blob: 0c6d5c549d840008b4dfff4e45e1aa6cbdcb88f8 [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>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
David Howells07f3f052006-09-30 20:52:18 +020038#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Nick Piggin789680d2011-01-07 17:49:30 +110040/*
41 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110042 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110044 * dcache_hash_bucket lock protects:
45 * - the dcache hash table
46 * s_anon bl list spinlock protects:
47 * - the s_anon list (see __d_drop)
Nick Piggin23044502011-01-07 17:49:31 +110048 * dcache_lru_lock protects:
49 * - the dcache lru lists and counters
50 * d_lock protects:
51 * - d_flags
52 * - d_name
53 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110054 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110055 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110056 * - d_parent and d_subdirs
57 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110058 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110059 *
60 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110061 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110062 * dentry->d_lock
63 * dcache_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110064 * dcache_hash_bucket lock
65 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110066 *
Nick Pigginda502952011-01-07 17:49:33 +110067 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock
69 * ...
70 * dentry->d_parent->d_lock
71 * dentry->d_lock
72 *
73 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110074 * if (dentry1 < dentry2)
75 * dentry1->d_lock
76 * dentry2->d_lock
77 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080078int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
80
Nick Piggin23044502011-01-07 17:49:31 +110081static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
Al Viro74c3cbe2007-07-22 08:04:18 -040082__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Nick Piggin949854d2011-01-07 17:49:37 +110084EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Christoph Lametere18b8902006-12-06 20:33:20 -080086static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/*
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
92 *
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
95 */
96#define D_HASHBITS d_hash_shift
97#define D_HASHMASK d_hash_mask
98
Eric Dumazetfa3536c2006-03-26 01:37:24 -080099static unsigned int d_hash_mask __read_mostly;
100static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100101
102struct dcache_hash_bucket {
103 struct hlist_bl_head head;
104};
105static struct dcache_hash_bucket *dentry_hashtable __read_mostly;
106
107static inline struct dcache_hash_bucket *d_hash(struct dentry *parent,
108 unsigned long hash)
109{
110 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
111 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
112 return dentry_hashtable + (hash & D_HASHMASK);
113}
114
115static inline void spin_lock_bucket(struct dcache_hash_bucket *b)
116{
117 bit_spin_lock(0, (unsigned long *)&b->head.first);
118}
119
120static inline void spin_unlock_bucket(struct dcache_hash_bucket *b)
121{
122 __bit_spin_unlock(0, (unsigned long *)&b->head.first);
123}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125/* Statistics gathering. */
126struct dentry_stat_t dentry_stat = {
127 .age_limit = 45,
128};
129
Nick Piggin3e880fb2011-01-07 17:49:19 +1100130static DEFINE_PER_CPU(unsigned int, nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400131
132#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100133static int get_nr_dentry(void)
134{
135 int i;
136 int sum = 0;
137 for_each_possible_cpu(i)
138 sum += per_cpu(nr_dentry, i);
139 return sum < 0 ? 0 : sum;
140}
141
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400142int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
143 size_t *lenp, loff_t *ppos)
144{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100145 dentry_stat.nr_dentry = get_nr_dentry();
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400146 return proc_dointvec(table, write, buffer, lenp, ppos);
147}
148#endif
149
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400150static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400152 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
153
Arjan van de Venfd217f42008-10-21 06:47:33 -0700154 WARN_ON(!list_empty(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (dname_external(dentry))
156 kfree(dentry->d_name.name);
157 kmem_cache_free(dentry_cache, dentry);
158}
159
160/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100161 * no locks, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 */
163static void d_free(struct dentry *dentry)
164{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100165 BUG_ON(dentry->d_count);
Nick Piggin3e880fb2011-01-07 17:49:19 +1100166 this_cpu_dec(nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 if (dentry->d_op && dentry->d_op->d_release)
168 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400169
Eric Dumazetb3423412006-12-06 20:38:48 -0800170 /* if dentry was never inserted into hash, immediate free is OK */
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100171 if (hlist_bl_unhashed(&dentry->d_hash))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400172 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800173 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400174 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
Nick Piggin31e6b012011-01-07 17:49:52 +1100177/**
178 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
179 * After this call, in-progress rcu-walk path lookup will fail. This
180 * should be called after unhashing, and after changing d_inode (if
181 * the dentry has not already been unhashed).
182 */
183static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
184{
185 assert_spin_locked(&dentry->d_lock);
186 /* Go through a barrier */
187 write_seqcount_barrier(&dentry->d_seq);
188}
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190/*
191 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100192 * d_iput() operation if defined. Dentry has no refcount
193 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800195static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200196 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100197 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
199 struct inode *inode = dentry->d_inode;
200 if (inode) {
201 dentry->d_inode = NULL;
202 list_del_init(&dentry->d_alias);
203 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100204 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700205 if (!inode->i_nlink)
206 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (dentry->d_op && dentry->d_op->d_iput)
208 dentry->d_op->d_iput(dentry, inode);
209 else
210 iput(inode);
211 } else {
212 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 }
214}
215
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700216/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100217 * Release the dentry's inode, using the filesystem
218 * d_iput() operation if defined. dentry remains in-use.
219 */
220static void dentry_unlink_inode(struct dentry * dentry)
221 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100222 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100223{
224 struct inode *inode = dentry->d_inode;
225 dentry->d_inode = NULL;
226 list_del_init(&dentry->d_alias);
227 dentry_rcuwalk_barrier(dentry);
228 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100229 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100230 if (!inode->i_nlink)
231 fsnotify_inoderemove(inode);
232 if (dentry->d_op && dentry->d_op->d_iput)
233 dentry->d_op->d_iput(dentry, inode);
234 else
235 iput(inode);
236}
237
238/*
Nick Piggin23044502011-01-07 17:49:31 +1100239 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700240 */
241static void dentry_lru_add(struct dentry *dentry)
242{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400243 if (list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100244 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400245 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
246 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100247 dentry_stat.nr_unused++;
Nick Piggin23044502011-01-07 17:49:31 +1100248 spin_unlock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400249 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700250}
251
Nick Piggin23044502011-01-07 17:49:31 +1100252static void __dentry_lru_del(struct dentry *dentry)
253{
254 list_del_init(&dentry->d_lru);
255 dentry->d_sb->s_nr_dentry_unused--;
256 dentry_stat.nr_unused--;
257}
258
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700259static void dentry_lru_del(struct dentry *dentry)
260{
261 if (!list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100262 spin_lock(&dcache_lru_lock);
263 __dentry_lru_del(dentry);
264 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700265 }
266}
267
Christoph Hellwiga4633352010-10-10 05:36:26 -0400268static void dentry_lru_move_tail(struct dentry *dentry)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700269{
Nick Piggin23044502011-01-07 17:49:31 +1100270 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400271 if (list_empty(&dentry->d_lru)) {
272 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
273 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100274 dentry_stat.nr_unused++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400275 } else {
276 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700277 }
Nick Piggin23044502011-01-07 17:49:31 +1100278 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700279}
280
Miklos Szeredid52b9082007-05-08 00:23:46 -0700281/**
282 * d_kill - kill dentry and return parent
283 * @dentry: dentry to kill
284 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200285 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700286 *
287 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100288 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100289 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
290 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700291 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100292static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200293 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100294 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100295 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700296{
Nick Piggin949854d2011-01-07 17:49:37 +1100297 dentry->d_parent = NULL;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700298 list_del(&dentry->d_u.d_child);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100299 if (parent)
300 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700301 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100302 /*
303 * dentry_iput drops the locks, at which point nobody (except
304 * transient RCU lookups) can reach this dentry.
305 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700306 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900307 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700308}
309
Nick Piggin789680d2011-01-07 17:49:30 +1100310/**
311 * d_drop - drop a dentry
312 * @dentry: dentry to drop
313 *
314 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
315 * be found through a VFS lookup any more. Note that this is different from
316 * deleting the dentry - d_delete will try to mark the dentry negative if
317 * possible, giving a successful _negative_ lookup, while d_drop will
318 * just make the cache lookup fail.
319 *
320 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
321 * reason (NFS timeouts or autofs deletes).
322 *
323 * __d_drop requires dentry->d_lock.
324 */
325void __d_drop(struct dentry *dentry)
326{
327 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100328 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) {
329 bit_spin_lock(0,
330 (unsigned long *)&dentry->d_sb->s_anon.first);
331 dentry->d_flags |= DCACHE_UNHASHED;
332 hlist_bl_del_init(&dentry->d_hash);
333 __bit_spin_unlock(0,
334 (unsigned long *)&dentry->d_sb->s_anon.first);
335 } else {
336 struct dcache_hash_bucket *b;
337 b = d_hash(dentry->d_parent, dentry->d_name.hash);
338 spin_lock_bucket(b);
339 /*
340 * We may not actually need to put DCACHE_UNHASHED
341 * manipulations under the hash lock, but follow
342 * the principle of least surprise.
343 */
344 dentry->d_flags |= DCACHE_UNHASHED;
345 hlist_bl_del_rcu(&dentry->d_hash);
346 spin_unlock_bucket(b);
347 dentry_rcuwalk_barrier(dentry);
348 }
Nick Piggin789680d2011-01-07 17:49:30 +1100349 }
350}
351EXPORT_SYMBOL(__d_drop);
352
353void d_drop(struct dentry *dentry)
354{
Nick Piggin789680d2011-01-07 17:49:30 +1100355 spin_lock(&dentry->d_lock);
356 __d_drop(dentry);
357 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100358}
359EXPORT_SYMBOL(d_drop);
360
Nick Piggin77812a12011-01-07 17:49:48 +1100361/*
362 * Finish off a dentry we've decided to kill.
363 * dentry->d_lock must be held, returns with it unlocked.
364 * If ref is non-zero, then decrement the refcount too.
365 * Returns dentry requiring refcount drop, or NULL if we're done.
366 */
367static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
368 __releases(dentry->d_lock)
369{
Nick Piggin873feea2011-01-07 17:50:06 +1100370 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100371 struct dentry *parent;
372
Nick Piggin873feea2011-01-07 17:50:06 +1100373 inode = dentry->d_inode;
374 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100375relock:
376 spin_unlock(&dentry->d_lock);
377 cpu_relax();
378 return dentry; /* try again with same dentry */
379 }
380 if (IS_ROOT(dentry))
381 parent = NULL;
382 else
383 parent = dentry->d_parent;
384 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100385 if (inode)
386 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100387 goto relock;
388 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100389
Nick Piggin77812a12011-01-07 17:49:48 +1100390 if (ref)
391 dentry->d_count--;
392 /* if dentry was on the d_lru list delete it from there */
393 dentry_lru_del(dentry);
394 /* if it was on the hash then remove it */
395 __d_drop(dentry);
396 return d_kill(dentry, parent);
397}
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399/*
400 * This is dput
401 *
402 * This is complicated by the fact that we do not want to put
403 * dentries that are no longer on any hash chain on the unused
404 * list: we'd much rather just get rid of them immediately.
405 *
406 * However, that implies that we have to traverse the dentry
407 * tree upwards to the parents which might _also_ now be
408 * scheduled for deletion (it may have been only waiting for
409 * its last child to go away).
410 *
411 * This tail recursion is done by hand as we don't want to depend
412 * on the compiler to always get this right (gcc generally doesn't).
413 * Real recursion would eat up our stack space.
414 */
415
416/*
417 * dput - release a dentry
418 * @dentry: dentry to release
419 *
420 * Release a dentry. This will drop the usage count and if appropriate
421 * call the dentry unlink method as well as removing it from the queues and
422 * releasing its resources. If the parent dentries were scheduled for release
423 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425void dput(struct dentry *dentry)
426{
427 if (!dentry)
428 return;
429
430repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100431 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100434 BUG_ON(!dentry->d_count);
435 if (dentry->d_count > 1) {
436 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 return;
439 }
440
Nick Pigginfb045ad2011-01-07 17:49:55 +1100441 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100443 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 }
Nick Piggin265ac902010-10-10 05:36:24 -0400445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 /* Unreachable? Get rid of it */
447 if (d_unhashed(dentry))
448 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400449
450 /* Otherwise leave it cached and ensure it's on the LRU */
451 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400452 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400453
Nick Piggin61f3dee2011-01-07 17:49:40 +1100454 dentry->d_count--;
455 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 return;
457
Miklos Szeredid52b9082007-05-08 00:23:46 -0700458kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100459 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700460 if (dentry)
461 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700463EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465/**
466 * d_invalidate - invalidate a dentry
467 * @dentry: dentry to invalidate
468 *
469 * Try to invalidate the dentry if it turns out to be
470 * possible. If there are other dentries that can be
471 * reached through this one we can't delete it and we
472 * return -EBUSY. On success we return 0.
473 *
474 * no dcache lock.
475 */
476
477int d_invalidate(struct dentry * dentry)
478{
479 /*
480 * If it's already been dropped, return OK.
481 */
Nick Pigginda502952011-01-07 17:49:33 +1100482 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100484 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return 0;
486 }
487 /*
488 * Check whether to do a partial shrink_dcache
489 * to get rid of unused child entries.
490 */
491 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100492 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100494 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496
497 /*
498 * Somebody else still using it?
499 *
500 * If it's a directory, we can't drop it
501 * for fear of somebody re-populating it
502 * with children (even though dropping it
503 * would make it unreachable from the root,
504 * we might still populate it if it was a
505 * working directory or similar).
506 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100507 if (dentry->d_count > 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
509 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return -EBUSY;
511 }
512 }
513
514 __d_drop(dentry);
515 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 return 0;
517}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700518EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100520/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100521static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100523 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
Nick Piggindc0474b2011-01-07 17:49:43 +1100526static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100527{
Nick Piggin23044502011-01-07 17:49:31 +1100528 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100529 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100530 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100531}
532
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100533struct dentry *dget_parent(struct dentry *dentry)
534{
535 struct dentry *ret;
536
537repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100538 /*
539 * Don't need rcu_dereference because we re-check it was correct under
540 * the lock.
541 */
542 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100543 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100544 if (!ret) {
545 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100546 goto out;
547 }
Nick Piggina734eb42011-01-07 17:49:44 +1100548 spin_lock(&ret->d_lock);
549 if (unlikely(ret != dentry->d_parent)) {
550 spin_unlock(&ret->d_lock);
551 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100552 goto repeat;
553 }
Nick Piggina734eb42011-01-07 17:49:44 +1100554 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100555 BUG_ON(!ret->d_count);
556 ret->d_count++;
557 spin_unlock(&ret->d_lock);
558out:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100559 return ret;
560}
561EXPORT_SYMBOL(dget_parent);
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/**
564 * d_find_alias - grab a hashed alias of inode
565 * @inode: inode in question
566 * @want_discon: flag, used by d_splice_alias, to request
567 * that only a DISCONNECTED alias be returned.
568 *
569 * If inode has a hashed alias, or is a directory and has any alias,
570 * acquire the reference to alias and return it. Otherwise return NULL.
571 * Notice that if inode is a directory there can be only one alias and
572 * it can be unhashed only if it has no children, or if it is the root
573 * of a filesystem.
574 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700575 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700577 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 */
Nick Pigginda502952011-01-07 17:49:33 +1100579static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Nick Pigginda502952011-01-07 17:49:33 +1100581 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Nick Pigginda502952011-01-07 17:49:33 +1100583again:
584 discon_alias = NULL;
585 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
586 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700588 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100589 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 discon_alias = alias;
Nick Pigginda502952011-01-07 17:49:33 +1100591 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100592 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100593 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 return alias;
595 }
596 }
Nick Pigginda502952011-01-07 17:49:33 +1100597 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 }
Nick Pigginda502952011-01-07 17:49:33 +1100599 if (discon_alias) {
600 alias = discon_alias;
601 spin_lock(&alias->d_lock);
602 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
603 if (IS_ROOT(alias) &&
604 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100605 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100606 spin_unlock(&alias->d_lock);
607 return alias;
608 }
609 }
610 spin_unlock(&alias->d_lock);
611 goto again;
612 }
613 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614}
615
Nick Pigginda502952011-01-07 17:49:33 +1100616struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
David Howells214fda12006-03-25 03:06:36 -0800618 struct dentry *de = NULL;
619
620 if (!list_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100621 spin_lock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800622 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100623 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 return de;
626}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700627EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629/*
630 * Try to kill dentries associated with this inode.
631 * WARNING: you must own a reference to inode.
632 */
633void d_prune_aliases(struct inode *inode)
634{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700635 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100637 spin_lock(&inode->i_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700638 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100640 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100641 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 __d_drop(dentry);
643 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100644 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 dput(dentry);
646 goto restart;
647 }
648 spin_unlock(&dentry->d_lock);
649 }
Nick Piggin873feea2011-01-07 17:50:06 +1100650 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700652EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654/*
Nick Piggin77812a12011-01-07 17:49:48 +1100655 * Try to throw away a dentry - free the inode, dput the parent.
656 * Requires dentry->d_lock is held, and dentry->d_count == 0.
657 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700658 *
Nick Piggin77812a12011-01-07 17:49:48 +1100659 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 */
Nick Piggin77812a12011-01-07 17:49:48 +1100661static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200662 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663{
Nick Piggin77812a12011-01-07 17:49:48 +1100664 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700665
Nick Piggin77812a12011-01-07 17:49:48 +1100666 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700667 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100668 * If dentry_kill returns NULL, we have nothing more to do.
669 * if it returns the same dentry, trylocks failed. In either
670 * case, just loop again.
671 *
672 * Otherwise, we need to prune ancestors too. This is necessary
673 * to prevent quadratic behavior of shrink_dcache_parent(), but
674 * is also expected to be beneficial in reducing dentry cache
675 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700676 */
Nick Piggin77812a12011-01-07 17:49:48 +1100677 if (!parent)
678 return;
679 if (parent == dentry)
680 return;
681
682 /* Prune ancestors. */
683 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700684 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100685 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100686 if (dentry->d_count > 1) {
687 dentry->d_count--;
688 spin_unlock(&dentry->d_lock);
689 return;
690 }
Nick Piggin77812a12011-01-07 17:49:48 +1100691 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700692 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400695static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700697 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700698
Nick Pigginec336792011-01-07 17:49:47 +1100699 rcu_read_lock();
700 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100701 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
702 if (&dentry->d_lru == list)
703 break; /* empty */
704 spin_lock(&dentry->d_lock);
705 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
706 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100707 continue;
708 }
709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 /*
711 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700712 * the LRU because of laziness during lookup. Do not free
713 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100715 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100716 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700717 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 continue;
719 }
Nick Pigginec336792011-01-07 17:49:47 +1100720
Nick Pigginec336792011-01-07 17:49:47 +1100721 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100722
723 try_prune_one_dentry(dentry);
724
Nick Pigginec336792011-01-07 17:49:47 +1100725 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
Nick Pigginec336792011-01-07 17:49:47 +1100727 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400728}
729
730/**
731 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
732 * @sb: superblock to shrink dentry LRU.
733 * @count: number of entries to prune
734 * @flags: flags to control the dentry processing
735 *
736 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
737 */
738static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
739{
740 /* called from prune_dcache() and shrink_dcache_parent() */
741 struct dentry *dentry;
742 LIST_HEAD(referenced);
743 LIST_HEAD(tmp);
744 int cnt = *count;
745
Nick Piggin23044502011-01-07 17:49:31 +1100746relock:
747 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400748 while (!list_empty(&sb->s_dentry_lru)) {
749 dentry = list_entry(sb->s_dentry_lru.prev,
750 struct dentry, d_lru);
751 BUG_ON(dentry->d_sb != sb);
752
Nick Piggin23044502011-01-07 17:49:31 +1100753 if (!spin_trylock(&dentry->d_lock)) {
754 spin_unlock(&dcache_lru_lock);
755 cpu_relax();
756 goto relock;
757 }
758
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400759 /*
760 * If we are honouring the DCACHE_REFERENCED flag and the
761 * dentry has this flag set, don't free it. Clear the flag
762 * and put it back on the LRU.
763 */
Nick Piggin23044502011-01-07 17:49:31 +1100764 if (flags & DCACHE_REFERENCED &&
765 dentry->d_flags & DCACHE_REFERENCED) {
766 dentry->d_flags &= ~DCACHE_REFERENCED;
767 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400768 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100769 } else {
770 list_move_tail(&dentry->d_lru, &tmp);
771 spin_unlock(&dentry->d_lock);
772 if (!--cnt)
773 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400774 }
Nick Pigginec336792011-01-07 17:49:47 +1100775 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400776 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700777 if (!list_empty(&referenced))
778 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100779 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100780
781 shrink_dentry_list(&tmp);
782
783 *count = cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700786/**
787 * prune_dcache - shrink the dcache
788 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700790 * Shrink the dcache. This is done when we need more memory, or simply when we
791 * need to unmount something (at which point we need to unuse all dentries).
792 *
793 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700795static void prune_dcache(int count)
796{
Al Virodca33252010-07-25 02:31:46 +0400797 struct super_block *sb, *p = NULL;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700798 int w_count;
Nick Piggin86c87492011-01-07 17:49:18 +1100799 int unused = dentry_stat.nr_unused;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700800 int prune_ratio;
801 int pruned;
802
803 if (unused == 0 || count == 0)
804 return;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700805 if (count >= unused)
806 prune_ratio = 1;
807 else
808 prune_ratio = unused / count;
809 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400810 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400811 if (list_empty(&sb->s_instances))
812 continue;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700813 if (sb->s_nr_dentry_unused == 0)
814 continue;
815 sb->s_count++;
816 /* Now, we reclaim unused dentrins with fairness.
817 * We reclaim them same percentage from each superblock.
818 * We calculate number of dentries to scan on this sb
819 * as follows, but the implementation is arranged to avoid
820 * overflows:
821 * number of dentries to scan on this sb =
822 * count * (number of dentries on this sb /
823 * number of dentries in the machine)
824 */
825 spin_unlock(&sb_lock);
826 if (prune_ratio != 1)
827 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
828 else
829 w_count = sb->s_nr_dentry_unused;
830 pruned = w_count;
831 /*
832 * We need to be sure this filesystem isn't being unmounted,
833 * otherwise we could race with generic_shutdown_super(), and
834 * end up holding a reference to an inode while the filesystem
835 * is unmounted. So we try to get s_umount, and make sure
836 * s_root isn't NULL.
837 */
838 if (down_read_trylock(&sb->s_umount)) {
839 if ((sb->s_root != NULL) &&
840 (!list_empty(&sb->s_dentry_lru))) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700841 __shrink_dcache_sb(sb, &w_count,
842 DCACHE_REFERENCED);
843 pruned -= w_count;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700844 }
845 up_read(&sb->s_umount);
846 }
847 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400848 if (p)
849 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700850 count -= pruned;
Al Virodca33252010-07-25 02:31:46 +0400851 p = sb;
Al Viro79893c12010-03-22 20:27:55 -0400852 /* more work left to do? */
853 if (count <= 0)
854 break;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700855 }
Al Virodca33252010-07-25 02:31:46 +0400856 if (p)
857 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700858 spin_unlock(&sb_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700859}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
861/**
862 * shrink_dcache_sb - shrink dcache for a superblock
863 * @sb: superblock
864 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400865 * Shrink the dcache for the specified super block. This is used to free
866 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400868void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400870 LIST_HEAD(tmp);
871
Nick Piggin23044502011-01-07 17:49:31 +1100872 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400873 while (!list_empty(&sb->s_dentry_lru)) {
874 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100875 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400876 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100877 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400878 }
Nick Piggin23044502011-01-07 17:49:31 +1100879 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700881EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
883/*
David Howellsc636ebd2006-10-11 01:22:19 -0700884 * destroy a single subtree of dentries for unmount
885 * - see the comments on shrink_dcache_for_umount() for a description of the
886 * locking
887 */
888static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
889{
890 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700891 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700892
893 BUG_ON(!IS_ROOT(dentry));
894
895 /* detach this root from the system */
Nick Piggin23044502011-01-07 17:49:31 +1100896 spin_lock(&dentry->d_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400897 dentry_lru_del(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700898 __d_drop(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100899 spin_unlock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700900
901 for (;;) {
902 /* descend to the first leaf in the current subtree */
903 while (!list_empty(&dentry->d_subdirs)) {
904 struct dentry *loop;
905
906 /* this is a branch with children - detach all of them
907 * from the system in one go */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100908 spin_lock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700909 list_for_each_entry(loop, &dentry->d_subdirs,
910 d_u.d_child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100911 spin_lock_nested(&loop->d_lock,
912 DENTRY_D_LOCK_NESTED);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400913 dentry_lru_del(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700914 __d_drop(loop);
Nick Pigginda502952011-01-07 17:49:33 +1100915 spin_unlock(&loop->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700916 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100917 spin_unlock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700918
919 /* move to the first child */
920 dentry = list_entry(dentry->d_subdirs.next,
921 struct dentry, d_u.d_child);
922 }
923
924 /* consume the dentries from this leaf up through its parents
925 * until we find one with children or run out altogether */
926 do {
927 struct inode *inode;
928
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100929 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700930 printk(KERN_ERR
931 "BUG: Dentry %p{i=%lx,n=%s}"
932 " still in use (%d)"
933 " [unmount of %s %s]\n",
934 dentry,
935 dentry->d_inode ?
936 dentry->d_inode->i_ino : 0UL,
937 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100938 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700939 dentry->d_sb->s_type->name,
940 dentry->d_sb->s_id);
941 BUG();
942 }
943
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100944 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700945 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100946 list_del(&dentry->d_u.d_child);
947 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900948 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100949 spin_lock(&parent->d_lock);
950 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100951 list_del(&dentry->d_u.d_child);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100952 spin_unlock(&parent->d_lock);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900953 }
David Howellsc636ebd2006-10-11 01:22:19 -0700954
David Howellsf8713572006-10-28 10:38:46 -0700955 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700956
957 inode = dentry->d_inode;
958 if (inode) {
959 dentry->d_inode = NULL;
960 list_del_init(&dentry->d_alias);
961 if (dentry->d_op && dentry->d_op->d_iput)
962 dentry->d_op->d_iput(dentry, inode);
963 else
964 iput(inode);
965 }
966
967 d_free(dentry);
968
969 /* finished when we fall off the top of the tree,
970 * otherwise we ascend to the parent and move to the
971 * next sibling if there is one */
972 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400973 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700974 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700975 } while (list_empty(&dentry->d_subdirs));
976
977 dentry = list_entry(dentry->d_subdirs.next,
978 struct dentry, d_u.d_child);
979 }
980}
981
982/*
983 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100984 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700985 * - the superblock is detached from all mountings and open files, so the
986 * dentry trees will not be rearranged by the VFS
987 * - s_umount is write-locked, so the memory pressure shrinker will ignore
988 * any dentries belonging to this superblock that it comes across
989 * - the filesystem itself is no longer permitted to rearrange the dentries
990 * in this superblock
991 */
992void shrink_dcache_for_umount(struct super_block *sb)
993{
994 struct dentry *dentry;
995
996 if (down_read_trylock(&sb->s_umount))
997 BUG();
998
999 dentry = sb->s_root;
1000 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001001 spin_lock(&dentry->d_lock);
1002 dentry->d_count--;
1003 spin_unlock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -07001004 shrink_dcache_for_umount_subtree(dentry);
1005
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001006 while (!hlist_bl_empty(&sb->s_anon)) {
1007 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001008 shrink_dcache_for_umount_subtree(dentry);
1009 }
1010}
1011
1012/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 * Search for at least 1 mount point in the dentry's subdirs.
1014 * We descend to the next level whenever the d_subdirs
1015 * list is non-empty and continue searching.
1016 */
1017
1018/**
1019 * have_submounts - check for mounts over a dentry
1020 * @parent: dentry to check.
1021 *
1022 * Return true if the parent or its subdirectories contain
1023 * a mount point
1024 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025int have_submounts(struct dentry *parent)
1026{
Nick Piggin949854d2011-01-07 17:49:37 +11001027 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001029 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001030 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001031
Nick Piggin949854d2011-01-07 17:49:37 +11001032 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001033again:
1034 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 if (d_mountpoint(parent))
1037 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001038 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039repeat:
1040 next = this_parent->d_subdirs.next;
1041resume:
1042 while (next != &this_parent->d_subdirs) {
1043 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001044 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001046
1047 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001049 if (d_mountpoint(dentry)) {
1050 spin_unlock(&dentry->d_lock);
1051 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001055 spin_unlock(&this_parent->d_lock);
1056 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001058 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 goto repeat;
1060 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001061 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 }
1063 /*
1064 * All done at this level ... ascend and resume the search.
1065 */
1066 if (this_parent != parent) {
Nick Piggin949854d2011-01-07 17:49:37 +11001067 struct dentry *tmp;
1068 struct dentry *child;
1069
1070 tmp = this_parent->d_parent;
1071 rcu_read_lock();
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001072 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001073 child = this_parent;
1074 this_parent = tmp;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001075 spin_lock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001076 /* might go back up the wrong parent if we have had a rename
1077 * or deletion */
1078 if (this_parent != child->d_parent ||
Nick Piggin58db63d2011-01-07 17:49:39 +11001079 (!locked && read_seqretry(&rename_lock, seq))) {
Nick Piggin949854d2011-01-07 17:49:37 +11001080 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001081 rcu_read_unlock();
1082 goto rename_retry;
1083 }
1084 rcu_read_unlock();
1085 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 goto resume;
1087 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001088 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001089 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001090 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001091 if (locked)
1092 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 return 0; /* No mount points found in tree */
1094positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001095 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001096 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001097 if (locked)
1098 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001100
1101rename_retry:
1102 locked = 1;
1103 write_seqlock(&rename_lock);
1104 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001106EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
1108/*
1109 * Search the dentry child list for the specified parent,
1110 * and move any unused dentries to the end of the unused
1111 * list for prune_dcache(). We descend to the next level
1112 * whenever the d_subdirs list is non-empty and continue
1113 * searching.
1114 *
1115 * It returns zero iff there are no unused children,
1116 * otherwise it returns the number of children moved to
1117 * the end of the unused list. This may not be the total
1118 * number of unused children, because select_parent can
1119 * drop the lock and return early due to latency
1120 * constraints.
1121 */
1122static int select_parent(struct dentry * parent)
1123{
Nick Piggin949854d2011-01-07 17:49:37 +11001124 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001126 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001128 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129
Nick Piggin949854d2011-01-07 17:49:37 +11001130 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001131again:
1132 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001133 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134repeat:
1135 next = this_parent->d_subdirs.next;
1136resume:
1137 while (next != &this_parent->d_subdirs) {
1138 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001139 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 next = tmp->next;
1141
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001142 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 /*
1145 * move only zero ref count dentries to the end
1146 * of the unused list for prune_dcache
1147 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001148 if (!dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001149 dentry_lru_move_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 found++;
Christoph Hellwiga4633352010-10-10 05:36:26 -04001151 } else {
1152 dentry_lru_del(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 }
1154
1155 /*
1156 * We can return to the caller if we have found some (this
1157 * ensures forward progress). We'll be coming back to find
1158 * the rest.
1159 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001160 if (found && need_resched()) {
1161 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 /*
1166 * Descend a level if the d_subdirs list is non-empty.
1167 */
1168 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001169 spin_unlock(&this_parent->d_lock);
1170 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001172 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 goto repeat;
1174 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001175
1176 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
1178 /*
1179 * All done at this level ... ascend and resume the search.
1180 */
1181 if (this_parent != parent) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001182 struct dentry *tmp;
Nick Piggin949854d2011-01-07 17:49:37 +11001183 struct dentry *child;
1184
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001185 tmp = this_parent->d_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11001186 rcu_read_lock();
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001187 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001188 child = this_parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001189 this_parent = tmp;
1190 spin_lock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001191 /* might go back up the wrong parent if we have had a rename
1192 * or deletion */
1193 if (this_parent != child->d_parent ||
Nick Piggin58db63d2011-01-07 17:49:39 +11001194 (!locked && read_seqretry(&rename_lock, seq))) {
Nick Piggin949854d2011-01-07 17:49:37 +11001195 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001196 rcu_read_unlock();
1197 goto rename_retry;
1198 }
1199 rcu_read_unlock();
1200 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 goto resume;
1202 }
1203out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001204 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001205 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001206 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001207 if (locked)
1208 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001210
1211rename_retry:
1212 if (found)
1213 return found;
1214 locked = 1;
1215 write_seqlock(&rename_lock);
1216 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
1219/**
1220 * shrink_dcache_parent - prune dcache
1221 * @parent: parent of entries to prune
1222 *
1223 * Prune the dcache to remove unused children of the parent dentry.
1224 */
1225
1226void shrink_dcache_parent(struct dentry * parent)
1227{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001228 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 int found;
1230
1231 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001232 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001234EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236/*
1237 * Scan `nr' dentries and return the number which remain.
1238 *
1239 * We need to avoid reentering the filesystem if the caller is performing a
1240 * GFP_NOFS allocation attempt. One example deadlock is:
1241 *
1242 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1243 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1244 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1245 *
1246 * In this case we return -1 to tell the caller that we baled.
1247 */
Dave Chinner7f8275d2010-07-19 14:56:17 +10001248static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249{
1250 if (nr) {
1251 if (!(gfp_mask & __GFP_FS))
1252 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001253 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001255
Nick Piggin86c87492011-01-07 17:49:18 +11001256 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
1258
Rusty Russell8e1f9362007-07-17 04:03:17 -07001259static struct shrinker dcache_shrinker = {
1260 .shrink = shrink_dcache_memory,
1261 .seeks = DEFAULT_SEEKS,
1262};
1263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264/**
1265 * d_alloc - allocate a dcache entry
1266 * @parent: parent of entry to allocate
1267 * @name: qstr of the name
1268 *
1269 * Allocates a dentry. It returns %NULL if there is insufficient memory
1270 * available. On a success the dentry is returned. The name passed in is
1271 * copied and the copy passed in may be reused after this call.
1272 */
1273
1274struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1275{
1276 struct dentry *dentry;
1277 char *dname;
1278
Mel Gormane12ba742007-10-16 01:25:52 -07001279 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 if (!dentry)
1281 return NULL;
1282
1283 if (name->len > DNAME_INLINE_LEN-1) {
1284 dname = kmalloc(name->len + 1, GFP_KERNEL);
1285 if (!dname) {
1286 kmem_cache_free(dentry_cache, dentry);
1287 return NULL;
1288 }
1289 } else {
1290 dname = dentry->d_iname;
1291 }
1292 dentry->d_name.name = dname;
1293
1294 dentry->d_name.len = name->len;
1295 dentry->d_name.hash = name->hash;
1296 memcpy(dname, name->name, name->len);
1297 dname[name->len] = 0;
1298
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001299 dentry->d_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 dentry->d_flags = DCACHE_UNHASHED;
1301 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001302 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 dentry->d_inode = NULL;
1304 dentry->d_parent = NULL;
1305 dentry->d_sb = NULL;
1306 dentry->d_op = NULL;
1307 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001308 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 INIT_LIST_HEAD(&dentry->d_lru);
1310 INIT_LIST_HEAD(&dentry->d_subdirs);
1311 INIT_LIST_HEAD(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001312 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314 if (parent) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001315 spin_lock(&parent->d_lock);
Nick Piggin89ad4852011-01-07 17:49:41 +11001316 /*
1317 * don't need child lock because it is not subject
1318 * to concurrency here
1319 */
Nick Piggindc0474b2011-01-07 17:49:43 +11001320 __dget_dlock(parent);
1321 dentry->d_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 dentry->d_sb = parent->d_sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001323 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Eric Dumazet5160ee62006-01-08 01:03:32 -08001324 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001325 spin_unlock(&parent->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Nick Piggin3e880fb2011-01-07 17:49:19 +11001328 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 return dentry;
1331}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001332EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Nick Piggin4b936882011-01-07 17:50:07 +11001334struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1335{
1336 struct dentry *dentry = d_alloc(NULL, name);
1337 if (dentry) {
1338 dentry->d_sb = sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001339 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Nick Piggin4b936882011-01-07 17:50:07 +11001340 dentry->d_parent = dentry;
1341 dentry->d_flags |= DCACHE_DISCONNECTED;
1342 }
1343 return dentry;
1344}
1345EXPORT_SYMBOL(d_alloc_pseudo);
1346
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1348{
1349 struct qstr q;
1350
1351 q.name = name;
1352 q.len = strlen(name);
1353 q.hash = full_name_hash(q.name, q.len);
1354 return d_alloc(parent, &q);
1355}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001356EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Nick Pigginfb045ad2011-01-07 17:49:55 +11001358void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1359{
1360 BUG_ON(dentry->d_op);
1361 BUG_ON(dentry->d_flags & (DCACHE_OP_HASH |
1362 DCACHE_OP_COMPARE |
1363 DCACHE_OP_REVALIDATE |
1364 DCACHE_OP_DELETE ));
1365 dentry->d_op = op;
1366 if (!op)
1367 return;
1368 if (op->d_hash)
1369 dentry->d_flags |= DCACHE_OP_HASH;
1370 if (op->d_compare)
1371 dentry->d_flags |= DCACHE_OP_COMPARE;
1372 if (op->d_revalidate)
1373 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1374 if (op->d_delete)
1375 dentry->d_flags |= DCACHE_OP_DELETE;
1376
1377}
1378EXPORT_SYMBOL(d_set_d_op);
1379
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001380static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1381{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001382 spin_lock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001383 if (inode)
1384 list_add(&dentry->d_alias, &inode->i_dentry);
1385 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001386 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001387 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001388 fsnotify_d_instantiate(dentry, inode);
1389}
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391/**
1392 * d_instantiate - fill in inode information for a dentry
1393 * @entry: dentry to complete
1394 * @inode: inode to attach to this dentry
1395 *
1396 * Fill in inode information in the entry.
1397 *
1398 * This turns negative dentries into productive full members
1399 * of society.
1400 *
1401 * NOTE! This assumes that the inode count has been incremented
1402 * (or otherwise set) by the caller to indicate that it is now
1403 * in use by the dcache.
1404 */
1405
1406void d_instantiate(struct dentry *entry, struct inode * inode)
1407{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001408 BUG_ON(!list_empty(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001409 if (inode)
1410 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001411 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001412 if (inode)
1413 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 security_d_instantiate(entry, inode);
1415}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001416EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
1418/**
1419 * d_instantiate_unique - instantiate a non-aliased dentry
1420 * @entry: dentry to instantiate
1421 * @inode: inode to attach to this dentry
1422 *
1423 * Fill in inode information in the entry. On success, it returns NULL.
1424 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001425 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 *
1427 * Note that in order to avoid conflicts with rename() etc, the caller
1428 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001429 *
1430 * This also assumes that the inode count has been incremented
1431 * (or otherwise set) by the caller to indicate that it is now
1432 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 */
David Howells770bfad2006-08-22 20:06:07 -04001434static struct dentry *__d_instantiate_unique(struct dentry *entry,
1435 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
1437 struct dentry *alias;
1438 int len = entry->d_name.len;
1439 const char *name = entry->d_name.name;
1440 unsigned int hash = entry->d_name.hash;
1441
David Howells770bfad2006-08-22 20:06:07 -04001442 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001443 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001444 return NULL;
1445 }
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1448 struct qstr *qstr = &alias->d_name;
1449
Nick Piggin9abca362011-01-07 17:49:36 +11001450 /*
1451 * Don't need alias->d_lock here, because aliases with
1452 * d_parent == entry->d_parent are not subject to name or
1453 * parent changes, because the parent inode i_mutex is held.
1454 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (qstr->hash != hash)
1456 continue;
1457 if (alias->d_parent != entry->d_parent)
1458 continue;
Nick Piggin9d55c362011-01-07 17:50:09 +11001459 if (dentry_cmp(qstr->name, qstr->len, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001461 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return alias;
1463 }
David Howells770bfad2006-08-22 20:06:07 -04001464
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001465 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 return NULL;
1467}
David Howells770bfad2006-08-22 20:06:07 -04001468
1469struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1470{
1471 struct dentry *result;
1472
1473 BUG_ON(!list_empty(&entry->d_alias));
1474
Nick Piggin873feea2011-01-07 17:50:06 +11001475 if (inode)
1476 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001477 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001478 if (inode)
1479 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001480
1481 if (!result) {
1482 security_d_instantiate(entry, inode);
1483 return NULL;
1484 }
1485
1486 BUG_ON(!d_unhashed(result));
1487 iput(inode);
1488 return result;
1489}
1490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491EXPORT_SYMBOL(d_instantiate_unique);
1492
1493/**
1494 * d_alloc_root - allocate root dentry
1495 * @root_inode: inode to allocate the root for
1496 *
1497 * Allocate a root ("/") dentry for the inode given. The inode is
1498 * instantiated and returned. %NULL is returned if there is insufficient
1499 * memory or the inode passed is %NULL.
1500 */
1501
1502struct dentry * d_alloc_root(struct inode * root_inode)
1503{
1504 struct dentry *res = NULL;
1505
1506 if (root_inode) {
1507 static const struct qstr name = { .name = "/", .len = 1 };
1508
1509 res = d_alloc(NULL, &name);
1510 if (res) {
1511 res->d_sb = root_inode->i_sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001512 d_set_d_op(res, res->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 res->d_parent = res;
1514 d_instantiate(res, root_inode);
1515 }
1516 }
1517 return res;
1518}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001519EXPORT_SYMBOL(d_alloc_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001522 * d_obtain_alias - find or allocate a dentry for a given inode
1523 * @inode: inode to allocate the dentry for
1524 *
1525 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1526 * similar open by handle operations. The returned dentry may be anonymous,
1527 * or may have a full name (if the inode was already in the cache).
1528 *
1529 * When called on a directory inode, we must ensure that the inode only ever
1530 * has one dentry. If a dentry is found, that is returned instead of
1531 * allocating a new one.
1532 *
1533 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001534 * to the dentry. In case of an error the reference on the inode is released.
1535 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1536 * be passed in and will be the error will be propagate to the return value,
1537 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001538 */
1539struct dentry *d_obtain_alias(struct inode *inode)
1540{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001541 static const struct qstr anonstring = { .name = "" };
1542 struct dentry *tmp;
1543 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001544
1545 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001546 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001547 if (IS_ERR(inode))
1548 return ERR_CAST(inode);
1549
Christoph Hellwig9308a612008-08-11 15:49:12 +02001550 res = d_find_alias(inode);
1551 if (res)
1552 goto out_iput;
1553
1554 tmp = d_alloc(NULL, &anonstring);
1555 if (!tmp) {
1556 res = ERR_PTR(-ENOMEM);
1557 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001558 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001559 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1560
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001561
Nick Piggin873feea2011-01-07 17:50:06 +11001562 spin_lock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001563 res = __d_find_alias(inode, 0);
1564 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001565 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001566 dput(tmp);
1567 goto out_iput;
1568 }
1569
1570 /* attach a disconnected dentry */
1571 spin_lock(&tmp->d_lock);
1572 tmp->d_sb = inode->i_sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001573 d_set_d_op(tmp, tmp->d_sb->s_d_op);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001574 tmp->d_inode = inode;
1575 tmp->d_flags |= DCACHE_DISCONNECTED;
Christoph Hellwig9308a612008-08-11 15:49:12 +02001576 list_add(&tmp->d_alias, &inode->i_dentry);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001577 bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1578 tmp->d_flags &= ~DCACHE_UNHASHED;
1579 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1580 __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001581 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001582 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001583
Christoph Hellwig9308a612008-08-11 15:49:12 +02001584 return tmp;
1585
1586 out_iput:
1587 iput(inode);
1588 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001589}
Benny Halevyadc48722009-02-27 14:02:59 -08001590EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591
1592/**
1593 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1594 * @inode: the inode which may have a disconnected dentry
1595 * @dentry: a negative dentry which we want to point to the inode.
1596 *
1597 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1598 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1599 * and return it, else simply d_add the inode to the dentry and return NULL.
1600 *
1601 * This is needed in the lookup routine of any filesystem that is exportable
1602 * (via knfsd) so that we can build dcache paths to directories effectively.
1603 *
1604 * If a dentry was found and moved, then it is returned. Otherwise NULL
1605 * is returned. This matches the expected return value of ->lookup.
1606 *
1607 */
1608struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1609{
1610 struct dentry *new = NULL;
1611
NeilBrown21c0d8f2006-10-04 02:16:16 -07001612 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001613 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 new = __d_find_alias(inode, 1);
1615 if (new) {
1616 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001617 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 d_move(new, dentry);
1620 iput(inode);
1621 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001622 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001623 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001624 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 security_d_instantiate(dentry, inode);
1626 d_rehash(dentry);
1627 }
1628 } else
1629 d_add(dentry, inode);
1630 return new;
1631}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001632EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
Barry Naujok94035402008-05-21 16:50:46 +10001634/**
1635 * d_add_ci - lookup or allocate new dentry with case-exact name
1636 * @inode: the inode case-insensitive lookup has found
1637 * @dentry: the negative dentry that was passed to the parent's lookup func
1638 * @name: the case-exact name to be associated with the returned dentry
1639 *
1640 * This is to avoid filling the dcache with case-insensitive names to the
1641 * same inode, only the actual correct case is stored in the dcache for
1642 * case-insensitive filesystems.
1643 *
1644 * For a case-insensitive lookup match and if the the case-exact dentry
1645 * already exists in in the dcache, use it and return it.
1646 *
1647 * If no entry exists with the exact case name, allocate new dentry with
1648 * the exact case, and return the spliced entry.
1649 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001650struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001651 struct qstr *name)
1652{
1653 int error;
1654 struct dentry *found;
1655 struct dentry *new;
1656
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001657 /*
1658 * First check if a dentry matching the name already exists,
1659 * if not go ahead and create it now.
1660 */
Barry Naujok94035402008-05-21 16:50:46 +10001661 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001662 if (!found) {
1663 new = d_alloc(dentry->d_parent, name);
1664 if (!new) {
1665 error = -ENOMEM;
1666 goto err_out;
1667 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001668
Barry Naujok94035402008-05-21 16:50:46 +10001669 found = d_splice_alias(inode, new);
1670 if (found) {
1671 dput(new);
1672 return found;
1673 }
1674 return new;
1675 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001676
1677 /*
1678 * If a matching dentry exists, and it's not negative use it.
1679 *
1680 * Decrement the reference count to balance the iget() done
1681 * earlier on.
1682 */
Barry Naujok94035402008-05-21 16:50:46 +10001683 if (found->d_inode) {
1684 if (unlikely(found->d_inode != inode)) {
1685 /* This can't happen because bad inodes are unhashed. */
1686 BUG_ON(!is_bad_inode(inode));
1687 BUG_ON(!is_bad_inode(found->d_inode));
1688 }
Barry Naujok94035402008-05-21 16:50:46 +10001689 iput(inode);
1690 return found;
1691 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001692
Barry Naujok94035402008-05-21 16:50:46 +10001693 /*
1694 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001695 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001696 */
Nick Piggin873feea2011-01-07 17:50:06 +11001697 spin_lock(&inode->i_lock);
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001698 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001699 __d_instantiate(found, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001700 spin_unlock(&inode->i_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001701 security_d_instantiate(found, inode);
1702 return found;
1703 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001704
Barry Naujok94035402008-05-21 16:50:46 +10001705 /*
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001706 * In case a directory already has a (disconnected) entry grab a
1707 * reference to it, move it in place and use it.
Barry Naujok94035402008-05-21 16:50:46 +10001708 */
1709 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
Nick Piggindc0474b2011-01-07 17:49:43 +11001710 __dget(new);
Nick Piggin873feea2011-01-07 17:50:06 +11001711 spin_unlock(&inode->i_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001712 security_d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001713 d_move(new, found);
Barry Naujok94035402008-05-21 16:50:46 +10001714 iput(inode);
Barry Naujok94035402008-05-21 16:50:46 +10001715 dput(found);
Barry Naujok94035402008-05-21 16:50:46 +10001716 return new;
1717
1718err_out:
1719 iput(inode);
1720 return ERR_PTR(error);
1721}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001722EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001725 * __d_lookup_rcu - search for a dentry (racy, store-free)
1726 * @parent: parent dentry
1727 * @name: qstr of name we wish to find
1728 * @seq: returns d_seq value at the point where the dentry was found
1729 * @inode: returns dentry->d_inode when the inode was found valid.
1730 * Returns: dentry, or NULL
1731 *
1732 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1733 * resolution (store-free path walking) design described in
1734 * Documentation/filesystems/path-lookup.txt.
1735 *
1736 * This is not to be used outside core vfs.
1737 *
1738 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1739 * held, and rcu_read_lock held. The returned dentry must not be stored into
1740 * without taking d_lock and checking d_seq sequence count against @seq
1741 * returned here.
1742 *
1743 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1744 * function.
1745 *
1746 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1747 * the returned dentry, so long as its parent's seqlock is checked after the
1748 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1749 * is formed, giving integrity down the path walk.
1750 */
1751struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1752 unsigned *seq, struct inode **inode)
1753{
1754 unsigned int len = name->len;
1755 unsigned int hash = name->hash;
1756 const unsigned char *str = name->name;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001757 struct dcache_hash_bucket *b = d_hash(parent, hash);
1758 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001759 struct dentry *dentry;
1760
1761 /*
1762 * Note: There is significant duplication with __d_lookup_rcu which is
1763 * required to prevent single threaded performance regressions
1764 * especially on architectures where smp_rmb (in seqcounts) are costly.
1765 * Keep the two functions in sync.
1766 */
1767
1768 /*
1769 * The hash list is protected using RCU.
1770 *
1771 * Carefully use d_seq when comparing a candidate dentry, to avoid
1772 * races with d_move().
1773 *
1774 * It is possible that concurrent renames can mess up our list
1775 * walk here and result in missing our dentry, resulting in the
1776 * false-negative result. d_lookup() protects against concurrent
1777 * renames using rename_lock seqlock.
1778 *
1779 * See Documentation/vfs/dcache-locking.txt for more details.
1780 */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001781 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001782 struct inode *i;
1783 const char *tname;
1784 int tlen;
1785
1786 if (dentry->d_name.hash != hash)
1787 continue;
1788
1789seqretry:
1790 *seq = read_seqcount_begin(&dentry->d_seq);
1791 if (dentry->d_parent != parent)
1792 continue;
1793 if (d_unhashed(dentry))
1794 continue;
1795 tlen = dentry->d_name.len;
1796 tname = dentry->d_name.name;
1797 i = dentry->d_inode;
Nick Piggine1bb5782011-01-07 17:50:08 +11001798 prefetch(tname);
1799 if (i)
1800 prefetch(i);
Nick Piggin31e6b012011-01-07 17:49:52 +11001801 /*
1802 * This seqcount check is required to ensure name and
1803 * len are loaded atomically, so as not to walk off the
1804 * edge of memory when walking. If we could load this
1805 * atomically some other way, we could drop this check.
1806 */
1807 if (read_seqcount_retry(&dentry->d_seq, *seq))
1808 goto seqretry;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001809 if (parent->d_flags & DCACHE_OP_COMPARE) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001810 if (parent->d_op->d_compare(parent, *inode,
1811 dentry, i,
1812 tlen, tname, name))
1813 continue;
1814 } else {
Nick Piggin9d55c362011-01-07 17:50:09 +11001815 if (dentry_cmp(tname, tlen, str, len))
Nick Piggin31e6b012011-01-07 17:49:52 +11001816 continue;
1817 }
1818 /*
1819 * No extra seqcount check is required after the name
1820 * compare. The caller must perform a seqcount check in
1821 * order to do anything useful with the returned dentry
1822 * anyway.
1823 */
1824 *inode = i;
1825 return dentry;
1826 }
1827 return NULL;
1828}
1829
1830/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 * d_lookup - search for a dentry
1832 * @parent: parent dentry
1833 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001834 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001836 * d_lookup searches the children of the parent dentry for the name in
1837 * question. If the dentry is found its reference count is incremented and the
1838 * dentry is returned. The caller must use dput to free the entry when it has
1839 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001841struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
Nick Piggin31e6b012011-01-07 17:49:52 +11001843 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001844 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
1846 do {
1847 seq = read_seqbegin(&rename_lock);
1848 dentry = __d_lookup(parent, name);
1849 if (dentry)
1850 break;
1851 } while (read_seqretry(&rename_lock, seq));
1852 return dentry;
1853}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001854EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Nick Piggin31e6b012011-01-07 17:49:52 +11001856/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001857 * __d_lookup - search for a dentry (racy)
1858 * @parent: parent dentry
1859 * @name: qstr of name we wish to find
1860 * Returns: dentry, or NULL
1861 *
1862 * __d_lookup is like d_lookup, however it may (rarely) return a
1863 * false-negative result due to unrelated rename activity.
1864 *
1865 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1866 * however it must be used carefully, eg. with a following d_lookup in
1867 * the case of failure.
1868 *
1869 * __d_lookup callers must be commented.
1870 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001871struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
1873 unsigned int len = name->len;
1874 unsigned int hash = name->hash;
1875 const unsigned char *str = name->name;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001876 struct dcache_hash_bucket *b = d_hash(parent, hash);
1877 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001878 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001879 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Nick Pigginb04f7842010-08-18 04:37:34 +10001881 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001882 * Note: There is significant duplication with __d_lookup_rcu which is
1883 * required to prevent single threaded performance regressions
1884 * especially on architectures where smp_rmb (in seqcounts) are costly.
1885 * Keep the two functions in sync.
1886 */
1887
1888 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001889 * The hash list is protected using RCU.
1890 *
1891 * Take d_lock when comparing a candidate dentry, to avoid races
1892 * with d_move().
1893 *
1894 * It is possible that concurrent renames can mess up our list
1895 * walk here and result in missing our dentry, resulting in the
1896 * false-negative result. d_lookup() protects against concurrent
1897 * renames using rename_lock seqlock.
1898 *
1899 * See Documentation/vfs/dcache-locking.txt for more details.
1900 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 rcu_read_lock();
1902
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001903 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001904 const char *tname;
1905 int tlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 if (dentry->d_name.hash != hash)
1908 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
1910 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 if (dentry->d_parent != parent)
1912 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001913 if (d_unhashed(dentry))
1914 goto next;
1915
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 /*
1917 * It is safe to compare names since d_move() cannot
1918 * change the qstr (protected by d_lock).
1919 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001920 tlen = dentry->d_name.len;
1921 tname = dentry->d_name.name;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001922 if (parent->d_flags & DCACHE_OP_COMPARE) {
Nick Piggin621e1552011-01-07 17:49:27 +11001923 if (parent->d_op->d_compare(parent, parent->d_inode,
1924 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001925 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 goto next;
1927 } else {
Nick Piggin9d55c362011-01-07 17:50:09 +11001928 if (dentry_cmp(tname, tlen, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 goto next;
1930 }
1931
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001932 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001933 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 spin_unlock(&dentry->d_lock);
1935 break;
1936next:
1937 spin_unlock(&dentry->d_lock);
1938 }
1939 rcu_read_unlock();
1940
1941 return found;
1942}
1943
1944/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001945 * d_hash_and_lookup - hash the qstr then search for a dentry
1946 * @dir: Directory to search in
1947 * @name: qstr of name we wish to find
1948 *
1949 * On hash failure or on lookup failure NULL is returned.
1950 */
1951struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1952{
1953 struct dentry *dentry = NULL;
1954
1955 /*
1956 * Check for a fs-specific hash function. Note that we must
1957 * calculate the standard hash first, as the d_op->d_hash()
1958 * routine may choose to leave the hash value unchanged.
1959 */
1960 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11001961 if (dir->d_flags & DCACHE_OP_HASH) {
Nick Pigginb1e6a012011-01-07 17:49:28 +11001962 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001963 goto out;
1964 }
1965 dentry = d_lookup(dir, name);
1966out:
1967 return dentry;
1968}
1969
1970/**
Nick Piggin786a5e12011-01-07 17:49:16 +11001971 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlap1c977542010-11-18 15:02:45 -08001973 * @parent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 *
1975 * An insecure source has sent us a dentry, here we verify it and dget() it.
1976 * This is used by ncpfs in its readdir implementation.
1977 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11001978 *
1979 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 */
Nick Piggind3a23e12011-01-05 20:01:21 +11001981int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
Nick Piggin786a5e12011-01-07 17:49:16 +11001983 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11001984
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001985 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11001986 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1987 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001988 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11001989 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001990 spin_unlock(&dentry->d_lock);
1991 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return 1;
1993 }
1994 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001995 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11001996
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return 0;
1998}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001999EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
2001/*
2002 * When a file is deleted, we have two options:
2003 * - turn this dentry into a negative dentry
2004 * - unhash this dentry and free it.
2005 *
2006 * Usually, we want to just turn this into
2007 * a negative dentry, but if anybody else is
2008 * currently using the dentry or the inode
2009 * we can't do that and we fall back on removing
2010 * it from the hash queues and waiting for
2011 * it to be deleted later when it has no users
2012 */
2013
2014/**
2015 * d_delete - delete a dentry
2016 * @dentry: The dentry to delete
2017 *
2018 * Turn the dentry into a negative dentry if possible, otherwise
2019 * remove it from the hash queues so it can be deleted later
2020 */
2021
2022void d_delete(struct dentry * dentry)
2023{
Nick Piggin873feea2011-01-07 17:50:06 +11002024 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002025 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 /*
2027 * Are we the only user?
2028 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002029again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002031 inode = dentry->d_inode;
2032 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002033 if (dentry->d_count == 1) {
Nick Piggin873feea2011-01-07 17:50:06 +11002034 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002035 spin_unlock(&dentry->d_lock);
2036 cpu_relax();
2037 goto again;
2038 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002039 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002040 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002041 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 return;
2043 }
2044
2045 if (!d_unhashed(dentry))
2046 __d_drop(dentry);
2047
2048 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002049
2050 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002052EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002054static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002056 BUG_ON(!d_unhashed(entry));
2057 spin_lock_bucket(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 entry->d_flags &= ~DCACHE_UNHASHED;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002059 hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
2060 spin_unlock_bucket(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061}
2062
David Howells770bfad2006-08-22 20:06:07 -04002063static void _d_rehash(struct dentry * entry)
2064{
2065 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2066}
2067
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068/**
2069 * d_rehash - add an entry back to the hash
2070 * @entry: dentry to add to the hash
2071 *
2072 * Adds a dentry to the hash according to its name.
2073 */
2074
2075void d_rehash(struct dentry * entry)
2076{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002078 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002081EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002083/**
2084 * dentry_update_name_case - update case insensitive dentry with a new name
2085 * @dentry: dentry to be updated
2086 * @name: new name
2087 *
2088 * Update a case insensitive dentry with new case of name.
2089 *
2090 * dentry must have been returned by d_lookup with name @name. Old and new
2091 * name lengths must match (ie. no d_compare which allows mismatched name
2092 * lengths).
2093 *
2094 * Parent inode i_mutex must be held over d_lookup and into this call (to
2095 * keep renames and concurrent inserts, and readdir(2) away).
2096 */
2097void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2098{
2099 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2100 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2101
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002102 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002103 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002104 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002105 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002106 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002107}
2108EXPORT_SYMBOL(dentry_update_name_case);
2109
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110static void switch_names(struct dentry *dentry, struct dentry *target)
2111{
2112 if (dname_external(target)) {
2113 if (dname_external(dentry)) {
2114 /*
2115 * Both external: swap the pointers
2116 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002117 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 } else {
2119 /*
2120 * dentry:internal, target:external. Steal target's
2121 * storage and make target internal.
2122 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002123 memcpy(target->d_iname, dentry->d_name.name,
2124 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 dentry->d_name.name = target->d_name.name;
2126 target->d_name.name = target->d_iname;
2127 }
2128 } else {
2129 if (dname_external(dentry)) {
2130 /*
2131 * dentry:external, target:internal. Give dentry's
2132 * storage to target and make dentry internal
2133 */
2134 memcpy(dentry->d_iname, target->d_name.name,
2135 target->d_name.len + 1);
2136 target->d_name.name = dentry->d_name.name;
2137 dentry->d_name.name = dentry->d_iname;
2138 } else {
2139 /*
2140 * Both are internal. Just copy target to dentry
2141 */
2142 memcpy(dentry->d_iname, target->d_name.name,
2143 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002144 dentry->d_name.len = target->d_name.len;
2145 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 }
2147 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002148 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149}
2150
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002151static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2152{
2153 /*
2154 * XXXX: do we really need to take target->d_lock?
2155 */
2156 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2157 spin_lock(&target->d_parent->d_lock);
2158 else {
2159 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2160 spin_lock(&dentry->d_parent->d_lock);
2161 spin_lock_nested(&target->d_parent->d_lock,
2162 DENTRY_D_LOCK_NESTED);
2163 } else {
2164 spin_lock(&target->d_parent->d_lock);
2165 spin_lock_nested(&dentry->d_parent->d_lock,
2166 DENTRY_D_LOCK_NESTED);
2167 }
2168 }
2169 if (target < dentry) {
2170 spin_lock_nested(&target->d_lock, 2);
2171 spin_lock_nested(&dentry->d_lock, 3);
2172 } else {
2173 spin_lock_nested(&dentry->d_lock, 2);
2174 spin_lock_nested(&target->d_lock, 3);
2175 }
2176}
2177
2178static void dentry_unlock_parents_for_move(struct dentry *dentry,
2179 struct dentry *target)
2180{
2181 if (target->d_parent != dentry->d_parent)
2182 spin_unlock(&dentry->d_parent->d_lock);
2183 if (target->d_parent != target)
2184 spin_unlock(&target->d_parent->d_lock);
2185}
2186
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002188 * When switching names, the actual string doesn't strictly have to
2189 * be preserved in the target - because we're dropping the target
2190 * anyway. As such, we can just do a simple memcpy() to copy over
2191 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002193 * Note that we have to be a lot more careful about getting the hash
2194 * switched - we have to switch the hash value properly even if it
2195 * then no longer matches the actual (corrupted) string of the target.
2196 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002198/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002199 * d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 * @dentry: entry to move
2201 * @target: new dentry
2202 *
2203 * Update the dcache to reflect the move of a file name. Negative
2204 * dcache entries should not be moved in this way.
2205 */
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002206void d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 if (!dentry->d_inode)
2209 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2210
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002211 BUG_ON(d_ancestor(dentry, target));
2212 BUG_ON(d_ancestor(target, dentry));
2213
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 write_seqlock(&rename_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002215
2216 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Nick Piggin31e6b012011-01-07 17:49:52 +11002218 write_seqcount_begin(&dentry->d_seq);
2219 write_seqcount_begin(&target->d_seq);
2220
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002221 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2222
2223 /*
2224 * Move the dentry to the target hash queue. Don't bother checking
2225 * for the same hash queue because of how unlikely it is.
2226 */
2227 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002228 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
2230 /* Unhash the target: dput() will then get rid of it */
2231 __d_drop(target);
2232
Eric Dumazet5160ee62006-01-08 01:03:32 -08002233 list_del(&dentry->d_u.d_child);
2234 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235
2236 /* Switch the names.. */
2237 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002238 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
2240 /* ... and switch the parents */
2241 if (IS_ROOT(dentry)) {
2242 dentry->d_parent = target->d_parent;
2243 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002244 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002246 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247
2248 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002249 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 }
2251
Eric Dumazet5160ee62006-01-08 01:03:32 -08002252 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002253
Nick Piggin31e6b012011-01-07 17:49:52 +11002254 write_seqcount_end(&target->d_seq);
2255 write_seqcount_end(&dentry->d_seq);
2256
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002257 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002259 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 spin_unlock(&dentry->d_lock);
2261 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002262}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002263EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002265/**
2266 * d_ancestor - search for an ancestor
2267 * @p1: ancestor dentry
2268 * @p2: child dentry
2269 *
2270 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2271 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002272 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002273struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002274{
2275 struct dentry *p;
2276
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002277 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002278 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002279 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002280 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002281 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002282}
2283
2284/*
2285 * This helper attempts to cope with remotely renamed directories
2286 *
2287 * It assumes that the caller is already holding
Nick Piggin873feea2011-01-07 17:50:06 +11002288 * dentry->d_parent->d_inode->i_mutex and the inode->i_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002289 *
2290 * Note: If ever the locking in lock_rename() changes, then please
2291 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002292 */
Nick Piggin873feea2011-01-07 17:50:06 +11002293static struct dentry *__d_unalias(struct inode *inode,
2294 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002295{
2296 struct mutex *m1 = NULL, *m2 = NULL;
2297 struct dentry *ret;
2298
2299 /* If alias and dentry share a parent, then no extra locks required */
2300 if (alias->d_parent == dentry->d_parent)
2301 goto out_unalias;
2302
2303 /* Check for loops */
2304 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002305 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002306 goto out_err;
2307
2308 /* See lock_rename() */
2309 ret = ERR_PTR(-EBUSY);
2310 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2311 goto out_err;
2312 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2313 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2314 goto out_err;
2315 m2 = &alias->d_parent->d_inode->i_mutex;
2316out_unalias:
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002317 d_move(alias, dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002318 ret = alias;
2319out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002320 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002321 if (m2)
2322 mutex_unlock(m2);
2323 if (m1)
2324 mutex_unlock(m1);
2325 return ret;
2326}
2327
2328/*
David Howells770bfad2006-08-22 20:06:07 -04002329 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2330 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002331 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002332 */
2333static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2334{
2335 struct dentry *dparent, *aparent;
2336
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002337 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002338
Nick Piggin31e6b012011-01-07 17:49:52 +11002339 write_seqcount_begin(&dentry->d_seq);
2340 write_seqcount_begin(&anon->d_seq);
2341
David Howells770bfad2006-08-22 20:06:07 -04002342 dparent = dentry->d_parent;
2343 aparent = anon->d_parent;
2344
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002345 switch_names(dentry, anon);
2346 swap(dentry->d_name.hash, anon->d_name.hash);
2347
David Howells770bfad2006-08-22 20:06:07 -04002348 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2349 list_del(&dentry->d_u.d_child);
2350 if (!IS_ROOT(dentry))
2351 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2352 else
2353 INIT_LIST_HEAD(&dentry->d_u.d_child);
2354
2355 anon->d_parent = (dparent == dentry) ? anon : dparent;
2356 list_del(&anon->d_u.d_child);
2357 if (!IS_ROOT(anon))
2358 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2359 else
2360 INIT_LIST_HEAD(&anon->d_u.d_child);
2361
Nick Piggin31e6b012011-01-07 17:49:52 +11002362 write_seqcount_end(&dentry->d_seq);
2363 write_seqcount_end(&anon->d_seq);
2364
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002365 dentry_unlock_parents_for_move(anon, dentry);
2366 spin_unlock(&dentry->d_lock);
2367
2368 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002369 anon->d_flags &= ~DCACHE_DISCONNECTED;
2370}
2371
2372/**
2373 * d_materialise_unique - introduce an inode into the tree
2374 * @dentry: candidate dentry
2375 * @inode: inode to bind to the dentry, to which aliases may be attached
2376 *
2377 * Introduces an dentry into the tree, substituting an extant disconnected
2378 * root directory alias in its place if there is one
2379 */
2380struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2381{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002382 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002383
2384 BUG_ON(!d_unhashed(dentry));
2385
David Howells770bfad2006-08-22 20:06:07 -04002386 if (!inode) {
2387 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002388 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002389 d_rehash(actual);
2390 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002391 }
2392
Nick Piggin873feea2011-01-07 17:50:06 +11002393 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002394
Trond Myklebust9eaef272006-10-21 10:24:20 -07002395 if (S_ISDIR(inode->i_mode)) {
2396 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002397
Trond Myklebust9eaef272006-10-21 10:24:20 -07002398 /* Does an aliased dentry already exist? */
2399 alias = __d_find_alias(inode, 0);
2400 if (alias) {
2401 actual = alias;
2402 /* Is this an anonymous mountpoint that we could splice
2403 * into our tree? */
2404 if (IS_ROOT(alias)) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002405 __d_materialise_dentry(dentry, alias);
2406 __d_drop(alias);
2407 goto found;
2408 }
2409 /* Nope, but we must(!) avoid directory aliasing */
Nick Piggin873feea2011-01-07 17:50:06 +11002410 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002411 if (IS_ERR(actual))
2412 dput(alias);
2413 goto out_nolock;
2414 }
David Howells770bfad2006-08-22 20:06:07 -04002415 }
2416
2417 /* Add a unique reference */
2418 actual = __d_instantiate_unique(dentry, inode);
2419 if (!actual)
2420 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002421 else
2422 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002423
David Howells770bfad2006-08-22 20:06:07 -04002424 spin_lock(&actual->d_lock);
2425found:
2426 _d_rehash(actual);
2427 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002428 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002429out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002430 if (actual == dentry) {
2431 security_d_instantiate(dentry, inode);
2432 return NULL;
2433 }
2434
2435 iput(inode);
2436 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002437}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002438EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002439
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002440static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002441{
2442 *buflen -= namelen;
2443 if (*buflen < 0)
2444 return -ENAMETOOLONG;
2445 *buffer -= namelen;
2446 memcpy(*buffer, str, namelen);
2447 return 0;
2448}
2449
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002450static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2451{
2452 return prepend(buffer, buflen, name->name, name->len);
2453}
2454
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002456 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002457 * @path: the dentry/vfsmount to report
2458 * @root: root vfsmnt/dentry (may be modified by this function)
2459 * @buffer: pointer to the end of the buffer
2460 * @buflen: pointer to buffer length
2461 *
Nick Piggin949854d2011-01-07 17:49:37 +11002462 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002463 *
2464 * If path is not reachable from the supplied root, then the value of
2465 * root is changed (without modifying refcounts).
2466 */
2467static int prepend_path(const struct path *path, struct path *root,
2468 char **buffer, int *buflen)
2469{
2470 struct dentry *dentry = path->dentry;
2471 struct vfsmount *vfsmnt = path->mnt;
2472 bool slash = false;
2473 int error = 0;
2474
Nick Piggin99b7db72010-08-18 04:37:39 +10002475 br_read_lock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002476 while (dentry != root->dentry || vfsmnt != root->mnt) {
2477 struct dentry * parent;
2478
2479 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2480 /* Global root? */
2481 if (vfsmnt->mnt_parent == vfsmnt) {
2482 goto global_root;
2483 }
2484 dentry = vfsmnt->mnt_mountpoint;
2485 vfsmnt = vfsmnt->mnt_parent;
2486 continue;
2487 }
2488 parent = dentry->d_parent;
2489 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002490 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002491 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002492 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002493 if (!error)
2494 error = prepend(buffer, buflen, "/", 1);
2495 if (error)
2496 break;
2497
2498 slash = true;
2499 dentry = parent;
2500 }
2501
2502out:
2503 if (!error && !slash)
2504 error = prepend(buffer, buflen, "/", 1);
2505
Nick Piggin99b7db72010-08-18 04:37:39 +10002506 br_read_unlock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002507 return error;
2508
2509global_root:
2510 /*
2511 * Filesystems needing to implement special "root names"
2512 * should do so with ->d_dname()
2513 */
2514 if (IS_ROOT(dentry) &&
2515 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2516 WARN(1, "Root dentry has weird name <%.*s>\n",
2517 (int) dentry->d_name.len, dentry->d_name.name);
2518 }
2519 root->mnt = vfsmnt;
2520 root->dentry = dentry;
2521 goto out;
2522}
2523
2524/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002525 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002526 * @path: the dentry/vfsmount to report
2527 * @root: root vfsmnt/dentry (may be modified by this function)
Randy Dunlapcd956a12010-08-14 13:05:31 -07002528 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 * @buflen: buffer length
2530 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002531 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002533 * Returns a pointer into the buffer or an error code if the
2534 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002535 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002536 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002537 *
2538 * If path is not reachable from the supplied root, then the value of
2539 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002541char *__d_path(const struct path *path, struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002542 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002544 char *res = buf + buflen;
2545 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002547 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002548 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002549 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002550 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002551
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002552 if (error)
2553 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002554 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555}
2556
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002557/*
2558 * same as __d_path but appends "(deleted)" for unlinked files.
2559 */
2560static int path_with_deleted(const struct path *path, struct path *root,
2561 char **buf, int *buflen)
2562{
2563 prepend(buf, buflen, "\0", 1);
2564 if (d_unlinked(path->dentry)) {
2565 int error = prepend(buf, buflen, " (deleted)", 10);
2566 if (error)
2567 return error;
2568 }
2569
2570 return prepend_path(path, root, buf, buflen);
2571}
2572
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002573static int prepend_unreachable(char **buffer, int *buflen)
2574{
2575 return prepend(buffer, buflen, "(unreachable)", 13);
2576}
2577
Jan Bluncka03a8a702008-02-14 19:38:32 -08002578/**
2579 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002580 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002581 * @buf: buffer to return value in
2582 * @buflen: buffer length
2583 *
2584 * Convert a dentry into an ASCII path name. If the entry has been deleted
2585 * the string " (deleted)" is appended. Note that this is ambiguous.
2586 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002587 * Returns a pointer into the buffer or an error code if the path was
2588 * too long. Note: Callers should use the returned pointer, not the passed
2589 * in buffer, to use the name! The implementation often starts at an offset
2590 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002591 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002592 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002593 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002594char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002596 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002597 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002598 struct path tmp;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002599 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002601 /*
2602 * We have various synthetic filesystems that never get mounted. On
2603 * these filesystems dentries are never used for lookup purposes, and
2604 * thus don't need to be hashed. They also don't need a name until a
2605 * user wants to identify the object in /proc/pid/fd/. The little hack
2606 * below allows us to generate a name for these objects on demand:
2607 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002608 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2609 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002610
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002611 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002612 write_seqlock(&rename_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002613 tmp = root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002614 error = path_with_deleted(path, &tmp, &res, &buflen);
2615 if (error)
2616 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002617 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002618 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 return res;
2620}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002621EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002623/**
2624 * d_path_with_unreachable - return the path of a dentry
2625 * @path: path to report
2626 * @buf: buffer to return value in
2627 * @buflen: buffer length
2628 *
2629 * The difference from d_path() is that this prepends "(unreachable)"
2630 * to paths which are unreachable from the current process' root.
2631 */
2632char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2633{
2634 char *res = buf + buflen;
2635 struct path root;
2636 struct path tmp;
2637 int error;
2638
2639 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2640 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2641
2642 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002643 write_seqlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002644 tmp = root;
2645 error = path_with_deleted(path, &tmp, &res, &buflen);
2646 if (!error && !path_equal(&tmp, &root))
2647 error = prepend_unreachable(&res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002648 write_sequnlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002649 path_put(&root);
2650 if (error)
2651 res = ERR_PTR(error);
2652
2653 return res;
2654}
2655
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002657 * Helper function for dentry_operations.d_dname() members
2658 */
2659char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2660 const char *fmt, ...)
2661{
2662 va_list args;
2663 char temp[64];
2664 int sz;
2665
2666 va_start(args, fmt);
2667 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2668 va_end(args);
2669
2670 if (sz > sizeof(temp) || sz > buflen)
2671 return ERR_PTR(-ENAMETOOLONG);
2672
2673 buffer += buflen - sz;
2674 return memcpy(buffer, temp, sz);
2675}
2676
2677/*
Ram Pai6092d042008-03-27 13:06:20 +01002678 * Write full pathname from the root of the filesystem into the buffer.
2679 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002680static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002681{
2682 char *end = buf + buflen;
2683 char *retval;
2684
Ram Pai6092d042008-03-27 13:06:20 +01002685 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002686 if (buflen < 1)
2687 goto Elong;
2688 /* Get '/' right */
2689 retval = end-1;
2690 *retval = '/';
2691
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002692 while (!IS_ROOT(dentry)) {
2693 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002694 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002695
Ram Pai6092d042008-03-27 13:06:20 +01002696 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002697 spin_lock(&dentry->d_lock);
2698 error = prepend_name(&end, &buflen, &dentry->d_name);
2699 spin_unlock(&dentry->d_lock);
2700 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002701 goto Elong;
2702
2703 retval = end;
2704 dentry = parent;
2705 }
Al Viroc1031352010-06-06 22:31:14 -04002706 return retval;
2707Elong:
2708 return ERR_PTR(-ENAMETOOLONG);
2709}
Nick Pigginec2447c2011-01-07 17:49:29 +11002710
2711char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2712{
2713 char *retval;
2714
Nick Piggin949854d2011-01-07 17:49:37 +11002715 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002716 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002717 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002718
2719 return retval;
2720}
2721EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002722
2723char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2724{
2725 char *p = NULL;
2726 char *retval;
2727
Nick Piggin949854d2011-01-07 17:49:37 +11002728 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002729 if (d_unlinked(dentry)) {
2730 p = buf + buflen;
2731 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2732 goto Elong;
2733 buflen++;
2734 }
2735 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002736 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002737 if (!IS_ERR(retval) && p)
2738 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002739 return retval;
2740Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002741 return ERR_PTR(-ENAMETOOLONG);
2742}
2743
2744/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 * NOTE! The user-level library version returns a
2746 * character pointer. The kernel system call just
2747 * returns the length of the buffer filled (which
2748 * includes the ending '\0' character), or a negative
2749 * error value. So libc would do something like
2750 *
2751 * char *getcwd(char * buf, size_t size)
2752 * {
2753 * int retval;
2754 *
2755 * retval = sys_getcwd(buf, size);
2756 * if (retval >= 0)
2757 * return buf;
2758 * errno = -retval;
2759 * return NULL;
2760 * }
2761 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002762SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763{
Linus Torvalds552ce542007-02-13 12:08:18 -08002764 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002765 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002766 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767
2768 if (!page)
2769 return -ENOMEM;
2770
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002771 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772
Linus Torvalds552ce542007-02-13 12:08:18 -08002773 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002774 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002775 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002776 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002777 struct path tmp = root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002778 char *cwd = page + PAGE_SIZE;
2779 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002781 prepend(&cwd, &buflen, "\0", 1);
2782 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002783 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002784
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002785 if (error)
Linus Torvalds552ce542007-02-13 12:08:18 -08002786 goto out;
2787
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002788 /* Unreachable from current root */
2789 if (!path_equal(&tmp, &root)) {
2790 error = prepend_unreachable(&cwd, &buflen);
2791 if (error)
2792 goto out;
2793 }
2794
Linus Torvalds552ce542007-02-13 12:08:18 -08002795 error = -ERANGE;
2796 len = PAGE_SIZE + page - cwd;
2797 if (len <= size) {
2798 error = len;
2799 if (copy_to_user(buf, cwd, len))
2800 error = -EFAULT;
2801 }
Nick Piggin949854d2011-01-07 17:49:37 +11002802 } else {
2803 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805
2806out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002807 path_put(&pwd);
2808 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 free_page((unsigned long) page);
2810 return error;
2811}
2812
2813/*
2814 * Test whether new_dentry is a subdirectory of old_dentry.
2815 *
2816 * Trivially implemented using the dcache structure
2817 */
2818
2819/**
2820 * is_subdir - is new dentry a subdirectory of old_dentry
2821 * @new_dentry: new dentry
2822 * @old_dentry: old dentry
2823 *
2824 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2825 * Returns 0 otherwise.
2826 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2827 */
2828
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002829int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830{
2831 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002832 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002834 if (new_dentry == old_dentry)
2835 return 1;
2836
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002837 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002840 /*
2841 * Need rcu_readlock to protect against the d_parent trashing
2842 * due to d_move
2843 */
2844 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002845 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002847 else
2848 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002849 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
2852 return result;
2853}
2854
Al Viro2096f752010-01-30 13:16:21 -05002855int path_is_under(struct path *path1, struct path *path2)
2856{
2857 struct vfsmount *mnt = path1->mnt;
2858 struct dentry *dentry = path1->dentry;
2859 int res;
Nick Piggin99b7db72010-08-18 04:37:39 +10002860
2861 br_read_lock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002862 if (mnt != path2->mnt) {
2863 for (;;) {
2864 if (mnt->mnt_parent == mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +10002865 br_read_unlock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002866 return 0;
2867 }
2868 if (mnt->mnt_parent == path2->mnt)
2869 break;
2870 mnt = mnt->mnt_parent;
2871 }
2872 dentry = mnt->mnt_mountpoint;
2873 }
2874 res = is_subdir(dentry, path2->dentry);
Nick Piggin99b7db72010-08-18 04:37:39 +10002875 br_read_unlock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002876 return res;
2877}
2878EXPORT_SYMBOL(path_is_under);
2879
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880void d_genocide(struct dentry *root)
2881{
Nick Piggin949854d2011-01-07 17:49:37 +11002882 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002884 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002885 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886
Nick Piggin949854d2011-01-07 17:49:37 +11002887 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002888again:
2889 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002890 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891repeat:
2892 next = this_parent->d_subdirs.next;
2893resume:
2894 while (next != &this_parent->d_subdirs) {
2895 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002896 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002898
Nick Pigginda502952011-01-07 17:49:33 +11002899 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2900 if (d_unhashed(dentry) || !dentry->d_inode) {
2901 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002905 spin_unlock(&this_parent->d_lock);
2906 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002908 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 goto repeat;
2910 }
Nick Piggin949854d2011-01-07 17:49:37 +11002911 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2912 dentry->d_flags |= DCACHE_GENOCIDE;
2913 dentry->d_count--;
2914 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002915 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 }
2917 if (this_parent != root) {
Nick Piggin949854d2011-01-07 17:49:37 +11002918 struct dentry *tmp;
2919 struct dentry *child;
2920
2921 tmp = this_parent->d_parent;
2922 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2923 this_parent->d_flags |= DCACHE_GENOCIDE;
2924 this_parent->d_count--;
2925 }
2926 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002927 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002928 child = this_parent;
2929 this_parent = tmp;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002930 spin_lock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002931 /* might go back up the wrong parent if we have had a rename
2932 * or deletion */
2933 if (this_parent != child->d_parent ||
Nick Piggin58db63d2011-01-07 17:49:39 +11002934 (!locked && read_seqretry(&rename_lock, seq))) {
Nick Piggin949854d2011-01-07 17:49:37 +11002935 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002936 rcu_read_unlock();
2937 goto rename_retry;
2938 }
2939 rcu_read_unlock();
2940 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 goto resume;
2942 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002943 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002944 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11002945 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11002946 if (locked)
2947 write_sequnlock(&rename_lock);
2948 return;
2949
2950rename_retry:
2951 locked = 1;
2952 write_seqlock(&rename_lock);
2953 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954}
2955
2956/**
2957 * find_inode_number - check for dentry with name
2958 * @dir: directory to check
2959 * @name: Name to find.
2960 *
2961 * Check whether a dentry already exists for the given name,
2962 * and return the inode number if it has an inode. Otherwise
2963 * 0 is returned.
2964 *
2965 * This routine is used to post-process directory listings for
2966 * filesystems using synthetic inode numbers, and is necessary
2967 * to keep getcwd() working.
2968 */
2969
2970ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2971{
2972 struct dentry * dentry;
2973 ino_t ino = 0;
2974
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002975 dentry = d_hash_and_lookup(dir, name);
2976 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 if (dentry->d_inode)
2978 ino = dentry->d_inode->i_ino;
2979 dput(dentry);
2980 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 return ino;
2982}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002983EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984
2985static __initdata unsigned long dhash_entries;
2986static int __init set_dhash_entries(char *str)
2987{
2988 if (!str)
2989 return 0;
2990 dhash_entries = simple_strtoul(str, &str, 0);
2991 return 1;
2992}
2993__setup("dhash_entries=", set_dhash_entries);
2994
2995static void __init dcache_init_early(void)
2996{
2997 int loop;
2998
2999 /* If hashes are distributed across NUMA nodes, defer
3000 * hash allocation until vmalloc space is available.
3001 */
3002 if (hashdist)
3003 return;
3004
3005 dentry_hashtable =
3006 alloc_large_system_hash("Dentry cache",
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003007 sizeof(struct dcache_hash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 dhash_entries,
3009 13,
3010 HASH_EARLY,
3011 &d_hash_shift,
3012 &d_hash_mask,
3013 0);
3014
3015 for (loop = 0; loop < (1 << d_hash_shift); loop++)
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003016 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017}
3018
Denis Cheng74bf17c2007-10-16 23:26:30 -07003019static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020{
3021 int loop;
3022
3023 /*
3024 * A constructor could be added for stable state like the lists,
3025 * but it is probably not worth it because of the cache nature
3026 * of the dcache.
3027 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003028 dentry_cache = KMEM_CACHE(dentry,
3029 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030
Rusty Russell8e1f9362007-07-17 04:03:17 -07003031 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032
3033 /* Hash may have been set up in dcache_init_early */
3034 if (!hashdist)
3035 return;
3036
3037 dentry_hashtable =
3038 alloc_large_system_hash("Dentry cache",
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003039 sizeof(struct dcache_hash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 dhash_entries,
3041 13,
3042 0,
3043 &d_hash_shift,
3044 &d_hash_mask,
3045 0);
3046
3047 for (loop = 0; loop < (1 << d_hash_shift); loop++)
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003048 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049}
3050
3051/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003052struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003053EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055EXPORT_SYMBOL(d_genocide);
3056
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057void __init vfs_caches_init_early(void)
3058{
3059 dcache_init_early();
3060 inode_init_early();
3061}
3062
3063void __init vfs_caches_init(unsigned long mempages)
3064{
3065 unsigned long reserve;
3066
3067 /* Base hash sizes on available memory, with a reserve equal to
3068 150% of current kernel size */
3069
3070 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3071 mempages -= reserve;
3072
3073 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003074 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075
Denis Cheng74bf17c2007-10-16 23:26:30 -07003076 dcache_init();
3077 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003079 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 bdev_cache_init();
3081 chrdev_init();
3082}