blob: 19153a0a810c1d12de751d5d416a5c1b6fe8c9bb [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 */
Miklos Szeredib161dfa62012-09-17 22:31:38 +0200392 dentry->d_flags |= DCACHE_DENTRY_KILLED;
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/*
458 * Finish off a dentry we've decided to kill.
459 * dentry->d_lock must be held, returns with it unlocked.
460 * If ref is non-zero, then decrement the refcount too.
461 * Returns dentry requiring refcount drop, or NULL if we're done.
462 */
463static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
464 __releases(dentry->d_lock)
465{
Nick Piggin873feea2011-01-07 17:50:06 +1100466 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100467 struct dentry *parent;
468
Nick Piggin873feea2011-01-07 17:50:06 +1100469 inode = dentry->d_inode;
470 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100471relock:
472 spin_unlock(&dentry->d_lock);
473 cpu_relax();
474 return dentry; /* try again with same dentry */
475 }
476 if (IS_ROOT(dentry))
477 parent = NULL;
478 else
479 parent = dentry->d_parent;
480 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100481 if (inode)
482 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100483 goto relock;
484 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100485
Nick Piggin77812a12011-01-07 17:49:48 +1100486 if (ref)
487 dentry->d_count--;
Sage Weilf0023bc2011-10-28 10:02:42 -0700488 /*
489 * if dentry was on the d_lru list delete it from there.
490 * inform the fs via d_prune that this dentry is about to be
491 * unhashed and destroyed.
492 */
493 dentry_lru_prune(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100494 /* if it was on the hash then remove it */
495 __d_drop(dentry);
496 return d_kill(dentry, parent);
497}
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499/*
500 * This is dput
501 *
502 * This is complicated by the fact that we do not want to put
503 * dentries that are no longer on any hash chain on the unused
504 * list: we'd much rather just get rid of them immediately.
505 *
506 * However, that implies that we have to traverse the dentry
507 * tree upwards to the parents which might _also_ now be
508 * scheduled for deletion (it may have been only waiting for
509 * its last child to go away).
510 *
511 * This tail recursion is done by hand as we don't want to depend
512 * on the compiler to always get this right (gcc generally doesn't).
513 * Real recursion would eat up our stack space.
514 */
515
516/*
517 * dput - release a dentry
518 * @dentry: dentry to release
519 *
520 * Release a dentry. This will drop the usage count and if appropriate
521 * call the dentry unlink method as well as removing it from the queues and
522 * releasing its resources. If the parent dentries were scheduled for release
523 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525void dput(struct dentry *dentry)
526{
527 if (!dentry)
528 return;
529
530repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100531 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100534 BUG_ON(!dentry->d_count);
535 if (dentry->d_count > 1) {
536 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return;
539 }
540
Nick Pigginfb045ad2011-01-07 17:49:55 +1100541 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100543 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 }
Nick Piggin265ac902010-10-10 05:36:24 -0400545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 /* Unreachable? Get rid of it */
547 if (d_unhashed(dentry))
548 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400549
Jeff Layton39e3c952012-11-28 11:30:53 -0500550 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400551 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400552
Nick Piggin61f3dee2011-01-07 17:49:40 +1100553 dentry->d_count--;
554 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 return;
556
Miklos Szeredid52b9082007-05-08 00:23:46 -0700557kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100558 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700559 if (dentry)
560 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700562EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564/**
565 * d_invalidate - invalidate a dentry
566 * @dentry: dentry to invalidate
567 *
568 * Try to invalidate the dentry if it turns out to be
569 * possible. If there are other dentries that can be
570 * reached through this one we can't delete it and we
571 * return -EBUSY. On success we return 0.
572 *
573 * no dcache lock.
574 */
575
576int d_invalidate(struct dentry * dentry)
577{
578 /*
579 * If it's already been dropped, return OK.
580 */
Nick Pigginda502952011-01-07 17:49:33 +1100581 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100583 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return 0;
585 }
586 /*
587 * Check whether to do a partial shrink_dcache
588 * to get rid of unused child entries.
589 */
590 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100591 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100593 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
596 /*
597 * Somebody else still using it?
598 *
599 * If it's a directory, we can't drop it
600 * for fear of somebody re-populating it
601 * with children (even though dropping it
602 * would make it unreachable from the root,
603 * we might still populate it if it was a
604 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000605 * We also need to leave mountpoints alone,
606 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Al Viro50e69632011-11-07 16:39:57 +0000608 if (dentry->d_count > 1 && dentry->d_inode) {
609 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 return -EBUSY;
612 }
613 }
614
615 __d_drop(dentry);
616 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 return 0;
618}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700619EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100621/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100622static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100624 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Nick Piggindc0474b2011-01-07 17:49:43 +1100627static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100628{
Nick Piggin23044502011-01-07 17:49:31 +1100629 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100630 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100631 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100632}
633
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100634struct dentry *dget_parent(struct dentry *dentry)
635{
636 struct dentry *ret;
637
638repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100639 /*
640 * Don't need rcu_dereference because we re-check it was correct under
641 * the lock.
642 */
643 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100644 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100645 spin_lock(&ret->d_lock);
646 if (unlikely(ret != dentry->d_parent)) {
647 spin_unlock(&ret->d_lock);
648 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100649 goto repeat;
650 }
Nick Piggina734eb42011-01-07 17:49:44 +1100651 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100652 BUG_ON(!ret->d_count);
653 ret->d_count++;
654 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100655 return ret;
656}
657EXPORT_SYMBOL(dget_parent);
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659/**
660 * d_find_alias - grab a hashed alias of inode
661 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700662 * @want_discon: flag, used by d_splice_alias, to request
663 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 *
665 * If inode has a hashed alias, or is a directory and has any alias,
666 * acquire the reference to alias and return it. Otherwise return NULL.
667 * Notice that if inode is a directory there can be only one alias and
668 * it can be unhashed only if it has no children, or if it is the root
669 * of a filesystem.
670 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700671 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700672 * any other hashed alias over that one unless @want_discon is set,
673 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700675static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
Nick Pigginda502952011-01-07 17:49:33 +1100677 struct dentry *alias, *discon_alias;
Al Virob3d9b7a2012-06-09 13:51:19 -0400678 struct hlist_node *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Nick Pigginda502952011-01-07 17:49:33 +1100680again:
681 discon_alias = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400682 hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100683 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700685 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100686 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700688 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100689 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100690 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 return alias;
692 }
693 }
Nick Pigginda502952011-01-07 17:49:33 +1100694 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
Nick Pigginda502952011-01-07 17:49:33 +1100696 if (discon_alias) {
697 alias = discon_alias;
698 spin_lock(&alias->d_lock);
699 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
700 if (IS_ROOT(alias) &&
701 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100702 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100703 spin_unlock(&alias->d_lock);
704 return alias;
705 }
706 }
707 spin_unlock(&alias->d_lock);
708 goto again;
709 }
710 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711}
712
Nick Pigginda502952011-01-07 17:49:33 +1100713struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714{
David Howells214fda12006-03-25 03:06:36 -0800715 struct dentry *de = NULL;
716
Al Virob3d9b7a2012-06-09 13:51:19 -0400717 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100718 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700719 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100720 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 return de;
723}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700724EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726/*
727 * Try to kill dentries associated with this inode.
728 * WARNING: you must own a reference to inode.
729 */
730void d_prune_aliases(struct inode *inode)
731{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700732 struct dentry *dentry;
Al Virob3d9b7a2012-06-09 13:51:19 -0400733 struct hlist_node *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100735 spin_lock(&inode->i_lock);
Al Virob3d9b7a2012-06-09 13:51:19 -0400736 hlist_for_each_entry(dentry, p, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100738 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100739 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 __d_drop(dentry);
741 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100742 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 dput(dentry);
744 goto restart;
745 }
746 spin_unlock(&dentry->d_lock);
747 }
Nick Piggin873feea2011-01-07 17:50:06 +1100748 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700750EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752/*
Nick Piggin77812a12011-01-07 17:49:48 +1100753 * Try to throw away a dentry - free the inode, dput the parent.
754 * Requires dentry->d_lock is held, and dentry->d_count == 0.
755 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700756 *
Nick Piggin77812a12011-01-07 17:49:48 +1100757 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 */
Nick Piggin77812a12011-01-07 17:49:48 +1100759static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200760 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
Nick Piggin77812a12011-01-07 17:49:48 +1100762 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700763
Nick Piggin77812a12011-01-07 17:49:48 +1100764 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700765 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100766 * If dentry_kill returns NULL, we have nothing more to do.
767 * if it returns the same dentry, trylocks failed. In either
768 * case, just loop again.
769 *
770 * Otherwise, we need to prune ancestors too. This is necessary
771 * to prevent quadratic behavior of shrink_dcache_parent(), but
772 * is also expected to be beneficial in reducing dentry cache
773 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700774 */
Nick Piggin77812a12011-01-07 17:49:48 +1100775 if (!parent)
776 return;
777 if (parent == dentry)
778 return;
779
780 /* Prune ancestors. */
781 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700782 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100783 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100784 if (dentry->d_count > 1) {
785 dentry->d_count--;
786 spin_unlock(&dentry->d_lock);
787 return;
788 }
Nick Piggin77812a12011-01-07 17:49:48 +1100789 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791}
792
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400793static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700795 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700796
Nick Pigginec336792011-01-07 17:49:47 +1100797 rcu_read_lock();
798 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100799 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
800 if (&dentry->d_lru == list)
801 break; /* empty */
802 spin_lock(&dentry->d_lock);
803 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
804 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100805 continue;
806 }
807
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 /*
809 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700810 * the LRU because of laziness during lookup. Do not free
811 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100813 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100814 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700815 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 continue;
817 }
Nick Pigginec336792011-01-07 17:49:47 +1100818
Nick Pigginec336792011-01-07 17:49:47 +1100819 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100820
821 try_prune_one_dentry(dentry);
822
Nick Pigginec336792011-01-07 17:49:47 +1100823 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 }
Nick Pigginec336792011-01-07 17:49:47 +1100825 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400826}
827
828/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000829 * prune_dcache_sb - shrink the dcache
830 * @sb: superblock
831 * @count: number of entries to try to free
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400832 *
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000833 * Attempt to shrink the superblock dcache LRU by @count entries. This is
834 * done when we need more memory an called from the superblock shrinker
835 * function.
836 *
837 * This function may fail to free any resources if all the dentries are in
838 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400839 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000840void prune_dcache_sb(struct super_block *sb, int count)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400841{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400842 struct dentry *dentry;
843 LIST_HEAD(referenced);
844 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400845
Nick Piggin23044502011-01-07 17:49:31 +1100846relock:
847 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400848 while (!list_empty(&sb->s_dentry_lru)) {
849 dentry = list_entry(sb->s_dentry_lru.prev,
850 struct dentry, d_lru);
851 BUG_ON(dentry->d_sb != sb);
852
Nick Piggin23044502011-01-07 17:49:31 +1100853 if (!spin_trylock(&dentry->d_lock)) {
854 spin_unlock(&dcache_lru_lock);
855 cpu_relax();
856 goto relock;
857 }
858
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000859 if (dentry->d_flags & DCACHE_REFERENCED) {
Nick Piggin23044502011-01-07 17:49:31 +1100860 dentry->d_flags &= ~DCACHE_REFERENCED;
861 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400862 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100863 } else {
864 list_move_tail(&dentry->d_lru, &tmp);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100865 dentry->d_flags |= DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100866 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000867 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100868 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400869 }
Nick Pigginec336792011-01-07 17:49:47 +1100870 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400871 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700872 if (!list_empty(&referenced))
873 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100874 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100875
876 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877}
878
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700879/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 * shrink_dcache_sb - shrink dcache for a superblock
881 * @sb: superblock
882 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400883 * Shrink the dcache for the specified super block. This is used to free
884 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400886void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400888 LIST_HEAD(tmp);
889
Nick Piggin23044502011-01-07 17:49:31 +1100890 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400891 while (!list_empty(&sb->s_dentry_lru)) {
892 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100893 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400894 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100895 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400896 }
Nick Piggin23044502011-01-07 17:49:31 +1100897 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700899EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901/*
David Howellsc636ebd2006-10-11 01:22:19 -0700902 * destroy a single subtree of dentries for unmount
903 * - see the comments on shrink_dcache_for_umount() for a description of the
904 * locking
905 */
906static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
907{
908 struct dentry *parent;
909
910 BUG_ON(!IS_ROOT(dentry));
911
David Howellsc636ebd2006-10-11 01:22:19 -0700912 for (;;) {
913 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100914 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700915 dentry = list_entry(dentry->d_subdirs.next,
916 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700917
918 /* consume the dentries from this leaf up through its parents
919 * until we find one with children or run out altogether */
920 do {
921 struct inode *inode;
922
Sage Weilf0023bc2011-10-28 10:02:42 -0700923 /*
924 * remove the dentry from the lru, and inform
925 * the fs that this dentry is about to be
926 * unhashed and destroyed.
927 */
928 dentry_lru_prune(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100929 __d_shrink(dentry);
930
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100931 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700932 printk(KERN_ERR
933 "BUG: Dentry %p{i=%lx,n=%s}"
934 " still in use (%d)"
935 " [unmount of %s %s]\n",
936 dentry,
937 dentry->d_inode ?
938 dentry->d_inode->i_ino : 0UL,
939 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100940 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700941 dentry->d_sb->s_type->name,
942 dentry->d_sb->s_id);
943 BUG();
944 }
945
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100946 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700947 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100948 list_del(&dentry->d_u.d_child);
949 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900950 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100951 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100952 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900953 }
David Howellsc636ebd2006-10-11 01:22:19 -0700954
David Howellsc636ebd2006-10-11 01:22:19 -0700955 inode = dentry->d_inode;
956 if (inode) {
957 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400958 hlist_del_init(&dentry->d_alias);
David Howellsc636ebd2006-10-11 01:22:19 -0700959 if (dentry->d_op && dentry->d_op->d_iput)
960 dentry->d_op->d_iput(dentry, inode);
961 else
962 iput(inode);
963 }
964
965 d_free(dentry);
966
967 /* finished when we fall off the top of the tree,
968 * otherwise we ascend to the parent and move to the
969 * next sibling if there is one */
970 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400971 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700972 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700973 } while (list_empty(&dentry->d_subdirs));
974
975 dentry = list_entry(dentry->d_subdirs.next,
976 struct dentry, d_u.d_child);
977 }
978}
979
980/*
981 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100982 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700983 * - the superblock is detached from all mountings and open files, so the
984 * dentry trees will not be rearranged by the VFS
985 * - s_umount is write-locked, so the memory pressure shrinker will ignore
986 * any dentries belonging to this superblock that it comes across
987 * - the filesystem itself is no longer permitted to rearrange the dentries
988 * in this superblock
989 */
990void shrink_dcache_for_umount(struct super_block *sb)
991{
992 struct dentry *dentry;
993
994 if (down_read_trylock(&sb->s_umount))
995 BUG();
996
997 dentry = sb->s_root;
998 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100999 dentry->d_count--;
David Howellsc636ebd2006-10-11 01:22:19 -07001000 shrink_dcache_for_umount_subtree(dentry);
1001
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001002 while (!hlist_bl_empty(&sb->s_anon)) {
1003 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001004 shrink_dcache_for_umount_subtree(dentry);
1005 }
1006}
1007
1008/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001009 * This tries to ascend one level of parenthood, but
1010 * we can race with renaming, so we need to re-check
1011 * the parenthood after dropping the lock and check
1012 * that the sequence number still matches.
1013 */
1014static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1015{
1016 struct dentry *new = old->d_parent;
1017
1018 rcu_read_lock();
1019 spin_unlock(&old->d_lock);
1020 spin_lock(&new->d_lock);
1021
1022 /*
1023 * might go back up the wrong parent if we have had a rename
1024 * or deletion
1025 */
1026 if (new != old->d_parent ||
Miklos Szeredib161dfa62012-09-17 22:31:38 +02001027 (old->d_flags & DCACHE_DENTRY_KILLED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001028 (!locked && read_seqretry(&rename_lock, seq))) {
1029 spin_unlock(&new->d_lock);
1030 new = NULL;
1031 }
1032 rcu_read_unlock();
1033 return new;
1034}
1035
1036
1037/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 * Search for at least 1 mount point in the dentry's subdirs.
1039 * We descend to the next level whenever the d_subdirs
1040 * list is non-empty and continue searching.
1041 */
1042
1043/**
1044 * have_submounts - check for mounts over a dentry
1045 * @parent: dentry to check.
1046 *
1047 * Return true if the parent or its subdirectories contain
1048 * a mount point
1049 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050int have_submounts(struct dentry *parent)
1051{
Nick Piggin949854d2011-01-07 17:49:37 +11001052 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001054 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001055 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001056
Nick Piggin949854d2011-01-07 17:49:37 +11001057 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001058again:
1059 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (d_mountpoint(parent))
1062 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001063 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064repeat:
1065 next = this_parent->d_subdirs.next;
1066resume:
1067 while (next != &this_parent->d_subdirs) {
1068 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001069 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001071
1072 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001074 if (d_mountpoint(dentry)) {
1075 spin_unlock(&dentry->d_lock);
1076 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001078 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001080 spin_unlock(&this_parent->d_lock);
1081 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001083 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 goto repeat;
1085 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001086 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
1088 /*
1089 * All done at this level ... ascend and resume the search.
1090 */
1091 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001092 struct dentry *child = this_parent;
1093 this_parent = try_to_ascend(this_parent, locked, seq);
1094 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001095 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001096 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 goto resume;
1098 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001099 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001100 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001101 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001102 if (locked)
1103 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 return 0; /* No mount points found in tree */
1105positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001106 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001107 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001108 if (locked)
1109 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001111
1112rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02001113 if (locked)
1114 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001115 locked = 1;
1116 write_seqlock(&rename_lock);
1117 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001119EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001122 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 * and move any unused dentries to the end of the unused
1124 * list for prune_dcache(). We descend to the next level
1125 * whenever the d_subdirs list is non-empty and continue
1126 * searching.
1127 *
1128 * It returns zero iff there are no unused children,
1129 * otherwise it returns the number of children moved to
1130 * the end of the unused list. This may not be the total
1131 * number of unused children, because select_parent can
1132 * drop the lock and return early due to latency
1133 * constraints.
1134 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001135static int select_parent(struct dentry *parent, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Nick Piggin949854d2011-01-07 17:49:37 +11001137 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001139 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001141 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Nick Piggin949854d2011-01-07 17:49:37 +11001143 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001144again:
1145 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001146 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147repeat:
1148 next = this_parent->d_subdirs.next;
1149resume:
1150 while (next != &this_parent->d_subdirs) {
1151 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001152 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 next = tmp->next;
1154
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001155 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001156
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001157 /*
1158 * move only zero ref count dentries to the dispose list.
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001159 *
1160 * Those which are presently on the shrink list, being processed
1161 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1162 * loop in shrink_dcache_parent() might not make any progress
1163 * and loop forever.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 */
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001165 if (dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001166 dentry_lru_del(dentry);
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001167 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1168 dentry_lru_move_list(dentry, dispose);
1169 dentry->d_flags |= DCACHE_SHRINK_LIST;
1170 found++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 /*
1173 * We can return to the caller if we have found some (this
1174 * ensures forward progress). We'll be coming back to find
1175 * the rest.
1176 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001177 if (found && need_resched()) {
1178 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 /*
1183 * Descend a level if the d_subdirs list is non-empty.
1184 */
1185 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001186 spin_unlock(&this_parent->d_lock);
1187 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001189 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 goto repeat;
1191 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001192
1193 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 }
1195 /*
1196 * All done at this level ... ascend and resume the search.
1197 */
1198 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001199 struct dentry *child = this_parent;
1200 this_parent = try_to_ascend(this_parent, locked, seq);
1201 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001202 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001203 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 goto resume;
1205 }
1206out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001207 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001208 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001209 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001210 if (locked)
1211 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001213
1214rename_retry:
1215 if (found)
1216 return found;
Miklos Szeredi8110e162012-09-17 22:23:30 +02001217 if (locked)
1218 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001219 locked = 1;
1220 write_seqlock(&rename_lock);
1221 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222}
1223
1224/**
1225 * shrink_dcache_parent - prune dcache
1226 * @parent: parent of entries to prune
1227 *
1228 * Prune the dcache to remove unused children of the parent dentry.
1229 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230void shrink_dcache_parent(struct dentry * parent)
1231{
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001232 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 int found;
1234
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001235 while ((found = select_parent(parent, &dispose)) != 0)
1236 shrink_dentry_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001238EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240/**
Al Viroa4464db2011-07-07 15:03:58 -04001241 * __d_alloc - allocate a dcache entry
1242 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 * @name: qstr of the name
1244 *
1245 * Allocates a dentry. It returns %NULL if there is insufficient memory
1246 * available. On a success the dentry is returned. The name passed in is
1247 * copied and the copy passed in may be reused after this call.
1248 */
1249
Al Viroa4464db2011-07-07 15:03:58 -04001250struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251{
1252 struct dentry *dentry;
1253 char *dname;
1254
Mel Gormane12ba742007-10-16 01:25:52 -07001255 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 if (!dentry)
1257 return NULL;
1258
Linus Torvalds6326c712012-05-21 16:14:04 -07001259 /*
1260 * We guarantee that the inline name is always NUL-terminated.
1261 * This way the memcpy() done by the name switching in rename
1262 * will still always have a NUL at the end, even if we might
1263 * be overwriting an internal NUL character
1264 */
1265 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 if (name->len > DNAME_INLINE_LEN-1) {
1267 dname = kmalloc(name->len + 1, GFP_KERNEL);
1268 if (!dname) {
1269 kmem_cache_free(dentry_cache, dentry);
1270 return NULL;
1271 }
1272 } else {
1273 dname = dentry->d_iname;
1274 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
1276 dentry->d_name.len = name->len;
1277 dentry->d_name.hash = name->hash;
1278 memcpy(dname, name->name, name->len);
1279 dname[name->len] = 0;
1280
Linus Torvalds6326c712012-05-21 16:14:04 -07001281 /* Make sure we always see the terminating NUL character */
1282 smp_wmb();
1283 dentry->d_name.name = dname;
1284
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001285 dentry->d_count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001286 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001288 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001290 dentry->d_parent = dentry;
1291 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 dentry->d_op = NULL;
1293 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001294 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 INIT_LIST_HEAD(&dentry->d_lru);
1296 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001297 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001298 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001299 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Nick Piggin3e880fb2011-01-07 17:49:19 +11001301 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 return dentry;
1304}
Al Viroa4464db2011-07-07 15:03:58 -04001305
1306/**
1307 * d_alloc - allocate a dcache entry
1308 * @parent: parent of entry to allocate
1309 * @name: qstr of the name
1310 *
1311 * Allocates a dentry. It returns %NULL if there is insufficient memory
1312 * available. On a success the dentry is returned. The name passed in is
1313 * copied and the copy passed in may be reused after this call.
1314 */
1315struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1316{
1317 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1318 if (!dentry)
1319 return NULL;
1320
1321 spin_lock(&parent->d_lock);
1322 /*
1323 * don't need child lock because it is not subject
1324 * to concurrency here
1325 */
1326 __dget_dlock(parent);
1327 dentry->d_parent = parent;
1328 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1329 spin_unlock(&parent->d_lock);
1330
1331 return dentry;
1332}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001333EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Nick Piggin4b936882011-01-07 17:50:07 +11001335struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1336{
Al Viroa4464db2011-07-07 15:03:58 -04001337 struct dentry *dentry = __d_alloc(sb, name);
1338 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001339 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001340 return dentry;
1341}
1342EXPORT_SYMBOL(d_alloc_pseudo);
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1345{
1346 struct qstr q;
1347
1348 q.name = name;
1349 q.len = strlen(name);
1350 q.hash = full_name_hash(q.name, q.len);
1351 return d_alloc(parent, &q);
1352}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001353EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Nick Pigginfb045ad2011-01-07 17:49:55 +11001355void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1356{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001357 WARN_ON_ONCE(dentry->d_op);
1358 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001359 DCACHE_OP_COMPARE |
1360 DCACHE_OP_REVALIDATE |
1361 DCACHE_OP_DELETE ));
1362 dentry->d_op = op;
1363 if (!op)
1364 return;
1365 if (op->d_hash)
1366 dentry->d_flags |= DCACHE_OP_HASH;
1367 if (op->d_compare)
1368 dentry->d_flags |= DCACHE_OP_COMPARE;
1369 if (op->d_revalidate)
1370 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1371 if (op->d_delete)
1372 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001373 if (op->d_prune)
1374 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001375
1376}
1377EXPORT_SYMBOL(d_set_d_op);
1378
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001379static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1380{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001381 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001382 if (inode) {
1383 if (unlikely(IS_AUTOMOUNT(inode)))
1384 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
Al Virob3d9b7a2012-06-09 13:51:19 -04001385 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001386 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001387 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001388 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001389 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001390 fsnotify_d_instantiate(dentry, inode);
1391}
1392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393/**
1394 * d_instantiate - fill in inode information for a dentry
1395 * @entry: dentry to complete
1396 * @inode: inode to attach to this dentry
1397 *
1398 * Fill in inode information in the entry.
1399 *
1400 * This turns negative dentries into productive full members
1401 * of society.
1402 *
1403 * NOTE! This assumes that the inode count has been incremented
1404 * (or otherwise set) by the caller to indicate that it is now
1405 * in use by the dcache.
1406 */
1407
1408void d_instantiate(struct dentry *entry, struct inode * inode)
1409{
Al Virob3d9b7a2012-06-09 13:51:19 -04001410 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001411 if (inode)
1412 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001413 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001414 if (inode)
1415 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 security_d_instantiate(entry, inode);
1417}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001418EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420/**
1421 * d_instantiate_unique - instantiate a non-aliased dentry
1422 * @entry: dentry to instantiate
1423 * @inode: inode to attach to this dentry
1424 *
1425 * Fill in inode information in the entry. On success, it returns NULL.
1426 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001427 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 *
1429 * Note that in order to avoid conflicts with rename() etc, the caller
1430 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001431 *
1432 * This also assumes that the inode count has been incremented
1433 * (or otherwise set) by the caller to indicate that it is now
1434 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 */
David Howells770bfad2006-08-22 20:06:07 -04001436static struct dentry *__d_instantiate_unique(struct dentry *entry,
1437 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438{
1439 struct dentry *alias;
1440 int len = entry->d_name.len;
1441 const char *name = entry->d_name.name;
1442 unsigned int hash = entry->d_name.hash;
Al Virob3d9b7a2012-06-09 13:51:19 -04001443 struct hlist_node *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
David Howells770bfad2006-08-22 20:06:07 -04001445 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001446 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001447 return NULL;
1448 }
1449
Al Virob3d9b7a2012-06-09 13:51:19 -04001450 hlist_for_each_entry(alias, p, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001451 /*
1452 * Don't need alias->d_lock here, because aliases with
1453 * d_parent == entry->d_parent are not subject to name or
1454 * parent changes, because the parent inode i_mutex is held.
1455 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001456 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 continue;
1458 if (alias->d_parent != entry->d_parent)
1459 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001460 if (alias->d_name.len != len)
1461 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001462 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001464 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return alias;
1466 }
David Howells770bfad2006-08-22 20:06:07 -04001467
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001468 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 return NULL;
1470}
David Howells770bfad2006-08-22 20:06:07 -04001471
1472struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1473{
1474 struct dentry *result;
1475
Al Virob3d9b7a2012-06-09 13:51:19 -04001476 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001477
Nick Piggin873feea2011-01-07 17:50:06 +11001478 if (inode)
1479 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001480 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001481 if (inode)
1482 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001483
1484 if (!result) {
1485 security_d_instantiate(entry, inode);
1486 return NULL;
1487 }
1488
1489 BUG_ON(!d_unhashed(result));
1490 iput(inode);
1491 return result;
1492}
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494EXPORT_SYMBOL(d_instantiate_unique);
1495
Al Viroadc0e912012-01-08 16:49:21 -05001496struct dentry *d_make_root(struct inode *root_inode)
1497{
1498 struct dentry *res = NULL;
1499
1500 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001501 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001502
1503 res = __d_alloc(root_inode->i_sb, &name);
1504 if (res)
1505 d_instantiate(res, root_inode);
1506 else
1507 iput(root_inode);
1508 }
1509 return res;
1510}
1511EXPORT_SYMBOL(d_make_root);
1512
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001513static struct dentry * __d_find_any_alias(struct inode *inode)
1514{
1515 struct dentry *alias;
1516
Al Virob3d9b7a2012-06-09 13:51:19 -04001517 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001518 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001519 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001520 __dget(alias);
1521 return alias;
1522}
1523
Sage Weil46f72b32012-01-10 09:04:37 -08001524/**
1525 * d_find_any_alias - find any alias for a given inode
1526 * @inode: inode to find an alias for
1527 *
1528 * If any aliases exist for the given inode, take and return a
1529 * reference for one of them. If no aliases exist, return %NULL.
1530 */
1531struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001532{
1533 struct dentry *de;
1534
1535 spin_lock(&inode->i_lock);
1536 de = __d_find_any_alias(inode);
1537 spin_unlock(&inode->i_lock);
1538 return de;
1539}
Sage Weil46f72b32012-01-10 09:04:37 -08001540EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001543 * d_obtain_alias - find or allocate a dentry for a given inode
1544 * @inode: inode to allocate the dentry for
1545 *
1546 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1547 * similar open by handle operations. The returned dentry may be anonymous,
1548 * or may have a full name (if the inode was already in the cache).
1549 *
1550 * When called on a directory inode, we must ensure that the inode only ever
1551 * has one dentry. If a dentry is found, that is returned instead of
1552 * allocating a new one.
1553 *
1554 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001555 * to the dentry. In case of an error the reference on the inode is released.
1556 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1557 * be passed in and will be the error will be propagate to the return value,
1558 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001559 */
1560struct dentry *d_obtain_alias(struct inode *inode)
1561{
NeilBrownb911a6b2012-11-08 16:09:37 -08001562 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001563 struct dentry *tmp;
1564 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001565
1566 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001567 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001568 if (IS_ERR(inode))
1569 return ERR_CAST(inode);
1570
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001571 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001572 if (res)
1573 goto out_iput;
1574
Al Viroa4464db2011-07-07 15:03:58 -04001575 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001576 if (!tmp) {
1577 res = ERR_PTR(-ENOMEM);
1578 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001579 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001580
Nick Piggin873feea2011-01-07 17:50:06 +11001581 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001582 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001583 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001584 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001585 dput(tmp);
1586 goto out_iput;
1587 }
1588
1589 /* attach a disconnected dentry */
1590 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001591 tmp->d_inode = inode;
1592 tmp->d_flags |= DCACHE_DISCONNECTED;
Al Virob3d9b7a2012-06-09 13:51:19 -04001593 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001594 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001595 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001596 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001597 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001598 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001599 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001600
Christoph Hellwig9308a612008-08-11 15:49:12 +02001601 return tmp;
1602
1603 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001604 if (res && !IS_ERR(res))
1605 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001606 iput(inode);
1607 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001608}
Benny Halevyadc48722009-02-27 14:02:59 -08001609EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611/**
1612 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1613 * @inode: the inode which may have a disconnected dentry
1614 * @dentry: a negative dentry which we want to point to the inode.
1615 *
1616 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1617 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1618 * and return it, else simply d_add the inode to the dentry and return NULL.
1619 *
1620 * This is needed in the lookup routine of any filesystem that is exportable
1621 * (via knfsd) so that we can build dcache paths to directories effectively.
1622 *
1623 * If a dentry was found and moved, then it is returned. Otherwise NULL
1624 * is returned. This matches the expected return value of ->lookup.
1625 *
1626 */
1627struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1628{
1629 struct dentry *new = NULL;
1630
Al Viroa9049372011-07-08 21:20:11 -04001631 if (IS_ERR(inode))
1632 return ERR_CAST(inode);
1633
NeilBrown21c0d8f2006-10-04 02:16:16 -07001634 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001635 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001636 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001638 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001639 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 d_move(new, dentry);
1642 iput(inode);
1643 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001644 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001645 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001646 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 security_d_instantiate(dentry, inode);
1648 d_rehash(dentry);
1649 }
1650 } else
1651 d_add(dentry, inode);
1652 return new;
1653}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001654EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Barry Naujok94035402008-05-21 16:50:46 +10001656/**
1657 * d_add_ci - lookup or allocate new dentry with case-exact name
1658 * @inode: the inode case-insensitive lookup has found
1659 * @dentry: the negative dentry that was passed to the parent's lookup func
1660 * @name: the case-exact name to be associated with the returned dentry
1661 *
1662 * This is to avoid filling the dcache with case-insensitive names to the
1663 * same inode, only the actual correct case is stored in the dcache for
1664 * case-insensitive filesystems.
1665 *
1666 * For a case-insensitive lookup match and if the the case-exact dentry
1667 * already exists in in the dcache, use it and return it.
1668 *
1669 * If no entry exists with the exact case name, allocate new dentry with
1670 * the exact case, and return the spliced entry.
1671 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001672struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001673 struct qstr *name)
1674{
1675 int error;
1676 struct dentry *found;
1677 struct dentry *new;
1678
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001679 /*
1680 * First check if a dentry matching the name already exists,
1681 * if not go ahead and create it now.
1682 */
Barry Naujok94035402008-05-21 16:50:46 +10001683 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10001684 if (!found) {
1685 new = d_alloc(dentry->d_parent, name);
1686 if (!new) {
1687 error = -ENOMEM;
1688 goto err_out;
1689 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001690
Barry Naujok94035402008-05-21 16:50:46 +10001691 found = d_splice_alias(inode, new);
1692 if (found) {
1693 dput(new);
1694 return found;
1695 }
1696 return new;
1697 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001698
1699 /*
1700 * If a matching dentry exists, and it's not negative use it.
1701 *
1702 * Decrement the reference count to balance the iget() done
1703 * earlier on.
1704 */
Barry Naujok94035402008-05-21 16:50:46 +10001705 if (found->d_inode) {
1706 if (unlikely(found->d_inode != inode)) {
1707 /* This can't happen because bad inodes are unhashed. */
1708 BUG_ON(!is_bad_inode(inode));
1709 BUG_ON(!is_bad_inode(found->d_inode));
1710 }
Barry Naujok94035402008-05-21 16:50:46 +10001711 iput(inode);
1712 return found;
1713 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001714
Barry Naujok94035402008-05-21 16:50:46 +10001715 /*
1716 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001717 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001718 */
Al Viro4513d892011-07-17 10:52:14 -04001719 new = d_splice_alias(inode, found);
1720 if (new) {
1721 dput(found);
1722 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001723 }
Al Viro4513d892011-07-17 10:52:14 -04001724 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001725
1726err_out:
1727 iput(inode);
1728 return ERR_PTR(error);
1729}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001730EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001732/*
1733 * Do the slow-case of the dentry name compare.
1734 *
1735 * Unlike the dentry_cmp() function, we need to atomically
1736 * load the name, length and inode information, so that the
1737 * filesystem can rely on them, and can use the 'name' and
1738 * 'len' information without worrying about walking off the
1739 * end of memory etc.
1740 *
1741 * Thus the read_seqcount_retry() and the "duplicate" info
1742 * in arguments (the low-level filesystem should not look
1743 * at the dentry inode or name contents directly, since
1744 * rename can change them while we're in RCU mode).
1745 */
1746enum slow_d_compare {
1747 D_COMP_OK,
1748 D_COMP_NOMATCH,
1749 D_COMP_SEQRETRY,
1750};
1751
1752static noinline enum slow_d_compare slow_dentry_cmp(
1753 const struct dentry *parent,
1754 struct inode *inode,
1755 struct dentry *dentry,
1756 unsigned int seq,
1757 const struct qstr *name)
1758{
1759 int tlen = dentry->d_name.len;
1760 const char *tname = dentry->d_name.name;
1761 struct inode *i = dentry->d_inode;
1762
1763 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1764 cpu_relax();
1765 return D_COMP_SEQRETRY;
1766 }
1767 if (parent->d_op->d_compare(parent, inode,
1768 dentry, i,
1769 tlen, tname, name))
1770 return D_COMP_NOMATCH;
1771 return D_COMP_OK;
1772}
1773
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001775 * __d_lookup_rcu - search for a dentry (racy, store-free)
1776 * @parent: parent dentry
1777 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001778 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001779 * @inode: returns dentry->d_inode when the inode was found valid.
1780 * Returns: dentry, or NULL
1781 *
1782 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1783 * resolution (store-free path walking) design described in
1784 * Documentation/filesystems/path-lookup.txt.
1785 *
1786 * This is not to be used outside core vfs.
1787 *
1788 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1789 * held, and rcu_read_lock held. The returned dentry must not be stored into
1790 * without taking d_lock and checking d_seq sequence count against @seq
1791 * returned here.
1792 *
1793 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1794 * function.
1795 *
1796 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1797 * the returned dentry, so long as its parent's seqlock is checked after the
1798 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1799 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001800 *
1801 * NOTE! The caller *has* to check the resulting dentry against the sequence
1802 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001803 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001804struct dentry *__d_lookup_rcu(const struct dentry *parent,
1805 const struct qstr *name,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001806 unsigned *seqp, struct inode *inode)
Nick Piggin31e6b012011-01-07 17:49:52 +11001807{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001808 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001809 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001810 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001811 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001812 struct dentry *dentry;
1813
1814 /*
1815 * Note: There is significant duplication with __d_lookup_rcu which is
1816 * required to prevent single threaded performance regressions
1817 * especially on architectures where smp_rmb (in seqcounts) are costly.
1818 * Keep the two functions in sync.
1819 */
1820
1821 /*
1822 * The hash list is protected using RCU.
1823 *
1824 * Carefully use d_seq when comparing a candidate dentry, to avoid
1825 * races with d_move().
1826 *
1827 * It is possible that concurrent renames can mess up our list
1828 * walk here and result in missing our dentry, resulting in the
1829 * false-negative result. d_lookup() protects against concurrent
1830 * renames using rename_lock seqlock.
1831 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001832 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11001833 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001834 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08001835 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001836
Nick Piggin31e6b012011-01-07 17:49:52 +11001837seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001838 /*
1839 * The dentry sequence count protects us from concurrent
1840 * renames, and thus protects inode, parent and name fields.
1841 *
1842 * The caller must perform a seqcount check in order
1843 * to do anything useful with the returned dentry,
1844 * including using the 'd_inode' pointer.
1845 *
1846 * NOTE! We do a "raw" seqcount_begin here. That means that
1847 * we don't wait for the sequence count to stabilize if it
1848 * is in the middle of a sequence change. If we do the slow
1849 * dentry compare, we will do seqretries until it is stable,
1850 * and if we end up with a successful lookup, we actually
1851 * want to exit RCU lookup anyway.
1852 */
1853 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001854 if (dentry->d_parent != parent)
1855 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07001856 if (d_unhashed(dentry))
1857 continue;
Linus Torvalds8966be92012-03-02 14:23:30 -08001858 *seqp = seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001859
1860 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001861 if (dentry->d_name.hash != hashlen_hash(hashlen))
1862 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001863 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1864 case D_COMP_OK:
1865 return dentry;
1866 case D_COMP_NOMATCH:
1867 continue;
1868 default:
1869 goto seqretry;
1870 }
1871 }
1872
Linus Torvalds26fe5752012-05-10 13:14:12 -07001873 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07001874 continue;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001875 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001876 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11001877 }
1878 return NULL;
1879}
1880
1881/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 * d_lookup - search for a dentry
1883 * @parent: parent dentry
1884 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001885 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001887 * d_lookup searches the children of the parent dentry for the name in
1888 * question. If the dentry is found its reference count is incremented and the
1889 * dentry is returned. The caller must use dput to free the entry when it has
1890 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001892struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
Nick Piggin31e6b012011-01-07 17:49:52 +11001894 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001895 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897 do {
1898 seq = read_seqbegin(&rename_lock);
1899 dentry = __d_lookup(parent, name);
1900 if (dentry)
1901 break;
1902 } while (read_seqretry(&rename_lock, seq));
1903 return dentry;
1904}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001905EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906
Nick Piggin31e6b012011-01-07 17:49:52 +11001907/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001908 * __d_lookup - search for a dentry (racy)
1909 * @parent: parent dentry
1910 * @name: qstr of name we wish to find
1911 * Returns: dentry, or NULL
1912 *
1913 * __d_lookup is like d_lookup, however it may (rarely) return a
1914 * false-negative result due to unrelated rename activity.
1915 *
1916 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1917 * however it must be used carefully, eg. with a following d_lookup in
1918 * the case of failure.
1919 *
1920 * __d_lookup callers must be commented.
1921 */
Nick Piggin31e6b012011-01-07 17:49:52 +11001922struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923{
1924 unsigned int len = name->len;
1925 unsigned int hash = name->hash;
1926 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001927 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001928 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001929 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001930 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
Nick Pigginb04f7842010-08-18 04:37:34 +10001932 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001933 * Note: There is significant duplication with __d_lookup_rcu which is
1934 * required to prevent single threaded performance regressions
1935 * especially on architectures where smp_rmb (in seqcounts) are costly.
1936 * Keep the two functions in sync.
1937 */
1938
1939 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001940 * The hash list is protected using RCU.
1941 *
1942 * Take d_lock when comparing a candidate dentry, to avoid races
1943 * with d_move().
1944 *
1945 * It is possible that concurrent renames can mess up our list
1946 * walk here and result in missing our dentry, resulting in the
1947 * false-negative result. d_lookup() protects against concurrent
1948 * renames using rename_lock seqlock.
1949 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001950 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001951 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 rcu_read_lock();
1953
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001954 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 if (dentry->d_name.hash != hash)
1957 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
1959 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 if (dentry->d_parent != parent)
1961 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001962 if (d_unhashed(dentry))
1963 goto next;
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 /*
1966 * It is safe to compare names since d_move() cannot
1967 * change the qstr (protected by d_lock).
1968 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11001969 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001970 int tlen = dentry->d_name.len;
1971 const char *tname = dentry->d_name.name;
Nick Piggin621e1552011-01-07 17:49:27 +11001972 if (parent->d_op->d_compare(parent, parent->d_inode,
1973 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001974 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 goto next;
1976 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07001977 if (dentry->d_name.len != len)
1978 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001979 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 goto next;
1981 }
1982
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001983 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001984 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 spin_unlock(&dentry->d_lock);
1986 break;
1987next:
1988 spin_unlock(&dentry->d_lock);
1989 }
1990 rcu_read_unlock();
1991
1992 return found;
1993}
1994
1995/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001996 * d_hash_and_lookup - hash the qstr then search for a dentry
1997 * @dir: Directory to search in
1998 * @name: qstr of name we wish to find
1999 *
2000 * On hash failure or on lookup failure NULL is returned.
2001 */
2002struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2003{
2004 struct dentry *dentry = NULL;
2005
2006 /*
2007 * Check for a fs-specific hash function. Note that we must
2008 * calculate the standard hash first, as the d_op->d_hash()
2009 * routine may choose to leave the hash value unchanged.
2010 */
2011 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002012 if (dir->d_flags & DCACHE_OP_HASH) {
Nick Pigginb1e6a012011-01-07 17:49:28 +11002013 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002014 goto out;
2015 }
2016 dentry = d_lookup(dir, name);
2017out:
2018 return dentry;
2019}
2020
2021/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002022 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002024 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 *
2026 * An insecure source has sent us a dentry, here we verify it and dget() it.
2027 * This is used by ncpfs in its readdir implementation.
2028 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002029 *
2030 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002032int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033{
Nick Piggin786a5e12011-01-07 17:49:16 +11002034 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002035
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002036 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002037 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2038 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002039 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002040 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002041 spin_unlock(&dentry->d_lock);
2042 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return 1;
2044 }
2045 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002046 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 return 0;
2049}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002050EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
2052/*
2053 * When a file is deleted, we have two options:
2054 * - turn this dentry into a negative dentry
2055 * - unhash this dentry and free it.
2056 *
2057 * Usually, we want to just turn this into
2058 * a negative dentry, but if anybody else is
2059 * currently using the dentry or the inode
2060 * we can't do that and we fall back on removing
2061 * it from the hash queues and waiting for
2062 * it to be deleted later when it has no users
2063 */
2064
2065/**
2066 * d_delete - delete a dentry
2067 * @dentry: The dentry to delete
2068 *
2069 * Turn the dentry into a negative dentry if possible, otherwise
2070 * remove it from the hash queues so it can be deleted later
2071 */
2072
2073void d_delete(struct dentry * dentry)
2074{
Nick Piggin873feea2011-01-07 17:50:06 +11002075 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002076 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 /*
2078 * Are we the only user?
2079 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002080again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002082 inode = dentry->d_inode;
2083 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002084 if (dentry->d_count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002085 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002086 spin_unlock(&dentry->d_lock);
2087 cpu_relax();
2088 goto again;
2089 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002090 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002091 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002092 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 return;
2094 }
2095
2096 if (!d_unhashed(dentry))
2097 __d_drop(dentry);
2098
2099 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002100
2101 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002103EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002105static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002107 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002108 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002109 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002110 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002111 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112}
2113
David Howells770bfad2006-08-22 20:06:07 -04002114static void _d_rehash(struct dentry * entry)
2115{
2116 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2117}
2118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119/**
2120 * d_rehash - add an entry back to the hash
2121 * @entry: dentry to add to the hash
2122 *
2123 * Adds a dentry to the hash according to its name.
2124 */
2125
2126void d_rehash(struct dentry * entry)
2127{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002129 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002132EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002134/**
2135 * dentry_update_name_case - update case insensitive dentry with a new name
2136 * @dentry: dentry to be updated
2137 * @name: new name
2138 *
2139 * Update a case insensitive dentry with new case of name.
2140 *
2141 * dentry must have been returned by d_lookup with name @name. Old and new
2142 * name lengths must match (ie. no d_compare which allows mismatched name
2143 * lengths).
2144 *
2145 * Parent inode i_mutex must be held over d_lookup and into this call (to
2146 * keep renames and concurrent inserts, and readdir(2) away).
2147 */
2148void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2149{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002150 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002151 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2152
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002153 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002154 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002155 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002156 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002157 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002158}
2159EXPORT_SYMBOL(dentry_update_name_case);
2160
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161static void switch_names(struct dentry *dentry, struct dentry *target)
2162{
2163 if (dname_external(target)) {
2164 if (dname_external(dentry)) {
2165 /*
2166 * Both external: swap the pointers
2167 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002168 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 } else {
2170 /*
2171 * dentry:internal, target:external. Steal target's
2172 * storage and make target internal.
2173 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002174 memcpy(target->d_iname, dentry->d_name.name,
2175 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 dentry->d_name.name = target->d_name.name;
2177 target->d_name.name = target->d_iname;
2178 }
2179 } else {
2180 if (dname_external(dentry)) {
2181 /*
2182 * dentry:external, target:internal. Give dentry's
2183 * storage to target and make dentry internal
2184 */
2185 memcpy(dentry->d_iname, target->d_name.name,
2186 target->d_name.len + 1);
2187 target->d_name.name = dentry->d_name.name;
2188 dentry->d_name.name = dentry->d_iname;
2189 } else {
2190 /*
2191 * Both are internal. Just copy target to dentry
2192 */
2193 memcpy(dentry->d_iname, target->d_name.name,
2194 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002195 dentry->d_name.len = target->d_name.len;
2196 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 }
2198 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002199 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200}
2201
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002202static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2203{
2204 /*
2205 * XXXX: do we really need to take target->d_lock?
2206 */
2207 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2208 spin_lock(&target->d_parent->d_lock);
2209 else {
2210 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2211 spin_lock(&dentry->d_parent->d_lock);
2212 spin_lock_nested(&target->d_parent->d_lock,
2213 DENTRY_D_LOCK_NESTED);
2214 } else {
2215 spin_lock(&target->d_parent->d_lock);
2216 spin_lock_nested(&dentry->d_parent->d_lock,
2217 DENTRY_D_LOCK_NESTED);
2218 }
2219 }
2220 if (target < dentry) {
2221 spin_lock_nested(&target->d_lock, 2);
2222 spin_lock_nested(&dentry->d_lock, 3);
2223 } else {
2224 spin_lock_nested(&dentry->d_lock, 2);
2225 spin_lock_nested(&target->d_lock, 3);
2226 }
2227}
2228
2229static void dentry_unlock_parents_for_move(struct dentry *dentry,
2230 struct dentry *target)
2231{
2232 if (target->d_parent != dentry->d_parent)
2233 spin_unlock(&dentry->d_parent->d_lock);
2234 if (target->d_parent != target)
2235 spin_unlock(&target->d_parent->d_lock);
2236}
2237
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002239 * When switching names, the actual string doesn't strictly have to
2240 * be preserved in the target - because we're dropping the target
2241 * anyway. As such, we can just do a simple memcpy() to copy over
2242 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002244 * Note that we have to be a lot more careful about getting the hash
2245 * switched - we have to switch the hash value properly even if it
2246 * then no longer matches the actual (corrupted) string of the target.
2247 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002249/*
Al Viro18367502011-07-12 21:42:24 -04002250 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 * @dentry: entry to move
2252 * @target: new dentry
2253 *
2254 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002255 * dcache entries should not be moved in this way. Caller must hold
2256 * rename_lock, the i_mutex of the source and target directories,
2257 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 */
Al Viro18367502011-07-12 21:42:24 -04002259static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 if (!dentry->d_inode)
2262 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2263
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002264 BUG_ON(d_ancestor(dentry, target));
2265 BUG_ON(d_ancestor(target, dentry));
2266
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002267 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268
Nick Piggin31e6b012011-01-07 17:49:52 +11002269 write_seqcount_begin(&dentry->d_seq);
2270 write_seqcount_begin(&target->d_seq);
2271
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002272 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2273
2274 /*
2275 * Move the dentry to the target hash queue. Don't bother checking
2276 * for the same hash queue because of how unlikely it is.
2277 */
2278 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002279 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 /* Unhash the target: dput() will then get rid of it */
2282 __d_drop(target);
2283
Eric Dumazet5160ee62006-01-08 01:03:32 -08002284 list_del(&dentry->d_u.d_child);
2285 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287 /* Switch the names.. */
2288 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002289 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 /* ... and switch the parents */
2292 if (IS_ROOT(dentry)) {
2293 dentry->d_parent = target->d_parent;
2294 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002295 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002297 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002300 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 }
2302
Eric Dumazet5160ee62006-01-08 01:03:32 -08002303 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002304
Nick Piggin31e6b012011-01-07 17:49:52 +11002305 write_seqcount_end(&target->d_seq);
2306 write_seqcount_end(&dentry->d_seq);
2307
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002308 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002310 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002312}
2313
2314/*
2315 * d_move - move a dentry
2316 * @dentry: entry to move
2317 * @target: new dentry
2318 *
2319 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002320 * dcache entries should not be moved in this way. See the locking
2321 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002322 */
2323void d_move(struct dentry *dentry, struct dentry *target)
2324{
2325 write_seqlock(&rename_lock);
2326 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002328}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002329EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002331/**
2332 * d_ancestor - search for an ancestor
2333 * @p1: ancestor dentry
2334 * @p2: child dentry
2335 *
2336 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2337 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002338 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002339struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002340{
2341 struct dentry *p;
2342
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002343 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002344 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002345 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002346 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002347 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002348}
2349
2350/*
2351 * This helper attempts to cope with remotely renamed directories
2352 *
2353 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002354 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002355 *
2356 * Note: If ever the locking in lock_rename() changes, then please
2357 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002358 */
Nick Piggin873feea2011-01-07 17:50:06 +11002359static struct dentry *__d_unalias(struct inode *inode,
2360 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002361{
2362 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002363 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002364
2365 /* If alias and dentry share a parent, then no extra locks required */
2366 if (alias->d_parent == dentry->d_parent)
2367 goto out_unalias;
2368
Trond Myklebust9eaef272006-10-21 10:24:20 -07002369 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002370 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2371 goto out_err;
2372 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2373 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2374 goto out_err;
2375 m2 = &alias->d_parent->d_inode->i_mutex;
2376out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002377 if (likely(!d_mountpoint(alias))) {
2378 __d_move(alias, dentry);
2379 ret = alias;
2380 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002381out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002382 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002383 if (m2)
2384 mutex_unlock(m2);
2385 if (m1)
2386 mutex_unlock(m1);
2387 return ret;
2388}
2389
2390/*
David Howells770bfad2006-08-22 20:06:07 -04002391 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2392 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002393 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002394 */
2395static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2396{
2397 struct dentry *dparent, *aparent;
2398
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002399 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002400
Nick Piggin31e6b012011-01-07 17:49:52 +11002401 write_seqcount_begin(&dentry->d_seq);
2402 write_seqcount_begin(&anon->d_seq);
2403
David Howells770bfad2006-08-22 20:06:07 -04002404 dparent = dentry->d_parent;
2405 aparent = anon->d_parent;
2406
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002407 switch_names(dentry, anon);
2408 swap(dentry->d_name.hash, anon->d_name.hash);
2409
David Howells770bfad2006-08-22 20:06:07 -04002410 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2411 list_del(&dentry->d_u.d_child);
2412 if (!IS_ROOT(dentry))
2413 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2414 else
2415 INIT_LIST_HEAD(&dentry->d_u.d_child);
2416
2417 anon->d_parent = (dparent == dentry) ? anon : dparent;
2418 list_del(&anon->d_u.d_child);
2419 if (!IS_ROOT(anon))
2420 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2421 else
2422 INIT_LIST_HEAD(&anon->d_u.d_child);
2423
Nick Piggin31e6b012011-01-07 17:49:52 +11002424 write_seqcount_end(&dentry->d_seq);
2425 write_seqcount_end(&anon->d_seq);
2426
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002427 dentry_unlock_parents_for_move(anon, dentry);
2428 spin_unlock(&dentry->d_lock);
2429
2430 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002431 anon->d_flags &= ~DCACHE_DISCONNECTED;
2432}
2433
2434/**
2435 * d_materialise_unique - introduce an inode into the tree
2436 * @dentry: candidate dentry
2437 * @inode: inode to bind to the dentry, to which aliases may be attached
2438 *
2439 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002440 * root directory alias in its place if there is one. Caller must hold the
2441 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002442 */
2443struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2444{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002445 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002446
2447 BUG_ON(!d_unhashed(dentry));
2448
David Howells770bfad2006-08-22 20:06:07 -04002449 if (!inode) {
2450 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002451 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002452 d_rehash(actual);
2453 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002454 }
2455
Nick Piggin873feea2011-01-07 17:50:06 +11002456 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002457
Trond Myklebust9eaef272006-10-21 10:24:20 -07002458 if (S_ISDIR(inode->i_mode)) {
2459 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002460
Trond Myklebust9eaef272006-10-21 10:24:20 -07002461 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002462 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002463 if (alias) {
2464 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002465 write_seqlock(&rename_lock);
2466
2467 if (d_ancestor(alias, dentry)) {
2468 /* Check for loops */
2469 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002470 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002471 } else if (IS_ROOT(alias)) {
2472 /* Is this an anonymous mountpoint that we
2473 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002474 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002475 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002476 __d_drop(alias);
2477 goto found;
Al Viro18367502011-07-12 21:42:24 -04002478 } else {
2479 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002480 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002481 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002482 }
Al Viro18367502011-07-12 21:42:24 -04002483 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002484 if (IS_ERR(actual)) {
2485 if (PTR_ERR(actual) == -ELOOP)
2486 pr_warn_ratelimited(
2487 "VFS: Lookup of '%s' in %s %s"
2488 " would have caused loop\n",
2489 dentry->d_name.name,
2490 inode->i_sb->s_type->name,
2491 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002492 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002493 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002494 goto out_nolock;
2495 }
David Howells770bfad2006-08-22 20:06:07 -04002496 }
2497
2498 /* Add a unique reference */
2499 actual = __d_instantiate_unique(dentry, inode);
2500 if (!actual)
2501 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002502 else
2503 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002504
David Howells770bfad2006-08-22 20:06:07 -04002505 spin_lock(&actual->d_lock);
2506found:
2507 _d_rehash(actual);
2508 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002509 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002510out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002511 if (actual == dentry) {
2512 security_d_instantiate(dentry, inode);
2513 return NULL;
2514 }
2515
2516 iput(inode);
2517 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002518}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002519EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002520
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002521static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002522{
2523 *buflen -= namelen;
2524 if (*buflen < 0)
2525 return -ENAMETOOLONG;
2526 *buffer -= namelen;
2527 memcpy(*buffer, str, namelen);
2528 return 0;
2529}
2530
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002531static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2532{
2533 return prepend(buffer, buflen, name->name, name->len);
2534}
2535
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002537 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002538 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002539 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002540 * @buffer: pointer to the end of the buffer
2541 * @buflen: pointer to buffer length
2542 *
Nick Piggin949854d2011-01-07 17:49:37 +11002543 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002544 */
Al Viro02125a82011-12-05 08:43:34 -05002545static int prepend_path(const struct path *path,
2546 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002547 char **buffer, int *buflen)
2548{
2549 struct dentry *dentry = path->dentry;
2550 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002551 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002552 bool slash = false;
2553 int error = 0;
2554
Andi Kleen962830d2012-05-08 13:32:02 +09302555 br_read_lock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002556 while (dentry != root->dentry || vfsmnt != root->mnt) {
2557 struct dentry * parent;
2558
2559 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2560 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002561 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002562 goto global_root;
Al Viroa73324d2011-11-24 22:25:07 -05002563 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002564 mnt = mnt->mnt_parent;
2565 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002566 continue;
2567 }
2568 parent = dentry->d_parent;
2569 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002570 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002571 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002572 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002573 if (!error)
2574 error = prepend(buffer, buflen, "/", 1);
2575 if (error)
2576 break;
2577
2578 slash = true;
2579 dentry = parent;
2580 }
2581
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002582 if (!error && !slash)
2583 error = prepend(buffer, buflen, "/", 1);
2584
Al Viro02125a82011-12-05 08:43:34 -05002585out:
Andi Kleen962830d2012-05-08 13:32:02 +09302586 br_read_unlock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002587 return error;
2588
2589global_root:
2590 /*
2591 * Filesystems needing to implement special "root names"
2592 * should do so with ->d_dname()
2593 */
2594 if (IS_ROOT(dentry) &&
2595 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2596 WARN(1, "Root dentry has weird name <%.*s>\n",
2597 (int) dentry->d_name.len, dentry->d_name.name);
2598 }
Al Viro02125a82011-12-05 08:43:34 -05002599 if (!slash)
2600 error = prepend(buffer, buflen, "/", 1);
2601 if (!error)
Al Virof7a99c52012-06-09 00:59:08 -04002602 error = is_mounted(vfsmnt) ? 1 : 2;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002603 goto out;
2604}
2605
2606/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002607 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002608 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002609 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002610 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 * @buflen: buffer length
2612 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002613 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002615 * Returns a pointer into the buffer or an error code if the
2616 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002617 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002618 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002619 *
Al Viro02125a82011-12-05 08:43:34 -05002620 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 */
Al Viro02125a82011-12-05 08:43:34 -05002622char *__d_path(const struct path *path,
2623 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002624 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002626 char *res = buf + buflen;
2627 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002629 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002630 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002631 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002632 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002633
Al Viro02125a82011-12-05 08:43:34 -05002634 if (error < 0)
2635 return ERR_PTR(error);
2636 if (error > 0)
2637 return NULL;
2638 return res;
2639}
2640
2641char *d_absolute_path(const struct path *path,
2642 char *buf, int buflen)
2643{
2644 struct path root = {};
2645 char *res = buf + buflen;
2646 int error;
2647
2648 prepend(&res, &buflen, "\0", 1);
2649 write_seqlock(&rename_lock);
2650 error = prepend_path(path, &root, &res, &buflen);
2651 write_sequnlock(&rename_lock);
2652
2653 if (error > 1)
2654 error = -EINVAL;
2655 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002656 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002657 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658}
2659
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002660/*
2661 * same as __d_path but appends "(deleted)" for unlinked files.
2662 */
Al Viro02125a82011-12-05 08:43:34 -05002663static int path_with_deleted(const struct path *path,
2664 const struct path *root,
2665 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002666{
2667 prepend(buf, buflen, "\0", 1);
2668 if (d_unlinked(path->dentry)) {
2669 int error = prepend(buf, buflen, " (deleted)", 10);
2670 if (error)
2671 return error;
2672 }
2673
2674 return prepend_path(path, root, buf, buflen);
2675}
2676
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002677static int prepend_unreachable(char **buffer, int *buflen)
2678{
2679 return prepend(buffer, buflen, "(unreachable)", 13);
2680}
2681
Jan Bluncka03a8a702008-02-14 19:38:32 -08002682/**
2683 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002684 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002685 * @buf: buffer to return value in
2686 * @buflen: buffer length
2687 *
2688 * Convert a dentry into an ASCII path name. If the entry has been deleted
2689 * the string " (deleted)" is appended. Note that this is ambiguous.
2690 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002691 * Returns a pointer into the buffer or an error code if the path was
2692 * too long. Note: Callers should use the returned pointer, not the passed
2693 * in buffer, to use the name! The implementation often starts at an offset
2694 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002695 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002696 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002697 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002698char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002700 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002701 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002702 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002704 /*
2705 * We have various synthetic filesystems that never get mounted. On
2706 * these filesystems dentries are never used for lookup purposes, and
2707 * thus don't need to be hashed. They also don't need a name until a
2708 * user wants to identify the object in /proc/pid/fd/. The little hack
2709 * below allows us to generate a name for these objects on demand:
2710 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002711 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2712 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002713
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002714 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002715 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002716 error = path_with_deleted(path, &root, &res, &buflen);
2717 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002718 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002719 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002720 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002721 return res;
2722}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002723EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002725/**
2726 * d_path_with_unreachable - return the path of a dentry
2727 * @path: path to report
2728 * @buf: buffer to return value in
2729 * @buflen: buffer length
2730 *
2731 * The difference from d_path() is that this prepends "(unreachable)"
2732 * to paths which are unreachable from the current process' root.
2733 */
2734char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2735{
2736 char *res = buf + buflen;
2737 struct path root;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002738 int error;
2739
2740 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2741 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2742
2743 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002744 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002745 error = path_with_deleted(path, &root, &res, &buflen);
2746 if (error > 0)
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002747 error = prepend_unreachable(&res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002748 write_sequnlock(&rename_lock);
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002749 path_put(&root);
2750 if (error)
2751 res = ERR_PTR(error);
2752
2753 return res;
2754}
2755
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002757 * Helper function for dentry_operations.d_dname() members
2758 */
2759char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2760 const char *fmt, ...)
2761{
2762 va_list args;
2763 char temp[64];
2764 int sz;
2765
2766 va_start(args, fmt);
2767 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2768 va_end(args);
2769
2770 if (sz > sizeof(temp) || sz > buflen)
2771 return ERR_PTR(-ENAMETOOLONG);
2772
2773 buffer += buflen - sz;
2774 return memcpy(buffer, temp, sz);
2775}
2776
2777/*
Ram Pai6092d042008-03-27 13:06:20 +01002778 * Write full pathname from the root of the filesystem into the buffer.
2779 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002780static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002781{
2782 char *end = buf + buflen;
2783 char *retval;
2784
Ram Pai6092d042008-03-27 13:06:20 +01002785 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002786 if (buflen < 1)
2787 goto Elong;
2788 /* Get '/' right */
2789 retval = end-1;
2790 *retval = '/';
2791
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002792 while (!IS_ROOT(dentry)) {
2793 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002794 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002795
Ram Pai6092d042008-03-27 13:06:20 +01002796 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002797 spin_lock(&dentry->d_lock);
2798 error = prepend_name(&end, &buflen, &dentry->d_name);
2799 spin_unlock(&dentry->d_lock);
2800 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002801 goto Elong;
2802
2803 retval = end;
2804 dentry = parent;
2805 }
Al Viroc1031352010-06-06 22:31:14 -04002806 return retval;
2807Elong:
2808 return ERR_PTR(-ENAMETOOLONG);
2809}
Nick Pigginec2447c2011-01-07 17:49:29 +11002810
2811char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2812{
2813 char *retval;
2814
Nick Piggin949854d2011-01-07 17:49:37 +11002815 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002816 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002817 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002818
2819 return retval;
2820}
2821EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002822
2823char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2824{
2825 char *p = NULL;
2826 char *retval;
2827
Nick Piggin949854d2011-01-07 17:49:37 +11002828 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002829 if (d_unlinked(dentry)) {
2830 p = buf + buflen;
2831 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2832 goto Elong;
2833 buflen++;
2834 }
2835 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002836 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002837 if (!IS_ERR(retval) && p)
2838 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002839 return retval;
2840Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002841 return ERR_PTR(-ENAMETOOLONG);
2842}
2843
2844/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845 * NOTE! The user-level library version returns a
2846 * character pointer. The kernel system call just
2847 * returns the length of the buffer filled (which
2848 * includes the ending '\0' character), or a negative
2849 * error value. So libc would do something like
2850 *
2851 * char *getcwd(char * buf, size_t size)
2852 * {
2853 * int retval;
2854 *
2855 * retval = sys_getcwd(buf, size);
2856 * if (retval >= 0)
2857 * return buf;
2858 * errno = -retval;
2859 * return NULL;
2860 * }
2861 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002862SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863{
Linus Torvalds552ce542007-02-13 12:08:18 -08002864 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002865 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002866 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867
2868 if (!page)
2869 return -ENOMEM;
2870
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002871 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
Linus Torvalds552ce542007-02-13 12:08:18 -08002873 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002874 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002875 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002876 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002877 char *cwd = page + PAGE_SIZE;
2878 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002880 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002881 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002882 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002883
Al Viro02125a82011-12-05 08:43:34 -05002884 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002885 goto out;
2886
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002887 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002888 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002889 error = prepend_unreachable(&cwd, &buflen);
2890 if (error)
2891 goto out;
2892 }
2893
Linus Torvalds552ce542007-02-13 12:08:18 -08002894 error = -ERANGE;
2895 len = PAGE_SIZE + page - cwd;
2896 if (len <= size) {
2897 error = len;
2898 if (copy_to_user(buf, cwd, len))
2899 error = -EFAULT;
2900 }
Nick Piggin949854d2011-01-07 17:49:37 +11002901 } else {
2902 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904
2905out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002906 path_put(&pwd);
2907 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 free_page((unsigned long) page);
2909 return error;
2910}
2911
2912/*
2913 * Test whether new_dentry is a subdirectory of old_dentry.
2914 *
2915 * Trivially implemented using the dcache structure
2916 */
2917
2918/**
2919 * is_subdir - is new dentry a subdirectory of old_dentry
2920 * @new_dentry: new dentry
2921 * @old_dentry: old dentry
2922 *
2923 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2924 * Returns 0 otherwise.
2925 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2926 */
2927
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002928int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929{
2930 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002931 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002933 if (new_dentry == old_dentry)
2934 return 1;
2935
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002936 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002939 /*
2940 * Need rcu_readlock to protect against the d_parent trashing
2941 * due to d_move
2942 */
2943 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002944 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002946 else
2947 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002948 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950
2951 return result;
2952}
2953
2954void d_genocide(struct dentry *root)
2955{
Nick Piggin949854d2011-01-07 17:49:37 +11002956 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002958 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002959 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960
Nick Piggin949854d2011-01-07 17:49:37 +11002961 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002962again:
2963 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002964 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965repeat:
2966 next = this_parent->d_subdirs.next;
2967resume:
2968 while (next != &this_parent->d_subdirs) {
2969 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002970 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002972
Nick Pigginda502952011-01-07 17:49:33 +11002973 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2974 if (d_unhashed(dentry) || !dentry->d_inode) {
2975 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002977 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002979 spin_unlock(&this_parent->d_lock);
2980 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002982 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 goto repeat;
2984 }
Nick Piggin949854d2011-01-07 17:49:37 +11002985 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2986 dentry->d_flags |= DCACHE_GENOCIDE;
2987 dentry->d_count--;
2988 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002989 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990 }
2991 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002992 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11002993 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2994 this_parent->d_flags |= DCACHE_GENOCIDE;
2995 this_parent->d_count--;
2996 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002997 this_parent = try_to_ascend(this_parent, locked, seq);
2998 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11002999 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11003000 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 goto resume;
3002 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11003003 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11003004 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11003005 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11003006 if (locked)
3007 write_sequnlock(&rename_lock);
3008 return;
3009
3010rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02003011 if (locked)
3012 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11003013 locked = 1;
3014 write_seqlock(&rename_lock);
3015 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016}
3017
3018/**
3019 * find_inode_number - check for dentry with name
3020 * @dir: directory to check
3021 * @name: Name to find.
3022 *
3023 * Check whether a dentry already exists for the given name,
3024 * and return the inode number if it has an inode. Otherwise
3025 * 0 is returned.
3026 *
3027 * This routine is used to post-process directory listings for
3028 * filesystems using synthetic inode numbers, and is necessary
3029 * to keep getcwd() working.
3030 */
3031
3032ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3033{
3034 struct dentry * dentry;
3035 ino_t ino = 0;
3036
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08003037 dentry = d_hash_and_lookup(dir, name);
3038 if (dentry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 if (dentry->d_inode)
3040 ino = dentry->d_inode->i_ino;
3041 dput(dentry);
3042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043 return ino;
3044}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003045EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046
3047static __initdata unsigned long dhash_entries;
3048static int __init set_dhash_entries(char *str)
3049{
3050 if (!str)
3051 return 0;
3052 dhash_entries = simple_strtoul(str, &str, 0);
3053 return 1;
3054}
3055__setup("dhash_entries=", set_dhash_entries);
3056
3057static void __init dcache_init_early(void)
3058{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003059 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
3061 /* If hashes are distributed across NUMA nodes, defer
3062 * hash allocation until vmalloc space is available.
3063 */
3064 if (hashdist)
3065 return;
3066
3067 dentry_hashtable =
3068 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003069 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 dhash_entries,
3071 13,
3072 HASH_EARLY,
3073 &d_hash_shift,
3074 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003075 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 0);
3077
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003078 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003079 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080}
3081
Denis Cheng74bf17c2007-10-16 23:26:30 -07003082static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003084 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085
3086 /*
3087 * A constructor could be added for stable state like the lists,
3088 * but it is probably not worth it because of the cache nature
3089 * of the dcache.
3090 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003091 dentry_cache = KMEM_CACHE(dentry,
3092 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
3094 /* Hash may have been set up in dcache_init_early */
3095 if (!hashdist)
3096 return;
3097
3098 dentry_hashtable =
3099 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003100 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 dhash_entries,
3102 13,
3103 0,
3104 &d_hash_shift,
3105 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003106 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 0);
3108
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003109 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003110 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111}
3112
3113/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003114struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003115EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117EXPORT_SYMBOL(d_genocide);
3118
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119void __init vfs_caches_init_early(void)
3120{
3121 dcache_init_early();
3122 inode_init_early();
3123}
3124
3125void __init vfs_caches_init(unsigned long mempages)
3126{
3127 unsigned long reserve;
3128
3129 /* Base hash sizes on available memory, with a reserve equal to
3130 150% of current kernel size */
3131
3132 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3133 mempages -= reserve;
3134
3135 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003136 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137
Denis Cheng74bf17c2007-10-16 23:26:30 -07003138 dcache_init();
3139 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003141 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 bdev_cache_init();
3143 chrdev_init();
3144}