
Trond Myklebust9eaef272006-10-21 10:24:20 -07001864 if (S_ISDIR(inode->i_mode)) {
1865 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04001866
Trond Myklebust9eaef272006-10-21 10:24:20 -07001867 /* Does an aliased dentry already exist? */
1868 alias = __d_find_alias(inode, 0);
1869 if (alias) {
1870 actual = alias;
1871 /* Is this an anonymous mountpoint that we could splice
1872 * into our tree? */
1873 if (IS_ROOT(alias)) {
1874 spin_lock(&alias->d_lock);
1875 __d_materialise_dentry(dentry, alias);
1876 __d_drop(alias);
1877 goto found;
1878 }
1879 /* Nope, but we must(!) avoid directory aliasing */
1880 actual = __d_unalias(dentry, alias);
1881 if (IS_ERR(actual))
1882 dput(alias);
1883 goto out_nolock;
1884 }
David Howells770bfad2006-08-22 20:06:07 -04001885 }
1886
1887 /* Add a unique reference */
1888 actual = __d_instantiate_unique(dentry, inode);
1889 if (!actual)
1890 actual = dentry;
1891 else if (unlikely(!d_unhashed(actual)))
1892 goto shouldnt_be_hashed;
1893
1894found_lock:
1895 spin_lock(&actual->d_lock);
1896found:
1897 _d_rehash(actual);
1898 spin_unlock(&actual->d_lock);
1899 spin_unlock(&dcache_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001900out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04001901 if (actual == dentry) {
1902 security_d_instantiate(dentry, inode);
1903 return NULL;
1904 }
1905
1906 iput(inode);
1907 return actual;
1908
David Howells770bfad2006-08-22 20:06:07 -04001909shouldnt_be_hashed:
1910 spin_unlock(&dcache_lock);
1911 BUG();
David Howells770bfad2006-08-22 20:06:07 -04001912}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001913EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04001914
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001915static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01001916{
1917 *buflen -= namelen;
1918 if (*buflen < 0)
1919 return -ENAMETOOLONG;
1920 *buffer -= namelen;
1921 memcpy(*buffer, str, namelen);
1922 return 0;
1923}
1924
Miklos Szeredicdd16d02008-06-23 18:11:53 +02001925static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1926{
1927 return prepend(buffer, buflen, name->name, name->len);
1928}
1929
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930/**
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001931 * Prepend path string to a buffer
1932 *
1933 * @path: the dentry/vfsmount to report
1934 * @root: root vfsmnt/dentry (may be modified by this function)
1935 * @buffer: pointer to the end of the buffer
1936 * @buflen: pointer to buffer length
1937 *
1938 * Caller holds the dcache_lock.
1939 *
1940 * If path is not reachable from the supplied root, then the value of
1941 * root is changed (without modifying refcounts).
1942 */
1943static int prepend_path(const struct path *path, struct path *root,
1944 char **buffer, int *buflen)
1945{
1946 struct dentry *dentry = path->dentry;
1947 struct vfsmount *vfsmnt = path->mnt;
1948 bool slash = false;
1949 int error = 0;
1950
Nick Piggin99b7db72010-08-18 04:37:39 +10001951 br_read_lock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001952 while (dentry != root->dentry || vfsmnt != root->mnt) {
1953 struct dentry * parent;
1954
1955 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1956 /* Global root? */
1957 if (vfsmnt->mnt_parent == vfsmnt) {
1958 goto global_root;
1959 }
1960 dentry = vfsmnt->mnt_mountpoint;
1961 vfsmnt = vfsmnt->mnt_parent;
1962 continue;
1963 }
1964 parent = dentry->d_parent;
1965 prefetch(parent);
1966 error = prepend_name(buffer, buflen, &dentry->d_name);
1967 if (!error)
1968 error = prepend(buffer, buflen, "/", 1);
1969 if (error)
1970 break;
1971
1972 slash = true;
1973 dentry = parent;
1974 }
1975
1976out:
1977 if (!error && !slash)
1978 error = prepend(buffer, buflen, "/", 1);
1979
Nick Piggin99b7db72010-08-18 04:37:39 +10001980 br_read_unlock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02001981 return error;
1982
1983global_root:
1984 /*
1985 * Filesystems needing to implement special "root names"
1986 * should do so with ->d_dname()
1987 */
1988 if (IS_ROOT(dentry) &&
1989 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
1990 WARN(1, "Root dentry has weird name <%.*s>\n",
1991 (int) dentry->d_name.len, dentry->d_name.name);
1992 }
1993 root->mnt = vfsmnt;
1994 root->dentry = dentry;
1995 goto out;
1996}
1997
1998/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02001999 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002000 * @path: the dentry/vfsmount to report
2001 * @root: root vfsmnt/dentry (may be modified by this function)
Randy Dunlapcd956a12010-08-14 13:05:31 -07002002 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 * @buflen: buffer length
2004 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002005 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002007 * Returns a pointer into the buffer or an error code if the
2008 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002009 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002010 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002011 *
2012 * If path is not reachable from the supplied root, then the value of
2013 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002015char *__d_path(const struct path *path, struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002016 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002018 char *res = buf + buflen;
2019 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002021 prepend(&res, &buflen, "\0", 1);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002022 spin_lock(&dcache_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002023 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002024 spin_unlock(&dcache_lock);
2025
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002026 if (error)
2027 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002028 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029}
2030
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002031/*
2032 * same as __d_path but appends "(deleted)" for unlinked files.
2033 */
2034static int path_with_deleted(const struct path *path, struct path *root,
2035 char **buf, int *buflen)
2036{
2037 prepend(buf, buflen, "\0", 1);
2038 if (d_unlinked(path->dentry)) {
2039 int error = prepend(buf, buflen, " (deleted)", 10);
2040 if (error)
2041 return error;
2042 }
2043
2044 return prepend_path(path, root, buf, buflen);
2045}
2046
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002047static int prepend_unreachable(char **buffer, int *buflen)
2048{
2049 return prepend(buffer, buflen, "(unreachable)", 13);
2050}
2051
Jan Bluncka03a8a702008-02-14 19:38:32 -08002052/**
2053 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002054 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002055 * @buf: buffer to return value in
2056 * @buflen: buffer length
2057 *
2058 * Convert a dentry into an ASCII path name. If the entry has been deleted
2059 * the string " (deleted)" is appended. Note that this is ambiguous.
2060 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002061 * Returns a pointer into the buffer or an error code if the path was
2062 * too long. Note: Callers should use the returned pointer, not the passed
2063 * in buffer, to use the name! The implementation often starts at an offset
2064 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002065 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002066 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002067 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002068char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002070 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002071 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002072 struct path tmp;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002073 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002075 /*
2076 * We have various synthetic filesystems that never get mounted. On
2077 * these filesystems dentries are never used for lookup purposes, and
2078 * thus don't need to be hashed. They also don't need a name until a
2079 * user wants to identify the object in /proc/pid/fd/. The little hack
2080 * below allows us to generate a name for these objects on demand:
2081 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002082 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2083 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002084
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002085 get_fs_root(current->fs, &root);
Linus Torvalds552ce542007-02-13 12:08:18 -08002086 spin_lock(&dcache_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002087 tmp = root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002088 error = path_with_deleted(path, &tmp, &res, &buflen);
2089 if (error)
2090 res = ERR_PTR(error);
Linus Torvalds552ce542007-02-13 12:08:18 -08002091 spin_unlock(&dcache_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002092 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 return res;
2094}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002095EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002097/**
2098 * d_path_with_unreachable - return the path of a dentry
2099 * @path: path to report
2100 * @buf: buffer to return value in
2101 * @buflen: buffer length
2102 *
2103 * The difference from d_path() is that this prepends "(unreachable)"
2104 * to paths which are unreachable from the current process' root.
2105 */
2106char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2107{
2108 char *res = buf + buflen;
2109 struct path root;
2110 struct path tmp;
2111 int error;
2112
2113 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2114 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2115
2116 get_fs_root(current->fs, &root);
2117 spin_lock(&dcache_lock);
2118 tmp = root;
2119 error = path_with_deleted(path, &tmp, &res, &buflen);
2120 if (!error && !path_equal(&tmp, &root))
2121 error = prepend_unreachable(&res, &buflen);
2122 spin_unlock(&dcache_lock);
2123 path_put(&root);
2124 if (error)
2125 res = ERR_PTR(error);
2126
2127 return res;
2128}
2129
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002131 * Helper function for dentry_operations.d_dname() members
2132 */
2133char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2134 const char *fmt, ...)
2135{
2136 va_list args;
2137 char temp[64];
2138 int sz;
2139
2140 va_start(args, fmt);
2141 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2142 va_end(args);
2143
2144 if (sz > sizeof(temp) || sz > buflen)
2145 return ERR_PTR(-ENAMETOOLONG);
2146
2147 buffer += buflen - sz;
2148 return memcpy(buffer, temp, sz);
2149}
2150
2151/*
Ram Pai6092d042008-03-27 13:06:20 +01002152 * Write full pathname from the root of the filesystem into the buffer.
2153 */
Al Viroc1031352010-06-06 22:31:14 -04002154char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002155{
2156 char *end = buf + buflen;
2157 char *retval;
2158
Ram Pai6092d042008-03-27 13:06:20 +01002159 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002160 if (buflen < 1)
2161 goto Elong;
2162 /* Get '/' right */
2163 retval = end-1;
2164 *retval = '/';
2165
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002166 while (!IS_ROOT(dentry)) {
2167 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01002168
Ram Pai6092d042008-03-27 13:06:20 +01002169 prefetch(parent);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002170 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
Ram Pai6092d042008-03-27 13:06:20 +01002171 (prepend(&end, &buflen, "/", 1) != 0))
2172 goto Elong;
2173
2174 retval = end;
2175 dentry = parent;
2176 }
Al Viroc1031352010-06-06 22:31:14 -04002177 return retval;
2178Elong:
2179 return ERR_PTR(-ENAMETOOLONG);
2180}
2181EXPORT_SYMBOL(__dentry_path);
2182
2183char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2184{
2185 char *p = NULL;
2186 char *retval;
2187
2188 spin_lock(&dcache_lock);
2189 if (d_unlinked(dentry)) {
2190 p = buf + buflen;
2191 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2192 goto Elong;
2193 buflen++;
2194 }
2195 retval = __dentry_path(dentry, buf, buflen);
Ram Pai6092d042008-03-27 13:06:20 +01002196 spin_unlock(&dcache_lock);
Al Viroc1031352010-06-06 22:31:14 -04002197 if (!IS_ERR(retval) && p)
2198 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002199 return retval;
2200Elong:
2201 spin_unlock(&dcache_lock);
2202 return ERR_PTR(-ENAMETOOLONG);
2203}
2204
2205/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 * NOTE! The user-level library version returns a
2207 * character pointer. The kernel system call just
2208 * returns the length of the buffer filled (which
2209 * includes the ending '\0' character), or a negative
2210 * error value. So libc would do something like
2211 *
2212 * char *getcwd(char * buf, size_t size)
2213 * {
2214 * int retval;
2215 *
2216 * retval = sys_getcwd(buf, size);
2217 * if (retval >= 0)
2218 * return buf;
2219 * errno = -retval;
2220 * return NULL;
2221 * }
2222 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002223SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224{
Linus Torvalds552ce542007-02-13 12:08:18 -08002225 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002226 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002227 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228
2229 if (!page)
2230 return -ENOMEM;
2231
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002232 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233
Linus Torvalds552ce542007-02-13 12:08:18 -08002234 error = -ENOENT;
Linus Torvalds552ce542007-02-13 12:08:18 -08002235 spin_lock(&dcache_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002236 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002237 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002238 struct path tmp = root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002239 char *cwd = page + PAGE_SIZE;
2240 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002242 prepend(&cwd, &buflen, "\0", 1);
2243 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
Linus Torvalds552ce542007-02-13 12:08:18 -08002244 spin_unlock(&dcache_lock);
2245
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002246 if (error)
Linus Torvalds552ce542007-02-13 12:08:18 -08002247 goto out;
2248
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002249 /* Unreachable from current root */
2250 if (!path_equal(&tmp, &root)) {
2251 error = prepend_unreachable(&cwd, &buflen);
2252 if (error)
2253 goto out;
2254 }
2255
Linus Torvalds552ce542007-02-13 12:08:18 -08002256 error = -ERANGE;
2257 len = PAGE_SIZE + page - cwd;
2258 if (len <= size) {
2259 error = len;
2260 if (copy_to_user(buf, cwd, len))
2261 error = -EFAULT;
2262 }
2263 } else
2264 spin_unlock(&dcache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002267 path_put(&pwd);
2268 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 free_page((unsigned long) page);
2270 return error;
2271}
2272
2273/*
2274 * Test whether new_dentry is a subdirectory of old_dentry.
2275 *
2276 * Trivially implemented using the dcache structure
2277 */
2278
2279/**
2280 * is_subdir - is new dentry a subdirectory of old_dentry
2281 * @new_dentry: new dentry
2282 * @old_dentry: old dentry
2283 *
2284 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2285 * Returns 0 otherwise.
2286 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2287 */
2288
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002289int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290{
2291 int result;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 unsigned long seq;
2293
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002294 if (new_dentry == old_dentry)
2295 return 1;
2296
2297 /*
2298 * Need rcu_readlock to protect against the d_parent trashing
2299 * due to d_move
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 */
2301 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002302 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 seq = read_seqbegin(&rename_lock);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002305 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002307 else
2308 result = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 } while (read_seqretry(&rename_lock, seq));
2310 rcu_read_unlock();
2311
2312 return result;
2313}
2314
Al Viro2096f752010-01-30 13:16:21 -05002315int path_is_under(struct path *path1, struct path *path2)
2316{
2317 struct vfsmount *mnt = path1->mnt;
2318 struct dentry *dentry = path1->dentry;
2319 int res;
Nick Piggin99b7db72010-08-18 04:37:39 +10002320
2321 br_read_lock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002322 if (mnt != path2->mnt) {
2323 for (;;) {
2324 if (mnt->mnt_parent == mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +10002325 br_read_unlock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002326 return 0;
2327 }
2328 if (mnt->mnt_parent == path2->mnt)
2329 break;
2330 mnt = mnt->mnt_parent;
2331 }
2332 dentry = mnt->mnt_mountpoint;
2333 }
2334 res = is_subdir(dentry, path2->dentry);
Nick Piggin99b7db72010-08-18 04:37:39 +10002335 br_read_unlock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002336 return res;
2337}
2338EXPORT_SYMBOL(path_is_under);
2339
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340void d_genocide(struct dentry *root)
2341{
2342 struct dentry *this_parent = root;
2343 struct list_head *next;
2344
2345 spin_lock(&dcache_lock);
2346repeat:
2347 next = this_parent->d_subdirs.next;
2348resume:
2349 while (next != &this_parent->d_subdirs) {
2350 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002351 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 next = tmp->next;
2353 if (d_unhashed(dentry)||!dentry->d_inode)
2354 continue;
2355 if (!list_empty(&dentry->d_subdirs)) {
2356 this_parent = dentry;
2357 goto repeat;
2358 }
2359 atomic_dec(&dentry->d_count);
2360 }
2361 if (this_parent != root) {
Eric Dumazet5160ee62006-01-08 01:03:32 -08002362 next = this_parent->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363 atomic_dec(&this_parent->d_count);
2364 this_parent = this_parent->d_parent;
2365 goto resume;
2366 }
2367 spin_unlock(&dcache_lock);
2368}
2369
2370/**
2371 * find_inode_number - check for dentry with name
2372 * @dir: directory to check
2373 * @name: Name to find.
2374 *
2375 * Check whether a dentry already exists for the given name,
2376 * and return the inode number if it has an inode. Otherwise
2377 * 0 is returned.
2378 *
2379 * This routine is used to post-process directory listings for
2380 * filesystems using synthetic inode numbers, and is necessary
2381 * to keep getcwd() working.
2382 */
2383
2384ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2385{
2386 struct dentry * dentry;
2387 ino_t ino = 0;
2388
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002389 dentry = d_hash_and_lookup(dir, name);
2390 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 if (dentry->d_inode)
2392 ino = dentry->d_inode->i_ino;
2393 dput(dentry);
2394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return ino;
2396}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002397EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
2399static __initdata unsigned long dhash_entries;
2400static int __init set_dhash_entries(char *str)
2401{
2402 if (!str)
2403 return 0;
2404 dhash_entries = simple_strtoul(str, &str, 0);
2405 return 1;
2406}
2407__setup("dhash_entries=", set_dhash_entries);
2408
2409static void __init dcache_init_early(void)
2410{
2411 int loop;
2412
2413 /* If hashes are distributed across NUMA nodes, defer
2414 * hash allocation until vmalloc space is available.
2415 */
2416 if (hashdist)
2417 return;
2418
2419 dentry_hashtable =
2420 alloc_large_system_hash("Dentry cache",
2421 sizeof(struct hlist_head),
2422 dhash_entries,
2423 13,
2424 HASH_EARLY,
2425 &d_hash_shift,
2426 &d_hash_mask,
2427 0);
2428
2429 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2430 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2431}
2432
Denis Cheng74bf17c2007-10-16 23:26:30 -07002433static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434{
2435 int loop;
2436
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04002437 percpu_counter_init(&nr_dentry, 0);
2438 percpu_counter_init(&nr_dentry_unused, 0);
2439
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 /*
2441 * A constructor could be added for stable state like the lists,
2442 * but it is probably not worth it because of the cache nature
2443 * of the dcache.
2444 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002445 dentry_cache = KMEM_CACHE(dentry,
2446 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Rusty Russell8e1f9362007-07-17 04:03:17 -07002448 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
2450 /* Hash may have been set up in dcache_init_early */
2451 if (!hashdist)
2452 return;
2453
2454 dentry_hashtable =
2455 alloc_large_system_hash("Dentry cache",
2456 sizeof(struct hlist_head),
2457 dhash_entries,
2458 13,
2459 0,
2460 &d_hash_shift,
2461 &d_hash_mask,
2462 0);
2463
2464 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2465 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2466}
2467
2468/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08002469struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002470EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472EXPORT_SYMBOL(d_genocide);
2473
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474void __init vfs_caches_init_early(void)
2475{
2476 dcache_init_early();
2477 inode_init_early();
2478}
2479
2480void __init vfs_caches_init(unsigned long mempages)
2481{
2482 unsigned long reserve;
2483
2484 /* Base hash sizes on available memory, with a reserve equal to
2485 150% of current kernel size */
2486
2487 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2488 mempages -= reserve;
2489
2490 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09002491 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
Denis Cheng74bf17c2007-10-16 23:26:30 -07002493 dcache_init();
2494 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07002496 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 bdev_cache_init();
2498 chrdev_init();
2499}