blob: 5a23073138dfe310fb076e81567bac97cd0be9b6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
David Howells07f3f052006-09-30 20:52:18 +020040#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050041#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Nick Piggin789680d2011-01-07 17:49:30 +110043/*
44 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110045 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110047 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
Nick Piggin23044502011-01-07 17:49:31 +110051 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
53 * d_lock protects:
54 * - d_flags
55 * - d_name
56 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110057 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110058 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110059 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110061 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110062 *
63 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110064 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110065 * dentry->d_lock
66 * dcache_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110067 * dcache_hash_bucket lock
68 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110069 *
Nick Pigginda502952011-01-07 17:49:33 +110070 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
72 * ...
73 * dentry->d_parent->d_lock
74 * dentry->d_lock
75 *
76 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110077 * if (dentry1 < dentry2)
78 * dentry1->d_lock
79 * dentry2->d_lock
80 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080081int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83
Nick Piggin23044502011-01-07 17:49:31 +110084static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
99#define D_HASHBITS d_hash_shift
100#define D_HASHMASK d_hash_mask
101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8966be92012-03-02 14:23:30 -0800107static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100109{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 hash = hash + (hash >> D_HASHBITS);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100112 return dentry_hashtable + (hash & D_HASHMASK);
113}
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115/* Statistics gathering. */
116struct dentry_stat_t dentry_stat = {
117 .age_limit = 45,
118};
119
Nick Piggin3e880fb2011-01-07 17:49:19 +1100120static DEFINE_PER_CPU(unsigned int, nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100123static int get_nr_dentry(void)
124{
125 int i;
126 int sum = 0;
127 for_each_possible_cpu(i)
128 sum += per_cpu(nr_dentry, i);
129 return sum < 0 ? 0 : sum;
130}
131
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400132int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
133 size_t *lenp, loff_t *ppos)
134{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100135 dentry_stat.nr_dentry = get_nr_dentry();
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400136 return proc_dointvec(table, write, buffer, lenp, ppos);
137}
138#endif
139
Linus Torvalds5483f182012-03-04 15:51:42 -0800140/*
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
143 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700144#ifdef CONFIG_DCACHE_WORD_ACCESS
145
146#include <asm/word-at-a-time.h>
147/*
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
152 *
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
155 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700156static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800157{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800158 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800159
160 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700161 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700162 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800163 if (tcount < sizeof(unsigned long))
164 break;
165 if (unlikely(a != b))
166 return 1;
167 cs += sizeof(unsigned long);
168 ct += sizeof(unsigned long);
169 tcount -= sizeof(unsigned long);
170 if (!tcount)
171 return 0;
172 }
173 mask = ~(~0ul << tcount*8);
174 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700175}
176
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800177#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700178
Linus Torvalds94753db52012-05-10 12:19:19 -0700179static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700180{
Linus Torvalds5483f182012-03-04 15:51:42 -0800181 do {
182 if (*cs != *ct)
183 return 1;
184 cs++;
185 ct++;
186 tcount--;
187 } while (tcount);
188 return 0;
189}
190
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700191#endif
192
Linus Torvalds94753db52012-05-10 12:19:19 -0700193static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
194{
Linus Torvalds6326c712012-05-21 16:14:04 -0700195 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700196 /*
197 * Be careful about RCU walk racing with rename:
198 * use ACCESS_ONCE to fetch the name pointer.
199 *
200 * NOTE! Even if a rename will mean that the length
201 * was not loaded atomically, we don't care. The
202 * RCU walk will check the sequence count eventually,
203 * and catch it. And we won't overrun the buffer,
204 * because we're reading the name pointer atomically,
205 * and a dentry name is guaranteed to be properly
206 * terminated with a NUL byte.
207 *
208 * End result: even if 'len' is wrong, we'll exit
209 * early because the data cannot match (there can
210 * be no NUL in the ct/tcount data)
211 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700212 cs = ACCESS_ONCE(dentry->d_name.name);
213 smp_read_barrier_depends();
214 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700215}
216
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400217static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400219 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
220
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
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000340static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700341{
Nick Piggin23044502011-01-07 17:49:31 +1100342 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400343 if (list_empty(&dentry->d_lru)) {
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000344 list_add_tail(&dentry->d_lru, list);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400345 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100346 dentry_stat.nr_unused++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400347 } else {
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000348 list_move_tail(&dentry->d_lru, list);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700349 }
Nick Piggin23044502011-01-07 17:49:31 +1100350 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700351}
352
Miklos Szeredid52b9082007-05-08 00:23:46 -0700353/**
354 * d_kill - kill dentry and return parent
355 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800356 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700357 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200358 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700359 *
360 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100361 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100362 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
363 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700364 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100365static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200366 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100367 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100368 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700369{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700370 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400371 /*
372 * Inform try_to_ascend() that we are no longer attached to the
373 * dentry tree
374 */
Miklos Szeredib161dfa62012-09-17 22:31:38 +0200375 dentry->d_flags |= DCACHE_DENTRY_KILLED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100376 if (parent)
377 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700378 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100379 /*
380 * dentry_iput drops the locks, at which point nobody (except
381 * transient RCU lookups) can reach this dentry.
382 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700383 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900384 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700385}
386
David Howellsc6627c62011-06-07 14:09:20 +0100387/*
388 * Unhash a dentry without inserting an RCU walk barrier or checking that
389 * dentry->d_lock is locked. The caller must take care of that, if
390 * appropriate.
391 */
392static void __d_shrink(struct dentry *dentry)
393{
394 if (!d_unhashed(dentry)) {
395 struct hlist_bl_head *b;
396 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
397 b = &dentry->d_sb->s_anon;
398 else
399 b = d_hash(dentry->d_parent, dentry->d_name.hash);
400
401 hlist_bl_lock(b);
402 __hlist_bl_del(&dentry->d_hash);
403 dentry->d_hash.pprev = NULL;
404 hlist_bl_unlock(b);
405 }
406}
407
Nick Piggin789680d2011-01-07 17:49:30 +1100408/**
409 * d_drop - drop a dentry
410 * @dentry: dentry to drop
411 *
412 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
413 * be found through a VFS lookup any more. Note that this is different from
414 * deleting the dentry - d_delete will try to mark the dentry negative if
415 * possible, giving a successful _negative_ lookup, while d_drop will
416 * just make the cache lookup fail.
417 *
418 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
419 * reason (NFS timeouts or autofs deletes).
420 *
421 * __d_drop requires dentry->d_lock.
422 */
423void __d_drop(struct dentry *dentry)
424{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700425 if (!d_unhashed(dentry)) {
David Howellsc6627c62011-06-07 14:09:20 +0100426 __d_shrink(dentry);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700427 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100428 }
429}
430EXPORT_SYMBOL(__d_drop);
431
432void d_drop(struct dentry *dentry)
433{
Nick Piggin789680d2011-01-07 17:49:30 +1100434 spin_lock(&dentry->d_lock);
435 __d_drop(dentry);
436 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100437}
438EXPORT_SYMBOL(d_drop);
439
Nick Piggin77812a12011-01-07 17:49:48 +1100440/*
441 * Finish off a dentry we've decided to kill.
442 * dentry->d_lock must be held, returns with it unlocked.
443 * If ref is non-zero, then decrement the refcount too.
444 * Returns dentry requiring refcount drop, or NULL if we're done.
445 */
446static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
447 __releases(dentry->d_lock)
448{
Nick Piggin873feea2011-01-07 17:50:06 +1100449 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100450 struct dentry *parent;
451
Nick Piggin873feea2011-01-07 17:50:06 +1100452 inode = dentry->d_inode;
453 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100454relock:
455 spin_unlock(&dentry->d_lock);
456 cpu_relax();
457 return dentry; /* try again with same dentry */
458 }
459 if (IS_ROOT(dentry))
460 parent = NULL;
461 else
462 parent = dentry->d_parent;
463 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100464 if (inode)
465 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100466 goto relock;
467 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100468
Nick Piggin77812a12011-01-07 17:49:48 +1100469 if (ref)
470 dentry->d_count--;
Sage Weilf0023bc2011-10-28 10:02:42 -0700471 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700472 * inform the fs via d_prune that this dentry is about to be
473 * unhashed and destroyed.
474 */
Yan, Zheng61572bb2013-04-15 14:13:21 +0800475 if (dentry->d_flags & DCACHE_OP_PRUNE)
476 dentry->d_op->d_prune(dentry);
477
478 dentry_lru_del(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100479 /* if it was on the hash then remove it */
480 __d_drop(dentry);
481 return d_kill(dentry, parent);
482}
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484/*
485 * This is dput
486 *
487 * This is complicated by the fact that we do not want to put
488 * dentries that are no longer on any hash chain on the unused
489 * list: we'd much rather just get rid of them immediately.
490 *
491 * However, that implies that we have to traverse the dentry
492 * tree upwards to the parents which might _also_ now be
493 * scheduled for deletion (it may have been only waiting for
494 * its last child to go away).
495 *
496 * This tail recursion is done by hand as we don't want to depend
497 * on the compiler to always get this right (gcc generally doesn't).
498 * Real recursion would eat up our stack space.
499 */
500
501/*
502 * dput - release a dentry
503 * @dentry: dentry to release
504 *
505 * Release a dentry. This will drop the usage count and if appropriate
506 * call the dentry unlink method as well as removing it from the queues and
507 * releasing its resources. If the parent dentries were scheduled for release
508 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510void dput(struct dentry *dentry)
511{
512 if (!dentry)
513 return;
514
515repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100516 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100519 BUG_ON(!dentry->d_count);
520 if (dentry->d_count > 1) {
521 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 return;
524 }
525
Nick Pigginfb045ad2011-01-07 17:49:55 +1100526 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100528 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 }
Nick Piggin265ac902010-10-10 05:36:24 -0400530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 /* Unreachable? Get rid of it */
532 if (d_unhashed(dentry))
533 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400534
Jeff Layton39e3c952012-11-28 11:30:53 -0500535 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400536 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400537
Nick Piggin61f3dee2011-01-07 17:49:40 +1100538 dentry->d_count--;
539 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 return;
541
Miklos Szeredid52b9082007-05-08 00:23:46 -0700542kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100543 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700544 if (dentry)
545 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700547EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548
549/**
550 * d_invalidate - invalidate a dentry
551 * @dentry: dentry to invalidate
552 *
553 * Try to invalidate the dentry if it turns out to be
554 * possible. If there are other dentries that can be
555 * reached through this one we can't delete it and we
556 * return -EBUSY. On success we return 0.
557 *
558 * no dcache lock.
559 */
560
561int d_invalidate(struct dentry * dentry)
562{
563 /*
564 * If it's already been dropped, return OK.
565 */
Nick Pigginda502952011-01-07 17:49:33 +1100566 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100568 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 return 0;
570 }
571 /*
572 * Check whether to do a partial shrink_dcache
573 * to get rid of unused child entries.
574 */
575 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100576 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100578 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580
581 /*
582 * Somebody else still using it?
583 *
584 * If it's a directory, we can't drop it
585 * for fear of somebody re-populating it
586 * with children (even though dropping it
587 * would make it unreachable from the root,
588 * we might still populate it if it was a
589 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000590 * We also need to leave mountpoints alone,
591 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 */
Al Viro50e69632011-11-07 16:39:57 +0000593 if (dentry->d_count > 1 && dentry->d_inode) {
594 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 return -EBUSY;
597 }
598 }
599
600 __d_drop(dentry);
601 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 return 0;
603}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700604EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100606/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100607static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100609 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
Nick Piggindc0474b2011-01-07 17:49:43 +1100612static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100613{
Nick Piggin23044502011-01-07 17:49:31 +1100614 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100615 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100616 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100617}
618
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100619struct dentry *dget_parent(struct dentry *dentry)
620{
621 struct dentry *ret;
622
623repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100624 /*
625 * Don't need rcu_dereference because we re-check it was correct under
626 * the lock.
627 */
628 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100629 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100630 spin_lock(&ret->d_lock);
631 if (unlikely(ret != dentry->d_parent)) {
632 spin_unlock(&ret->d_lock);
633 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100634 goto repeat;
635 }
Nick Piggina734eb42011-01-07 17:49:44 +1100636 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100637 BUG_ON(!ret->d_count);
638 ret->d_count++;
639 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100640 return ret;
641}
642EXPORT_SYMBOL(dget_parent);
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644/**
645 * d_find_alias - grab a hashed alias of inode
646 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700647 * @want_discon: flag, used by d_splice_alias, to request
648 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 *
650 * If inode has a hashed alias, or is a directory and has any alias,
651 * acquire the reference to alias and return it. Otherwise return NULL.
652 * Notice that if inode is a directory there can be only one alias and
653 * it can be unhashed only if it has no children, or if it is the root
654 * of a filesystem.
655 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700656 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700657 * any other hashed alias over that one unless @want_discon is set,
658 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700660static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Nick Pigginda502952011-01-07 17:49:33 +1100662 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Nick Pigginda502952011-01-07 17:49:33 +1100664again:
665 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800666 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100667 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700669 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100670 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700672 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100673 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100674 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 return alias;
676 }
677 }
Nick Pigginda502952011-01-07 17:49:33 +1100678 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
Nick Pigginda502952011-01-07 17:49:33 +1100680 if (discon_alias) {
681 alias = discon_alias;
682 spin_lock(&alias->d_lock);
683 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
684 if (IS_ROOT(alias) &&
685 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100686 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100687 spin_unlock(&alias->d_lock);
688 return alias;
689 }
690 }
691 spin_unlock(&alias->d_lock);
692 goto again;
693 }
694 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695}
696
Nick Pigginda502952011-01-07 17:49:33 +1100697struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
David Howells214fda12006-03-25 03:06:36 -0800699 struct dentry *de = NULL;
700
Al Virob3d9b7a2012-06-09 13:51:19 -0400701 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100702 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700703 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100704 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 return de;
707}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700708EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
710/*
711 * Try to kill dentries associated with this inode.
712 * WARNING: you must own a reference to inode.
713 */
714void d_prune_aliases(struct inode *inode)
715{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700716 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100718 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800719 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100721 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100722 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 __d_drop(dentry);
724 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100725 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 dput(dentry);
727 goto restart;
728 }
729 spin_unlock(&dentry->d_lock);
730 }
Nick Piggin873feea2011-01-07 17:50:06 +1100731 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700733EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
735/*
Nick Piggin77812a12011-01-07 17:49:48 +1100736 * Try to throw away a dentry - free the inode, dput the parent.
737 * Requires dentry->d_lock is held, and dentry->d_count == 0.
738 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700739 *
Nick Piggin77812a12011-01-07 17:49:48 +1100740 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 */
Nick Piggin77812a12011-01-07 17:49:48 +1100742static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200743 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Nick Piggin77812a12011-01-07 17:49:48 +1100745 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700746
Nick Piggin77812a12011-01-07 17:49:48 +1100747 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700748 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100749 * If dentry_kill returns NULL, we have nothing more to do.
750 * if it returns the same dentry, trylocks failed. In either
751 * case, just loop again.
752 *
753 * Otherwise, we need to prune ancestors too. This is necessary
754 * to prevent quadratic behavior of shrink_dcache_parent(), but
755 * is also expected to be beneficial in reducing dentry cache
756 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700757 */
Nick Piggin77812a12011-01-07 17:49:48 +1100758 if (!parent)
759 return;
760 if (parent == dentry)
761 return;
762
763 /* Prune ancestors. */
764 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700765 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100766 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100767 if (dentry->d_count > 1) {
768 dentry->d_count--;
769 spin_unlock(&dentry->d_lock);
770 return;
771 }
Nick Piggin77812a12011-01-07 17:49:48 +1100772 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774}
775
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400776static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700778 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700779
Nick Pigginec336792011-01-07 17:49:47 +1100780 rcu_read_lock();
781 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100782 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
783 if (&dentry->d_lru == list)
784 break; /* empty */
785 spin_lock(&dentry->d_lock);
786 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
787 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100788 continue;
789 }
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 /*
792 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700793 * the LRU because of laziness during lookup. Do not free
794 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100796 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100797 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700798 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 continue;
800 }
Nick Pigginec336792011-01-07 17:49:47 +1100801
Nick Pigginec336792011-01-07 17:49:47 +1100802 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100803
804 try_prune_one_dentry(dentry);
805
Nick Pigginec336792011-01-07 17:49:47 +1100806 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
Nick Pigginec336792011-01-07 17:49:47 +1100808 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400809}
810
811/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000812 * prune_dcache_sb - shrink the dcache
813 * @sb: superblock
814 * @count: number of entries to try to free
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400815 *
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000816 * Attempt to shrink the superblock dcache LRU by @count entries. This is
817 * done when we need more memory an called from the superblock shrinker
818 * function.
819 *
820 * This function may fail to free any resources if all the dentries are in
821 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400822 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000823void prune_dcache_sb(struct super_block *sb, int count)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400824{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400825 struct dentry *dentry;
826 LIST_HEAD(referenced);
827 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400828
Nick Piggin23044502011-01-07 17:49:31 +1100829relock:
830 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400831 while (!list_empty(&sb->s_dentry_lru)) {
832 dentry = list_entry(sb->s_dentry_lru.prev,
833 struct dentry, d_lru);
834 BUG_ON(dentry->d_sb != sb);
835
Nick Piggin23044502011-01-07 17:49:31 +1100836 if (!spin_trylock(&dentry->d_lock)) {
837 spin_unlock(&dcache_lru_lock);
838 cpu_relax();
839 goto relock;
840 }
841
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000842 if (dentry->d_flags & DCACHE_REFERENCED) {
Nick Piggin23044502011-01-07 17:49:31 +1100843 dentry->d_flags &= ~DCACHE_REFERENCED;
844 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400845 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100846 } else {
847 list_move_tail(&dentry->d_lru, &tmp);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100848 dentry->d_flags |= DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100849 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000850 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100851 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400852 }
Nick Pigginec336792011-01-07 17:49:47 +1100853 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400854 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700855 if (!list_empty(&referenced))
856 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100857 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100858
859 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860}
861
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700862/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 * shrink_dcache_sb - shrink dcache for a superblock
864 * @sb: superblock
865 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400866 * Shrink the dcache for the specified super block. This is used to free
867 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400869void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400871 LIST_HEAD(tmp);
872
Nick Piggin23044502011-01-07 17:49:31 +1100873 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400874 while (!list_empty(&sb->s_dentry_lru)) {
875 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100876 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400877 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100878 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400879 }
Nick Piggin23044502011-01-07 17:49:31 +1100880 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700882EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884/*
David Howellsc636ebd2006-10-11 01:22:19 -0700885 * destroy a single subtree of dentries for unmount
886 * - see the comments on shrink_dcache_for_umount() for a description of the
887 * locking
888 */
889static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
890{
891 struct dentry *parent;
892
893 BUG_ON(!IS_ROOT(dentry));
894
David Howellsc636ebd2006-10-11 01:22:19 -0700895 for (;;) {
896 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100897 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700898 dentry = list_entry(dentry->d_subdirs.next,
899 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700900
901 /* consume the dentries from this leaf up through its parents
902 * until we find one with children or run out altogether */
903 do {
904 struct inode *inode;
905
Sage Weilf0023bc2011-10-28 10:02:42 -0700906 /*
Yan, Zheng61572bb2013-04-15 14:13:21 +0800907 * inform the fs that this dentry is about to be
Sage Weilf0023bc2011-10-28 10:02:42 -0700908 * unhashed and destroyed.
909 */
Yan, Zheng61572bb2013-04-15 14:13:21 +0800910 if (dentry->d_flags & DCACHE_OP_PRUNE)
911 dentry->d_op->d_prune(dentry);
912
913 dentry_lru_del(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100914 __d_shrink(dentry);
915
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100916 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700917 printk(KERN_ERR
918 "BUG: Dentry %p{i=%lx,n=%s}"
919 " still in use (%d)"
920 " [unmount of %s %s]\n",
921 dentry,
922 dentry->d_inode ?
923 dentry->d_inode->i_ino : 0UL,
924 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100925 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700926 dentry->d_sb->s_type->name,
927 dentry->d_sb->s_id);
928 BUG();
929 }
930
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100931 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700932 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100933 list_del(&dentry->d_u.d_child);
934 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900935 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100936 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100937 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900938 }
David Howellsc636ebd2006-10-11 01:22:19 -0700939
David Howellsc636ebd2006-10-11 01:22:19 -0700940 inode = dentry->d_inode;
941 if (inode) {
942 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400943 hlist_del_init(&dentry->d_alias);
David Howellsc636ebd2006-10-11 01:22:19 -0700944 if (dentry->d_op && dentry->d_op->d_iput)
945 dentry->d_op->d_iput(dentry, inode);
946 else
947 iput(inode);
948 }
949
950 d_free(dentry);
951
952 /* finished when we fall off the top of the tree,
953 * otherwise we ascend to the parent and move to the
954 * next sibling if there is one */
955 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400956 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700957 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700958 } while (list_empty(&dentry->d_subdirs));
959
960 dentry = list_entry(dentry->d_subdirs.next,
961 struct dentry, d_u.d_child);
962 }
963}
964
965/*
966 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100967 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700968 * - the superblock is detached from all mountings and open files, so the
969 * dentry trees will not be rearranged by the VFS
970 * - s_umount is write-locked, so the memory pressure shrinker will ignore
971 * any dentries belonging to this superblock that it comes across
972 * - the filesystem itself is no longer permitted to rearrange the dentries
973 * in this superblock
974 */
975void shrink_dcache_for_umount(struct super_block *sb)
976{
977 struct dentry *dentry;
978
979 if (down_read_trylock(&sb->s_umount))
980 BUG();
981
982 dentry = sb->s_root;
983 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100984 dentry->d_count--;
David Howellsc636ebd2006-10-11 01:22:19 -0700985 shrink_dcache_for_umount_subtree(dentry);
986
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100987 while (!hlist_bl_empty(&sb->s_anon)) {
988 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -0700989 shrink_dcache_for_umount_subtree(dentry);
990 }
991}
992
993/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -0700994 * This tries to ascend one level of parenthood, but
995 * we can race with renaming, so we need to re-check
996 * the parenthood after dropping the lock and check
997 * that the sequence number still matches.
998 */
999static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1000{
1001 struct dentry *new = old->d_parent;
1002
1003 rcu_read_lock();
1004 spin_unlock(&old->d_lock);
1005 spin_lock(&new->d_lock);
1006
1007 /*
1008 * might go back up the wrong parent if we have had a rename
1009 * or deletion
1010 */
1011 if (new != old->d_parent ||
Miklos Szeredib161dfa62012-09-17 22:31:38 +02001012 (old->d_flags & DCACHE_DENTRY_KILLED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001013 (!locked && read_seqretry(&rename_lock, seq))) {
1014 spin_unlock(&new->d_lock);
1015 new = NULL;
1016 }
1017 rcu_read_unlock();
1018 return new;
1019}
1020
1021
1022/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 * Search for at least 1 mount point in the dentry's subdirs.
1024 * We descend to the next level whenever the d_subdirs
1025 * list is non-empty and continue searching.
1026 */
1027
1028/**
1029 * have_submounts - check for mounts over a dentry
1030 * @parent: dentry to check.
1031 *
1032 * Return true if the parent or its subdirectories contain
1033 * a mount point
1034 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035int have_submounts(struct dentry *parent)
1036{
Nick Piggin949854d2011-01-07 17:49:37 +11001037 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001039 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001040 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001041
Nick Piggin949854d2011-01-07 17:49:37 +11001042 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001043again:
1044 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 if (d_mountpoint(parent))
1047 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001048 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049repeat:
1050 next = this_parent->d_subdirs.next;
1051resume:
1052 while (next != &this_parent->d_subdirs) {
1053 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001054 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001056
1057 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001059 if (d_mountpoint(dentry)) {
1060 spin_unlock(&dentry->d_lock);
1061 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001065 spin_unlock(&this_parent->d_lock);
1066 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001068 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 goto repeat;
1070 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001071 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073 /*
1074 * All done at this level ... ascend and resume the search.
1075 */
1076 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001077 struct dentry *child = this_parent;
1078 this_parent = try_to_ascend(this_parent, locked, seq);
1079 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001080 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001081 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 goto resume;
1083 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001084 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001085 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001086 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001087 if (locked)
1088 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 return 0; /* No mount points found in tree */
1090positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001091 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001092 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001093 if (locked)
1094 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001096
1097rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02001098 if (locked)
1099 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001100 locked = 1;
1101 write_seqlock(&rename_lock);
1102 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001104EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001107 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 * and move any unused dentries to the end of the unused
1109 * list for prune_dcache(). We descend to the next level
1110 * whenever the d_subdirs list is non-empty and continue
1111 * searching.
1112 *
1113 * It returns zero iff there are no unused children,
1114 * otherwise it returns the number of children moved to
1115 * the end of the unused list. This may not be the total
1116 * number of unused children, because select_parent can
1117 * drop the lock and return early due to latency
1118 * constraints.
1119 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001120static int select_parent(struct dentry *parent, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
Nick Piggin949854d2011-01-07 17:49:37 +11001122 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001124 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001126 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Nick Piggin949854d2011-01-07 17:49:37 +11001128 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001129again:
1130 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001131 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132repeat:
1133 next = this_parent->d_subdirs.next;
1134resume:
1135 while (next != &this_parent->d_subdirs) {
1136 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001137 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 next = tmp->next;
1139
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001140 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001141
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001142 /*
1143 * move only zero ref count dentries to the dispose list.
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001144 *
1145 * Those which are presently on the shrink list, being processed
1146 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1147 * loop in shrink_dcache_parent() might not make any progress
1148 * and loop forever.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 */
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001150 if (dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001151 dentry_lru_del(dentry);
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001152 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1153 dentry_lru_move_list(dentry, dispose);
1154 dentry->d_flags |= DCACHE_SHRINK_LIST;
1155 found++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 /*
1158 * We can return to the caller if we have found some (this
1159 * ensures forward progress). We'll be coming back to find
1160 * the rest.
1161 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001162 if (found && need_resched()) {
1163 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
1167 /*
1168 * Descend a level if the d_subdirs list is non-empty.
1169 */
1170 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001171 spin_unlock(&this_parent->d_lock);
1172 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001174 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 goto repeat;
1176 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001177
1178 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 }
1180 /*
1181 * All done at this level ... ascend and resume the search.
1182 */
1183 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001184 struct dentry *child = this_parent;
1185 this_parent = try_to_ascend(this_parent, locked, seq);
1186 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001187 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001188 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 goto resume;
1190 }
1191out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001192 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001193 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001194 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001195 if (locked)
1196 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001198
1199rename_retry:
1200 if (found)
1201 return found;
Miklos Szeredi8110e162012-09-17 22:23:30 +02001202 if (locked)
1203 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001204 locked = 1;
1205 write_seqlock(&rename_lock);
1206 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
1209/**
1210 * shrink_dcache_parent - prune dcache
1211 * @parent: parent of entries to prune
1212 *
1213 * Prune the dcache to remove unused children of the parent dentry.
1214 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215void shrink_dcache_parent(struct dentry * parent)
1216{
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001217 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 int found;
1219
Greg Thelen421348f2013-04-30 15:26:48 -07001220 while ((found = select_parent(parent, &dispose)) != 0) {
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001221 shrink_dentry_list(&dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001222 cond_resched();
1223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001225EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227/**
Al Viroa4464db2011-07-07 15:03:58 -04001228 * __d_alloc - allocate a dcache entry
1229 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 * @name: qstr of the name
1231 *
1232 * Allocates a dentry. It returns %NULL if there is insufficient memory
1233 * available. On a success the dentry is returned. The name passed in is
1234 * copied and the copy passed in may be reused after this call.
1235 */
1236
Al Viroa4464db2011-07-07 15:03:58 -04001237struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238{
1239 struct dentry *dentry;
1240 char *dname;
1241
Mel Gormane12ba742007-10-16 01:25:52 -07001242 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (!dentry)
1244 return NULL;
1245
Linus Torvalds6326c712012-05-21 16:14:04 -07001246 /*
1247 * We guarantee that the inline name is always NUL-terminated.
1248 * This way the memcpy() done by the name switching in rename
1249 * will still always have a NUL at the end, even if we might
1250 * be overwriting an internal NUL character
1251 */
1252 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 if (name->len > DNAME_INLINE_LEN-1) {
1254 dname = kmalloc(name->len + 1, GFP_KERNEL);
1255 if (!dname) {
1256 kmem_cache_free(dentry_cache, dentry);
1257 return NULL;
1258 }
1259 } else {
1260 dname = dentry->d_iname;
1261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262
1263 dentry->d_name.len = name->len;
1264 dentry->d_name.hash = name->hash;
1265 memcpy(dname, name->name, name->len);
1266 dname[name->len] = 0;
1267
Linus Torvalds6326c712012-05-21 16:14:04 -07001268 /* Make sure we always see the terminating NUL character */
1269 smp_wmb();
1270 dentry->d_name.name = dname;
1271
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001272 dentry->d_count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001273 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001275 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001277 dentry->d_parent = dentry;
1278 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 dentry->d_op = NULL;
1280 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001281 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 INIT_LIST_HEAD(&dentry->d_lru);
1283 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001284 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001285 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001286 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
Nick Piggin3e880fb2011-01-07 17:49:19 +11001288 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001289
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 return dentry;
1291}
Al Viroa4464db2011-07-07 15:03:58 -04001292
1293/**
1294 * d_alloc - allocate a dcache entry
1295 * @parent: parent of entry to allocate
1296 * @name: qstr of the name
1297 *
1298 * Allocates a dentry. It returns %NULL if there is insufficient memory
1299 * available. On a success the dentry is returned. The name passed in is
1300 * copied and the copy passed in may be reused after this call.
1301 */
1302struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1303{
1304 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1305 if (!dentry)
1306 return NULL;
1307
1308 spin_lock(&parent->d_lock);
1309 /*
1310 * don't need child lock because it is not subject
1311 * to concurrency here
1312 */
1313 __dget_dlock(parent);
1314 dentry->d_parent = parent;
1315 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1316 spin_unlock(&parent->d_lock);
1317
1318 return dentry;
1319}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001320EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Nick Piggin4b936882011-01-07 17:50:07 +11001322struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1323{
Al Viroa4464db2011-07-07 15:03:58 -04001324 struct dentry *dentry = __d_alloc(sb, name);
1325 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001326 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001327 return dentry;
1328}
1329EXPORT_SYMBOL(d_alloc_pseudo);
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1332{
1333 struct qstr q;
1334
1335 q.name = name;
1336 q.len = strlen(name);
1337 q.hash = full_name_hash(q.name, q.len);
1338 return d_alloc(parent, &q);
1339}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001340EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Nick Pigginfb045ad2011-01-07 17:49:55 +11001342void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1343{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001344 WARN_ON_ONCE(dentry->d_op);
1345 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001346 DCACHE_OP_COMPARE |
1347 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001348 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001349 DCACHE_OP_DELETE ));
1350 dentry->d_op = op;
1351 if (!op)
1352 return;
1353 if (op->d_hash)
1354 dentry->d_flags |= DCACHE_OP_HASH;
1355 if (op->d_compare)
1356 dentry->d_flags |= DCACHE_OP_COMPARE;
1357 if (op->d_revalidate)
1358 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001359 if (op->d_weak_revalidate)
1360 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001361 if (op->d_delete)
1362 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001363 if (op->d_prune)
1364 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001365
1366}
1367EXPORT_SYMBOL(d_set_d_op);
1368
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001369static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1370{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001371 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001372 if (inode) {
1373 if (unlikely(IS_AUTOMOUNT(inode)))
1374 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
Al Virob3d9b7a2012-06-09 13:51:19 -04001375 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001376 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001377 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001378 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001379 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001380 fsnotify_d_instantiate(dentry, inode);
1381}
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383/**
1384 * d_instantiate - fill in inode information for a dentry
1385 * @entry: dentry to complete
1386 * @inode: inode to attach to this dentry
1387 *
1388 * Fill in inode information in the entry.
1389 *
1390 * This turns negative dentries into productive full members
1391 * of society.
1392 *
1393 * NOTE! This assumes that the inode count has been incremented
1394 * (or otherwise set) by the caller to indicate that it is now
1395 * in use by the dcache.
1396 */
1397
1398void d_instantiate(struct dentry *entry, struct inode * inode)
1399{
Al Virob3d9b7a2012-06-09 13:51:19 -04001400 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001401 if (inode)
1402 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001403 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001404 if (inode)
1405 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 security_d_instantiate(entry, inode);
1407}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001408EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
1410/**
1411 * d_instantiate_unique - instantiate a non-aliased dentry
1412 * @entry: dentry to instantiate
1413 * @inode: inode to attach to this dentry
1414 *
1415 * Fill in inode information in the entry. On success, it returns NULL.
1416 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001417 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 *
1419 * Note that in order to avoid conflicts with rename() etc, the caller
1420 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001421 *
1422 * This also assumes that the inode count has been incremented
1423 * (or otherwise set) by the caller to indicate that it is now
1424 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 */
David Howells770bfad2006-08-22 20:06:07 -04001426static struct dentry *__d_instantiate_unique(struct dentry *entry,
1427 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
1429 struct dentry *alias;
1430 int len = entry->d_name.len;
1431 const char *name = entry->d_name.name;
1432 unsigned int hash = entry->d_name.hash;
1433
David Howells770bfad2006-08-22 20:06:07 -04001434 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001435 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001436 return NULL;
1437 }
1438
Sasha Levinb67bfe02013-02-27 17:06:00 -08001439 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001440 /*
1441 * Don't need alias->d_lock here, because aliases with
1442 * d_parent == entry->d_parent are not subject to name or
1443 * parent changes, because the parent inode i_mutex is held.
1444 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001445 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 continue;
1447 if (alias->d_parent != entry->d_parent)
1448 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001449 if (alias->d_name.len != len)
1450 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001451 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001453 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 return alias;
1455 }
David Howells770bfad2006-08-22 20:06:07 -04001456
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001457 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 return NULL;
1459}
David Howells770bfad2006-08-22 20:06:07 -04001460
1461struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1462{
1463 struct dentry *result;
1464
Al Virob3d9b7a2012-06-09 13:51:19 -04001465 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001466
Nick Piggin873feea2011-01-07 17:50:06 +11001467 if (inode)
1468 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001469 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001470 if (inode)
1471 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001472
1473 if (!result) {
1474 security_d_instantiate(entry, inode);
1475 return NULL;
1476 }
1477
1478 BUG_ON(!d_unhashed(result));
1479 iput(inode);
1480 return result;
1481}
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483EXPORT_SYMBOL(d_instantiate_unique);
1484
Al Viroadc0e912012-01-08 16:49:21 -05001485struct dentry *d_make_root(struct inode *root_inode)
1486{
1487 struct dentry *res = NULL;
1488
1489 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001490 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001491
1492 res = __d_alloc(root_inode->i_sb, &name);
1493 if (res)
1494 d_instantiate(res, root_inode);
1495 else
1496 iput(root_inode);
1497 }
1498 return res;
1499}
1500EXPORT_SYMBOL(d_make_root);
1501
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001502static struct dentry * __d_find_any_alias(struct inode *inode)
1503{
1504 struct dentry *alias;
1505
Al Virob3d9b7a2012-06-09 13:51:19 -04001506 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001507 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001508 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001509 __dget(alias);
1510 return alias;
1511}
1512
Sage Weil46f72b32012-01-10 09:04:37 -08001513/**
1514 * d_find_any_alias - find any alias for a given inode
1515 * @inode: inode to find an alias for
1516 *
1517 * If any aliases exist for the given inode, take and return a
1518 * reference for one of them. If no aliases exist, return %NULL.
1519 */
1520struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001521{
1522 struct dentry *de;
1523
1524 spin_lock(&inode->i_lock);
1525 de = __d_find_any_alias(inode);
1526 spin_unlock(&inode->i_lock);
1527 return de;
1528}
Sage Weil46f72b32012-01-10 09:04:37 -08001529EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001530
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001532 * d_obtain_alias - find or allocate a dentry for a given inode
1533 * @inode: inode to allocate the dentry for
1534 *
1535 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1536 * similar open by handle operations. The returned dentry may be anonymous,
1537 * or may have a full name (if the inode was already in the cache).
1538 *
1539 * When called on a directory inode, we must ensure that the inode only ever
1540 * has one dentry. If a dentry is found, that is returned instead of
1541 * allocating a new one.
1542 *
1543 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001544 * to the dentry. In case of an error the reference on the inode is released.
1545 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1546 * be passed in and will be the error will be propagate to the return value,
1547 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001548 */
1549struct dentry *d_obtain_alias(struct inode *inode)
1550{
NeilBrownb911a6b2012-11-08 16:09:37 -08001551 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001552 struct dentry *tmp;
1553 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001554
1555 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001556 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001557 if (IS_ERR(inode))
1558 return ERR_CAST(inode);
1559
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001560 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001561 if (res)
1562 goto out_iput;
1563
Al Viroa4464db2011-07-07 15:03:58 -04001564 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001565 if (!tmp) {
1566 res = ERR_PTR(-ENOMEM);
1567 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001568 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001569
Nick Piggin873feea2011-01-07 17:50:06 +11001570 spin_lock(&inode->i_lock);
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) {
Nick Piggin873feea2011-01-07 17:50:06 +11001573 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001574 dput(tmp);
1575 goto out_iput;
1576 }
1577
1578 /* attach a disconnected dentry */
1579 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001580 tmp->d_inode = inode;
1581 tmp->d_flags |= DCACHE_DISCONNECTED;
Al Virob3d9b7a2012-06-09 13:51:19 -04001582 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001583 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001584 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001585 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001586 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001587 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001588 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001589
Christoph Hellwig9308a612008-08-11 15:49:12 +02001590 return tmp;
1591
1592 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001593 if (res && !IS_ERR(res))
1594 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001595 iput(inode);
1596 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001597}
Benny Halevyadc48722009-02-27 14:02:59 -08001598EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600/**
1601 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1602 * @inode: the inode which may have a disconnected dentry
1603 * @dentry: a negative dentry which we want to point to the inode.
1604 *
1605 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1606 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1607 * and return it, else simply d_add the inode to the dentry and return NULL.
1608 *
1609 * This is needed in the lookup routine of any filesystem that is exportable
1610 * (via knfsd) so that we can build dcache paths to directories effectively.
1611 *
1612 * If a dentry was found and moved, then it is returned. Otherwise NULL
1613 * is returned. This matches the expected return value of ->lookup.
1614 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001615 * Cluster filesystems may call this function with a negative, hashed dentry.
1616 * In that case, we know that the inode will be a regular file, and also this
1617 * will only occur during atomic_open. So we need to check for the dentry
1618 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 */
1620struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1621{
1622 struct dentry *new = NULL;
1623
Al Viroa9049372011-07-08 21:20:11 -04001624 if (IS_ERR(inode))
1625 return ERR_CAST(inode);
1626
NeilBrown21c0d8f2006-10-04 02:16:16 -07001627 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001628 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001629 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001631 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001632 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 d_move(new, dentry);
1635 iput(inode);
1636 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001637 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001638 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001639 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 security_d_instantiate(dentry, inode);
1641 d_rehash(dentry);
1642 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001643 } else {
1644 d_instantiate(dentry, inode);
1645 if (d_unhashed(dentry))
1646 d_rehash(dentry);
1647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 return new;
1649}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001650EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
Barry Naujok94035402008-05-21 16:50:46 +10001652/**
1653 * d_add_ci - lookup or allocate new dentry with case-exact name
1654 * @inode: the inode case-insensitive lookup has found
1655 * @dentry: the negative dentry that was passed to the parent's lookup func
1656 * @name: the case-exact name to be associated with the returned dentry
1657 *
1658 * This is to avoid filling the dcache with case-insensitive names to the
1659 * same inode, only the actual correct case is stored in the dcache for
1660 * case-insensitive filesystems.
1661 *
1662 * For a case-insensitive lookup match and if the the case-exact dentry
1663 * already exists in in the dcache, use it and return it.
1664 *
1665 * If no entry exists with the exact case name, allocate new dentry with
1666 * the exact case, and return the spliced entry.
1667 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001668struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001669 struct qstr *name)
1670{
Barry Naujok94035402008-05-21 16:50:46 +10001671 struct dentry *found;
1672 struct dentry *new;
1673
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001674 /*
1675 * First check if a dentry matching the name already exists,
1676 * if not go ahead and create it now.
1677 */
Barry Naujok94035402008-05-21 16:50:46 +10001678 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001679 if (unlikely(IS_ERR(found)))
1680 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001681 if (!found) {
1682 new = d_alloc(dentry->d_parent, name);
1683 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001684 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001685 goto err_out;
1686 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001687
Barry Naujok94035402008-05-21 16:50:46 +10001688 found = d_splice_alias(inode, new);
1689 if (found) {
1690 dput(new);
1691 return found;
1692 }
1693 return new;
1694 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001695
1696 /*
1697 * If a matching dentry exists, and it's not negative use it.
1698 *
1699 * Decrement the reference count to balance the iget() done
1700 * earlier on.
1701 */
Barry Naujok94035402008-05-21 16:50:46 +10001702 if (found->d_inode) {
1703 if (unlikely(found->d_inode != inode)) {
1704 /* This can't happen because bad inodes are unhashed. */
1705 BUG_ON(!is_bad_inode(inode));
1706 BUG_ON(!is_bad_inode(found->d_inode));
1707 }
Barry Naujok94035402008-05-21 16:50:46 +10001708 iput(inode);
1709 return found;
1710 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001711
Barry Naujok94035402008-05-21 16:50:46 +10001712 /*
1713 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001714 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001715 */
Al Viro4513d892011-07-17 10:52:14 -04001716 new = d_splice_alias(inode, found);
1717 if (new) {
1718 dput(found);
1719 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001720 }
Al Viro4513d892011-07-17 10:52:14 -04001721 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001722
1723err_out:
1724 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001725 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001726}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001727EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001729/*
1730 * Do the slow-case of the dentry name compare.
1731 *
1732 * Unlike the dentry_cmp() function, we need to atomically
1733 * load the name, length and inode information, so that the
1734 * filesystem can rely on them, and can use the 'name' and
1735 * 'len' information without worrying about walking off the
1736 * end of memory etc.
1737 *
1738 * Thus the read_seqcount_retry() and the "duplicate" info
1739 * in arguments (the low-level filesystem should not look
1740 * at the dentry inode or name contents directly, since
1741 * rename can change them while we're in RCU mode).
1742 */
1743enum slow_d_compare {
1744 D_COMP_OK,
1745 D_COMP_NOMATCH,
1746 D_COMP_SEQRETRY,
1747};
1748
1749static noinline enum slow_d_compare slow_dentry_cmp(
1750 const struct dentry *parent,
1751 struct inode *inode,
1752 struct dentry *dentry,
1753 unsigned int seq,
1754 const struct qstr *name)
1755{
1756 int tlen = dentry->d_name.len;
1757 const char *tname = dentry->d_name.name;
1758 struct inode *i = dentry->d_inode;
1759
1760 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1761 cpu_relax();
1762 return D_COMP_SEQRETRY;
1763 }
1764 if (parent->d_op->d_compare(parent, inode,
1765 dentry, i,
1766 tlen, tname, name))
1767 return D_COMP_NOMATCH;
1768 return D_COMP_OK;
1769}
1770
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001772 * __d_lookup_rcu - search for a dentry (racy, store-free)
1773 * @parent: parent dentry
1774 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001775 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001776 * @inode: returns dentry->d_inode when the inode was found valid.
1777 * Returns: dentry, or NULL
1778 *
1779 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1780 * resolution (store-free path walking) design described in
1781 * Documentation/filesystems/path-lookup.txt.
1782 *
1783 * This is not to be used outside core vfs.
1784 *
1785 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1786 * held, and rcu_read_lock held. The returned dentry must not be stored into
1787 * without taking d_lock and checking d_seq sequence count against @seq
1788 * returned here.
1789 *
1790 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1791 * function.
1792 *
1793 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1794 * the returned dentry, so long as its parent's seqlock is checked after the
1795 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1796 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001797 *
1798 * NOTE! The caller *has* to check the resulting dentry against the sequence
1799 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001800 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001801struct dentry *__d_lookup_rcu(const struct dentry *parent,
1802 const struct qstr *name,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001803 unsigned *seqp, struct inode *inode)
Nick Piggin31e6b012011-01-07 17:49:52 +11001804{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001805 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001806 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001807 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001808 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001809 struct dentry *dentry;
1810
1811 /*
1812 * Note: There is significant duplication with __d_lookup_rcu which is
1813 * required to prevent single threaded performance regressions
1814 * especially on architectures where smp_rmb (in seqcounts) are costly.
1815 * Keep the two functions in sync.
1816 */
1817
1818 /*
1819 * The hash list is protected using RCU.
1820 *
1821 * Carefully use d_seq when comparing a candidate dentry, to avoid
1822 * races with d_move().
1823 *
1824 * It is possible that concurrent renames can mess up our list
1825 * walk here and result in missing our dentry, resulting in the
1826 * false-negative result. d_lookup() protects against concurrent
1827 * renames using rename_lock seqlock.
1828 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001829 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11001830 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001831 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08001832 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001833
Nick Piggin31e6b012011-01-07 17:49:52 +11001834seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001835 /*
1836 * The dentry sequence count protects us from concurrent
1837 * renames, and thus protects inode, parent and name fields.
1838 *
1839 * The caller must perform a seqcount check in order
1840 * to do anything useful with the returned dentry,
1841 * including using the 'd_inode' pointer.
1842 *
1843 * NOTE! We do a "raw" seqcount_begin here. That means that
1844 * we don't wait for the sequence count to stabilize if it
1845 * is in the middle of a sequence change. If we do the slow
1846 * dentry compare, we will do seqretries until it is stable,
1847 * and if we end up with a successful lookup, we actually
1848 * want to exit RCU lookup anyway.
1849 */
1850 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001851 if (dentry->d_parent != parent)
1852 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07001853 if (d_unhashed(dentry))
1854 continue;
Linus Torvalds8966be92012-03-02 14:23:30 -08001855 *seqp = seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001856
1857 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001858 if (dentry->d_name.hash != hashlen_hash(hashlen))
1859 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001860 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1861 case D_COMP_OK:
1862 return dentry;
1863 case D_COMP_NOMATCH:
1864 continue;
1865 default:
1866 goto seqretry;
1867 }
1868 }
1869
Linus Torvalds26fe5752012-05-10 13:14:12 -07001870 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07001871 continue;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001872 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001873 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11001874 }
1875 return NULL;
1876}
1877
1878/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 * d_lookup - search for a dentry
1880 * @parent: parent dentry
1881 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001882 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001884 * d_lookup searches the children of the parent dentry for the name in
1885 * question. If the dentry is found its reference count is incremented and the
1886 * dentry is returned. The caller must use dput to free the entry when it has
1887 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 */
Al Viroda2d8452013-01-24 18:29:34 -05001889struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890{
Nick Piggin31e6b012011-01-07 17:49:52 +11001891 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001892 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
1894 do {
1895 seq = read_seqbegin(&rename_lock);
1896 dentry = __d_lookup(parent, name);
1897 if (dentry)
1898 break;
1899 } while (read_seqretry(&rename_lock, seq));
1900 return dentry;
1901}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001902EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903
Nick Piggin31e6b012011-01-07 17:49:52 +11001904/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001905 * __d_lookup - search for a dentry (racy)
1906 * @parent: parent dentry
1907 * @name: qstr of name we wish to find
1908 * Returns: dentry, or NULL
1909 *
1910 * __d_lookup is like d_lookup, however it may (rarely) return a
1911 * false-negative result due to unrelated rename activity.
1912 *
1913 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1914 * however it must be used carefully, eg. with a following d_lookup in
1915 * the case of failure.
1916 *
1917 * __d_lookup callers must be commented.
1918 */
Al Viroa713ca22013-01-24 18:27:00 -05001919struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920{
1921 unsigned int len = name->len;
1922 unsigned int hash = name->hash;
1923 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001924 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001925 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001926 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001927 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928
Nick Pigginb04f7842010-08-18 04:37:34 +10001929 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001930 * Note: There is significant duplication with __d_lookup_rcu which is
1931 * required to prevent single threaded performance regressions
1932 * especially on architectures where smp_rmb (in seqcounts) are costly.
1933 * Keep the two functions in sync.
1934 */
1935
1936 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001937 * The hash list is protected using RCU.
1938 *
1939 * Take d_lock when comparing a candidate dentry, to avoid races
1940 * with d_move().
1941 *
1942 * It is possible that concurrent renames can mess up our list
1943 * walk here and result in missing our dentry, resulting in the
1944 * false-negative result. d_lookup() protects against concurrent
1945 * renames using rename_lock seqlock.
1946 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001947 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001948 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 rcu_read_lock();
1950
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001951 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 if (dentry->d_name.hash != hash)
1954 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955
1956 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (dentry->d_parent != parent)
1958 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001959 if (d_unhashed(dentry))
1960 goto next;
1961
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 /*
1963 * It is safe to compare names since d_move() cannot
1964 * change the qstr (protected by d_lock).
1965 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11001966 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001967 int tlen = dentry->d_name.len;
1968 const char *tname = dentry->d_name.name;
Nick Piggin621e1552011-01-07 17:49:27 +11001969 if (parent->d_op->d_compare(parent, parent->d_inode,
1970 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001971 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 goto next;
1973 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07001974 if (dentry->d_name.len != len)
1975 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001976 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 goto next;
1978 }
1979
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001980 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001981 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 spin_unlock(&dentry->d_lock);
1983 break;
1984next:
1985 spin_unlock(&dentry->d_lock);
1986 }
1987 rcu_read_unlock();
1988
1989 return found;
1990}
1991
1992/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001993 * d_hash_and_lookup - hash the qstr then search for a dentry
1994 * @dir: Directory to search in
1995 * @name: qstr of name we wish to find
1996 *
Al Viro4f522a22013-02-11 23:20:37 -05001997 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001998 */
1999struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2000{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002001 /*
2002 * Check for a fs-specific hash function. Note that we must
2003 * calculate the standard hash first, as the d_op->d_hash()
2004 * routine may choose to leave the hash value unchanged.
2005 */
2006 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002007 if (dir->d_flags & DCACHE_OP_HASH) {
Al Viro4f522a22013-02-11 23:20:37 -05002008 int err = dir->d_op->d_hash(dir, dir->d_inode, name);
2009 if (unlikely(err < 0))
2010 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002011 }
Al Viro4f522a22013-02-11 23:20:37 -05002012 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002013}
Al Viro4f522a22013-02-11 23:20:37 -05002014EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002015
2016/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002017 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002019 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 *
2021 * An insecure source has sent us a dentry, here we verify it and dget() it.
2022 * This is used by ncpfs in its readdir implementation.
2023 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002024 *
2025 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002027int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028{
Nick Piggin786a5e12011-01-07 17:49:16 +11002029 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002030
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002031 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002032 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2033 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002034 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002035 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002036 spin_unlock(&dentry->d_lock);
2037 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 return 1;
2039 }
2040 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002041 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002042
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return 0;
2044}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002045EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047/*
2048 * When a file is deleted, we have two options:
2049 * - turn this dentry into a negative dentry
2050 * - unhash this dentry and free it.
2051 *
2052 * Usually, we want to just turn this into
2053 * a negative dentry, but if anybody else is
2054 * currently using the dentry or the inode
2055 * we can't do that and we fall back on removing
2056 * it from the hash queues and waiting for
2057 * it to be deleted later when it has no users
2058 */
2059
2060/**
2061 * d_delete - delete a dentry
2062 * @dentry: The dentry to delete
2063 *
2064 * Turn the dentry into a negative dentry if possible, otherwise
2065 * remove it from the hash queues so it can be deleted later
2066 */
2067
2068void d_delete(struct dentry * dentry)
2069{
Nick Piggin873feea2011-01-07 17:50:06 +11002070 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002071 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 /*
2073 * Are we the only user?
2074 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002075again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002077 inode = dentry->d_inode;
2078 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002079 if (dentry->d_count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002080 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002081 spin_unlock(&dentry->d_lock);
2082 cpu_relax();
2083 goto again;
2084 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002085 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002086 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002087 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 return;
2089 }
2090
2091 if (!d_unhashed(dentry))
2092 __d_drop(dentry);
2093
2094 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002095
2096 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002098EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002100static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002102 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002103 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002104 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002105 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002106 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107}
2108
David Howells770bfad2006-08-22 20:06:07 -04002109static void _d_rehash(struct dentry * entry)
2110{
2111 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2112}
2113
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114/**
2115 * d_rehash - add an entry back to the hash
2116 * @entry: dentry to add to the hash
2117 *
2118 * Adds a dentry to the hash according to its name.
2119 */
2120
2121void d_rehash(struct dentry * entry)
2122{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002124 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002127EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002129/**
2130 * dentry_update_name_case - update case insensitive dentry with a new name
2131 * @dentry: dentry to be updated
2132 * @name: new name
2133 *
2134 * Update a case insensitive dentry with new case of name.
2135 *
2136 * dentry must have been returned by d_lookup with name @name. Old and new
2137 * name lengths must match (ie. no d_compare which allows mismatched name
2138 * lengths).
2139 *
2140 * Parent inode i_mutex must be held over d_lookup and into this call (to
2141 * keep renames and concurrent inserts, and readdir(2) away).
2142 */
2143void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2144{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002145 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002146 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2147
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002148 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002149 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002150 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002151 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002152 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002153}
2154EXPORT_SYMBOL(dentry_update_name_case);
2155
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156static void switch_names(struct dentry *dentry, struct dentry *target)
2157{
2158 if (dname_external(target)) {
2159 if (dname_external(dentry)) {
2160 /*
2161 * Both external: swap the pointers
2162 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002163 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 } else {
2165 /*
2166 * dentry:internal, target:external. Steal target's
2167 * storage and make target internal.
2168 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002169 memcpy(target->d_iname, dentry->d_name.name,
2170 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 dentry->d_name.name = target->d_name.name;
2172 target->d_name.name = target->d_iname;
2173 }
2174 } else {
2175 if (dname_external(dentry)) {
2176 /*
2177 * dentry:external, target:internal. Give dentry's
2178 * storage to target and make dentry internal
2179 */
2180 memcpy(dentry->d_iname, target->d_name.name,
2181 target->d_name.len + 1);
2182 target->d_name.name = dentry->d_name.name;
2183 dentry->d_name.name = dentry->d_iname;
2184 } else {
2185 /*
2186 * Both are internal. Just copy target to dentry
2187 */
2188 memcpy(dentry->d_iname, target->d_name.name,
2189 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002190 dentry->d_name.len = target->d_name.len;
2191 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 }
2193 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002194 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195}
2196
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002197static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2198{
2199 /*
2200 * XXXX: do we really need to take target->d_lock?
2201 */
2202 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2203 spin_lock(&target->d_parent->d_lock);
2204 else {
2205 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2206 spin_lock(&dentry->d_parent->d_lock);
2207 spin_lock_nested(&target->d_parent->d_lock,
2208 DENTRY_D_LOCK_NESTED);
2209 } else {
2210 spin_lock(&target->d_parent->d_lock);
2211 spin_lock_nested(&dentry->d_parent->d_lock,
2212 DENTRY_D_LOCK_NESTED);
2213 }
2214 }
2215 if (target < dentry) {
2216 spin_lock_nested(&target->d_lock, 2);
2217 spin_lock_nested(&dentry->d_lock, 3);
2218 } else {
2219 spin_lock_nested(&dentry->d_lock, 2);
2220 spin_lock_nested(&target->d_lock, 3);
2221 }
2222}
2223
2224static void dentry_unlock_parents_for_move(struct dentry *dentry,
2225 struct dentry *target)
2226{
2227 if (target->d_parent != dentry->d_parent)
2228 spin_unlock(&dentry->d_parent->d_lock);
2229 if (target->d_parent != target)
2230 spin_unlock(&target->d_parent->d_lock);
2231}
2232
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002234 * When switching names, the actual string doesn't strictly have to
2235 * be preserved in the target - because we're dropping the target
2236 * anyway. As such, we can just do a simple memcpy() to copy over
2237 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002239 * Note that we have to be a lot more careful about getting the hash
2240 * switched - we have to switch the hash value properly even if it
2241 * then no longer matches the actual (corrupted) string of the target.
2242 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002244/*
Al Viro18367502011-07-12 21:42:24 -04002245 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 * @dentry: entry to move
2247 * @target: new dentry
2248 *
2249 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002250 * dcache entries should not be moved in this way. Caller must hold
2251 * rename_lock, the i_mutex of the source and target directories,
2252 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 */
Al Viro18367502011-07-12 21:42:24 -04002254static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 if (!dentry->d_inode)
2257 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2258
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002259 BUG_ON(d_ancestor(dentry, target));
2260 BUG_ON(d_ancestor(target, dentry));
2261
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002262 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
Nick Piggin31e6b012011-01-07 17:49:52 +11002264 write_seqcount_begin(&dentry->d_seq);
2265 write_seqcount_begin(&target->d_seq);
2266
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002267 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2268
2269 /*
2270 * Move the dentry to the target hash queue. Don't bother checking
2271 * for the same hash queue because of how unlikely it is.
2272 */
2273 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002274 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275
2276 /* Unhash the target: dput() will then get rid of it */
2277 __d_drop(target);
2278
Eric Dumazet5160ee62006-01-08 01:03:32 -08002279 list_del(&dentry->d_u.d_child);
2280 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
2282 /* Switch the names.. */
2283 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002284 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
2286 /* ... and switch the parents */
2287 if (IS_ROOT(dentry)) {
2288 dentry->d_parent = target->d_parent;
2289 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002290 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002292 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293
2294 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002295 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 }
2297
Eric Dumazet5160ee62006-01-08 01:03:32 -08002298 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002299
Nick Piggin31e6b012011-01-07 17:49:52 +11002300 write_seqcount_end(&target->d_seq);
2301 write_seqcount_end(&dentry->d_seq);
2302
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002303 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002305 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002307}
2308
2309/*
2310 * d_move - move a dentry
2311 * @dentry: entry to move
2312 * @target: new dentry
2313 *
2314 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002315 * dcache entries should not be moved in this way. See the locking
2316 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002317 */
2318void d_move(struct dentry *dentry, struct dentry *target)
2319{
2320 write_seqlock(&rename_lock);
2321 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002323}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002324EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002326/**
2327 * d_ancestor - search for an ancestor
2328 * @p1: ancestor dentry
2329 * @p2: child dentry
2330 *
2331 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2332 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002333 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002334struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002335{
2336 struct dentry *p;
2337
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002338 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002339 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002340 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002341 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002342 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002343}
2344
2345/*
2346 * This helper attempts to cope with remotely renamed directories
2347 *
2348 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002349 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002350 *
2351 * Note: If ever the locking in lock_rename() changes, then please
2352 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002353 */
Nick Piggin873feea2011-01-07 17:50:06 +11002354static struct dentry *__d_unalias(struct inode *inode,
2355 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002356{
2357 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002358 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002359
2360 /* If alias and dentry share a parent, then no extra locks required */
2361 if (alias->d_parent == dentry->d_parent)
2362 goto out_unalias;
2363
Trond Myklebust9eaef272006-10-21 10:24:20 -07002364 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002365 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2366 goto out_err;
2367 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2368 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2369 goto out_err;
2370 m2 = &alias->d_parent->d_inode->i_mutex;
2371out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002372 if (likely(!d_mountpoint(alias))) {
2373 __d_move(alias, dentry);
2374 ret = alias;
2375 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002376out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002377 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002378 if (m2)
2379 mutex_unlock(m2);
2380 if (m1)
2381 mutex_unlock(m1);
2382 return ret;
2383}
2384
2385/*
David Howells770bfad2006-08-22 20:06:07 -04002386 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2387 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002388 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002389 */
2390static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2391{
Al Viro740da422013-01-30 10:13:38 -05002392 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002393
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002394 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002395
Nick Piggin31e6b012011-01-07 17:49:52 +11002396 write_seqcount_begin(&dentry->d_seq);
2397 write_seqcount_begin(&anon->d_seq);
2398
David Howells770bfad2006-08-22 20:06:07 -04002399 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002400
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002401 switch_names(dentry, anon);
2402 swap(dentry->d_name.hash, anon->d_name.hash);
2403
Al Viro740da422013-01-30 10:13:38 -05002404 dentry->d_parent = dentry;
2405 list_del_init(&dentry->d_u.d_child);
2406 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002407 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002408
Nick Piggin31e6b012011-01-07 17:49:52 +11002409 write_seqcount_end(&dentry->d_seq);
2410 write_seqcount_end(&anon->d_seq);
2411
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002412 dentry_unlock_parents_for_move(anon, dentry);
2413 spin_unlock(&dentry->d_lock);
2414
2415 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002416 anon->d_flags &= ~DCACHE_DISCONNECTED;
2417}
2418
2419/**
2420 * d_materialise_unique - introduce an inode into the tree
2421 * @dentry: candidate dentry
2422 * @inode: inode to bind to the dentry, to which aliases may be attached
2423 *
2424 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002425 * root directory alias in its place if there is one. Caller must hold the
2426 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002427 */
2428struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2429{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002430 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002431
2432 BUG_ON(!d_unhashed(dentry));
2433
David Howells770bfad2006-08-22 20:06:07 -04002434 if (!inode) {
2435 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002436 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002437 d_rehash(actual);
2438 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002439 }
2440
Nick Piggin873feea2011-01-07 17:50:06 +11002441 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002442
Trond Myklebust9eaef272006-10-21 10:24:20 -07002443 if (S_ISDIR(inode->i_mode)) {
2444 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002445
Trond Myklebust9eaef272006-10-21 10:24:20 -07002446 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002447 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002448 if (alias) {
2449 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002450 write_seqlock(&rename_lock);
2451
2452 if (d_ancestor(alias, dentry)) {
2453 /* Check for loops */
2454 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002455 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002456 } else if (IS_ROOT(alias)) {
2457 /* Is this an anonymous mountpoint that we
2458 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002459 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002460 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002461 __d_drop(alias);
2462 goto found;
Al Viro18367502011-07-12 21:42:24 -04002463 } else {
2464 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002465 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002466 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002467 }
Al Viro18367502011-07-12 21:42:24 -04002468 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002469 if (IS_ERR(actual)) {
2470 if (PTR_ERR(actual) == -ELOOP)
2471 pr_warn_ratelimited(
2472 "VFS: Lookup of '%s' in %s %s"
2473 " would have caused loop\n",
2474 dentry->d_name.name,
2475 inode->i_sb->s_type->name,
2476 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002477 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002478 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002479 goto out_nolock;
2480 }
David Howells770bfad2006-08-22 20:06:07 -04002481 }
2482
2483 /* Add a unique reference */
2484 actual = __d_instantiate_unique(dentry, inode);
2485 if (!actual)
2486 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002487 else
2488 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002489
David Howells770bfad2006-08-22 20:06:07 -04002490 spin_lock(&actual->d_lock);
2491found:
2492 _d_rehash(actual);
2493 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002494 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002495out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002496 if (actual == dentry) {
2497 security_d_instantiate(dentry, inode);
2498 return NULL;
2499 }
2500
2501 iput(inode);
2502 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002503}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002504EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002505
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002506static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002507{
2508 *buflen -= namelen;
2509 if (*buflen < 0)
2510 return -ENAMETOOLONG;
2511 *buffer -= namelen;
2512 memcpy(*buffer, str, namelen);
2513 return 0;
2514}
2515
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002516static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2517{
2518 return prepend(buffer, buflen, name->name, name->len);
2519}
2520
Linus Torvalds1da177e2005-04-16 15:20:36 -07002521/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002522 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002523 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002524 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002525 * @buffer: pointer to the end of the buffer
2526 * @buflen: pointer to buffer length
2527 *
Nick Piggin949854d2011-01-07 17:49:37 +11002528 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002529 */
Al Viro02125a82011-12-05 08:43:34 -05002530static int prepend_path(const struct path *path,
2531 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002532 char **buffer, int *buflen)
2533{
2534 struct dentry *dentry = path->dentry;
2535 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002536 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002537 bool slash = false;
2538 int error = 0;
2539
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002540 while (dentry != root->dentry || vfsmnt != root->mnt) {
2541 struct dentry * parent;
2542
2543 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2544 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002545 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002546 goto global_root;
Al Viroa73324d2011-11-24 22:25:07 -05002547 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002548 mnt = mnt->mnt_parent;
2549 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002550 continue;
2551 }
2552 parent = dentry->d_parent;
2553 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002554 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002555 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002556 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002557 if (!error)
2558 error = prepend(buffer, buflen, "/", 1);
2559 if (error)
2560 break;
2561
2562 slash = true;
2563 dentry = parent;
2564 }
2565
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002566 if (!error && !slash)
2567 error = prepend(buffer, buflen, "/", 1);
2568
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002569 return error;
2570
2571global_root:
2572 /*
2573 * Filesystems needing to implement special "root names"
2574 * should do so with ->d_dname()
2575 */
2576 if (IS_ROOT(dentry) &&
2577 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2578 WARN(1, "Root dentry has weird name <%.*s>\n",
2579 (int) dentry->d_name.len, dentry->d_name.name);
2580 }
Al Viro02125a82011-12-05 08:43:34 -05002581 if (!slash)
2582 error = prepend(buffer, buflen, "/", 1);
2583 if (!error)
Al Virof7a99c52012-06-09 00:59:08 -04002584 error = is_mounted(vfsmnt) ? 1 : 2;
Al Viro7ea600b2013-03-26 18:25:57 -04002585 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002586}
2587
2588/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002589 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002590 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002591 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002592 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 * @buflen: buffer length
2594 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002595 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002597 * Returns a pointer into the buffer or an error code if the
2598 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002599 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002600 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002601 *
Al Viro02125a82011-12-05 08:43:34 -05002602 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 */
Al Viro02125a82011-12-05 08:43:34 -05002604char *__d_path(const struct path *path,
2605 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002606 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002608 char *res = buf + buflen;
2609 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002611 prepend(&res, &buflen, "\0", 1);
Al Viro7ea600b2013-03-26 18:25:57 -04002612 br_read_lock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002613 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002614 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002615 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002616 br_read_unlock(&vfsmount_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002617
Al Viro02125a82011-12-05 08:43:34 -05002618 if (error < 0)
2619 return ERR_PTR(error);
2620 if (error > 0)
2621 return NULL;
2622 return res;
2623}
2624
2625char *d_absolute_path(const struct path *path,
2626 char *buf, int buflen)
2627{
2628 struct path root = {};
2629 char *res = buf + buflen;
2630 int error;
2631
2632 prepend(&res, &buflen, "\0", 1);
Al Viro7ea600b2013-03-26 18:25:57 -04002633 br_read_lock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002634 write_seqlock(&rename_lock);
2635 error = prepend_path(path, &root, &res, &buflen);
2636 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002637 br_read_unlock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002638
2639 if (error > 1)
2640 error = -EINVAL;
2641 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002642 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002643 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644}
2645
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002646/*
2647 * same as __d_path but appends "(deleted)" for unlinked files.
2648 */
Al Viro02125a82011-12-05 08:43:34 -05002649static int path_with_deleted(const struct path *path,
2650 const struct path *root,
2651 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002652{
2653 prepend(buf, buflen, "\0", 1);
2654 if (d_unlinked(path->dentry)) {
2655 int error = prepend(buf, buflen, " (deleted)", 10);
2656 if (error)
2657 return error;
2658 }
2659
2660 return prepend_path(path, root, buf, buflen);
2661}
2662
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002663static int prepend_unreachable(char **buffer, int *buflen)
2664{
2665 return prepend(buffer, buflen, "(unreachable)", 13);
2666}
2667
Jan Bluncka03a8a702008-02-14 19:38:32 -08002668/**
2669 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002670 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002671 * @buf: buffer to return value in
2672 * @buflen: buffer length
2673 *
2674 * Convert a dentry into an ASCII path name. If the entry has been deleted
2675 * the string " (deleted)" is appended. Note that this is ambiguous.
2676 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002677 * Returns a pointer into the buffer or an error code if the path was
2678 * too long. Note: Callers should use the returned pointer, not the passed
2679 * in buffer, to use the name! The implementation often starts at an offset
2680 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002681 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002682 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002683 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002684char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002686 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002687 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002688 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002690 /*
2691 * We have various synthetic filesystems that never get mounted. On
2692 * these filesystems dentries are never used for lookup purposes, and
2693 * thus don't need to be hashed. They also don't need a name until a
2694 * user wants to identify the object in /proc/pid/fd/. The little hack
2695 * below allows us to generate a name for these objects on demand:
2696 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002697 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2698 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002699
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002700 get_fs_root(current->fs, &root);
Al Viro7ea600b2013-03-26 18:25:57 -04002701 br_read_lock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002702 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002703 error = path_with_deleted(path, &root, &res, &buflen);
Al Viro7ea600b2013-03-26 18:25:57 -04002704 write_sequnlock(&rename_lock);
2705 br_read_unlock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002706 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002707 res = ERR_PTR(error);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002708 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 return res;
2710}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002711EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
2713/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002714 * Helper function for dentry_operations.d_dname() members
2715 */
2716char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2717 const char *fmt, ...)
2718{
2719 va_list args;
2720 char temp[64];
2721 int sz;
2722
2723 va_start(args, fmt);
2724 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2725 va_end(args);
2726
2727 if (sz > sizeof(temp) || sz > buflen)
2728 return ERR_PTR(-ENAMETOOLONG);
2729
2730 buffer += buflen - sz;
2731 return memcpy(buffer, temp, sz);
2732}
2733
2734/*
Ram Pai6092d042008-03-27 13:06:20 +01002735 * Write full pathname from the root of the filesystem into the buffer.
2736 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002737static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002738{
2739 char *end = buf + buflen;
2740 char *retval;
2741
Ram Pai6092d042008-03-27 13:06:20 +01002742 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002743 if (buflen < 1)
2744 goto Elong;
2745 /* Get '/' right */
2746 retval = end-1;
2747 *retval = '/';
2748
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002749 while (!IS_ROOT(dentry)) {
2750 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002751 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002752
Ram Pai6092d042008-03-27 13:06:20 +01002753 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002754 spin_lock(&dentry->d_lock);
2755 error = prepend_name(&end, &buflen, &dentry->d_name);
2756 spin_unlock(&dentry->d_lock);
2757 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002758 goto Elong;
2759
2760 retval = end;
2761 dentry = parent;
2762 }
Al Viroc1031352010-06-06 22:31:14 -04002763 return retval;
2764Elong:
2765 return ERR_PTR(-ENAMETOOLONG);
2766}
Nick Pigginec2447c2011-01-07 17:49:29 +11002767
2768char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2769{
2770 char *retval;
2771
Nick Piggin949854d2011-01-07 17:49:37 +11002772 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002773 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002774 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002775
2776 return retval;
2777}
2778EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002779
2780char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2781{
2782 char *p = NULL;
2783 char *retval;
2784
Nick Piggin949854d2011-01-07 17:49:37 +11002785 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002786 if (d_unlinked(dentry)) {
2787 p = buf + buflen;
2788 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2789 goto Elong;
2790 buflen++;
2791 }
2792 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002793 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002794 if (!IS_ERR(retval) && p)
2795 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002796 return retval;
2797Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002798 return ERR_PTR(-ENAMETOOLONG);
2799}
2800
2801/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 * NOTE! The user-level library version returns a
2803 * character pointer. The kernel system call just
2804 * returns the length of the buffer filled (which
2805 * includes the ending '\0' character), or a negative
2806 * error value. So libc would do something like
2807 *
2808 * char *getcwd(char * buf, size_t size)
2809 * {
2810 * int retval;
2811 *
2812 * retval = sys_getcwd(buf, size);
2813 * if (retval >= 0)
2814 * return buf;
2815 * errno = -retval;
2816 * return NULL;
2817 * }
2818 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002819SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820{
Linus Torvalds552ce542007-02-13 12:08:18 -08002821 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002822 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002823 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824
2825 if (!page)
2826 return -ENOMEM;
2827
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002828 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
Linus Torvalds552ce542007-02-13 12:08:18 -08002830 error = -ENOENT;
Al Viro7ea600b2013-03-26 18:25:57 -04002831 br_read_lock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002832 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002833 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002834 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002835 char *cwd = page + PAGE_SIZE;
2836 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002838 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002839 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002840 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002841 br_read_unlock(&vfsmount_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002842
Al Viro02125a82011-12-05 08:43:34 -05002843 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002844 goto out;
2845
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002846 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002847 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002848 error = prepend_unreachable(&cwd, &buflen);
2849 if (error)
2850 goto out;
2851 }
2852
Linus Torvalds552ce542007-02-13 12:08:18 -08002853 error = -ERANGE;
2854 len = PAGE_SIZE + page - cwd;
2855 if (len <= size) {
2856 error = len;
2857 if (copy_to_user(buf, cwd, len))
2858 error = -EFAULT;
2859 }
Nick Piggin949854d2011-01-07 17:49:37 +11002860 } else {
2861 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002862 br_read_unlock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864
2865out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002866 path_put(&pwd);
2867 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 free_page((unsigned long) page);
2869 return error;
2870}
2871
2872/*
2873 * Test whether new_dentry is a subdirectory of old_dentry.
2874 *
2875 * Trivially implemented using the dcache structure
2876 */
2877
2878/**
2879 * is_subdir - is new dentry a subdirectory of old_dentry
2880 * @new_dentry: new dentry
2881 * @old_dentry: old dentry
2882 *
2883 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2884 * Returns 0 otherwise.
2885 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2886 */
2887
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002888int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889{
2890 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002891 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002893 if (new_dentry == old_dentry)
2894 return 1;
2895
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002896 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002899 /*
2900 * Need rcu_readlock to protect against the d_parent trashing
2901 * due to d_move
2902 */
2903 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002904 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002906 else
2907 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002908 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910
2911 return result;
2912}
2913
2914void d_genocide(struct dentry *root)
2915{
Nick Piggin949854d2011-01-07 17:49:37 +11002916 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002918 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002919 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920
Nick Piggin949854d2011-01-07 17:49:37 +11002921 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002922again:
2923 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002924 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925repeat:
2926 next = this_parent->d_subdirs.next;
2927resume:
2928 while (next != &this_parent->d_subdirs) {
2929 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002930 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002932
Nick Pigginda502952011-01-07 17:49:33 +11002933 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2934 if (d_unhashed(dentry) || !dentry->d_inode) {
2935 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002937 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002939 spin_unlock(&this_parent->d_lock);
2940 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002942 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 goto repeat;
2944 }
Nick Piggin949854d2011-01-07 17:49:37 +11002945 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2946 dentry->d_flags |= DCACHE_GENOCIDE;
2947 dentry->d_count--;
2948 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002949 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 }
2951 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002952 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11002953 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2954 this_parent->d_flags |= DCACHE_GENOCIDE;
2955 this_parent->d_count--;
2956 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002957 this_parent = try_to_ascend(this_parent, locked, seq);
2958 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11002959 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11002960 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 goto resume;
2962 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002963 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002964 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11002965 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11002966 if (locked)
2967 write_sequnlock(&rename_lock);
2968 return;
2969
2970rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02002971 if (locked)
2972 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11002973 locked = 1;
2974 write_seqlock(&rename_lock);
2975 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976}
2977
2978/**
2979 * find_inode_number - check for dentry with name
2980 * @dir: directory to check
2981 * @name: Name to find.
2982 *
2983 * Check whether a dentry already exists for the given name,
2984 * and return the inode number if it has an inode. Otherwise
2985 * 0 is returned.
2986 *
2987 * This routine is used to post-process directory listings for
2988 * filesystems using synthetic inode numbers, and is necessary
2989 * to keep getcwd() working.
2990 */
2991
2992ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2993{
2994 struct dentry * dentry;
2995 ino_t ino = 0;
2996
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002997 dentry = d_hash_and_lookup(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002998 if (!IS_ERR_OR_NULL(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 if (dentry->d_inode)
3000 ino = dentry->d_inode->i_ino;
3001 dput(dentry);
3002 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 return ino;
3004}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003005EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006
3007static __initdata unsigned long dhash_entries;
3008static int __init set_dhash_entries(char *str)
3009{
3010 if (!str)
3011 return 0;
3012 dhash_entries = simple_strtoul(str, &str, 0);
3013 return 1;
3014}
3015__setup("dhash_entries=", set_dhash_entries);
3016
3017static void __init dcache_init_early(void)
3018{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003019 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020
3021 /* If hashes are distributed across NUMA nodes, defer
3022 * hash allocation until vmalloc space is available.
3023 */
3024 if (hashdist)
3025 return;
3026
3027 dentry_hashtable =
3028 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003029 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 dhash_entries,
3031 13,
3032 HASH_EARLY,
3033 &d_hash_shift,
3034 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003035 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 0);
3037
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003038 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003039 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040}
3041
Denis Cheng74bf17c2007-10-16 23:26:30 -07003042static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003044 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
3046 /*
3047 * A constructor could be added for stable state like the lists,
3048 * but it is probably not worth it because of the cache nature
3049 * of the dcache.
3050 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003051 dentry_cache = KMEM_CACHE(dentry,
3052 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053
3054 /* Hash may have been set up in dcache_init_early */
3055 if (!hashdist)
3056 return;
3057
3058 dentry_hashtable =
3059 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003060 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 dhash_entries,
3062 13,
3063 0,
3064 &d_hash_shift,
3065 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003066 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 0);
3068
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003069 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003070 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071}
3072
3073/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003074struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003075EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077EXPORT_SYMBOL(d_genocide);
3078
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079void __init vfs_caches_init_early(void)
3080{
3081 dcache_init_early();
3082 inode_init_early();
3083}
3084
3085void __init vfs_caches_init(unsigned long mempages)
3086{
3087 unsigned long reserve;
3088
3089 /* Base hash sizes on available memory, with a reserve equal to
3090 150% of current kernel size */
3091
3092 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3093 mempages -= reserve;
3094
3095 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003096 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097
Denis Cheng74bf17c2007-10-16 23:26:30 -07003098 dcache_init();
3099 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003101 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 bdev_cache_init();
3103 chrdev_init();
3104}