blob: 2a6bd9a4ae975fdc3eedf9723b551fc9e5f65836 [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
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800179 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100180 * After this call, in-progress rcu-walk path lookup will fail. This
181 * should be called after unhashing, and after changing d_inode (if
182 * the dentry has not already been unhashed).
183 */
184static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
185{
186 assert_spin_locked(&dentry->d_lock);
187 /* Go through a barrier */
188 write_seqcount_barrier(&dentry->d_seq);
189}
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191/*
192 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100193 * d_iput() operation if defined. Dentry has no refcount
194 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800196static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200197 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100198 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
200 struct inode *inode = dentry->d_inode;
201 if (inode) {
202 dentry->d_inode = NULL;
203 list_del_init(&dentry->d_alias);
204 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100205 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700206 if (!inode->i_nlink)
207 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 if (dentry->d_op && dentry->d_op->d_iput)
209 dentry->d_op->d_iput(dentry, inode);
210 else
211 iput(inode);
212 } else {
213 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 }
215}
216
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700217/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100218 * Release the dentry's inode, using the filesystem
219 * d_iput() operation if defined. dentry remains in-use.
220 */
221static void dentry_unlink_inode(struct dentry * dentry)
222 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100223 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100224{
225 struct inode *inode = dentry->d_inode;
226 dentry->d_inode = NULL;
227 list_del_init(&dentry->d_alias);
228 dentry_rcuwalk_barrier(dentry);
229 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100230 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100231 if (!inode->i_nlink)
232 fsnotify_inoderemove(inode);
233 if (dentry->d_op && dentry->d_op->d_iput)
234 dentry->d_op->d_iput(dentry, inode);
235 else
236 iput(inode);
237}
238
239/*
Nick Piggin23044502011-01-07 17:49:31 +1100240 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700241 */
242static void dentry_lru_add(struct dentry *dentry)
243{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400244 if (list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100245 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400246 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
247 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100248 dentry_stat.nr_unused++;
Nick Piggin23044502011-01-07 17:49:31 +1100249 spin_unlock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400250 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700251}
252
Nick Piggin23044502011-01-07 17:49:31 +1100253static void __dentry_lru_del(struct dentry *dentry)
254{
255 list_del_init(&dentry->d_lru);
256 dentry->d_sb->s_nr_dentry_unused--;
257 dentry_stat.nr_unused--;
258}
259
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700260static void dentry_lru_del(struct dentry *dentry)
261{
262 if (!list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100263 spin_lock(&dcache_lru_lock);
264 __dentry_lru_del(dentry);
265 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700266 }
267}
268
Christoph Hellwiga4633352010-10-10 05:36:26 -0400269static void dentry_lru_move_tail(struct dentry *dentry)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700270{
Nick Piggin23044502011-01-07 17:49:31 +1100271 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400272 if (list_empty(&dentry->d_lru)) {
273 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
274 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100275 dentry_stat.nr_unused++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400276 } else {
277 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700278 }
Nick Piggin23044502011-01-07 17:49:31 +1100279 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700280}
281
Miklos Szeredid52b9082007-05-08 00:23:46 -0700282/**
283 * d_kill - kill dentry and return parent
284 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800285 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700286 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200287 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700288 *
289 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100290 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100291 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
292 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700293 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100294static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200295 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100296 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100297 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700298{
Nick Piggin949854d2011-01-07 17:49:37 +1100299 dentry->d_parent = NULL;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700300 list_del(&dentry->d_u.d_child);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100301 if (parent)
302 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700303 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100304 /*
305 * dentry_iput drops the locks, at which point nobody (except
306 * transient RCU lookups) can reach this dentry.
307 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700308 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900309 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700310}
311
Nick Piggin789680d2011-01-07 17:49:30 +1100312/**
313 * d_drop - drop a dentry
314 * @dentry: dentry to drop
315 *
316 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
317 * be found through a VFS lookup any more. Note that this is different from
318 * deleting the dentry - d_delete will try to mark the dentry negative if
319 * possible, giving a successful _negative_ lookup, while d_drop will
320 * just make the cache lookup fail.
321 *
322 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
323 * reason (NFS timeouts or autofs deletes).
324 *
325 * __d_drop requires dentry->d_lock.
326 */
327void __d_drop(struct dentry *dentry)
328{
329 if (!(dentry->d_flags & DCACHE_UNHASHED)) {
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100330 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) {
331 bit_spin_lock(0,
332 (unsigned long *)&dentry->d_sb->s_anon.first);
333 dentry->d_flags |= DCACHE_UNHASHED;
334 hlist_bl_del_init(&dentry->d_hash);
335 __bit_spin_unlock(0,
336 (unsigned long *)&dentry->d_sb->s_anon.first);
337 } else {
338 struct dcache_hash_bucket *b;
339 b = d_hash(dentry->d_parent, dentry->d_name.hash);
340 spin_lock_bucket(b);
341 /*
342 * We may not actually need to put DCACHE_UNHASHED
343 * manipulations under the hash lock, but follow
344 * the principle of least surprise.
345 */
346 dentry->d_flags |= DCACHE_UNHASHED;
347 hlist_bl_del_rcu(&dentry->d_hash);
348 spin_unlock_bucket(b);
349 dentry_rcuwalk_barrier(dentry);
350 }
Nick Piggin789680d2011-01-07 17:49:30 +1100351 }
352}
353EXPORT_SYMBOL(__d_drop);
354
355void d_drop(struct dentry *dentry)
356{
Nick Piggin789680d2011-01-07 17:49:30 +1100357 spin_lock(&dentry->d_lock);
358 __d_drop(dentry);
359 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100360}
361EXPORT_SYMBOL(d_drop);
362
Nick Piggin77812a12011-01-07 17:49:48 +1100363/*
364 * Finish off a dentry we've decided to kill.
365 * dentry->d_lock must be held, returns with it unlocked.
366 * If ref is non-zero, then decrement the refcount too.
367 * Returns dentry requiring refcount drop, or NULL if we're done.
368 */
369static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
370 __releases(dentry->d_lock)
371{
Nick Piggin873feea2011-01-07 17:50:06 +1100372 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100373 struct dentry *parent;
374
Nick Piggin873feea2011-01-07 17:50:06 +1100375 inode = dentry->d_inode;
376 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100377relock:
378 spin_unlock(&dentry->d_lock);
379 cpu_relax();
380 return dentry; /* try again with same dentry */
381 }
382 if (IS_ROOT(dentry))
383 parent = NULL;
384 else
385 parent = dentry->d_parent;
386 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100387 if (inode)
388 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100389 goto relock;
390 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100391
Nick Piggin77812a12011-01-07 17:49:48 +1100392 if (ref)
393 dentry->d_count--;
394 /* if dentry was on the d_lru list delete it from there */
395 dentry_lru_del(dentry);
396 /* if it was on the hash then remove it */
397 __d_drop(dentry);
398 return d_kill(dentry, parent);
399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401/*
402 * This is dput
403 *
404 * This is complicated by the fact that we do not want to put
405 * dentries that are no longer on any hash chain on the unused
406 * list: we'd much rather just get rid of them immediately.
407 *
408 * However, that implies that we have to traverse the dentry
409 * tree upwards to the parents which might _also_ now be
410 * scheduled for deletion (it may have been only waiting for
411 * its last child to go away).
412 *
413 * This tail recursion is done by hand as we don't want to depend
414 * on the compiler to always get this right (gcc generally doesn't).
415 * Real recursion would eat up our stack space.
416 */
417
418/*
419 * dput - release a dentry
420 * @dentry: dentry to release
421 *
422 * Release a dentry. This will drop the usage count and if appropriate
423 * call the dentry unlink method as well as removing it from the queues and
424 * releasing its resources. If the parent dentries were scheduled for release
425 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427void dput(struct dentry *dentry)
428{
429 if (!dentry)
430 return;
431
432repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100433 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100436 BUG_ON(!dentry->d_count);
437 if (dentry->d_count > 1) {
438 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 return;
441 }
442
Nick Pigginfb045ad2011-01-07 17:49:55 +1100443 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100445 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 }
Nick Piggin265ac902010-10-10 05:36:24 -0400447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 /* Unreachable? Get rid of it */
449 if (d_unhashed(dentry))
450 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400451
452 /* Otherwise leave it cached and ensure it's on the LRU */
453 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400454 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400455
Nick Piggin61f3dee2011-01-07 17:49:40 +1100456 dentry->d_count--;
457 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 return;
459
Miklos Szeredid52b9082007-05-08 00:23:46 -0700460kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100461 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700462 if (dentry)
463 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700465EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
467/**
468 * d_invalidate - invalidate a dentry
469 * @dentry: dentry to invalidate
470 *
471 * Try to invalidate the dentry if it turns out to be
472 * possible. If there are other dentries that can be
473 * reached through this one we can't delete it and we
474 * return -EBUSY. On success we return 0.
475 *
476 * no dcache lock.
477 */
478
479int d_invalidate(struct dentry * dentry)
480{
481 /*
482 * If it's already been dropped, return OK.
483 */
Nick Pigginda502952011-01-07 17:49:33 +1100484 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100486 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return 0;
488 }
489 /*
490 * Check whether to do a partial shrink_dcache
491 * to get rid of unused child entries.
492 */
493 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100494 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100496 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 }
498
499 /*
500 * Somebody else still using it?
501 *
502 * If it's a directory, we can't drop it
503 * for fear of somebody re-populating it
504 * with children (even though dropping it
505 * would make it unreachable from the root,
506 * we might still populate it if it was a
507 * working directory or similar).
508 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100509 if (dentry->d_count > 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
511 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return -EBUSY;
513 }
514 }
515
516 __d_drop(dentry);
517 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return 0;
519}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700520EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100522/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100523static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100525 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
Nick Piggindc0474b2011-01-07 17:49:43 +1100528static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100529{
Nick Piggin23044502011-01-07 17:49:31 +1100530 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100531 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100532 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100533}
534
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100535struct dentry *dget_parent(struct dentry *dentry)
536{
537 struct dentry *ret;
538
539repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100540 /*
541 * Don't need rcu_dereference because we re-check it was correct under
542 * the lock.
543 */
544 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100545 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100546 if (!ret) {
547 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100548 goto out;
549 }
Nick Piggina734eb42011-01-07 17:49:44 +1100550 spin_lock(&ret->d_lock);
551 if (unlikely(ret != dentry->d_parent)) {
552 spin_unlock(&ret->d_lock);
553 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100554 goto repeat;
555 }
Nick Piggina734eb42011-01-07 17:49:44 +1100556 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100557 BUG_ON(!ret->d_count);
558 ret->d_count++;
559 spin_unlock(&ret->d_lock);
560out:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100561 return ret;
562}
563EXPORT_SYMBOL(dget_parent);
564
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565/**
566 * d_find_alias - grab a hashed alias of inode
567 * @inode: inode in question
568 * @want_discon: flag, used by d_splice_alias, to request
569 * that only a DISCONNECTED alias be returned.
570 *
571 * If inode has a hashed alias, or is a directory and has any alias,
572 * acquire the reference to alias and return it. Otherwise return NULL.
573 * Notice that if inode is a directory there can be only one alias and
574 * it can be unhashed only if it has no children, or if it is the root
575 * of a filesystem.
576 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700577 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700579 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 */
Nick Pigginda502952011-01-07 17:49:33 +1100581static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
Nick Pigginda502952011-01-07 17:49:33 +1100583 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Nick Pigginda502952011-01-07 17:49:33 +1100585again:
586 discon_alias = NULL;
587 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
588 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700590 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100591 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 discon_alias = alias;
Nick Pigginda502952011-01-07 17:49:33 +1100593 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100594 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100595 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return alias;
597 }
598 }
Nick Pigginda502952011-01-07 17:49:33 +1100599 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 }
Nick Pigginda502952011-01-07 17:49:33 +1100601 if (discon_alias) {
602 alias = discon_alias;
603 spin_lock(&alias->d_lock);
604 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
605 if (IS_ROOT(alias) &&
606 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100607 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100608 spin_unlock(&alias->d_lock);
609 return alias;
610 }
611 }
612 spin_unlock(&alias->d_lock);
613 goto again;
614 }
615 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616}
617
Nick Pigginda502952011-01-07 17:49:33 +1100618struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
David Howells214fda12006-03-25 03:06:36 -0800620 struct dentry *de = NULL;
621
622 if (!list_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100623 spin_lock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800624 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100625 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return de;
628}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700629EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
631/*
632 * Try to kill dentries associated with this inode.
633 * WARNING: you must own a reference to inode.
634 */
635void d_prune_aliases(struct inode *inode)
636{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700637 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100639 spin_lock(&inode->i_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700640 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100642 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100643 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 __d_drop(dentry);
645 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100646 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 dput(dentry);
648 goto restart;
649 }
650 spin_unlock(&dentry->d_lock);
651 }
Nick Piggin873feea2011-01-07 17:50:06 +1100652 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700654EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
656/*
Nick Piggin77812a12011-01-07 17:49:48 +1100657 * Try to throw away a dentry - free the inode, dput the parent.
658 * Requires dentry->d_lock is held, and dentry->d_count == 0.
659 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700660 *
Nick Piggin77812a12011-01-07 17:49:48 +1100661 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 */
Nick Piggin77812a12011-01-07 17:49:48 +1100663static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200664 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
Nick Piggin77812a12011-01-07 17:49:48 +1100666 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700667
Nick Piggin77812a12011-01-07 17:49:48 +1100668 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700669 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100670 * If dentry_kill returns NULL, we have nothing more to do.
671 * if it returns the same dentry, trylocks failed. In either
672 * case, just loop again.
673 *
674 * Otherwise, we need to prune ancestors too. This is necessary
675 * to prevent quadratic behavior of shrink_dcache_parent(), but
676 * is also expected to be beneficial in reducing dentry cache
677 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700678 */
Nick Piggin77812a12011-01-07 17:49:48 +1100679 if (!parent)
680 return;
681 if (parent == dentry)
682 return;
683
684 /* Prune ancestors. */
685 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700686 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100687 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100688 if (dentry->d_count > 1) {
689 dentry->d_count--;
690 spin_unlock(&dentry->d_lock);
691 return;
692 }
Nick Piggin77812a12011-01-07 17:49:48 +1100693 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400697static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700699 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700700
Nick Pigginec336792011-01-07 17:49:47 +1100701 rcu_read_lock();
702 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100703 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
704 if (&dentry->d_lru == list)
705 break; /* empty */
706 spin_lock(&dentry->d_lock);
707 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
708 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100709 continue;
710 }
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 /*
713 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700714 * the LRU because of laziness during lookup. Do not free
715 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100717 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100718 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700719 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 continue;
721 }
Nick Pigginec336792011-01-07 17:49:47 +1100722
Nick Pigginec336792011-01-07 17:49:47 +1100723 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100724
725 try_prune_one_dentry(dentry);
726
Nick Pigginec336792011-01-07 17:49:47 +1100727 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Nick Pigginec336792011-01-07 17:49:47 +1100729 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400730}
731
732/**
733 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
734 * @sb: superblock to shrink dentry LRU.
735 * @count: number of entries to prune
736 * @flags: flags to control the dentry processing
737 *
738 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
739 */
740static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
741{
742 /* called from prune_dcache() and shrink_dcache_parent() */
743 struct dentry *dentry;
744 LIST_HEAD(referenced);
745 LIST_HEAD(tmp);
746 int cnt = *count;
747
Nick Piggin23044502011-01-07 17:49:31 +1100748relock:
749 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400750 while (!list_empty(&sb->s_dentry_lru)) {
751 dentry = list_entry(sb->s_dentry_lru.prev,
752 struct dentry, d_lru);
753 BUG_ON(dentry->d_sb != sb);
754
Nick Piggin23044502011-01-07 17:49:31 +1100755 if (!spin_trylock(&dentry->d_lock)) {
756 spin_unlock(&dcache_lru_lock);
757 cpu_relax();
758 goto relock;
759 }
760
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400761 /*
762 * If we are honouring the DCACHE_REFERENCED flag and the
763 * dentry has this flag set, don't free it. Clear the flag
764 * and put it back on the LRU.
765 */
Nick Piggin23044502011-01-07 17:49:31 +1100766 if (flags & DCACHE_REFERENCED &&
767 dentry->d_flags & DCACHE_REFERENCED) {
768 dentry->d_flags &= ~DCACHE_REFERENCED;
769 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400770 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100771 } else {
772 list_move_tail(&dentry->d_lru, &tmp);
773 spin_unlock(&dentry->d_lock);
774 if (!--cnt)
775 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400776 }
Nick Pigginec336792011-01-07 17:49:47 +1100777 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400778 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700779 if (!list_empty(&referenced))
780 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100781 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100782
783 shrink_dentry_list(&tmp);
784
785 *count = cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786}
787
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700788/**
789 * prune_dcache - shrink the dcache
790 * @count: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 *
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700792 * Shrink the dcache. This is done when we need more memory, or simply when we
793 * need to unmount something (at which point we need to unuse all dentries).
794 *
795 * This function may fail to free any resources if all the dentries are in use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700797static void prune_dcache(int count)
798{
Al Virodca33252010-07-25 02:31:46 +0400799 struct super_block *sb, *p = NULL;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700800 int w_count;
Nick Piggin86c87492011-01-07 17:49:18 +1100801 int unused = dentry_stat.nr_unused;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700802 int prune_ratio;
803 int pruned;
804
805 if (unused == 0 || count == 0)
806 return;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700807 if (count >= unused)
808 prune_ratio = 1;
809 else
810 prune_ratio = unused / count;
811 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400812 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viro551de6f2010-03-22 19:36:35 -0400813 if (list_empty(&sb->s_instances))
814 continue;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700815 if (sb->s_nr_dentry_unused == 0)
816 continue;
817 sb->s_count++;
818 /* Now, we reclaim unused dentrins with fairness.
819 * We reclaim them same percentage from each superblock.
820 * We calculate number of dentries to scan on this sb
821 * as follows, but the implementation is arranged to avoid
822 * overflows:
823 * number of dentries to scan on this sb =
824 * count * (number of dentries on this sb /
825 * number of dentries in the machine)
826 */
827 spin_unlock(&sb_lock);
828 if (prune_ratio != 1)
829 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
830 else
831 w_count = sb->s_nr_dentry_unused;
832 pruned = w_count;
833 /*
834 * We need to be sure this filesystem isn't being unmounted,
835 * otherwise we could race with generic_shutdown_super(), and
836 * end up holding a reference to an inode while the filesystem
837 * is unmounted. So we try to get s_umount, and make sure
838 * s_root isn't NULL.
839 */
840 if (down_read_trylock(&sb->s_umount)) {
841 if ((sb->s_root != NULL) &&
842 (!list_empty(&sb->s_dentry_lru))) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700843 __shrink_dcache_sb(sb, &w_count,
844 DCACHE_REFERENCED);
845 pruned -= w_count;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700846 }
847 up_read(&sb->s_umount);
848 }
849 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400850 if (p)
851 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700852 count -= pruned;
Al Virodca33252010-07-25 02:31:46 +0400853 p = sb;
Al Viro79893c12010-03-22 20:27:55 -0400854 /* more work left to do? */
855 if (count <= 0)
856 break;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700857 }
Al Virodca33252010-07-25 02:31:46 +0400858 if (p)
859 __put_super(p);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700860 spin_unlock(&sb_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700861}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863/**
864 * shrink_dcache_sb - shrink dcache for a superblock
865 * @sb: superblock
866 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400867 * Shrink the dcache for the specified super block. This is used to free
868 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400870void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400872 LIST_HEAD(tmp);
873
Nick Piggin23044502011-01-07 17:49:31 +1100874 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400875 while (!list_empty(&sb->s_dentry_lru)) {
876 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100877 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400878 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100879 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400880 }
Nick Piggin23044502011-01-07 17:49:31 +1100881 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700883EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885/*
David Howellsc636ebd2006-10-11 01:22:19 -0700886 * destroy a single subtree of dentries for unmount
887 * - see the comments on shrink_dcache_for_umount() for a description of the
888 * locking
889 */
890static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
891{
892 struct dentry *parent;
David Howellsf8713572006-10-28 10:38:46 -0700893 unsigned detached = 0;
David Howellsc636ebd2006-10-11 01:22:19 -0700894
895 BUG_ON(!IS_ROOT(dentry));
896
897 /* detach this root from the system */
Nick Piggin23044502011-01-07 17:49:31 +1100898 spin_lock(&dentry->d_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400899 dentry_lru_del(dentry);
David Howellsc636ebd2006-10-11 01:22:19 -0700900 __d_drop(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100901 spin_unlock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700902
903 for (;;) {
904 /* descend to the first leaf in the current subtree */
905 while (!list_empty(&dentry->d_subdirs)) {
906 struct dentry *loop;
907
908 /* this is a branch with children - detach all of them
909 * from the system in one go */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100910 spin_lock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700911 list_for_each_entry(loop, &dentry->d_subdirs,
912 d_u.d_child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100913 spin_lock_nested(&loop->d_lock,
914 DENTRY_D_LOCK_NESTED);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400915 dentry_lru_del(loop);
David Howellsc636ebd2006-10-11 01:22:19 -0700916 __d_drop(loop);
Nick Pigginda502952011-01-07 17:49:33 +1100917 spin_unlock(&loop->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700918 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100919 spin_unlock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -0700920
921 /* move to the first child */
922 dentry = list_entry(dentry->d_subdirs.next,
923 struct dentry, d_u.d_child);
924 }
925
926 /* consume the dentries from this leaf up through its parents
927 * until we find one with children or run out altogether */
928 do {
929 struct inode *inode;
930
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100931 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700932 printk(KERN_ERR
933 "BUG: Dentry %p{i=%lx,n=%s}"
934 " still in use (%d)"
935 " [unmount of %s %s]\n",
936 dentry,
937 dentry->d_inode ?
938 dentry->d_inode->i_ino : 0UL,
939 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100940 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700941 dentry->d_sb->s_type->name,
942 dentry->d_sb->s_id);
943 BUG();
944 }
945
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100946 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700947 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100948 list_del(&dentry->d_u.d_child);
949 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900950 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100951 spin_lock(&parent->d_lock);
952 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100953 list_del(&dentry->d_u.d_child);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100954 spin_unlock(&parent->d_lock);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900955 }
David Howellsc636ebd2006-10-11 01:22:19 -0700956
David Howellsf8713572006-10-28 10:38:46 -0700957 detached++;
David Howellsc636ebd2006-10-11 01:22:19 -0700958
959 inode = dentry->d_inode;
960 if (inode) {
961 dentry->d_inode = NULL;
962 list_del_init(&dentry->d_alias);
963 if (dentry->d_op && dentry->d_op->d_iput)
964 dentry->d_op->d_iput(dentry, inode);
965 else
966 iput(inode);
967 }
968
969 d_free(dentry);
970
971 /* finished when we fall off the top of the tree,
972 * otherwise we ascend to the parent and move to the
973 * next sibling if there is one */
974 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400975 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700976 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700977 } while (list_empty(&dentry->d_subdirs));
978
979 dentry = list_entry(dentry->d_subdirs.next,
980 struct dentry, d_u.d_child);
981 }
982}
983
984/*
985 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100986 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700987 * - the superblock is detached from all mountings and open files, so the
988 * dentry trees will not be rearranged by the VFS
989 * - s_umount is write-locked, so the memory pressure shrinker will ignore
990 * any dentries belonging to this superblock that it comes across
991 * - the filesystem itself is no longer permitted to rearrange the dentries
992 * in this superblock
993 */
994void shrink_dcache_for_umount(struct super_block *sb)
995{
996 struct dentry *dentry;
997
998 if (down_read_trylock(&sb->s_umount))
999 BUG();
1000
1001 dentry = sb->s_root;
1002 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001003 spin_lock(&dentry->d_lock);
1004 dentry->d_count--;
1005 spin_unlock(&dentry->d_lock);
David Howellsc636ebd2006-10-11 01:22:19 -07001006 shrink_dcache_for_umount_subtree(dentry);
1007
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001008 while (!hlist_bl_empty(&sb->s_anon)) {
1009 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001010 shrink_dcache_for_umount_subtree(dentry);
1011 }
1012}
1013
1014/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 * Search for at least 1 mount point in the dentry's subdirs.
1016 * We descend to the next level whenever the d_subdirs
1017 * list is non-empty and continue searching.
1018 */
1019
1020/**
1021 * have_submounts - check for mounts over a dentry
1022 * @parent: dentry to check.
1023 *
1024 * Return true if the parent or its subdirectories contain
1025 * a mount point
1026 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027int have_submounts(struct dentry *parent)
1028{
Nick Piggin949854d2011-01-07 17:49:37 +11001029 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001031 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001032 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001033
Nick Piggin949854d2011-01-07 17:49:37 +11001034 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001035again:
1036 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if (d_mountpoint(parent))
1039 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001040 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041repeat:
1042 next = this_parent->d_subdirs.next;
1043resume:
1044 while (next != &this_parent->d_subdirs) {
1045 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001046 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001048
1049 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001051 if (d_mountpoint(dentry)) {
1052 spin_unlock(&dentry->d_lock);
1053 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001057 spin_unlock(&this_parent->d_lock);
1058 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001060 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 goto repeat;
1062 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001063 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 }
1065 /*
1066 * All done at this level ... ascend and resume the search.
1067 */
1068 if (this_parent != parent) {
Nick Piggin949854d2011-01-07 17:49:37 +11001069 struct dentry *tmp;
1070 struct dentry *child;
1071
1072 tmp = this_parent->d_parent;
1073 rcu_read_lock();
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001074 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001075 child = this_parent;
1076 this_parent = tmp;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001077 spin_lock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001078 /* might go back up the wrong parent if we have had a rename
1079 * or deletion */
1080 if (this_parent != child->d_parent ||
Nick Piggin58db63d2011-01-07 17:49:39 +11001081 (!locked && read_seqretry(&rename_lock, seq))) {
Nick Piggin949854d2011-01-07 17:49:37 +11001082 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001083 rcu_read_unlock();
1084 goto rename_retry;
1085 }
1086 rcu_read_unlock();
1087 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 goto resume;
1089 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001090 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001091 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001092 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001093 if (locked)
1094 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 return 0; /* No mount points found in tree */
1096positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001097 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001098 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001099 if (locked)
1100 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001102
1103rename_retry:
1104 locked = 1;
1105 write_seqlock(&rename_lock);
1106 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001108EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
1110/*
1111 * Search the dentry child list for the specified parent,
1112 * and move any unused dentries to the end of the unused
1113 * list for prune_dcache(). We descend to the next level
1114 * whenever the d_subdirs list is non-empty and continue
1115 * searching.
1116 *
1117 * It returns zero iff there are no unused children,
1118 * otherwise it returns the number of children moved to
1119 * the end of the unused list. This may not be the total
1120 * number of unused children, because select_parent can
1121 * drop the lock and return early due to latency
1122 * constraints.
1123 */
1124static int select_parent(struct dentry * parent)
1125{
Nick Piggin949854d2011-01-07 17:49:37 +11001126 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001128 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001130 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Nick Piggin949854d2011-01-07 17:49:37 +11001132 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001133again:
1134 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001135 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136repeat:
1137 next = this_parent->d_subdirs.next;
1138resume:
1139 while (next != &this_parent->d_subdirs) {
1140 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001141 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 next = tmp->next;
1143
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001144 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001145
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 /*
1147 * move only zero ref count dentries to the end
1148 * of the unused list for prune_dcache
1149 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001150 if (!dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001151 dentry_lru_move_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 found++;
Christoph Hellwiga4633352010-10-10 05:36:26 -04001153 } else {
1154 dentry_lru_del(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
1156
1157 /*
1158 * We can return to the caller if we have found some (this
1159 * ensures forward progress). We'll be coming back to find
1160 * the rest.
1161 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001162 if (found && need_resched()) {
1163 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
1167 /*
1168 * Descend a level if the d_subdirs list is non-empty.
1169 */
1170 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001171 spin_unlock(&this_parent->d_lock);
1172 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001174 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 goto repeat;
1176 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001177
1178 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
1180 /*
1181 * All done at this level ... ascend and resume the search.
1182 */
1183 if (this_parent != parent) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001184 struct dentry *tmp;
Nick Piggin949854d2011-01-07 17:49:37 +11001185 struct dentry *child;
1186
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001187 tmp = this_parent->d_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11001188 rcu_read_lock();
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001189 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001190 child = this_parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001191 this_parent = tmp;
1192 spin_lock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001193 /* might go back up the wrong parent if we have had a rename
1194 * or deletion */
1195 if (this_parent != child->d_parent ||
Nick Piggin58db63d2011-01-07 17:49:39 +11001196 (!locked && read_seqretry(&rename_lock, seq))) {
Nick Piggin949854d2011-01-07 17:49:37 +11001197 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11001198 rcu_read_unlock();
1199 goto rename_retry;
1200 }
1201 rcu_read_unlock();
1202 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 goto resume;
1204 }
1205out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001206 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001207 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001208 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001209 if (locked)
1210 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001212
1213rename_retry:
1214 if (found)
1215 return found;
1216 locked = 1;
1217 write_seqlock(&rename_lock);
1218 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
1221/**
1222 * shrink_dcache_parent - prune dcache
1223 * @parent: parent of entries to prune
1224 *
1225 * Prune the dcache to remove unused children of the parent dentry.
1226 */
1227
1228void shrink_dcache_parent(struct dentry * parent)
1229{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001230 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 int found;
1232
1233 while ((found = select_parent(parent)) != 0)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001234 __shrink_dcache_sb(sb, &found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001236EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238/*
1239 * Scan `nr' dentries and return the number which remain.
1240 *
1241 * We need to avoid reentering the filesystem if the caller is performing a
1242 * GFP_NOFS allocation attempt. One example deadlock is:
1243 *
1244 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1245 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1246 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1247 *
1248 * In this case we return -1 to tell the caller that we baled.
1249 */
Dave Chinner7f8275d2010-07-19 14:56:17 +10001250static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
1252 if (nr) {
1253 if (!(gfp_mask & __GFP_FS))
1254 return -1;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001255 prune_dcache(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 }
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001257
Nick Piggin86c87492011-01-07 17:49:18 +11001258 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259}
1260
Rusty Russell8e1f9362007-07-17 04:03:17 -07001261static struct shrinker dcache_shrinker = {
1262 .shrink = shrink_dcache_memory,
1263 .seeks = DEFAULT_SEEKS,
1264};
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266/**
1267 * d_alloc - allocate a dcache entry
1268 * @parent: parent of entry to allocate
1269 * @name: qstr of the name
1270 *
1271 * Allocates a dentry. It returns %NULL if there is insufficient memory
1272 * available. On a success the dentry is returned. The name passed in is
1273 * copied and the copy passed in may be reused after this call.
1274 */
1275
1276struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1277{
1278 struct dentry *dentry;
1279 char *dname;
1280
Mel Gormane12ba742007-10-16 01:25:52 -07001281 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (!dentry)
1283 return NULL;
1284
1285 if (name->len > DNAME_INLINE_LEN-1) {
1286 dname = kmalloc(name->len + 1, GFP_KERNEL);
1287 if (!dname) {
1288 kmem_cache_free(dentry_cache, dentry);
1289 return NULL;
1290 }
1291 } else {
1292 dname = dentry->d_iname;
1293 }
1294 dentry->d_name.name = dname;
1295
1296 dentry->d_name.len = name->len;
1297 dentry->d_name.hash = name->hash;
1298 memcpy(dname, name->name, name->len);
1299 dname[name->len] = 0;
1300
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001301 dentry->d_count = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 dentry->d_flags = DCACHE_UNHASHED;
1303 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001304 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 dentry->d_inode = NULL;
1306 dentry->d_parent = NULL;
1307 dentry->d_sb = NULL;
1308 dentry->d_op = NULL;
1309 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001310 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 INIT_LIST_HEAD(&dentry->d_lru);
1312 INIT_LIST_HEAD(&dentry->d_subdirs);
1313 INIT_LIST_HEAD(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001314 INIT_LIST_HEAD(&dentry->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
1316 if (parent) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001317 spin_lock(&parent->d_lock);
Nick Piggin89ad4852011-01-07 17:49:41 +11001318 /*
1319 * don't need child lock because it is not subject
1320 * to concurrency here
1321 */
Nick Piggindc0474b2011-01-07 17:49:43 +11001322 __dget_dlock(parent);
1323 dentry->d_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 dentry->d_sb = parent->d_sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001325 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Eric Dumazet5160ee62006-01-08 01:03:32 -08001326 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001327 spin_unlock(&parent->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Nick Piggin3e880fb2011-01-07 17:49:19 +11001330 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return dentry;
1333}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001334EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Nick Piggin4b936882011-01-07 17:50:07 +11001336struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1337{
1338 struct dentry *dentry = d_alloc(NULL, name);
1339 if (dentry) {
1340 dentry->d_sb = sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001341 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Nick Piggin4b936882011-01-07 17:50:07 +11001342 dentry->d_parent = dentry;
1343 dentry->d_flags |= DCACHE_DISCONNECTED;
1344 }
1345 return dentry;
1346}
1347EXPORT_SYMBOL(d_alloc_pseudo);
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1350{
1351 struct qstr q;
1352
1353 q.name = name;
1354 q.len = strlen(name);
1355 q.hash = full_name_hash(q.name, q.len);
1356 return d_alloc(parent, &q);
1357}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001358EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Nick Pigginfb045ad2011-01-07 17:49:55 +11001360void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1361{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001362 WARN_ON_ONCE(dentry->d_op);
1363 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001364 DCACHE_OP_COMPARE |
1365 DCACHE_OP_REVALIDATE |
1366 DCACHE_OP_DELETE ));
1367 dentry->d_op = op;
1368 if (!op)
1369 return;
1370 if (op->d_hash)
1371 dentry->d_flags |= DCACHE_OP_HASH;
1372 if (op->d_compare)
1373 dentry->d_flags |= DCACHE_OP_COMPARE;
1374 if (op->d_revalidate)
1375 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1376 if (op->d_delete)
1377 dentry->d_flags |= DCACHE_OP_DELETE;
1378
1379}
1380EXPORT_SYMBOL(d_set_d_op);
1381
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001382static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1383{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001384 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001385 if (inode) {
1386 if (unlikely(IS_AUTOMOUNT(inode)))
1387 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001388 list_add(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001389 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001390 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001391 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001392 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001393 fsnotify_d_instantiate(dentry, inode);
1394}
1395
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396/**
1397 * d_instantiate - fill in inode information for a dentry
1398 * @entry: dentry to complete
1399 * @inode: inode to attach to this dentry
1400 *
1401 * Fill in inode information in the entry.
1402 *
1403 * This turns negative dentries into productive full members
1404 * of society.
1405 *
1406 * NOTE! This assumes that the inode count has been incremented
1407 * (or otherwise set) by the caller to indicate that it is now
1408 * in use by the dcache.
1409 */
1410
1411void d_instantiate(struct dentry *entry, struct inode * inode)
1412{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001413 BUG_ON(!list_empty(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001414 if (inode)
1415 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001416 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001417 if (inode)
1418 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 security_d_instantiate(entry, inode);
1420}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001421EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
1423/**
1424 * d_instantiate_unique - instantiate a non-aliased dentry
1425 * @entry: dentry to instantiate
1426 * @inode: inode to attach to this dentry
1427 *
1428 * Fill in inode information in the entry. On success, it returns NULL.
1429 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001430 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 *
1432 * Note that in order to avoid conflicts with rename() etc, the caller
1433 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001434 *
1435 * This also assumes that the inode count has been incremented
1436 * (or otherwise set) by the caller to indicate that it is now
1437 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 */
David Howells770bfad2006-08-22 20:06:07 -04001439static struct dentry *__d_instantiate_unique(struct dentry *entry,
1440 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441{
1442 struct dentry *alias;
1443 int len = entry->d_name.len;
1444 const char *name = entry->d_name.name;
1445 unsigned int hash = entry->d_name.hash;
1446
David Howells770bfad2006-08-22 20:06:07 -04001447 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001448 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001449 return NULL;
1450 }
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1453 struct qstr *qstr = &alias->d_name;
1454
Nick Piggin9abca362011-01-07 17:49:36 +11001455 /*
1456 * Don't need alias->d_lock here, because aliases with
1457 * d_parent == entry->d_parent are not subject to name or
1458 * parent changes, because the parent inode i_mutex is held.
1459 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 if (qstr->hash != hash)
1461 continue;
1462 if (alias->d_parent != entry->d_parent)
1463 continue;
Nick Piggin9d55c362011-01-07 17:50:09 +11001464 if (dentry_cmp(qstr->name, qstr->len, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001466 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 return alias;
1468 }
David Howells770bfad2006-08-22 20:06:07 -04001469
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001470 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return NULL;
1472}
David Howells770bfad2006-08-22 20:06:07 -04001473
1474struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1475{
1476 struct dentry *result;
1477
1478 BUG_ON(!list_empty(&entry->d_alias));
1479
Nick Piggin873feea2011-01-07 17:50:06 +11001480 if (inode)
1481 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001482 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001483 if (inode)
1484 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001485
1486 if (!result) {
1487 security_d_instantiate(entry, inode);
1488 return NULL;
1489 }
1490
1491 BUG_ON(!d_unhashed(result));
1492 iput(inode);
1493 return result;
1494}
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496EXPORT_SYMBOL(d_instantiate_unique);
1497
1498/**
1499 * d_alloc_root - allocate root dentry
1500 * @root_inode: inode to allocate the root for
1501 *
1502 * Allocate a root ("/") dentry for the inode given. The inode is
1503 * instantiated and returned. %NULL is returned if there is insufficient
1504 * memory or the inode passed is %NULL.
1505 */
1506
1507struct dentry * d_alloc_root(struct inode * root_inode)
1508{
1509 struct dentry *res = NULL;
1510
1511 if (root_inode) {
1512 static const struct qstr name = { .name = "/", .len = 1 };
1513
1514 res = d_alloc(NULL, &name);
1515 if (res) {
1516 res->d_sb = root_inode->i_sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001517 d_set_d_op(res, res->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 res->d_parent = res;
1519 d_instantiate(res, root_inode);
1520 }
1521 }
1522 return res;
1523}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001524EXPORT_SYMBOL(d_alloc_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001527 * d_obtain_alias - find or allocate a dentry for a given inode
1528 * @inode: inode to allocate the dentry for
1529 *
1530 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1531 * similar open by handle operations. The returned dentry may be anonymous,
1532 * or may have a full name (if the inode was already in the cache).
1533 *
1534 * When called on a directory inode, we must ensure that the inode only ever
1535 * has one dentry. If a dentry is found, that is returned instead of
1536 * allocating a new one.
1537 *
1538 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001539 * to the dentry. In case of an error the reference on the inode is released.
1540 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1541 * be passed in and will be the error will be propagate to the return value,
1542 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001543 */
1544struct dentry *d_obtain_alias(struct inode *inode)
1545{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001546 static const struct qstr anonstring = { .name = "" };
1547 struct dentry *tmp;
1548 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001549
1550 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001551 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001552 if (IS_ERR(inode))
1553 return ERR_CAST(inode);
1554
Christoph Hellwig9308a612008-08-11 15:49:12 +02001555 res = d_find_alias(inode);
1556 if (res)
1557 goto out_iput;
1558
1559 tmp = d_alloc(NULL, &anonstring);
1560 if (!tmp) {
1561 res = ERR_PTR(-ENOMEM);
1562 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001563 }
Christoph Hellwig9308a612008-08-11 15:49:12 +02001564 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1565
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001566
Nick Piggin873feea2011-01-07 17:50:06 +11001567 spin_lock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001568 res = __d_find_alias(inode, 0);
1569 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001570 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001571 dput(tmp);
1572 goto out_iput;
1573 }
1574
1575 /* attach a disconnected dentry */
1576 spin_lock(&tmp->d_lock);
1577 tmp->d_sb = inode->i_sb;
Al Viroc8aebb02010-12-18 10:22:30 -05001578 d_set_d_op(tmp, tmp->d_sb->s_d_op);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001579 tmp->d_inode = inode;
1580 tmp->d_flags |= DCACHE_DISCONNECTED;
Christoph Hellwig9308a612008-08-11 15:49:12 +02001581 list_add(&tmp->d_alias, &inode->i_dentry);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001582 bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
1583 tmp->d_flags &= ~DCACHE_UNHASHED;
1584 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1585 __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001586 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001587 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001588
Christoph Hellwig9308a612008-08-11 15:49:12 +02001589 return tmp;
1590
1591 out_iput:
1592 iput(inode);
1593 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001594}
Benny Halevyadc48722009-02-27 14:02:59 -08001595EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597/**
1598 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1599 * @inode: the inode which may have a disconnected dentry
1600 * @dentry: a negative dentry which we want to point to the inode.
1601 *
1602 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1603 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1604 * and return it, else simply d_add the inode to the dentry and return NULL.
1605 *
1606 * This is needed in the lookup routine of any filesystem that is exportable
1607 * (via knfsd) so that we can build dcache paths to directories effectively.
1608 *
1609 * If a dentry was found and moved, then it is returned. Otherwise NULL
1610 * is returned. This matches the expected return value of ->lookup.
1611 *
1612 */
1613struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1614{
1615 struct dentry *new = NULL;
1616
NeilBrown21c0d8f2006-10-04 02:16:16 -07001617 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001618 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 new = __d_find_alias(inode, 1);
1620 if (new) {
1621 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001622 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 d_move(new, dentry);
1625 iput(inode);
1626 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001627 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001628 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001629 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 security_d_instantiate(dentry, inode);
1631 d_rehash(dentry);
1632 }
1633 } else
1634 d_add(dentry, inode);
1635 return new;
1636}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001637EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
Barry Naujok94035402008-05-21 16:50:46 +10001639/**
1640 * d_add_ci - lookup or allocate new dentry with case-exact name
1641 * @inode: the inode case-insensitive lookup has found
1642 * @dentry: the negative dentry that was passed to the parent's lookup func
1643 * @name: the case-exact name to be associated with the returned dentry
1644 *
1645 * This is to avoid filling the dcache with case-insensitive names to the
1646 * same inode, only the actual correct case is stored in the dcache for
1647 * case-insensitive filesystems.
1648 *
1649 * For a case-insensitive lookup match and if the the case-exact dentry
1650 * already exists in in the dcache, use it and return it.
1651 *
1652 * If no entry exists with the exact case name, allocate new dentry with
1653 * the exact case, and return the spliced entry.
1654 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001655struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001656 struct qstr *name)
1657{
1658 int error;
1659 struct dentry *found;
1660 struct dentry *new;
1661
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001662 /*
1663 * First check if a dentry matching the name already exists,
1664 * if not go ahead and create it now.
1665 */
Barry Naujok94035402008-05-21 16:50:46 +10001666 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001667 if (!found) {
1668 new = d_alloc(dentry->d_parent, name);
1669 if (!new) {
1670 error = -ENOMEM;
1671 goto err_out;
1672 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001673
Barry Naujok94035402008-05-21 16:50:46 +10001674 found = d_splice_alias(inode, new);
1675 if (found) {
1676 dput(new);
1677 return found;
1678 }
1679 return new;
1680 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001681
1682 /*
1683 * If a matching dentry exists, and it's not negative use it.
1684 *
1685 * Decrement the reference count to balance the iget() done
1686 * earlier on.
1687 */
Barry Naujok94035402008-05-21 16:50:46 +10001688 if (found->d_inode) {
1689 if (unlikely(found->d_inode != inode)) {
1690 /* This can't happen because bad inodes are unhashed. */
1691 BUG_ON(!is_bad_inode(inode));
1692 BUG_ON(!is_bad_inode(found->d_inode));
1693 }
Barry Naujok94035402008-05-21 16:50:46 +10001694 iput(inode);
1695 return found;
1696 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001697
Barry Naujok94035402008-05-21 16:50:46 +10001698 /*
1699 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001700 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001701 */
Nick Piggin873feea2011-01-07 17:50:06 +11001702 spin_lock(&inode->i_lock);
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001703 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001704 __d_instantiate(found, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001705 spin_unlock(&inode->i_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001706 security_d_instantiate(found, inode);
1707 return found;
1708 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001709
Barry Naujok94035402008-05-21 16:50:46 +10001710 /*
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001711 * In case a directory already has a (disconnected) entry grab a
1712 * reference to it, move it in place and use it.
Barry Naujok94035402008-05-21 16:50:46 +10001713 */
1714 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
Nick Piggindc0474b2011-01-07 17:49:43 +11001715 __dget(new);
Nick Piggin873feea2011-01-07 17:50:06 +11001716 spin_unlock(&inode->i_lock);
Barry Naujok94035402008-05-21 16:50:46 +10001717 security_d_instantiate(found, inode);
Barry Naujok94035402008-05-21 16:50:46 +10001718 d_move(new, found);
Barry Naujok94035402008-05-21 16:50:46 +10001719 iput(inode);
Barry Naujok94035402008-05-21 16:50:46 +10001720 dput(found);
Barry Naujok94035402008-05-21 16:50:46 +10001721 return new;
1722
1723err_out:
1724 iput(inode);
1725 return ERR_PTR(error);
1726}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001727EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001730 * __d_lookup_rcu - search for a dentry (racy, store-free)
1731 * @parent: parent dentry
1732 * @name: qstr of name we wish to find
1733 * @seq: returns d_seq value at the point where the dentry was found
1734 * @inode: returns dentry->d_inode when the inode was found valid.
1735 * Returns: dentry, or NULL
1736 *
1737 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1738 * resolution (store-free path walking) design described in
1739 * Documentation/filesystems/path-lookup.txt.
1740 *
1741 * This is not to be used outside core vfs.
1742 *
1743 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1744 * held, and rcu_read_lock held. The returned dentry must not be stored into
1745 * without taking d_lock and checking d_seq sequence count against @seq
1746 * returned here.
1747 *
1748 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1749 * function.
1750 *
1751 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1752 * the returned dentry, so long as its parent's seqlock is checked after the
1753 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1754 * is formed, giving integrity down the path walk.
1755 */
1756struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1757 unsigned *seq, struct inode **inode)
1758{
1759 unsigned int len = name->len;
1760 unsigned int hash = name->hash;
1761 const unsigned char *str = name->name;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001762 struct dcache_hash_bucket *b = d_hash(parent, hash);
1763 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001764 struct dentry *dentry;
1765
1766 /*
1767 * Note: There is significant duplication with __d_lookup_rcu which is
1768 * required to prevent single threaded performance regressions
1769 * especially on architectures where smp_rmb (in seqcounts) are costly.
1770 * Keep the two functions in sync.
1771 */
1772
1773 /*
1774 * The hash list is protected using RCU.
1775 *
1776 * Carefully use d_seq when comparing a candidate dentry, to avoid
1777 * races with d_move().
1778 *
1779 * It is possible that concurrent renames can mess up our list
1780 * walk here and result in missing our dentry, resulting in the
1781 * false-negative result. d_lookup() protects against concurrent
1782 * renames using rename_lock seqlock.
1783 *
1784 * See Documentation/vfs/dcache-locking.txt for more details.
1785 */
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001786 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001787 struct inode *i;
1788 const char *tname;
1789 int tlen;
1790
1791 if (dentry->d_name.hash != hash)
1792 continue;
1793
1794seqretry:
1795 *seq = read_seqcount_begin(&dentry->d_seq);
1796 if (dentry->d_parent != parent)
1797 continue;
1798 if (d_unhashed(dentry))
1799 continue;
1800 tlen = dentry->d_name.len;
1801 tname = dentry->d_name.name;
1802 i = dentry->d_inode;
Nick Piggine1bb5782011-01-07 17:50:08 +11001803 prefetch(tname);
1804 if (i)
1805 prefetch(i);
Nick Piggin31e6b012011-01-07 17:49:52 +11001806 /*
1807 * This seqcount check is required to ensure name and
1808 * len are loaded atomically, so as not to walk off the
1809 * edge of memory when walking. If we could load this
1810 * atomically some other way, we could drop this check.
1811 */
1812 if (read_seqcount_retry(&dentry->d_seq, *seq))
1813 goto seqretry;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001814 if (parent->d_flags & DCACHE_OP_COMPARE) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001815 if (parent->d_op->d_compare(parent, *inode,
1816 dentry, i,
1817 tlen, tname, name))
1818 continue;
1819 } else {
Nick Piggin9d55c362011-01-07 17:50:09 +11001820 if (dentry_cmp(tname, tlen, str, len))
Nick Piggin31e6b012011-01-07 17:49:52 +11001821 continue;
1822 }
1823 /*
1824 * No extra seqcount check is required after the name
1825 * compare. The caller must perform a seqcount check in
1826 * order to do anything useful with the returned dentry
1827 * anyway.
1828 */
1829 *inode = i;
1830 return dentry;
1831 }
1832 return NULL;
1833}
1834
1835/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 * d_lookup - search for a dentry
1837 * @parent: parent dentry
1838 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001839 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001841 * d_lookup searches the children of the parent dentry for the name in
1842 * question. If the dentry is found its reference count is incremented and the
1843 * dentry is returned. The caller must use dput to free the entry when it has
1844 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001846struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847{
Nick Piggin31e6b012011-01-07 17:49:52 +11001848 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001849 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
1851 do {
1852 seq = read_seqbegin(&rename_lock);
1853 dentry = __d_lookup(parent, name);
1854 if (dentry)
1855 break;
1856 } while (read_seqretry(&rename_lock, seq));
1857 return dentry;
1858}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001859EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860
Nick Piggin31e6b012011-01-07 17:49:52 +11001861/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001862 * __d_lookup - search for a dentry (racy)
1863 * @parent: parent dentry
1864 * @name: qstr of name we wish to find
1865 * Returns: dentry, or NULL
1866 *
1867 * __d_lookup is like d_lookup, however it may (rarely) return a
1868 * false-negative result due to unrelated rename activity.
1869 *
1870 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1871 * however it must be used carefully, eg. with a following d_lookup in
1872 * the case of failure.
1873 *
1874 * __d_lookup callers must be commented.
1875 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001876struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877{
1878 unsigned int len = name->len;
1879 unsigned int hash = name->hash;
1880 const unsigned char *str = name->name;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001881 struct dcache_hash_bucket *b = d_hash(parent, hash);
1882 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001883 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001884 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Nick Pigginb04f7842010-08-18 04:37:34 +10001886 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001887 * Note: There is significant duplication with __d_lookup_rcu which is
1888 * required to prevent single threaded performance regressions
1889 * especially on architectures where smp_rmb (in seqcounts) are costly.
1890 * Keep the two functions in sync.
1891 */
1892
1893 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001894 * The hash list is protected using RCU.
1895 *
1896 * Take d_lock when comparing a candidate dentry, to avoid races
1897 * with d_move().
1898 *
1899 * It is possible that concurrent renames can mess up our list
1900 * walk here and result in missing our dentry, resulting in the
1901 * false-negative result. d_lookup() protects against concurrent
1902 * renames using rename_lock seqlock.
1903 *
1904 * See Documentation/vfs/dcache-locking.txt for more details.
1905 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 rcu_read_lock();
1907
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001908 hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001909 const char *tname;
1910 int tlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 if (dentry->d_name.hash != hash)
1913 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
1915 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 if (dentry->d_parent != parent)
1917 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001918 if (d_unhashed(dentry))
1919 goto next;
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 /*
1922 * It is safe to compare names since d_move() cannot
1923 * change the qstr (protected by d_lock).
1924 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001925 tlen = dentry->d_name.len;
1926 tname = dentry->d_name.name;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001927 if (parent->d_flags & DCACHE_OP_COMPARE) {
Nick Piggin621e1552011-01-07 17:49:27 +11001928 if (parent->d_op->d_compare(parent, parent->d_inode,
1929 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001930 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 goto next;
1932 } else {
Nick Piggin9d55c362011-01-07 17:50:09 +11001933 if (dentry_cmp(tname, tlen, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 goto next;
1935 }
1936
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001937 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001938 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 spin_unlock(&dentry->d_lock);
1940 break;
1941next:
1942 spin_unlock(&dentry->d_lock);
1943 }
1944 rcu_read_unlock();
1945
1946 return found;
1947}
1948
1949/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001950 * d_hash_and_lookup - hash the qstr then search for a dentry
1951 * @dir: Directory to search in
1952 * @name: qstr of name we wish to find
1953 *
1954 * On hash failure or on lookup failure NULL is returned.
1955 */
1956struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1957{
1958 struct dentry *dentry = NULL;
1959
1960 /*
1961 * Check for a fs-specific hash function. Note that we must
1962 * calculate the standard hash first, as the d_op->d_hash()
1963 * routine may choose to leave the hash value unchanged.
1964 */
1965 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11001966 if (dir->d_flags & DCACHE_OP_HASH) {
Nick Pigginb1e6a012011-01-07 17:49:28 +11001967 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001968 goto out;
1969 }
1970 dentry = d_lookup(dir, name);
1971out:
1972 return dentry;
1973}
1974
1975/**
Nick Piggin786a5e12011-01-07 17:49:16 +11001976 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08001978 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 *
1980 * An insecure source has sent us a dentry, here we verify it and dget() it.
1981 * This is used by ncpfs in its readdir implementation.
1982 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11001983 *
1984 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 */
Nick Piggind3a23e12011-01-05 20:01:21 +11001986int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987{
Nick Piggin786a5e12011-01-07 17:49:16 +11001988 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11001989
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001990 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11001991 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1992 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001993 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11001994 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001995 spin_unlock(&dentry->d_lock);
1996 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 return 1;
1998 }
1999 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002000 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002001
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 return 0;
2003}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002004EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
2006/*
2007 * When a file is deleted, we have two options:
2008 * - turn this dentry into a negative dentry
2009 * - unhash this dentry and free it.
2010 *
2011 * Usually, we want to just turn this into
2012 * a negative dentry, but if anybody else is
2013 * currently using the dentry or the inode
2014 * we can't do that and we fall back on removing
2015 * it from the hash queues and waiting for
2016 * it to be deleted later when it has no users
2017 */
2018
2019/**
2020 * d_delete - delete a dentry
2021 * @dentry: The dentry to delete
2022 *
2023 * Turn the dentry into a negative dentry if possible, otherwise
2024 * remove it from the hash queues so it can be deleted later
2025 */
2026
2027void d_delete(struct dentry * dentry)
2028{
Nick Piggin873feea2011-01-07 17:50:06 +11002029 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002030 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 /*
2032 * Are we the only user?
2033 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002034again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002036 inode = dentry->d_inode;
2037 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002038 if (dentry->d_count == 1) {
Nick Piggin873feea2011-01-07 17:50:06 +11002039 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002040 spin_unlock(&dentry->d_lock);
2041 cpu_relax();
2042 goto again;
2043 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002044 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002045 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002046 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return;
2048 }
2049
2050 if (!d_unhashed(dentry))
2051 __d_drop(dentry);
2052
2053 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002054
2055 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002057EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002059static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002061 BUG_ON(!d_unhashed(entry));
2062 spin_lock_bucket(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 entry->d_flags &= ~DCACHE_UNHASHED;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002064 hlist_bl_add_head_rcu(&entry->d_hash, &b->head);
2065 spin_unlock_bucket(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066}
2067
David Howells770bfad2006-08-22 20:06:07 -04002068static void _d_rehash(struct dentry * entry)
2069{
2070 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2071}
2072
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073/**
2074 * d_rehash - add an entry back to the hash
2075 * @entry: dentry to add to the hash
2076 *
2077 * Adds a dentry to the hash according to its name.
2078 */
2079
2080void d_rehash(struct dentry * entry)
2081{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002083 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002086EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002088/**
2089 * dentry_update_name_case - update case insensitive dentry with a new name
2090 * @dentry: dentry to be updated
2091 * @name: new name
2092 *
2093 * Update a case insensitive dentry with new case of name.
2094 *
2095 * dentry must have been returned by d_lookup with name @name. Old and new
2096 * name lengths must match (ie. no d_compare which allows mismatched name
2097 * lengths).
2098 *
2099 * Parent inode i_mutex must be held over d_lookup and into this call (to
2100 * keep renames and concurrent inserts, and readdir(2) away).
2101 */
2102void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2103{
2104 BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2105 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2106
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002107 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002108 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002109 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002110 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002111 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002112}
2113EXPORT_SYMBOL(dentry_update_name_case);
2114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115static void switch_names(struct dentry *dentry, struct dentry *target)
2116{
2117 if (dname_external(target)) {
2118 if (dname_external(dentry)) {
2119 /*
2120 * Both external: swap the pointers
2121 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002122 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 } else {
2124 /*
2125 * dentry:internal, target:external. Steal target's
2126 * storage and make target internal.
2127 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002128 memcpy(target->d_iname, dentry->d_name.name,
2129 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 dentry->d_name.name = target->d_name.name;
2131 target->d_name.name = target->d_iname;
2132 }
2133 } else {
2134 if (dname_external(dentry)) {
2135 /*
2136 * dentry:external, target:internal. Give dentry's
2137 * storage to target and make dentry internal
2138 */
2139 memcpy(dentry->d_iname, target->d_name.name,
2140 target->d_name.len + 1);
2141 target->d_name.name = dentry->d_name.name;
2142 dentry->d_name.name = dentry->d_iname;
2143 } else {
2144 /*
2145 * Both are internal. Just copy target to dentry
2146 */
2147 memcpy(dentry->d_iname, target->d_name.name,
2148 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002149 dentry->d_name.len = target->d_name.len;
2150 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 }
2152 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002153 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154}
2155
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002156static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2157{
2158 /*
2159 * XXXX: do we really need to take target->d_lock?
2160 */
2161 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2162 spin_lock(&target->d_parent->d_lock);
2163 else {
2164 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2165 spin_lock(&dentry->d_parent->d_lock);
2166 spin_lock_nested(&target->d_parent->d_lock,
2167 DENTRY_D_LOCK_NESTED);
2168 } else {
2169 spin_lock(&target->d_parent->d_lock);
2170 spin_lock_nested(&dentry->d_parent->d_lock,
2171 DENTRY_D_LOCK_NESTED);
2172 }
2173 }
2174 if (target < dentry) {
2175 spin_lock_nested(&target->d_lock, 2);
2176 spin_lock_nested(&dentry->d_lock, 3);
2177 } else {
2178 spin_lock_nested(&dentry->d_lock, 2);
2179 spin_lock_nested(&target->d_lock, 3);
2180 }
2181}
2182
2183static void dentry_unlock_parents_for_move(struct dentry *dentry,
2184 struct dentry *target)
2185{
2186 if (target->d_parent != dentry->d_parent)
2187 spin_unlock(&dentry->d_parent->d_lock);
2188 if (target->d_parent != target)
2189 spin_unlock(&target->d_parent->d_lock);
2190}
2191
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002193 * When switching names, the actual string doesn't strictly have to
2194 * be preserved in the target - because we're dropping the target
2195 * anyway. As such, we can just do a simple memcpy() to copy over
2196 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002198 * Note that we have to be a lot more careful about getting the hash
2199 * switched - we have to switch the hash value properly even if it
2200 * then no longer matches the actual (corrupted) string of the target.
2201 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002203/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002204 * d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 * @dentry: entry to move
2206 * @target: new dentry
2207 *
2208 * Update the dcache to reflect the move of a file name. Negative
2209 * dcache entries should not be moved in this way.
2210 */
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002211void d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 if (!dentry->d_inode)
2214 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2215
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002216 BUG_ON(d_ancestor(dentry, target));
2217 BUG_ON(d_ancestor(target, dentry));
2218
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 write_seqlock(&rename_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002220
2221 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
Nick Piggin31e6b012011-01-07 17:49:52 +11002223 write_seqcount_begin(&dentry->d_seq);
2224 write_seqcount_begin(&target->d_seq);
2225
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002226 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2227
2228 /*
2229 * Move the dentry to the target hash queue. Don't bother checking
2230 * for the same hash queue because of how unlikely it is.
2231 */
2232 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002233 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234
2235 /* Unhash the target: dput() will then get rid of it */
2236 __d_drop(target);
2237
Eric Dumazet5160ee62006-01-08 01:03:32 -08002238 list_del(&dentry->d_u.d_child);
2239 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240
2241 /* Switch the names.. */
2242 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002243 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 /* ... and switch the parents */
2246 if (IS_ROOT(dentry)) {
2247 dentry->d_parent = target->d_parent;
2248 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002249 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002251 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
2253 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002254 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 }
2256
Eric Dumazet5160ee62006-01-08 01:03:32 -08002257 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002258
Nick Piggin31e6b012011-01-07 17:49:52 +11002259 write_seqcount_end(&target->d_seq);
2260 write_seqcount_end(&dentry->d_seq);
2261
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002262 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002264 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 spin_unlock(&dentry->d_lock);
2266 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002267}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002268EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002270/**
2271 * d_ancestor - search for an ancestor
2272 * @p1: ancestor dentry
2273 * @p2: child dentry
2274 *
2275 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2276 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002277 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002278struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002279{
2280 struct dentry *p;
2281
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002282 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002283 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002284 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002285 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002286 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002287}
2288
2289/*
2290 * This helper attempts to cope with remotely renamed directories
2291 *
2292 * It assumes that the caller is already holding
Nick Piggin873feea2011-01-07 17:50:06 +11002293 * dentry->d_parent->d_inode->i_mutex and the inode->i_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002294 *
2295 * Note: If ever the locking in lock_rename() changes, then please
2296 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002297 */
Nick Piggin873feea2011-01-07 17:50:06 +11002298static struct dentry *__d_unalias(struct inode *inode,
2299 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002300{
2301 struct mutex *m1 = NULL, *m2 = NULL;
2302 struct dentry *ret;
2303
2304 /* If alias and dentry share a parent, then no extra locks required */
2305 if (alias->d_parent == dentry->d_parent)
2306 goto out_unalias;
2307
2308 /* Check for loops */
2309 ret = ERR_PTR(-ELOOP);
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002310 if (d_ancestor(alias, dentry))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002311 goto out_err;
2312
2313 /* See lock_rename() */
2314 ret = ERR_PTR(-EBUSY);
2315 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2316 goto out_err;
2317 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2318 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2319 goto out_err;
2320 m2 = &alias->d_parent->d_inode->i_mutex;
2321out_unalias:
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002322 d_move(alias, dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002323 ret = alias;
2324out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002325 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002326 if (m2)
2327 mutex_unlock(m2);
2328 if (m1)
2329 mutex_unlock(m1);
2330 return ret;
2331}
2332
2333/*
David Howells770bfad2006-08-22 20:06:07 -04002334 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2335 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002336 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002337 */
2338static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2339{
2340 struct dentry *dparent, *aparent;
2341
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002342 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002343
Nick Piggin31e6b012011-01-07 17:49:52 +11002344 write_seqcount_begin(&dentry->d_seq);
2345 write_seqcount_begin(&anon->d_seq);
2346
David Howells770bfad2006-08-22 20:06:07 -04002347 dparent = dentry->d_parent;
2348 aparent = anon->d_parent;
2349
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002350 switch_names(dentry, anon);
2351 swap(dentry->d_name.hash, anon->d_name.hash);
2352
David Howells770bfad2006-08-22 20:06:07 -04002353 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2354 list_del(&dentry->d_u.d_child);
2355 if (!IS_ROOT(dentry))
2356 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2357 else
2358 INIT_LIST_HEAD(&dentry->d_u.d_child);
2359
2360 anon->d_parent = (dparent == dentry) ? anon : dparent;
2361 list_del(&anon->d_u.d_child);
2362 if (!IS_ROOT(anon))
2363 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2364 else
2365 INIT_LIST_HEAD(&anon->d_u.d_child);
2366
Nick Piggin31e6b012011-01-07 17:49:52 +11002367 write_seqcount_end(&dentry->d_seq);
2368 write_seqcount_end(&anon->d_seq);
2369
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002370 dentry_unlock_parents_for_move(anon, dentry);
2371 spin_unlock(&dentry->d_lock);
2372
2373 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002374 anon->d_flags &= ~DCACHE_DISCONNECTED;
2375}
2376
2377/**
2378 * d_materialise_unique - introduce an inode into the tree
2379 * @dentry: candidate dentry
2380 * @inode: inode to bind to the dentry, to which aliases may be attached
2381 *
2382 * Introduces an dentry into the tree, substituting an extant disconnected
2383 * root directory alias in its place if there is one
2384 */
2385struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2386{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002387 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002388
2389 BUG_ON(!d_unhashed(dentry));
2390
David Howells770bfad2006-08-22 20:06:07 -04002391 if (!inode) {
2392 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002393 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002394 d_rehash(actual);
2395 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002396 }
2397
Nick Piggin873feea2011-01-07 17:50:06 +11002398 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002399
Trond Myklebust9eaef272006-10-21 10:24:20 -07002400 if (S_ISDIR(inode->i_mode)) {
2401 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002402
Trond Myklebust9eaef272006-10-21 10:24:20 -07002403 /* Does an aliased dentry already exist? */
2404 alias = __d_find_alias(inode, 0);
2405 if (alias) {
2406 actual = alias;
2407 /* Is this an anonymous mountpoint that we could splice
2408 * into our tree? */
2409 if (IS_ROOT(alias)) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002410 __d_materialise_dentry(dentry, alias);
2411 __d_drop(alias);
2412 goto found;
2413 }
2414 /* Nope, but we must(!) avoid directory aliasing */
Nick Piggin873feea2011-01-07 17:50:06 +11002415 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002416 if (IS_ERR(actual))
2417 dput(alias);
2418 goto out_nolock;
2419 }
David Howells770bfad2006-08-22 20:06:07 -04002420 }
2421
2422 /* Add a unique reference */
2423 actual = __d_instantiate_unique(dentry, inode);
2424 if (!actual)
2425 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002426 else
2427 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002428
David Howells770bfad2006-08-22 20:06:07 -04002429 spin_lock(&actual->d_lock);
2430found:
2431 _d_rehash(actual);
2432 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002433 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002434out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002435 if (actual == dentry) {
2436 security_d_instantiate(dentry, inode);
2437 return NULL;
2438 }
2439
2440 iput(inode);
2441 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002442}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002443EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002444
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002445static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002446{
2447 *buflen -= namelen;
2448 if (*buflen < 0)
2449 return -ENAMETOOLONG;
2450 *buffer -= namelen;
2451 memcpy(*buffer, str, namelen);
2452 return 0;
2453}
2454
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002455static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2456{
2457 return prepend(buffer, buflen, name->name, name->len);
2458}
2459
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002461 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002462 * @path: the dentry/vfsmount to report
2463 * @root: root vfsmnt/dentry (may be modified by this function)
2464 * @buffer: pointer to the end of the buffer
2465 * @buflen: pointer to buffer length
2466 *
Nick Piggin949854d2011-01-07 17:49:37 +11002467 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002468 *
2469 * If path is not reachable from the supplied root, then the value of
2470 * root is changed (without modifying refcounts).
2471 */
2472static int prepend_path(const struct path *path, struct path *root,
2473 char **buffer, int *buflen)
2474{
2475 struct dentry *dentry = path->dentry;
2476 struct vfsmount *vfsmnt = path->mnt;
2477 bool slash = false;
2478 int error = 0;
2479
Nick Piggin99b7db72010-08-18 04:37:39 +10002480 br_read_lock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002481 while (dentry != root->dentry || vfsmnt != root->mnt) {
2482 struct dentry * parent;
2483
2484 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2485 /* Global root? */
2486 if (vfsmnt->mnt_parent == vfsmnt) {
2487 goto global_root;
2488 }
2489 dentry = vfsmnt->mnt_mountpoint;
2490 vfsmnt = vfsmnt->mnt_parent;
2491 continue;
2492 }
2493 parent = dentry->d_parent;
2494 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002495 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002496 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002497 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002498 if (!error)
2499 error = prepend(buffer, buflen, "/", 1);
2500 if (error)
2501 break;
2502
2503 slash = true;
2504 dentry = parent;
2505 }
2506
2507out:
2508 if (!error && !slash)
2509 error = prepend(buffer, buflen, "/", 1);
2510
Nick Piggin99b7db72010-08-18 04:37:39 +10002511 br_read_unlock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002512 return error;
2513
2514global_root:
2515 /*
2516 * Filesystems needing to implement special "root names"
2517 * should do so with ->d_dname()
2518 */
2519 if (IS_ROOT(dentry) &&
2520 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2521 WARN(1, "Root dentry has weird name <%.*s>\n",
2522 (int) dentry->d_name.len, dentry->d_name.name);
2523 }
2524 root->mnt = vfsmnt;
2525 root->dentry = dentry;
2526 goto out;
2527}
2528
2529/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002530 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002531 * @path: the dentry/vfsmount to report
2532 * @root: root vfsmnt/dentry (may be modified by this function)
Randy Dunlapcd956a12010-08-14 13:05:31 -07002533 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 * @buflen: buffer length
2535 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002536 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002538 * Returns a pointer into the buffer or an error code if the
2539 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002540 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002541 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002542 *
2543 * If path is not reachable from the supplied root, then the value of
2544 * root is changed (without modifying refcounts).
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 */
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002546char *__d_path(const struct path *path, struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002547 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002549 char *res = buf + buflen;
2550 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002552 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002553 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002554 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002555 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002556
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002557 if (error)
2558 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002559 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
2561
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002562/*
2563 * same as __d_path but appends "(deleted)" for unlinked files.
2564 */
2565static int path_with_deleted(const struct path *path, struct path *root,
2566 char **buf, int *buflen)
2567{
2568 prepend(buf, buflen, "\0", 1);
2569 if (d_unlinked(path->dentry)) {
2570 int error = prepend(buf, buflen, " (deleted)", 10);
2571 if (error)
2572 return error;
2573 }
2574
2575 return prepend_path(path, root, buf, buflen);
2576}
2577
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002578static int prepend_unreachable(char **buffer, int *buflen)
2579{
2580 return prepend(buffer, buflen, "(unreachable)", 13);
2581}
2582
Jan Bluncka03a8a702008-02-14 19:38:32 -08002583/**
2584 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002585 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002586 * @buf: buffer to return value in
2587 * @buflen: buffer length
2588 *
2589 * Convert a dentry into an ASCII path name. If the entry has been deleted
2590 * the string " (deleted)" is appended. Note that this is ambiguous.
2591 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002592 * Returns a pointer into the buffer or an error code if the path was
2593 * too long. Note: Callers should use the returned pointer, not the passed
2594 * in buffer, to use the name! The implementation often starts at an offset
2595 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002596 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002597 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002598 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002599char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002601 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002602 struct path root;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002603 struct path tmp;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002604 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002606 /*
2607 * We have various synthetic filesystems that never get mounted. On
2608 * these filesystems dentries are never used for lookup purposes, and
2609 * thus don't need to be hashed. They also don't need a name until a
2610 * user wants to identify the object in /proc/pid/fd/. The little hack
2611 * below allows us to generate a name for these objects on demand:
2612 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002613 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2614 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002615
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002616 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002617 write_seqlock(&rename_lock);
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002618 tmp = root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002619 error = path_with_deleted(path, &tmp, &res, &buflen);
2620 if (error)
2621 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002622 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002623 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 return res;
2625}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002626EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002628/**
2629 * d_path_with_unreachable - return the path of a dentry
2630 * @path: path to report
2631 * @buf: buffer to return value in
2632 * @buflen: buffer length
2633 *
2634 * The difference from d_path() is that this prepends "(unreachable)"
2635 * to paths which are unreachable from the current process' root.
2636 */
2637char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2638{
2639 char *res = buf + buflen;
2640 struct path root;
2641 struct path tmp;
2642 int error;
2643
2644 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2645 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2646
2647 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002648 write_seqlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002649 tmp = root;
2650 error = path_with_deleted(path, &tmp, &res, &buflen);
2651 if (!error && !path_equal(&tmp, &root))
2652 error = prepend_unreachable(&res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002653 write_sequnlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002654 path_put(&root);
2655 if (error)
2656 res = ERR_PTR(error);
2657
2658 return res;
2659}
2660
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002662 * Helper function for dentry_operations.d_dname() members
2663 */
2664char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2665 const char *fmt, ...)
2666{
2667 va_list args;
2668 char temp[64];
2669 int sz;
2670
2671 va_start(args, fmt);
2672 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2673 va_end(args);
2674
2675 if (sz > sizeof(temp) || sz > buflen)
2676 return ERR_PTR(-ENAMETOOLONG);
2677
2678 buffer += buflen - sz;
2679 return memcpy(buffer, temp, sz);
2680}
2681
2682/*
Ram Pai6092d042008-03-27 13:06:20 +01002683 * Write full pathname from the root of the filesystem into the buffer.
2684 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002685static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002686{
2687 char *end = buf + buflen;
2688 char *retval;
2689
Ram Pai6092d042008-03-27 13:06:20 +01002690 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002691 if (buflen < 1)
2692 goto Elong;
2693 /* Get '/' right */
2694 retval = end-1;
2695 *retval = '/';
2696
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002697 while (!IS_ROOT(dentry)) {
2698 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002699 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002700
Ram Pai6092d042008-03-27 13:06:20 +01002701 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002702 spin_lock(&dentry->d_lock);
2703 error = prepend_name(&end, &buflen, &dentry->d_name);
2704 spin_unlock(&dentry->d_lock);
2705 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002706 goto Elong;
2707
2708 retval = end;
2709 dentry = parent;
2710 }
Al Viroc1031352010-06-06 22:31:14 -04002711 return retval;
2712Elong:
2713 return ERR_PTR(-ENAMETOOLONG);
2714}
Nick Pigginec2447c2011-01-07 17:49:29 +11002715
2716char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2717{
2718 char *retval;
2719
Nick Piggin949854d2011-01-07 17:49:37 +11002720 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002721 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002722 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002723
2724 return retval;
2725}
2726EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002727
2728char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2729{
2730 char *p = NULL;
2731 char *retval;
2732
Nick Piggin949854d2011-01-07 17:49:37 +11002733 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002734 if (d_unlinked(dentry)) {
2735 p = buf + buflen;
2736 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2737 goto Elong;
2738 buflen++;
2739 }
2740 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002741 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002742 if (!IS_ERR(retval) && p)
2743 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002744 return retval;
2745Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002746 return ERR_PTR(-ENAMETOOLONG);
2747}
2748
2749/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 * NOTE! The user-level library version returns a
2751 * character pointer. The kernel system call just
2752 * returns the length of the buffer filled (which
2753 * includes the ending '\0' character), or a negative
2754 * error value. So libc would do something like
2755 *
2756 * char *getcwd(char * buf, size_t size)
2757 * {
2758 * int retval;
2759 *
2760 * retval = sys_getcwd(buf, size);
2761 * if (retval >= 0)
2762 * return buf;
2763 * errno = -retval;
2764 * return NULL;
2765 * }
2766 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002767SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768{
Linus Torvalds552ce542007-02-13 12:08:18 -08002769 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002770 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002771 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772
2773 if (!page)
2774 return -ENOMEM;
2775
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002776 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777
Linus Torvalds552ce542007-02-13 12:08:18 -08002778 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002779 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002780 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002781 unsigned long len;
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002782 struct path tmp = root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002783 char *cwd = page + PAGE_SIZE;
2784 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002786 prepend(&cwd, &buflen, "\0", 1);
2787 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002788 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002789
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002790 if (error)
Linus Torvalds552ce542007-02-13 12:08:18 -08002791 goto out;
2792
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002793 /* Unreachable from current root */
2794 if (!path_equal(&tmp, &root)) {
2795 error = prepend_unreachable(&cwd, &buflen);
2796 if (error)
2797 goto out;
2798 }
2799
Linus Torvalds552ce542007-02-13 12:08:18 -08002800 error = -ERANGE;
2801 len = PAGE_SIZE + page - cwd;
2802 if (len <= size) {
2803 error = len;
2804 if (copy_to_user(buf, cwd, len))
2805 error = -EFAULT;
2806 }
Nick Piggin949854d2011-01-07 17:49:37 +11002807 } else {
2808 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
2811out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002812 path_put(&pwd);
2813 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 free_page((unsigned long) page);
2815 return error;
2816}
2817
2818/*
2819 * Test whether new_dentry is a subdirectory of old_dentry.
2820 *
2821 * Trivially implemented using the dcache structure
2822 */
2823
2824/**
2825 * is_subdir - is new dentry a subdirectory of old_dentry
2826 * @new_dentry: new dentry
2827 * @old_dentry: old dentry
2828 *
2829 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2830 * Returns 0 otherwise.
2831 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2832 */
2833
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002834int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835{
2836 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002837 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002839 if (new_dentry == old_dentry)
2840 return 1;
2841
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002842 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002845 /*
2846 * Need rcu_readlock to protect against the d_parent trashing
2847 * due to d_move
2848 */
2849 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002850 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002852 else
2853 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002854 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856
2857 return result;
2858}
2859
Al Viro2096f752010-01-30 13:16:21 -05002860int path_is_under(struct path *path1, struct path *path2)
2861{
2862 struct vfsmount *mnt = path1->mnt;
2863 struct dentry *dentry = path1->dentry;
2864 int res;
Nick Piggin99b7db72010-08-18 04:37:39 +10002865
2866 br_read_lock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002867 if (mnt != path2->mnt) {
2868 for (;;) {
2869 if (mnt->mnt_parent == mnt) {
Nick Piggin99b7db72010-08-18 04:37:39 +10002870 br_read_unlock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002871 return 0;
2872 }
2873 if (mnt->mnt_parent == path2->mnt)
2874 break;
2875 mnt = mnt->mnt_parent;
2876 }
2877 dentry = mnt->mnt_mountpoint;
2878 }
2879 res = is_subdir(dentry, path2->dentry);
Nick Piggin99b7db72010-08-18 04:37:39 +10002880 br_read_unlock(vfsmount_lock);
Al Viro2096f752010-01-30 13:16:21 -05002881 return res;
2882}
2883EXPORT_SYMBOL(path_is_under);
2884
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885void d_genocide(struct dentry *root)
2886{
Nick Piggin949854d2011-01-07 17:49:37 +11002887 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002889 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002890 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
Nick Piggin949854d2011-01-07 17:49:37 +11002892 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002893again:
2894 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002895 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896repeat:
2897 next = this_parent->d_subdirs.next;
2898resume:
2899 while (next != &this_parent->d_subdirs) {
2900 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002901 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002903
Nick Pigginda502952011-01-07 17:49:33 +11002904 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2905 if (d_unhashed(dentry) || !dentry->d_inode) {
2906 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002907 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002908 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002910 spin_unlock(&this_parent->d_lock);
2911 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002913 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 goto repeat;
2915 }
Nick Piggin949854d2011-01-07 17:49:37 +11002916 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2917 dentry->d_flags |= DCACHE_GENOCIDE;
2918 dentry->d_count--;
2919 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002920 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 }
2922 if (this_parent != root) {
Nick Piggin949854d2011-01-07 17:49:37 +11002923 struct dentry *tmp;
2924 struct dentry *child;
2925
2926 tmp = this_parent->d_parent;
2927 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2928 this_parent->d_flags |= DCACHE_GENOCIDE;
2929 this_parent->d_count--;
2930 }
2931 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002932 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002933 child = this_parent;
2934 this_parent = tmp;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002935 spin_lock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002936 /* might go back up the wrong parent if we have had a rename
2937 * or deletion */
2938 if (this_parent != child->d_parent ||
Nick Piggin58db63d2011-01-07 17:49:39 +11002939 (!locked && read_seqretry(&rename_lock, seq))) {
Nick Piggin949854d2011-01-07 17:49:37 +11002940 spin_unlock(&this_parent->d_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002941 rcu_read_unlock();
2942 goto rename_retry;
2943 }
2944 rcu_read_unlock();
2945 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 goto resume;
2947 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002948 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002949 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11002950 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11002951 if (locked)
2952 write_sequnlock(&rename_lock);
2953 return;
2954
2955rename_retry:
2956 locked = 1;
2957 write_seqlock(&rename_lock);
2958 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959}
2960
2961/**
2962 * find_inode_number - check for dentry with name
2963 * @dir: directory to check
2964 * @name: Name to find.
2965 *
2966 * Check whether a dentry already exists for the given name,
2967 * and return the inode number if it has an inode. Otherwise
2968 * 0 is returned.
2969 *
2970 * This routine is used to post-process directory listings for
2971 * filesystems using synthetic inode numbers, and is necessary
2972 * to keep getcwd() working.
2973 */
2974
2975ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2976{
2977 struct dentry * dentry;
2978 ino_t ino = 0;
2979
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002980 dentry = d_hash_and_lookup(dir, name);
2981 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 if (dentry->d_inode)
2983 ino = dentry->d_inode->i_ino;
2984 dput(dentry);
2985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 return ino;
2987}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002988EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989
2990static __initdata unsigned long dhash_entries;
2991static int __init set_dhash_entries(char *str)
2992{
2993 if (!str)
2994 return 0;
2995 dhash_entries = simple_strtoul(str, &str, 0);
2996 return 1;
2997}
2998__setup("dhash_entries=", set_dhash_entries);
2999
3000static void __init dcache_init_early(void)
3001{
3002 int loop;
3003
3004 /* If hashes are distributed across NUMA nodes, defer
3005 * hash allocation until vmalloc space is available.
3006 */
3007 if (hashdist)
3008 return;
3009
3010 dentry_hashtable =
3011 alloc_large_system_hash("Dentry cache",
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003012 sizeof(struct dcache_hash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 dhash_entries,
3014 13,
3015 HASH_EARLY,
3016 &d_hash_shift,
3017 &d_hash_mask,
3018 0);
3019
3020 for (loop = 0; loop < (1 << d_hash_shift); loop++)
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003021 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022}
3023
Denis Cheng74bf17c2007-10-16 23:26:30 -07003024static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025{
3026 int loop;
3027
3028 /*
3029 * A constructor could be added for stable state like the lists,
3030 * but it is probably not worth it because of the cache nature
3031 * of the dcache.
3032 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003033 dentry_cache = KMEM_CACHE(dentry,
3034 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035
Rusty Russell8e1f9362007-07-17 04:03:17 -07003036 register_shrinker(&dcache_shrinker);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037
3038 /* Hash may have been set up in dcache_init_early */
3039 if (!hashdist)
3040 return;
3041
3042 dentry_hashtable =
3043 alloc_large_system_hash("Dentry cache",
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003044 sizeof(struct dcache_hash_bucket),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 dhash_entries,
3046 13,
3047 0,
3048 &d_hash_shift,
3049 &d_hash_mask,
3050 0);
3051
3052 for (loop = 0; loop < (1 << d_hash_shift); loop++)
Nick Pigginceb5bdc2011-01-07 17:50:05 +11003053 INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054}
3055
3056/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003057struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003058EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060EXPORT_SYMBOL(d_genocide);
3061
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062void __init vfs_caches_init_early(void)
3063{
3064 dcache_init_early();
3065 inode_init_early();
3066}
3067
3068void __init vfs_caches_init(unsigned long mempages)
3069{
3070 unsigned long reserve;
3071
3072 /* Base hash sizes on available memory, with a reserve equal to
3073 150% of current kernel size */
3074
3075 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3076 mempages -= reserve;
3077
3078 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003079 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
Denis Cheng74bf17c2007-10-16 23:26:30 -07003081 dcache_init();
3082 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003084 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085 bdev_cache_init();
3086 chrdev_init();
3087}