blob: 2d244227999d151de8f12c423fed5d7faa99d770 [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{
Waiman Long98474232013-08-28 18:24:59 -0700232 BUG_ON(dentry->d_lockref.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)
Waiman Long98474232013-08-28 18:24:59 -0700470 dentry->d_lockref.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:
Waiman Long98474232013-08-28 18:24:59 -0700516 if (dentry->d_lockref.count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 might_sleep();
Waiman Long98474232013-08-28 18:24:59 -0700518 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
Nick Pigginfb045ad2011-01-07 17:49:55 +1100521 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100523 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 }
Nick Piggin265ac902010-10-10 05:36:24 -0400525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 /* Unreachable? Get rid of it */
527 if (d_unhashed(dentry))
528 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400529
Jeff Layton39e3c952012-11-28 11:30:53 -0500530 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400531 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400532
Waiman Long98474232013-08-28 18:24:59 -0700533 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100534 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return;
536
Miklos Szeredid52b9082007-05-08 00:23:46 -0700537kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100538 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700539 if (dentry)
540 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700542EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543
544/**
545 * d_invalidate - invalidate a dentry
546 * @dentry: dentry to invalidate
547 *
548 * Try to invalidate the dentry if it turns out to be
549 * possible. If there are other dentries that can be
550 * reached through this one we can't delete it and we
551 * return -EBUSY. On success we return 0.
552 *
553 * no dcache lock.
554 */
555
556int d_invalidate(struct dentry * dentry)
557{
558 /*
559 * If it's already been dropped, return OK.
560 */
Nick Pigginda502952011-01-07 17:49:33 +1100561 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100563 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return 0;
565 }
566 /*
567 * Check whether to do a partial shrink_dcache
568 * to get rid of unused child entries.
569 */
570 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100571 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100573 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
575
576 /*
577 * Somebody else still using it?
578 *
579 * If it's a directory, we can't drop it
580 * for fear of somebody re-populating it
581 * with children (even though dropping it
582 * would make it unreachable from the root,
583 * we might still populate it if it was a
584 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000585 * We also need to leave mountpoints alone,
586 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
Waiman Long98474232013-08-28 18:24:59 -0700588 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000589 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 return -EBUSY;
592 }
593 }
594
595 __d_drop(dentry);
596 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 return 0;
598}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700599EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100601/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100602static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
Waiman Long98474232013-08-28 18:24:59 -0700604 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
Nick Piggindc0474b2011-01-07 17:49:43 +1100607static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100608{
Waiman Long98474232013-08-28 18:24:59 -0700609 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100610}
611
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100612struct dentry *dget_parent(struct dentry *dentry)
613{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700614 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100615 struct dentry *ret;
616
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700617 /*
618 * Do optimistic parent lookup without any
619 * locking.
620 */
621 rcu_read_lock();
622 ret = ACCESS_ONCE(dentry->d_parent);
623 gotref = lockref_get_not_zero(&ret->d_lockref);
624 rcu_read_unlock();
625 if (likely(gotref)) {
626 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
627 return ret;
628 dput(ret);
629 }
630
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100631repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100632 /*
633 * Don't need rcu_dereference because we re-check it was correct under
634 * the lock.
635 */
636 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100637 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100638 spin_lock(&ret->d_lock);
639 if (unlikely(ret != dentry->d_parent)) {
640 spin_unlock(&ret->d_lock);
641 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100642 goto repeat;
643 }
Nick Piggina734eb42011-01-07 17:49:44 +1100644 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700645 BUG_ON(!ret->d_lockref.count);
646 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100647 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100648 return ret;
649}
650EXPORT_SYMBOL(dget_parent);
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652/**
653 * d_find_alias - grab a hashed alias of inode
654 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700655 * @want_discon: flag, used by d_splice_alias, to request
656 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 *
658 * If inode has a hashed alias, or is a directory and has any alias,
659 * acquire the reference to alias and return it. Otherwise return NULL.
660 * Notice that if inode is a directory there can be only one alias and
661 * it can be unhashed only if it has no children, or if it is the root
662 * of a filesystem.
663 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700664 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700665 * any other hashed alias over that one unless @want_discon is set,
666 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700668static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
Nick Pigginda502952011-01-07 17:49:33 +1100670 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671
Nick Pigginda502952011-01-07 17:49:33 +1100672again:
673 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800674 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100675 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700677 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100678 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700680 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100681 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100682 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 return alias;
684 }
685 }
Nick Pigginda502952011-01-07 17:49:33 +1100686 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
Nick Pigginda502952011-01-07 17:49:33 +1100688 if (discon_alias) {
689 alias = discon_alias;
690 spin_lock(&alias->d_lock);
691 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
692 if (IS_ROOT(alias) &&
693 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100694 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100695 spin_unlock(&alias->d_lock);
696 return alias;
697 }
698 }
699 spin_unlock(&alias->d_lock);
700 goto again;
701 }
702 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703}
704
Nick Pigginda502952011-01-07 17:49:33 +1100705struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
David Howells214fda12006-03-25 03:06:36 -0800707 struct dentry *de = NULL;
708
Al Virob3d9b7a2012-06-09 13:51:19 -0400709 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100710 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700711 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100712 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800713 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 return de;
715}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700716EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
718/*
719 * Try to kill dentries associated with this inode.
720 * WARNING: you must own a reference to inode.
721 */
722void d_prune_aliases(struct inode *inode)
723{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700724 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100726 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800727 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700729 if (!dentry->d_lockref.count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100730 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 __d_drop(dentry);
732 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100733 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 dput(dentry);
735 goto restart;
736 }
737 spin_unlock(&dentry->d_lock);
738 }
Nick Piggin873feea2011-01-07 17:50:06 +1100739 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700741EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742
743/*
Nick Piggin77812a12011-01-07 17:49:48 +1100744 * Try to throw away a dentry - free the inode, dput the parent.
745 * Requires dentry->d_lock is held, and dentry->d_count == 0.
746 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700747 *
Nick Piggin77812a12011-01-07 17:49:48 +1100748 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 */
Nick Piggin77812a12011-01-07 17:49:48 +1100750static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200751 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
Nick Piggin77812a12011-01-07 17:49:48 +1100753 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700754
Nick Piggin77812a12011-01-07 17:49:48 +1100755 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700756 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100757 * If dentry_kill returns NULL, we have nothing more to do.
758 * if it returns the same dentry, trylocks failed. In either
759 * case, just loop again.
760 *
761 * Otherwise, we need to prune ancestors too. This is necessary
762 * to prevent quadratic behavior of shrink_dcache_parent(), but
763 * is also expected to be beneficial in reducing dentry cache
764 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700765 */
Nick Piggin77812a12011-01-07 17:49:48 +1100766 if (!parent)
767 return;
768 if (parent == dentry)
769 return;
770
771 /* Prune ancestors. */
772 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700773 while (dentry) {
Waiman Long98474232013-08-28 18:24:59 -0700774 if (lockref_put_or_lock(&dentry->d_lockref))
Nick Piggin89e60542011-01-07 17:49:45 +1100775 return;
Nick Piggin77812a12011-01-07 17:49:48 +1100776 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
779
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400780static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700782 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700783
Nick Pigginec336792011-01-07 17:49:47 +1100784 rcu_read_lock();
785 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100786 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
787 if (&dentry->d_lru == list)
788 break; /* empty */
789 spin_lock(&dentry->d_lock);
790 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
791 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100792 continue;
793 }
794
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 /*
796 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700797 * the LRU because of laziness during lookup. Do not free
798 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Waiman Long98474232013-08-28 18:24:59 -0700800 if (dentry->d_lockref.count) {
Nick Pigginec336792011-01-07 17:49:47 +1100801 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700802 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 continue;
804 }
Nick Pigginec336792011-01-07 17:49:47 +1100805
Nick Pigginec336792011-01-07 17:49:47 +1100806 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100807
808 try_prune_one_dentry(dentry);
809
Nick Pigginec336792011-01-07 17:49:47 +1100810 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 }
Nick Pigginec336792011-01-07 17:49:47 +1100812 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400813}
814
815/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000816 * prune_dcache_sb - shrink the dcache
817 * @sb: superblock
818 * @count: number of entries to try to free
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400819 *
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000820 * Attempt to shrink the superblock dcache LRU by @count entries. This is
821 * done when we need more memory an called from the superblock shrinker
822 * function.
823 *
824 * This function may fail to free any resources if all the dentries are in
825 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400826 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000827void prune_dcache_sb(struct super_block *sb, int count)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400828{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400829 struct dentry *dentry;
830 LIST_HEAD(referenced);
831 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400832
Nick Piggin23044502011-01-07 17:49:31 +1100833relock:
834 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400835 while (!list_empty(&sb->s_dentry_lru)) {
836 dentry = list_entry(sb->s_dentry_lru.prev,
837 struct dentry, d_lru);
838 BUG_ON(dentry->d_sb != sb);
839
Nick Piggin23044502011-01-07 17:49:31 +1100840 if (!spin_trylock(&dentry->d_lock)) {
841 spin_unlock(&dcache_lru_lock);
842 cpu_relax();
843 goto relock;
844 }
845
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000846 if (dentry->d_flags & DCACHE_REFERENCED) {
Nick Piggin23044502011-01-07 17:49:31 +1100847 dentry->d_flags &= ~DCACHE_REFERENCED;
848 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400849 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100850 } else {
851 list_move_tail(&dentry->d_lru, &tmp);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100852 dentry->d_flags |= DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100853 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000854 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100855 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400856 }
Nick Pigginec336792011-01-07 17:49:47 +1100857 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400858 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700859 if (!list_empty(&referenced))
860 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100861 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100862
863 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864}
865
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700866/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 * shrink_dcache_sb - shrink dcache for a superblock
868 * @sb: superblock
869 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400870 * Shrink the dcache for the specified super block. This is used to free
871 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400873void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400875 LIST_HEAD(tmp);
876
Nick Piggin23044502011-01-07 17:49:31 +1100877 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400878 while (!list_empty(&sb->s_dentry_lru)) {
879 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100880 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400881 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100882 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400883 }
Nick Piggin23044502011-01-07 17:49:31 +1100884 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700886EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
888/*
David Howellsc636ebd2006-10-11 01:22:19 -0700889 * destroy a single subtree of dentries for unmount
890 * - see the comments on shrink_dcache_for_umount() for a description of the
891 * locking
892 */
893static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
894{
895 struct dentry *parent;
896
897 BUG_ON(!IS_ROOT(dentry));
898
David Howellsc636ebd2006-10-11 01:22:19 -0700899 for (;;) {
900 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100901 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700902 dentry = list_entry(dentry->d_subdirs.next,
903 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700904
905 /* consume the dentries from this leaf up through its parents
906 * until we find one with children or run out altogether */
907 do {
908 struct inode *inode;
909
Sage Weilf0023bc2011-10-28 10:02:42 -0700910 /*
Yan, Zheng61572bb2013-04-15 14:13:21 +0800911 * inform the fs that this dentry is about to be
Sage Weilf0023bc2011-10-28 10:02:42 -0700912 * unhashed and destroyed.
913 */
Yan, Zheng61572bb2013-04-15 14:13:21 +0800914 if (dentry->d_flags & DCACHE_OP_PRUNE)
915 dentry->d_op->d_prune(dentry);
916
917 dentry_lru_del(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100918 __d_shrink(dentry);
919
Waiman Long98474232013-08-28 18:24:59 -0700920 if (dentry->d_lockref.count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700921 printk(KERN_ERR
922 "BUG: Dentry %p{i=%lx,n=%s}"
923 " still in use (%d)"
924 " [unmount of %s %s]\n",
925 dentry,
926 dentry->d_inode ?
927 dentry->d_inode->i_ino : 0UL,
928 dentry->d_name.name,
Waiman Long98474232013-08-28 18:24:59 -0700929 dentry->d_lockref.count,
David Howellsc636ebd2006-10-11 01:22:19 -0700930 dentry->d_sb->s_type->name,
931 dentry->d_sb->s_id);
932 BUG();
933 }
934
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100935 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700936 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100937 list_del(&dentry->d_u.d_child);
938 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900939 parent = dentry->d_parent;
Waiman Long98474232013-08-28 18:24:59 -0700940 parent->d_lockref.count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100941 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900942 }
David Howellsc636ebd2006-10-11 01:22:19 -0700943
David Howellsc636ebd2006-10-11 01:22:19 -0700944 inode = dentry->d_inode;
945 if (inode) {
946 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400947 hlist_del_init(&dentry->d_alias);
David Howellsc636ebd2006-10-11 01:22:19 -0700948 if (dentry->d_op && dentry->d_op->d_iput)
949 dentry->d_op->d_iput(dentry, inode);
950 else
951 iput(inode);
952 }
953
954 d_free(dentry);
955
956 /* finished when we fall off the top of the tree,
957 * otherwise we ascend to the parent and move to the
958 * next sibling if there is one */
959 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400960 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700961 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700962 } while (list_empty(&dentry->d_subdirs));
963
964 dentry = list_entry(dentry->d_subdirs.next,
965 struct dentry, d_u.d_child);
966 }
967}
968
969/*
970 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100971 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700972 * - the superblock is detached from all mountings and open files, so the
973 * dentry trees will not be rearranged by the VFS
974 * - s_umount is write-locked, so the memory pressure shrinker will ignore
975 * any dentries belonging to this superblock that it comes across
976 * - the filesystem itself is no longer permitted to rearrange the dentries
977 * in this superblock
978 */
979void shrink_dcache_for_umount(struct super_block *sb)
980{
981 struct dentry *dentry;
982
983 if (down_read_trylock(&sb->s_umount))
984 BUG();
985
986 dentry = sb->s_root;
987 sb->s_root = NULL;
Waiman Long98474232013-08-28 18:24:59 -0700988 dentry->d_lockref.count--;
David Howellsc636ebd2006-10-11 01:22:19 -0700989 shrink_dcache_for_umount_subtree(dentry);
990
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100991 while (!hlist_bl_empty(&sb->s_anon)) {
992 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -0700993 shrink_dcache_for_umount_subtree(dentry);
994 }
995}
996
997/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -0700998 * This tries to ascend one level of parenthood, but
999 * we can race with renaming, so we need to re-check
1000 * the parenthood after dropping the lock and check
1001 * that the sequence number still matches.
1002 */
1003static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1004{
1005 struct dentry *new = old->d_parent;
1006
1007 rcu_read_lock();
1008 spin_unlock(&old->d_lock);
1009 spin_lock(&new->d_lock);
1010
1011 /*
1012 * might go back up the wrong parent if we have had a rename
1013 * or deletion
1014 */
1015 if (new != old->d_parent ||
Miklos Szeredib161dfa62012-09-17 22:31:38 +02001016 (old->d_flags & DCACHE_DENTRY_KILLED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001017 (!locked && read_seqretry(&rename_lock, seq))) {
1018 spin_unlock(&new->d_lock);
1019 new = NULL;
1020 }
1021 rcu_read_unlock();
1022 return new;
1023}
1024
1025
1026/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 * Search for at least 1 mount point in the dentry's subdirs.
1028 * We descend to the next level whenever the d_subdirs
1029 * list is non-empty and continue searching.
1030 */
1031
1032/**
1033 * have_submounts - check for mounts over a dentry
1034 * @parent: dentry to check.
1035 *
1036 * Return true if the parent or its subdirectories contain
1037 * a mount point
1038 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039int have_submounts(struct dentry *parent)
1040{
Nick Piggin949854d2011-01-07 17:49:37 +11001041 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001043 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001044 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001045
Nick Piggin949854d2011-01-07 17:49:37 +11001046 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001047again:
1048 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (d_mountpoint(parent))
1051 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001052 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053repeat:
1054 next = this_parent->d_subdirs.next;
1055resume:
1056 while (next != &this_parent->d_subdirs) {
1057 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001058 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001060
1061 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001063 if (d_mountpoint(dentry)) {
1064 spin_unlock(&dentry->d_lock);
1065 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001069 spin_unlock(&this_parent->d_lock);
1070 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001072 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 goto repeat;
1074 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001075 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 }
1077 /*
1078 * All done at this level ... ascend and resume the search.
1079 */
1080 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001081 struct dentry *child = this_parent;
1082 this_parent = try_to_ascend(this_parent, locked, seq);
1083 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001084 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001085 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 goto resume;
1087 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001088 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001089 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001090 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001091 if (locked)
1092 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 return 0; /* No mount points found in tree */
1094positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001095 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001096 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001097 if (locked)
1098 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001100
1101rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02001102 if (locked)
1103 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001104 locked = 1;
1105 write_seqlock(&rename_lock);
1106 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001108EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
1110/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001111 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 * and move any unused dentries to the end of the unused
1113 * list for prune_dcache(). We descend to the next level
1114 * whenever the d_subdirs list is non-empty and continue
1115 * searching.
1116 *
1117 * It returns zero iff there are no unused children,
1118 * otherwise it returns the number of children moved to
1119 * the end of the unused list. This may not be the total
1120 * number of unused children, because select_parent can
1121 * drop the lock and return early due to latency
1122 * constraints.
1123 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001124static int select_parent(struct dentry *parent, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
Nick Piggin949854d2011-01-07 17:49:37 +11001126 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001128 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001130 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131
Nick Piggin949854d2011-01-07 17:49:37 +11001132 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001133again:
1134 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001135 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136repeat:
1137 next = this_parent->d_subdirs.next;
1138resume:
1139 while (next != &this_parent->d_subdirs) {
1140 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001141 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 next = tmp->next;
1143
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001144 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001145
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001146 /*
1147 * move only zero ref count dentries to the dispose list.
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001148 *
1149 * Those which are presently on the shrink list, being processed
1150 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1151 * loop in shrink_dcache_parent() might not make any progress
1152 * and loop forever.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 */
Waiman Long98474232013-08-28 18:24:59 -07001154 if (dentry->d_lockref.count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001155 dentry_lru_del(dentry);
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001156 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1157 dentry_lru_move_list(dentry, dispose);
1158 dentry->d_flags |= DCACHE_SHRINK_LIST;
1159 found++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 /*
1162 * We can return to the caller if we have found some (this
1163 * ensures forward progress). We'll be coming back to find
1164 * the rest.
1165 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001166 if (found && need_resched()) {
1167 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
1171 /*
1172 * Descend a level if the d_subdirs list is non-empty.
1173 */
1174 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001175 spin_unlock(&this_parent->d_lock);
1176 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001178 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 goto repeat;
1180 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001181
1182 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 }
1184 /*
1185 * All done at this level ... ascend and resume the search.
1186 */
1187 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001188 struct dentry *child = this_parent;
1189 this_parent = try_to_ascend(this_parent, locked, seq);
1190 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001191 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001192 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 goto resume;
1194 }
1195out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001196 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001197 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001198 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001199 if (locked)
1200 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001202
1203rename_retry:
1204 if (found)
1205 return found;
Miklos Szeredi8110e162012-09-17 22:23:30 +02001206 if (locked)
1207 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001208 locked = 1;
1209 write_seqlock(&rename_lock);
1210 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
1213/**
1214 * shrink_dcache_parent - prune dcache
1215 * @parent: parent of entries to prune
1216 *
1217 * Prune the dcache to remove unused children of the parent dentry.
1218 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219void shrink_dcache_parent(struct dentry * parent)
1220{
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001221 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 int found;
1223
Greg Thelen421348f2013-04-30 15:26:48 -07001224 while ((found = select_parent(parent, &dispose)) != 0) {
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001225 shrink_dentry_list(&dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001226 cond_resched();
1227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001229EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231/**
Al Viroa4464db2011-07-07 15:03:58 -04001232 * __d_alloc - allocate a dcache entry
1233 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 * @name: qstr of the name
1235 *
1236 * Allocates a dentry. It returns %NULL if there is insufficient memory
1237 * available. On a success the dentry is returned. The name passed in is
1238 * copied and the copy passed in may be reused after this call.
1239 */
1240
Al Viroa4464db2011-07-07 15:03:58 -04001241struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
1243 struct dentry *dentry;
1244 char *dname;
1245
Mel Gormane12ba742007-10-16 01:25:52 -07001246 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 if (!dentry)
1248 return NULL;
1249
Linus Torvalds6326c712012-05-21 16:14:04 -07001250 /*
1251 * We guarantee that the inline name is always NUL-terminated.
1252 * This way the memcpy() done by the name switching in rename
1253 * will still always have a NUL at the end, even if we might
1254 * be overwriting an internal NUL character
1255 */
1256 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 if (name->len > DNAME_INLINE_LEN-1) {
1258 dname = kmalloc(name->len + 1, GFP_KERNEL);
1259 if (!dname) {
1260 kmem_cache_free(dentry_cache, dentry);
1261 return NULL;
1262 }
1263 } else {
1264 dname = dentry->d_iname;
1265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
1267 dentry->d_name.len = name->len;
1268 dentry->d_name.hash = name->hash;
1269 memcpy(dname, name->name, name->len);
1270 dname[name->len] = 0;
1271
Linus Torvalds6326c712012-05-21 16:14:04 -07001272 /* Make sure we always see the terminating NUL character */
1273 smp_wmb();
1274 dentry->d_name.name = dname;
1275
Waiman Long98474232013-08-28 18:24:59 -07001276 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001277 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001279 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001281 dentry->d_parent = dentry;
1282 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 dentry->d_op = NULL;
1284 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001285 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 INIT_LIST_HEAD(&dentry->d_lru);
1287 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001288 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001289 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001290 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
Nick Piggin3e880fb2011-01-07 17:49:19 +11001292 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 return dentry;
1295}
Al Viroa4464db2011-07-07 15:03:58 -04001296
1297/**
1298 * d_alloc - allocate a dcache entry
1299 * @parent: parent of entry to allocate
1300 * @name: qstr of the name
1301 *
1302 * Allocates a dentry. It returns %NULL if there is insufficient memory
1303 * available. On a success the dentry is returned. The name passed in is
1304 * copied and the copy passed in may be reused after this call.
1305 */
1306struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1307{
1308 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1309 if (!dentry)
1310 return NULL;
1311
1312 spin_lock(&parent->d_lock);
1313 /*
1314 * don't need child lock because it is not subject
1315 * to concurrency here
1316 */
1317 __dget_dlock(parent);
1318 dentry->d_parent = parent;
1319 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1320 spin_unlock(&parent->d_lock);
1321
1322 return dentry;
1323}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001324EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Nick Piggin4b936882011-01-07 17:50:07 +11001326struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1327{
Al Viroa4464db2011-07-07 15:03:58 -04001328 struct dentry *dentry = __d_alloc(sb, name);
1329 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001330 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001331 return dentry;
1332}
1333EXPORT_SYMBOL(d_alloc_pseudo);
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1336{
1337 struct qstr q;
1338
1339 q.name = name;
1340 q.len = strlen(name);
1341 q.hash = full_name_hash(q.name, q.len);
1342 return d_alloc(parent, &q);
1343}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001344EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Nick Pigginfb045ad2011-01-07 17:49:55 +11001346void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1347{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001348 WARN_ON_ONCE(dentry->d_op);
1349 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001350 DCACHE_OP_COMPARE |
1351 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001352 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001353 DCACHE_OP_DELETE ));
1354 dentry->d_op = op;
1355 if (!op)
1356 return;
1357 if (op->d_hash)
1358 dentry->d_flags |= DCACHE_OP_HASH;
1359 if (op->d_compare)
1360 dentry->d_flags |= DCACHE_OP_COMPARE;
1361 if (op->d_revalidate)
1362 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001363 if (op->d_weak_revalidate)
1364 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001365 if (op->d_delete)
1366 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001367 if (op->d_prune)
1368 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001369
1370}
1371EXPORT_SYMBOL(d_set_d_op);
1372
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001373static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1374{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001375 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001376 if (inode) {
1377 if (unlikely(IS_AUTOMOUNT(inode)))
1378 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
Al Virob3d9b7a2012-06-09 13:51:19 -04001379 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001380 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001381 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001382 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001383 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001384 fsnotify_d_instantiate(dentry, inode);
1385}
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387/**
1388 * d_instantiate - fill in inode information for a dentry
1389 * @entry: dentry to complete
1390 * @inode: inode to attach to this dentry
1391 *
1392 * Fill in inode information in the entry.
1393 *
1394 * This turns negative dentries into productive full members
1395 * of society.
1396 *
1397 * NOTE! This assumes that the inode count has been incremented
1398 * (or otherwise set) by the caller to indicate that it is now
1399 * in use by the dcache.
1400 */
1401
1402void d_instantiate(struct dentry *entry, struct inode * inode)
1403{
Al Virob3d9b7a2012-06-09 13:51:19 -04001404 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001405 if (inode)
1406 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001407 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001408 if (inode)
1409 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 security_d_instantiate(entry, inode);
1411}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001412EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414/**
1415 * d_instantiate_unique - instantiate a non-aliased dentry
1416 * @entry: dentry to instantiate
1417 * @inode: inode to attach to this dentry
1418 *
1419 * Fill in inode information in the entry. On success, it returns NULL.
1420 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001421 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 *
1423 * Note that in order to avoid conflicts with rename() etc, the caller
1424 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001425 *
1426 * This also assumes that the inode count has been incremented
1427 * (or otherwise set) by the caller to indicate that it is now
1428 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 */
David Howells770bfad2006-08-22 20:06:07 -04001430static struct dentry *__d_instantiate_unique(struct dentry *entry,
1431 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432{
1433 struct dentry *alias;
1434 int len = entry->d_name.len;
1435 const char *name = entry->d_name.name;
1436 unsigned int hash = entry->d_name.hash;
1437
David Howells770bfad2006-08-22 20:06:07 -04001438 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001439 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001440 return NULL;
1441 }
1442
Sasha Levinb67bfe02013-02-27 17:06:00 -08001443 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001444 /*
1445 * Don't need alias->d_lock here, because aliases with
1446 * d_parent == entry->d_parent are not subject to name or
1447 * parent changes, because the parent inode i_mutex is held.
1448 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001449 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 continue;
1451 if (alias->d_parent != entry->d_parent)
1452 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001453 if (alias->d_name.len != len)
1454 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001455 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001457 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 return alias;
1459 }
David Howells770bfad2006-08-22 20:06:07 -04001460
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001461 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return NULL;
1463}
David Howells770bfad2006-08-22 20:06:07 -04001464
1465struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1466{
1467 struct dentry *result;
1468
Al Virob3d9b7a2012-06-09 13:51:19 -04001469 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001470
Nick Piggin873feea2011-01-07 17:50:06 +11001471 if (inode)
1472 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001473 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001474 if (inode)
1475 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001476
1477 if (!result) {
1478 security_d_instantiate(entry, inode);
1479 return NULL;
1480 }
1481
1482 BUG_ON(!d_unhashed(result));
1483 iput(inode);
1484 return result;
1485}
1486
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487EXPORT_SYMBOL(d_instantiate_unique);
1488
Al Viroadc0e912012-01-08 16:49:21 -05001489struct dentry *d_make_root(struct inode *root_inode)
1490{
1491 struct dentry *res = NULL;
1492
1493 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001494 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001495
1496 res = __d_alloc(root_inode->i_sb, &name);
1497 if (res)
1498 d_instantiate(res, root_inode);
1499 else
1500 iput(root_inode);
1501 }
1502 return res;
1503}
1504EXPORT_SYMBOL(d_make_root);
1505
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001506static struct dentry * __d_find_any_alias(struct inode *inode)
1507{
1508 struct dentry *alias;
1509
Al Virob3d9b7a2012-06-09 13:51:19 -04001510 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001511 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001512 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001513 __dget(alias);
1514 return alias;
1515}
1516
Sage Weil46f72b32012-01-10 09:04:37 -08001517/**
1518 * d_find_any_alias - find any alias for a given inode
1519 * @inode: inode to find an alias for
1520 *
1521 * If any aliases exist for the given inode, take and return a
1522 * reference for one of them. If no aliases exist, return %NULL.
1523 */
1524struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001525{
1526 struct dentry *de;
1527
1528 spin_lock(&inode->i_lock);
1529 de = __d_find_any_alias(inode);
1530 spin_unlock(&inode->i_lock);
1531 return de;
1532}
Sage Weil46f72b32012-01-10 09:04:37 -08001533EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001536 * d_obtain_alias - find or allocate a dentry for a given inode
1537 * @inode: inode to allocate the dentry for
1538 *
1539 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1540 * similar open by handle operations. The returned dentry may be anonymous,
1541 * or may have a full name (if the inode was already in the cache).
1542 *
1543 * When called on a directory inode, we must ensure that the inode only ever
1544 * has one dentry. If a dentry is found, that is returned instead of
1545 * allocating a new one.
1546 *
1547 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001548 * to the dentry. In case of an error the reference on the inode is released.
1549 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1550 * be passed in and will be the error will be propagate to the return value,
1551 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001552 */
1553struct dentry *d_obtain_alias(struct inode *inode)
1554{
NeilBrownb911a6b2012-11-08 16:09:37 -08001555 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001556 struct dentry *tmp;
1557 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001558
1559 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001560 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001561 if (IS_ERR(inode))
1562 return ERR_CAST(inode);
1563
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001564 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001565 if (res)
1566 goto out_iput;
1567
Al Viroa4464db2011-07-07 15:03:58 -04001568 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001569 if (!tmp) {
1570 res = ERR_PTR(-ENOMEM);
1571 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001572 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001573
Nick Piggin873feea2011-01-07 17:50:06 +11001574 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001575 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001576 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001577 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001578 dput(tmp);
1579 goto out_iput;
1580 }
1581
1582 /* attach a disconnected dentry */
1583 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001584 tmp->d_inode = inode;
1585 tmp->d_flags |= DCACHE_DISCONNECTED;
Al Virob3d9b7a2012-06-09 13:51:19 -04001586 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001587 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001588 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001589 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001590 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001591 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001592 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001593
Christoph Hellwig9308a612008-08-11 15:49:12 +02001594 return tmp;
1595
1596 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001597 if (res && !IS_ERR(res))
1598 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001599 iput(inode);
1600 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001601}
Benny Halevyadc48722009-02-27 14:02:59 -08001602EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
1604/**
1605 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1606 * @inode: the inode which may have a disconnected dentry
1607 * @dentry: a negative dentry which we want to point to the inode.
1608 *
1609 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1610 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1611 * and return it, else simply d_add the inode to the dentry and return NULL.
1612 *
1613 * This is needed in the lookup routine of any filesystem that is exportable
1614 * (via knfsd) so that we can build dcache paths to directories effectively.
1615 *
1616 * If a dentry was found and moved, then it is returned. Otherwise NULL
1617 * is returned. This matches the expected return value of ->lookup.
1618 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001619 * Cluster filesystems may call this function with a negative, hashed dentry.
1620 * In that case, we know that the inode will be a regular file, and also this
1621 * will only occur during atomic_open. So we need to check for the dentry
1622 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 */
1624struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1625{
1626 struct dentry *new = NULL;
1627
Al Viroa9049372011-07-08 21:20:11 -04001628 if (IS_ERR(inode))
1629 return ERR_CAST(inode);
1630
NeilBrown21c0d8f2006-10-04 02:16:16 -07001631 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001632 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001633 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001635 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001636 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 d_move(new, dentry);
1639 iput(inode);
1640 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001641 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001642 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001643 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 security_d_instantiate(dentry, inode);
1645 d_rehash(dentry);
1646 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001647 } else {
1648 d_instantiate(dentry, inode);
1649 if (d_unhashed(dentry))
1650 d_rehash(dentry);
1651 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 return new;
1653}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001654EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Barry Naujok94035402008-05-21 16:50:46 +10001656/**
1657 * d_add_ci - lookup or allocate new dentry with case-exact name
1658 * @inode: the inode case-insensitive lookup has found
1659 * @dentry: the negative dentry that was passed to the parent's lookup func
1660 * @name: the case-exact name to be associated with the returned dentry
1661 *
1662 * This is to avoid filling the dcache with case-insensitive names to the
1663 * same inode, only the actual correct case is stored in the dcache for
1664 * case-insensitive filesystems.
1665 *
1666 * For a case-insensitive lookup match and if the the case-exact dentry
1667 * already exists in in the dcache, use it and return it.
1668 *
1669 * If no entry exists with the exact case name, allocate new dentry with
1670 * the exact case, and return the spliced entry.
1671 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001672struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001673 struct qstr *name)
1674{
Barry Naujok94035402008-05-21 16:50:46 +10001675 struct dentry *found;
1676 struct dentry *new;
1677
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001678 /*
1679 * First check if a dentry matching the name already exists,
1680 * if not go ahead and create it now.
1681 */
Barry Naujok94035402008-05-21 16:50:46 +10001682 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001683 if (unlikely(IS_ERR(found)))
1684 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10001685 if (!found) {
1686 new = d_alloc(dentry->d_parent, name);
1687 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05001688 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10001689 goto err_out;
1690 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001691
Barry Naujok94035402008-05-21 16:50:46 +10001692 found = d_splice_alias(inode, new);
1693 if (found) {
1694 dput(new);
1695 return found;
1696 }
1697 return new;
1698 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001699
1700 /*
1701 * If a matching dentry exists, and it's not negative use it.
1702 *
1703 * Decrement the reference count to balance the iget() done
1704 * earlier on.
1705 */
Barry Naujok94035402008-05-21 16:50:46 +10001706 if (found->d_inode) {
1707 if (unlikely(found->d_inode != inode)) {
1708 /* This can't happen because bad inodes are unhashed. */
1709 BUG_ON(!is_bad_inode(inode));
1710 BUG_ON(!is_bad_inode(found->d_inode));
1711 }
Barry Naujok94035402008-05-21 16:50:46 +10001712 iput(inode);
1713 return found;
1714 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001715
Barry Naujok94035402008-05-21 16:50:46 +10001716 /*
1717 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001718 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10001719 */
Al Viro4513d892011-07-17 10:52:14 -04001720 new = d_splice_alias(inode, found);
1721 if (new) {
1722 dput(found);
1723 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10001724 }
Al Viro4513d892011-07-17 10:52:14 -04001725 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001726
1727err_out:
1728 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05001729 return found;
Barry Naujok94035402008-05-21 16:50:46 +10001730}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001731EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001733/*
1734 * Do the slow-case of the dentry name compare.
1735 *
1736 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07001737 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001738 * filesystem can rely on them, and can use the 'name' and
1739 * 'len' information without worrying about walking off the
1740 * end of memory etc.
1741 *
1742 * Thus the read_seqcount_retry() and the "duplicate" info
1743 * in arguments (the low-level filesystem should not look
1744 * at the dentry inode or name contents directly, since
1745 * rename can change them while we're in RCU mode).
1746 */
1747enum slow_d_compare {
1748 D_COMP_OK,
1749 D_COMP_NOMATCH,
1750 D_COMP_SEQRETRY,
1751};
1752
1753static noinline enum slow_d_compare slow_dentry_cmp(
1754 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001755 struct dentry *dentry,
1756 unsigned int seq,
1757 const struct qstr *name)
1758{
1759 int tlen = dentry->d_name.len;
1760 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001761
1762 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1763 cpu_relax();
1764 return D_COMP_SEQRETRY;
1765 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07001766 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001767 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 * Returns: dentry, or NULL
1777 *
1778 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1779 * resolution (store-free path walking) design described in
1780 * Documentation/filesystems/path-lookup.txt.
1781 *
1782 * This is not to be used outside core vfs.
1783 *
1784 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1785 * held, and rcu_read_lock held. The returned dentry must not be stored into
1786 * without taking d_lock and checking d_seq sequence count against @seq
1787 * returned here.
1788 *
1789 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1790 * function.
1791 *
1792 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1793 * the returned dentry, so long as its parent's seqlock is checked after the
1794 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1795 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001796 *
1797 * NOTE! The caller *has* to check the resulting dentry against the sequence
1798 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001799 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001800struct dentry *__d_lookup_rcu(const struct dentry *parent,
1801 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07001802 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11001803{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001804 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001805 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001806 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001807 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001808 struct dentry *dentry;
1809
1810 /*
1811 * Note: There is significant duplication with __d_lookup_rcu which is
1812 * required to prevent single threaded performance regressions
1813 * especially on architectures where smp_rmb (in seqcounts) are costly.
1814 * Keep the two functions in sync.
1815 */
1816
1817 /*
1818 * The hash list is protected using RCU.
1819 *
1820 * Carefully use d_seq when comparing a candidate dentry, to avoid
1821 * races with d_move().
1822 *
1823 * It is possible that concurrent renames can mess up our list
1824 * walk here and result in missing our dentry, resulting in the
1825 * false-negative result. d_lookup() protects against concurrent
1826 * renames using rename_lock seqlock.
1827 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001828 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11001829 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001830 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08001831 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001832
Nick Piggin31e6b012011-01-07 17:49:52 +11001833seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001834 /*
1835 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07001836 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001837 *
1838 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07001839 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001840 *
1841 * NOTE! We do a "raw" seqcount_begin here. That means that
1842 * we don't wait for the sequence count to stabilize if it
1843 * is in the middle of a sequence change. If we do the slow
1844 * dentry compare, we will do seqretries until it is stable,
1845 * and if we end up with a successful lookup, we actually
1846 * want to exit RCU lookup anyway.
1847 */
1848 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001849 if (dentry->d_parent != parent)
1850 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07001851 if (d_unhashed(dentry))
1852 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001853
1854 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001855 if (dentry->d_name.hash != hashlen_hash(hashlen))
1856 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001857 *seqp = seq;
1858 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001859 case D_COMP_OK:
1860 return dentry;
1861 case D_COMP_NOMATCH:
1862 continue;
1863 default:
1864 goto seqretry;
1865 }
1866 }
1867
Linus Torvalds26fe5752012-05-10 13:14:12 -07001868 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07001869 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001870 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001871 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001872 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11001873 }
1874 return NULL;
1875}
1876
1877/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 * d_lookup - search for a dentry
1879 * @parent: parent dentry
1880 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001881 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001883 * d_lookup searches the children of the parent dentry for the name in
1884 * question. If the dentry is found its reference count is incremented and the
1885 * dentry is returned. The caller must use dput to free the entry when it has
1886 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 */
Al Viroda2d8452013-01-24 18:29:34 -05001888struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889{
Nick Piggin31e6b012011-01-07 17:49:52 +11001890 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001891 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
1893 do {
1894 seq = read_seqbegin(&rename_lock);
1895 dentry = __d_lookup(parent, name);
1896 if (dentry)
1897 break;
1898 } while (read_seqretry(&rename_lock, seq));
1899 return dentry;
1900}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001901EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902
Nick Piggin31e6b012011-01-07 17:49:52 +11001903/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001904 * __d_lookup - search for a dentry (racy)
1905 * @parent: parent dentry
1906 * @name: qstr of name we wish to find
1907 * Returns: dentry, or NULL
1908 *
1909 * __d_lookup is like d_lookup, however it may (rarely) return a
1910 * false-negative result due to unrelated rename activity.
1911 *
1912 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1913 * however it must be used carefully, eg. with a following d_lookup in
1914 * the case of failure.
1915 *
1916 * __d_lookup callers must be commented.
1917 */
Al Viroa713ca22013-01-24 18:27:00 -05001918struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919{
1920 unsigned int len = name->len;
1921 unsigned int hash = name->hash;
1922 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001923 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001924 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001925 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001926 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Nick Pigginb04f7842010-08-18 04:37:34 +10001928 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001929 * Note: There is significant duplication with __d_lookup_rcu which is
1930 * required to prevent single threaded performance regressions
1931 * especially on architectures where smp_rmb (in seqcounts) are costly.
1932 * Keep the two functions in sync.
1933 */
1934
1935 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001936 * The hash list is protected using RCU.
1937 *
1938 * Take d_lock when comparing a candidate dentry, to avoid races
1939 * with d_move().
1940 *
1941 * It is possible that concurrent renames can mess up our list
1942 * walk here and result in missing our dentry, resulting in the
1943 * false-negative result. d_lookup() protects against concurrent
1944 * renames using rename_lock seqlock.
1945 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001946 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001947 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 rcu_read_lock();
1949
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001950 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 if (dentry->d_name.hash != hash)
1953 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 if (dentry->d_parent != parent)
1957 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001958 if (d_unhashed(dentry))
1959 goto next;
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 /*
1962 * It is safe to compare names since d_move() cannot
1963 * change the qstr (protected by d_lock).
1964 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11001965 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001966 int tlen = dentry->d_name.len;
1967 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07001968 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 goto next;
1970 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07001971 if (dentry->d_name.len != len)
1972 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001973 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 goto next;
1975 }
1976
Waiman Long98474232013-08-28 18:24:59 -07001977 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001978 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 spin_unlock(&dentry->d_lock);
1980 break;
1981next:
1982 spin_unlock(&dentry->d_lock);
1983 }
1984 rcu_read_unlock();
1985
1986 return found;
1987}
1988
1989/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001990 * d_hash_and_lookup - hash the qstr then search for a dentry
1991 * @dir: Directory to search in
1992 * @name: qstr of name we wish to find
1993 *
Al Viro4f522a22013-02-11 23:20:37 -05001994 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001995 */
1996struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1997{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001998 /*
1999 * Check for a fs-specific hash function. Note that we must
2000 * calculate the standard hash first, as the d_op->d_hash()
2001 * routine may choose to leave the hash value unchanged.
2002 */
2003 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002004 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002005 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002006 if (unlikely(err < 0))
2007 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002008 }
Al Viro4f522a22013-02-11 23:20:37 -05002009 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002010}
Al Viro4f522a22013-02-11 23:20:37 -05002011EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002012
2013/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002014 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002016 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 *
2018 * An insecure source has sent us a dentry, here we verify it and dget() it.
2019 * This is used by ncpfs in its readdir implementation.
2020 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002021 *
2022 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002024int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025{
Nick Piggin786a5e12011-01-07 17:49:16 +11002026 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002027
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002028 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002029 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2030 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002031 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002032 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002033 spin_unlock(&dentry->d_lock);
2034 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 return 1;
2036 }
2037 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002038 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 return 0;
2041}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002042EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
2044/*
2045 * When a file is deleted, we have two options:
2046 * - turn this dentry into a negative dentry
2047 * - unhash this dentry and free it.
2048 *
2049 * Usually, we want to just turn this into
2050 * a negative dentry, but if anybody else is
2051 * currently using the dentry or the inode
2052 * we can't do that and we fall back on removing
2053 * it from the hash queues and waiting for
2054 * it to be deleted later when it has no users
2055 */
2056
2057/**
2058 * d_delete - delete a dentry
2059 * @dentry: The dentry to delete
2060 *
2061 * Turn the dentry into a negative dentry if possible, otherwise
2062 * remove it from the hash queues so it can be deleted later
2063 */
2064
2065void d_delete(struct dentry * dentry)
2066{
Nick Piggin873feea2011-01-07 17:50:06 +11002067 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002068 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 /*
2070 * Are we the only user?
2071 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002072again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002074 inode = dentry->d_inode;
2075 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002076 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002077 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002078 spin_unlock(&dentry->d_lock);
2079 cpu_relax();
2080 goto again;
2081 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002082 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002083 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002084 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return;
2086 }
2087
2088 if (!d_unhashed(dentry))
2089 __d_drop(dentry);
2090
2091 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002092
2093 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002095EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002097static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002099 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002100 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002101 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002102 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002103 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104}
2105
David Howells770bfad2006-08-22 20:06:07 -04002106static void _d_rehash(struct dentry * entry)
2107{
2108 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2109}
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111/**
2112 * d_rehash - add an entry back to the hash
2113 * @entry: dentry to add to the hash
2114 *
2115 * Adds a dentry to the hash according to its name.
2116 */
2117
2118void d_rehash(struct dentry * entry)
2119{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002121 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002124EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002126/**
2127 * dentry_update_name_case - update case insensitive dentry with a new name
2128 * @dentry: dentry to be updated
2129 * @name: new name
2130 *
2131 * Update a case insensitive dentry with new case of name.
2132 *
2133 * dentry must have been returned by d_lookup with name @name. Old and new
2134 * name lengths must match (ie. no d_compare which allows mismatched name
2135 * lengths).
2136 *
2137 * Parent inode i_mutex must be held over d_lookup and into this call (to
2138 * keep renames and concurrent inserts, and readdir(2) away).
2139 */
2140void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2141{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002142 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002143 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2144
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002145 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002146 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002147 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002148 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002149 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002150}
2151EXPORT_SYMBOL(dentry_update_name_case);
2152
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153static void switch_names(struct dentry *dentry, struct dentry *target)
2154{
2155 if (dname_external(target)) {
2156 if (dname_external(dentry)) {
2157 /*
2158 * Both external: swap the pointers
2159 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002160 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 } else {
2162 /*
2163 * dentry:internal, target:external. Steal target's
2164 * storage and make target internal.
2165 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002166 memcpy(target->d_iname, dentry->d_name.name,
2167 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 dentry->d_name.name = target->d_name.name;
2169 target->d_name.name = target->d_iname;
2170 }
2171 } else {
2172 if (dname_external(dentry)) {
2173 /*
2174 * dentry:external, target:internal. Give dentry's
2175 * storage to target and make dentry internal
2176 */
2177 memcpy(dentry->d_iname, target->d_name.name,
2178 target->d_name.len + 1);
2179 target->d_name.name = dentry->d_name.name;
2180 dentry->d_name.name = dentry->d_iname;
2181 } else {
2182 /*
2183 * Both are internal. Just copy target to dentry
2184 */
2185 memcpy(dentry->d_iname, target->d_name.name,
2186 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002187 dentry->d_name.len = target->d_name.len;
2188 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 }
2190 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002191 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192}
2193
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002194static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2195{
2196 /*
2197 * XXXX: do we really need to take target->d_lock?
2198 */
2199 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2200 spin_lock(&target->d_parent->d_lock);
2201 else {
2202 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2203 spin_lock(&dentry->d_parent->d_lock);
2204 spin_lock_nested(&target->d_parent->d_lock,
2205 DENTRY_D_LOCK_NESTED);
2206 } else {
2207 spin_lock(&target->d_parent->d_lock);
2208 spin_lock_nested(&dentry->d_parent->d_lock,
2209 DENTRY_D_LOCK_NESTED);
2210 }
2211 }
2212 if (target < dentry) {
2213 spin_lock_nested(&target->d_lock, 2);
2214 spin_lock_nested(&dentry->d_lock, 3);
2215 } else {
2216 spin_lock_nested(&dentry->d_lock, 2);
2217 spin_lock_nested(&target->d_lock, 3);
2218 }
2219}
2220
2221static void dentry_unlock_parents_for_move(struct dentry *dentry,
2222 struct dentry *target)
2223{
2224 if (target->d_parent != dentry->d_parent)
2225 spin_unlock(&dentry->d_parent->d_lock);
2226 if (target->d_parent != target)
2227 spin_unlock(&target->d_parent->d_lock);
2228}
2229
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002231 * When switching names, the actual string doesn't strictly have to
2232 * be preserved in the target - because we're dropping the target
2233 * anyway. As such, we can just do a simple memcpy() to copy over
2234 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002236 * Note that we have to be a lot more careful about getting the hash
2237 * switched - we have to switch the hash value properly even if it
2238 * then no longer matches the actual (corrupted) string of the target.
2239 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002241/*
Al Viro18367502011-07-12 21:42:24 -04002242 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 * @dentry: entry to move
2244 * @target: new dentry
2245 *
2246 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002247 * dcache entries should not be moved in this way. Caller must hold
2248 * rename_lock, the i_mutex of the source and target directories,
2249 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 */
Al Viro18367502011-07-12 21:42:24 -04002251static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (!dentry->d_inode)
2254 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2255
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002256 BUG_ON(d_ancestor(dentry, target));
2257 BUG_ON(d_ancestor(target, dentry));
2258
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002259 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
Nick Piggin31e6b012011-01-07 17:49:52 +11002261 write_seqcount_begin(&dentry->d_seq);
2262 write_seqcount_begin(&target->d_seq);
2263
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002264 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2265
2266 /*
2267 * Move the dentry to the target hash queue. Don't bother checking
2268 * for the same hash queue because of how unlikely it is.
2269 */
2270 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002271 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
2273 /* Unhash the target: dput() will then get rid of it */
2274 __d_drop(target);
2275
Eric Dumazet5160ee62006-01-08 01:03:32 -08002276 list_del(&dentry->d_u.d_child);
2277 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278
2279 /* Switch the names.. */
2280 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002281 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282
2283 /* ... and switch the parents */
2284 if (IS_ROOT(dentry)) {
2285 dentry->d_parent = target->d_parent;
2286 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002287 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002289 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
2291 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002292 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 }
2294
Eric Dumazet5160ee62006-01-08 01:03:32 -08002295 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002296
Nick Piggin31e6b012011-01-07 17:49:52 +11002297 write_seqcount_end(&target->d_seq);
2298 write_seqcount_end(&dentry->d_seq);
2299
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002300 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002302 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002304}
2305
2306/*
2307 * d_move - move a dentry
2308 * @dentry: entry to move
2309 * @target: new dentry
2310 *
2311 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002312 * dcache entries should not be moved in this way. See the locking
2313 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002314 */
2315void d_move(struct dentry *dentry, struct dentry *target)
2316{
2317 write_seqlock(&rename_lock);
2318 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002320}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002321EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002323/**
2324 * d_ancestor - search for an ancestor
2325 * @p1: ancestor dentry
2326 * @p2: child dentry
2327 *
2328 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2329 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002330 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002331struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002332{
2333 struct dentry *p;
2334
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002335 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002336 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002337 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002338 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002339 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002340}
2341
2342/*
2343 * This helper attempts to cope with remotely renamed directories
2344 *
2345 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002346 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002347 *
2348 * Note: If ever the locking in lock_rename() changes, then please
2349 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002350 */
Nick Piggin873feea2011-01-07 17:50:06 +11002351static struct dentry *__d_unalias(struct inode *inode,
2352 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002353{
2354 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002355 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002356
2357 /* If alias and dentry share a parent, then no extra locks required */
2358 if (alias->d_parent == dentry->d_parent)
2359 goto out_unalias;
2360
Trond Myklebust9eaef272006-10-21 10:24:20 -07002361 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002362 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2363 goto out_err;
2364 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2365 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2366 goto out_err;
2367 m2 = &alias->d_parent->d_inode->i_mutex;
2368out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002369 if (likely(!d_mountpoint(alias))) {
2370 __d_move(alias, dentry);
2371 ret = alias;
2372 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002373out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002374 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002375 if (m2)
2376 mutex_unlock(m2);
2377 if (m1)
2378 mutex_unlock(m1);
2379 return ret;
2380}
2381
2382/*
David Howells770bfad2006-08-22 20:06:07 -04002383 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2384 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002385 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002386 */
2387static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2388{
Al Viro740da422013-01-30 10:13:38 -05002389 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002390
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002391 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002392
Nick Piggin31e6b012011-01-07 17:49:52 +11002393 write_seqcount_begin(&dentry->d_seq);
2394 write_seqcount_begin(&anon->d_seq);
2395
David Howells770bfad2006-08-22 20:06:07 -04002396 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002397
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002398 switch_names(dentry, anon);
2399 swap(dentry->d_name.hash, anon->d_name.hash);
2400
Al Viro740da422013-01-30 10:13:38 -05002401 dentry->d_parent = dentry;
2402 list_del_init(&dentry->d_u.d_child);
2403 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002404 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002405
Nick Piggin31e6b012011-01-07 17:49:52 +11002406 write_seqcount_end(&dentry->d_seq);
2407 write_seqcount_end(&anon->d_seq);
2408
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002409 dentry_unlock_parents_for_move(anon, dentry);
2410 spin_unlock(&dentry->d_lock);
2411
2412 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002413 anon->d_flags &= ~DCACHE_DISCONNECTED;
2414}
2415
2416/**
2417 * d_materialise_unique - introduce an inode into the tree
2418 * @dentry: candidate dentry
2419 * @inode: inode to bind to the dentry, to which aliases may be attached
2420 *
2421 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002422 * root directory alias in its place if there is one. Caller must hold the
2423 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002424 */
2425struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2426{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002427 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002428
2429 BUG_ON(!d_unhashed(dentry));
2430
David Howells770bfad2006-08-22 20:06:07 -04002431 if (!inode) {
2432 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002433 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002434 d_rehash(actual);
2435 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002436 }
2437
Nick Piggin873feea2011-01-07 17:50:06 +11002438 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002439
Trond Myklebust9eaef272006-10-21 10:24:20 -07002440 if (S_ISDIR(inode->i_mode)) {
2441 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002442
Trond Myklebust9eaef272006-10-21 10:24:20 -07002443 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002444 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002445 if (alias) {
2446 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002447 write_seqlock(&rename_lock);
2448
2449 if (d_ancestor(alias, dentry)) {
2450 /* Check for loops */
2451 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002452 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002453 } else if (IS_ROOT(alias)) {
2454 /* Is this an anonymous mountpoint that we
2455 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002456 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002457 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002458 __d_drop(alias);
2459 goto found;
Al Viro18367502011-07-12 21:42:24 -04002460 } else {
2461 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002462 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002463 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002464 }
Al Viro18367502011-07-12 21:42:24 -04002465 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002466 if (IS_ERR(actual)) {
2467 if (PTR_ERR(actual) == -ELOOP)
2468 pr_warn_ratelimited(
2469 "VFS: Lookup of '%s' in %s %s"
2470 " would have caused loop\n",
2471 dentry->d_name.name,
2472 inode->i_sb->s_type->name,
2473 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002474 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002475 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002476 goto out_nolock;
2477 }
David Howells770bfad2006-08-22 20:06:07 -04002478 }
2479
2480 /* Add a unique reference */
2481 actual = __d_instantiate_unique(dentry, inode);
2482 if (!actual)
2483 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002484 else
2485 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002486
David Howells770bfad2006-08-22 20:06:07 -04002487 spin_lock(&actual->d_lock);
2488found:
2489 _d_rehash(actual);
2490 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002491 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002492out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002493 if (actual == dentry) {
2494 security_d_instantiate(dentry, inode);
2495 return NULL;
2496 }
2497
2498 iput(inode);
2499 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002500}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002501EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002502
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002503static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002504{
2505 *buflen -= namelen;
2506 if (*buflen < 0)
2507 return -ENAMETOOLONG;
2508 *buffer -= namelen;
2509 memcpy(*buffer, str, namelen);
2510 return 0;
2511}
2512
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002513static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2514{
2515 return prepend(buffer, buflen, name->name, name->len);
2516}
2517
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002519 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002520 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002521 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002522 * @buffer: pointer to the end of the buffer
2523 * @buflen: pointer to buffer length
2524 *
Nick Piggin949854d2011-01-07 17:49:37 +11002525 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002526 */
Al Viro02125a82011-12-05 08:43:34 -05002527static int prepend_path(const struct path *path,
2528 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002529 char **buffer, int *buflen)
2530{
2531 struct dentry *dentry = path->dentry;
2532 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002533 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002534 bool slash = false;
2535 int error = 0;
2536
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002537 while (dentry != root->dentry || vfsmnt != root->mnt) {
2538 struct dentry * parent;
2539
2540 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2541 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002542 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002543 goto global_root;
Al Viroa73324d2011-11-24 22:25:07 -05002544 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002545 mnt = mnt->mnt_parent;
2546 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002547 continue;
2548 }
2549 parent = dentry->d_parent;
2550 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002551 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002552 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002553 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002554 if (!error)
2555 error = prepend(buffer, buflen, "/", 1);
2556 if (error)
2557 break;
2558
2559 slash = true;
2560 dentry = parent;
2561 }
2562
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002563 if (!error && !slash)
2564 error = prepend(buffer, buflen, "/", 1);
2565
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002566 return error;
2567
2568global_root:
2569 /*
2570 * Filesystems needing to implement special "root names"
2571 * should do so with ->d_dname()
2572 */
2573 if (IS_ROOT(dentry) &&
2574 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2575 WARN(1, "Root dentry has weird name <%.*s>\n",
2576 (int) dentry->d_name.len, dentry->d_name.name);
2577 }
Al Viro02125a82011-12-05 08:43:34 -05002578 if (!slash)
2579 error = prepend(buffer, buflen, "/", 1);
2580 if (!error)
Al Virof7a99c52012-06-09 00:59:08 -04002581 error = is_mounted(vfsmnt) ? 1 : 2;
Al Viro7ea600b2013-03-26 18:25:57 -04002582 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002583}
2584
2585/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002586 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002587 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002588 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002589 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 * @buflen: buffer length
2591 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002592 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002594 * Returns a pointer into the buffer or an error code if the
2595 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002596 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002597 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002598 *
Al Viro02125a82011-12-05 08:43:34 -05002599 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 */
Al Viro02125a82011-12-05 08:43:34 -05002601char *__d_path(const struct path *path,
2602 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002603 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002605 char *res = buf + buflen;
2606 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002608 prepend(&res, &buflen, "\0", 1);
Al Viro7ea600b2013-03-26 18:25:57 -04002609 br_read_lock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002610 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002611 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002612 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002613 br_read_unlock(&vfsmount_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002614
Al Viro02125a82011-12-05 08:43:34 -05002615 if (error < 0)
2616 return ERR_PTR(error);
2617 if (error > 0)
2618 return NULL;
2619 return res;
2620}
2621
2622char *d_absolute_path(const struct path *path,
2623 char *buf, int buflen)
2624{
2625 struct path root = {};
2626 char *res = buf + buflen;
2627 int error;
2628
2629 prepend(&res, &buflen, "\0", 1);
Al Viro7ea600b2013-03-26 18:25:57 -04002630 br_read_lock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002631 write_seqlock(&rename_lock);
2632 error = prepend_path(path, &root, &res, &buflen);
2633 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002634 br_read_unlock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002635
2636 if (error > 1)
2637 error = -EINVAL;
2638 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002639 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002640 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641}
2642
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002643/*
2644 * same as __d_path but appends "(deleted)" for unlinked files.
2645 */
Al Viro02125a82011-12-05 08:43:34 -05002646static int path_with_deleted(const struct path *path,
2647 const struct path *root,
2648 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002649{
2650 prepend(buf, buflen, "\0", 1);
2651 if (d_unlinked(path->dentry)) {
2652 int error = prepend(buf, buflen, " (deleted)", 10);
2653 if (error)
2654 return error;
2655 }
2656
2657 return prepend_path(path, root, buf, buflen);
2658}
2659
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002660static int prepend_unreachable(char **buffer, int *buflen)
2661{
2662 return prepend(buffer, buflen, "(unreachable)", 13);
2663}
2664
Jan Bluncka03a8a702008-02-14 19:38:32 -08002665/**
2666 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002667 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002668 * @buf: buffer to return value in
2669 * @buflen: buffer length
2670 *
2671 * Convert a dentry into an ASCII path name. If the entry has been deleted
2672 * the string " (deleted)" is appended. Note that this is ambiguous.
2673 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002674 * Returns a pointer into the buffer or an error code if the path was
2675 * too long. Note: Callers should use the returned pointer, not the passed
2676 * in buffer, to use the name! The implementation often starts at an offset
2677 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002678 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002679 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002680 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002681char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002683 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002684 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002685 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002687 /*
2688 * We have various synthetic filesystems that never get mounted. On
2689 * these filesystems dentries are never used for lookup purposes, and
2690 * thus don't need to be hashed. They also don't need a name until a
2691 * user wants to identify the object in /proc/pid/fd/. The little hack
2692 * below allows us to generate a name for these objects on demand:
2693 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002694 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2695 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002696
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002697 get_fs_root(current->fs, &root);
Al Viro7ea600b2013-03-26 18:25:57 -04002698 br_read_lock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002699 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002700 error = path_with_deleted(path, &root, &res, &buflen);
Al Viro7ea600b2013-03-26 18:25:57 -04002701 write_sequnlock(&rename_lock);
2702 br_read_unlock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002703 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002704 res = ERR_PTR(error);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002705 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 return res;
2707}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002708EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709
2710/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002711 * Helper function for dentry_operations.d_dname() members
2712 */
2713char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2714 const char *fmt, ...)
2715{
2716 va_list args;
2717 char temp[64];
2718 int sz;
2719
2720 va_start(args, fmt);
2721 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2722 va_end(args);
2723
2724 if (sz > sizeof(temp) || sz > buflen)
2725 return ERR_PTR(-ENAMETOOLONG);
2726
2727 buffer += buflen - sz;
2728 return memcpy(buffer, temp, sz);
2729}
2730
Al Viro118b2302013-08-24 12:08:17 -04002731char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2732{
2733 char *end = buffer + buflen;
2734 /* these dentries are never renamed, so d_lock is not needed */
2735 if (prepend(&end, &buflen, " (deleted)", 11) ||
2736 prepend_name(&end, &buflen, &dentry->d_name) ||
2737 prepend(&end, &buflen, "/", 1))
2738 end = ERR_PTR(-ENAMETOOLONG);
2739 return end;
2740}
2741
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002742/*
Ram Pai6092d042008-03-27 13:06:20 +01002743 * Write full pathname from the root of the filesystem into the buffer.
2744 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002745static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002746{
2747 char *end = buf + buflen;
2748 char *retval;
2749
Ram Pai6092d042008-03-27 13:06:20 +01002750 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002751 if (buflen < 1)
2752 goto Elong;
2753 /* Get '/' right */
2754 retval = end-1;
2755 *retval = '/';
2756
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002757 while (!IS_ROOT(dentry)) {
2758 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002759 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002760
Ram Pai6092d042008-03-27 13:06:20 +01002761 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002762 spin_lock(&dentry->d_lock);
2763 error = prepend_name(&end, &buflen, &dentry->d_name);
2764 spin_unlock(&dentry->d_lock);
2765 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002766 goto Elong;
2767
2768 retval = end;
2769 dentry = parent;
2770 }
Al Viroc1031352010-06-06 22:31:14 -04002771 return retval;
2772Elong:
2773 return ERR_PTR(-ENAMETOOLONG);
2774}
Nick Pigginec2447c2011-01-07 17:49:29 +11002775
2776char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2777{
2778 char *retval;
2779
Nick Piggin949854d2011-01-07 17:49:37 +11002780 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002781 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002782 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002783
2784 return retval;
2785}
2786EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002787
2788char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2789{
2790 char *p = NULL;
2791 char *retval;
2792
Nick Piggin949854d2011-01-07 17:49:37 +11002793 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002794 if (d_unlinked(dentry)) {
2795 p = buf + buflen;
2796 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2797 goto Elong;
2798 buflen++;
2799 }
2800 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002801 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002802 if (!IS_ERR(retval) && p)
2803 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002804 return retval;
2805Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002806 return ERR_PTR(-ENAMETOOLONG);
2807}
2808
2809/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 * NOTE! The user-level library version returns a
2811 * character pointer. The kernel system call just
2812 * returns the length of the buffer filled (which
2813 * includes the ending '\0' character), or a negative
2814 * error value. So libc would do something like
2815 *
2816 * char *getcwd(char * buf, size_t size)
2817 * {
2818 * int retval;
2819 *
2820 * retval = sys_getcwd(buf, size);
2821 * if (retval >= 0)
2822 * return buf;
2823 * errno = -retval;
2824 * return NULL;
2825 * }
2826 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002827SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002828{
Linus Torvalds552ce542007-02-13 12:08:18 -08002829 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002830 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002831 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832
2833 if (!page)
2834 return -ENOMEM;
2835
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002836 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837
Linus Torvalds552ce542007-02-13 12:08:18 -08002838 error = -ENOENT;
Al Viro7ea600b2013-03-26 18:25:57 -04002839 br_read_lock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002840 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002841 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002842 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002843 char *cwd = page + PAGE_SIZE;
2844 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002846 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002847 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002848 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002849 br_read_unlock(&vfsmount_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002850
Al Viro02125a82011-12-05 08:43:34 -05002851 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002852 goto out;
2853
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002854 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002855 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002856 error = prepend_unreachable(&cwd, &buflen);
2857 if (error)
2858 goto out;
2859 }
2860
Linus Torvalds552ce542007-02-13 12:08:18 -08002861 error = -ERANGE;
2862 len = PAGE_SIZE + page - cwd;
2863 if (len <= size) {
2864 error = len;
2865 if (copy_to_user(buf, cwd, len))
2866 error = -EFAULT;
2867 }
Nick Piggin949854d2011-01-07 17:49:37 +11002868 } else {
2869 write_sequnlock(&rename_lock);
Al Viro7ea600b2013-03-26 18:25:57 -04002870 br_read_unlock(&vfsmount_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002871 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
2873out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002874 path_put(&pwd);
2875 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 free_page((unsigned long) page);
2877 return error;
2878}
2879
2880/*
2881 * Test whether new_dentry is a subdirectory of old_dentry.
2882 *
2883 * Trivially implemented using the dcache structure
2884 */
2885
2886/**
2887 * is_subdir - is new dentry a subdirectory of old_dentry
2888 * @new_dentry: new dentry
2889 * @old_dentry: old dentry
2890 *
2891 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2892 * Returns 0 otherwise.
2893 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2894 */
2895
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002896int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897{
2898 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002899 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002901 if (new_dentry == old_dentry)
2902 return 1;
2903
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002904 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002907 /*
2908 * Need rcu_readlock to protect against the d_parent trashing
2909 * due to d_move
2910 */
2911 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002912 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002914 else
2915 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002916 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918
2919 return result;
2920}
2921
2922void d_genocide(struct dentry *root)
2923{
Nick Piggin949854d2011-01-07 17:49:37 +11002924 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002926 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002927 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
Nick Piggin949854d2011-01-07 17:49:37 +11002929 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002930again:
2931 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002932 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933repeat:
2934 next = this_parent->d_subdirs.next;
2935resume:
2936 while (next != &this_parent->d_subdirs) {
2937 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002938 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002940
Nick Pigginda502952011-01-07 17:49:33 +11002941 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2942 if (d_unhashed(dentry) || !dentry->d_inode) {
2943 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002944 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002947 spin_unlock(&this_parent->d_lock);
2948 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002950 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 goto repeat;
2952 }
Nick Piggin949854d2011-01-07 17:49:37 +11002953 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2954 dentry->d_flags |= DCACHE_GENOCIDE;
Waiman Long98474232013-08-28 18:24:59 -07002955 dentry->d_lockref.count--;
Nick Piggin949854d2011-01-07 17:49:37 +11002956 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002957 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 }
2959 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002960 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11002961 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2962 this_parent->d_flags |= DCACHE_GENOCIDE;
Waiman Long98474232013-08-28 18:24:59 -07002963 this_parent->d_lockref.count--;
Nick Piggin949854d2011-01-07 17:49:37 +11002964 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002965 this_parent = try_to_ascend(this_parent, locked, seq);
2966 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11002967 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11002968 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 goto resume;
2970 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002971 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002972 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11002973 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11002974 if (locked)
2975 write_sequnlock(&rename_lock);
2976 return;
2977
2978rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02002979 if (locked)
2980 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11002981 locked = 1;
2982 write_seqlock(&rename_lock);
2983 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984}
2985
Al Viro60545d02013-06-07 01:20:27 -04002986void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987{
Al Viro60545d02013-06-07 01:20:27 -04002988 inode_dec_link_count(inode);
2989 BUG_ON(dentry->d_name.name != dentry->d_iname ||
2990 !hlist_unhashed(&dentry->d_alias) ||
2991 !d_unlinked(dentry));
2992 spin_lock(&dentry->d_parent->d_lock);
2993 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2994 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
2995 (unsigned long long)inode->i_ino);
2996 spin_unlock(&dentry->d_lock);
2997 spin_unlock(&dentry->d_parent->d_lock);
2998 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999}
Al Viro60545d02013-06-07 01:20:27 -04003000EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001
3002static __initdata unsigned long dhash_entries;
3003static int __init set_dhash_entries(char *str)
3004{
3005 if (!str)
3006 return 0;
3007 dhash_entries = simple_strtoul(str, &str, 0);
3008 return 1;
3009}
3010__setup("dhash_entries=", set_dhash_entries);
3011
3012static void __init dcache_init_early(void)
3013{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003014 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
3016 /* If hashes are distributed across NUMA nodes, defer
3017 * hash allocation until vmalloc space is available.
3018 */
3019 if (hashdist)
3020 return;
3021
3022 dentry_hashtable =
3023 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003024 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 dhash_entries,
3026 13,
3027 HASH_EARLY,
3028 &d_hash_shift,
3029 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003030 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031 0);
3032
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003033 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003034 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035}
3036
Denis Cheng74bf17c2007-10-16 23:26:30 -07003037static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003039 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
3041 /*
3042 * A constructor could be added for stable state like the lists,
3043 * but it is probably not worth it because of the cache nature
3044 * of the dcache.
3045 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003046 dentry_cache = KMEM_CACHE(dentry,
3047 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048
3049 /* Hash may have been set up in dcache_init_early */
3050 if (!hashdist)
3051 return;
3052
3053 dentry_hashtable =
3054 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003055 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 dhash_entries,
3057 13,
3058 0,
3059 &d_hash_shift,
3060 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003061 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062 0);
3063
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003064 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003065 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066}
3067
3068/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003069struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003070EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072EXPORT_SYMBOL(d_genocide);
3073
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074void __init vfs_caches_init_early(void)
3075{
3076 dcache_init_early();
3077 inode_init_early();
3078}
3079
3080void __init vfs_caches_init(unsigned long mempages)
3081{
3082 unsigned long reserve;
3083
3084 /* Base hash sizes on available memory, with a reserve equal to
3085 150% of current kernel size */
3086
3087 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3088 mempages -= reserve;
3089
3090 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003091 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092
Denis Cheng74bf17c2007-10-16 23:26:30 -07003093 dcache_init();
3094 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003096 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 bdev_cache_init();
3098 chrdev_init();
3099}