blob: 8086636bf796ab328219d4a898608af1be2fdb2b [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 Howells07f3f05c2006-09-30 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
Al Virob3d9b7a2012-06-09 13:51:19 -0400221 WARN_ON(!hlist_unhashed(&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;
Al Virob3d9b7a2012-06-09 13:51:19 -0400270 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 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;
Al Virob3d9b7a2012-06-09 13:51:19 -0400294 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100295 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;
Al Virob3d9b7a2012-06-09 13:51:19 -0400702 struct hlist_node *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Nick Pigginda502952011-01-07 17:49:33 +1100704again:
705 discon_alias = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400706 hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100707 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700709 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100710 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700712 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100713 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100714 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 return alias;
716 }
717 }
Nick Pigginda502952011-01-07 17:49:33 +1100718 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
Nick Pigginda502952011-01-07 17:49:33 +1100720 if (discon_alias) {
721 alias = discon_alias;
722 spin_lock(&alias->d_lock);
723 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
724 if (IS_ROOT(alias) &&
725 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100726 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100727 spin_unlock(&alias->d_lock);
728 return alias;
729 }
730 }
731 spin_unlock(&alias->d_lock);
732 goto again;
733 }
734 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
736
Nick Pigginda502952011-01-07 17:49:33 +1100737struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738{
David Howells214fda12006-03-25 03:06:36 -0800739 struct dentry *de = NULL;
740
Al Virob3d9b7a2012-06-09 13:51:19 -0400741 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100742 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700743 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100744 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 return de;
747}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700748EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750/*
751 * Try to kill dentries associated with this inode.
752 * WARNING: you must own a reference to inode.
753 */
754void d_prune_aliases(struct inode *inode)
755{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700756 struct dentry *dentry;
Al Virob3d9b7a2012-06-09 13:51:19 -0400757 struct hlist_node *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100759 spin_lock(&inode->i_lock);
Al Virob3d9b7a2012-06-09 13:51:19 -0400760 hlist_for_each_entry(dentry, p, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100762 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100763 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 __d_drop(dentry);
765 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100766 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 dput(dentry);
768 goto restart;
769 }
770 spin_unlock(&dentry->d_lock);
771 }
Nick Piggin873feea2011-01-07 17:50:06 +1100772 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700774EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776/*
Nick Piggin77812a12011-01-07 17:49:48 +1100777 * Try to throw away a dentry - free the inode, dput the parent.
778 * Requires dentry->d_lock is held, and dentry->d_count == 0.
779 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700780 *
Nick Piggin77812a12011-01-07 17:49:48 +1100781 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
Nick Piggin77812a12011-01-07 17:49:48 +1100783static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200784 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
Nick Piggin77812a12011-01-07 17:49:48 +1100786 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700787
Nick Piggin77812a12011-01-07 17:49:48 +1100788 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700789 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100790 * If dentry_kill returns NULL, we have nothing more to do.
791 * if it returns the same dentry, trylocks failed. In either
792 * case, just loop again.
793 *
794 * Otherwise, we need to prune ancestors too. This is necessary
795 * to prevent quadratic behavior of shrink_dcache_parent(), but
796 * is also expected to be beneficial in reducing dentry cache
797 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700798 */
Nick Piggin77812a12011-01-07 17:49:48 +1100799 if (!parent)
800 return;
801 if (parent == dentry)
802 return;
803
804 /* Prune ancestors. */
805 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700806 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100807 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100808 if (dentry->d_count > 1) {
809 dentry->d_count--;
810 spin_unlock(&dentry->d_lock);
811 return;
812 }
Nick Piggin77812a12011-01-07 17:49:48 +1100813 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700814 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815}
816
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400817static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700819 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700820
Nick Pigginec336792011-01-07 17:49:47 +1100821 rcu_read_lock();
822 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100823 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
824 if (&dentry->d_lru == list)
825 break; /* empty */
826 spin_lock(&dentry->d_lock);
827 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
828 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100829 continue;
830 }
831
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 /*
833 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700834 * the LRU because of laziness during lookup. Do not free
835 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100837 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100838 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700839 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 continue;
841 }
Nick Pigginec336792011-01-07 17:49:47 +1100842
Nick Pigginec336792011-01-07 17:49:47 +1100843 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100844
845 try_prune_one_dentry(dentry);
846
Nick Pigginec336792011-01-07 17:49:47 +1100847 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 }
Nick Pigginec336792011-01-07 17:49:47 +1100849 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400850}
851
852/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000853 * prune_dcache_sb - shrink the dcache
854 * @sb: superblock
855 * @count: number of entries to try to free
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400856 *
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000857 * Attempt to shrink the superblock dcache LRU by @count entries. This is
858 * done when we need more memory an called from the superblock shrinker
859 * function.
860 *
861 * This function may fail to free any resources if all the dentries are in
862 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400863 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000864void prune_dcache_sb(struct super_block *sb, int count)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400865{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400866 struct dentry *dentry;
867 LIST_HEAD(referenced);
868 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400869
Nick Piggin23044502011-01-07 17:49:31 +1100870relock:
871 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400872 while (!list_empty(&sb->s_dentry_lru)) {
873 dentry = list_entry(sb->s_dentry_lru.prev,
874 struct dentry, d_lru);
875 BUG_ON(dentry->d_sb != sb);
876
Nick Piggin23044502011-01-07 17:49:31 +1100877 if (!spin_trylock(&dentry->d_lock)) {
878 spin_unlock(&dcache_lru_lock);
879 cpu_relax();
880 goto relock;
881 }
882
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000883 if (dentry->d_flags & DCACHE_REFERENCED) {
Nick Piggin23044502011-01-07 17:49:31 +1100884 dentry->d_flags &= ~DCACHE_REFERENCED;
885 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400886 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100887 } else {
888 list_move_tail(&dentry->d_lru, &tmp);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100889 dentry->d_flags |= DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100890 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000891 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100892 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400893 }
Nick Pigginec336792011-01-07 17:49:47 +1100894 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400895 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700896 if (!list_empty(&referenced))
897 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100898 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100899
900 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901}
902
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700903/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 * shrink_dcache_sb - shrink dcache for a superblock
905 * @sb: superblock
906 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400907 * Shrink the dcache for the specified super block. This is used to free
908 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400910void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400912 LIST_HEAD(tmp);
913
Nick Piggin23044502011-01-07 17:49:31 +1100914 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400915 while (!list_empty(&sb->s_dentry_lru)) {
916 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100917 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400918 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100919 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400920 }
Nick Piggin23044502011-01-07 17:49:31 +1100921 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700923EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
925/*
David Howellsc636ebd2006-10-11 01:22:19 -0700926 * destroy a single subtree of dentries for unmount
927 * - see the comments on shrink_dcache_for_umount() for a description of the
928 * locking
929 */
930static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
931{
932 struct dentry *parent;
933
934 BUG_ON(!IS_ROOT(dentry));
935
David Howellsc636ebd2006-10-11 01:22:19 -0700936 for (;;) {
937 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100938 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700939 dentry = list_entry(dentry->d_subdirs.next,
940 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700941
942 /* consume the dentries from this leaf up through its parents
943 * until we find one with children or run out altogether */
944 do {
945 struct inode *inode;
946
Sage Weilf0023bc2011-10-28 10:02:42 -0700947 /*
948 * remove the dentry from the lru, and inform
949 * the fs that this dentry is about to be
950 * unhashed and destroyed.
951 */
952 dentry_lru_prune(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100953 __d_shrink(dentry);
954
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100955 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700956 printk(KERN_ERR
957 "BUG: Dentry %p{i=%lx,n=%s}"
958 " still in use (%d)"
959 " [unmount of %s %s]\n",
960 dentry,
961 dentry->d_inode ?
962 dentry->d_inode->i_ino : 0UL,
963 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100964 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700965 dentry->d_sb->s_type->name,
966 dentry->d_sb->s_id);
967 BUG();
968 }
969
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100970 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700971 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100972 list_del(&dentry->d_u.d_child);
973 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900974 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100975 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100976 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900977 }
David Howellsc636ebd2006-10-11 01:22:19 -0700978
David Howellsc636ebd2006-10-11 01:22:19 -0700979 inode = dentry->d_inode;
980 if (inode) {
981 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400982 hlist_del_init(&dentry->d_alias);
David Howellsc636ebd2006-10-11 01:22:19 -0700983 if (dentry->d_op && dentry->d_op->d_iput)
984 dentry->d_op->d_iput(dentry, inode);
985 else
986 iput(inode);
987 }
988
989 d_free(dentry);
990
991 /* finished when we fall off the top of the tree,
992 * otherwise we ascend to the parent and move to the
993 * next sibling if there is one */
994 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400995 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700996 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700997 } while (list_empty(&dentry->d_subdirs));
998
999 dentry = list_entry(dentry->d_subdirs.next,
1000 struct dentry, d_u.d_child);
1001 }
1002}
1003
1004/*
1005 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001006 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -07001007 * - the superblock is detached from all mountings and open files, so the
1008 * dentry trees will not be rearranged by the VFS
1009 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1010 * any dentries belonging to this superblock that it comes across
1011 * - the filesystem itself is no longer permitted to rearrange the dentries
1012 * in this superblock
1013 */
1014void shrink_dcache_for_umount(struct super_block *sb)
1015{
1016 struct dentry *dentry;
1017
1018 if (down_read_trylock(&sb->s_umount))
1019 BUG();
1020
1021 dentry = sb->s_root;
1022 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001023 dentry->d_count--;
David Howellsc636ebd2006-10-11 01:22:19 -07001024 shrink_dcache_for_umount_subtree(dentry);
1025
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001026 while (!hlist_bl_empty(&sb->s_anon)) {
1027 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001028 shrink_dcache_for_umount_subtree(dentry);
1029 }
1030}
1031
1032/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001033 * This tries to ascend one level of parenthood, but
1034 * we can race with renaming, so we need to re-check
1035 * the parenthood after dropping the lock and check
1036 * that the sequence number still matches.
1037 */
1038static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1039{
1040 struct dentry *new = old->d_parent;
1041
1042 rcu_read_lock();
1043 spin_unlock(&old->d_lock);
1044 spin_lock(&new->d_lock);
1045
1046 /*
1047 * might go back up the wrong parent if we have had a rename
1048 * or deletion
1049 */
1050 if (new != old->d_parent ||
Trond Myklebustc83ce982011-03-15 13:36:43 -04001051 (old->d_flags & DCACHE_DISCONNECTED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001052 (!locked && read_seqretry(&rename_lock, seq))) {
1053 spin_unlock(&new->d_lock);
1054 new = NULL;
1055 }
1056 rcu_read_unlock();
1057 return new;
1058}
1059
1060
1061/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 * Search for at least 1 mount point in the dentry's subdirs.
1063 * We descend to the next level whenever the d_subdirs
1064 * list is non-empty and continue searching.
1065 */
1066
1067/**
1068 * have_submounts - check for mounts over a dentry
1069 * @parent: dentry to check.
1070 *
1071 * Return true if the parent or its subdirectories contain
1072 * a mount point
1073 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074int have_submounts(struct dentry *parent)
1075{
Nick Piggin949854d2011-01-07 17:49:37 +11001076 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001078 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001079 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001080
Nick Piggin949854d2011-01-07 17:49:37 +11001081 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001082again:
1083 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 if (d_mountpoint(parent))
1086 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001087 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088repeat:
1089 next = this_parent->d_subdirs.next;
1090resume:
1091 while (next != &this_parent->d_subdirs) {
1092 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001093 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001095
1096 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001098 if (d_mountpoint(dentry)) {
1099 spin_unlock(&dentry->d_lock);
1100 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001104 spin_unlock(&this_parent->d_lock);
1105 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001107 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 goto repeat;
1109 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001110 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 }
1112 /*
1113 * All done at this level ... ascend and resume the search.
1114 */
1115 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001116 struct dentry *child = this_parent;
1117 this_parent = try_to_ascend(this_parent, locked, seq);
1118 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001119 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001120 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 goto resume;
1122 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001123 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001124 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001125 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001126 if (locked)
1127 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 return 0; /* No mount points found in tree */
1129positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001130 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001131 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001132 if (locked)
1133 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001135
1136rename_retry:
1137 locked = 1;
1138 write_seqlock(&rename_lock);
1139 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001141EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
1143/*
1144 * Search the dentry child list for the specified parent,
1145 * and move any unused dentries to the end of the unused
1146 * list for prune_dcache(). We descend to the next level
1147 * whenever the d_subdirs list is non-empty and continue
1148 * searching.
1149 *
1150 * It returns zero iff there are no unused children,
1151 * otherwise it returns the number of children moved to
1152 * the end of the unused list. This may not be the total
1153 * number of unused children, because select_parent can
1154 * drop the lock and return early due to latency
1155 * constraints.
1156 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001157static int select_parent(struct dentry *parent, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
Nick Piggin949854d2011-01-07 17:49:37 +11001159 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001161 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001163 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
Nick Piggin949854d2011-01-07 17:49:37 +11001165 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001166again:
1167 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001168 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169repeat:
1170 next = this_parent->d_subdirs.next;
1171resume:
1172 while (next != &this_parent->d_subdirs) {
1173 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001174 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 next = tmp->next;
1176
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001177 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001178
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001179 /*
1180 * move only zero ref count dentries to the dispose list.
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001181 *
1182 * Those which are presently on the shrink list, being processed
1183 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1184 * loop in shrink_dcache_parent() might not make any progress
1185 * and loop forever.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 */
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001187 if (dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001188 dentry_lru_del(dentry);
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001189 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1190 dentry_lru_move_list(dentry, dispose);
1191 dentry->d_flags |= DCACHE_SHRINK_LIST;
1192 found++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 /*
1195 * We can return to the caller if we have found some (this
1196 * ensures forward progress). We'll be coming back to find
1197 * the rest.
1198 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001199 if (found && need_resched()) {
1200 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 /*
1205 * Descend a level if the d_subdirs list is non-empty.
1206 */
1207 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001208 spin_unlock(&this_parent->d_lock);
1209 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001211 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 goto repeat;
1213 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001214
1215 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
1217 /*
1218 * All done at this level ... ascend and resume the search.
1219 */
1220 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001221 struct dentry *child = this_parent;
1222 this_parent = try_to_ascend(this_parent, locked, seq);
1223 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001224 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001225 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 goto resume;
1227 }
1228out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001229 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001230 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001231 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001232 if (locked)
1233 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001235
1236rename_retry:
1237 if (found)
1238 return found;
1239 locked = 1;
1240 write_seqlock(&rename_lock);
1241 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242}
1243
1244/**
1245 * shrink_dcache_parent - prune dcache
1246 * @parent: parent of entries to prune
1247 *
1248 * Prune the dcache to remove unused children of the parent dentry.
1249 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250void shrink_dcache_parent(struct dentry * parent)
1251{
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001252 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 int found;
1254
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001255 while ((found = select_parent(parent, &dispose)) != 0)
1256 shrink_dentry_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001258EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260/**
Al Viroa4464db2011-07-07 15:03:58 -04001261 * __d_alloc - allocate a dcache entry
1262 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 * @name: qstr of the name
1264 *
1265 * Allocates a dentry. It returns %NULL if there is insufficient memory
1266 * available. On a success the dentry is returned. The name passed in is
1267 * copied and the copy passed in may be reused after this call.
1268 */
1269
Al Viroa4464db2011-07-07 15:03:58 -04001270struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271{
1272 struct dentry *dentry;
1273 char *dname;
1274
Mel Gormane12ba742007-10-16 01:25:52 -07001275 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 if (!dentry)
1277 return NULL;
1278
Linus Torvalds6326c712012-05-21 16:14:04 -07001279 /*
1280 * We guarantee that the inline name is always NUL-terminated.
1281 * This way the memcpy() done by the name switching in rename
1282 * will still always have a NUL at the end, even if we might
1283 * be overwriting an internal NUL character
1284 */
1285 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (name->len > DNAME_INLINE_LEN-1) {
1287 dname = kmalloc(name->len + 1, GFP_KERNEL);
1288 if (!dname) {
1289 kmem_cache_free(dentry_cache, dentry);
1290 return NULL;
1291 }
1292 } else {
1293 dname = dentry->d_iname;
1294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 dentry->d_name.len = name->len;
1297 dentry->d_name.hash = name->hash;
1298 memcpy(dname, name->name, name->len);
1299 dname[name->len] = 0;
1300
Linus Torvalds6326c712012-05-21 16:14:04 -07001301 /* Make sure we always see the terminating NUL character */
1302 smp_wmb();
1303 dentry->d_name.name = dname;
1304
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001305 dentry->d_count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001306 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001308 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001310 dentry->d_parent = dentry;
1311 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 dentry->d_op = NULL;
1313 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001314 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 INIT_LIST_HEAD(&dentry->d_lru);
1316 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001317 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001318 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001319 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Nick Piggin3e880fb2011-01-07 17:49:19 +11001321 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 return dentry;
1324}
Al Viroa4464db2011-07-07 15:03:58 -04001325
1326/**
1327 * d_alloc - allocate a dcache entry
1328 * @parent: parent of entry to allocate
1329 * @name: qstr of the name
1330 *
1331 * Allocates a dentry. It returns %NULL if there is insufficient memory
1332 * available. On a success the dentry is returned. The name passed in is
1333 * copied and the copy passed in may be reused after this call.
1334 */
1335struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1336{
1337 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1338 if (!dentry)
1339 return NULL;
1340
1341 spin_lock(&parent->d_lock);
1342 /*
1343 * don't need child lock because it is not subject
1344 * to concurrency here
1345 */
1346 __dget_dlock(parent);
1347 dentry->d_parent = parent;
1348 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1349 spin_unlock(&parent->d_lock);
1350
1351 return dentry;
1352}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001353EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Nick Piggin4b936882011-01-07 17:50:07 +11001355struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1356{
Al Viroa4464db2011-07-07 15:03:58 -04001357 struct dentry *dentry = __d_alloc(sb, name);
1358 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001359 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001360 return dentry;
1361}
1362EXPORT_SYMBOL(d_alloc_pseudo);
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1365{
1366 struct qstr q;
1367
1368 q.name = name;
1369 q.len = strlen(name);
1370 q.hash = full_name_hash(q.name, q.len);
1371 return d_alloc(parent, &q);
1372}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001373EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
Nick Pigginfb045ad2011-01-07 17:49:55 +11001375void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1376{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001377 WARN_ON_ONCE(dentry->d_op);
1378 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001379 DCACHE_OP_COMPARE |
1380 DCACHE_OP_REVALIDATE |
1381 DCACHE_OP_DELETE ));
1382 dentry->d_op = op;
1383 if (!op)
1384 return;
1385 if (op->d_hash)
1386 dentry->d_flags |= DCACHE_OP_HASH;
1387 if (op->d_compare)
1388 dentry->d_flags |= DCACHE_OP_COMPARE;
1389 if (op->d_revalidate)
1390 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1391 if (op->d_delete)
1392 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001393 if (op->d_prune)
1394 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001395
1396}
1397EXPORT_SYMBOL(d_set_d_op);
1398
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001399static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1400{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001401 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001402 if (inode) {
1403 if (unlikely(IS_AUTOMOUNT(inode)))
1404 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
Al Virob3d9b7a2012-06-09 13:51:19 -04001405 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001406 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001407 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001408 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001409 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001410 fsnotify_d_instantiate(dentry, inode);
1411}
1412
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413/**
1414 * d_instantiate - fill in inode information for a dentry
1415 * @entry: dentry to complete
1416 * @inode: inode to attach to this dentry
1417 *
1418 * Fill in inode information in the entry.
1419 *
1420 * This turns negative dentries into productive full members
1421 * of society.
1422 *
1423 * NOTE! This assumes that the inode count has been incremented
1424 * (or otherwise set) by the caller to indicate that it is now
1425 * in use by the dcache.
1426 */
1427
1428void d_instantiate(struct dentry *entry, struct inode * inode)
1429{
Al Virob3d9b7a2012-06-09 13:51:19 -04001430 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001431 if (inode)
1432 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001433 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001434 if (inode)
1435 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 security_d_instantiate(entry, inode);
1437}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001438EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
1440/**
1441 * d_instantiate_unique - instantiate a non-aliased dentry
1442 * @entry: dentry to instantiate
1443 * @inode: inode to attach to this dentry
1444 *
1445 * Fill in inode information in the entry. On success, it returns NULL.
1446 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001447 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 *
1449 * Note that in order to avoid conflicts with rename() etc, the caller
1450 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001451 *
1452 * This also assumes that the inode count has been incremented
1453 * (or otherwise set) by the caller to indicate that it is now
1454 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 */
David Howells770bfad2006-08-22 20:06:07 -04001456static struct dentry *__d_instantiate_unique(struct dentry *entry,
1457 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
1459 struct dentry *alias;
1460 int len = entry->d_name.len;
1461 const char *name = entry->d_name.name;
1462 unsigned int hash = entry->d_name.hash;
Al Virob3d9b7a2012-06-09 13:51:19 -04001463 struct hlist_node *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
David Howells770bfad2006-08-22 20:06:07 -04001465 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001466 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001467 return NULL;
1468 }
1469
Al Virob3d9b7a2012-06-09 13:51:19 -04001470 hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001471 /*
1472 * Don't need alias->d_lock here, because aliases with
1473 * d_parent == entry->d_parent are not subject to name or
1474 * parent changes, because the parent inode i_mutex is held.
1475 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001476 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 continue;
1478 if (alias->d_parent != entry->d_parent)
1479 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001480 if (alias->d_name.len != len)
1481 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001482 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001484 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 return alias;
1486 }
David Howells770bfad2006-08-22 20:06:07 -04001487
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001488 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 return NULL;
1490}
David Howells770bfad2006-08-22 20:06:07 -04001491
1492struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1493{
1494 struct dentry *result;
1495
Al Virob3d9b7a2012-06-09 13:51:19 -04001496 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001497
Nick Piggin873feea2011-01-07 17:50:06 +11001498 if (inode)
1499 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001500 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001501 if (inode)
1502 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001503
1504 if (!result) {
1505 security_d_instantiate(entry, inode);
1506 return NULL;
1507 }
1508
1509 BUG_ON(!d_unhashed(result));
1510 iput(inode);
1511 return result;
1512}
1513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514EXPORT_SYMBOL(d_instantiate_unique);
1515
Al Viroadc0e912012-01-08 16:49:21 -05001516struct dentry *d_make_root(struct inode *root_inode)
1517{
1518 struct dentry *res = NULL;
1519
1520 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001521 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001522
1523 res = __d_alloc(root_inode->i_sb, &name);
1524 if (res)
1525 d_instantiate(res, root_inode);
1526 else
1527 iput(root_inode);
1528 }
1529 return res;
1530}
1531EXPORT_SYMBOL(d_make_root);
1532
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001533static struct dentry * __d_find_any_alias(struct inode *inode)
1534{
1535 struct dentry *alias;
1536
Al Virob3d9b7a2012-06-09 13:51:19 -04001537 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001538 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001539 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001540 __dget(alias);
1541 return alias;
1542}
1543
Sage Weil46f72b32012-01-10 09:04:37 -08001544/**
1545 * d_find_any_alias - find any alias for a given inode
1546 * @inode: inode to find an alias for
1547 *
1548 * If any aliases exist for the given inode, take and return a
1549 * reference for one of them. If no aliases exist, return %NULL.
1550 */
1551struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001552{
1553 struct dentry *de;
1554
1555 spin_lock(&inode->i_lock);
1556 de = __d_find_any_alias(inode);
1557 spin_unlock(&inode->i_lock);
1558 return de;
1559}
Sage Weil46f72b32012-01-10 09:04:37 -08001560EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001563 * d_obtain_alias - find or allocate a dentry for a given inode
1564 * @inode: inode to allocate the dentry for
1565 *
1566 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1567 * similar open by handle operations. The returned dentry may be anonymous,
1568 * or may have a full name (if the inode was already in the cache).
1569 *
1570 * When called on a directory inode, we must ensure that the inode only ever
1571 * has one dentry. If a dentry is found, that is returned instead of
1572 * allocating a new one.
1573 *
1574 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001575 * to the dentry. In case of an error the reference on the inode is released.
1576 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1577 * be passed in and will be the error will be propagate to the return value,
1578 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001579 */
1580struct dentry *d_obtain_alias(struct inode *inode)
1581{
Christoph Hellwig9308a612008-08-11 15:49:12 +02001582 static const struct qstr anonstring = { .name = "" };
1583 struct dentry *tmp;
1584 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001585
1586 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001587 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001588 if (IS_ERR(inode))
1589 return ERR_CAST(inode);
1590
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001591 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001592 if (res)
1593 goto out_iput;
1594
Al Viroa4464db2011-07-07 15:03:58 -04001595 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001596 if (!tmp) {
1597 res = ERR_PTR(-ENOMEM);
1598 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001599 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001600
Nick Piggin873feea2011-01-07 17:50:06 +11001601 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001602 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001603 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001604 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001605 dput(tmp);
1606 goto out_iput;
1607 }
1608
1609 /* attach a disconnected dentry */
1610 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001611 tmp->d_inode = inode;
1612 tmp->d_flags |= DCACHE_DISCONNECTED;
Al Virob3d9b7a2012-06-09 13:51:19 -04001613 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001614 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001615 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001616 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001617 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001618 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001619 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001620
Christoph Hellwig9308a612008-08-11 15:49:12 +02001621 return tmp;
1622
1623 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001624 if (res && !IS_ERR(res))
1625 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001626 iput(inode);
1627 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001628}
Benny Halevyadc48722009-02-27 14:02:59 -08001629EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631/**
1632 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1633 * @inode: the inode which may have a disconnected dentry
1634 * @dentry: a negative dentry which we want to point to the inode.
1635 *
1636 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1637 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1638 * and return it, else simply d_add the inode to the dentry and return NULL.
1639 *
1640 * This is needed in the lookup routine of any filesystem that is exportable
1641 * (via knfsd) so that we can build dcache paths to directories effectively.
1642 *
1643 * If a dentry was found and moved, then it is returned. Otherwise NULL
1644 * is returned. This matches the expected return value of ->lookup.
1645 *
1646 */
1647struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1648{
1649 struct dentry *new = NULL;
1650
Al Viroa9049372011-07-08 21:20:11 -04001651 if (IS_ERR(inode))
1652 return ERR_CAST(inode);
1653
NeilBrown21c0d8f2006-10-04 02:16:16 -07001654 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001655 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001656 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001658 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001659 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 d_move(new, dentry);
1662 iput(inode);
1663 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001664 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001665 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001666 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 security_d_instantiate(dentry, inode);
1668 d_rehash(dentry);
1669 }
1670 } else
1671 d_add(dentry, inode);
1672 return new;
1673}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001674EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
Barry Naujok94035402008-05-21 16:50:46 +10001676/**
1677 * d_add_ci - lookup or allocate new dentry with case-exact name
1678 * @inode: the inode case-insensitive lookup has found
1679 * @dentry: the negative dentry that was passed to the parent's lookup func
1680 * @name: the case-exact name to be associated with the returned dentry
1681 *
1682 * This is to avoid filling the dcache with case-insensitive names to the
1683 * same inode, only the actual correct case is stored in the dcache for
1684 * case-insensitive filesystems.
1685 *
1686 * For a case-insensitive lookup match and if the the case-exact dentry
1687 * already exists in in the dcache, use it and return it.
1688 *
1689 * If no entry exists with the exact case name, allocate new dentry with
1690 * the exact case, and return the spliced entry.
1691 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001692struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001693 struct qstr *name)
1694{
1695 int error;
1696 struct dentry *found;
1697 struct dentry *new;
1698
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001699 /*
1700 * First check if a dentry matching the name already exists,
1701 * if not go ahead and create it now.
1702 */
Barry Naujok94035402008-05-21 16:50:46 +10001703 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001704 if (!found) {
1705 new = d_alloc(dentry->d_parent, name);
1706 if (!new) {
1707 error = -ENOMEM;
1708 goto err_out;
1709 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001710
Barry Naujok94035402008-05-21 16:50:46 +10001711 found = d_splice_alias(inode, new);
1712 if (found) {
1713 dput(new);
1714 return found;
1715 }
1716 return new;
1717 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001718
1719 /*
1720 * If a matching dentry exists, and it's not negative use it.
1721 *
1722 * Decrement the reference count to balance the iget() done
1723 * earlier on.
1724 */
Barry Naujok94035402008-05-21 16:50:46 +10001725 if (found->d_inode) {
1726 if (unlikely(found->d_inode != inode)) {
1727 /* This can't happen because bad inodes are unhashed. */
1728 BUG_ON(!is_bad_inode(inode));
1729 BUG_ON(!is_bad_inode(found->d_inode));
1730 }
Barry Naujok94035402008-05-21 16:50:46 +10001731 iput(inode);
1732 return found;
1733 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001734
Barry Naujok94035402008-05-21 16:50:46 +10001735 /*
Josef Bacik44396f42011-05-31 11:58:49 -04001736 * We are going to instantiate this dentry, unhash it and clear the
1737 * lookup flag so we can do that.
1738 */
1739 if (unlikely(d_need_lookup(found)))
1740 d_clear_need_lookup(found);
1741
1742 /*
Barry Naujok94035402008-05-21 16:50:46 +10001743 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001744 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001745 */
Al Viro4513d892011-07-17 10:52:14 -04001746 new = d_splice_alias(inode, found);
1747 if (new) {
1748 dput(found);
1749 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001750 }
Al Viro4513d892011-07-17 10:52:14 -04001751 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001752
1753err_out:
1754 iput(inode);
1755 return ERR_PTR(error);
1756}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001757EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001759/*
1760 * Do the slow-case of the dentry name compare.
1761 *
1762 * Unlike the dentry_cmp() function, we need to atomically
1763 * load the name, length and inode information, so that the
1764 * filesystem can rely on them, and can use the 'name' and
1765 * 'len' information without worrying about walking off the
1766 * end of memory etc.
1767 *
1768 * Thus the read_seqcount_retry() and the "duplicate" info
1769 * in arguments (the low-level filesystem should not look
1770 * at the dentry inode or name contents directly, since
1771 * rename can change them while we're in RCU mode).
1772 */
1773enum slow_d_compare {
1774 D_COMP_OK,
1775 D_COMP_NOMATCH,
1776 D_COMP_SEQRETRY,
1777};
1778
1779static noinline enum slow_d_compare slow_dentry_cmp(
1780 const struct dentry *parent,
1781 struct inode *inode,
1782 struct dentry *dentry,
1783 unsigned int seq,
1784 const struct qstr *name)
1785{
1786 int tlen = dentry->d_name.len;
1787 const char *tname = dentry->d_name.name;
1788 struct inode *i = dentry->d_inode;
1789
1790 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1791 cpu_relax();
1792 return D_COMP_SEQRETRY;
1793 }
1794 if (parent->d_op->d_compare(parent, inode,
1795 dentry, i,
1796 tlen, tname, name))
1797 return D_COMP_NOMATCH;
1798 return D_COMP_OK;
1799}
1800
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001802 * __d_lookup_rcu - search for a dentry (racy, store-free)
1803 * @parent: parent dentry
1804 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001805 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001806 * @inode: returns dentry->d_inode when the inode was found valid.
1807 * Returns: dentry, or NULL
1808 *
1809 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1810 * resolution (store-free path walking) design described in
1811 * Documentation/filesystems/path-lookup.txt.
1812 *
1813 * This is not to be used outside core vfs.
1814 *
1815 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1816 * held, and rcu_read_lock held. The returned dentry must not be stored into
1817 * without taking d_lock and checking d_seq sequence count against @seq
1818 * returned here.
1819 *
1820 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1821 * function.
1822 *
1823 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1824 * the returned dentry, so long as its parent's seqlock is checked after the
1825 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1826 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001827 *
1828 * NOTE! The caller *has* to check the resulting dentry against the sequence
1829 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001830 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001831struct dentry *__d_lookup_rcu(const struct dentry *parent,
1832 const struct qstr *name,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001833 unsigned *seqp, struct inode *inode)
Nick Piggin31e6b012011-01-07 17:49:52 +11001834{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001835 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001836 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001837 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001838 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001839 struct dentry *dentry;
1840
1841 /*
1842 * Note: There is significant duplication with __d_lookup_rcu which is
1843 * required to prevent single threaded performance regressions
1844 * especially on architectures where smp_rmb (in seqcounts) are costly.
1845 * Keep the two functions in sync.
1846 */
1847
1848 /*
1849 * The hash list is protected using RCU.
1850 *
1851 * Carefully use d_seq when comparing a candidate dentry, to avoid
1852 * races with d_move().
1853 *
1854 * It is possible that concurrent renames can mess up our list
1855 * walk here and result in missing our dentry, resulting in the
1856 * false-negative result. d_lookup() protects against concurrent
1857 * renames using rename_lock seqlock.
1858 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001859 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11001860 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001861 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08001862 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001863
Nick Piggin31e6b012011-01-07 17:49:52 +11001864seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001865 /*
1866 * The dentry sequence count protects us from concurrent
1867 * renames, and thus protects inode, parent and name fields.
1868 *
1869 * The caller must perform a seqcount check in order
1870 * to do anything useful with the returned dentry,
1871 * including using the 'd_inode' pointer.
1872 *
1873 * NOTE! We do a "raw" seqcount_begin here. That means that
1874 * we don't wait for the sequence count to stabilize if it
1875 * is in the middle of a sequence change. If we do the slow
1876 * dentry compare, we will do seqretries until it is stable,
1877 * and if we end up with a successful lookup, we actually
1878 * want to exit RCU lookup anyway.
1879 */
1880 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001881 if (dentry->d_parent != parent)
1882 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07001883 if (d_unhashed(dentry))
1884 continue;
Linus Torvalds8966be92012-03-02 14:23:30 -08001885 *seqp = seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001886
1887 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001888 if (dentry->d_name.hash != hashlen_hash(hashlen))
1889 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001890 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1891 case D_COMP_OK:
1892 return dentry;
1893 case D_COMP_NOMATCH:
1894 continue;
1895 default:
1896 goto seqretry;
1897 }
1898 }
1899
Linus Torvalds26fe5752012-05-10 13:14:12 -07001900 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07001901 continue;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001902 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001903 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11001904 }
1905 return NULL;
1906}
1907
1908/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 * d_lookup - search for a dentry
1910 * @parent: parent dentry
1911 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001912 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001914 * d_lookup searches the children of the parent dentry for the name in
1915 * question. If the dentry is found its reference count is incremented and the
1916 * dentry is returned. The caller must use dput to free the entry when it has
1917 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001919struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920{
Nick Piggin31e6b012011-01-07 17:49:52 +11001921 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001922 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 do {
1925 seq = read_seqbegin(&rename_lock);
1926 dentry = __d_lookup(parent, name);
1927 if (dentry)
1928 break;
1929 } while (read_seqretry(&rename_lock, seq));
1930 return dentry;
1931}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001932EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933
Nick Piggin31e6b012011-01-07 17:49:52 +11001934/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001935 * __d_lookup - search for a dentry (racy)
1936 * @parent: parent dentry
1937 * @name: qstr of name we wish to find
1938 * Returns: dentry, or NULL
1939 *
1940 * __d_lookup is like d_lookup, however it may (rarely) return a
1941 * false-negative result due to unrelated rename activity.
1942 *
1943 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1944 * however it must be used carefully, eg. with a following d_lookup in
1945 * the case of failure.
1946 *
1947 * __d_lookup callers must be commented.
1948 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001949struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950{
1951 unsigned int len = name->len;
1952 unsigned int hash = name->hash;
1953 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001954 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001955 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001956 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001957 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
Nick Pigginb04f7842010-08-18 04:37:34 +10001959 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001960 * Note: There is significant duplication with __d_lookup_rcu which is
1961 * required to prevent single threaded performance regressions
1962 * especially on architectures where smp_rmb (in seqcounts) are costly.
1963 * Keep the two functions in sync.
1964 */
1965
1966 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001967 * The hash list is protected using RCU.
1968 *
1969 * Take d_lock when comparing a candidate dentry, to avoid races
1970 * with d_move().
1971 *
1972 * It is possible that concurrent renames can mess up our list
1973 * walk here and result in missing our dentry, resulting in the
1974 * false-negative result. d_lookup() protects against concurrent
1975 * renames using rename_lock seqlock.
1976 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001977 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001978 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 rcu_read_lock();
1980
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001981 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 if (dentry->d_name.hash != hash)
1984 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
1986 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 if (dentry->d_parent != parent)
1988 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001989 if (d_unhashed(dentry))
1990 goto next;
1991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 /*
1993 * It is safe to compare names since d_move() cannot
1994 * change the qstr (protected by d_lock).
1995 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11001996 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001997 int tlen = dentry->d_name.len;
1998 const char *tname = dentry->d_name.name;
Nick Piggin621e1552011-01-07 17:49:27 +11001999 if (parent->d_op->d_compare(parent, parent->d_inode,
2000 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11002001 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 goto next;
2003 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002004 if (dentry->d_name.len != len)
2005 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002006 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 goto next;
2008 }
2009
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002010 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002011 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 spin_unlock(&dentry->d_lock);
2013 break;
2014next:
2015 spin_unlock(&dentry->d_lock);
2016 }
2017 rcu_read_unlock();
2018
2019 return found;
2020}
2021
2022/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002023 * d_hash_and_lookup - hash the qstr then search for a dentry
2024 * @dir: Directory to search in
2025 * @name: qstr of name we wish to find
2026 *
2027 * On hash failure or on lookup failure NULL is returned.
2028 */
2029struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2030{
2031 struct dentry *dentry = NULL;
2032
2033 /*
2034 * Check for a fs-specific hash function. Note that we must
2035 * calculate the standard hash first, as the d_op->d_hash()
2036 * routine may choose to leave the hash value unchanged.
2037 */
2038 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002039 if (dir->d_flags & DCACHE_OP_HASH) {
Nick Pigginb1e6a012011-01-07 17:49:28 +11002040 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002041 goto out;
2042 }
2043 dentry = d_lookup(dir, name);
2044out:
2045 return dentry;
2046}
2047
2048/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002049 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002051 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 *
2053 * An insecure source has sent us a dentry, here we verify it and dget() it.
2054 * This is used by ncpfs in its readdir implementation.
2055 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002056 *
2057 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002059int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060{
Nick Piggin786a5e12011-01-07 17:49:16 +11002061 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002062
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002063 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002064 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2065 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002066 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002067 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002068 spin_unlock(&dentry->d_lock);
2069 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 return 1;
2071 }
2072 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002073 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 return 0;
2076}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002077EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
2079/*
2080 * When a file is deleted, we have two options:
2081 * - turn this dentry into a negative dentry
2082 * - unhash this dentry and free it.
2083 *
2084 * Usually, we want to just turn this into
2085 * a negative dentry, but if anybody else is
2086 * currently using the dentry or the inode
2087 * we can't do that and we fall back on removing
2088 * it from the hash queues and waiting for
2089 * it to be deleted later when it has no users
2090 */
2091
2092/**
2093 * d_delete - delete a dentry
2094 * @dentry: The dentry to delete
2095 *
2096 * Turn the dentry into a negative dentry if possible, otherwise
2097 * remove it from the hash queues so it can be deleted later
2098 */
2099
2100void d_delete(struct dentry * dentry)
2101{
Nick Piggin873feea2011-01-07 17:50:06 +11002102 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002103 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 /*
2105 * Are we the only user?
2106 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002107again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002109 inode = dentry->d_inode;
2110 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002111 if (dentry->d_count == 1) {
Nick Piggin873feea2011-01-07 17:50:06 +11002112 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002113 spin_unlock(&dentry->d_lock);
2114 cpu_relax();
2115 goto again;
2116 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002117 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002118 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002119 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 return;
2121 }
2122
2123 if (!d_unhashed(dentry))
2124 __d_drop(dentry);
2125
2126 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002127
2128 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002130EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002132static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002134 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002135 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002136 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002137 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002138 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139}
2140
David Howells770bfad2006-08-22 20:06:07 -04002141static void _d_rehash(struct dentry * entry)
2142{
2143 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2144}
2145
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146/**
2147 * d_rehash - add an entry back to the hash
2148 * @entry: dentry to add to the hash
2149 *
2150 * Adds a dentry to the hash according to its name.
2151 */
2152
2153void d_rehash(struct dentry * entry)
2154{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002156 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002159EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002161/**
2162 * dentry_update_name_case - update case insensitive dentry with a new name
2163 * @dentry: dentry to be updated
2164 * @name: new name
2165 *
2166 * Update a case insensitive dentry with new case of name.
2167 *
2168 * dentry must have been returned by d_lookup with name @name. Old and new
2169 * name lengths must match (ie. no d_compare which allows mismatched name
2170 * lengths).
2171 *
2172 * Parent inode i_mutex must be held over d_lookup and into this call (to
2173 * keep renames and concurrent inserts, and readdir(2) away).
2174 */
2175void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2176{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002177 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002178 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2179
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002180 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002181 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002182 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002183 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002184 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002185}
2186EXPORT_SYMBOL(dentry_update_name_case);
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188static void switch_names(struct dentry *dentry, struct dentry *target)
2189{
2190 if (dname_external(target)) {
2191 if (dname_external(dentry)) {
2192 /*
2193 * Both external: swap the pointers
2194 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002195 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 } else {
2197 /*
2198 * dentry:internal, target:external. Steal target's
2199 * storage and make target internal.
2200 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002201 memcpy(target->d_iname, dentry->d_name.name,
2202 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 dentry->d_name.name = target->d_name.name;
2204 target->d_name.name = target->d_iname;
2205 }
2206 } else {
2207 if (dname_external(dentry)) {
2208 /*
2209 * dentry:external, target:internal. Give dentry's
2210 * storage to target and make dentry internal
2211 */
2212 memcpy(dentry->d_iname, target->d_name.name,
2213 target->d_name.len + 1);
2214 target->d_name.name = dentry->d_name.name;
2215 dentry->d_name.name = dentry->d_iname;
2216 } else {
2217 /*
2218 * Both are internal. Just copy target to dentry
2219 */
2220 memcpy(dentry->d_iname, target->d_name.name,
2221 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002222 dentry->d_name.len = target->d_name.len;
2223 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 }
2225 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002226 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227}
2228
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002229static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2230{
2231 /*
2232 * XXXX: do we really need to take target->d_lock?
2233 */
2234 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2235 spin_lock(&target->d_parent->d_lock);
2236 else {
2237 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2238 spin_lock(&dentry->d_parent->d_lock);
2239 spin_lock_nested(&target->d_parent->d_lock,
2240 DENTRY_D_LOCK_NESTED);
2241 } else {
2242 spin_lock(&target->d_parent->d_lock);
2243 spin_lock_nested(&dentry->d_parent->d_lock,
2244 DENTRY_D_LOCK_NESTED);
2245 }
2246 }
2247 if (target < dentry) {
2248 spin_lock_nested(&target->d_lock, 2);
2249 spin_lock_nested(&dentry->d_lock, 3);
2250 } else {
2251 spin_lock_nested(&dentry->d_lock, 2);
2252 spin_lock_nested(&target->d_lock, 3);
2253 }
2254}
2255
2256static void dentry_unlock_parents_for_move(struct dentry *dentry,
2257 struct dentry *target)
2258{
2259 if (target->d_parent != dentry->d_parent)
2260 spin_unlock(&dentry->d_parent->d_lock);
2261 if (target->d_parent != target)
2262 spin_unlock(&target->d_parent->d_lock);
2263}
2264
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002266 * When switching names, the actual string doesn't strictly have to
2267 * be preserved in the target - because we're dropping the target
2268 * anyway. As such, we can just do a simple memcpy() to copy over
2269 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002271 * Note that we have to be a lot more careful about getting the hash
2272 * switched - we have to switch the hash value properly even if it
2273 * then no longer matches the actual (corrupted) string of the target.
2274 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002276/*
Al Viro18367502011-07-12 21:42:24 -04002277 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 * @dentry: entry to move
2279 * @target: new dentry
2280 *
2281 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002282 * dcache entries should not be moved in this way. Caller must hold
2283 * rename_lock, the i_mutex of the source and target directories,
2284 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 */
Al Viro18367502011-07-12 21:42:24 -04002286static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (!dentry->d_inode)
2289 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2290
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002291 BUG_ON(d_ancestor(dentry, target));
2292 BUG_ON(d_ancestor(target, dentry));
2293
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002294 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Nick Piggin31e6b012011-01-07 17:49:52 +11002296 write_seqcount_begin(&dentry->d_seq);
2297 write_seqcount_begin(&target->d_seq);
2298
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002299 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2300
2301 /*
2302 * Move the dentry to the target hash queue. Don't bother checking
2303 * for the same hash queue because of how unlikely it is.
2304 */
2305 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002306 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307
2308 /* Unhash the target: dput() will then get rid of it */
2309 __d_drop(target);
2310
Eric Dumazet5160ee62006-01-08 01:03:32 -08002311 list_del(&dentry->d_u.d_child);
2312 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
2314 /* Switch the names.. */
2315 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002316 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
2318 /* ... and switch the parents */
2319 if (IS_ROOT(dentry)) {
2320 dentry->d_parent = target->d_parent;
2321 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002322 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002324 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
2326 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002327 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 }
2329
Eric Dumazet5160ee62006-01-08 01:03:32 -08002330 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002331
Nick Piggin31e6b012011-01-07 17:49:52 +11002332 write_seqcount_end(&target->d_seq);
2333 write_seqcount_end(&dentry->d_seq);
2334
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002335 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002337 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002339}
2340
2341/*
2342 * d_move - move a dentry
2343 * @dentry: entry to move
2344 * @target: new dentry
2345 *
2346 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002347 * dcache entries should not be moved in this way. See the locking
2348 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002349 */
2350void d_move(struct dentry *dentry, struct dentry *target)
2351{
2352 write_seqlock(&rename_lock);
2353 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002355}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002356EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002358/**
2359 * d_ancestor - search for an ancestor
2360 * @p1: ancestor dentry
2361 * @p2: child dentry
2362 *
2363 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2364 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002365 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002366struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002367{
2368 struct dentry *p;
2369
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002370 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002371 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002372 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002373 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002374 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002375}
2376
2377/*
2378 * This helper attempts to cope with remotely renamed directories
2379 *
2380 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002381 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002382 *
2383 * Note: If ever the locking in lock_rename() changes, then please
2384 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002385 */
Nick Piggin873feea2011-01-07 17:50:06 +11002386static struct dentry *__d_unalias(struct inode *inode,
2387 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002388{
2389 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002390 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002391
2392 /* If alias and dentry share a parent, then no extra locks required */
2393 if (alias->d_parent == dentry->d_parent)
2394 goto out_unalias;
2395
Trond Myklebust9eaef272006-10-21 10:24:20 -07002396 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002397 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2398 goto out_err;
2399 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2400 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2401 goto out_err;
2402 m2 = &alias->d_parent->d_inode->i_mutex;
2403out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002404 if (likely(!d_mountpoint(alias))) {
2405 __d_move(alias, dentry);
2406 ret = alias;
2407 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002408out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002409 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002410 if (m2)
2411 mutex_unlock(m2);
2412 if (m1)
2413 mutex_unlock(m1);
2414 return ret;
2415}
2416
2417/*
David Howells770bfad2006-08-22 20:06:07 -04002418 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2419 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002420 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002421 */
2422static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2423{
2424 struct dentry *dparent, *aparent;
2425
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002426 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002427
Nick Piggin31e6b012011-01-07 17:49:52 +11002428 write_seqcount_begin(&dentry->d_seq);
2429 write_seqcount_begin(&anon->d_seq);
2430
David Howells770bfad2006-08-22 20:06:07 -04002431 dparent = dentry->d_parent;
2432 aparent = anon->d_parent;
2433
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002434 switch_names(dentry, anon);
2435 swap(dentry->d_name.hash, anon->d_name.hash);
2436
David Howells770bfad2006-08-22 20:06:07 -04002437 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2438 list_del(&dentry->d_u.d_child);
2439 if (!IS_ROOT(dentry))
2440 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2441 else
2442 INIT_LIST_HEAD(&dentry->d_u.d_child);
2443
2444 anon->d_parent = (dparent == dentry) ? anon : dparent;
2445 list_del(&anon->d_u.d_child);
2446 if (!IS_ROOT(anon))
2447 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2448 else
2449 INIT_LIST_HEAD(&anon->d_u.d_child);
2450
Nick Piggin31e6b012011-01-07 17:49:52 +11002451 write_seqcount_end(&dentry->d_seq);
2452 write_seqcount_end(&anon->d_seq);
2453
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002454 dentry_unlock_parents_for_move(anon, dentry);
2455 spin_unlock(&dentry->d_lock);
2456
2457 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002458 anon->d_flags &= ~DCACHE_DISCONNECTED;
2459}
2460
2461/**
2462 * d_materialise_unique - introduce an inode into the tree
2463 * @dentry: candidate dentry
2464 * @inode: inode to bind to the dentry, to which aliases may be attached
2465 *
2466 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002467 * root directory alias in its place if there is one. Caller must hold the
2468 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002469 */
2470struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2471{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002472 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002473
2474 BUG_ON(!d_unhashed(dentry));
2475
David Howells770bfad2006-08-22 20:06:07 -04002476 if (!inode) {
2477 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002478 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002479 d_rehash(actual);
2480 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002481 }
2482
Nick Piggin873feea2011-01-07 17:50:06 +11002483 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002484
Trond Myklebust9eaef272006-10-21 10:24:20 -07002485 if (S_ISDIR(inode->i_mode)) {
2486 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002487
Trond Myklebust9eaef272006-10-21 10:24:20 -07002488 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002489 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002490 if (alias) {
2491 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002492 write_seqlock(&rename_lock);
2493
2494 if (d_ancestor(alias, dentry)) {
2495 /* Check for loops */
2496 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002497 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002498 } else if (IS_ROOT(alias)) {
2499 /* Is this an anonymous mountpoint that we
2500 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002501 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002502 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002503 __d_drop(alias);
2504 goto found;
Al Viro18367502011-07-12 21:42:24 -04002505 } else {
2506 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002507 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002508 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002509 }
Al Viro18367502011-07-12 21:42:24 -04002510 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002511 if (IS_ERR(actual)) {
2512 if (PTR_ERR(actual) == -ELOOP)
2513 pr_warn_ratelimited(
2514 "VFS: Lookup of '%s' in %s %s"
2515 " would have caused loop\n",
2516 dentry->d_name.name,
2517 inode->i_sb->s_type->name,
2518 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002519 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002520 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002521 goto out_nolock;
2522 }
David Howells770bfad2006-08-22 20:06:07 -04002523 }
2524
2525 /* Add a unique reference */
2526 actual = __d_instantiate_unique(dentry, inode);
2527 if (!actual)
2528 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002529 else
2530 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002531
David Howells770bfad2006-08-22 20:06:07 -04002532 spin_lock(&actual->d_lock);
2533found:
2534 _d_rehash(actual);
2535 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002536 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002537out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002538 if (actual == dentry) {
2539 security_d_instantiate(dentry, inode);
2540 return NULL;
2541 }
2542
2543 iput(inode);
2544 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002545}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002546EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002547
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002548static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002549{
2550 *buflen -= namelen;
2551 if (*buflen < 0)
2552 return -ENAMETOOLONG;
2553 *buffer -= namelen;
2554 memcpy(*buffer, str, namelen);
2555 return 0;
2556}
2557
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002558static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2559{
2560 return prepend(buffer, buflen, name->name, name->len);
2561}
2562
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002564 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002565 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002566 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002567 * @buffer: pointer to the end of the buffer
2568 * @buflen: pointer to buffer length
2569 *
Nick Piggin949854d2011-01-07 17:49:37 +11002570 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002571 */
Al Viro02125a82011-12-05 08:43:34 -05002572static int prepend_path(const struct path *path,
2573 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002574 char **buffer, int *buflen)
2575{
2576 struct dentry *dentry = path->dentry;
2577 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002578 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002579 bool slash = false;
2580 int error = 0;
2581
Andi Kleen962830d2012-05-08 13:32:02 +09302582 br_read_lock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002583 while (dentry != root->dentry || vfsmnt != root->mnt) {
2584 struct dentry * parent;
2585
2586 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2587 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002588 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002589 goto global_root;
Al Viroa73324d2011-11-24 22:25:07 -05002590 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002591 mnt = mnt->mnt_parent;
2592 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002593 continue;
2594 }
2595 parent = dentry->d_parent;
2596 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002597 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002598 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002599 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002600 if (!error)
2601 error = prepend(buffer, buflen, "/", 1);
2602 if (error)
2603 break;
2604
2605 slash = true;
2606 dentry = parent;
2607 }
2608
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002609 if (!error && !slash)
2610 error = prepend(buffer, buflen, "/", 1);
2611
Al Viro02125a82011-12-05 08:43:34 -05002612out:
Andi Kleen962830d2012-05-08 13:32:02 +09302613 br_read_unlock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002614 return error;
2615
2616global_root:
2617 /*
2618 * Filesystems needing to implement special "root names"
2619 * should do so with ->d_dname()
2620 */
2621 if (IS_ROOT(dentry) &&
2622 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2623 WARN(1, "Root dentry has weird name <%.*s>\n",
2624 (int) dentry->d_name.len, dentry->d_name.name);
2625 }
Al Viro02125a82011-12-05 08:43:34 -05002626 if (!slash)
2627 error = prepend(buffer, buflen, "/", 1);
2628 if (!error)
Al Virof7a99c52012-06-09 00:59:08 -04002629 error = is_mounted(vfsmnt) ? 1 : 2;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002630 goto out;
2631}
2632
2633/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002634 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002635 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002636 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002637 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 * @buflen: buffer length
2639 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002640 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002642 * Returns a pointer into the buffer or an error code if the
2643 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002644 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002645 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002646 *
Al Viro02125a82011-12-05 08:43:34 -05002647 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 */
Al Viro02125a82011-12-05 08:43:34 -05002649char *__d_path(const struct path *path,
2650 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002651 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002653 char *res = buf + buflen;
2654 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002656 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002657 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002658 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002659 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002660
Al Viro02125a82011-12-05 08:43:34 -05002661 if (error < 0)
2662 return ERR_PTR(error);
2663 if (error > 0)
2664 return NULL;
2665 return res;
2666}
2667
2668char *d_absolute_path(const struct path *path,
2669 char *buf, int buflen)
2670{
2671 struct path root = {};
2672 char *res = buf + buflen;
2673 int error;
2674
2675 prepend(&res, &buflen, "\0", 1);
2676 write_seqlock(&rename_lock);
2677 error = prepend_path(path, &root, &res, &buflen);
2678 write_sequnlock(&rename_lock);
2679
2680 if (error > 1)
2681 error = -EINVAL;
2682 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002683 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002684 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685}
2686
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002687/*
2688 * same as __d_path but appends "(deleted)" for unlinked files.
2689 */
Al Viro02125a82011-12-05 08:43:34 -05002690static int path_with_deleted(const struct path *path,
2691 const struct path *root,
2692 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002693{
2694 prepend(buf, buflen, "\0", 1);
2695 if (d_unlinked(path->dentry)) {
2696 int error = prepend(buf, buflen, " (deleted)", 10);
2697 if (error)
2698 return error;
2699 }
2700
2701 return prepend_path(path, root, buf, buflen);
2702}
2703
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002704static int prepend_unreachable(char **buffer, int *buflen)
2705{
2706 return prepend(buffer, buflen, "(unreachable)", 13);
2707}
2708
Jan Bluncka03a8a702008-02-14 19:38:32 -08002709/**
2710 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002711 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002712 * @buf: buffer to return value in
2713 * @buflen: buffer length
2714 *
2715 * Convert a dentry into an ASCII path name. If the entry has been deleted
2716 * the string " (deleted)" is appended. Note that this is ambiguous.
2717 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002718 * Returns a pointer into the buffer or an error code if the path was
2719 * too long. Note: Callers should use the returned pointer, not the passed
2720 * in buffer, to use the name! The implementation often starts at an offset
2721 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002722 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002723 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002724 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002725char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002727 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002728 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002729 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002731 /*
2732 * We have various synthetic filesystems that never get mounted. On
2733 * these filesystems dentries are never used for lookup purposes, and
2734 * thus don't need to be hashed. They also don't need a name until a
2735 * user wants to identify the object in /proc/pid/fd/. The little hack
2736 * below allows us to generate a name for these objects on demand:
2737 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002738 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2739 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002740
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002741 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002742 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002743 error = path_with_deleted(path, &root, &res, &buflen);
2744 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002745 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002746 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002747 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 return res;
2749}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002750EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002752/**
2753 * d_path_with_unreachable - return the path of a dentry
2754 * @path: path to report
2755 * @buf: buffer to return value in
2756 * @buflen: buffer length
2757 *
2758 * The difference from d_path() is that this prepends "(unreachable)"
2759 * to paths which are unreachable from the current process' root.
2760 */
2761char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2762{
2763 char *res = buf + buflen;
2764 struct path root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002765 int error;
2766
2767 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2768 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2769
2770 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002771 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002772 error = path_with_deleted(path, &root, &res, &buflen);
2773 if (error > 0)
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002774 error = prepend_unreachable(&res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002775 write_sequnlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002776 path_put(&root);
2777 if (error)
2778 res = ERR_PTR(error);
2779
2780 return res;
2781}
2782
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002784 * Helper function for dentry_operations.d_dname() members
2785 */
2786char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2787 const char *fmt, ...)
2788{
2789 va_list args;
2790 char temp[64];
2791 int sz;
2792
2793 va_start(args, fmt);
2794 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2795 va_end(args);
2796
2797 if (sz > sizeof(temp) || sz > buflen)
2798 return ERR_PTR(-ENAMETOOLONG);
2799
2800 buffer += buflen - sz;
2801 return memcpy(buffer, temp, sz);
2802}
2803
2804/*
Ram Pai6092d042008-03-27 13:06:20 +01002805 * Write full pathname from the root of the filesystem into the buffer.
2806 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002807static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002808{
2809 char *end = buf + buflen;
2810 char *retval;
2811
Ram Pai6092d042008-03-27 13:06:20 +01002812 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002813 if (buflen < 1)
2814 goto Elong;
2815 /* Get '/' right */
2816 retval = end-1;
2817 *retval = '/';
2818
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002819 while (!IS_ROOT(dentry)) {
2820 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002821 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002822
Ram Pai6092d042008-03-27 13:06:20 +01002823 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002824 spin_lock(&dentry->d_lock);
2825 error = prepend_name(&end, &buflen, &dentry->d_name);
2826 spin_unlock(&dentry->d_lock);
2827 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002828 goto Elong;
2829
2830 retval = end;
2831 dentry = parent;
2832 }
Al Viroc1031352010-06-06 22:31:14 -04002833 return retval;
2834Elong:
2835 return ERR_PTR(-ENAMETOOLONG);
2836}
Nick Pigginec2447c2011-01-07 17:49:29 +11002837
2838char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2839{
2840 char *retval;
2841
Nick Piggin949854d2011-01-07 17:49:37 +11002842 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002843 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002844 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002845
2846 return retval;
2847}
2848EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002849
2850char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2851{
2852 char *p = NULL;
2853 char *retval;
2854
Nick Piggin949854d2011-01-07 17:49:37 +11002855 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002856 if (d_unlinked(dentry)) {
2857 p = buf + buflen;
2858 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2859 goto Elong;
2860 buflen++;
2861 }
2862 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002863 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002864 if (!IS_ERR(retval) && p)
2865 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002866 return retval;
2867Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002868 return ERR_PTR(-ENAMETOOLONG);
2869}
2870
2871/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 * NOTE! The user-level library version returns a
2873 * character pointer. The kernel system call just
2874 * returns the length of the buffer filled (which
2875 * includes the ending '\0' character), or a negative
2876 * error value. So libc would do something like
2877 *
2878 * char *getcwd(char * buf, size_t size)
2879 * {
2880 * int retval;
2881 *
2882 * retval = sys_getcwd(buf, size);
2883 * if (retval >= 0)
2884 * return buf;
2885 * errno = -retval;
2886 * return NULL;
2887 * }
2888 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002889SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890{
Linus Torvalds552ce542007-02-13 12:08:18 -08002891 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002892 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002893 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894
2895 if (!page)
2896 return -ENOMEM;
2897
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002898 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899
Linus Torvalds552ce542007-02-13 12:08:18 -08002900 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002901 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002902 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002903 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002904 char *cwd = page + PAGE_SIZE;
2905 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002907 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002908 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002909 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002910
Al Viro02125a82011-12-05 08:43:34 -05002911 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002912 goto out;
2913
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002914 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002915 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002916 error = prepend_unreachable(&cwd, &buflen);
2917 if (error)
2918 goto out;
2919 }
2920
Linus Torvalds552ce542007-02-13 12:08:18 -08002921 error = -ERANGE;
2922 len = PAGE_SIZE + page - cwd;
2923 if (len <= size) {
2924 error = len;
2925 if (copy_to_user(buf, cwd, len))
2926 error = -EFAULT;
2927 }
Nick Piggin949854d2011-01-07 17:49:37 +11002928 } else {
2929 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
2932out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002933 path_put(&pwd);
2934 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 free_page((unsigned long) page);
2936 return error;
2937}
2938
2939/*
2940 * Test whether new_dentry is a subdirectory of old_dentry.
2941 *
2942 * Trivially implemented using the dcache structure
2943 */
2944
2945/**
2946 * is_subdir - is new dentry a subdirectory of old_dentry
2947 * @new_dentry: new dentry
2948 * @old_dentry: old dentry
2949 *
2950 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2951 * Returns 0 otherwise.
2952 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2953 */
2954
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002955int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956{
2957 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002958 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002960 if (new_dentry == old_dentry)
2961 return 1;
2962
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002963 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002966 /*
2967 * Need rcu_readlock to protect against the d_parent trashing
2968 * due to d_move
2969 */
2970 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002971 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002973 else
2974 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002975 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977
2978 return result;
2979}
2980
2981void d_genocide(struct dentry *root)
2982{
Nick Piggin949854d2011-01-07 17:49:37 +11002983 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002985 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002986 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
Nick Piggin949854d2011-01-07 17:49:37 +11002988 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002989again:
2990 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002991 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992repeat:
2993 next = this_parent->d_subdirs.next;
2994resume:
2995 while (next != &this_parent->d_subdirs) {
2996 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002997 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002999
Nick Pigginda502952011-01-07 17:49:33 +11003000 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3001 if (d_unhashed(dentry) || !dentry->d_inode) {
3002 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 continue;
Nick Pigginda502952011-01-07 17:49:33 +11003004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003006 spin_unlock(&this_parent->d_lock);
3007 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003009 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 goto repeat;
3011 }
Nick Piggin949854d2011-01-07 17:49:37 +11003012 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3013 dentry->d_flags |= DCACHE_GENOCIDE;
3014 dentry->d_count--;
3015 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11003016 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 }
3018 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07003019 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11003020 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
3021 this_parent->d_flags |= DCACHE_GENOCIDE;
3022 this_parent->d_count--;
3023 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07003024 this_parent = try_to_ascend(this_parent, locked, seq);
3025 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11003026 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11003027 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028 goto resume;
3029 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003030 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11003031 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11003032 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11003033 if (locked)
3034 write_sequnlock(&rename_lock);
3035 return;
3036
3037rename_retry:
3038 locked = 1;
3039 write_seqlock(&rename_lock);
3040 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041}
3042
3043/**
3044 * find_inode_number - check for dentry with name
3045 * @dir: directory to check
3046 * @name: Name to find.
3047 *
3048 * Check whether a dentry already exists for the given name,
3049 * and return the inode number if it has an inode. Otherwise
3050 * 0 is returned.
3051 *
3052 * This routine is used to post-process directory listings for
3053 * filesystems using synthetic inode numbers, and is necessary
3054 * to keep getcwd() working.
3055 */
3056
3057ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3058{
3059 struct dentry * dentry;
3060 ino_t ino = 0;
3061
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08003062 dentry = d_hash_and_lookup(dir, name);
3063 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 if (dentry->d_inode)
3065 ino = dentry->d_inode->i_ino;
3066 dput(dentry);
3067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 return ino;
3069}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003070EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
3072static __initdata unsigned long dhash_entries;
3073static int __init set_dhash_entries(char *str)
3074{
3075 if (!str)
3076 return 0;
3077 dhash_entries = simple_strtoul(str, &str, 0);
3078 return 1;
3079}
3080__setup("dhash_entries=", set_dhash_entries);
3081
3082static void __init dcache_init_early(void)
3083{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003084 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085
3086 /* If hashes are distributed across NUMA nodes, defer
3087 * hash allocation until vmalloc space is available.
3088 */
3089 if (hashdist)
3090 return;
3091
3092 dentry_hashtable =
3093 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003094 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 dhash_entries,
3096 13,
3097 HASH_EARLY,
3098 &d_hash_shift,
3099 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003100 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 0);
3102
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003103 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003104 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003105}
3106
Denis Cheng74bf17c2007-10-16 23:26:30 -07003107static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003109 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
3111 /*
3112 * A constructor could be added for stable state like the lists,
3113 * but it is probably not worth it because of the cache nature
3114 * of the dcache.
3115 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003116 dentry_cache = KMEM_CACHE(dentry,
3117 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118
3119 /* Hash may have been set up in dcache_init_early */
3120 if (!hashdist)
3121 return;
3122
3123 dentry_hashtable =
3124 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003125 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 dhash_entries,
3127 13,
3128 0,
3129 &d_hash_shift,
3130 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003131 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 0);
3133
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003134 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003135 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136}
3137
3138/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003139struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003140EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003141
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142EXPORT_SYMBOL(d_genocide);
3143
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144void __init vfs_caches_init_early(void)
3145{
3146 dcache_init_early();
3147 inode_init_early();
3148}
3149
3150void __init vfs_caches_init(unsigned long mempages)
3151{
3152 unsigned long reserve;
3153
3154 /* Base hash sizes on available memory, with a reserve equal to
3155 150% of current kernel size */
3156
3157 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3158 mempages -= reserve;
3159
3160 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003161 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162
Denis Cheng74bf17c2007-10-16 23:26:30 -07003163 dcache_init();
3164 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003166 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 bdev_cache_init();
3168 chrdev_init();
3169}