blob: 40469044088def2607d19540ef0cb8dcce78ac91 [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>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#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 Torvalds8966be92012-03-02 14:23:30 -0800107static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100109{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 hash = hash + (hash >> D_HASHBITS);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100112 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
Linus Torvalds5483f182012-03-04 15:51:42 -0800140/*
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
143 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700144#ifdef CONFIG_DCACHE_WORD_ACCESS
145
146#include <asm/word-at-a-time.h>
147/*
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
152 *
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
155 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700156static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800157{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800158 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800159
160 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700161 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700162 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800163 if (tcount < sizeof(unsigned long))
164 break;
165 if (unlikely(a != b))
166 return 1;
167 cs += sizeof(unsigned long);
168 ct += sizeof(unsigned long);
169 tcount -= sizeof(unsigned long);
170 if (!tcount)
171 return 0;
172 }
173 mask = ~(~0ul << tcount*8);
174 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700175}
176
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800177#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700178
Linus Torvalds94753db52012-05-10 12:19:19 -0700179static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700180{
Linus Torvalds5483f182012-03-04 15:51:42 -0800181 do {
182 if (*cs != *ct)
183 return 1;
184 cs++;
185 ct++;
186 tcount--;
187 } while (tcount);
188 return 0;
189}
190
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700191#endif
192
Linus Torvalds94753db52012-05-10 12:19:19 -0700193static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
194{
Linus Torvalds6326c712012-05-21 16:14:04 -0700195 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700196 /*
197 * Be careful about RCU walk racing with rename:
198 * use ACCESS_ONCE to fetch the name pointer.
199 *
200 * NOTE! Even if a rename will mean that the length
201 * was not loaded atomically, we don't care. The
202 * RCU walk will check the sequence count eventually,
203 * and catch it. And we won't overrun the buffer,
204 * because we're reading the name pointer atomically,
205 * and a dentry name is guaranteed to be properly
206 * terminated with a NUL byte.
207 *
208 * End result: even if 'len' is wrong, we'll exit
209 * early because the data cannot match (there can
210 * be no NUL in the ct/tcount data)
211 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700212 cs = ACCESS_ONCE(dentry->d_name.name);
213 smp_read_barrier_depends();
214 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700215}
216
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400217static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400219 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
220
Arjan van de Venfd217f42008-10-21 06:47:33 -0700221 WARN_ON(!list_empty(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (dname_external(dentry))
223 kfree(dentry->d_name.name);
224 kmem_cache_free(dentry_cache, dentry);
225}
226
227/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100228 * no locks, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
230static void d_free(struct dentry *dentry)
231{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100232 BUG_ON(dentry->d_count);
Nick Piggin3e880fb2011-01-07 17:49:19 +1100233 this_cpu_dec(nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 if (dentry->d_op && dentry->d_op->d_release)
235 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400236
Linus Torvaldsdea36672011-04-24 07:58:46 -0700237 /* if dentry was never visible to RCU, immediate free is OK */
238 if (!(dentry->d_flags & DCACHE_RCUACCESS))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400239 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800240 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400241 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Nick Piggin31e6b012011-01-07 17:49:52 +1100244/**
245 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800246 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100247 * After this call, in-progress rcu-walk path lookup will fail. This
248 * should be called after unhashing, and after changing d_inode (if
249 * the dentry has not already been unhashed).
250 */
251static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
252{
253 assert_spin_locked(&dentry->d_lock);
254 /* Go through a barrier */
255 write_seqcount_barrier(&dentry->d_seq);
256}
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258/*
259 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100260 * d_iput() operation if defined. Dentry has no refcount
261 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800263static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200264 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100265 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
267 struct inode *inode = dentry->d_inode;
268 if (inode) {
269 dentry->d_inode = NULL;
270 list_del_init(&dentry->d_alias);
271 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100272 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700273 if (!inode->i_nlink)
274 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (dentry->d_op && dentry->d_op->d_iput)
276 dentry->d_op->d_iput(dentry, inode);
277 else
278 iput(inode);
279 } else {
280 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 }
282}
283
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700284/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100285 * Release the dentry's inode, using the filesystem
286 * d_iput() operation if defined. dentry remains in-use.
287 */
288static void dentry_unlink_inode(struct dentry * dentry)
289 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100290 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100291{
292 struct inode *inode = dentry->d_inode;
293 dentry->d_inode = NULL;
294 list_del_init(&dentry->d_alias);
295 dentry_rcuwalk_barrier(dentry);
296 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100297 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100298 if (!inode->i_nlink)
299 fsnotify_inoderemove(inode);
300 if (dentry->d_op && dentry->d_op->d_iput)
301 dentry->d_op->d_iput(dentry, inode);
302 else
303 iput(inode);
304}
305
306/*
Sage Weilf0023bc2011-10-28 10:02:42 -0700307 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700308 */
309static void dentry_lru_add(struct dentry *dentry)
310{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400311 if (list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100312 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400313 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
314 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100315 dentry_stat.nr_unused++;
Nick Piggin23044502011-01-07 17:49:31 +1100316 spin_unlock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400317 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700318}
319
Nick Piggin23044502011-01-07 17:49:31 +1100320static void __dentry_lru_del(struct dentry *dentry)
321{
322 list_del_init(&dentry->d_lru);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100323 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100324 dentry->d_sb->s_nr_dentry_unused--;
325 dentry_stat.nr_unused--;
326}
327
Sage Weilf0023bc2011-10-28 10:02:42 -0700328/*
329 * Remove a dentry with references from the LRU.
330 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700331static void dentry_lru_del(struct dentry *dentry)
332{
333 if (!list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100334 spin_lock(&dcache_lru_lock);
335 __dentry_lru_del(dentry);
336 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700337 }
338}
339
Sage Weilf0023bc2011-10-28 10:02:42 -0700340/*
341 * Remove a dentry that is unreferenced and about to be pruned
342 * (unhashed and destroyed) from the LRU, and inform the file system.
343 * This wrapper should be called _prior_ to unhashing a victim dentry.
344 */
345static void dentry_lru_prune(struct dentry *dentry)
346{
347 if (!list_empty(&dentry->d_lru)) {
348 if (dentry->d_flags & DCACHE_OP_PRUNE)
349 dentry->d_op->d_prune(dentry);
350
351 spin_lock(&dcache_lru_lock);
352 __dentry_lru_del(dentry);
353 spin_unlock(&dcache_lru_lock);
354 }
355}
356
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000357static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700358{
Nick Piggin23044502011-01-07 17:49:31 +1100359 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400360 if (list_empty(&dentry->d_lru)) {
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000361 list_add_tail(&dentry->d_lru, list);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400362 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100363 dentry_stat.nr_unused++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400364 } else {
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000365 list_move_tail(&dentry->d_lru, list);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700366 }
Nick Piggin23044502011-01-07 17:49:31 +1100367 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700368}
369
Miklos Szeredid52b9082007-05-08 00:23:46 -0700370/**
371 * d_kill - kill dentry and return parent
372 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800373 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700374 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200375 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700376 *
377 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100378 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100379 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
380 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700381 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100382static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200383 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100384 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100385 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700386{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700387 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400388 /*
389 * Inform try_to_ascend() that we are no longer attached to the
390 * dentry tree
391 */
392 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100393 if (parent)
394 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700395 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100396 /*
397 * dentry_iput drops the locks, at which point nobody (except
398 * transient RCU lookups) can reach this dentry.
399 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700400 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900401 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700402}
403
David Howellsc6627c62011-06-07 14:09:20 +0100404/*
405 * Unhash a dentry without inserting an RCU walk barrier or checking that
406 * dentry->d_lock is locked. The caller must take care of that, if
407 * appropriate.
408 */
409static void __d_shrink(struct dentry *dentry)
410{
411 if (!d_unhashed(dentry)) {
412 struct hlist_bl_head *b;
413 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
414 b = &dentry->d_sb->s_anon;
415 else
416 b = d_hash(dentry->d_parent, dentry->d_name.hash);
417
418 hlist_bl_lock(b);
419 __hlist_bl_del(&dentry->d_hash);
420 dentry->d_hash.pprev = NULL;
421 hlist_bl_unlock(b);
422 }
423}
424
Nick Piggin789680d2011-01-07 17:49:30 +1100425/**
426 * d_drop - drop a dentry
427 * @dentry: dentry to drop
428 *
429 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
430 * be found through a VFS lookup any more. Note that this is different from
431 * deleting the dentry - d_delete will try to mark the dentry negative if
432 * possible, giving a successful _negative_ lookup, while d_drop will
433 * just make the cache lookup fail.
434 *
435 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
436 * reason (NFS timeouts or autofs deletes).
437 *
438 * __d_drop requires dentry->d_lock.
439 */
440void __d_drop(struct dentry *dentry)
441{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700442 if (!d_unhashed(dentry)) {
David Howellsc6627c62011-06-07 14:09:20 +0100443 __d_shrink(dentry);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700444 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100445 }
446}
447EXPORT_SYMBOL(__d_drop);
448
449void d_drop(struct dentry *dentry)
450{
Nick Piggin789680d2011-01-07 17:49:30 +1100451 spin_lock(&dentry->d_lock);
452 __d_drop(dentry);
453 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100454}
455EXPORT_SYMBOL(d_drop);
456
Nick Piggin77812a12011-01-07 17:49:48 +1100457/*
Josef Bacik44396f42011-05-31 11:58:49 -0400458 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
459 * @dentry: dentry to drop
460 *
461 * This is called when we do a lookup on a placeholder dentry that needed to be
462 * looked up. The dentry should have been hashed in order for it to be found by
463 * the lookup code, but now needs to be unhashed while we do the actual lookup
464 * and clear the DCACHE_NEED_LOOKUP flag.
465 */
466void d_clear_need_lookup(struct dentry *dentry)
467{
468 spin_lock(&dentry->d_lock);
469 __d_drop(dentry);
470 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
471 spin_unlock(&dentry->d_lock);
472}
473EXPORT_SYMBOL(d_clear_need_lookup);
474
475/*
Nick Piggin77812a12011-01-07 17:49:48 +1100476 * Finish off a dentry we've decided to kill.
477 * dentry->d_lock must be held, returns with it unlocked.
478 * If ref is non-zero, then decrement the refcount too.
479 * Returns dentry requiring refcount drop, or NULL if we're done.
480 */
481static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
482 __releases(dentry->d_lock)
483{
Nick Piggin873feea2011-01-07 17:50:06 +1100484 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100485 struct dentry *parent;
486
Nick Piggin873feea2011-01-07 17:50:06 +1100487 inode = dentry->d_inode;
488 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100489relock:
490 spin_unlock(&dentry->d_lock);
491 cpu_relax();
492 return dentry; /* try again with same dentry */
493 }
494 if (IS_ROOT(dentry))
495 parent = NULL;
496 else
497 parent = dentry->d_parent;
498 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100499 if (inode)
500 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100501 goto relock;
502 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100503
Nick Piggin77812a12011-01-07 17:49:48 +1100504 if (ref)
505 dentry->d_count--;
Sage Weilf0023bc2011-10-28 10:02:42 -0700506 /*
507 * if dentry was on the d_lru list delete it from there.
508 * inform the fs via d_prune that this dentry is about to be
509 * unhashed and destroyed.
510 */
511 dentry_lru_prune(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100512 /* if it was on the hash then remove it */
513 __d_drop(dentry);
514 return d_kill(dentry, parent);
515}
516
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517/*
518 * This is dput
519 *
520 * This is complicated by the fact that we do not want to put
521 * dentries that are no longer on any hash chain on the unused
522 * list: we'd much rather just get rid of them immediately.
523 *
524 * However, that implies that we have to traverse the dentry
525 * tree upwards to the parents which might _also_ now be
526 * scheduled for deletion (it may have been only waiting for
527 * its last child to go away).
528 *
529 * This tail recursion is done by hand as we don't want to depend
530 * on the compiler to always get this right (gcc generally doesn't).
531 * Real recursion would eat up our stack space.
532 */
533
534/*
535 * dput - release a dentry
536 * @dentry: dentry to release
537 *
538 * Release a dentry. This will drop the usage count and if appropriate
539 * call the dentry unlink method as well as removing it from the queues and
540 * releasing its resources. If the parent dentries were scheduled for release
541 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543void dput(struct dentry *dentry)
544{
545 if (!dentry)
546 return;
547
548repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100549 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100552 BUG_ON(!dentry->d_count);
553 if (dentry->d_count > 1) {
554 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return;
557 }
558
Nick Pigginfb045ad2011-01-07 17:49:55 +1100559 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100561 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
Nick Piggin265ac902010-10-10 05:36:24 -0400563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 /* Unreachable? Get rid of it */
565 if (d_unhashed(dentry))
566 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400567
Josef Bacik44396f42011-05-31 11:58:49 -0400568 /*
569 * If this dentry needs lookup, don't set the referenced flag so that it
570 * is more likely to be cleaned up by the dcache shrinker in case of
571 * memory pressure.
572 */
573 if (!d_need_lookup(dentry))
574 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400575 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400576
Nick Piggin61f3dee2011-01-07 17:49:40 +1100577 dentry->d_count--;
578 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 return;
580
Miklos Szeredid52b9082007-05-08 00:23:46 -0700581kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100582 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700583 if (dentry)
584 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700586EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588/**
589 * d_invalidate - invalidate a dentry
590 * @dentry: dentry to invalidate
591 *
592 * Try to invalidate the dentry if it turns out to be
593 * possible. If there are other dentries that can be
594 * reached through this one we can't delete it and we
595 * return -EBUSY. On success we return 0.
596 *
597 * no dcache lock.
598 */
599
600int d_invalidate(struct dentry * dentry)
601{
602 /*
603 * If it's already been dropped, return OK.
604 */
Nick Pigginda502952011-01-07 17:49:33 +1100605 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100607 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return 0;
609 }
610 /*
611 * Check whether to do a partial shrink_dcache
612 * to get rid of unused child entries.
613 */
614 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100615 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100617 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 }
619
620 /*
621 * Somebody else still using it?
622 *
623 * If it's a directory, we can't drop it
624 * for fear of somebody re-populating it
625 * with children (even though dropping it
626 * would make it unreachable from the root,
627 * we might still populate it if it was a
628 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000629 * We also need to leave mountpoints alone,
630 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 */
Al Viro50e69632011-11-07 16:39:57 +0000632 if (dentry->d_count > 1 && dentry->d_inode) {
633 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 return -EBUSY;
636 }
637 }
638
639 __d_drop(dentry);
640 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 return 0;
642}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700643EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100645/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100646static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100648 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Nick Piggindc0474b2011-01-07 17:49:43 +1100651static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100652{
Nick Piggin23044502011-01-07 17:49:31 +1100653 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100654 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100655 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100656}
657
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100658struct dentry *dget_parent(struct dentry *dentry)
659{
660 struct dentry *ret;
661
662repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100663 /*
664 * Don't need rcu_dereference because we re-check it was correct under
665 * the lock.
666 */
667 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100668 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100669 spin_lock(&ret->d_lock);
670 if (unlikely(ret != dentry->d_parent)) {
671 spin_unlock(&ret->d_lock);
672 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100673 goto repeat;
674 }
Nick Piggina734eb42011-01-07 17:49:44 +1100675 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100676 BUG_ON(!ret->d_count);
677 ret->d_count++;
678 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100679 return ret;
680}
681EXPORT_SYMBOL(dget_parent);
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683/**
684 * d_find_alias - grab a hashed alias of inode
685 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700686 * @want_discon: flag, used by d_splice_alias, to request
687 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 *
689 * If inode has a hashed alias, or is a directory and has any alias,
690 * acquire the reference to alias and return it. Otherwise return NULL.
691 * Notice that if inode is a directory there can be only one alias and
692 * it can be unhashed only if it has no children, or if it is the root
693 * of a filesystem.
694 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700695 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700696 * any other hashed alias over that one unless @want_discon is set,
697 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700699static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700{
Nick Pigginda502952011-01-07 17:49:33 +1100701 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Nick Pigginda502952011-01-07 17:49:33 +1100703again:
704 discon_alias = NULL;
705 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
706 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700708 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100709 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700711 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100712 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100713 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 return alias;
715 }
716 }
Nick Pigginda502952011-01-07 17:49:33 +1100717 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 }
Nick Pigginda502952011-01-07 17:49:33 +1100719 if (discon_alias) {
720 alias = discon_alias;
721 spin_lock(&alias->d_lock);
722 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
723 if (IS_ROOT(alias) &&
724 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100725 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100726 spin_unlock(&alias->d_lock);
727 return alias;
728 }
729 }
730 spin_unlock(&alias->d_lock);
731 goto again;
732 }
733 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734}
735
Nick Pigginda502952011-01-07 17:49:33 +1100736struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
David Howells214fda12006-03-25 03:06:36 -0800738 struct dentry *de = NULL;
739
740 if (!list_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100741 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700742 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100743 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return de;
746}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700747EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
749/*
750 * Try to kill dentries associated with this inode.
751 * WARNING: you must own a reference to inode.
752 */
753void d_prune_aliases(struct inode *inode)
754{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700755 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100757 spin_lock(&inode->i_lock);
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700758 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100760 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100761 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 __d_drop(dentry);
763 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100764 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 dput(dentry);
766 goto restart;
767 }
768 spin_unlock(&dentry->d_lock);
769 }
Nick Piggin873feea2011-01-07 17:50:06 +1100770 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700772EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
774/*
Nick Piggin77812a12011-01-07 17:49:48 +1100775 * Try to throw away a dentry - free the inode, dput the parent.
776 * Requires dentry->d_lock is held, and dentry->d_count == 0.
777 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700778 *
Nick Piggin77812a12011-01-07 17:49:48 +1100779 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 */
Nick Piggin77812a12011-01-07 17:49:48 +1100781static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200782 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
Nick Piggin77812a12011-01-07 17:49:48 +1100784 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700785
Nick Piggin77812a12011-01-07 17:49:48 +1100786 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700787 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100788 * If dentry_kill returns NULL, we have nothing more to do.
789 * if it returns the same dentry, trylocks failed. In either
790 * case, just loop again.
791 *
792 * Otherwise, we need to prune ancestors too. This is necessary
793 * to prevent quadratic behavior of shrink_dcache_parent(), but
794 * is also expected to be beneficial in reducing dentry cache
795 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700796 */
Nick Piggin77812a12011-01-07 17:49:48 +1100797 if (!parent)
798 return;
799 if (parent == dentry)
800 return;
801
802 /* Prune ancestors. */
803 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700804 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100805 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100806 if (dentry->d_count > 1) {
807 dentry->d_count--;
808 spin_unlock(&dentry->d_lock);
809 return;
810 }
Nick Piggin77812a12011-01-07 17:49:48 +1100811 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813}
814
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400815static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700817 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700818
Nick Pigginec336792011-01-07 17:49:47 +1100819 rcu_read_lock();
820 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100821 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
822 if (&dentry->d_lru == list)
823 break; /* empty */
824 spin_lock(&dentry->d_lock);
825 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
826 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100827 continue;
828 }
829
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 /*
831 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700832 * the LRU because of laziness during lookup. Do not free
833 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100835 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100836 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700837 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 continue;
839 }
Nick Pigginec336792011-01-07 17:49:47 +1100840
Nick Pigginec336792011-01-07 17:49:47 +1100841 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100842
843 try_prune_one_dentry(dentry);
844
Nick Pigginec336792011-01-07 17:49:47 +1100845 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 }
Nick Pigginec336792011-01-07 17:49:47 +1100847 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400848}
849
850/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000851 * prune_dcache_sb - shrink the dcache
852 * @sb: superblock
853 * @count: number of entries to try to free
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400854 *
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000855 * Attempt to shrink the superblock dcache LRU by @count entries. This is
856 * done when we need more memory an called from the superblock shrinker
857 * function.
858 *
859 * This function may fail to free any resources if all the dentries are in
860 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400861 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000862void prune_dcache_sb(struct super_block *sb, int count)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400863{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400864 struct dentry *dentry;
865 LIST_HEAD(referenced);
866 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400867
Nick Piggin23044502011-01-07 17:49:31 +1100868relock:
869 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400870 while (!list_empty(&sb->s_dentry_lru)) {
871 dentry = list_entry(sb->s_dentry_lru.prev,
872 struct dentry, d_lru);
873 BUG_ON(dentry->d_sb != sb);
874
Nick Piggin23044502011-01-07 17:49:31 +1100875 if (!spin_trylock(&dentry->d_lock)) {
876 spin_unlock(&dcache_lru_lock);
877 cpu_relax();
878 goto relock;
879 }
880
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000881 if (dentry->d_flags & DCACHE_REFERENCED) {
Nick Piggin23044502011-01-07 17:49:31 +1100882 dentry->d_flags &= ~DCACHE_REFERENCED;
883 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400884 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100885 } else {
886 list_move_tail(&dentry->d_lru, &tmp);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100887 dentry->d_flags |= DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100888 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000889 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100890 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400891 }
Nick Pigginec336792011-01-07 17:49:47 +1100892 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400893 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700894 if (!list_empty(&referenced))
895 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100896 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100897
898 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
900
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700901/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 * shrink_dcache_sb - shrink dcache for a superblock
903 * @sb: superblock
904 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400905 * Shrink the dcache for the specified super block. This is used to free
906 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400908void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400910 LIST_HEAD(tmp);
911
Nick Piggin23044502011-01-07 17:49:31 +1100912 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400913 while (!list_empty(&sb->s_dentry_lru)) {
914 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100915 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400916 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100917 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400918 }
Nick Piggin23044502011-01-07 17:49:31 +1100919 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700921EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923/*
David Howellsc636ebd2006-10-11 01:22:19 -0700924 * destroy a single subtree of dentries for unmount
925 * - see the comments on shrink_dcache_for_umount() for a description of the
926 * locking
927 */
928static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
929{
930 struct dentry *parent;
931
932 BUG_ON(!IS_ROOT(dentry));
933
David Howellsc636ebd2006-10-11 01:22:19 -0700934 for (;;) {
935 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100936 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700937 dentry = list_entry(dentry->d_subdirs.next,
938 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700939
940 /* consume the dentries from this leaf up through its parents
941 * until we find one with children or run out altogether */
942 do {
943 struct inode *inode;
944
Sage Weilf0023bc2011-10-28 10:02:42 -0700945 /*
946 * remove the dentry from the lru, and inform
947 * the fs that this dentry is about to be
948 * unhashed and destroyed.
949 */
950 dentry_lru_prune(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100951 __d_shrink(dentry);
952
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100953 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700954 printk(KERN_ERR
955 "BUG: Dentry %p{i=%lx,n=%s}"
956 " still in use (%d)"
957 " [unmount of %s %s]\n",
958 dentry,
959 dentry->d_inode ?
960 dentry->d_inode->i_ino : 0UL,
961 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100962 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700963 dentry->d_sb->s_type->name,
964 dentry->d_sb->s_id);
965 BUG();
966 }
967
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100968 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700969 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100970 list_del(&dentry->d_u.d_child);
971 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900972 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100973 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100974 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900975 }
David Howellsc636ebd2006-10-11 01:22:19 -0700976
David Howellsc636ebd2006-10-11 01:22:19 -0700977 inode = dentry->d_inode;
978 if (inode) {
979 dentry->d_inode = NULL;
980 list_del_init(&dentry->d_alias);
981 if (dentry->d_op && dentry->d_op->d_iput)
982 dentry->d_op->d_iput(dentry, inode);
983 else
984 iput(inode);
985 }
986
987 d_free(dentry);
988
989 /* finished when we fall off the top of the tree,
990 * otherwise we ascend to the parent and move to the
991 * next sibling if there is one */
992 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400993 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700994 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700995 } while (list_empty(&dentry->d_subdirs));
996
997 dentry = list_entry(dentry->d_subdirs.next,
998 struct dentry, d_u.d_child);
999 }
1000}
1001
1002/*
1003 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001004 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -07001005 * - the superblock is detached from all mountings and open files, so the
1006 * dentry trees will not be rearranged by the VFS
1007 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1008 * any dentries belonging to this superblock that it comes across
1009 * - the filesystem itself is no longer permitted to rearrange the dentries
1010 * in this superblock
1011 */
1012void shrink_dcache_for_umount(struct super_block *sb)
1013{
1014 struct dentry *dentry;
1015
1016 if (down_read_trylock(&sb->s_umount))
1017 BUG();
1018
1019 dentry = sb->s_root;
1020 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001021 dentry->d_count--;
David Howellsc636ebd2006-10-11 01:22:19 -07001022 shrink_dcache_for_umount_subtree(dentry);
1023
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001024 while (!hlist_bl_empty(&sb->s_anon)) {
1025 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001026 shrink_dcache_for_umount_subtree(dentry);
1027 }
1028}
1029
1030/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001031 * This tries to ascend one level of parenthood, but
1032 * we can race with renaming, so we need to re-check
1033 * the parenthood after dropping the lock and check
1034 * that the sequence number still matches.
1035 */
1036static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1037{
1038 struct dentry *new = old->d_parent;
1039
1040 rcu_read_lock();
1041 spin_unlock(&old->d_lock);
1042 spin_lock(&new->d_lock);
1043
1044 /*
1045 * might go back up the wrong parent if we have had a rename
1046 * or deletion
1047 */
1048 if (new != old->d_parent ||
Trond Myklebustc83ce982011-03-15 13:36:43 -04001049 (old->d_flags & DCACHE_DISCONNECTED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001050 (!locked && read_seqretry(&rename_lock, seq))) {
1051 spin_unlock(&new->d_lock);
1052 new = NULL;
1053 }
1054 rcu_read_unlock();
1055 return new;
1056}
1057
1058
1059/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 * Search for at least 1 mount point in the dentry's subdirs.
1061 * We descend to the next level whenever the d_subdirs
1062 * list is non-empty and continue searching.
1063 */
1064
1065/**
1066 * have_submounts - check for mounts over a dentry
1067 * @parent: dentry to check.
1068 *
1069 * Return true if the parent or its subdirectories contain
1070 * a mount point
1071 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072int have_submounts(struct dentry *parent)
1073{
Nick Piggin949854d2011-01-07 17:49:37 +11001074 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001076 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001077 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001078
Nick Piggin949854d2011-01-07 17:49:37 +11001079 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001080again:
1081 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 if (d_mountpoint(parent))
1084 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001085 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086repeat:
1087 next = this_parent->d_subdirs.next;
1088resume:
1089 while (next != &this_parent->d_subdirs) {
1090 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001091 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001093
1094 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001096 if (d_mountpoint(dentry)) {
1097 spin_unlock(&dentry->d_lock);
1098 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001100 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001102 spin_unlock(&this_parent->d_lock);
1103 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001105 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 goto repeat;
1107 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001108 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 }
1110 /*
1111 * All done at this level ... ascend and resume the search.
1112 */
1113 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001114 struct dentry *child = this_parent;
1115 this_parent = try_to_ascend(this_parent, locked, seq);
1116 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001117 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001118 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 goto resume;
1120 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001121 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001122 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001123 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001124 if (locked)
1125 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 return 0; /* No mount points found in tree */
1127positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001128 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001129 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001130 if (locked)
1131 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001133
1134rename_retry:
1135 locked = 1;
1136 write_seqlock(&rename_lock);
1137 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001139EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
1141/*
1142 * Search the dentry child list for the specified parent,
1143 * and move any unused dentries to the end of the unused
1144 * list for prune_dcache(). We descend to the next level
1145 * whenever the d_subdirs list is non-empty and continue
1146 * searching.
1147 *
1148 * It returns zero iff there are no unused children,
1149 * otherwise it returns the number of children moved to
1150 * the end of the unused list. This may not be the total
1151 * number of unused children, because select_parent can
1152 * drop the lock and return early due to latency
1153 * constraints.
1154 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001155static int select_parent(struct dentry *parent, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
Nick Piggin949854d2011-01-07 17:49:37 +11001157 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001159 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001161 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Nick Piggin949854d2011-01-07 17:49:37 +11001163 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001164again:
1165 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001166 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167repeat:
1168 next = this_parent->d_subdirs.next;
1169resume:
1170 while (next != &this_parent->d_subdirs) {
1171 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001172 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 next = tmp->next;
1174
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001175 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001176
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001177 /*
1178 * move only zero ref count dentries to the dispose list.
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001179 *
1180 * Those which are presently on the shrink list, being processed
1181 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1182 * loop in shrink_dcache_parent() might not make any progress
1183 * and loop forever.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 */
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001185 if (dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001186 dentry_lru_del(dentry);
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001187 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1188 dentry_lru_move_list(dentry, dispose);
1189 dentry->d_flags |= DCACHE_SHRINK_LIST;
1190 found++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 /*
1193 * We can return to the caller if we have found some (this
1194 * ensures forward progress). We'll be coming back to find
1195 * the rest.
1196 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001197 if (found && need_resched()) {
1198 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202 /*
1203 * Descend a level if the d_subdirs list is non-empty.
1204 */
1205 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001206 spin_unlock(&this_parent->d_lock);
1207 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001209 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 goto repeat;
1211 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001212
1213 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215 /*
1216 * All done at this level ... ascend and resume the search.
1217 */
1218 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001219 struct dentry *child = this_parent;
1220 this_parent = try_to_ascend(this_parent, locked, seq);
1221 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001222 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001223 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 goto resume;
1225 }
1226out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001227 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001228 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001229 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001230 if (locked)
1231 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001233
1234rename_retry:
1235 if (found)
1236 return found;
1237 locked = 1;
1238 write_seqlock(&rename_lock);
1239 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240}
1241
1242/**
1243 * shrink_dcache_parent - prune dcache
1244 * @parent: parent of entries to prune
1245 *
1246 * Prune the dcache to remove unused children of the parent dentry.
1247 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248void shrink_dcache_parent(struct dentry * parent)
1249{
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001250 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 int found;
1252
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001253 while ((found = select_parent(parent, &dispose)) != 0)
1254 shrink_dentry_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001256EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258/**
Al Viroa4464db2011-07-07 15:03:58 -04001259 * __d_alloc - allocate a dcache entry
1260 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 * @name: qstr of the name
1262 *
1263 * Allocates a dentry. It returns %NULL if there is insufficient memory
1264 * available. On a success the dentry is returned. The name passed in is
1265 * copied and the copy passed in may be reused after this call.
1266 */
1267
Al Viroa4464db2011-07-07 15:03:58 -04001268struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269{
1270 struct dentry *dentry;
1271 char *dname;
1272
Mel Gormane12ba742007-10-16 01:25:52 -07001273 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 if (!dentry)
1275 return NULL;
1276
Linus Torvalds6326c712012-05-21 16:14:04 -07001277 /*
1278 * We guarantee that the inline name is always NUL-terminated.
1279 * This way the memcpy() done by the name switching in rename
1280 * will still always have a NUL at the end, even if we might
1281 * be overwriting an internal NUL character
1282 */
1283 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (name->len > DNAME_INLINE_LEN-1) {
1285 dname = kmalloc(name->len + 1, GFP_KERNEL);
1286 if (!dname) {
1287 kmem_cache_free(dentry_cache, dentry);
1288 return NULL;
1289 }
1290 } else {
1291 dname = dentry->d_iname;
1292 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
1294 dentry->d_name.len = name->len;
1295 dentry->d_name.hash = name->hash;
1296 memcpy(dname, name->name, name->len);
1297 dname[name->len] = 0;
1298
Linus Torvalds6326c712012-05-21 16:14:04 -07001299 /* Make sure we always see the terminating NUL character */
1300 smp_wmb();
1301 dentry->d_name.name = dname;
1302
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001303 dentry->d_count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001304 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001306 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001308 dentry->d_parent = dentry;
1309 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 dentry->d_op = NULL;
1311 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001312 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 INIT_LIST_HEAD(&dentry->d_lru);
1314 INIT_LIST_HEAD(&dentry->d_subdirs);
1315 INIT_LIST_HEAD(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001316 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001317 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Nick Piggin3e880fb2011-01-07 17:49:19 +11001319 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 return dentry;
1322}
Al Viroa4464db2011-07-07 15:03:58 -04001323
1324/**
1325 * d_alloc - allocate a dcache entry
1326 * @parent: parent of entry to allocate
1327 * @name: qstr of the name
1328 *
1329 * Allocates a dentry. It returns %NULL if there is insufficient memory
1330 * available. On a success the dentry is returned. The name passed in is
1331 * copied and the copy passed in may be reused after this call.
1332 */
1333struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1334{
1335 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1336 if (!dentry)
1337 return NULL;
1338
1339 spin_lock(&parent->d_lock);
1340 /*
1341 * don't need child lock because it is not subject
1342 * to concurrency here
1343 */
1344 __dget_dlock(parent);
1345 dentry->d_parent = parent;
1346 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1347 spin_unlock(&parent->d_lock);
1348
1349 return dentry;
1350}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001351EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Nick Piggin4b936882011-01-07 17:50:07 +11001353struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1354{
Al Viroa4464db2011-07-07 15:03:58 -04001355 struct dentry *dentry = __d_alloc(sb, name);
1356 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001357 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001358 return dentry;
1359}
1360EXPORT_SYMBOL(d_alloc_pseudo);
1361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1363{
1364 struct qstr q;
1365
1366 q.name = name;
1367 q.len = strlen(name);
1368 q.hash = full_name_hash(q.name, q.len);
1369 return d_alloc(parent, &q);
1370}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001371EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
Nick Pigginfb045ad2011-01-07 17:49:55 +11001373void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1374{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001375 WARN_ON_ONCE(dentry->d_op);
1376 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001377 DCACHE_OP_COMPARE |
1378 DCACHE_OP_REVALIDATE |
1379 DCACHE_OP_DELETE ));
1380 dentry->d_op = op;
1381 if (!op)
1382 return;
1383 if (op->d_hash)
1384 dentry->d_flags |= DCACHE_OP_HASH;
1385 if (op->d_compare)
1386 dentry->d_flags |= DCACHE_OP_COMPARE;
1387 if (op->d_revalidate)
1388 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1389 if (op->d_delete)
1390 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001391 if (op->d_prune)
1392 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001393
1394}
1395EXPORT_SYMBOL(d_set_d_op);
1396
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001397static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1398{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001399 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001400 if (inode) {
1401 if (unlikely(IS_AUTOMOUNT(inode)))
1402 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001403 list_add(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001404 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001405 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001406 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001407 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001408 fsnotify_d_instantiate(dentry, inode);
1409}
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411/**
1412 * d_instantiate - fill in inode information for a dentry
1413 * @entry: dentry to complete
1414 * @inode: inode to attach to this dentry
1415 *
1416 * Fill in inode information in the entry.
1417 *
1418 * This turns negative dentries into productive full members
1419 * of society.
1420 *
1421 * NOTE! This assumes that the inode count has been incremented
1422 * (or otherwise set) by the caller to indicate that it is now
1423 * in use by the dcache.
1424 */
1425
1426void d_instantiate(struct dentry *entry, struct inode * inode)
1427{
Eric Sesterhenn28133c72006-03-26 18:25:39 +02001428 BUG_ON(!list_empty(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001429 if (inode)
1430 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001431 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001432 if (inode)
1433 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 security_d_instantiate(entry, inode);
1435}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001436EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
1438/**
1439 * d_instantiate_unique - instantiate a non-aliased dentry
1440 * @entry: dentry to instantiate
1441 * @inode: inode to attach to this dentry
1442 *
1443 * Fill in inode information in the entry. On success, it returns NULL.
1444 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001445 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 *
1447 * Note that in order to avoid conflicts with rename() etc, the caller
1448 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001449 *
1450 * This also assumes that the inode count has been incremented
1451 * (or otherwise set) by the caller to indicate that it is now
1452 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 */
David Howells770bfad2006-08-22 20:06:07 -04001454static struct dentry *__d_instantiate_unique(struct dentry *entry,
1455 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
1457 struct dentry *alias;
1458 int len = entry->d_name.len;
1459 const char *name = entry->d_name.name;
1460 unsigned int hash = entry->d_name.hash;
1461
David Howells770bfad2006-08-22 20:06:07 -04001462 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001463 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001464 return NULL;
1465 }
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001468 /*
1469 * Don't need alias->d_lock here, because aliases with
1470 * d_parent == entry->d_parent are not subject to name or
1471 * parent changes, because the parent inode i_mutex is held.
1472 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001473 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 continue;
1475 if (alias->d_parent != entry->d_parent)
1476 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001477 if (alias->d_name.len != len)
1478 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001479 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001481 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 return alias;
1483 }
David Howells770bfad2006-08-22 20:06:07 -04001484
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001485 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return NULL;
1487}
David Howells770bfad2006-08-22 20:06:07 -04001488
1489struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1490{
1491 struct dentry *result;
1492
1493 BUG_ON(!list_empty(&entry->d_alias));
1494
Nick Piggin873feea2011-01-07 17:50:06 +11001495 if (inode)
1496 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001497 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001498 if (inode)
1499 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001500
1501 if (!result) {
1502 security_d_instantiate(entry, inode);
1503 return NULL;
1504 }
1505
1506 BUG_ON(!d_unhashed(result));
1507 iput(inode);
1508 return result;
1509}
1510
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511EXPORT_SYMBOL(d_instantiate_unique);
1512
Al Viroadc0e912012-01-08 16:49:21 -05001513struct dentry *d_make_root(struct inode *root_inode)
1514{
1515 struct dentry *res = NULL;
1516
1517 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001518 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001519
1520 res = __d_alloc(root_inode->i_sb, &name);
1521 if (res)
1522 d_instantiate(res, root_inode);
1523 else
1524 iput(root_inode);
1525 }
1526 return res;
1527}
1528EXPORT_SYMBOL(d_make_root);
1529
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001530static struct dentry * __d_find_any_alias(struct inode *inode)
1531{
1532 struct dentry *alias;
1533
1534 if (list_empty(&inode->i_dentry))
1535 return NULL;
1536 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1537 __dget(alias);
1538 return alias;
1539}
1540
Sage Weil46f72b32012-01-10 09:04:37 -08001541/**
1542 * d_find_any_alias - find any alias for a given inode
1543 * @inode: inode to find an alias for
1544 *
1545 * If any aliases exist for the given inode, take and return a
1546 * reference for one of them. If no aliases exist, return %NULL.
1547 */
1548struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001549{
1550 struct dentry *de;
1551
1552 spin_lock(&inode->i_lock);
1553 de = __d_find_any_alias(inode);
1554 spin_unlock(&inode->i_lock);
1555 return de;
1556}
Sage Weil46f72b32012-01-10 09:04:37 -08001557EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001560 * d_obtain_alias - find or allocate a dentry for a given inode
1561 * @inode: inode to allocate the dentry for
1562 *
1563 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1564 * similar open by handle operations. The returned dentry may be anonymous,
1565 * or may have a full name (if the inode was already in the cache).
1566 *
1567 * When called on a directory inode, we must ensure that the inode only ever
1568 * has one dentry. If a dentry is found, that is returned instead of
1569 * allocating a new one.
1570 *
1571 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001572 * to the dentry. In case of an error the reference on the inode is released.
1573 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1574 * be passed in and will be the error will be propagate to the return value,
1575 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001576 */
1577struct dentry *d_obtain_alias(struct inode *inode)
1578{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001579 static const struct qstr anonstring = { .name = "" };
1580 struct dentry *tmp;
1581 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001582
1583 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001584 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001585 if (IS_ERR(inode))
1586 return ERR_CAST(inode);
1587
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001588 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001589 if (res)
1590 goto out_iput;
1591
Al Viroa4464db2011-07-07 15:03:58 -04001592 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001593 if (!tmp) {
1594 res = ERR_PTR(-ENOMEM);
1595 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001596 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001597
Nick Piggin873feea2011-01-07 17:50:06 +11001598 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001599 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001600 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001601 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001602 dput(tmp);
1603 goto out_iput;
1604 }
1605
1606 /* attach a disconnected dentry */
1607 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001608 tmp->d_inode = inode;
1609 tmp->d_flags |= DCACHE_DISCONNECTED;
Christoph Hellwig9308a612008-08-11 15:49:12 +02001610 list_add(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001611 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001612 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001613 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001614 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001615 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001616 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001617
Christoph Hellwig9308a612008-08-11 15:49:12 +02001618 return tmp;
1619
1620 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001621 if (res && !IS_ERR(res))
1622 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001623 iput(inode);
1624 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001625}
Benny Halevyadc48722009-02-27 14:02:59 -08001626EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628/**
1629 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1630 * @inode: the inode which may have a disconnected dentry
1631 * @dentry: a negative dentry which we want to point to the inode.
1632 *
1633 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1634 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1635 * and return it, else simply d_add the inode to the dentry and return NULL.
1636 *
1637 * This is needed in the lookup routine of any filesystem that is exportable
1638 * (via knfsd) so that we can build dcache paths to directories effectively.
1639 *
1640 * If a dentry was found and moved, then it is returned. Otherwise NULL
1641 * is returned. This matches the expected return value of ->lookup.
1642 *
1643 */
1644struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1645{
1646 struct dentry *new = NULL;
1647
Al Viroa9049372011-07-08 21:20:11 -04001648 if (IS_ERR(inode))
1649 return ERR_CAST(inode);
1650
NeilBrown21c0d8f2006-10-04 02:16:16 -07001651 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001652 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001653 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001655 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001656 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 d_move(new, dentry);
1659 iput(inode);
1660 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001661 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001662 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001663 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 security_d_instantiate(dentry, inode);
1665 d_rehash(dentry);
1666 }
1667 } else
1668 d_add(dentry, inode);
1669 return new;
1670}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001671EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
Barry Naujok94035402008-05-21 16:50:46 +10001673/**
1674 * d_add_ci - lookup or allocate new dentry with case-exact name
1675 * @inode: the inode case-insensitive lookup has found
1676 * @dentry: the negative dentry that was passed to the parent's lookup func
1677 * @name: the case-exact name to be associated with the returned dentry
1678 *
1679 * This is to avoid filling the dcache with case-insensitive names to the
1680 * same inode, only the actual correct case is stored in the dcache for
1681 * case-insensitive filesystems.
1682 *
1683 * For a case-insensitive lookup match and if the the case-exact dentry
1684 * already exists in in the dcache, use it and return it.
1685 *
1686 * If no entry exists with the exact case name, allocate new dentry with
1687 * the exact case, and return the spliced entry.
1688 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001689struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001690 struct qstr *name)
1691{
1692 int error;
1693 struct dentry *found;
1694 struct dentry *new;
1695
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001696 /*
1697 * First check if a dentry matching the name already exists,
1698 * if not go ahead and create it now.
1699 */
Barry Naujok94035402008-05-21 16:50:46 +10001700 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001701 if (!found) {
1702 new = d_alloc(dentry->d_parent, name);
1703 if (!new) {
1704 error = -ENOMEM;
1705 goto err_out;
1706 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001707
Barry Naujok94035402008-05-21 16:50:46 +10001708 found = d_splice_alias(inode, new);
1709 if (found) {
1710 dput(new);
1711 return found;
1712 }
1713 return new;
1714 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001715
1716 /*
1717 * If a matching dentry exists, and it's not negative use it.
1718 *
1719 * Decrement the reference count to balance the iget() done
1720 * earlier on.
1721 */
Barry Naujok94035402008-05-21 16:50:46 +10001722 if (found->d_inode) {
1723 if (unlikely(found->d_inode != inode)) {
1724 /* This can't happen because bad inodes are unhashed. */
1725 BUG_ON(!is_bad_inode(inode));
1726 BUG_ON(!is_bad_inode(found->d_inode));
1727 }
Barry Naujok94035402008-05-21 16:50:46 +10001728 iput(inode);
1729 return found;
1730 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001731
Barry Naujok94035402008-05-21 16:50:46 +10001732 /*
Josef Bacik44396f42011-05-31 11:58:49 -04001733 * We are going to instantiate this dentry, unhash it and clear the
1734 * lookup flag so we can do that.
1735 */
1736 if (unlikely(d_need_lookup(found)))
1737 d_clear_need_lookup(found);
1738
1739 /*
Barry Naujok94035402008-05-21 16:50:46 +10001740 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001741 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001742 */
Al Viro4513d892011-07-17 10:52:14 -04001743 new = d_splice_alias(inode, found);
1744 if (new) {
1745 dput(found);
1746 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001747 }
Al Viro4513d892011-07-17 10:52:14 -04001748 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001749
1750err_out:
1751 iput(inode);
1752 return ERR_PTR(error);
1753}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001754EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001756/*
1757 * Do the slow-case of the dentry name compare.
1758 *
1759 * Unlike the dentry_cmp() function, we need to atomically
1760 * load the name, length and inode information, so that the
1761 * filesystem can rely on them, and can use the 'name' and
1762 * 'len' information without worrying about walking off the
1763 * end of memory etc.
1764 *
1765 * Thus the read_seqcount_retry() and the "duplicate" info
1766 * in arguments (the low-level filesystem should not look
1767 * at the dentry inode or name contents directly, since
1768 * rename can change them while we're in RCU mode).
1769 */
1770enum slow_d_compare {
1771 D_COMP_OK,
1772 D_COMP_NOMATCH,
1773 D_COMP_SEQRETRY,
1774};
1775
1776static noinline enum slow_d_compare slow_dentry_cmp(
1777 const struct dentry *parent,
1778 struct inode *inode,
1779 struct dentry *dentry,
1780 unsigned int seq,
1781 const struct qstr *name)
1782{
1783 int tlen = dentry->d_name.len;
1784 const char *tname = dentry->d_name.name;
1785 struct inode *i = dentry->d_inode;
1786
1787 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1788 cpu_relax();
1789 return D_COMP_SEQRETRY;
1790 }
1791 if (parent->d_op->d_compare(parent, inode,
1792 dentry, i,
1793 tlen, tname, name))
1794 return D_COMP_NOMATCH;
1795 return D_COMP_OK;
1796}
1797
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001799 * __d_lookup_rcu - search for a dentry (racy, store-free)
1800 * @parent: parent dentry
1801 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001802 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001803 * @inode: returns dentry->d_inode when the inode was found valid.
1804 * Returns: dentry, or NULL
1805 *
1806 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1807 * resolution (store-free path walking) design described in
1808 * Documentation/filesystems/path-lookup.txt.
1809 *
1810 * This is not to be used outside core vfs.
1811 *
1812 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1813 * held, and rcu_read_lock held. The returned dentry must not be stored into
1814 * without taking d_lock and checking d_seq sequence count against @seq
1815 * returned here.
1816 *
1817 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1818 * function.
1819 *
1820 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1821 * the returned dentry, so long as its parent's seqlock is checked after the
1822 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1823 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001824 *
1825 * NOTE! The caller *has* to check the resulting dentry against the sequence
1826 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001827 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001828struct dentry *__d_lookup_rcu(const struct dentry *parent,
1829 const struct qstr *name,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001830 unsigned *seqp, struct inode *inode)
Nick Piggin31e6b012011-01-07 17:49:52 +11001831{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001832 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001833 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001834 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001835 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001836 struct dentry *dentry;
1837
1838 /*
1839 * 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 /*
1846 * The hash list is protected using RCU.
1847 *
1848 * Carefully use d_seq when comparing a candidate dentry, to avoid
1849 * races 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 Piggin31e6b012011-01-07 17:49:52 +11001857 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001858 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08001859 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001860
Nick Piggin31e6b012011-01-07 17:49:52 +11001861seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001862 /*
1863 * The dentry sequence count protects us from concurrent
1864 * renames, and thus protects inode, parent and name fields.
1865 *
1866 * The caller must perform a seqcount check in order
1867 * to do anything useful with the returned dentry,
1868 * including using the 'd_inode' pointer.
1869 *
1870 * NOTE! We do a "raw" seqcount_begin here. That means that
1871 * we don't wait for the sequence count to stabilize if it
1872 * is in the middle of a sequence change. If we do the slow
1873 * dentry compare, we will do seqretries until it is stable,
1874 * and if we end up with a successful lookup, we actually
1875 * want to exit RCU lookup anyway.
1876 */
1877 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001878 if (dentry->d_parent != parent)
1879 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07001880 if (d_unhashed(dentry))
1881 continue;
Linus Torvalds8966be92012-03-02 14:23:30 -08001882 *seqp = seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001883
1884 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001885 if (dentry->d_name.hash != hashlen_hash(hashlen))
1886 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001887 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1888 case D_COMP_OK:
1889 return dentry;
1890 case D_COMP_NOMATCH:
1891 continue;
1892 default:
1893 goto seqretry;
1894 }
1895 }
1896
Linus Torvalds26fe5752012-05-10 13:14:12 -07001897 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07001898 continue;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001899 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001900 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11001901 }
1902 return NULL;
1903}
1904
1905/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 * d_lookup - search for a dentry
1907 * @parent: parent dentry
1908 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001909 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001911 * d_lookup searches the children of the parent dentry for the name in
1912 * question. If the dentry is found its reference count is incremented and the
1913 * dentry is returned. The caller must use dput to free the entry when it has
1914 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001916struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917{
Nick Piggin31e6b012011-01-07 17:49:52 +11001918 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001919 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
1921 do {
1922 seq = read_seqbegin(&rename_lock);
1923 dentry = __d_lookup(parent, name);
1924 if (dentry)
1925 break;
1926 } while (read_seqretry(&rename_lock, seq));
1927 return dentry;
1928}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001929EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
Nick Piggin31e6b012011-01-07 17:49:52 +11001931/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001932 * __d_lookup - search for a dentry (racy)
1933 * @parent: parent dentry
1934 * @name: qstr of name we wish to find
1935 * Returns: dentry, or NULL
1936 *
1937 * __d_lookup is like d_lookup, however it may (rarely) return a
1938 * false-negative result due to unrelated rename activity.
1939 *
1940 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1941 * however it must be used carefully, eg. with a following d_lookup in
1942 * the case of failure.
1943 *
1944 * __d_lookup callers must be commented.
1945 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001946struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947{
1948 unsigned int len = name->len;
1949 unsigned int hash = name->hash;
1950 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001951 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001952 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001953 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001954 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Nick Pigginb04f7842010-08-18 04:37:34 +10001956 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001957 * Note: There is significant duplication with __d_lookup_rcu which is
1958 * required to prevent single threaded performance regressions
1959 * especially on architectures where smp_rmb (in seqcounts) are costly.
1960 * Keep the two functions in sync.
1961 */
1962
1963 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001964 * The hash list is protected using RCU.
1965 *
1966 * Take d_lock when comparing a candidate dentry, to avoid races
1967 * with d_move().
1968 *
1969 * It is possible that concurrent renames can mess up our list
1970 * walk here and result in missing our dentry, resulting in the
1971 * false-negative result. d_lookup() protects against concurrent
1972 * renames using rename_lock seqlock.
1973 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001974 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001975 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 rcu_read_lock();
1977
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001978 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 if (dentry->d_name.hash != hash)
1981 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
1983 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 if (dentry->d_parent != parent)
1985 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001986 if (d_unhashed(dentry))
1987 goto next;
1988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 /*
1990 * It is safe to compare names since d_move() cannot
1991 * change the qstr (protected by d_lock).
1992 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11001993 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001994 int tlen = dentry->d_name.len;
1995 const char *tname = dentry->d_name.name;
Nick Piggin621e1552011-01-07 17:49:27 +11001996 if (parent->d_op->d_compare(parent, parent->d_inode,
1997 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001998 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 goto next;
2000 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002001 if (dentry->d_name.len != len)
2002 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002003 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 goto next;
2005 }
2006
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002007 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002008 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 spin_unlock(&dentry->d_lock);
2010 break;
2011next:
2012 spin_unlock(&dentry->d_lock);
2013 }
2014 rcu_read_unlock();
2015
2016 return found;
2017}
2018
2019/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002020 * d_hash_and_lookup - hash the qstr then search for a dentry
2021 * @dir: Directory to search in
2022 * @name: qstr of name we wish to find
2023 *
2024 * On hash failure or on lookup failure NULL is returned.
2025 */
2026struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2027{
2028 struct dentry *dentry = NULL;
2029
2030 /*
2031 * Check for a fs-specific hash function. Note that we must
2032 * calculate the standard hash first, as the d_op->d_hash()
2033 * routine may choose to leave the hash value unchanged.
2034 */
2035 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002036 if (dir->d_flags & DCACHE_OP_HASH) {
Nick Pigginb1e6a012011-01-07 17:49:28 +11002037 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002038 goto out;
2039 }
2040 dentry = d_lookup(dir, name);
2041out:
2042 return dentry;
2043}
2044
2045/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002046 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002048 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 *
2050 * An insecure source has sent us a dentry, here we verify it and dget() it.
2051 * This is used by ncpfs in its readdir implementation.
2052 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002053 *
2054 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002056int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057{
Nick Piggin786a5e12011-01-07 17:49:16 +11002058 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002059
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002060 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002061 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2062 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002063 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002064 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002065 spin_unlock(&dentry->d_lock);
2066 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 return 1;
2068 }
2069 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002070 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002071
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return 0;
2073}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002074EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075
2076/*
2077 * When a file is deleted, we have two options:
2078 * - turn this dentry into a negative dentry
2079 * - unhash this dentry and free it.
2080 *
2081 * Usually, we want to just turn this into
2082 * a negative dentry, but if anybody else is
2083 * currently using the dentry or the inode
2084 * we can't do that and we fall back on removing
2085 * it from the hash queues and waiting for
2086 * it to be deleted later when it has no users
2087 */
2088
2089/**
2090 * d_delete - delete a dentry
2091 * @dentry: The dentry to delete
2092 *
2093 * Turn the dentry into a negative dentry if possible, otherwise
2094 * remove it from the hash queues so it can be deleted later
2095 */
2096
2097void d_delete(struct dentry * dentry)
2098{
Nick Piggin873feea2011-01-07 17:50:06 +11002099 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002100 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 /*
2102 * Are we the only user?
2103 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002104again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002106 inode = dentry->d_inode;
2107 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002108 if (dentry->d_count == 1) {
Nick Piggin873feea2011-01-07 17:50:06 +11002109 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002110 spin_unlock(&dentry->d_lock);
2111 cpu_relax();
2112 goto again;
2113 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002114 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002115 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002116 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 return;
2118 }
2119
2120 if (!d_unhashed(dentry))
2121 __d_drop(dentry);
2122
2123 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002124
2125 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002127EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002129static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002131 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002132 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002133 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002134 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002135 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136}
2137
David Howells770bfad2006-08-22 20:06:07 -04002138static void _d_rehash(struct dentry * entry)
2139{
2140 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2141}
2142
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143/**
2144 * d_rehash - add an entry back to the hash
2145 * @entry: dentry to add to the hash
2146 *
2147 * Adds a dentry to the hash according to its name.
2148 */
2149
2150void d_rehash(struct dentry * entry)
2151{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002153 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002156EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002158/**
2159 * dentry_update_name_case - update case insensitive dentry with a new name
2160 * @dentry: dentry to be updated
2161 * @name: new name
2162 *
2163 * Update a case insensitive dentry with new case of name.
2164 *
2165 * dentry must have been returned by d_lookup with name @name. Old and new
2166 * name lengths must match (ie. no d_compare which allows mismatched name
2167 * lengths).
2168 *
2169 * Parent inode i_mutex must be held over d_lookup and into this call (to
2170 * keep renames and concurrent inserts, and readdir(2) away).
2171 */
2172void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2173{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002174 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002175 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2176
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002177 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002178 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002179 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002180 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002181 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002182}
2183EXPORT_SYMBOL(dentry_update_name_case);
2184
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185static void switch_names(struct dentry *dentry, struct dentry *target)
2186{
2187 if (dname_external(target)) {
2188 if (dname_external(dentry)) {
2189 /*
2190 * Both external: swap the pointers
2191 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002192 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 } else {
2194 /*
2195 * dentry:internal, target:external. Steal target's
2196 * storage and make target internal.
2197 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002198 memcpy(target->d_iname, dentry->d_name.name,
2199 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 dentry->d_name.name = target->d_name.name;
2201 target->d_name.name = target->d_iname;
2202 }
2203 } else {
2204 if (dname_external(dentry)) {
2205 /*
2206 * dentry:external, target:internal. Give dentry's
2207 * storage to target and make dentry internal
2208 */
2209 memcpy(dentry->d_iname, target->d_name.name,
2210 target->d_name.len + 1);
2211 target->d_name.name = dentry->d_name.name;
2212 dentry->d_name.name = dentry->d_iname;
2213 } else {
2214 /*
2215 * Both are internal. Just copy target to dentry
2216 */
2217 memcpy(dentry->d_iname, target->d_name.name,
2218 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002219 dentry->d_name.len = target->d_name.len;
2220 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 }
2222 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002223 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224}
2225
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002226static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2227{
2228 /*
2229 * XXXX: do we really need to take target->d_lock?
2230 */
2231 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2232 spin_lock(&target->d_parent->d_lock);
2233 else {
2234 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2235 spin_lock(&dentry->d_parent->d_lock);
2236 spin_lock_nested(&target->d_parent->d_lock,
2237 DENTRY_D_LOCK_NESTED);
2238 } else {
2239 spin_lock(&target->d_parent->d_lock);
2240 spin_lock_nested(&dentry->d_parent->d_lock,
2241 DENTRY_D_LOCK_NESTED);
2242 }
2243 }
2244 if (target < dentry) {
2245 spin_lock_nested(&target->d_lock, 2);
2246 spin_lock_nested(&dentry->d_lock, 3);
2247 } else {
2248 spin_lock_nested(&dentry->d_lock, 2);
2249 spin_lock_nested(&target->d_lock, 3);
2250 }
2251}
2252
2253static void dentry_unlock_parents_for_move(struct dentry *dentry,
2254 struct dentry *target)
2255{
2256 if (target->d_parent != dentry->d_parent)
2257 spin_unlock(&dentry->d_parent->d_lock);
2258 if (target->d_parent != target)
2259 spin_unlock(&target->d_parent->d_lock);
2260}
2261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002263 * When switching names, the actual string doesn't strictly have to
2264 * be preserved in the target - because we're dropping the target
2265 * anyway. As such, we can just do a simple memcpy() to copy over
2266 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002268 * Note that we have to be a lot more careful about getting the hash
2269 * switched - we have to switch the hash value properly even if it
2270 * then no longer matches the actual (corrupted) string of the target.
2271 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002273/*
Al Viro18367502011-07-12 21:42:24 -04002274 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 * @dentry: entry to move
2276 * @target: new dentry
2277 *
2278 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002279 * dcache entries should not be moved in this way. Caller must hold
2280 * rename_lock, the i_mutex of the source and target directories,
2281 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282 */
Al Viro18367502011-07-12 21:42:24 -04002283static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 if (!dentry->d_inode)
2286 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2287
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002288 BUG_ON(d_ancestor(dentry, target));
2289 BUG_ON(d_ancestor(target, dentry));
2290
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002291 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292
Nick Piggin31e6b012011-01-07 17:49:52 +11002293 write_seqcount_begin(&dentry->d_seq);
2294 write_seqcount_begin(&target->d_seq);
2295
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002296 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2297
2298 /*
2299 * Move the dentry to the target hash queue. Don't bother checking
2300 * for the same hash queue because of how unlikely it is.
2301 */
2302 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002303 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305 /* Unhash the target: dput() will then get rid of it */
2306 __d_drop(target);
2307
Eric Dumazet5160ee62006-01-08 01:03:32 -08002308 list_del(&dentry->d_u.d_child);
2309 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
2311 /* Switch the names.. */
2312 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002313 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
2315 /* ... and switch the parents */
2316 if (IS_ROOT(dentry)) {
2317 dentry->d_parent = target->d_parent;
2318 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002319 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002321 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
2323 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002324 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 }
2326
Eric Dumazet5160ee62006-01-08 01:03:32 -08002327 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002328
Nick Piggin31e6b012011-01-07 17:49:52 +11002329 write_seqcount_end(&target->d_seq);
2330 write_seqcount_end(&dentry->d_seq);
2331
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002332 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002334 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002336}
2337
2338/*
2339 * d_move - move a dentry
2340 * @dentry: entry to move
2341 * @target: new dentry
2342 *
2343 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002344 * dcache entries should not be moved in this way. See the locking
2345 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002346 */
2347void d_move(struct dentry *dentry, struct dentry *target)
2348{
2349 write_seqlock(&rename_lock);
2350 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002352}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002353EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002355/**
2356 * d_ancestor - search for an ancestor
2357 * @p1: ancestor dentry
2358 * @p2: child dentry
2359 *
2360 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2361 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002362 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002363struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002364{
2365 struct dentry *p;
2366
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002367 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002368 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002369 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002370 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002371 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002372}
2373
2374/*
2375 * This helper attempts to cope with remotely renamed directories
2376 *
2377 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002378 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002379 *
2380 * Note: If ever the locking in lock_rename() changes, then please
2381 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002382 */
Nick Piggin873feea2011-01-07 17:50:06 +11002383static struct dentry *__d_unalias(struct inode *inode,
2384 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002385{
2386 struct mutex *m1 = NULL, *m2 = NULL;
2387 struct dentry *ret;
2388
2389 /* If alias and dentry share a parent, then no extra locks required */
2390 if (alias->d_parent == dentry->d_parent)
2391 goto out_unalias;
2392
Trond Myklebust9eaef272006-10-21 10:24:20 -07002393 /* See lock_rename() */
2394 ret = ERR_PTR(-EBUSY);
2395 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2396 goto out_err;
2397 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2398 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2399 goto out_err;
2400 m2 = &alias->d_parent->d_inode->i_mutex;
2401out_unalias:
Al Viro18367502011-07-12 21:42:24 -04002402 __d_move(alias, dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002403 ret = alias;
2404out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002405 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002406 if (m2)
2407 mutex_unlock(m2);
2408 if (m1)
2409 mutex_unlock(m1);
2410 return ret;
2411}
2412
2413/*
David Howells770bfad2006-08-22 20:06:07 -04002414 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2415 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002416 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002417 */
2418static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2419{
2420 struct dentry *dparent, *aparent;
2421
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002422 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002423
Nick Piggin31e6b012011-01-07 17:49:52 +11002424 write_seqcount_begin(&dentry->d_seq);
2425 write_seqcount_begin(&anon->d_seq);
2426
David Howells770bfad2006-08-22 20:06:07 -04002427 dparent = dentry->d_parent;
2428 aparent = anon->d_parent;
2429
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002430 switch_names(dentry, anon);
2431 swap(dentry->d_name.hash, anon->d_name.hash);
2432
David Howells770bfad2006-08-22 20:06:07 -04002433 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2434 list_del(&dentry->d_u.d_child);
2435 if (!IS_ROOT(dentry))
2436 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2437 else
2438 INIT_LIST_HEAD(&dentry->d_u.d_child);
2439
2440 anon->d_parent = (dparent == dentry) ? anon : dparent;
2441 list_del(&anon->d_u.d_child);
2442 if (!IS_ROOT(anon))
2443 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2444 else
2445 INIT_LIST_HEAD(&anon->d_u.d_child);
2446
Nick Piggin31e6b012011-01-07 17:49:52 +11002447 write_seqcount_end(&dentry->d_seq);
2448 write_seqcount_end(&anon->d_seq);
2449
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002450 dentry_unlock_parents_for_move(anon, dentry);
2451 spin_unlock(&dentry->d_lock);
2452
2453 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002454 anon->d_flags &= ~DCACHE_DISCONNECTED;
2455}
2456
2457/**
2458 * d_materialise_unique - introduce an inode into the tree
2459 * @dentry: candidate dentry
2460 * @inode: inode to bind to the dentry, to which aliases may be attached
2461 *
2462 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002463 * root directory alias in its place if there is one. Caller must hold the
2464 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002465 */
2466struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2467{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002468 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002469
2470 BUG_ON(!d_unhashed(dentry));
2471
David Howells770bfad2006-08-22 20:06:07 -04002472 if (!inode) {
2473 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002474 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002475 d_rehash(actual);
2476 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002477 }
2478
Nick Piggin873feea2011-01-07 17:50:06 +11002479 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002480
Trond Myklebust9eaef272006-10-21 10:24:20 -07002481 if (S_ISDIR(inode->i_mode)) {
2482 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002483
Trond Myklebust9eaef272006-10-21 10:24:20 -07002484 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002485 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002486 if (alias) {
2487 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002488 write_seqlock(&rename_lock);
2489
2490 if (d_ancestor(alias, dentry)) {
2491 /* Check for loops */
2492 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002493 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002494 } else if (IS_ROOT(alias)) {
2495 /* Is this an anonymous mountpoint that we
2496 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002497 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002498 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002499 __d_drop(alias);
2500 goto found;
Al Viro18367502011-07-12 21:42:24 -04002501 } else {
2502 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002503 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002504 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002505 }
Al Viro18367502011-07-12 21:42:24 -04002506 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002507 if (IS_ERR(actual)) {
2508 if (PTR_ERR(actual) == -ELOOP)
2509 pr_warn_ratelimited(
2510 "VFS: Lookup of '%s' in %s %s"
2511 " would have caused loop\n",
2512 dentry->d_name.name,
2513 inode->i_sb->s_type->name,
2514 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002515 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002516 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002517 goto out_nolock;
2518 }
David Howells770bfad2006-08-22 20:06:07 -04002519 }
2520
2521 /* Add a unique reference */
2522 actual = __d_instantiate_unique(dentry, inode);
2523 if (!actual)
2524 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002525 else
2526 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002527
David Howells770bfad2006-08-22 20:06:07 -04002528 spin_lock(&actual->d_lock);
2529found:
2530 _d_rehash(actual);
2531 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002532 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002533out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002534 if (actual == dentry) {
2535 security_d_instantiate(dentry, inode);
2536 return NULL;
2537 }
2538
2539 iput(inode);
2540 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002541}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002542EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002543
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002544static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002545{
2546 *buflen -= namelen;
2547 if (*buflen < 0)
2548 return -ENAMETOOLONG;
2549 *buffer -= namelen;
2550 memcpy(*buffer, str, namelen);
2551 return 0;
2552}
2553
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002554static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2555{
2556 return prepend(buffer, buflen, name->name, name->len);
2557}
2558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002560 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002561 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002562 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002563 * @buffer: pointer to the end of the buffer
2564 * @buflen: pointer to buffer length
2565 *
Nick Piggin949854d2011-01-07 17:49:37 +11002566 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002567 */
Al Viro02125a82011-12-05 08:43:34 -05002568static int prepend_path(const struct path *path,
2569 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002570 char **buffer, int *buflen)
2571{
2572 struct dentry *dentry = path->dentry;
2573 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002574 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002575 bool slash = false;
2576 int error = 0;
2577
Andi Kleen962830d2012-05-08 13:32:02 +09302578 br_read_lock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002579 while (dentry != root->dentry || vfsmnt != root->mnt) {
2580 struct dentry * parent;
2581
2582 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2583 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002584 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002585 goto global_root;
Al Viroa73324d2011-11-24 22:25:07 -05002586 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002587 mnt = mnt->mnt_parent;
2588 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002589 continue;
2590 }
2591 parent = dentry->d_parent;
2592 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002593 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002594 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002595 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002596 if (!error)
2597 error = prepend(buffer, buflen, "/", 1);
2598 if (error)
2599 break;
2600
2601 slash = true;
2602 dentry = parent;
2603 }
2604
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002605 if (!error && !slash)
2606 error = prepend(buffer, buflen, "/", 1);
2607
Al Viro02125a82011-12-05 08:43:34 -05002608out:
Andi Kleen962830d2012-05-08 13:32:02 +09302609 br_read_unlock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002610 return error;
2611
2612global_root:
2613 /*
2614 * Filesystems needing to implement special "root names"
2615 * should do so with ->d_dname()
2616 */
2617 if (IS_ROOT(dentry) &&
2618 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2619 WARN(1, "Root dentry has weird name <%.*s>\n",
2620 (int) dentry->d_name.len, dentry->d_name.name);
2621 }
Al Viro02125a82011-12-05 08:43:34 -05002622 if (!slash)
2623 error = prepend(buffer, buflen, "/", 1);
2624 if (!error)
Al Viro143c8c92011-11-25 00:46:35 -05002625 error = real_mount(vfsmnt)->mnt_ns ? 1 : 2;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002626 goto out;
2627}
2628
2629/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002630 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002631 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002632 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002633 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 * @buflen: buffer length
2635 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002636 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002638 * Returns a pointer into the buffer or an error code if the
2639 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002640 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002641 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002642 *
Al Viro02125a82011-12-05 08:43:34 -05002643 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 */
Al Viro02125a82011-12-05 08:43:34 -05002645char *__d_path(const struct path *path,
2646 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002647 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002649 char *res = buf + buflen;
2650 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002652 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002653 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002654 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002655 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002656
Al Viro02125a82011-12-05 08:43:34 -05002657 if (error < 0)
2658 return ERR_PTR(error);
2659 if (error > 0)
2660 return NULL;
2661 return res;
2662}
2663
2664char *d_absolute_path(const struct path *path,
2665 char *buf, int buflen)
2666{
2667 struct path root = {};
2668 char *res = buf + buflen;
2669 int error;
2670
2671 prepend(&res, &buflen, "\0", 1);
2672 write_seqlock(&rename_lock);
2673 error = prepend_path(path, &root, &res, &buflen);
2674 write_sequnlock(&rename_lock);
2675
2676 if (error > 1)
2677 error = -EINVAL;
2678 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002679 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002680 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681}
2682
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002683/*
2684 * same as __d_path but appends "(deleted)" for unlinked files.
2685 */
Al Viro02125a82011-12-05 08:43:34 -05002686static int path_with_deleted(const struct path *path,
2687 const struct path *root,
2688 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002689{
2690 prepend(buf, buflen, "\0", 1);
2691 if (d_unlinked(path->dentry)) {
2692 int error = prepend(buf, buflen, " (deleted)", 10);
2693 if (error)
2694 return error;
2695 }
2696
2697 return prepend_path(path, root, buf, buflen);
2698}
2699
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002700static int prepend_unreachable(char **buffer, int *buflen)
2701{
2702 return prepend(buffer, buflen, "(unreachable)", 13);
2703}
2704
Jan Bluncka03a8a702008-02-14 19:38:32 -08002705/**
2706 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002707 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002708 * @buf: buffer to return value in
2709 * @buflen: buffer length
2710 *
2711 * Convert a dentry into an ASCII path name. If the entry has been deleted
2712 * the string " (deleted)" is appended. Note that this is ambiguous.
2713 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002714 * Returns a pointer into the buffer or an error code if the path was
2715 * too long. Note: Callers should use the returned pointer, not the passed
2716 * in buffer, to use the name! The implementation often starts at an offset
2717 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002718 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002719 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002720 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002721char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002723 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002724 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002725 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002727 /*
2728 * We have various synthetic filesystems that never get mounted. On
2729 * these filesystems dentries are never used for lookup purposes, and
2730 * thus don't need to be hashed. They also don't need a name until a
2731 * user wants to identify the object in /proc/pid/fd/. The little hack
2732 * below allows us to generate a name for these objects on demand:
2733 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002734 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2735 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002736
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002737 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002738 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002739 error = path_with_deleted(path, &root, &res, &buflen);
2740 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002741 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002742 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002743 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 return res;
2745}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002746EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002748/**
2749 * d_path_with_unreachable - return the path of a dentry
2750 * @path: path to report
2751 * @buf: buffer to return value in
2752 * @buflen: buffer length
2753 *
2754 * The difference from d_path() is that this prepends "(unreachable)"
2755 * to paths which are unreachable from the current process' root.
2756 */
2757char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2758{
2759 char *res = buf + buflen;
2760 struct path root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002761 int error;
2762
2763 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2764 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2765
2766 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002767 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002768 error = path_with_deleted(path, &root, &res, &buflen);
2769 if (error > 0)
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002770 error = prepend_unreachable(&res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002771 write_sequnlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002772 path_put(&root);
2773 if (error)
2774 res = ERR_PTR(error);
2775
2776 return res;
2777}
2778
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002780 * Helper function for dentry_operations.d_dname() members
2781 */
2782char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2783 const char *fmt, ...)
2784{
2785 va_list args;
2786 char temp[64];
2787 int sz;
2788
2789 va_start(args, fmt);
2790 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2791 va_end(args);
2792
2793 if (sz > sizeof(temp) || sz > buflen)
2794 return ERR_PTR(-ENAMETOOLONG);
2795
2796 buffer += buflen - sz;
2797 return memcpy(buffer, temp, sz);
2798}
2799
2800/*
Ram Pai6092d042008-03-27 13:06:20 +01002801 * Write full pathname from the root of the filesystem into the buffer.
2802 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002803static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002804{
2805 char *end = buf + buflen;
2806 char *retval;
2807
Ram Pai6092d042008-03-27 13:06:20 +01002808 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002809 if (buflen < 1)
2810 goto Elong;
2811 /* Get '/' right */
2812 retval = end-1;
2813 *retval = '/';
2814
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002815 while (!IS_ROOT(dentry)) {
2816 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002817 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002818
Ram Pai6092d042008-03-27 13:06:20 +01002819 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002820 spin_lock(&dentry->d_lock);
2821 error = prepend_name(&end, &buflen, &dentry->d_name);
2822 spin_unlock(&dentry->d_lock);
2823 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002824 goto Elong;
2825
2826 retval = end;
2827 dentry = parent;
2828 }
Al Viroc1031352010-06-06 22:31:14 -04002829 return retval;
2830Elong:
2831 return ERR_PTR(-ENAMETOOLONG);
2832}
Nick Pigginec2447c2011-01-07 17:49:29 +11002833
2834char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2835{
2836 char *retval;
2837
Nick Piggin949854d2011-01-07 17:49:37 +11002838 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002839 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002840 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002841
2842 return retval;
2843}
2844EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002845
2846char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2847{
2848 char *p = NULL;
2849 char *retval;
2850
Nick Piggin949854d2011-01-07 17:49:37 +11002851 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002852 if (d_unlinked(dentry)) {
2853 p = buf + buflen;
2854 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2855 goto Elong;
2856 buflen++;
2857 }
2858 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002859 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002860 if (!IS_ERR(retval) && p)
2861 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002862 return retval;
2863Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002864 return ERR_PTR(-ENAMETOOLONG);
2865}
2866
2867/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 * NOTE! The user-level library version returns a
2869 * character pointer. The kernel system call just
2870 * returns the length of the buffer filled (which
2871 * includes the ending '\0' character), or a negative
2872 * error value. So libc would do something like
2873 *
2874 * char *getcwd(char * buf, size_t size)
2875 * {
2876 * int retval;
2877 *
2878 * retval = sys_getcwd(buf, size);
2879 * if (retval >= 0)
2880 * return buf;
2881 * errno = -retval;
2882 * return NULL;
2883 * }
2884 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002885SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886{
Linus Torvalds552ce542007-02-13 12:08:18 -08002887 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002888 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002889 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
2891 if (!page)
2892 return -ENOMEM;
2893
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002894 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895
Linus Torvalds552ce542007-02-13 12:08:18 -08002896 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002897 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002898 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002899 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002900 char *cwd = page + PAGE_SIZE;
2901 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002903 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002904 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002905 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002906
Al Viro02125a82011-12-05 08:43:34 -05002907 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002908 goto out;
2909
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002910 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002911 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002912 error = prepend_unreachable(&cwd, &buflen);
2913 if (error)
2914 goto out;
2915 }
2916
Linus Torvalds552ce542007-02-13 12:08:18 -08002917 error = -ERANGE;
2918 len = PAGE_SIZE + page - cwd;
2919 if (len <= size) {
2920 error = len;
2921 if (copy_to_user(buf, cwd, len))
2922 error = -EFAULT;
2923 }
Nick Piggin949854d2011-01-07 17:49:37 +11002924 } else {
2925 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927
2928out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002929 path_put(&pwd);
2930 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 free_page((unsigned long) page);
2932 return error;
2933}
2934
2935/*
2936 * Test whether new_dentry is a subdirectory of old_dentry.
2937 *
2938 * Trivially implemented using the dcache structure
2939 */
2940
2941/**
2942 * is_subdir - is new dentry a subdirectory of old_dentry
2943 * @new_dentry: new dentry
2944 * @old_dentry: old dentry
2945 *
2946 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2947 * Returns 0 otherwise.
2948 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2949 */
2950
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002951int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952{
2953 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002954 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002956 if (new_dentry == old_dentry)
2957 return 1;
2958
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002959 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002962 /*
2963 * Need rcu_readlock to protect against the d_parent trashing
2964 * due to d_move
2965 */
2966 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002967 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002969 else
2970 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002971 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973
2974 return result;
2975}
2976
2977void d_genocide(struct dentry *root)
2978{
Nick Piggin949854d2011-01-07 17:49:37 +11002979 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002981 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002982 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983
Nick Piggin949854d2011-01-07 17:49:37 +11002984 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002985again:
2986 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002987 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988repeat:
2989 next = this_parent->d_subdirs.next;
2990resume:
2991 while (next != &this_parent->d_subdirs) {
2992 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002993 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002995
Nick Pigginda502952011-01-07 17:49:33 +11002996 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2997 if (d_unhashed(dentry) || !dentry->d_inode) {
2998 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 continue;
Nick Pigginda502952011-01-07 17:49:33 +11003000 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003002 spin_unlock(&this_parent->d_lock);
3003 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003005 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 goto repeat;
3007 }
Nick Piggin949854d2011-01-07 17:49:37 +11003008 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3009 dentry->d_flags |= DCACHE_GENOCIDE;
3010 dentry->d_count--;
3011 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11003012 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 }
3014 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07003015 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11003016 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
3017 this_parent->d_flags |= DCACHE_GENOCIDE;
3018 this_parent->d_count--;
3019 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07003020 this_parent = try_to_ascend(this_parent, locked, seq);
3021 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11003022 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11003023 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 goto resume;
3025 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003026 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11003027 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11003028 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11003029 if (locked)
3030 write_sequnlock(&rename_lock);
3031 return;
3032
3033rename_retry:
3034 locked = 1;
3035 write_seqlock(&rename_lock);
3036 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037}
3038
3039/**
3040 * find_inode_number - check for dentry with name
3041 * @dir: directory to check
3042 * @name: Name to find.
3043 *
3044 * Check whether a dentry already exists for the given name,
3045 * and return the inode number if it has an inode. Otherwise
3046 * 0 is returned.
3047 *
3048 * This routine is used to post-process directory listings for
3049 * filesystems using synthetic inode numbers, and is necessary
3050 * to keep getcwd() working.
3051 */
3052
3053ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3054{
3055 struct dentry * dentry;
3056 ino_t ino = 0;
3057
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08003058 dentry = d_hash_and_lookup(dir, name);
3059 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 if (dentry->d_inode)
3061 ino = dentry->d_inode->i_ino;
3062 dput(dentry);
3063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 return ino;
3065}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003066EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067
3068static __initdata unsigned long dhash_entries;
3069static int __init set_dhash_entries(char *str)
3070{
3071 if (!str)
3072 return 0;
3073 dhash_entries = simple_strtoul(str, &str, 0);
3074 return 1;
3075}
3076__setup("dhash_entries=", set_dhash_entries);
3077
3078static void __init dcache_init_early(void)
3079{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003080 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081
3082 /* If hashes are distributed across NUMA nodes, defer
3083 * hash allocation until vmalloc space is available.
3084 */
3085 if (hashdist)
3086 return;
3087
3088 dentry_hashtable =
3089 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003090 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 dhash_entries,
3092 13,
3093 HASH_EARLY,
3094 &d_hash_shift,
3095 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003096 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 0);
3098
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003099 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003100 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101}
3102
Denis Cheng74bf17c2007-10-16 23:26:30 -07003103static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003105 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106
3107 /*
3108 * A constructor could be added for stable state like the lists,
3109 * but it is probably not worth it because of the cache nature
3110 * of the dcache.
3111 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003112 dentry_cache = KMEM_CACHE(dentry,
3113 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114
3115 /* Hash may have been set up in dcache_init_early */
3116 if (!hashdist)
3117 return;
3118
3119 dentry_hashtable =
3120 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003121 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 dhash_entries,
3123 13,
3124 0,
3125 &d_hash_shift,
3126 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003127 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003128 0);
3129
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003130 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003131 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132}
3133
3134/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003135struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003136EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138EXPORT_SYMBOL(d_genocide);
3139
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140void __init vfs_caches_init_early(void)
3141{
3142 dcache_init_early();
3143 inode_init_early();
3144}
3145
3146void __init vfs_caches_init(unsigned long mempages)
3147{
3148 unsigned long reserve;
3149
3150 /* Base hash sizes on available memory, with a reserve equal to
3151 150% of current kernel size */
3152
3153 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3154 mempages -= reserve;
3155
3156 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003157 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158
Denis Cheng74bf17c2007-10-16 23:26:30 -07003159 dcache_init();
3160 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003162 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 bdev_cache_init();
3164 chrdev_init();
3165}