blob: 98b48753f77beca2b4aae725939dc68b4b7bba51 [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>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
David Howells07f3f052006-09-30 20:52:18 +020040#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050041#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Nick Piggin789680d2011-01-07 17:49:30 +110043/*
44 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110045 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110047 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
Nick Piggin23044502011-01-07 17:49:31 +110051 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
53 * d_lock protects:
54 * - d_flags
55 * - d_name
56 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110057 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110058 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110059 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110061 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110062 *
63 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110064 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110065 * dentry->d_lock
66 * dcache_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110067 * dcache_hash_bucket lock
68 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110069 *
Nick Pigginda502952011-01-07 17:49:33 +110070 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
72 * ...
73 * dentry->d_parent->d_lock
74 * dentry->d_lock
75 *
76 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110077 * if (dentry1 < dentry2)
78 * dentry1->d_lock
79 * dentry2->d_lock
80 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080081int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83
Nick Piggin23044502011-01-07 17:49:31 +110084static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
99#define D_HASHBITS d_hash_shift
100#define D_HASHMASK d_hash_mask
101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700107static inline struct hlist_bl_head *d_hash(struct dentry *parent,
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100108 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115/* Statistics gathering. */
116struct dentry_stat_t dentry_stat = {
117 .age_limit = 45,
118};
119
Nick Piggin3e880fb2011-01-07 17:49:19 +1100120static DEFINE_PER_CPU(unsigned int, nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100123static int get_nr_dentry(void)
124{
125 int i;
126 int sum = 0;
127 for_each_possible_cpu(i)
128 sum += per_cpu(nr_dentry, i);
129 return sum < 0 ? 0 : sum;
130}
131
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400132int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
133 size_t *lenp, loff_t *ppos)
134{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100135 dentry_stat.nr_dentry = get_nr_dentry();
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400136 return proc_dointvec(table, write, buffer, lenp, ppos);
137}
138#endif
139
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400140static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400142 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
143
Arjan van de Venfd217f42008-10-21 06:47:33 -0700144 WARN_ON(!list_empty(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (dname_external(dentry))
146 kfree(dentry->d_name.name);
147 kmem_cache_free(dentry_cache, dentry);
148}
149
150/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100151 * no locks, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 */
153static void d_free(struct dentry *dentry)
154{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100155 BUG_ON(dentry->d_count);
Nick Piggin3e880fb2011-01-07 17:49:19 +1100156 this_cpu_dec(nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 if (dentry->d_op && dentry->d_op->d_release)
158 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400159
Linus Torvaldsdea36672011-04-24 07:58:46 -0700160 /* if dentry was never visible to RCU, immediate free is OK */
161 if (!(dentry->d_flags & DCACHE_RCUACCESS))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400162 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800163 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400164 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165}
166
Nick Piggin31e6b012011-01-07 17:49:52 +1100167/**
168 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800169 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100170 * After this call, in-progress rcu-walk path lookup will fail. This
171 * should be called after unhashing, and after changing d_inode (if
172 * the dentry has not already been unhashed).
173 */
174static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
175{
176 assert_spin_locked(&dentry->d_lock);
177 /* Go through a barrier */
178 write_seqcount_barrier(&dentry->d_seq);
179}
180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181/*
182 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100183 * d_iput() operation if defined. Dentry has no refcount
184 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800186static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200187 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100188 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
190 struct inode *inode = dentry->d_inode;
191 if (inode) {
192 dentry->d_inode = NULL;
193 list_del_init(&dentry->d_alias);
194 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100195 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700196 if (!inode->i_nlink)
197 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 if (dentry->d_op && dentry->d_op->d_iput)
199 dentry->d_op->d_iput(dentry, inode);
200 else
201 iput(inode);
202 } else {
203 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 }
205}
206
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700207/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100208 * Release the dentry's inode, using the filesystem
209 * d_iput() operation if defined. dentry remains in-use.
210 */
211static void dentry_unlink_inode(struct dentry * dentry)
212 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100213 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100214{
215 struct inode *inode = dentry->d_inode;
216 dentry->d_inode = NULL;
217 list_del_init(&dentry->d_alias);
218 dentry_rcuwalk_barrier(dentry);
219 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100220 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100221 if (!inode->i_nlink)
222 fsnotify_inoderemove(inode);
223 if (dentry->d_op && dentry->d_op->d_iput)
224 dentry->d_op->d_iput(dentry, inode);
225 else
226 iput(inode);
227}
228
229/*
Sage Weilf0023bc2011-10-28 10:02:42 -0700230 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700231 */
232static void dentry_lru_add(struct dentry *dentry)
233{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400234 if (list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100235 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400236 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
237 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100238 dentry_stat.nr_unused++;
Nick Piggin23044502011-01-07 17:49:31 +1100239 spin_unlock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400240 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700241}
242
Nick Piggin23044502011-01-07 17:49:31 +1100243static void __dentry_lru_del(struct dentry *dentry)
244{
245 list_del_init(&dentry->d_lru);
246 dentry->d_sb->s_nr_dentry_unused--;
247 dentry_stat.nr_unused--;
248}
249
Sage Weilf0023bc2011-10-28 10:02:42 -0700250/*
251 * Remove a dentry with references from the LRU.
252 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700253static void dentry_lru_del(struct dentry *dentry)
254{
255 if (!list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100256 spin_lock(&dcache_lru_lock);
257 __dentry_lru_del(dentry);
258 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700259 }
260}
261
Sage Weilf0023bc2011-10-28 10:02:42 -0700262/*
263 * Remove a dentry that is unreferenced and about to be pruned
264 * (unhashed and destroyed) from the LRU, and inform the file system.
265 * This wrapper should be called _prior_ to unhashing a victim dentry.
266 */
267static void dentry_lru_prune(struct dentry *dentry)
268{
269 if (!list_empty(&dentry->d_lru)) {
270 if (dentry->d_flags & DCACHE_OP_PRUNE)
271 dentry->d_op->d_prune(dentry);
272
273 spin_lock(&dcache_lru_lock);
274 __dentry_lru_del(dentry);
275 spin_unlock(&dcache_lru_lock);
276 }
277}
278
Christoph Hellwiga4633352010-10-10 05:36:26 -0400279static void dentry_lru_move_tail(struct dentry *dentry)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700280{
Nick Piggin23044502011-01-07 17:49:31 +1100281 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400282 if (list_empty(&dentry->d_lru)) {
283 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
284 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100285 dentry_stat.nr_unused++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400286 } else {
287 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700288 }
Nick Piggin23044502011-01-07 17:49:31 +1100289 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700290}
291
Miklos Szeredid52b9082007-05-08 00:23:46 -0700292/**
293 * d_kill - kill dentry and return parent
294 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800295 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700296 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200297 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700298 *
299 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100300 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100301 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
302 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700303 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100304static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200305 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100306 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100307 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700308{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700309 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400310 /*
311 * Inform try_to_ascend() that we are no longer attached to the
312 * dentry tree
313 */
314 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100315 if (parent)
316 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700317 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100318 /*
319 * dentry_iput drops the locks, at which point nobody (except
320 * transient RCU lookups) can reach this dentry.
321 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700322 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900323 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700324}
325
David Howellsc6627c62011-06-07 14:09:20 +0100326/*
327 * Unhash a dentry without inserting an RCU walk barrier or checking that
328 * dentry->d_lock is locked. The caller must take care of that, if
329 * appropriate.
330 */
331static void __d_shrink(struct dentry *dentry)
332{
333 if (!d_unhashed(dentry)) {
334 struct hlist_bl_head *b;
335 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
336 b = &dentry->d_sb->s_anon;
337 else
338 b = d_hash(dentry->d_parent, dentry->d_name.hash);
339
340 hlist_bl_lock(b);
341 __hlist_bl_del(&dentry->d_hash);
342 dentry->d_hash.pprev = NULL;
343 hlist_bl_unlock(b);
344 }
345}
346
Nick Piggin789680d2011-01-07 17:49:30 +1100347/**
348 * d_drop - drop a dentry
349 * @dentry: dentry to drop
350 *
351 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
352 * be found through a VFS lookup any more. Note that this is different from
353 * deleting the dentry - d_delete will try to mark the dentry negative if
354 * possible, giving a successful _negative_ lookup, while d_drop will
355 * just make the cache lookup fail.
356 *
357 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
358 * reason (NFS timeouts or autofs deletes).
359 *
360 * __d_drop requires dentry->d_lock.
361 */
362void __d_drop(struct dentry *dentry)
363{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700364 if (!d_unhashed(dentry)) {
David Howellsc6627c62011-06-07 14:09:20 +0100365 __d_shrink(dentry);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700366 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100367 }
368}
369EXPORT_SYMBOL(__d_drop);
370
371void d_drop(struct dentry *dentry)
372{
Nick Piggin789680d2011-01-07 17:49:30 +1100373 spin_lock(&dentry->d_lock);
374 __d_drop(dentry);
375 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100376}
377EXPORT_SYMBOL(d_drop);
378
Nick Piggin77812a12011-01-07 17:49:48 +1100379/*
Josef Bacik44396f42011-05-31 11:58:49 -0400380 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
381 * @dentry: dentry to drop
382 *
383 * This is called when we do a lookup on a placeholder dentry that needed to be
384 * looked up. The dentry should have been hashed in order for it to be found by
385 * the lookup code, but now needs to be unhashed while we do the actual lookup
386 * and clear the DCACHE_NEED_LOOKUP flag.
387 */
388void d_clear_need_lookup(struct dentry *dentry)
389{
390 spin_lock(&dentry->d_lock);
391 __d_drop(dentry);
392 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
393 spin_unlock(&dentry->d_lock);
394}
395EXPORT_SYMBOL(d_clear_need_lookup);
396
397/*
Nick Piggin77812a12011-01-07 17:49:48 +1100398 * Finish off a dentry we've decided to kill.
399 * dentry->d_lock must be held, returns with it unlocked.
400 * If ref is non-zero, then decrement the refcount too.
401 * Returns dentry requiring refcount drop, or NULL if we're done.
402 */
403static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
404 __releases(dentry->d_lock)
405{
Nick Piggin873feea2011-01-07 17:50:06 +1100406 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100407 struct dentry *parent;
408
Nick Piggin873feea2011-01-07 17:50:06 +1100409 inode = dentry->d_inode;
410 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100411relock:
412 spin_unlock(&dentry->d_lock);
413 cpu_relax();
414 return dentry; /* try again with same dentry */
415 }
416 if (IS_ROOT(dentry))
417 parent = NULL;
418 else
419 parent = dentry->d_parent;
420 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100421 if (inode)
422 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100423 goto relock;
424 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100425
Nick Piggin77812a12011-01-07 17:49:48 +1100426 if (ref)
427 dentry->d_count--;
Sage Weilf0023bc2011-10-28 10:02:42 -0700428 /*
429 * if dentry was on the d_lru list delete it from there.
430 * inform the fs via d_prune that this dentry is about to be
431 * unhashed and destroyed.
432 */
433 dentry_lru_prune(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100434 /* if it was on the hash then remove it */
435 __d_drop(dentry);
436 return d_kill(dentry, parent);
437}
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439/*
440 * This is dput
441 *
442 * This is complicated by the fact that we do not want to put
443 * dentries that are no longer on any hash chain on the unused
444 * list: we'd much rather just get rid of them immediately.
445 *
446 * However, that implies that we have to traverse the dentry
447 * tree upwards to the parents which might _also_ now be
448 * scheduled for deletion (it may have been only waiting for
449 * its last child to go away).
450 *
451 * This tail recursion is done by hand as we don't want to depend
452 * on the compiler to always get this right (gcc generally doesn't).
453 * Real recursion would eat up our stack space.
454 */
455
456/*
457 * dput - release a dentry
458 * @dentry: dentry to release
459 *
460 * Release a dentry. This will drop the usage count and if appropriate
461 * call the dentry unlink method as well as removing it from the queues and
462 * releasing its resources. If the parent dentries were scheduled for release
463 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465void dput(struct dentry *dentry)
466{
467 if (!dentry)
468 return;
469
470repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100471 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100474 BUG_ON(!dentry->d_count);
475 if (dentry->d_count > 1) {
476 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 return;
479 }
480
Nick Pigginfb045ad2011-01-07 17:49:55 +1100481 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100483 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 }
Nick Piggin265ac902010-10-10 05:36:24 -0400485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 /* Unreachable? Get rid of it */
487 if (d_unhashed(dentry))
488 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400489
Josef Bacik44396f42011-05-31 11:58:49 -0400490 /*
491 * If this dentry needs lookup, don't set the referenced flag so that it
492 * is more likely to be cleaned up by the dcache shrinker in case of
493 * memory pressure.
494 */
495 if (!d_need_lookup(dentry))
496 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400497 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400498
Nick Piggin61f3dee2011-01-07 17:49:40 +1100499 dentry->d_count--;
500 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return;
502
Miklos Szeredid52b9082007-05-08 00:23:46 -0700503kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100504 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700505 if (dentry)
506 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700508EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
510/**
511 * d_invalidate - invalidate a dentry
512 * @dentry: dentry to invalidate
513 *
514 * Try to invalidate the dentry if it turns out to be
515 * possible. If there are other dentries that can be
516 * reached through this one we can't delete it and we
517 * return -EBUSY. On success we return 0.
518 *
519 * no dcache lock.
520 */
521
522int d_invalidate(struct dentry * dentry)
523{
524 /*
525 * If it's already been dropped, return OK.
526 */
Nick Pigginda502952011-01-07 17:49:33 +1100527 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100529 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 return 0;
531 }
532 /*
533 * Check whether to do a partial shrink_dcache
534 * to get rid of unused child entries.
535 */
536 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100537 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100539 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 }
541
542 /*
543 * Somebody else still using it?
544 *
545 * If it's a directory, we can't drop it
546 * for fear of somebody re-populating it
547 * with children (even though dropping it
548 * would make it unreachable from the root,
549 * we might still populate it if it was a
550 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000551 * We also need to leave mountpoints alone,
552 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 */
Al Viro50e69632011-11-07 16:39:57 +0000554 if (dentry->d_count > 1 && dentry->d_inode) {
555 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return -EBUSY;
558 }
559 }
560
561 __d_drop(dentry);
562 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 return 0;
564}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700565EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100567/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100568static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100570 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
Nick Piggindc0474b2011-01-07 17:49:43 +1100573static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100574{
Nick Piggin23044502011-01-07 17:49:31 +1100575 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100576 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100577 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100578}
579
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100580struct dentry *dget_parent(struct dentry *dentry)
581{
582 struct dentry *ret;
583
584repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100585 /*
586 * Don't need rcu_dereference because we re-check it was correct under
587 * the lock.
588 */
589 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100590 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100591 spin_lock(&ret->d_lock);
592 if (unlikely(ret != dentry->d_parent)) {
593 spin_unlock(&ret->d_lock);
594 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100595 goto repeat;
596 }
Nick Piggina734eb42011-01-07 17:49:44 +1100597 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100598 BUG_ON(!ret->d_count);
599 ret->d_count++;
600 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100601 return ret;
602}
603EXPORT_SYMBOL(dget_parent);
604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605/**
606 * d_find_alias - grab a hashed alias of inode
607 * @inode: inode in question
608 * @want_discon: flag, used by d_splice_alias, to request
609 * that only a DISCONNECTED alias be returned.
610 *
611 * If inode has a hashed alias, or is a directory and has any alias,
612 * acquire the reference to alias and return it. Otherwise return NULL.
613 * Notice that if inode is a directory there can be only one alias and
614 * it can be unhashed only if it has no children, or if it is the root
615 * of a filesystem.
616 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700617 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 * any other hashed alias over that one unless @want_discon is set,
NeilBrown21c0d8f2006-10-04 02:16:16 -0700619 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 */
Nick Pigginda502952011-01-07 17:49:33 +1100621static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Nick Pigginda502952011-01-07 17:49:33 +1100623 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Nick Pigginda502952011-01-07 17:49:33 +1100625again:
626 discon_alias = NULL;
627 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
628 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700630 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100631 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 discon_alias = alias;
Nick Pigginda502952011-01-07 17:49:33 +1100633 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100634 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100635 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return alias;
637 }
638 }
Nick Pigginda502952011-01-07 17:49:33 +1100639 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 }
Nick Pigginda502952011-01-07 17:49:33 +1100641 if (discon_alias) {
642 alias = discon_alias;
643 spin_lock(&alias->d_lock);
644 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
645 if (IS_ROOT(alias) &&
646 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100647 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100648 spin_unlock(&alias->d_lock);
649 return alias;
650 }
651 }
652 spin_unlock(&alias->d_lock);
653 goto again;
654 }
655 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656}
657
Nick Pigginda502952011-01-07 17:49:33 +1100658struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
David Howells214fda12006-03-25 03:06:36 -0800660 struct dentry *de = NULL;
661
662 if (!list_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100663 spin_lock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800664 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100665 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 return de;
668}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700669EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671/*
672 * Try to kill dentries associated with this inode.
673 * WARNING: you must own a reference to inode.
674 */
675void d_prune_aliases(struct inode *inode)
676{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700677 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100679 spin_lock(&inode->i_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700680 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100682 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100683 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 __d_drop(dentry);
685 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100686 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 dput(dentry);
688 goto restart;
689 }
690 spin_unlock(&dentry->d_lock);
691 }
Nick Piggin873feea2011-01-07 17:50:06 +1100692 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700694EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696/*
Nick Piggin77812a12011-01-07 17:49:48 +1100697 * Try to throw away a dentry - free the inode, dput the parent.
698 * Requires dentry->d_lock is held, and dentry->d_count == 0.
699 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700700 *
Nick Piggin77812a12011-01-07 17:49:48 +1100701 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 */
Nick Piggin77812a12011-01-07 17:49:48 +1100703static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200704 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Nick Piggin77812a12011-01-07 17:49:48 +1100706 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700707
Nick Piggin77812a12011-01-07 17:49:48 +1100708 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700709 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100710 * If dentry_kill returns NULL, we have nothing more to do.
711 * if it returns the same dentry, trylocks failed. In either
712 * case, just loop again.
713 *
714 * Otherwise, we need to prune ancestors too. This is necessary
715 * to prevent quadratic behavior of shrink_dcache_parent(), but
716 * is also expected to be beneficial in reducing dentry cache
717 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700718 */
Nick Piggin77812a12011-01-07 17:49:48 +1100719 if (!parent)
720 return;
721 if (parent == dentry)
722 return;
723
724 /* Prune ancestors. */
725 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700726 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100727 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100728 if (dentry->d_count > 1) {
729 dentry->d_count--;
730 spin_unlock(&dentry->d_lock);
731 return;
732 }
Nick Piggin77812a12011-01-07 17:49:48 +1100733 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700734 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
736
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400737static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700739 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700740
Nick Pigginec336792011-01-07 17:49:47 +1100741 rcu_read_lock();
742 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100743 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
744 if (&dentry->d_lru == list)
745 break; /* empty */
746 spin_lock(&dentry->d_lock);
747 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
748 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100749 continue;
750 }
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 /*
753 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700754 * the LRU because of laziness during lookup. Do not free
755 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100757 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100758 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700759 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 continue;
761 }
Nick Pigginec336792011-01-07 17:49:47 +1100762
Nick Pigginec336792011-01-07 17:49:47 +1100763 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100764
765 try_prune_one_dentry(dentry);
766
Nick Pigginec336792011-01-07 17:49:47 +1100767 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
Nick Pigginec336792011-01-07 17:49:47 +1100769 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400770}
771
772/**
773 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
774 * @sb: superblock to shrink dentry LRU.
775 * @count: number of entries to prune
776 * @flags: flags to control the dentry processing
777 *
778 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
779 */
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000780static void __shrink_dcache_sb(struct super_block *sb, int count, int flags)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400781{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400782 struct dentry *dentry;
783 LIST_HEAD(referenced);
784 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400785
Nick Piggin23044502011-01-07 17:49:31 +1100786relock:
787 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400788 while (!list_empty(&sb->s_dentry_lru)) {
789 dentry = list_entry(sb->s_dentry_lru.prev,
790 struct dentry, d_lru);
791 BUG_ON(dentry->d_sb != sb);
792
Nick Piggin23044502011-01-07 17:49:31 +1100793 if (!spin_trylock(&dentry->d_lock)) {
794 spin_unlock(&dcache_lru_lock);
795 cpu_relax();
796 goto relock;
797 }
798
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400799 /*
800 * If we are honouring the DCACHE_REFERENCED flag and the
801 * dentry has this flag set, don't free it. Clear the flag
802 * and put it back on the LRU.
803 */
Nick Piggin23044502011-01-07 17:49:31 +1100804 if (flags & DCACHE_REFERENCED &&
805 dentry->d_flags & DCACHE_REFERENCED) {
806 dentry->d_flags &= ~DCACHE_REFERENCED;
807 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400808 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100809 } else {
810 list_move_tail(&dentry->d_lru, &tmp);
811 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000812 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100813 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400814 }
Nick Pigginec336792011-01-07 17:49:47 +1100815 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400816 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700817 if (!list_empty(&referenced))
818 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100819 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100820
821 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822}
823
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700824/**
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000825 * prune_dcache_sb - shrink the dcache
Randy Dunlap2af14162011-08-03 16:21:07 -0700826 * @sb: superblock
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000827 * @nr_to_scan: number of entries to try to free
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 *
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000829 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
830 * done when we need more memory an called from the superblock shrinker
831 * function.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700832 *
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000833 * This function may fail to free any resources if all the dentries are in
834 * use.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 */
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000836void prune_dcache_sb(struct super_block *sb, int nr_to_scan)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700837{
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000838 __shrink_dcache_sb(sb, nr_to_scan, DCACHE_REFERENCED);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700839}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841/**
842 * shrink_dcache_sb - shrink dcache for a superblock
843 * @sb: superblock
844 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400845 * Shrink the dcache for the specified super block. This is used to free
846 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400848void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400850 LIST_HEAD(tmp);
851
Nick Piggin23044502011-01-07 17:49:31 +1100852 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400853 while (!list_empty(&sb->s_dentry_lru)) {
854 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100855 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400856 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100857 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400858 }
Nick Piggin23044502011-01-07 17:49:31 +1100859 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700861EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863/*
David Howellsc636ebd2006-10-11 01:22:19 -0700864 * destroy a single subtree of dentries for unmount
865 * - see the comments on shrink_dcache_for_umount() for a description of the
866 * locking
867 */
868static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
869{
870 struct dentry *parent;
871
872 BUG_ON(!IS_ROOT(dentry));
873
David Howellsc636ebd2006-10-11 01:22:19 -0700874 for (;;) {
875 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100876 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700877 dentry = list_entry(dentry->d_subdirs.next,
878 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700879
880 /* consume the dentries from this leaf up through its parents
881 * until we find one with children or run out altogether */
882 do {
883 struct inode *inode;
884
Sage Weilf0023bc2011-10-28 10:02:42 -0700885 /*
886 * remove the dentry from the lru, and inform
887 * the fs that this dentry is about to be
888 * unhashed and destroyed.
889 */
890 dentry_lru_prune(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100891 __d_shrink(dentry);
892
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100893 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700894 printk(KERN_ERR
895 "BUG: Dentry %p{i=%lx,n=%s}"
896 " still in use (%d)"
897 " [unmount of %s %s]\n",
898 dentry,
899 dentry->d_inode ?
900 dentry->d_inode->i_ino : 0UL,
901 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100902 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700903 dentry->d_sb->s_type->name,
904 dentry->d_sb->s_id);
905 BUG();
906 }
907
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100908 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700909 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100910 list_del(&dentry->d_u.d_child);
911 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900912 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100913 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100914 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900915 }
David Howellsc636ebd2006-10-11 01:22:19 -0700916
David Howellsc636ebd2006-10-11 01:22:19 -0700917 inode = dentry->d_inode;
918 if (inode) {
919 dentry->d_inode = NULL;
920 list_del_init(&dentry->d_alias);
921 if (dentry->d_op && dentry->d_op->d_iput)
922 dentry->d_op->d_iput(dentry, inode);
923 else
924 iput(inode);
925 }
926
927 d_free(dentry);
928
929 /* finished when we fall off the top of the tree,
930 * otherwise we ascend to the parent and move to the
931 * next sibling if there is one */
932 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400933 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700934 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700935 } while (list_empty(&dentry->d_subdirs));
936
937 dentry = list_entry(dentry->d_subdirs.next,
938 struct dentry, d_u.d_child);
939 }
940}
941
942/*
943 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100944 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700945 * - the superblock is detached from all mountings and open files, so the
946 * dentry trees will not be rearranged by the VFS
947 * - s_umount is write-locked, so the memory pressure shrinker will ignore
948 * any dentries belonging to this superblock that it comes across
949 * - the filesystem itself is no longer permitted to rearrange the dentries
950 * in this superblock
951 */
952void shrink_dcache_for_umount(struct super_block *sb)
953{
954 struct dentry *dentry;
955
956 if (down_read_trylock(&sb->s_umount))
957 BUG();
958
959 dentry = sb->s_root;
960 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100961 dentry->d_count--;
David Howellsc636ebd2006-10-11 01:22:19 -0700962 shrink_dcache_for_umount_subtree(dentry);
963
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100964 while (!hlist_bl_empty(&sb->s_anon)) {
965 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -0700966 shrink_dcache_for_umount_subtree(dentry);
967 }
968}
969
970/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -0700971 * This tries to ascend one level of parenthood, but
972 * we can race with renaming, so we need to re-check
973 * the parenthood after dropping the lock and check
974 * that the sequence number still matches.
975 */
976static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
977{
978 struct dentry *new = old->d_parent;
979
980 rcu_read_lock();
981 spin_unlock(&old->d_lock);
982 spin_lock(&new->d_lock);
983
984 /*
985 * might go back up the wrong parent if we have had a rename
986 * or deletion
987 */
988 if (new != old->d_parent ||
Trond Myklebustc83ce982011-03-15 13:36:43 -0400989 (old->d_flags & DCACHE_DISCONNECTED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -0700990 (!locked && read_seqretry(&rename_lock, seq))) {
991 spin_unlock(&new->d_lock);
992 new = NULL;
993 }
994 rcu_read_unlock();
995 return new;
996}
997
998
999/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 * Search for at least 1 mount point in the dentry's subdirs.
1001 * We descend to the next level whenever the d_subdirs
1002 * list is non-empty and continue searching.
1003 */
1004
1005/**
1006 * have_submounts - check for mounts over a dentry
1007 * @parent: dentry to check.
1008 *
1009 * Return true if the parent or its subdirectories contain
1010 * a mount point
1011 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012int have_submounts(struct dentry *parent)
1013{
Nick Piggin949854d2011-01-07 17:49:37 +11001014 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001016 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001017 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001018
Nick Piggin949854d2011-01-07 17:49:37 +11001019 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001020again:
1021 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (d_mountpoint(parent))
1024 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001025 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026repeat:
1027 next = this_parent->d_subdirs.next;
1028resume:
1029 while (next != &this_parent->d_subdirs) {
1030 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001031 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001033
1034 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001036 if (d_mountpoint(dentry)) {
1037 spin_unlock(&dentry->d_lock);
1038 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001042 spin_unlock(&this_parent->d_lock);
1043 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001045 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 goto repeat;
1047 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001048 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 }
1050 /*
1051 * All done at this level ... ascend and resume the search.
1052 */
1053 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001054 struct dentry *child = this_parent;
1055 this_parent = try_to_ascend(this_parent, locked, seq);
1056 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001057 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001058 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 goto resume;
1060 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001061 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001062 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001063 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001064 if (locked)
1065 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 return 0; /* No mount points found in tree */
1067positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001068 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001069 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001070 if (locked)
1071 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001073
1074rename_retry:
1075 locked = 1;
1076 write_seqlock(&rename_lock);
1077 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001079EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081/*
1082 * Search the dentry child list for the specified parent,
1083 * and move any unused dentries to the end of the unused
1084 * list for prune_dcache(). We descend to the next level
1085 * whenever the d_subdirs list is non-empty and continue
1086 * searching.
1087 *
1088 * It returns zero iff there are no unused children,
1089 * otherwise it returns the number of children moved to
1090 * the end of the unused list. This may not be the total
1091 * number of unused children, because select_parent can
1092 * drop the lock and return early due to latency
1093 * constraints.
1094 */
1095static int select_parent(struct dentry * parent)
1096{
Nick Piggin949854d2011-01-07 17:49:37 +11001097 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001099 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001101 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Nick Piggin949854d2011-01-07 17:49:37 +11001103 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001104again:
1105 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001106 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107repeat:
1108 next = this_parent->d_subdirs.next;
1109resume:
1110 while (next != &this_parent->d_subdirs) {
1111 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001112 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 next = tmp->next;
1114
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001115 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 /*
1118 * move only zero ref count dentries to the end
1119 * of the unused list for prune_dcache
1120 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001121 if (!dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001122 dentry_lru_move_tail(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 found++;
Christoph Hellwiga4633352010-10-10 05:36:26 -04001124 } else {
1125 dentry_lru_del(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 }
1127
1128 /*
1129 * We can return to the caller if we have found some (this
1130 * ensures forward progress). We'll be coming back to find
1131 * the rest.
1132 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001133 if (found && need_resched()) {
1134 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
1138 /*
1139 * Descend a level if the d_subdirs list is non-empty.
1140 */
1141 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001142 spin_unlock(&this_parent->d_lock);
1143 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001145 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 goto repeat;
1147 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001148
1149 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 }
1151 /*
1152 * All done at this level ... ascend and resume the search.
1153 */
1154 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001155 struct dentry *child = this_parent;
1156 this_parent = try_to_ascend(this_parent, locked, seq);
1157 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001158 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001159 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 goto resume;
1161 }
1162out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001163 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001164 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001165 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001166 if (locked)
1167 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001169
1170rename_retry:
1171 if (found)
1172 return found;
1173 locked = 1;
1174 write_seqlock(&rename_lock);
1175 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176}
1177
1178/**
1179 * shrink_dcache_parent - prune dcache
1180 * @parent: parent of entries to prune
1181 *
1182 * Prune the dcache to remove unused children of the parent dentry.
1183 */
1184
1185void shrink_dcache_parent(struct dentry * parent)
1186{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001187 struct super_block *sb = parent->d_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 int found;
1189
1190 while ((found = select_parent(parent)) != 0)
Dave Chinnerb0d40c92011-07-08 14:14:42 +10001191 __shrink_dcache_sb(sb, found, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001193EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195/**
Al Viroa4464db2011-07-07 15:03:58 -04001196 * __d_alloc - allocate a dcache entry
1197 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 * @name: qstr of the name
1199 *
1200 * Allocates a dentry. It returns %NULL if there is insufficient memory
1201 * available. On a success the dentry is returned. The name passed in is
1202 * copied and the copy passed in may be reused after this call.
1203 */
1204
Al Viroa4464db2011-07-07 15:03:58 -04001205struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206{
1207 struct dentry *dentry;
1208 char *dname;
1209
Mel Gormane12ba742007-10-16 01:25:52 -07001210 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 if (!dentry)
1212 return NULL;
1213
1214 if (name->len > DNAME_INLINE_LEN-1) {
1215 dname = kmalloc(name->len + 1, GFP_KERNEL);
1216 if (!dname) {
1217 kmem_cache_free(dentry_cache, dentry);
1218 return NULL;
1219 }
1220 } else {
1221 dname = dentry->d_iname;
1222 }
1223 dentry->d_name.name = dname;
1224
1225 dentry->d_name.len = name->len;
1226 dentry->d_name.hash = name->hash;
1227 memcpy(dname, name->name, name->len);
1228 dname[name->len] = 0;
1229
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001230 dentry->d_count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001231 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001233 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001235 dentry->d_parent = dentry;
1236 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 dentry->d_op = NULL;
1238 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001239 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 INIT_LIST_HEAD(&dentry->d_lru);
1241 INIT_LIST_HEAD(&dentry->d_subdirs);
1242 INIT_LIST_HEAD(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001243 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001244 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
Nick Piggin3e880fb2011-01-07 17:49:19 +11001246 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 return dentry;
1249}
Al Viroa4464db2011-07-07 15:03:58 -04001250
1251/**
1252 * d_alloc - allocate a dcache entry
1253 * @parent: parent of entry to allocate
1254 * @name: qstr of the name
1255 *
1256 * Allocates a dentry. It returns %NULL if there is insufficient memory
1257 * available. On a success the dentry is returned. The name passed in is
1258 * copied and the copy passed in may be reused after this call.
1259 */
1260struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1261{
1262 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1263 if (!dentry)
1264 return NULL;
1265
1266 spin_lock(&parent->d_lock);
1267 /*
1268 * don't need child lock because it is not subject
1269 * to concurrency here
1270 */
1271 __dget_dlock(parent);
1272 dentry->d_parent = parent;
1273 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1274 spin_unlock(&parent->d_lock);
1275
1276 return dentry;
1277}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001278EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
Nick Piggin4b936882011-01-07 17:50:07 +11001280struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1281{
Al Viroa4464db2011-07-07 15:03:58 -04001282 struct dentry *dentry = __d_alloc(sb, name);
1283 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001284 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001285 return dentry;
1286}
1287EXPORT_SYMBOL(d_alloc_pseudo);
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1290{
1291 struct qstr q;
1292
1293 q.name = name;
1294 q.len = strlen(name);
1295 q.hash = full_name_hash(q.name, q.len);
1296 return d_alloc(parent, &q);
1297}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001298EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Nick Pigginfb045ad2011-01-07 17:49:55 +11001300void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1301{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001302 WARN_ON_ONCE(dentry->d_op);
1303 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001304 DCACHE_OP_COMPARE |
1305 DCACHE_OP_REVALIDATE |
1306 DCACHE_OP_DELETE ));
1307 dentry->d_op = op;
1308 if (!op)
1309 return;
1310 if (op->d_hash)
1311 dentry->d_flags |= DCACHE_OP_HASH;
1312 if (op->d_compare)
1313 dentry->d_flags |= DCACHE_OP_COMPARE;
1314 if (op->d_revalidate)
1315 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1316 if (op->d_delete)
1317 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001318 if (op->d_prune)
1319 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001320
1321}
1322EXPORT_SYMBOL(d_set_d_op);
1323
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001324static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1325{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001326 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001327 if (inode) {
1328 if (unlikely(IS_AUTOMOUNT(inode)))
1329 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001330 list_add(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001331 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001332 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001333 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001334 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001335 fsnotify_d_instantiate(dentry, inode);
1336}
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338/**
1339 * d_instantiate - fill in inode information for a dentry
1340 * @entry: dentry to complete
1341 * @inode: inode to attach to this dentry
1342 *
1343 * Fill in inode information in the entry.
1344 *
1345 * This turns negative dentries into productive full members
1346 * of society.
1347 *
1348 * NOTE! This assumes that the inode count has been incremented
1349 * (or otherwise set) by the caller to indicate that it is now
1350 * in use by the dcache.
1351 */
1352
1353void d_instantiate(struct dentry *entry, struct inode * inode)
1354{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001355 BUG_ON(!list_empty(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001356 if (inode)
1357 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001358 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001359 if (inode)
1360 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 security_d_instantiate(entry, inode);
1362}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001363EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365/**
1366 * d_instantiate_unique - instantiate a non-aliased dentry
1367 * @entry: dentry to instantiate
1368 * @inode: inode to attach to this dentry
1369 *
1370 * Fill in inode information in the entry. On success, it returns NULL.
1371 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001372 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 *
1374 * Note that in order to avoid conflicts with rename() etc, the caller
1375 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001376 *
1377 * This also assumes that the inode count has been incremented
1378 * (or otherwise set) by the caller to indicate that it is now
1379 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 */
David Howells770bfad2006-08-22 20:06:07 -04001381static struct dentry *__d_instantiate_unique(struct dentry *entry,
1382 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
1384 struct dentry *alias;
1385 int len = entry->d_name.len;
1386 const char *name = entry->d_name.name;
1387 unsigned int hash = entry->d_name.hash;
1388
David Howells770bfad2006-08-22 20:06:07 -04001389 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001390 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001391 return NULL;
1392 }
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1395 struct qstr *qstr = &alias->d_name;
1396
Nick Piggin9abca362011-01-07 17:49:36 +11001397 /*
1398 * Don't need alias->d_lock here, because aliases with
1399 * d_parent == entry->d_parent are not subject to name or
1400 * parent changes, because the parent inode i_mutex is held.
1401 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 if (qstr->hash != hash)
1403 continue;
1404 if (alias->d_parent != entry->d_parent)
1405 continue;
Nick Piggin9d55c362011-01-07 17:50:09 +11001406 if (dentry_cmp(qstr->name, qstr->len, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001408 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 return alias;
1410 }
David Howells770bfad2006-08-22 20:06:07 -04001411
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001412 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return NULL;
1414}
David Howells770bfad2006-08-22 20:06:07 -04001415
1416struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1417{
1418 struct dentry *result;
1419
1420 BUG_ON(!list_empty(&entry->d_alias));
1421
Nick Piggin873feea2011-01-07 17:50:06 +11001422 if (inode)
1423 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001424 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001425 if (inode)
1426 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001427
1428 if (!result) {
1429 security_d_instantiate(entry, inode);
1430 return NULL;
1431 }
1432
1433 BUG_ON(!d_unhashed(result));
1434 iput(inode);
1435 return result;
1436}
1437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438EXPORT_SYMBOL(d_instantiate_unique);
1439
1440/**
1441 * d_alloc_root - allocate root dentry
1442 * @root_inode: inode to allocate the root for
1443 *
1444 * Allocate a root ("/") dentry for the inode given. The inode is
1445 * instantiated and returned. %NULL is returned if there is insufficient
1446 * memory or the inode passed is %NULL.
1447 */
1448
1449struct dentry * d_alloc_root(struct inode * root_inode)
1450{
1451 struct dentry *res = NULL;
1452
1453 if (root_inode) {
1454 static const struct qstr name = { .name = "/", .len = 1 };
1455
Al Viroa4464db2011-07-07 15:03:58 -04001456 res = __d_alloc(root_inode->i_sb, &name);
1457 if (res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 d_instantiate(res, root_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
1460 return res;
1461}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001462EXPORT_SYMBOL(d_alloc_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001464static struct dentry * __d_find_any_alias(struct inode *inode)
1465{
1466 struct dentry *alias;
1467
1468 if (list_empty(&inode->i_dentry))
1469 return NULL;
1470 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1471 __dget(alias);
1472 return alias;
1473}
1474
1475static struct dentry * d_find_any_alias(struct inode *inode)
1476{
1477 struct dentry *de;
1478
1479 spin_lock(&inode->i_lock);
1480 de = __d_find_any_alias(inode);
1481 spin_unlock(&inode->i_lock);
1482 return de;
1483}
1484
1485
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001487 * d_obtain_alias - find or allocate a dentry for a given inode
1488 * @inode: inode to allocate the dentry for
1489 *
1490 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1491 * similar open by handle operations. The returned dentry may be anonymous,
1492 * or may have a full name (if the inode was already in the cache).
1493 *
1494 * When called on a directory inode, we must ensure that the inode only ever
1495 * has one dentry. If a dentry is found, that is returned instead of
1496 * allocating a new one.
1497 *
1498 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001499 * to the dentry. In case of an error the reference on the inode is released.
1500 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1501 * be passed in and will be the error will be propagate to the return value,
1502 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001503 */
1504struct dentry *d_obtain_alias(struct inode *inode)
1505{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001506 static const struct qstr anonstring = { .name = "" };
1507 struct dentry *tmp;
1508 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001509
1510 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001511 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001512 if (IS_ERR(inode))
1513 return ERR_CAST(inode);
1514
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001515 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001516 if (res)
1517 goto out_iput;
1518
Al Viroa4464db2011-07-07 15:03:58 -04001519 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001520 if (!tmp) {
1521 res = ERR_PTR(-ENOMEM);
1522 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001523 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001524
Nick Piggin873feea2011-01-07 17:50:06 +11001525 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001526 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001527 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001528 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001529 dput(tmp);
1530 goto out_iput;
1531 }
1532
1533 /* attach a disconnected dentry */
1534 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001535 tmp->d_inode = inode;
1536 tmp->d_flags |= DCACHE_DISCONNECTED;
Christoph Hellwig9308a612008-08-11 15:49:12 +02001537 list_add(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001538 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001539 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001540 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001541 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001542 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001543 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001544
Christoph Hellwig9308a612008-08-11 15:49:12 +02001545 return tmp;
1546
1547 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001548 if (res && !IS_ERR(res))
1549 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001550 iput(inode);
1551 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001552}
Benny Halevyadc48722009-02-27 14:02:59 -08001553EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
1555/**
1556 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1557 * @inode: the inode which may have a disconnected dentry
1558 * @dentry: a negative dentry which we want to point to the inode.
1559 *
1560 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1561 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1562 * and return it, else simply d_add the inode to the dentry and return NULL.
1563 *
1564 * This is needed in the lookup routine of any filesystem that is exportable
1565 * (via knfsd) so that we can build dcache paths to directories effectively.
1566 *
1567 * If a dentry was found and moved, then it is returned. Otherwise NULL
1568 * is returned. This matches the expected return value of ->lookup.
1569 *
1570 */
1571struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1572{
1573 struct dentry *new = NULL;
1574
Al Viroa9049372011-07-08 21:20:11 -04001575 if (IS_ERR(inode))
1576 return ERR_CAST(inode);
1577
NeilBrown21c0d8f2006-10-04 02:16:16 -07001578 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001579 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 new = __d_find_alias(inode, 1);
1581 if (new) {
1582 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001583 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 d_move(new, dentry);
1586 iput(inode);
1587 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001588 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001589 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001590 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 security_d_instantiate(dentry, inode);
1592 d_rehash(dentry);
1593 }
1594 } else
1595 d_add(dentry, inode);
1596 return new;
1597}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001598EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
Barry Naujok94035402008-05-21 16:50:46 +10001600/**
1601 * d_add_ci - lookup or allocate new dentry with case-exact name
1602 * @inode: the inode case-insensitive lookup has found
1603 * @dentry: the negative dentry that was passed to the parent's lookup func
1604 * @name: the case-exact name to be associated with the returned dentry
1605 *
1606 * This is to avoid filling the dcache with case-insensitive names to the
1607 * same inode, only the actual correct case is stored in the dcache for
1608 * case-insensitive filesystems.
1609 *
1610 * For a case-insensitive lookup match and if the the case-exact dentry
1611 * already exists in in the dcache, use it and return it.
1612 *
1613 * If no entry exists with the exact case name, allocate new dentry with
1614 * the exact case, and return the spliced entry.
1615 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001616struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001617 struct qstr *name)
1618{
1619 int error;
1620 struct dentry *found;
1621 struct dentry *new;
1622
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001623 /*
1624 * First check if a dentry matching the name already exists,
1625 * if not go ahead and create it now.
1626 */
Barry Naujok94035402008-05-21 16:50:46 +10001627 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001628 if (!found) {
1629 new = d_alloc(dentry->d_parent, name);
1630 if (!new) {
1631 error = -ENOMEM;
1632 goto err_out;
1633 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001634
Barry Naujok94035402008-05-21 16:50:46 +10001635 found = d_splice_alias(inode, new);
1636 if (found) {
1637 dput(new);
1638 return found;
1639 }
1640 return new;
1641 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001642
1643 /*
1644 * If a matching dentry exists, and it's not negative use it.
1645 *
1646 * Decrement the reference count to balance the iget() done
1647 * earlier on.
1648 */
Barry Naujok94035402008-05-21 16:50:46 +10001649 if (found->d_inode) {
1650 if (unlikely(found->d_inode != inode)) {
1651 /* This can't happen because bad inodes are unhashed. */
1652 BUG_ON(!is_bad_inode(inode));
1653 BUG_ON(!is_bad_inode(found->d_inode));
1654 }
Barry Naujok94035402008-05-21 16:50:46 +10001655 iput(inode);
1656 return found;
1657 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001658
Barry Naujok94035402008-05-21 16:50:46 +10001659 /*
Josef Bacik44396f42011-05-31 11:58:49 -04001660 * We are going to instantiate this dentry, unhash it and clear the
1661 * lookup flag so we can do that.
1662 */
1663 if (unlikely(d_need_lookup(found)))
1664 d_clear_need_lookup(found);
1665
1666 /*
Barry Naujok94035402008-05-21 16:50:46 +10001667 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001668 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001669 */
Al Viro4513d892011-07-17 10:52:14 -04001670 new = d_splice_alias(inode, found);
1671 if (new) {
1672 dput(found);
1673 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001674 }
Al Viro4513d892011-07-17 10:52:14 -04001675 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001676
1677err_out:
1678 iput(inode);
1679 return ERR_PTR(error);
1680}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001681EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001684 * __d_lookup_rcu - search for a dentry (racy, store-free)
1685 * @parent: parent dentry
1686 * @name: qstr of name we wish to find
1687 * @seq: returns d_seq value at the point where the dentry was found
1688 * @inode: returns dentry->d_inode when the inode was found valid.
1689 * Returns: dentry, or NULL
1690 *
1691 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1692 * resolution (store-free path walking) design described in
1693 * Documentation/filesystems/path-lookup.txt.
1694 *
1695 * This is not to be used outside core vfs.
1696 *
1697 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1698 * held, and rcu_read_lock held. The returned dentry must not be stored into
1699 * without taking d_lock and checking d_seq sequence count against @seq
1700 * returned here.
1701 *
1702 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1703 * function.
1704 *
1705 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1706 * the returned dentry, so long as its parent's seqlock is checked after the
1707 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1708 * is formed, giving integrity down the path walk.
1709 */
1710struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1711 unsigned *seq, struct inode **inode)
1712{
1713 unsigned int len = name->len;
1714 unsigned int hash = name->hash;
1715 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001716 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001717 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001718 struct dentry *dentry;
1719
1720 /*
1721 * Note: There is significant duplication with __d_lookup_rcu which is
1722 * required to prevent single threaded performance regressions
1723 * especially on architectures where smp_rmb (in seqcounts) are costly.
1724 * Keep the two functions in sync.
1725 */
1726
1727 /*
1728 * The hash list is protected using RCU.
1729 *
1730 * Carefully use d_seq when comparing a candidate dentry, to avoid
1731 * races with d_move().
1732 *
1733 * It is possible that concurrent renames can mess up our list
1734 * walk here and result in missing our dentry, resulting in the
1735 * false-negative result. d_lookup() protects against concurrent
1736 * renames using rename_lock seqlock.
1737 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001738 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11001739 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001740 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001741 struct inode *i;
1742 const char *tname;
1743 int tlen;
1744
1745 if (dentry->d_name.hash != hash)
1746 continue;
1747
1748seqretry:
1749 *seq = read_seqcount_begin(&dentry->d_seq);
1750 if (dentry->d_parent != parent)
1751 continue;
1752 if (d_unhashed(dentry))
1753 continue;
1754 tlen = dentry->d_name.len;
1755 tname = dentry->d_name.name;
1756 i = dentry->d_inode;
Nick Piggine1bb5782011-01-07 17:50:08 +11001757 prefetch(tname);
Nick Piggin31e6b012011-01-07 17:49:52 +11001758 /*
1759 * This seqcount check is required to ensure name and
1760 * len are loaded atomically, so as not to walk off the
1761 * edge of memory when walking. If we could load this
1762 * atomically some other way, we could drop this check.
1763 */
1764 if (read_seqcount_retry(&dentry->d_seq, *seq))
1765 goto seqretry;
Linus Torvalds830c0f02011-08-06 22:41:50 -07001766 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001767 if (parent->d_op->d_compare(parent, *inode,
1768 dentry, i,
1769 tlen, tname, name))
1770 continue;
1771 } else {
Nick Piggin9d55c362011-01-07 17:50:09 +11001772 if (dentry_cmp(tname, tlen, str, len))
Nick Piggin31e6b012011-01-07 17:49:52 +11001773 continue;
1774 }
1775 /*
1776 * No extra seqcount check is required after the name
1777 * compare. The caller must perform a seqcount check in
1778 * order to do anything useful with the returned dentry
1779 * anyway.
1780 */
1781 *inode = i;
1782 return dentry;
1783 }
1784 return NULL;
1785}
1786
1787/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 * d_lookup - search for a dentry
1789 * @parent: parent dentry
1790 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001791 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001793 * d_lookup searches the children of the parent dentry for the name in
1794 * question. If the dentry is found its reference count is incremented and the
1795 * dentry is returned. The caller must use dput to free the entry when it has
1796 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001798struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799{
Nick Piggin31e6b012011-01-07 17:49:52 +11001800 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001801 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803 do {
1804 seq = read_seqbegin(&rename_lock);
1805 dentry = __d_lookup(parent, name);
1806 if (dentry)
1807 break;
1808 } while (read_seqretry(&rename_lock, seq));
1809 return dentry;
1810}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001811EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Nick Piggin31e6b012011-01-07 17:49:52 +11001813/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001814 * __d_lookup - search for a dentry (racy)
1815 * @parent: parent dentry
1816 * @name: qstr of name we wish to find
1817 * Returns: dentry, or NULL
1818 *
1819 * __d_lookup is like d_lookup, however it may (rarely) return a
1820 * false-negative result due to unrelated rename activity.
1821 *
1822 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1823 * however it must be used carefully, eg. with a following d_lookup in
1824 * the case of failure.
1825 *
1826 * __d_lookup callers must be commented.
1827 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001828struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829{
1830 unsigned int len = name->len;
1831 unsigned int hash = name->hash;
1832 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001833 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001834 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001835 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001836 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Nick Pigginb04f7842010-08-18 04:37:34 +10001838 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001839 * Note: There is significant duplication with __d_lookup_rcu which is
1840 * required to prevent single threaded performance regressions
1841 * especially on architectures where smp_rmb (in seqcounts) are costly.
1842 * Keep the two functions in sync.
1843 */
1844
1845 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001846 * The hash list is protected using RCU.
1847 *
1848 * Take d_lock when comparing a candidate dentry, to avoid races
1849 * with d_move().
1850 *
1851 * It is possible that concurrent renames can mess up our list
1852 * walk here and result in missing our dentry, resulting in the
1853 * false-negative result. d_lookup() protects against concurrent
1854 * renames using rename_lock seqlock.
1855 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001856 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001857 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 rcu_read_lock();
1859
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001860 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Nick Piggin31e6b012011-01-07 17:49:52 +11001861 const char *tname;
1862 int tlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 if (dentry->d_name.hash != hash)
1865 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866
1867 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 if (dentry->d_parent != parent)
1869 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001870 if (d_unhashed(dentry))
1871 goto next;
1872
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 /*
1874 * It is safe to compare names since d_move() cannot
1875 * change the qstr (protected by d_lock).
1876 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001877 tlen = dentry->d_name.len;
1878 tname = dentry->d_name.name;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001879 if (parent->d_flags & DCACHE_OP_COMPARE) {
Nick Piggin621e1552011-01-07 17:49:27 +11001880 if (parent->d_op->d_compare(parent, parent->d_inode,
1881 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001882 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 goto next;
1884 } else {
Nick Piggin9d55c362011-01-07 17:50:09 +11001885 if (dentry_cmp(tname, tlen, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 goto next;
1887 }
1888
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001889 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001890 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 spin_unlock(&dentry->d_lock);
1892 break;
1893next:
1894 spin_unlock(&dentry->d_lock);
1895 }
1896 rcu_read_unlock();
1897
1898 return found;
1899}
1900
1901/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001902 * d_hash_and_lookup - hash the qstr then search for a dentry
1903 * @dir: Directory to search in
1904 * @name: qstr of name we wish to find
1905 *
1906 * On hash failure or on lookup failure NULL is returned.
1907 */
1908struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1909{
1910 struct dentry *dentry = NULL;
1911
1912 /*
1913 * Check for a fs-specific hash function. Note that we must
1914 * calculate the standard hash first, as the d_op->d_hash()
1915 * routine may choose to leave the hash value unchanged.
1916 */
1917 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11001918 if (dir->d_flags & DCACHE_OP_HASH) {
Nick Pigginb1e6a012011-01-07 17:49:28 +11001919 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001920 goto out;
1921 }
1922 dentry = d_lookup(dir, name);
1923out:
1924 return dentry;
1925}
1926
1927/**
Nick Piggin786a5e12011-01-07 17:49:16 +11001928 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08001930 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 *
1932 * An insecure source has sent us a dentry, here we verify it and dget() it.
1933 * This is used by ncpfs in its readdir implementation.
1934 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11001935 *
1936 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 */
Nick Piggind3a23e12011-01-05 20:01:21 +11001938int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939{
Nick Piggin786a5e12011-01-07 17:49:16 +11001940 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11001941
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001942 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11001943 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1944 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001945 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11001946 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001947 spin_unlock(&dentry->d_lock);
1948 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 return 1;
1950 }
1951 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001952 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11001953
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return 0;
1955}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001956EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958/*
1959 * When a file is deleted, we have two options:
1960 * - turn this dentry into a negative dentry
1961 * - unhash this dentry and free it.
1962 *
1963 * Usually, we want to just turn this into
1964 * a negative dentry, but if anybody else is
1965 * currently using the dentry or the inode
1966 * we can't do that and we fall back on removing
1967 * it from the hash queues and waiting for
1968 * it to be deleted later when it has no users
1969 */
1970
1971/**
1972 * d_delete - delete a dentry
1973 * @dentry: The dentry to delete
1974 *
1975 * Turn the dentry into a negative dentry if possible, otherwise
1976 * remove it from the hash queues so it can be deleted later
1977 */
1978
1979void d_delete(struct dentry * dentry)
1980{
Nick Piggin873feea2011-01-07 17:50:06 +11001981 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04001982 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 /*
1984 * Are we the only user?
1985 */
Nick Piggin357f8e62011-01-07 17:49:42 +11001986again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001988 inode = dentry->d_inode;
1989 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001990 if (dentry->d_count == 1) {
Nick Piggin873feea2011-01-07 17:50:06 +11001991 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11001992 spin_unlock(&dentry->d_lock);
1993 cpu_relax();
1994 goto again;
1995 }
Al Viro13e3c5e2010-05-21 16:11:04 -04001996 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11001997 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04001998 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 return;
2000 }
2001
2002 if (!d_unhashed(dentry))
2003 __d_drop(dentry);
2004
2005 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002006
2007 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002009EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002011static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002013 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002014 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002015 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002016 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002017 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018}
2019
David Howells770bfad2006-08-22 20:06:07 -04002020static void _d_rehash(struct dentry * entry)
2021{
2022 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2023}
2024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025/**
2026 * d_rehash - add an entry back to the hash
2027 * @entry: dentry to add to the hash
2028 *
2029 * Adds a dentry to the hash according to its name.
2030 */
2031
2032void d_rehash(struct dentry * entry)
2033{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002035 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002038EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002040/**
2041 * dentry_update_name_case - update case insensitive dentry with a new name
2042 * @dentry: dentry to be updated
2043 * @name: new name
2044 *
2045 * Update a case insensitive dentry with new case of name.
2046 *
2047 * dentry must have been returned by d_lookup with name @name. Old and new
2048 * name lengths must match (ie. no d_compare which allows mismatched name
2049 * lengths).
2050 *
2051 * Parent inode i_mutex must be held over d_lookup and into this call (to
2052 * keep renames and concurrent inserts, and readdir(2) away).
2053 */
2054void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2055{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002056 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002057 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2058
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002059 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002060 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002061 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002062 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002063 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002064}
2065EXPORT_SYMBOL(dentry_update_name_case);
2066
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067static void switch_names(struct dentry *dentry, struct dentry *target)
2068{
2069 if (dname_external(target)) {
2070 if (dname_external(dentry)) {
2071 /*
2072 * Both external: swap the pointers
2073 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002074 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 } else {
2076 /*
2077 * dentry:internal, target:external. Steal target's
2078 * storage and make target internal.
2079 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002080 memcpy(target->d_iname, dentry->d_name.name,
2081 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 dentry->d_name.name = target->d_name.name;
2083 target->d_name.name = target->d_iname;
2084 }
2085 } else {
2086 if (dname_external(dentry)) {
2087 /*
2088 * dentry:external, target:internal. Give dentry's
2089 * storage to target and make dentry internal
2090 */
2091 memcpy(dentry->d_iname, target->d_name.name,
2092 target->d_name.len + 1);
2093 target->d_name.name = dentry->d_name.name;
2094 dentry->d_name.name = dentry->d_iname;
2095 } else {
2096 /*
2097 * Both are internal. Just copy target to dentry
2098 */
2099 memcpy(dentry->d_iname, target->d_name.name,
2100 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002101 dentry->d_name.len = target->d_name.len;
2102 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 }
2104 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002105 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
2107
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002108static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2109{
2110 /*
2111 * XXXX: do we really need to take target->d_lock?
2112 */
2113 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2114 spin_lock(&target->d_parent->d_lock);
2115 else {
2116 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2117 spin_lock(&dentry->d_parent->d_lock);
2118 spin_lock_nested(&target->d_parent->d_lock,
2119 DENTRY_D_LOCK_NESTED);
2120 } else {
2121 spin_lock(&target->d_parent->d_lock);
2122 spin_lock_nested(&dentry->d_parent->d_lock,
2123 DENTRY_D_LOCK_NESTED);
2124 }
2125 }
2126 if (target < dentry) {
2127 spin_lock_nested(&target->d_lock, 2);
2128 spin_lock_nested(&dentry->d_lock, 3);
2129 } else {
2130 spin_lock_nested(&dentry->d_lock, 2);
2131 spin_lock_nested(&target->d_lock, 3);
2132 }
2133}
2134
2135static void dentry_unlock_parents_for_move(struct dentry *dentry,
2136 struct dentry *target)
2137{
2138 if (target->d_parent != dentry->d_parent)
2139 spin_unlock(&dentry->d_parent->d_lock);
2140 if (target->d_parent != target)
2141 spin_unlock(&target->d_parent->d_lock);
2142}
2143
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002145 * When switching names, the actual string doesn't strictly have to
2146 * be preserved in the target - because we're dropping the target
2147 * anyway. As such, we can just do a simple memcpy() to copy over
2148 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002150 * Note that we have to be a lot more careful about getting the hash
2151 * switched - we have to switch the hash value properly even if it
2152 * then no longer matches the actual (corrupted) string of the target.
2153 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002155/*
Al Viro18367502011-07-12 21:42:24 -04002156 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 * @dentry: entry to move
2158 * @target: new dentry
2159 *
2160 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002161 * dcache entries should not be moved in this way. Caller must hold
2162 * rename_lock, the i_mutex of the source and target directories,
2163 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 */
Al Viro18367502011-07-12 21:42:24 -04002165static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 if (!dentry->d_inode)
2168 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2169
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002170 BUG_ON(d_ancestor(dentry, target));
2171 BUG_ON(d_ancestor(target, dentry));
2172
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002173 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Nick Piggin31e6b012011-01-07 17:49:52 +11002175 write_seqcount_begin(&dentry->d_seq);
2176 write_seqcount_begin(&target->d_seq);
2177
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002178 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2179
2180 /*
2181 * Move the dentry to the target hash queue. Don't bother checking
2182 * for the same hash queue because of how unlikely it is.
2183 */
2184 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002185 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186
2187 /* Unhash the target: dput() will then get rid of it */
2188 __d_drop(target);
2189
Eric Dumazet5160ee62006-01-08 01:03:32 -08002190 list_del(&dentry->d_u.d_child);
2191 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192
2193 /* Switch the names.. */
2194 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002195 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196
2197 /* ... and switch the parents */
2198 if (IS_ROOT(dentry)) {
2199 dentry->d_parent = target->d_parent;
2200 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002201 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002203 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204
2205 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002206 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 }
2208
Eric Dumazet5160ee62006-01-08 01:03:32 -08002209 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002210
Nick Piggin31e6b012011-01-07 17:49:52 +11002211 write_seqcount_end(&target->d_seq);
2212 write_seqcount_end(&dentry->d_seq);
2213
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002214 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002216 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002218}
2219
2220/*
2221 * d_move - move a dentry
2222 * @dentry: entry to move
2223 * @target: new dentry
2224 *
2225 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002226 * dcache entries should not be moved in this way. See the locking
2227 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002228 */
2229void d_move(struct dentry *dentry, struct dentry *target)
2230{
2231 write_seqlock(&rename_lock);
2232 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002234}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002235EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002237/**
2238 * d_ancestor - search for an ancestor
2239 * @p1: ancestor dentry
2240 * @p2: child dentry
2241 *
2242 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2243 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002244 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002245struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002246{
2247 struct dentry *p;
2248
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002249 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002250 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002251 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002252 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002253 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002254}
2255
2256/*
2257 * This helper attempts to cope with remotely renamed directories
2258 *
2259 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002260 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002261 *
2262 * Note: If ever the locking in lock_rename() changes, then please
2263 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002264 */
Nick Piggin873feea2011-01-07 17:50:06 +11002265static struct dentry *__d_unalias(struct inode *inode,
2266 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002267{
2268 struct mutex *m1 = NULL, *m2 = NULL;
2269 struct dentry *ret;
2270
2271 /* If alias and dentry share a parent, then no extra locks required */
2272 if (alias->d_parent == dentry->d_parent)
2273 goto out_unalias;
2274
Trond Myklebust9eaef272006-10-21 10:24:20 -07002275 /* See lock_rename() */
2276 ret = ERR_PTR(-EBUSY);
2277 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2278 goto out_err;
2279 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2280 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2281 goto out_err;
2282 m2 = &alias->d_parent->d_inode->i_mutex;
2283out_unalias:
Al Viro18367502011-07-12 21:42:24 -04002284 __d_move(alias, dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002285 ret = alias;
2286out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002287 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002288 if (m2)
2289 mutex_unlock(m2);
2290 if (m1)
2291 mutex_unlock(m1);
2292 return ret;
2293}
2294
2295/*
David Howells770bfad2006-08-22 20:06:07 -04002296 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2297 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002298 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002299 */
2300static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2301{
2302 struct dentry *dparent, *aparent;
2303
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002304 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002305
Nick Piggin31e6b012011-01-07 17:49:52 +11002306 write_seqcount_begin(&dentry->d_seq);
2307 write_seqcount_begin(&anon->d_seq);
2308
David Howells770bfad2006-08-22 20:06:07 -04002309 dparent = dentry->d_parent;
2310 aparent = anon->d_parent;
2311
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002312 switch_names(dentry, anon);
2313 swap(dentry->d_name.hash, anon->d_name.hash);
2314
David Howells770bfad2006-08-22 20:06:07 -04002315 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2316 list_del(&dentry->d_u.d_child);
2317 if (!IS_ROOT(dentry))
2318 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2319 else
2320 INIT_LIST_HEAD(&dentry->d_u.d_child);
2321
2322 anon->d_parent = (dparent == dentry) ? anon : dparent;
2323 list_del(&anon->d_u.d_child);
2324 if (!IS_ROOT(anon))
2325 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2326 else
2327 INIT_LIST_HEAD(&anon->d_u.d_child);
2328
Nick Piggin31e6b012011-01-07 17:49:52 +11002329 write_seqcount_end(&dentry->d_seq);
2330 write_seqcount_end(&anon->d_seq);
2331
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002332 dentry_unlock_parents_for_move(anon, dentry);
2333 spin_unlock(&dentry->d_lock);
2334
2335 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002336 anon->d_flags &= ~DCACHE_DISCONNECTED;
2337}
2338
2339/**
2340 * d_materialise_unique - introduce an inode into the tree
2341 * @dentry: candidate dentry
2342 * @inode: inode to bind to the dentry, to which aliases may be attached
2343 *
2344 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002345 * root directory alias in its place if there is one. Caller must hold the
2346 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002347 */
2348struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2349{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002350 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002351
2352 BUG_ON(!d_unhashed(dentry));
2353
David Howells770bfad2006-08-22 20:06:07 -04002354 if (!inode) {
2355 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002356 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002357 d_rehash(actual);
2358 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002359 }
2360
Nick Piggin873feea2011-01-07 17:50:06 +11002361 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002362
Trond Myklebust9eaef272006-10-21 10:24:20 -07002363 if (S_ISDIR(inode->i_mode)) {
2364 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002365
Trond Myklebust9eaef272006-10-21 10:24:20 -07002366 /* Does an aliased dentry already exist? */
2367 alias = __d_find_alias(inode, 0);
2368 if (alias) {
2369 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002370 write_seqlock(&rename_lock);
2371
2372 if (d_ancestor(alias, dentry)) {
2373 /* Check for loops */
2374 actual = ERR_PTR(-ELOOP);
2375 } else if (IS_ROOT(alias)) {
2376 /* Is this an anonymous mountpoint that we
2377 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002378 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002379 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002380 __d_drop(alias);
2381 goto found;
Al Viro18367502011-07-12 21:42:24 -04002382 } else {
2383 /* Nope, but we must(!) avoid directory
2384 * aliasing */
2385 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002386 }
Al Viro18367502011-07-12 21:42:24 -04002387 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002388 if (IS_ERR(actual)) {
2389 if (PTR_ERR(actual) == -ELOOP)
2390 pr_warn_ratelimited(
2391 "VFS: Lookup of '%s' in %s %s"
2392 " would have caused loop\n",
2393 dentry->d_name.name,
2394 inode->i_sb->s_type->name,
2395 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002396 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002397 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002398 goto out_nolock;
2399 }
David Howells770bfad2006-08-22 20:06:07 -04002400 }
2401
2402 /* Add a unique reference */
2403 actual = __d_instantiate_unique(dentry, inode);
2404 if (!actual)
2405 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002406 else
2407 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002408
David Howells770bfad2006-08-22 20:06:07 -04002409 spin_lock(&actual->d_lock);
2410found:
2411 _d_rehash(actual);
2412 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002413 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002414out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002415 if (actual == dentry) {
2416 security_d_instantiate(dentry, inode);
2417 return NULL;
2418 }
2419
2420 iput(inode);
2421 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002422}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002423EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002424
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002425static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002426{
2427 *buflen -= namelen;
2428 if (*buflen < 0)
2429 return -ENAMETOOLONG;
2430 *buffer -= namelen;
2431 memcpy(*buffer, str, namelen);
2432 return 0;
2433}
2434
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002435static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2436{
2437 return prepend(buffer, buflen, name->name, name->len);
2438}
2439
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002441 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002442 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002443 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002444 * @buffer: pointer to the end of the buffer
2445 * @buflen: pointer to buffer length
2446 *
Nick Piggin949854d2011-01-07 17:49:37 +11002447 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002448 */
Al Viro02125a82011-12-05 08:43:34 -05002449static int prepend_path(const struct path *path,
2450 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002451 char **buffer, int *buflen)
2452{
2453 struct dentry *dentry = path->dentry;
2454 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002455 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002456 bool slash = false;
2457 int error = 0;
2458
Nick Piggin99b7db72010-08-18 04:37:39 +10002459 br_read_lock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002460 while (dentry != root->dentry || vfsmnt != root->mnt) {
2461 struct dentry * parent;
2462
2463 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2464 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002465 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002466 goto global_root;
Al Viro0714a532011-11-24 22:19:58 -05002467 dentry = mnt->mnt.mnt_mountpoint;
2468 mnt = mnt->mnt_parent;
2469 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002470 continue;
2471 }
2472 parent = dentry->d_parent;
2473 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002474 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002475 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002476 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002477 if (!error)
2478 error = prepend(buffer, buflen, "/", 1);
2479 if (error)
2480 break;
2481
2482 slash = true;
2483 dentry = parent;
2484 }
2485
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002486 if (!error && !slash)
2487 error = prepend(buffer, buflen, "/", 1);
2488
Al Viro02125a82011-12-05 08:43:34 -05002489out:
Nick Piggin99b7db72010-08-18 04:37:39 +10002490 br_read_unlock(vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002491 return error;
2492
2493global_root:
2494 /*
2495 * Filesystems needing to implement special "root names"
2496 * should do so with ->d_dname()
2497 */
2498 if (IS_ROOT(dentry) &&
2499 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2500 WARN(1, "Root dentry has weird name <%.*s>\n",
2501 (int) dentry->d_name.len, dentry->d_name.name);
2502 }
Al Viro02125a82011-12-05 08:43:34 -05002503 if (!slash)
2504 error = prepend(buffer, buflen, "/", 1);
2505 if (!error)
2506 error = vfsmnt->mnt_ns ? 1 : 2;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002507 goto out;
2508}
2509
2510/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002511 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002512 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002513 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002514 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 * @buflen: buffer length
2516 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002517 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002519 * Returns a pointer into the buffer or an error code if the
2520 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002521 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002522 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002523 *
Al Viro02125a82011-12-05 08:43:34 -05002524 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 */
Al Viro02125a82011-12-05 08:43:34 -05002526char *__d_path(const struct path *path,
2527 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002528 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002530 char *res = buf + buflen;
2531 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002533 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002534 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002535 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002536 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002537
Al Viro02125a82011-12-05 08:43:34 -05002538 if (error < 0)
2539 return ERR_PTR(error);
2540 if (error > 0)
2541 return NULL;
2542 return res;
2543}
2544
2545char *d_absolute_path(const struct path *path,
2546 char *buf, int buflen)
2547{
2548 struct path root = {};
2549 char *res = buf + buflen;
2550 int error;
2551
2552 prepend(&res, &buflen, "\0", 1);
2553 write_seqlock(&rename_lock);
2554 error = prepend_path(path, &root, &res, &buflen);
2555 write_sequnlock(&rename_lock);
2556
2557 if (error > 1)
2558 error = -EINVAL;
2559 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002560 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002561 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562}
2563
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002564/*
2565 * same as __d_path but appends "(deleted)" for unlinked files.
2566 */
Al Viro02125a82011-12-05 08:43:34 -05002567static int path_with_deleted(const struct path *path,
2568 const struct path *root,
2569 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002570{
2571 prepend(buf, buflen, "\0", 1);
2572 if (d_unlinked(path->dentry)) {
2573 int error = prepend(buf, buflen, " (deleted)", 10);
2574 if (error)
2575 return error;
2576 }
2577
2578 return prepend_path(path, root, buf, buflen);
2579}
2580
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002581static int prepend_unreachable(char **buffer, int *buflen)
2582{
2583 return prepend(buffer, buflen, "(unreachable)", 13);
2584}
2585
Jan Bluncka03a8a702008-02-14 19:38:32 -08002586/**
2587 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002588 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002589 * @buf: buffer to return value in
2590 * @buflen: buffer length
2591 *
2592 * Convert a dentry into an ASCII path name. If the entry has been deleted
2593 * the string " (deleted)" is appended. Note that this is ambiguous.
2594 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002595 * Returns a pointer into the buffer or an error code if the path was
2596 * too long. Note: Callers should use the returned pointer, not the passed
2597 * in buffer, to use the name! The implementation often starts at an offset
2598 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002599 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002600 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002601 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002602char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002604 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002605 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002606 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002608 /*
2609 * We have various synthetic filesystems that never get mounted. On
2610 * these filesystems dentries are never used for lookup purposes, and
2611 * thus don't need to be hashed. They also don't need a name until a
2612 * user wants to identify the object in /proc/pid/fd/. The little hack
2613 * below allows us to generate a name for these objects on demand:
2614 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002615 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2616 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002617
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002618 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002619 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002620 error = path_with_deleted(path, &root, &res, &buflen);
2621 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002622 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002623 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002624 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 return res;
2626}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002627EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002629/**
2630 * d_path_with_unreachable - return the path of a dentry
2631 * @path: path to report
2632 * @buf: buffer to return value in
2633 * @buflen: buffer length
2634 *
2635 * The difference from d_path() is that this prepends "(unreachable)"
2636 * to paths which are unreachable from the current process' root.
2637 */
2638char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2639{
2640 char *res = buf + buflen;
2641 struct path root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002642 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);
Al Viro02125a82011-12-05 08:43:34 -05002649 error = path_with_deleted(path, &root, &res, &buflen);
2650 if (error > 0)
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002651 error = prepend_unreachable(&res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002652 write_sequnlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002653 path_put(&root);
2654 if (error)
2655 res = ERR_PTR(error);
2656
2657 return res;
2658}
2659
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002661 * Helper function for dentry_operations.d_dname() members
2662 */
2663char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2664 const char *fmt, ...)
2665{
2666 va_list args;
2667 char temp[64];
2668 int sz;
2669
2670 va_start(args, fmt);
2671 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2672 va_end(args);
2673
2674 if (sz > sizeof(temp) || sz > buflen)
2675 return ERR_PTR(-ENAMETOOLONG);
2676
2677 buffer += buflen - sz;
2678 return memcpy(buffer, temp, sz);
2679}
2680
2681/*
Ram Pai6092d042008-03-27 13:06:20 +01002682 * Write full pathname from the root of the filesystem into the buffer.
2683 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002684static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002685{
2686 char *end = buf + buflen;
2687 char *retval;
2688
Ram Pai6092d042008-03-27 13:06:20 +01002689 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002690 if (buflen < 1)
2691 goto Elong;
2692 /* Get '/' right */
2693 retval = end-1;
2694 *retval = '/';
2695
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002696 while (!IS_ROOT(dentry)) {
2697 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002698 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002699
Ram Pai6092d042008-03-27 13:06:20 +01002700 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002701 spin_lock(&dentry->d_lock);
2702 error = prepend_name(&end, &buflen, &dentry->d_name);
2703 spin_unlock(&dentry->d_lock);
2704 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002705 goto Elong;
2706
2707 retval = end;
2708 dentry = parent;
2709 }
Al Viroc1031352010-06-06 22:31:14 -04002710 return retval;
2711Elong:
2712 return ERR_PTR(-ENAMETOOLONG);
2713}
Nick Pigginec2447c2011-01-07 17:49:29 +11002714
2715char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2716{
2717 char *retval;
2718
Nick Piggin949854d2011-01-07 17:49:37 +11002719 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002720 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002721 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002722
2723 return retval;
2724}
2725EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002726
2727char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2728{
2729 char *p = NULL;
2730 char *retval;
2731
Nick Piggin949854d2011-01-07 17:49:37 +11002732 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002733 if (d_unlinked(dentry)) {
2734 p = buf + buflen;
2735 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2736 goto Elong;
2737 buflen++;
2738 }
2739 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002740 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002741 if (!IS_ERR(retval) && p)
2742 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002743 return retval;
2744Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002745 return ERR_PTR(-ENAMETOOLONG);
2746}
2747
2748/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749 * NOTE! The user-level library version returns a
2750 * character pointer. The kernel system call just
2751 * returns the length of the buffer filled (which
2752 * includes the ending '\0' character), or a negative
2753 * error value. So libc would do something like
2754 *
2755 * char *getcwd(char * buf, size_t size)
2756 * {
2757 * int retval;
2758 *
2759 * retval = sys_getcwd(buf, size);
2760 * if (retval >= 0)
2761 * return buf;
2762 * errno = -retval;
2763 * return NULL;
2764 * }
2765 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002766SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767{
Linus Torvalds552ce542007-02-13 12:08:18 -08002768 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002769 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002770 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771
2772 if (!page)
2773 return -ENOMEM;
2774
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002775 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Linus Torvalds552ce542007-02-13 12:08:18 -08002777 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002778 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002779 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002780 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002781 char *cwd = page + PAGE_SIZE;
2782 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002784 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002785 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002786 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002787
Al Viro02125a82011-12-05 08:43:34 -05002788 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002789 goto out;
2790
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002791 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002792 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002793 error = prepend_unreachable(&cwd, &buflen);
2794 if (error)
2795 goto out;
2796 }
2797
Linus Torvalds552ce542007-02-13 12:08:18 -08002798 error = -ERANGE;
2799 len = PAGE_SIZE + page - cwd;
2800 if (len <= size) {
2801 error = len;
2802 if (copy_to_user(buf, cwd, len))
2803 error = -EFAULT;
2804 }
Nick Piggin949854d2011-01-07 17:49:37 +11002805 } else {
2806 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808
2809out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002810 path_put(&pwd);
2811 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 free_page((unsigned long) page);
2813 return error;
2814}
2815
2816/*
2817 * Test whether new_dentry is a subdirectory of old_dentry.
2818 *
2819 * Trivially implemented using the dcache structure
2820 */
2821
2822/**
2823 * is_subdir - is new dentry a subdirectory of old_dentry
2824 * @new_dentry: new dentry
2825 * @old_dentry: old dentry
2826 *
2827 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2828 * Returns 0 otherwise.
2829 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2830 */
2831
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002832int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833{
2834 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002835 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002837 if (new_dentry == old_dentry)
2838 return 1;
2839
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002840 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002843 /*
2844 * Need rcu_readlock to protect against the d_parent trashing
2845 * due to d_move
2846 */
2847 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002848 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002850 else
2851 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002852 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
2855 return result;
2856}
2857
2858void d_genocide(struct dentry *root)
2859{
Nick Piggin949854d2011-01-07 17:49:37 +11002860 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002862 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002863 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
Nick Piggin949854d2011-01-07 17:49:37 +11002865 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002866again:
2867 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002868 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869repeat:
2870 next = this_parent->d_subdirs.next;
2871resume:
2872 while (next != &this_parent->d_subdirs) {
2873 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002874 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002876
Nick Pigginda502952011-01-07 17:49:33 +11002877 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2878 if (d_unhashed(dentry) || !dentry->d_inode) {
2879 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002883 spin_unlock(&this_parent->d_lock);
2884 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002886 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 goto repeat;
2888 }
Nick Piggin949854d2011-01-07 17:49:37 +11002889 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2890 dentry->d_flags |= DCACHE_GENOCIDE;
2891 dentry->d_count--;
2892 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002893 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 }
2895 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002896 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11002897 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2898 this_parent->d_flags |= DCACHE_GENOCIDE;
2899 this_parent->d_count--;
2900 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002901 this_parent = try_to_ascend(this_parent, locked, seq);
2902 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11002903 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11002904 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 goto resume;
2906 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002907 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002908 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11002909 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11002910 if (locked)
2911 write_sequnlock(&rename_lock);
2912 return;
2913
2914rename_retry:
2915 locked = 1;
2916 write_seqlock(&rename_lock);
2917 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918}
2919
2920/**
2921 * find_inode_number - check for dentry with name
2922 * @dir: directory to check
2923 * @name: Name to find.
2924 *
2925 * Check whether a dentry already exists for the given name,
2926 * and return the inode number if it has an inode. Otherwise
2927 * 0 is returned.
2928 *
2929 * This routine is used to post-process directory listings for
2930 * filesystems using synthetic inode numbers, and is necessary
2931 * to keep getcwd() working.
2932 */
2933
2934ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2935{
2936 struct dentry * dentry;
2937 ino_t ino = 0;
2938
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002939 dentry = d_hash_and_lookup(dir, name);
2940 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 if (dentry->d_inode)
2942 ino = dentry->d_inode->i_ino;
2943 dput(dentry);
2944 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 return ino;
2946}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002947EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948
2949static __initdata unsigned long dhash_entries;
2950static int __init set_dhash_entries(char *str)
2951{
2952 if (!str)
2953 return 0;
2954 dhash_entries = simple_strtoul(str, &str, 0);
2955 return 1;
2956}
2957__setup("dhash_entries=", set_dhash_entries);
2958
2959static void __init dcache_init_early(void)
2960{
2961 int loop;
2962
2963 /* If hashes are distributed across NUMA nodes, defer
2964 * hash allocation until vmalloc space is available.
2965 */
2966 if (hashdist)
2967 return;
2968
2969 dentry_hashtable =
2970 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002971 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 dhash_entries,
2973 13,
2974 HASH_EARLY,
2975 &d_hash_shift,
2976 &d_hash_mask,
2977 0);
2978
2979 for (loop = 0; loop < (1 << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002980 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981}
2982
Denis Cheng74bf17c2007-10-16 23:26:30 -07002983static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984{
2985 int loop;
2986
2987 /*
2988 * A constructor could be added for stable state like the lists,
2989 * but it is probably not worth it because of the cache nature
2990 * of the dcache.
2991 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07002992 dentry_cache = KMEM_CACHE(dentry,
2993 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994
2995 /* Hash may have been set up in dcache_init_early */
2996 if (!hashdist)
2997 return;
2998
2999 dentry_hashtable =
3000 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003001 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 dhash_entries,
3003 13,
3004 0,
3005 &d_hash_shift,
3006 &d_hash_mask,
3007 0);
3008
3009 for (loop = 0; loop < (1 << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003010 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011}
3012
3013/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003014struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003015EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017EXPORT_SYMBOL(d_genocide);
3018
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019void __init vfs_caches_init_early(void)
3020{
3021 dcache_init_early();
3022 inode_init_early();
3023}
3024
3025void __init vfs_caches_init(unsigned long mempages)
3026{
3027 unsigned long reserve;
3028
3029 /* Base hash sizes on available memory, with a reserve equal to
3030 150% of current kernel size */
3031
3032 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3033 mempages -= reserve;
3034
3035 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003036 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037
Denis Cheng74bf17c2007-10-16 23:26:30 -07003038 dcache_init();
3039 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003041 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 bdev_cache_init();
3043 chrdev_init();
3044}