blob: fbfae008ba44e3e8223f42d4823cfeae2a673ac9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * fs/dcache.c
3 *
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
8
9/*
10 * Notes on the allocation strategy:
11 *
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
15 */
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040021#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
23#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/hash.h>
25#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050026#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mount.h>
28#include <linux/file.h>
29#include <asm/uaccess.h>
30#include <linux/security.h>
31#include <linux/seqlock.h>
32#include <linux/swap.h>
33#include <linux/bootmem.h>
Al Viro5ad4e532009-03-29 19:50:06 -040034#include <linux/fs_struct.h>
Frederic Weisbecker613afbf2009-07-16 15:44:29 +020035#include <linux/hardirq.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110036#include <linux/bit_spinlock.h>
37#include <linux/rculist_bl.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070038#include <linux/prefetch.h>
David Howellsdd179942011-08-16 15:31:30 +010039#include <linux/ratelimit.h>
David Howells07f3f052006-09-30 20:52:18 +020040#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050041#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Nick Piggin789680d2011-01-07 17:49:30 +110043/*
44 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110045 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110047 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
Nick Piggin23044502011-01-07 17:49:31 +110051 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
53 * d_lock protects:
54 * - d_flags
55 * - d_name
56 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110057 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110058 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110059 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110061 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110062 *
63 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110064 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110065 * dentry->d_lock
66 * dcache_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110067 * dcache_hash_bucket lock
68 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110069 *
Nick Pigginda502952011-01-07 17:49:33 +110070 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
72 * ...
73 * dentry->d_parent->d_lock
74 * dentry->d_lock
75 *
76 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110077 * if (dentry1 < dentry2)
78 * dentry1->d_lock
79 * dentry2->d_lock
80 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080081int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83
Nick Piggin23044502011-01-07 17:49:31 +110084static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
Al Viro74c3cbe2007-07-22 08:04:18 -040085__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Nick Piggin949854d2011-01-07 17:49:37 +110087EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Christoph Lametere18b8902006-12-06 20:33:20 -080089static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Linus Torvalds1da177e2005-04-16 15:20:36 -070091/*
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
95 *
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
98 */
99#define D_HASHBITS d_hash_shift
100#define D_HASHMASK d_hash_mask
101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8966be92012-03-02 14:23:30 -0800107static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100109{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 hash = hash + (hash >> D_HASHBITS);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100112 return dentry_hashtable + (hash & D_HASHMASK);
113}
114
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115/* Statistics gathering. */
116struct dentry_stat_t dentry_stat = {
117 .age_limit = 45,
118};
119
Nick Piggin3e880fb2011-01-07 17:49:19 +1100120static DEFINE_PER_CPU(unsigned int, nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100123static int get_nr_dentry(void)
124{
125 int i;
126 int sum = 0;
127 for_each_possible_cpu(i)
128 sum += per_cpu(nr_dentry, i);
129 return sum < 0 ? 0 : sum;
130}
131
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400132int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
133 size_t *lenp, loff_t *ppos)
134{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100135 dentry_stat.nr_dentry = get_nr_dentry();
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400136 return proc_dointvec(table, write, buffer, lenp, ppos);
137}
138#endif
139
Linus Torvalds5483f182012-03-04 15:51:42 -0800140/*
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
143 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700144#ifdef CONFIG_DCACHE_WORD_ACCESS
145
146#include <asm/word-at-a-time.h>
147/*
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
152 *
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
155 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700156static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800157{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800158 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800159
160 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700161 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700162 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800163 if (tcount < sizeof(unsigned long))
164 break;
165 if (unlikely(a != b))
166 return 1;
167 cs += sizeof(unsigned long);
168 ct += sizeof(unsigned long);
169 tcount -= sizeof(unsigned long);
170 if (!tcount)
171 return 0;
172 }
173 mask = ~(~0ul << tcount*8);
174 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700175}
176
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800177#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700178
Linus Torvalds94753db52012-05-10 12:19:19 -0700179static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700180{
Linus Torvalds5483f182012-03-04 15:51:42 -0800181 do {
182 if (*cs != *ct)
183 return 1;
184 cs++;
185 ct++;
186 tcount--;
187 } while (tcount);
188 return 0;
189}
190
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700191#endif
192
Linus Torvalds94753db52012-05-10 12:19:19 -0700193static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
194{
Linus Torvalds6326c712012-05-21 16:14:04 -0700195 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700196 /*
197 * Be careful about RCU walk racing with rename:
198 * use ACCESS_ONCE to fetch the name pointer.
199 *
200 * NOTE! Even if a rename will mean that the length
201 * was not loaded atomically, we don't care. The
202 * RCU walk will check the sequence count eventually,
203 * and catch it. And we won't overrun the buffer,
204 * because we're reading the name pointer atomically,
205 * and a dentry name is guaranteed to be properly
206 * terminated with a NUL byte.
207 *
208 * End result: even if 'len' is wrong, we'll exit
209 * early because the data cannot match (there can
210 * be no NUL in the ct/tcount data)
211 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700212 cs = ACCESS_ONCE(dentry->d_name.name);
213 smp_read_barrier_depends();
214 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700215}
216
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400217static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400219 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
220
Al Virob3d9b7a2012-06-09 13:51:19 -0400221 WARN_ON(!hlist_unhashed(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 if (dname_external(dentry))
223 kfree(dentry->d_name.name);
224 kmem_cache_free(dentry_cache, dentry);
225}
226
227/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100228 * no locks, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
230static void d_free(struct dentry *dentry)
231{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100232 BUG_ON(dentry->d_count);
Nick Piggin3e880fb2011-01-07 17:49:19 +1100233 this_cpu_dec(nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 if (dentry->d_op && dentry->d_op->d_release)
235 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400236
Linus Torvaldsdea36672011-04-24 07:58:46 -0700237 /* if dentry was never visible to RCU, immediate free is OK */
238 if (!(dentry->d_flags & DCACHE_RCUACCESS))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400239 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800240 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400241 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Nick Piggin31e6b012011-01-07 17:49:52 +1100244/**
245 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800246 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100247 * After this call, in-progress rcu-walk path lookup will fail. This
248 * should be called after unhashing, and after changing d_inode (if
249 * the dentry has not already been unhashed).
250 */
251static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
252{
253 assert_spin_locked(&dentry->d_lock);
254 /* Go through a barrier */
255 write_seqcount_barrier(&dentry->d_seq);
256}
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258/*
259 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100260 * d_iput() operation if defined. Dentry has no refcount
261 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800263static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200264 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100265 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
267 struct inode *inode = dentry->d_inode;
268 if (inode) {
269 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400270 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100272 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700273 if (!inode->i_nlink)
274 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (dentry->d_op && dentry->d_op->d_iput)
276 dentry->d_op->d_iput(dentry, inode);
277 else
278 iput(inode);
279 } else {
280 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 }
282}
283
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700284/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100285 * Release the dentry's inode, using the filesystem
286 * d_iput() operation if defined. dentry remains in-use.
287 */
288static void dentry_unlink_inode(struct dentry * dentry)
289 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100290 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100291{
292 struct inode *inode = dentry->d_inode;
293 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400294 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100295 dentry_rcuwalk_barrier(dentry);
296 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100297 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100298 if (!inode->i_nlink)
299 fsnotify_inoderemove(inode);
300 if (dentry->d_op && dentry->d_op->d_iput)
301 dentry->d_op->d_iput(dentry, inode);
302 else
303 iput(inode);
304}
305
306/*
Sage Weilf0023bc2011-10-28 10:02:42 -0700307 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700308 */
309static void dentry_lru_add(struct dentry *dentry)
310{
Christoph Hellwiga4633352010-10-10 05:36:26 -0400311 if (list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100312 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400313 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
314 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100315 dentry_stat.nr_unused++;
Nick Piggin23044502011-01-07 17:49:31 +1100316 spin_unlock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400317 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700318}
319
Nick Piggin23044502011-01-07 17:49:31 +1100320static void __dentry_lru_del(struct dentry *dentry)
321{
322 list_del_init(&dentry->d_lru);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100323 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100324 dentry->d_sb->s_nr_dentry_unused--;
325 dentry_stat.nr_unused--;
326}
327
Sage Weilf0023bc2011-10-28 10:02:42 -0700328/*
329 * Remove a dentry with references from the LRU.
330 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700331static void dentry_lru_del(struct dentry *dentry)
332{
333 if (!list_empty(&dentry->d_lru)) {
Nick Piggin23044502011-01-07 17:49:31 +1100334 spin_lock(&dcache_lru_lock);
335 __dentry_lru_del(dentry);
336 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700337 }
338}
339
Sage Weilf0023bc2011-10-28 10:02:42 -0700340/*
341 * Remove a dentry that is unreferenced and about to be pruned
342 * (unhashed and destroyed) from the LRU, and inform the file system.
343 * This wrapper should be called _prior_ to unhashing a victim dentry.
344 */
345static void dentry_lru_prune(struct dentry *dentry)
346{
347 if (!list_empty(&dentry->d_lru)) {
348 if (dentry->d_flags & DCACHE_OP_PRUNE)
349 dentry->d_op->d_prune(dentry);
350
351 spin_lock(&dcache_lru_lock);
352 __dentry_lru_del(dentry);
353 spin_unlock(&dcache_lru_lock);
354 }
355}
356
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000357static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700358{
Nick Piggin23044502011-01-07 17:49:31 +1100359 spin_lock(&dcache_lru_lock);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400360 if (list_empty(&dentry->d_lru)) {
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000361 list_add_tail(&dentry->d_lru, list);
Christoph Hellwiga4633352010-10-10 05:36:26 -0400362 dentry->d_sb->s_nr_dentry_unused++;
Nick Piggin86c87492011-01-07 17:49:18 +1100363 dentry_stat.nr_unused++;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400364 } else {
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000365 list_move_tail(&dentry->d_lru, list);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700366 }
Nick Piggin23044502011-01-07 17:49:31 +1100367 spin_unlock(&dcache_lru_lock);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700368}
369
Miklos Szeredid52b9082007-05-08 00:23:46 -0700370/**
371 * d_kill - kill dentry and return parent
372 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800373 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700374 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200375 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700376 *
377 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100378 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100379 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
380 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700381 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100382static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200383 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100384 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100385 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700386{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700387 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400388 /*
389 * Inform try_to_ascend() that we are no longer attached to the
390 * dentry tree
391 */
Miklos Szeredib161dfa62012-09-17 22:31:38 +0200392 dentry->d_flags |= DCACHE_DENTRY_KILLED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100393 if (parent)
394 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700395 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100396 /*
397 * dentry_iput drops the locks, at which point nobody (except
398 * transient RCU lookups) can reach this dentry.
399 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700400 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900401 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700402}
403
David Howellsc6627c62011-06-07 14:09:20 +0100404/*
405 * Unhash a dentry without inserting an RCU walk barrier or checking that
406 * dentry->d_lock is locked. The caller must take care of that, if
407 * appropriate.
408 */
409static void __d_shrink(struct dentry *dentry)
410{
411 if (!d_unhashed(dentry)) {
412 struct hlist_bl_head *b;
413 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
414 b = &dentry->d_sb->s_anon;
415 else
416 b = d_hash(dentry->d_parent, dentry->d_name.hash);
417
418 hlist_bl_lock(b);
419 __hlist_bl_del(&dentry->d_hash);
420 dentry->d_hash.pprev = NULL;
421 hlist_bl_unlock(b);
422 }
423}
424
Nick Piggin789680d2011-01-07 17:49:30 +1100425/**
426 * d_drop - drop a dentry
427 * @dentry: dentry to drop
428 *
429 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
430 * be found through a VFS lookup any more. Note that this is different from
431 * deleting the dentry - d_delete will try to mark the dentry negative if
432 * possible, giving a successful _negative_ lookup, while d_drop will
433 * just make the cache lookup fail.
434 *
435 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
436 * reason (NFS timeouts or autofs deletes).
437 *
438 * __d_drop requires dentry->d_lock.
439 */
440void __d_drop(struct dentry *dentry)
441{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700442 if (!d_unhashed(dentry)) {
David Howellsc6627c62011-06-07 14:09:20 +0100443 __d_shrink(dentry);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700444 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100445 }
446}
447EXPORT_SYMBOL(__d_drop);
448
449void d_drop(struct dentry *dentry)
450{
Nick Piggin789680d2011-01-07 17:49:30 +1100451 spin_lock(&dentry->d_lock);
452 __d_drop(dentry);
453 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100454}
455EXPORT_SYMBOL(d_drop);
456
Nick Piggin77812a12011-01-07 17:49:48 +1100457/*
458 * Finish off a dentry we've decided to kill.
459 * dentry->d_lock must be held, returns with it unlocked.
460 * If ref is non-zero, then decrement the refcount too.
461 * Returns dentry requiring refcount drop, or NULL if we're done.
462 */
463static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
464 __releases(dentry->d_lock)
465{
Nick Piggin873feea2011-01-07 17:50:06 +1100466 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100467 struct dentry *parent;
468
Nick Piggin873feea2011-01-07 17:50:06 +1100469 inode = dentry->d_inode;
470 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100471relock:
472 spin_unlock(&dentry->d_lock);
473 cpu_relax();
474 return dentry; /* try again with same dentry */
475 }
476 if (IS_ROOT(dentry))
477 parent = NULL;
478 else
479 parent = dentry->d_parent;
480 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100481 if (inode)
482 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100483 goto relock;
484 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100485
Nick Piggin77812a12011-01-07 17:49:48 +1100486 if (ref)
487 dentry->d_count--;
Sage Weilf0023bc2011-10-28 10:02:42 -0700488 /*
489 * if dentry was on the d_lru list delete it from there.
490 * inform the fs via d_prune that this dentry is about to be
491 * unhashed and destroyed.
492 */
493 dentry_lru_prune(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100494 /* if it was on the hash then remove it */
495 __d_drop(dentry);
496 return d_kill(dentry, parent);
497}
498
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499/*
500 * This is dput
501 *
502 * This is complicated by the fact that we do not want to put
503 * dentries that are no longer on any hash chain on the unused
504 * list: we'd much rather just get rid of them immediately.
505 *
506 * However, that implies that we have to traverse the dentry
507 * tree upwards to the parents which might _also_ now be
508 * scheduled for deletion (it may have been only waiting for
509 * its last child to go away).
510 *
511 * This tail recursion is done by hand as we don't want to depend
512 * on the compiler to always get this right (gcc generally doesn't).
513 * Real recursion would eat up our stack space.
514 */
515
516/*
517 * dput - release a dentry
518 * @dentry: dentry to release
519 *
520 * Release a dentry. This will drop the usage count and if appropriate
521 * call the dentry unlink method as well as removing it from the queues and
522 * releasing its resources. If the parent dentries were scheduled for release
523 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525void dput(struct dentry *dentry)
526{
527 if (!dentry)
528 return;
529
530repeat:
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100531 if (dentry->d_count == 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 spin_lock(&dentry->d_lock);
Nick Piggin61f3dee2011-01-07 17:49:40 +1100534 BUG_ON(!dentry->d_count);
535 if (dentry->d_count > 1) {
536 dentry->d_count--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return;
539 }
540
Nick Pigginfb045ad2011-01-07 17:49:55 +1100541 if (dentry->d_flags & DCACHE_OP_DELETE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100543 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 }
Nick Piggin265ac902010-10-10 05:36:24 -0400545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 /* Unreachable? Get rid of it */
547 if (d_unhashed(dentry))
548 goto kill_it;
Nick Piggin265ac902010-10-10 05:36:24 -0400549
Jeff Layton39e3c952012-11-28 11:30:53 -0500550 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400551 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400552
Nick Piggin61f3dee2011-01-07 17:49:40 +1100553 dentry->d_count--;
554 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 return;
556
Miklos Szeredid52b9082007-05-08 00:23:46 -0700557kill_it:
Nick Piggin77812a12011-01-07 17:49:48 +1100558 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700559 if (dentry)
560 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700562EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564/**
565 * d_invalidate - invalidate a dentry
566 * @dentry: dentry to invalidate
567 *
568 * Try to invalidate the dentry if it turns out to be
569 * possible. If there are other dentries that can be
570 * reached through this one we can't delete it and we
571 * return -EBUSY. On success we return 0.
572 *
573 * no dcache lock.
574 */
575
576int d_invalidate(struct dentry * dentry)
577{
578 /*
579 * If it's already been dropped, return OK.
580 */
Nick Pigginda502952011-01-07 17:49:33 +1100581 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100583 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return 0;
585 }
586 /*
587 * Check whether to do a partial shrink_dcache
588 * to get rid of unused child entries.
589 */
590 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100591 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100593 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
595
596 /*
597 * Somebody else still using it?
598 *
599 * If it's a directory, we can't drop it
600 * for fear of somebody re-populating it
601 * with children (even though dropping it
602 * would make it unreachable from the root,
603 * we might still populate it if it was a
604 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000605 * We also need to leave mountpoints alone,
606 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Al Viro50e69632011-11-07 16:39:57 +0000608 if (dentry->d_count > 1 && dentry->d_inode) {
609 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 return -EBUSY;
612 }
613 }
614
615 __d_drop(dentry);
616 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 return 0;
618}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700619EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100621/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100622static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100624 dentry->d_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Nick Piggindc0474b2011-01-07 17:49:43 +1100627static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100628{
Nick Piggin23044502011-01-07 17:49:31 +1100629 spin_lock(&dentry->d_lock);
Nick Piggindc0474b2011-01-07 17:49:43 +1100630 __dget_dlock(dentry);
Nick Piggin23044502011-01-07 17:49:31 +1100631 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100632}
633
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100634struct dentry *dget_parent(struct dentry *dentry)
635{
636 struct dentry *ret;
637
638repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100639 /*
640 * Don't need rcu_dereference because we re-check it was correct under
641 * the lock.
642 */
643 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100644 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100645 spin_lock(&ret->d_lock);
646 if (unlikely(ret != dentry->d_parent)) {
647 spin_unlock(&ret->d_lock);
648 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100649 goto repeat;
650 }
Nick Piggina734eb42011-01-07 17:49:44 +1100651 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100652 BUG_ON(!ret->d_count);
653 ret->d_count++;
654 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100655 return ret;
656}
657EXPORT_SYMBOL(dget_parent);
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659/**
660 * d_find_alias - grab a hashed alias of inode
661 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700662 * @want_discon: flag, used by d_splice_alias, to request
663 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 *
665 * If inode has a hashed alias, or is a directory and has any alias,
666 * acquire the reference to alias and return it. Otherwise return NULL.
667 * Notice that if inode is a directory there can be only one alias and
668 * it can be unhashed only if it has no children, or if it is the root
669 * of a filesystem.
670 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700671 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700672 * any other hashed alias over that one unless @want_discon is set,
673 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700675static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676{
Nick Pigginda502952011-01-07 17:49:33 +1100677 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Nick Pigginda502952011-01-07 17:49:33 +1100679again:
680 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800681 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100682 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700684 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100685 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700687 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100688 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100689 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 return alias;
691 }
692 }
Nick Pigginda502952011-01-07 17:49:33 +1100693 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Nick Pigginda502952011-01-07 17:49:33 +1100695 if (discon_alias) {
696 alias = discon_alias;
697 spin_lock(&alias->d_lock);
698 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
699 if (IS_ROOT(alias) &&
700 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100701 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100702 spin_unlock(&alias->d_lock);
703 return alias;
704 }
705 }
706 spin_unlock(&alias->d_lock);
707 goto again;
708 }
709 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
Nick Pigginda502952011-01-07 17:49:33 +1100712struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
David Howells214fda12006-03-25 03:06:36 -0800714 struct dentry *de = NULL;
715
Al Virob3d9b7a2012-06-09 13:51:19 -0400716 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100717 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700718 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100719 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 return de;
722}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700723EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
725/*
726 * Try to kill dentries associated with this inode.
727 * WARNING: you must own a reference to inode.
728 */
729void d_prune_aliases(struct inode *inode)
730{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700731 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100733 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800734 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100736 if (!dentry->d_count) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100737 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 __d_drop(dentry);
739 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100740 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 dput(dentry);
742 goto restart;
743 }
744 spin_unlock(&dentry->d_lock);
745 }
Nick Piggin873feea2011-01-07 17:50:06 +1100746 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700748EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750/*
Nick Piggin77812a12011-01-07 17:49:48 +1100751 * Try to throw away a dentry - free the inode, dput the parent.
752 * Requires dentry->d_lock is held, and dentry->d_count == 0.
753 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700754 *
Nick Piggin77812a12011-01-07 17:49:48 +1100755 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 */
Nick Piggin77812a12011-01-07 17:49:48 +1100757static void try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200758 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Nick Piggin77812a12011-01-07 17:49:48 +1100760 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700761
Nick Piggin77812a12011-01-07 17:49:48 +1100762 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700763 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100764 * If dentry_kill returns NULL, we have nothing more to do.
765 * if it returns the same dentry, trylocks failed. In either
766 * case, just loop again.
767 *
768 * Otherwise, we need to prune ancestors too. This is necessary
769 * to prevent quadratic behavior of shrink_dcache_parent(), but
770 * is also expected to be beneficial in reducing dentry cache
771 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700772 */
Nick Piggin77812a12011-01-07 17:49:48 +1100773 if (!parent)
774 return;
775 if (parent == dentry)
776 return;
777
778 /* Prune ancestors. */
779 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700780 while (dentry) {
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100781 spin_lock(&dentry->d_lock);
Nick Piggin89e60542011-01-07 17:49:45 +1100782 if (dentry->d_count > 1) {
783 dentry->d_count--;
784 spin_unlock(&dentry->d_lock);
785 return;
786 }
Nick Piggin77812a12011-01-07 17:49:48 +1100787 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700788 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
790
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400791static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700793 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700794
Nick Pigginec336792011-01-07 17:49:47 +1100795 rcu_read_lock();
796 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100797 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
798 if (&dentry->d_lru == list)
799 break; /* empty */
800 spin_lock(&dentry->d_lock);
801 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
802 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100803 continue;
804 }
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 /*
807 * We found an inuse dentry which was not removed from
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700808 * the LRU because of laziness during lookup. Do not free
809 * it - just keep it off the LRU list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 */
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100811 if (dentry->d_count) {
Nick Pigginec336792011-01-07 17:49:47 +1100812 dentry_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700813 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 continue;
815 }
Nick Pigginec336792011-01-07 17:49:47 +1100816
Nick Pigginec336792011-01-07 17:49:47 +1100817 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100818
819 try_prune_one_dentry(dentry);
820
Nick Pigginec336792011-01-07 17:49:47 +1100821 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 }
Nick Pigginec336792011-01-07 17:49:47 +1100823 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400824}
825
826/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000827 * prune_dcache_sb - shrink the dcache
828 * @sb: superblock
829 * @count: number of entries to try to free
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400830 *
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000831 * Attempt to shrink the superblock dcache LRU by @count entries. This is
832 * done when we need more memory an called from the superblock shrinker
833 * function.
834 *
835 * This function may fail to free any resources if all the dentries are in
836 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400837 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000838void prune_dcache_sb(struct super_block *sb, int count)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400839{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400840 struct dentry *dentry;
841 LIST_HEAD(referenced);
842 LIST_HEAD(tmp);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400843
Nick Piggin23044502011-01-07 17:49:31 +1100844relock:
845 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400846 while (!list_empty(&sb->s_dentry_lru)) {
847 dentry = list_entry(sb->s_dentry_lru.prev,
848 struct dentry, d_lru);
849 BUG_ON(dentry->d_sb != sb);
850
Nick Piggin23044502011-01-07 17:49:31 +1100851 if (!spin_trylock(&dentry->d_lock)) {
852 spin_unlock(&dcache_lru_lock);
853 cpu_relax();
854 goto relock;
855 }
856
Dave Chinnerb48f03b2011-08-23 18:56:24 +1000857 if (dentry->d_flags & DCACHE_REFERENCED) {
Nick Piggin23044502011-01-07 17:49:31 +1100858 dentry->d_flags &= ~DCACHE_REFERENCED;
859 list_move(&dentry->d_lru, &referenced);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400860 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100861 } else {
862 list_move_tail(&dentry->d_lru, &tmp);
Miklos Szeredieaf5f902012-01-10 18:22:25 +0100863 dentry->d_flags |= DCACHE_SHRINK_LIST;
Nick Piggin23044502011-01-07 17:49:31 +1100864 spin_unlock(&dentry->d_lock);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000865 if (!--count)
Nick Piggin23044502011-01-07 17:49:31 +1100866 break;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400867 }
Nick Pigginec336792011-01-07 17:49:47 +1100868 cond_resched_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400869 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700870 if (!list_empty(&referenced))
871 list_splice(&referenced, &sb->s_dentry_lru);
Nick Piggin23044502011-01-07 17:49:31 +1100872 spin_unlock(&dcache_lru_lock);
Nick Pigginec336792011-01-07 17:49:47 +1100873
874 shrink_dentry_list(&tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700877/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 * shrink_dcache_sb - shrink dcache for a superblock
879 * @sb: superblock
880 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400881 * Shrink the dcache for the specified super block. This is used to free
882 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400884void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400886 LIST_HEAD(tmp);
887
Nick Piggin23044502011-01-07 17:49:31 +1100888 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400889 while (!list_empty(&sb->s_dentry_lru)) {
890 list_splice_init(&sb->s_dentry_lru, &tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100891 spin_unlock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400892 shrink_dentry_list(&tmp);
Nick Pigginec336792011-01-07 17:49:47 +1100893 spin_lock(&dcache_lru_lock);
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400894 }
Nick Piggin23044502011-01-07 17:49:31 +1100895 spin_unlock(&dcache_lru_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700897EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899/*
David Howellsc636ebd2006-10-11 01:22:19 -0700900 * destroy a single subtree of dentries for unmount
901 * - see the comments on shrink_dcache_for_umount() for a description of the
902 * locking
903 */
904static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
905{
906 struct dentry *parent;
907
908 BUG_ON(!IS_ROOT(dentry));
909
David Howellsc636ebd2006-10-11 01:22:19 -0700910 for (;;) {
911 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +0100912 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -0700913 dentry = list_entry(dentry->d_subdirs.next,
914 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -0700915
916 /* consume the dentries from this leaf up through its parents
917 * until we find one with children or run out altogether */
918 do {
919 struct inode *inode;
920
Sage Weilf0023bc2011-10-28 10:02:42 -0700921 /*
922 * remove the dentry from the lru, and inform
923 * the fs that this dentry is about to be
924 * unhashed and destroyed.
925 */
926 dentry_lru_prune(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +0100927 __d_shrink(dentry);
928
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100929 if (dentry->d_count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -0700930 printk(KERN_ERR
931 "BUG: Dentry %p{i=%lx,n=%s}"
932 " still in use (%d)"
933 " [unmount of %s %s]\n",
934 dentry,
935 dentry->d_inode ?
936 dentry->d_inode->i_ino : 0UL,
937 dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100938 dentry->d_count,
David Howellsc636ebd2006-10-11 01:22:19 -0700939 dentry->d_sb->s_type->name,
940 dentry->d_sb->s_id);
941 BUG();
942 }
943
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100944 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -0700945 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100946 list_del(&dentry->d_u.d_child);
947 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900948 parent = dentry->d_parent;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100949 parent->d_count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100950 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900951 }
David Howellsc636ebd2006-10-11 01:22:19 -0700952
David Howellsc636ebd2006-10-11 01:22:19 -0700953 inode = dentry->d_inode;
954 if (inode) {
955 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400956 hlist_del_init(&dentry->d_alias);
David Howellsc636ebd2006-10-11 01:22:19 -0700957 if (dentry->d_op && dentry->d_op->d_iput)
958 dentry->d_op->d_iput(dentry, inode);
959 else
960 iput(inode);
961 }
962
963 d_free(dentry);
964
965 /* finished when we fall off the top of the tree,
966 * otherwise we ascend to the parent and move to the
967 * next sibling if there is one */
968 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400969 return;
David Howellsc636ebd2006-10-11 01:22:19 -0700970 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -0700971 } while (list_empty(&dentry->d_subdirs));
972
973 dentry = list_entry(dentry->d_subdirs.next,
974 struct dentry, d_u.d_child);
975 }
976}
977
978/*
979 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100980 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -0700981 * - the superblock is detached from all mountings and open files, so the
982 * dentry trees will not be rearranged by the VFS
983 * - s_umount is write-locked, so the memory pressure shrinker will ignore
984 * any dentries belonging to this superblock that it comes across
985 * - the filesystem itself is no longer permitted to rearrange the dentries
986 * in this superblock
987 */
988void shrink_dcache_for_umount(struct super_block *sb)
989{
990 struct dentry *dentry;
991
992 if (down_read_trylock(&sb->s_umount))
993 BUG();
994
995 dentry = sb->s_root;
996 sb->s_root = NULL;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100997 dentry->d_count--;
David Howellsc636ebd2006-10-11 01:22:19 -0700998 shrink_dcache_for_umount_subtree(dentry);
999
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001000 while (!hlist_bl_empty(&sb->s_anon)) {
1001 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001002 shrink_dcache_for_umount_subtree(dentry);
1003 }
1004}
1005
1006/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001007 * This tries to ascend one level of parenthood, but
1008 * we can race with renaming, so we need to re-check
1009 * the parenthood after dropping the lock and check
1010 * that the sequence number still matches.
1011 */
1012static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1013{
1014 struct dentry *new = old->d_parent;
1015
1016 rcu_read_lock();
1017 spin_unlock(&old->d_lock);
1018 spin_lock(&new->d_lock);
1019
1020 /*
1021 * might go back up the wrong parent if we have had a rename
1022 * or deletion
1023 */
1024 if (new != old->d_parent ||
Miklos Szeredib161dfa62012-09-17 22:31:38 +02001025 (old->d_flags & DCACHE_DENTRY_KILLED) ||
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001026 (!locked && read_seqretry(&rename_lock, seq))) {
1027 spin_unlock(&new->d_lock);
1028 new = NULL;
1029 }
1030 rcu_read_unlock();
1031 return new;
1032}
1033
1034
1035/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 * Search for at least 1 mount point in the dentry's subdirs.
1037 * We descend to the next level whenever the d_subdirs
1038 * list is non-empty and continue searching.
1039 */
1040
1041/**
1042 * have_submounts - check for mounts over a dentry
1043 * @parent: dentry to check.
1044 *
1045 * Return true if the parent or its subdirectories contain
1046 * a mount point
1047 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048int have_submounts(struct dentry *parent)
1049{
Nick Piggin949854d2011-01-07 17:49:37 +11001050 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001052 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11001053 int locked = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11001054
Nick Piggin949854d2011-01-07 17:49:37 +11001055 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001056again:
1057 this_parent = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (d_mountpoint(parent))
1060 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001061 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062repeat:
1063 next = this_parent->d_subdirs.next;
1064resume:
1065 while (next != &this_parent->d_subdirs) {
1066 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001067 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001069
1070 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 /* Have we found a mount point ? */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001072 if (d_mountpoint(dentry)) {
1073 spin_unlock(&dentry->d_lock);
1074 spin_unlock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 goto positive;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001076 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001078 spin_unlock(&this_parent->d_lock);
1079 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001081 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 goto repeat;
1083 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001084 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086 /*
1087 * All done at this level ... ascend and resume the search.
1088 */
1089 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001090 struct dentry *child = this_parent;
1091 this_parent = try_to_ascend(this_parent, locked, seq);
1092 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001093 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001094 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 goto resume;
1096 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001097 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001098 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001099 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001100 if (locked)
1101 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 return 0; /* No mount points found in tree */
1103positive:
Nick Piggin58db63d2011-01-07 17:49:39 +11001104 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001105 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001106 if (locked)
1107 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 return 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001109
1110rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02001111 if (locked)
1112 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001113 locked = 1;
1114 write_seqlock(&rename_lock);
1115 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001117EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001120 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 * and move any unused dentries to the end of the unused
1122 * list for prune_dcache(). We descend to the next level
1123 * whenever the d_subdirs list is non-empty and continue
1124 * searching.
1125 *
1126 * It returns zero iff there are no unused children,
1127 * otherwise it returns the number of children moved to
1128 * the end of the unused list. This may not be the total
1129 * number of unused children, because select_parent can
1130 * drop the lock and return early due to latency
1131 * constraints.
1132 */
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001133static int select_parent(struct dentry *parent, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134{
Nick Piggin949854d2011-01-07 17:49:37 +11001135 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11001137 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 int found = 0;
Nick Piggin58db63d2011-01-07 17:49:39 +11001139 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Nick Piggin949854d2011-01-07 17:49:37 +11001141 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001142again:
1143 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001144 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145repeat:
1146 next = this_parent->d_subdirs.next;
1147resume:
1148 while (next != &this_parent->d_subdirs) {
1149 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001150 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 next = tmp->next;
1152
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001153 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggin23044502011-01-07 17:49:31 +11001154
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001155 /*
1156 * move only zero ref count dentries to the dispose list.
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001157 *
1158 * Those which are presently on the shrink list, being processed
1159 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1160 * loop in shrink_dcache_parent() might not make any progress
1161 * and loop forever.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 */
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001163 if (dentry->d_count) {
Christoph Hellwiga4633352010-10-10 05:36:26 -04001164 dentry_lru_del(dentry);
Miklos Szeredieaf5f902012-01-10 18:22:25 +01001165 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1166 dentry_lru_move_list(dentry, dispose);
1167 dentry->d_flags |= DCACHE_SHRINK_LIST;
1168 found++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 /*
1171 * We can return to the caller if we have found some (this
1172 * ensures forward progress). We'll be coming back to find
1173 * the rest.
1174 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001175 if (found && need_resched()) {
1176 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 goto out;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
1180 /*
1181 * Descend a level if the d_subdirs list is non-empty.
1182 */
1183 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001184 spin_unlock(&this_parent->d_lock);
1185 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001187 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 goto repeat;
1189 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001190
1191 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 }
1193 /*
1194 * All done at this level ... ascend and resume the search.
1195 */
1196 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001197 struct dentry *child = this_parent;
1198 this_parent = try_to_ascend(this_parent, locked, seq);
1199 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001200 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001201 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 goto resume;
1203 }
1204out:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001205 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11001206 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001207 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11001208 if (locked)
1209 write_sequnlock(&rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 return found;
Nick Piggin58db63d2011-01-07 17:49:39 +11001211
1212rename_retry:
1213 if (found)
1214 return found;
Miklos Szeredi8110e162012-09-17 22:23:30 +02001215 if (locked)
1216 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11001217 locked = 1;
1218 write_seqlock(&rename_lock);
1219 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220}
1221
1222/**
1223 * shrink_dcache_parent - prune dcache
1224 * @parent: parent of entries to prune
1225 *
1226 * Prune the dcache to remove unused children of the parent dentry.
1227 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228void shrink_dcache_parent(struct dentry * parent)
1229{
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001230 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 int found;
1232
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001233 while ((found = select_parent(parent, &dispose)) != 0)
1234 shrink_dentry_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001236EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238/**
Al Viroa4464db2011-07-07 15:03:58 -04001239 * __d_alloc - allocate a dcache entry
1240 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 * @name: qstr of the name
1242 *
1243 * Allocates a dentry. It returns %NULL if there is insufficient memory
1244 * available. On a success the dentry is returned. The name passed in is
1245 * copied and the copy passed in may be reused after this call.
1246 */
1247
Al Viroa4464db2011-07-07 15:03:58 -04001248struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249{
1250 struct dentry *dentry;
1251 char *dname;
1252
Mel Gormane12ba742007-10-16 01:25:52 -07001253 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (!dentry)
1255 return NULL;
1256
Linus Torvalds6326c712012-05-21 16:14:04 -07001257 /*
1258 * We guarantee that the inline name is always NUL-terminated.
1259 * This way the memcpy() done by the name switching in rename
1260 * will still always have a NUL at the end, even if we might
1261 * be overwriting an internal NUL character
1262 */
1263 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 if (name->len > DNAME_INLINE_LEN-1) {
1265 dname = kmalloc(name->len + 1, GFP_KERNEL);
1266 if (!dname) {
1267 kmem_cache_free(dentry_cache, dentry);
1268 return NULL;
1269 }
1270 } else {
1271 dname = dentry->d_iname;
1272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
1274 dentry->d_name.len = name->len;
1275 dentry->d_name.hash = name->hash;
1276 memcpy(dname, name->name, name->len);
1277 dname[name->len] = 0;
1278
Linus Torvalds6326c712012-05-21 16:14:04 -07001279 /* Make sure we always see the terminating NUL character */
1280 smp_wmb();
1281 dentry->d_name.name = dname;
1282
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001283 dentry->d_count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001284 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001286 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001288 dentry->d_parent = dentry;
1289 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 dentry->d_op = NULL;
1291 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001292 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 INIT_LIST_HEAD(&dentry->d_lru);
1294 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001295 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001296 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001297 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Nick Piggin3e880fb2011-01-07 17:49:19 +11001299 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 return dentry;
1302}
Al Viroa4464db2011-07-07 15:03:58 -04001303
1304/**
1305 * d_alloc - allocate a dcache entry
1306 * @parent: parent of entry to allocate
1307 * @name: qstr of the name
1308 *
1309 * Allocates a dentry. It returns %NULL if there is insufficient memory
1310 * available. On a success the dentry is returned. The name passed in is
1311 * copied and the copy passed in may be reused after this call.
1312 */
1313struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1314{
1315 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1316 if (!dentry)
1317 return NULL;
1318
1319 spin_lock(&parent->d_lock);
1320 /*
1321 * don't need child lock because it is not subject
1322 * to concurrency here
1323 */
1324 __dget_dlock(parent);
1325 dentry->d_parent = parent;
1326 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1327 spin_unlock(&parent->d_lock);
1328
1329 return dentry;
1330}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001331EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Nick Piggin4b936882011-01-07 17:50:07 +11001333struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1334{
Al Viroa4464db2011-07-07 15:03:58 -04001335 struct dentry *dentry = __d_alloc(sb, name);
1336 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001337 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001338 return dentry;
1339}
1340EXPORT_SYMBOL(d_alloc_pseudo);
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1343{
1344 struct qstr q;
1345
1346 q.name = name;
1347 q.len = strlen(name);
1348 q.hash = full_name_hash(q.name, q.len);
1349 return d_alloc(parent, &q);
1350}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001351EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Nick Pigginfb045ad2011-01-07 17:49:55 +11001353void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1354{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001355 WARN_ON_ONCE(dentry->d_op);
1356 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001357 DCACHE_OP_COMPARE |
1358 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001359 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001360 DCACHE_OP_DELETE ));
1361 dentry->d_op = op;
1362 if (!op)
1363 return;
1364 if (op->d_hash)
1365 dentry->d_flags |= DCACHE_OP_HASH;
1366 if (op->d_compare)
1367 dentry->d_flags |= DCACHE_OP_COMPARE;
1368 if (op->d_revalidate)
1369 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001370 if (op->d_weak_revalidate)
1371 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001372 if (op->d_delete)
1373 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001374 if (op->d_prune)
1375 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001376
1377}
1378EXPORT_SYMBOL(d_set_d_op);
1379
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001380static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1381{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001382 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001383 if (inode) {
1384 if (unlikely(IS_AUTOMOUNT(inode)))
1385 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
Al Virob3d9b7a2012-06-09 13:51:19 -04001386 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001387 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001388 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001389 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001390 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001391 fsnotify_d_instantiate(dentry, inode);
1392}
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394/**
1395 * d_instantiate - fill in inode information for a dentry
1396 * @entry: dentry to complete
1397 * @inode: inode to attach to this dentry
1398 *
1399 * Fill in inode information in the entry.
1400 *
1401 * This turns negative dentries into productive full members
1402 * of society.
1403 *
1404 * NOTE! This assumes that the inode count has been incremented
1405 * (or otherwise set) by the caller to indicate that it is now
1406 * in use by the dcache.
1407 */
1408
1409void d_instantiate(struct dentry *entry, struct inode * inode)
1410{
Al Virob3d9b7a2012-06-09 13:51:19 -04001411 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001412 if (inode)
1413 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001414 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001415 if (inode)
1416 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 security_d_instantiate(entry, inode);
1418}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001419EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421/**
1422 * d_instantiate_unique - instantiate a non-aliased dentry
1423 * @entry: dentry to instantiate
1424 * @inode: inode to attach to this dentry
1425 *
1426 * Fill in inode information in the entry. On success, it returns NULL.
1427 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001428 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 *
1430 * Note that in order to avoid conflicts with rename() etc, the caller
1431 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001432 *
1433 * This also assumes that the inode count has been incremented
1434 * (or otherwise set) by the caller to indicate that it is now
1435 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 */
David Howells770bfad2006-08-22 20:06:07 -04001437static struct dentry *__d_instantiate_unique(struct dentry *entry,
1438 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439{
1440 struct dentry *alias;
1441 int len = entry->d_name.len;
1442 const char *name = entry->d_name.name;
1443 unsigned int hash = entry->d_name.hash;
1444
David Howells770bfad2006-08-22 20:06:07 -04001445 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001446 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001447 return NULL;
1448 }
1449
Sasha Levinb67bfe02013-02-27 17:06:00 -08001450 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001451 /*
1452 * Don't need alias->d_lock here, because aliases with
1453 * d_parent == entry->d_parent are not subject to name or
1454 * parent changes, because the parent inode i_mutex is held.
1455 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001456 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 continue;
1458 if (alias->d_parent != entry->d_parent)
1459 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001460 if (alias->d_name.len != len)
1461 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001462 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001464 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 return alias;
1466 }
David Howells770bfad2006-08-22 20:06:07 -04001467
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001468 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 return NULL;
1470}
David Howells770bfad2006-08-22 20:06:07 -04001471
1472struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1473{
1474 struct dentry *result;
1475
Al Virob3d9b7a2012-06-09 13:51:19 -04001476 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001477
Nick Piggin873feea2011-01-07 17:50:06 +11001478 if (inode)
1479 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001480 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001481 if (inode)
1482 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001483
1484 if (!result) {
1485 security_d_instantiate(entry, inode);
1486 return NULL;
1487 }
1488
1489 BUG_ON(!d_unhashed(result));
1490 iput(inode);
1491 return result;
1492}
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494EXPORT_SYMBOL(d_instantiate_unique);
1495
Al Viroadc0e912012-01-08 16:49:21 -05001496struct dentry *d_make_root(struct inode *root_inode)
1497{
1498 struct dentry *res = NULL;
1499
1500 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001501 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001502
1503 res = __d_alloc(root_inode->i_sb, &name);
1504 if (res)
1505 d_instantiate(res, root_inode);
1506 else
1507 iput(root_inode);
1508 }
1509 return res;
1510}
1511EXPORT_SYMBOL(d_make_root);
1512
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001513static struct dentry * __d_find_any_alias(struct inode *inode)
1514{
1515 struct dentry *alias;
1516
Al Virob3d9b7a2012-06-09 13:51:19 -04001517 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001518 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001519 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001520 __dget(alias);
1521 return alias;
1522}
1523
Sage Weil46f72b32012-01-10 09:04:37 -08001524/**
1525 * d_find_any_alias - find any alias for a given inode
1526 * @inode: inode to find an alias for
1527 *
1528 * If any aliases exist for the given inode, take and return a
1529 * reference for one of them. If no aliases exist, return %NULL.
1530 */
1531struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001532{
1533 struct dentry *de;
1534
1535 spin_lock(&inode->i_lock);
1536 de = __d_find_any_alias(inode);
1537 spin_unlock(&inode->i_lock);
1538 return de;
1539}
Sage Weil46f72b32012-01-10 09:04:37 -08001540EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001543 * d_obtain_alias - find or allocate a dentry for a given inode
1544 * @inode: inode to allocate the dentry for
1545 *
1546 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1547 * similar open by handle operations. The returned dentry may be anonymous,
1548 * or may have a full name (if the inode was already in the cache).
1549 *
1550 * When called on a directory inode, we must ensure that the inode only ever
1551 * has one dentry. If a dentry is found, that is returned instead of
1552 * allocating a new one.
1553 *
1554 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001555 * to the dentry. In case of an error the reference on the inode is released.
1556 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1557 * be passed in and will be the error will be propagate to the return value,
1558 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001559 */
1560struct dentry *d_obtain_alias(struct inode *inode)
1561{
NeilBrownb911a6b2012-11-08 16:09:37 -08001562 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001563 struct dentry *tmp;
1564 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001565
1566 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001567 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001568 if (IS_ERR(inode))
1569 return ERR_CAST(inode);
1570
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001571 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001572 if (res)
1573 goto out_iput;
1574
Al Viroa4464db2011-07-07 15:03:58 -04001575 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001576 if (!tmp) {
1577 res = ERR_PTR(-ENOMEM);
1578 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001579 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001580
Nick Piggin873feea2011-01-07 17:50:06 +11001581 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001582 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001583 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001584 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001585 dput(tmp);
1586 goto out_iput;
1587 }
1588
1589 /* attach a disconnected dentry */
1590 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001591 tmp->d_inode = inode;
1592 tmp->d_flags |= DCACHE_DISCONNECTED;
Al Virob3d9b7a2012-06-09 13:51:19 -04001593 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001594 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001595 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001596 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001597 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001598 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001599 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001600
Christoph Hellwig9308a612008-08-11 15:49:12 +02001601 return tmp;
1602
1603 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001604 if (res && !IS_ERR(res))
1605 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001606 iput(inode);
1607 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001608}
Benny Halevyadc48722009-02-27 14:02:59 -08001609EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611/**
1612 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1613 * @inode: the inode which may have a disconnected dentry
1614 * @dentry: a negative dentry which we want to point to the inode.
1615 *
1616 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1617 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1618 * and return it, else simply d_add the inode to the dentry and return NULL.
1619 *
1620 * This is needed in the lookup routine of any filesystem that is exportable
1621 * (via knfsd) so that we can build dcache paths to directories effectively.
1622 *
1623 * If a dentry was found and moved, then it is returned. Otherwise NULL
1624 * is returned. This matches the expected return value of ->lookup.
1625 *
1626 */
1627struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1628{
1629 struct dentry *new = NULL;
1630
Al Viroa9049372011-07-08 21:20:11 -04001631 if (IS_ERR(inode))
1632 return ERR_CAST(inode);
1633
NeilBrown21c0d8f2006-10-04 02:16:16 -07001634 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001635 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001636 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001638 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001639 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 d_move(new, dentry);
1642 iput(inode);
1643 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001644 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001645 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001646 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 security_d_instantiate(dentry, inode);
1648 d_rehash(dentry);
1649 }
1650 } else
1651 d_add(dentry, inode);
1652 return new;
1653}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001654EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Barry Naujok94035402008-05-21 16:50:46 +10001656/**
1657 * d_add_ci - lookup or allocate new dentry with case-exact name
1658 * @inode: the inode case-insensitive lookup has found
1659 * @dentry: the negative dentry that was passed to the parent's lookup func
1660 * @name: the case-exact name to be associated with the returned dentry
1661 *
1662 * This is to avoid filling the dcache with case-insensitive names to the
1663 * same inode, only the actual correct case is stored in the dcache for
1664 * case-insensitive filesystems.
1665 *
1666 * For a case-insensitive lookup match and if the the case-exact dentry
1667 * already exists in in the dcache, use it and return it.
1668 *
1669 * If no entry exists with the exact case name, allocate new dentry with
1670 * the exact case, and return the spliced entry.
1671 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001672struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001673 struct qstr *name)
1674{
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
1737 * load the name, length and inode information, so that the
1738 * 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,
1755 struct inode *inode,
1756 struct dentry *dentry,
1757 unsigned int seq,
1758 const struct qstr *name)
1759{
1760 int tlen = dentry->d_name.len;
1761 const char *tname = dentry->d_name.name;
1762 struct inode *i = dentry->d_inode;
1763
1764 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1765 cpu_relax();
1766 return D_COMP_SEQRETRY;
1767 }
1768 if (parent->d_op->d_compare(parent, inode,
1769 dentry, i,
1770 tlen, tname, name))
1771 return D_COMP_NOMATCH;
1772 return D_COMP_OK;
1773}
1774
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775/**
Nick Piggin31e6b012011-01-07 17:49:52 +11001776 * __d_lookup_rcu - search for a dentry (racy, store-free)
1777 * @parent: parent dentry
1778 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07001779 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11001780 * @inode: returns dentry->d_inode when the inode was found valid.
1781 * Returns: dentry, or NULL
1782 *
1783 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1784 * resolution (store-free path walking) design described in
1785 * Documentation/filesystems/path-lookup.txt.
1786 *
1787 * This is not to be used outside core vfs.
1788 *
1789 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1790 * held, and rcu_read_lock held. The returned dentry must not be stored into
1791 * without taking d_lock and checking d_seq sequence count against @seq
1792 * returned here.
1793 *
1794 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1795 * function.
1796 *
1797 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1798 * the returned dentry, so long as its parent's seqlock is checked after the
1799 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1800 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001801 *
1802 * NOTE! The caller *has* to check the resulting dentry against the sequence
1803 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11001804 */
Linus Torvalds8966be92012-03-02 14:23:30 -08001805struct dentry *__d_lookup_rcu(const struct dentry *parent,
1806 const struct qstr *name,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001807 unsigned *seqp, struct inode *inode)
Nick Piggin31e6b012011-01-07 17:49:52 +11001808{
Linus Torvalds26fe5752012-05-10 13:14:12 -07001809 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11001810 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001811 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001812 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001813 struct dentry *dentry;
1814
1815 /*
1816 * Note: There is significant duplication with __d_lookup_rcu which is
1817 * required to prevent single threaded performance regressions
1818 * especially on architectures where smp_rmb (in seqcounts) are costly.
1819 * Keep the two functions in sync.
1820 */
1821
1822 /*
1823 * The hash list is protected using RCU.
1824 *
1825 * Carefully use d_seq when comparing a candidate dentry, to avoid
1826 * races with d_move().
1827 *
1828 * It is possible that concurrent renames can mess up our list
1829 * walk here and result in missing our dentry, resulting in the
1830 * false-negative result. d_lookup() protects against concurrent
1831 * renames using rename_lock seqlock.
1832 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001833 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11001834 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001835 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08001836 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11001837
Nick Piggin31e6b012011-01-07 17:49:52 +11001838seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001839 /*
1840 * The dentry sequence count protects us from concurrent
1841 * renames, and thus protects inode, parent and name fields.
1842 *
1843 * The caller must perform a seqcount check in order
1844 * to do anything useful with the returned dentry,
1845 * including using the 'd_inode' pointer.
1846 *
1847 * NOTE! We do a "raw" seqcount_begin here. That means that
1848 * we don't wait for the sequence count to stabilize if it
1849 * is in the middle of a sequence change. If we do the slow
1850 * dentry compare, we will do seqretries until it is stable,
1851 * and if we end up with a successful lookup, we actually
1852 * want to exit RCU lookup anyway.
1853 */
1854 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11001855 if (dentry->d_parent != parent)
1856 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07001857 if (d_unhashed(dentry))
1858 continue;
Linus Torvalds8966be92012-03-02 14:23:30 -08001859 *seqp = seq;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001860
1861 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001862 if (dentry->d_name.hash != hashlen_hash(hashlen))
1863 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001864 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1865 case D_COMP_OK:
1866 return dentry;
1867 case D_COMP_NOMATCH:
1868 continue;
1869 default:
1870 goto seqretry;
1871 }
1872 }
1873
Linus Torvalds26fe5752012-05-10 13:14:12 -07001874 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07001875 continue;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001876 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001877 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11001878 }
1879 return NULL;
1880}
1881
1882/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 * d_lookup - search for a dentry
1884 * @parent: parent dentry
1885 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10001886 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 *
Nick Pigginb04f7842010-08-18 04:37:34 +10001888 * d_lookup searches the children of the parent dentry for the name in
1889 * question. If the dentry is found its reference count is incremented and the
1890 * dentry is returned. The caller must use dput to free the entry when it has
1891 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 */
Al Viroda2d8452013-01-24 18:29:34 -05001893struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894{
Nick Piggin31e6b012011-01-07 17:49:52 +11001895 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11001896 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
1898 do {
1899 seq = read_seqbegin(&rename_lock);
1900 dentry = __d_lookup(parent, name);
1901 if (dentry)
1902 break;
1903 } while (read_seqretry(&rename_lock, seq));
1904 return dentry;
1905}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001906EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907
Nick Piggin31e6b012011-01-07 17:49:52 +11001908/**
Nick Pigginb04f7842010-08-18 04:37:34 +10001909 * __d_lookup - search for a dentry (racy)
1910 * @parent: parent dentry
1911 * @name: qstr of name we wish to find
1912 * Returns: dentry, or NULL
1913 *
1914 * __d_lookup is like d_lookup, however it may (rarely) return a
1915 * false-negative result due to unrelated rename activity.
1916 *
1917 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1918 * however it must be used carefully, eg. with a following d_lookup in
1919 * the case of failure.
1920 *
1921 * __d_lookup callers must be commented.
1922 */
Al Viroa713ca22013-01-24 18:27:00 -05001923struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924{
1925 unsigned int len = name->len;
1926 unsigned int hash = name->hash;
1927 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001928 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001929 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11001930 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08001931 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932
Nick Pigginb04f7842010-08-18 04:37:34 +10001933 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11001934 * Note: There is significant duplication with __d_lookup_rcu which is
1935 * required to prevent single threaded performance regressions
1936 * especially on architectures where smp_rmb (in seqcounts) are costly.
1937 * Keep the two functions in sync.
1938 */
1939
1940 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10001941 * The hash list is protected using RCU.
1942 *
1943 * Take d_lock when comparing a candidate dentry, to avoid races
1944 * with d_move().
1945 *
1946 * It is possible that concurrent renames can mess up our list
1947 * walk here and result in missing our dentry, resulting in the
1948 * false-negative result. d_lookup() protects against concurrent
1949 * renames using rename_lock seqlock.
1950 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09001951 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10001952 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 rcu_read_lock();
1954
Linus Torvaldsb07ad992011-04-23 22:32:03 -07001955 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 if (dentry->d_name.hash != hash)
1958 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 if (dentry->d_parent != parent)
1962 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001963 if (d_unhashed(dentry))
1964 goto next;
1965
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 /*
1967 * It is safe to compare names since d_move() cannot
1968 * change the qstr (protected by d_lock).
1969 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11001970 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001971 int tlen = dentry->d_name.len;
1972 const char *tname = dentry->d_name.name;
Nick Piggin621e1552011-01-07 17:49:27 +11001973 if (parent->d_op->d_compare(parent, parent->d_inode,
1974 dentry, dentry->d_inode,
Nick Piggin31e6b012011-01-07 17:49:52 +11001975 tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 goto next;
1977 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07001978 if (dentry->d_name.len != len)
1979 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001980 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 goto next;
1982 }
1983
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001984 dentry->d_count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07001985 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 spin_unlock(&dentry->d_lock);
1987 break;
1988next:
1989 spin_unlock(&dentry->d_lock);
1990 }
1991 rcu_read_unlock();
1992
1993 return found;
1994}
1995
1996/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08001997 * d_hash_and_lookup - hash the qstr then search for a dentry
1998 * @dir: Directory to search in
1999 * @name: qstr of name we wish to find
2000 *
Al Viro4f522a22013-02-11 23:20:37 -05002001 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002002 */
2003struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2004{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002005 /*
2006 * Check for a fs-specific hash function. Note that we must
2007 * calculate the standard hash first, as the d_op->d_hash()
2008 * routine may choose to leave the hash value unchanged.
2009 */
2010 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002011 if (dir->d_flags & DCACHE_OP_HASH) {
Al Viro4f522a22013-02-11 23:20:37 -05002012 int err = dir->d_op->d_hash(dir, dir->d_inode, name);
2013 if (unlikely(err < 0))
2014 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002015 }
Al Viro4f522a22013-02-11 23:20:37 -05002016 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002017}
Al Viro4f522a22013-02-11 23:20:37 -05002018EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002019
2020/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002021 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002023 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 *
2025 * An insecure source has sent us a dentry, here we verify it and dget() it.
2026 * This is used by ncpfs in its readdir implementation.
2027 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002028 *
2029 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002031int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032{
Nick Piggin786a5e12011-01-07 17:49:16 +11002033 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002034
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002035 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002036 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2037 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002038 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002039 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002040 spin_unlock(&dentry->d_lock);
2041 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 return 1;
2043 }
2044 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002045 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 return 0;
2048}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002049EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
2051/*
2052 * When a file is deleted, we have two options:
2053 * - turn this dentry into a negative dentry
2054 * - unhash this dentry and free it.
2055 *
2056 * Usually, we want to just turn this into
2057 * a negative dentry, but if anybody else is
2058 * currently using the dentry or the inode
2059 * we can't do that and we fall back on removing
2060 * it from the hash queues and waiting for
2061 * it to be deleted later when it has no users
2062 */
2063
2064/**
2065 * d_delete - delete a dentry
2066 * @dentry: The dentry to delete
2067 *
2068 * Turn the dentry into a negative dentry if possible, otherwise
2069 * remove it from the hash queues so it can be deleted later
2070 */
2071
2072void d_delete(struct dentry * dentry)
2073{
Nick Piggin873feea2011-01-07 17:50:06 +11002074 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002075 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 /*
2077 * Are we the only user?
2078 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002079again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002081 inode = dentry->d_inode;
2082 isdir = S_ISDIR(inode->i_mode);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002083 if (dentry->d_count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002084 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002085 spin_unlock(&dentry->d_lock);
2086 cpu_relax();
2087 goto again;
2088 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002089 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002090 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002091 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 return;
2093 }
2094
2095 if (!d_unhashed(dentry))
2096 __d_drop(dentry);
2097
2098 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002099
2100 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002102EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002104static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002106 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002107 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002108 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002109 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002110 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111}
2112
David Howells770bfad2006-08-22 20:06:07 -04002113static void _d_rehash(struct dentry * entry)
2114{
2115 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2116}
2117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118/**
2119 * d_rehash - add an entry back to the hash
2120 * @entry: dentry to add to the hash
2121 *
2122 * Adds a dentry to the hash according to its name.
2123 */
2124
2125void d_rehash(struct dentry * entry)
2126{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002128 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002131EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002133/**
2134 * dentry_update_name_case - update case insensitive dentry with a new name
2135 * @dentry: dentry to be updated
2136 * @name: new name
2137 *
2138 * Update a case insensitive dentry with new case of name.
2139 *
2140 * dentry must have been returned by d_lookup with name @name. Old and new
2141 * name lengths must match (ie. no d_compare which allows mismatched name
2142 * lengths).
2143 *
2144 * Parent inode i_mutex must be held over d_lookup and into this call (to
2145 * keep renames and concurrent inserts, and readdir(2) away).
2146 */
2147void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2148{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002149 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002150 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2151
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002152 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002153 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002154 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002155 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002156 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002157}
2158EXPORT_SYMBOL(dentry_update_name_case);
2159
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160static void switch_names(struct dentry *dentry, struct dentry *target)
2161{
2162 if (dname_external(target)) {
2163 if (dname_external(dentry)) {
2164 /*
2165 * Both external: swap the pointers
2166 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002167 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 } else {
2169 /*
2170 * dentry:internal, target:external. Steal target's
2171 * storage and make target internal.
2172 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002173 memcpy(target->d_iname, dentry->d_name.name,
2174 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 dentry->d_name.name = target->d_name.name;
2176 target->d_name.name = target->d_iname;
2177 }
2178 } else {
2179 if (dname_external(dentry)) {
2180 /*
2181 * dentry:external, target:internal. Give dentry's
2182 * storage to target and make dentry internal
2183 */
2184 memcpy(dentry->d_iname, target->d_name.name,
2185 target->d_name.len + 1);
2186 target->d_name.name = dentry->d_name.name;
2187 dentry->d_name.name = dentry->d_iname;
2188 } else {
2189 /*
2190 * Both are internal. Just copy target to dentry
2191 */
2192 memcpy(dentry->d_iname, target->d_name.name,
2193 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002194 dentry->d_name.len = target->d_name.len;
2195 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 }
2197 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002198 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199}
2200
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002201static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2202{
2203 /*
2204 * XXXX: do we really need to take target->d_lock?
2205 */
2206 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2207 spin_lock(&target->d_parent->d_lock);
2208 else {
2209 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2210 spin_lock(&dentry->d_parent->d_lock);
2211 spin_lock_nested(&target->d_parent->d_lock,
2212 DENTRY_D_LOCK_NESTED);
2213 } else {
2214 spin_lock(&target->d_parent->d_lock);
2215 spin_lock_nested(&dentry->d_parent->d_lock,
2216 DENTRY_D_LOCK_NESTED);
2217 }
2218 }
2219 if (target < dentry) {
2220 spin_lock_nested(&target->d_lock, 2);
2221 spin_lock_nested(&dentry->d_lock, 3);
2222 } else {
2223 spin_lock_nested(&dentry->d_lock, 2);
2224 spin_lock_nested(&target->d_lock, 3);
2225 }
2226}
2227
2228static void dentry_unlock_parents_for_move(struct dentry *dentry,
2229 struct dentry *target)
2230{
2231 if (target->d_parent != dentry->d_parent)
2232 spin_unlock(&dentry->d_parent->d_lock);
2233 if (target->d_parent != target)
2234 spin_unlock(&target->d_parent->d_lock);
2235}
2236
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002238 * When switching names, the actual string doesn't strictly have to
2239 * be preserved in the target - because we're dropping the target
2240 * anyway. As such, we can just do a simple memcpy() to copy over
2241 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002243 * Note that we have to be a lot more careful about getting the hash
2244 * switched - we have to switch the hash value properly even if it
2245 * then no longer matches the actual (corrupted) string of the target.
2246 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002248/*
Al Viro18367502011-07-12 21:42:24 -04002249 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 * @dentry: entry to move
2251 * @target: new dentry
2252 *
2253 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002254 * dcache entries should not be moved in this way. Caller must hold
2255 * rename_lock, the i_mutex of the source and target directories,
2256 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 */
Al Viro18367502011-07-12 21:42:24 -04002258static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 if (!dentry->d_inode)
2261 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2262
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002263 BUG_ON(d_ancestor(dentry, target));
2264 BUG_ON(d_ancestor(target, dentry));
2265
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002266 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
Nick Piggin31e6b012011-01-07 17:49:52 +11002268 write_seqcount_begin(&dentry->d_seq);
2269 write_seqcount_begin(&target->d_seq);
2270
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002271 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2272
2273 /*
2274 * Move the dentry to the target hash queue. Don't bother checking
2275 * for the same hash queue because of how unlikely it is.
2276 */
2277 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002278 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
2280 /* Unhash the target: dput() will then get rid of it */
2281 __d_drop(target);
2282
Eric Dumazet5160ee62006-01-08 01:03:32 -08002283 list_del(&dentry->d_u.d_child);
2284 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
2286 /* Switch the names.. */
2287 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002288 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
2290 /* ... and switch the parents */
2291 if (IS_ROOT(dentry)) {
2292 dentry->d_parent = target->d_parent;
2293 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002294 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002296 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297
2298 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002299 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 }
2301
Eric Dumazet5160ee62006-01-08 01:03:32 -08002302 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002303
Nick Piggin31e6b012011-01-07 17:49:52 +11002304 write_seqcount_end(&target->d_seq);
2305 write_seqcount_end(&dentry->d_seq);
2306
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002307 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002309 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002311}
2312
2313/*
2314 * d_move - move a dentry
2315 * @dentry: entry to move
2316 * @target: new dentry
2317 *
2318 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002319 * dcache entries should not be moved in this way. See the locking
2320 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002321 */
2322void d_move(struct dentry *dentry, struct dentry *target)
2323{
2324 write_seqlock(&rename_lock);
2325 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002327}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002328EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002330/**
2331 * d_ancestor - search for an ancestor
2332 * @p1: ancestor dentry
2333 * @p2: child dentry
2334 *
2335 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2336 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002337 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002338struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002339{
2340 struct dentry *p;
2341
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002342 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002343 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002344 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002345 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002346 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002347}
2348
2349/*
2350 * This helper attempts to cope with remotely renamed directories
2351 *
2352 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002353 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002354 *
2355 * Note: If ever the locking in lock_rename() changes, then please
2356 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002357 */
Nick Piggin873feea2011-01-07 17:50:06 +11002358static struct dentry *__d_unalias(struct inode *inode,
2359 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002360{
2361 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002362 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002363
2364 /* If alias and dentry share a parent, then no extra locks required */
2365 if (alias->d_parent == dentry->d_parent)
2366 goto out_unalias;
2367
Trond Myklebust9eaef272006-10-21 10:24:20 -07002368 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002369 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2370 goto out_err;
2371 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2372 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2373 goto out_err;
2374 m2 = &alias->d_parent->d_inode->i_mutex;
2375out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002376 if (likely(!d_mountpoint(alias))) {
2377 __d_move(alias, dentry);
2378 ret = alias;
2379 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002380out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002381 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002382 if (m2)
2383 mutex_unlock(m2);
2384 if (m1)
2385 mutex_unlock(m1);
2386 return ret;
2387}
2388
2389/*
David Howells770bfad2006-08-22 20:06:07 -04002390 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2391 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002392 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002393 */
2394static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2395{
Al Viro740da422013-01-30 10:13:38 -05002396 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002397
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002398 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002399
Nick Piggin31e6b012011-01-07 17:49:52 +11002400 write_seqcount_begin(&dentry->d_seq);
2401 write_seqcount_begin(&anon->d_seq);
2402
David Howells770bfad2006-08-22 20:06:07 -04002403 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002404
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002405 switch_names(dentry, anon);
2406 swap(dentry->d_name.hash, anon->d_name.hash);
2407
Al Viro740da422013-01-30 10:13:38 -05002408 dentry->d_parent = dentry;
2409 list_del_init(&dentry->d_u.d_child);
2410 anon->d_parent = dparent;
David Howells770bfad2006-08-22 20:06:07 -04002411 list_del(&anon->d_u.d_child);
Al Viro740da422013-01-30 10:13:38 -05002412 list_add(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002413
Nick Piggin31e6b012011-01-07 17:49:52 +11002414 write_seqcount_end(&dentry->d_seq);
2415 write_seqcount_end(&anon->d_seq);
2416
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002417 dentry_unlock_parents_for_move(anon, dentry);
2418 spin_unlock(&dentry->d_lock);
2419
2420 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002421 anon->d_flags &= ~DCACHE_DISCONNECTED;
2422}
2423
2424/**
2425 * d_materialise_unique - introduce an inode into the tree
2426 * @dentry: candidate dentry
2427 * @inode: inode to bind to the dentry, to which aliases may be attached
2428 *
2429 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002430 * root directory alias in its place if there is one. Caller must hold the
2431 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002432 */
2433struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2434{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002435 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002436
2437 BUG_ON(!d_unhashed(dentry));
2438
David Howells770bfad2006-08-22 20:06:07 -04002439 if (!inode) {
2440 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002441 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002442 d_rehash(actual);
2443 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002444 }
2445
Nick Piggin873feea2011-01-07 17:50:06 +11002446 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002447
Trond Myklebust9eaef272006-10-21 10:24:20 -07002448 if (S_ISDIR(inode->i_mode)) {
2449 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002450
Trond Myklebust9eaef272006-10-21 10:24:20 -07002451 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002452 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002453 if (alias) {
2454 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002455 write_seqlock(&rename_lock);
2456
2457 if (d_ancestor(alias, dentry)) {
2458 /* Check for loops */
2459 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002460 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002461 } else if (IS_ROOT(alias)) {
2462 /* Is this an anonymous mountpoint that we
2463 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002464 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002465 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002466 __d_drop(alias);
2467 goto found;
Al Viro18367502011-07-12 21:42:24 -04002468 } else {
2469 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002470 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002471 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002472 }
Al Viro18367502011-07-12 21:42:24 -04002473 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002474 if (IS_ERR(actual)) {
2475 if (PTR_ERR(actual) == -ELOOP)
2476 pr_warn_ratelimited(
2477 "VFS: Lookup of '%s' in %s %s"
2478 " would have caused loop\n",
2479 dentry->d_name.name,
2480 inode->i_sb->s_type->name,
2481 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002482 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002483 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002484 goto out_nolock;
2485 }
David Howells770bfad2006-08-22 20:06:07 -04002486 }
2487
2488 /* Add a unique reference */
2489 actual = __d_instantiate_unique(dentry, inode);
2490 if (!actual)
2491 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002492 else
2493 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002494
David Howells770bfad2006-08-22 20:06:07 -04002495 spin_lock(&actual->d_lock);
2496found:
2497 _d_rehash(actual);
2498 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002499 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002500out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002501 if (actual == dentry) {
2502 security_d_instantiate(dentry, inode);
2503 return NULL;
2504 }
2505
2506 iput(inode);
2507 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002508}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002509EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002510
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002511static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002512{
2513 *buflen -= namelen;
2514 if (*buflen < 0)
2515 return -ENAMETOOLONG;
2516 *buffer -= namelen;
2517 memcpy(*buffer, str, namelen);
2518 return 0;
2519}
2520
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002521static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2522{
2523 return prepend(buffer, buflen, name->name, name->len);
2524}
2525
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002527 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002528 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002529 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002530 * @buffer: pointer to the end of the buffer
2531 * @buflen: pointer to buffer length
2532 *
Nick Piggin949854d2011-01-07 17:49:37 +11002533 * Caller holds the rename_lock.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002534 */
Al Viro02125a82011-12-05 08:43:34 -05002535static int prepend_path(const struct path *path,
2536 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002537 char **buffer, int *buflen)
2538{
2539 struct dentry *dentry = path->dentry;
2540 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002541 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002542 bool slash = false;
2543 int error = 0;
2544
Andi Kleen962830d2012-05-08 13:32:02 +09302545 br_read_lock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002546 while (dentry != root->dentry || vfsmnt != root->mnt) {
2547 struct dentry * parent;
2548
2549 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2550 /* Global root? */
Al Viro676da582011-11-24 21:47:05 -05002551 if (!mnt_has_parent(mnt))
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002552 goto global_root;
Al Viroa73324d2011-11-24 22:25:07 -05002553 dentry = mnt->mnt_mountpoint;
Al Viro0714a532011-11-24 22:19:58 -05002554 mnt = mnt->mnt_parent;
2555 vfsmnt = &mnt->mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002556 continue;
2557 }
2558 parent = dentry->d_parent;
2559 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002560 spin_lock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002561 error = prepend_name(buffer, buflen, &dentry->d_name);
Nick Piggin9abca362011-01-07 17:49:36 +11002562 spin_unlock(&dentry->d_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002563 if (!error)
2564 error = prepend(buffer, buflen, "/", 1);
2565 if (error)
2566 break;
2567
2568 slash = true;
2569 dentry = parent;
2570 }
2571
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002572 if (!error && !slash)
2573 error = prepend(buffer, buflen, "/", 1);
2574
Al Viro02125a82011-12-05 08:43:34 -05002575out:
Andi Kleen962830d2012-05-08 13:32:02 +09302576 br_read_unlock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002577 return error;
2578
2579global_root:
2580 /*
2581 * Filesystems needing to implement special "root names"
2582 * should do so with ->d_dname()
2583 */
2584 if (IS_ROOT(dentry) &&
2585 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2586 WARN(1, "Root dentry has weird name <%.*s>\n",
2587 (int) dentry->d_name.len, dentry->d_name.name);
2588 }
Al Viro02125a82011-12-05 08:43:34 -05002589 if (!slash)
2590 error = prepend(buffer, buflen, "/", 1);
2591 if (!error)
Al Virof7a99c52012-06-09 00:59:08 -04002592 error = is_mounted(vfsmnt) ? 1 : 2;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002593 goto out;
2594}
2595
2596/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002597 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002598 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002599 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002600 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 * @buflen: buffer length
2602 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002603 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002605 * Returns a pointer into the buffer or an error code if the
2606 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002607 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002608 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002609 *
Al Viro02125a82011-12-05 08:43:34 -05002610 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611 */
Al Viro02125a82011-12-05 08:43:34 -05002612char *__d_path(const struct path *path,
2613 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002614 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002616 char *res = buf + buflen;
2617 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002619 prepend(&res, &buflen, "\0", 1);
Nick Piggin949854d2011-01-07 17:49:37 +11002620 write_seqlock(&rename_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002621 error = prepend_path(path, root, &res, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002622 write_sequnlock(&rename_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002623
Al Viro02125a82011-12-05 08:43:34 -05002624 if (error < 0)
2625 return ERR_PTR(error);
2626 if (error > 0)
2627 return NULL;
2628 return res;
2629}
2630
2631char *d_absolute_path(const struct path *path,
2632 char *buf, int buflen)
2633{
2634 struct path root = {};
2635 char *res = buf + buflen;
2636 int error;
2637
2638 prepend(&res, &buflen, "\0", 1);
2639 write_seqlock(&rename_lock);
2640 error = prepend_path(path, &root, &res, &buflen);
2641 write_sequnlock(&rename_lock);
2642
2643 if (error > 1)
2644 error = -EINVAL;
2645 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002646 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002647 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648}
2649
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002650/*
2651 * same as __d_path but appends "(deleted)" for unlinked files.
2652 */
Al Viro02125a82011-12-05 08:43:34 -05002653static int path_with_deleted(const struct path *path,
2654 const struct path *root,
2655 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002656{
2657 prepend(buf, buflen, "\0", 1);
2658 if (d_unlinked(path->dentry)) {
2659 int error = prepend(buf, buflen, " (deleted)", 10);
2660 if (error)
2661 return error;
2662 }
2663
2664 return prepend_path(path, root, buf, buflen);
2665}
2666
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002667static int prepend_unreachable(char **buffer, int *buflen)
2668{
2669 return prepend(buffer, buflen, "(unreachable)", 13);
2670}
2671
Jan Bluncka03a8a702008-02-14 19:38:32 -08002672/**
2673 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08002674 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08002675 * @buf: buffer to return value in
2676 * @buflen: buffer length
2677 *
2678 * Convert a dentry into an ASCII path name. If the entry has been deleted
2679 * the string " (deleted)" is appended. Note that this is ambiguous.
2680 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002681 * Returns a pointer into the buffer or an error code if the path was
2682 * too long. Note: Callers should use the returned pointer, not the passed
2683 * in buffer, to use the name! The implementation often starts at an offset
2684 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002685 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002686 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08002687 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07002688char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002690 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002691 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002692 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002694 /*
2695 * We have various synthetic filesystems that never get mounted. On
2696 * these filesystems dentries are never used for lookup purposes, and
2697 * thus don't need to be hashed. They also don't need a name until a
2698 * user wants to identify the object in /proc/pid/fd/. The little hack
2699 * below allows us to generate a name for these objects on demand:
2700 */
Jan Blunckcf28b482008-02-14 19:38:44 -08002701 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2702 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002703
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002704 get_fs_root(current->fs, &root);
Nick Piggin949854d2011-01-07 17:49:37 +11002705 write_seqlock(&rename_lock);
Al Viro02125a82011-12-05 08:43:34 -05002706 error = path_with_deleted(path, &root, &res, &buflen);
2707 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002708 res = ERR_PTR(error);
Nick Piggin949854d2011-01-07 17:49:37 +11002709 write_sequnlock(&rename_lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08002710 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 return res;
2712}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002713EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714
2715/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07002716 * Helper function for dentry_operations.d_dname() members
2717 */
2718char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2719 const char *fmt, ...)
2720{
2721 va_list args;
2722 char temp[64];
2723 int sz;
2724
2725 va_start(args, fmt);
2726 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2727 va_end(args);
2728
2729 if (sz > sizeof(temp) || sz > buflen)
2730 return ERR_PTR(-ENAMETOOLONG);
2731
2732 buffer += buflen - sz;
2733 return memcpy(buffer, temp, sz);
2734}
2735
2736/*
Ram Pai6092d042008-03-27 13:06:20 +01002737 * Write full pathname from the root of the filesystem into the buffer.
2738 */
Nick Pigginec2447c2011-01-07 17:49:29 +11002739static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01002740{
2741 char *end = buf + buflen;
2742 char *retval;
2743
Ram Pai6092d042008-03-27 13:06:20 +01002744 prepend(&end, &buflen, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01002745 if (buflen < 1)
2746 goto Elong;
2747 /* Get '/' right */
2748 retval = end-1;
2749 *retval = '/';
2750
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002751 while (!IS_ROOT(dentry)) {
2752 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11002753 int error;
Ram Pai6092d042008-03-27 13:06:20 +01002754
Ram Pai6092d042008-03-27 13:06:20 +01002755 prefetch(parent);
Nick Piggin9abca362011-01-07 17:49:36 +11002756 spin_lock(&dentry->d_lock);
2757 error = prepend_name(&end, &buflen, &dentry->d_name);
2758 spin_unlock(&dentry->d_lock);
2759 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
Ram Pai6092d042008-03-27 13:06:20 +01002760 goto Elong;
2761
2762 retval = end;
2763 dentry = parent;
2764 }
Al Viroc1031352010-06-06 22:31:14 -04002765 return retval;
2766Elong:
2767 return ERR_PTR(-ENAMETOOLONG);
2768}
Nick Pigginec2447c2011-01-07 17:49:29 +11002769
2770char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2771{
2772 char *retval;
2773
Nick Piggin949854d2011-01-07 17:49:37 +11002774 write_seqlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002775 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002776 write_sequnlock(&rename_lock);
Nick Pigginec2447c2011-01-07 17:49:29 +11002777
2778 return retval;
2779}
2780EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04002781
2782char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2783{
2784 char *p = NULL;
2785 char *retval;
2786
Nick Piggin949854d2011-01-07 17:49:37 +11002787 write_seqlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002788 if (d_unlinked(dentry)) {
2789 p = buf + buflen;
2790 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2791 goto Elong;
2792 buflen++;
2793 }
2794 retval = __dentry_path(dentry, buf, buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002795 write_sequnlock(&rename_lock);
Al Viroc1031352010-06-06 22:31:14 -04002796 if (!IS_ERR(retval) && p)
2797 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01002798 return retval;
2799Elong:
Ram Pai6092d042008-03-27 13:06:20 +01002800 return ERR_PTR(-ENAMETOOLONG);
2801}
2802
2803/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 * NOTE! The user-level library version returns a
2805 * character pointer. The kernel system call just
2806 * returns the length of the buffer filled (which
2807 * includes the ending '\0' character), or a negative
2808 * error value. So libc would do something like
2809 *
2810 * char *getcwd(char * buf, size_t size)
2811 * {
2812 * int retval;
2813 *
2814 * retval = sys_getcwd(buf, size);
2815 * if (retval >= 0)
2816 * return buf;
2817 * errno = -retval;
2818 * return NULL;
2819 * }
2820 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01002821SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822{
Linus Torvalds552ce542007-02-13 12:08:18 -08002823 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002824 struct path pwd, root;
Linus Torvalds552ce542007-02-13 12:08:18 -08002825 char *page = (char *) __get_free_page(GFP_USER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826
2827 if (!page)
2828 return -ENOMEM;
2829
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002830 get_fs_root_and_pwd(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831
Linus Torvalds552ce542007-02-13 12:08:18 -08002832 error = -ENOENT;
Nick Piggin949854d2011-01-07 17:49:37 +11002833 write_seqlock(&rename_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002834 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08002835 unsigned long len;
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002836 char *cwd = page + PAGE_SIZE;
2837 int buflen = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002839 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05002840 error = prepend_path(&pwd, &root, &cwd, &buflen);
Nick Piggin949854d2011-01-07 17:49:37 +11002841 write_sequnlock(&rename_lock);
Linus Torvalds552ce542007-02-13 12:08:18 -08002842
Al Viro02125a82011-12-05 08:43:34 -05002843 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08002844 goto out;
2845
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002846 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05002847 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02002848 error = prepend_unreachable(&cwd, &buflen);
2849 if (error)
2850 goto out;
2851 }
2852
Linus Torvalds552ce542007-02-13 12:08:18 -08002853 error = -ERANGE;
2854 len = PAGE_SIZE + page - cwd;
2855 if (len <= size) {
2856 error = len;
2857 if (copy_to_user(buf, cwd, len))
2858 error = -EFAULT;
2859 }
Nick Piggin949854d2011-01-07 17:49:37 +11002860 } else {
2861 write_sequnlock(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002862 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863
2864out:
Jan Blunck6ac08c32008-02-14 19:34:38 -08002865 path_put(&pwd);
2866 path_put(&root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 free_page((unsigned long) page);
2868 return error;
2869}
2870
2871/*
2872 * Test whether new_dentry is a subdirectory of old_dentry.
2873 *
2874 * Trivially implemented using the dcache structure
2875 */
2876
2877/**
2878 * is_subdir - is new dentry a subdirectory of old_dentry
2879 * @new_dentry: new dentry
2880 * @old_dentry: old dentry
2881 *
2882 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2883 * Returns 0 otherwise.
2884 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2885 */
2886
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002887int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888{
2889 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11002890 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002892 if (new_dentry == old_dentry)
2893 return 1;
2894
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002895 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11002898 /*
2899 * Need rcu_readlock to protect against the d_parent trashing
2900 * due to d_move
2901 */
2902 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002903 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002905 else
2906 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11002907 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909
2910 return result;
2911}
2912
2913void d_genocide(struct dentry *root)
2914{
Nick Piggin949854d2011-01-07 17:49:37 +11002915 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 struct list_head *next;
Nick Piggin949854d2011-01-07 17:49:37 +11002917 unsigned seq;
Nick Piggin58db63d2011-01-07 17:49:39 +11002918 int locked = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
Nick Piggin949854d2011-01-07 17:49:37 +11002920 seq = read_seqbegin(&rename_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002921again:
2922 this_parent = root;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002923 spin_lock(&this_parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924repeat:
2925 next = this_parent->d_subdirs.next;
2926resume:
2927 while (next != &this_parent->d_subdirs) {
2928 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002929 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 next = tmp->next;
Nick Piggin949854d2011-01-07 17:49:37 +11002931
Nick Pigginda502952011-01-07 17:49:33 +11002932 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2933 if (d_unhashed(dentry) || !dentry->d_inode) {
2934 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 continue;
Nick Pigginda502952011-01-07 17:49:33 +11002936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002938 spin_unlock(&this_parent->d_lock);
2939 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002941 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 goto repeat;
2943 }
Nick Piggin949854d2011-01-07 17:49:37 +11002944 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2945 dentry->d_flags |= DCACHE_GENOCIDE;
2946 dentry->d_count--;
2947 }
Nick Pigginb7ab39f2011-01-07 17:49:32 +11002948 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 }
2950 if (this_parent != root) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002951 struct dentry *child = this_parent;
Nick Piggin949854d2011-01-07 17:49:37 +11002952 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2953 this_parent->d_flags |= DCACHE_GENOCIDE;
2954 this_parent->d_count--;
2955 }
Linus Torvaldsc826cb72011-03-15 15:29:21 -07002956 this_parent = try_to_ascend(this_parent, locked, seq);
2957 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11002958 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11002959 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 goto resume;
2961 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002962 spin_unlock(&this_parent->d_lock);
Nick Piggin58db63d2011-01-07 17:49:39 +11002963 if (!locked && read_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11002964 goto rename_retry;
Nick Piggin58db63d2011-01-07 17:49:39 +11002965 if (locked)
2966 write_sequnlock(&rename_lock);
2967 return;
2968
2969rename_retry:
Miklos Szeredi8110e162012-09-17 22:23:30 +02002970 if (locked)
2971 goto again;
Nick Piggin58db63d2011-01-07 17:49:39 +11002972 locked = 1;
2973 write_seqlock(&rename_lock);
2974 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975}
2976
2977/**
2978 * find_inode_number - check for dentry with name
2979 * @dir: directory to check
2980 * @name: Name to find.
2981 *
2982 * Check whether a dentry already exists for the given name,
2983 * and return the inode number if it has an inode. Otherwise
2984 * 0 is returned.
2985 *
2986 * This routine is used to post-process directory listings for
2987 * filesystems using synthetic inode numbers, and is necessary
2988 * to keep getcwd() working.
2989 */
2990
2991ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2992{
2993 struct dentry * dentry;
2994 ino_t ino = 0;
2995
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002996 dentry = d_hash_and_lookup(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002997 if (!IS_ERR_OR_NULL(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 if (dentry->d_inode)
2999 ino = dentry->d_inode->i_ino;
3000 dput(dentry);
3001 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 return ino;
3003}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003004EXPORT_SYMBOL(find_inode_number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005
3006static __initdata unsigned long dhash_entries;
3007static int __init set_dhash_entries(char *str)
3008{
3009 if (!str)
3010 return 0;
3011 dhash_entries = simple_strtoul(str, &str, 0);
3012 return 1;
3013}
3014__setup("dhash_entries=", set_dhash_entries);
3015
3016static void __init dcache_init_early(void)
3017{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003018 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019
3020 /* If hashes are distributed across NUMA nodes, defer
3021 * hash allocation until vmalloc space is available.
3022 */
3023 if (hashdist)
3024 return;
3025
3026 dentry_hashtable =
3027 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003028 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 dhash_entries,
3030 13,
3031 HASH_EARLY,
3032 &d_hash_shift,
3033 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003034 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 0);
3036
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003037 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003038 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039}
3040
Denis Cheng74bf17c2007-10-16 23:26:30 -07003041static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003043 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
3045 /*
3046 * A constructor could be added for stable state like the lists,
3047 * but it is probably not worth it because of the cache nature
3048 * of the dcache.
3049 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003050 dentry_cache = KMEM_CACHE(dentry,
3051 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
3053 /* Hash may have been set up in dcache_init_early */
3054 if (!hashdist)
3055 return;
3056
3057 dentry_hashtable =
3058 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003059 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 dhash_entries,
3061 13,
3062 0,
3063 &d_hash_shift,
3064 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003065 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 0);
3067
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003068 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003069 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070}
3071
3072/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003073struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003074EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076EXPORT_SYMBOL(d_genocide);
3077
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078void __init vfs_caches_init_early(void)
3079{
3080 dcache_init_early();
3081 inode_init_early();
3082}
3083
3084void __init vfs_caches_init(unsigned long mempages)
3085{
3086 unsigned long reserve;
3087
3088 /* Base hash sizes on available memory, with a reserve equal to
3089 150% of current kernel size */
3090
3091 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3092 mempages -= reserve;
3093
3094 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003095 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096
Denis Cheng74bf17c2007-10-16 23:26:30 -07003097 dcache_init();
3098 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003100 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 bdev_cache_init();
3102 chrdev_init();
3103}