blob: 41000305d716ea51c47ed52ddb5abe024045e958 [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>
Dave Chinnerf6041562013-08-28 10:18:00 +100040#include <linux/list_lru.h>
David Howells07f3f052006-09-30 20:52:18 +020041#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050042#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Nick Piggin789680d2011-01-07 17:49:30 +110044/*
45 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110046 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110048 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100052 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110053 * - the dcache lru lists and counters
54 * d_lock protects:
55 * - d_flags
56 * - d_name
57 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110058 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110059 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110060 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
Nick Pigginb23fb0a2011-01-07 17:49:35 +110062 * - d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110063 *
64 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110065 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110066 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100067 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110068 * dcache_hash_bucket lock
69 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110070 *
Nick Pigginda502952011-01-07 17:49:33 +110071 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
73 * ...
74 * dentry->d_parent->d_lock
75 * dentry->d_lock
76 *
77 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110078 * if (dentry1 < dentry2)
79 * dentry1->d_lock
80 * dentry2->d_lock
81 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080082int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84
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
Waiman Long232d2d62013-09-09 12:18:13 -040091/**
92 * read_seqbegin_or_lock - begin a sequence number check or locking block
Waiman Long18129972013-09-12 10:55:35 -040093 * @lock: sequence lock
94 * @seq : sequence number to be checked
Waiman Long232d2d62013-09-09 12:18:13 -040095 *
96 * First try it once optimistically without taking the lock. If that fails,
97 * take the lock. The sequence number is also used as a marker for deciding
98 * whether to be a reader (even) or writer (odd).
99 * N.B. seq must be initialized to an even number to begin with.
100 */
101static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
102{
Al Viro48f5ec22013-09-09 15:22:25 -0400103 if (!(*seq & 1)) /* Even */
Waiman Long232d2d62013-09-09 12:18:13 -0400104 *seq = read_seqbegin(lock);
Al Viro48f5ec22013-09-09 15:22:25 -0400105 else /* Odd */
Waiman Long18129972013-09-12 10:55:35 -0400106 read_seqlock_excl(lock);
Waiman Long232d2d62013-09-09 12:18:13 -0400107}
108
Al Viro48f5ec22013-09-09 15:22:25 -0400109static inline int need_seqretry(seqlock_t *lock, int seq)
Waiman Long232d2d62013-09-09 12:18:13 -0400110{
Al Viro48f5ec22013-09-09 15:22:25 -0400111 return !(seq & 1) && read_seqretry(lock, seq);
112}
113
114static inline void done_seqretry(seqlock_t *lock, int seq)
115{
116 if (seq & 1)
Waiman Long18129972013-09-12 10:55:35 -0400117 read_sequnlock_excl(lock);
Waiman Long232d2d62013-09-09 12:18:13 -0400118}
119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120/*
121 * This is the single most critical data structure when it comes
122 * to the dcache: the hashtable for lookups. Somebody should try
123 * to make this good - I've just made it work.
124 *
125 * This hash-function tries to avoid losing too many bits of hash
126 * information, yet avoid using a prime hash-size or similar.
127 */
128#define D_HASHBITS d_hash_shift
129#define D_HASHMASK d_hash_mask
130
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800131static unsigned int d_hash_mask __read_mostly;
132static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100133
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700134static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100135
Linus Torvalds8966be92012-03-02 14:23:30 -0800136static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700137 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100138{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700139 hash += (unsigned long) parent / L1_CACHE_BYTES;
140 hash = hash + (hash >> D_HASHBITS);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100141 return dentry_hashtable + (hash & D_HASHMASK);
142}
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144/* Statistics gathering. */
145struct dentry_stat_t dentry_stat = {
146 .age_limit = 45,
147};
148
Glauber Costa3942c072013-08-28 10:17:53 +1000149static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000150static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400151
152#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000153
154/*
155 * Here we resort to our own counters instead of using generic per-cpu counters
156 * for consistency with what the vfs inode code does. We are expected to harvest
157 * better code and performance by having our own specialized counters.
158 *
159 * Please note that the loop is done over all possible CPUs, not over all online
160 * CPUs. The reason for this is that we don't want to play games with CPUs going
161 * on and off. If one of them goes off, we will just keep their counters.
162 *
163 * glommer: See cffbc8a for details, and if you ever intend to change this,
164 * please update all vfs counters to match.
165 */
Glauber Costa3942c072013-08-28 10:17:53 +1000166static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100167{
168 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000169 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100170 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry, i);
172 return sum < 0 ? 0 : sum;
173}
174
Dave Chinner62d36c72013-08-28 10:17:54 +1000175static long get_nr_dentry_unused(void)
176{
177 int i;
178 long sum = 0;
179 for_each_possible_cpu(i)
180 sum += per_cpu(nr_dentry_unused, i);
181 return sum < 0 ? 0 : sum;
182}
183
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400184int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
185 size_t *lenp, loff_t *ppos)
186{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100187 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000188 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000189 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400190}
191#endif
192
Linus Torvalds5483f182012-03-04 15:51:42 -0800193/*
194 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
195 * The strings are both count bytes long, and count is non-zero.
196 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700197#ifdef CONFIG_DCACHE_WORD_ACCESS
198
199#include <asm/word-at-a-time.h>
200/*
201 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
202 * aligned allocation for this particular component. We don't
203 * strictly need the load_unaligned_zeropad() safety, but it
204 * doesn't hurt either.
205 *
206 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
207 * need the careful unaligned handling.
208 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700209static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800210{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800211 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800212
213 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700214 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700215 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800216 if (tcount < sizeof(unsigned long))
217 break;
218 if (unlikely(a != b))
219 return 1;
220 cs += sizeof(unsigned long);
221 ct += sizeof(unsigned long);
222 tcount -= sizeof(unsigned long);
223 if (!tcount)
224 return 0;
225 }
226 mask = ~(~0ul << tcount*8);
227 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700228}
229
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800230#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700231
Linus Torvalds94753db52012-05-10 12:19:19 -0700232static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700233{
Linus Torvalds5483f182012-03-04 15:51:42 -0800234 do {
235 if (*cs != *ct)
236 return 1;
237 cs++;
238 ct++;
239 tcount--;
240 } while (tcount);
241 return 0;
242}
243
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700244#endif
245
Linus Torvalds94753db52012-05-10 12:19:19 -0700246static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
247{
Linus Torvalds6326c712012-05-21 16:14:04 -0700248 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700249 /*
250 * Be careful about RCU walk racing with rename:
251 * use ACCESS_ONCE to fetch the name pointer.
252 *
253 * NOTE! Even if a rename will mean that the length
254 * was not loaded atomically, we don't care. The
255 * RCU walk will check the sequence count eventually,
256 * and catch it. And we won't overrun the buffer,
257 * because we're reading the name pointer atomically,
258 * and a dentry name is guaranteed to be properly
259 * terminated with a NUL byte.
260 *
261 * End result: even if 'len' is wrong, we'll exit
262 * early because the data cannot match (there can
263 * be no NUL in the ct/tcount data)
264 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700265 cs = ACCESS_ONCE(dentry->d_name.name);
266 smp_read_barrier_depends();
267 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700268}
269
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400270static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400272 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
273
Al Virob3d9b7a2012-06-09 13:51:19 -0400274 WARN_ON(!hlist_unhashed(&dentry->d_alias));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (dname_external(dentry))
276 kfree(dentry->d_name.name);
277 kmem_cache_free(dentry_cache, dentry);
278}
279
280/*
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100281 * no locks, please.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 */
283static void d_free(struct dentry *dentry)
284{
Linus Torvalds0d984392013-09-08 13:46:52 -0700285 BUG_ON((int)dentry->d_lockref.count > 0);
Nick Piggin3e880fb2011-01-07 17:49:19 +1100286 this_cpu_dec(nr_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 if (dentry->d_op && dentry->d_op->d_release)
288 dentry->d_op->d_release(dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400289
Linus Torvaldsdea36672011-04-24 07:58:46 -0700290 /* if dentry was never visible to RCU, immediate free is OK */
291 if (!(dentry->d_flags & DCACHE_RCUACCESS))
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400292 __d_free(&dentry->d_u.d_rcu);
Eric Dumazetb3423412006-12-06 20:38:48 -0800293 else
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400294 call_rcu(&dentry->d_u.d_rcu, __d_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
Nick Piggin31e6b012011-01-07 17:49:52 +1100297/**
298 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800299 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100300 * After this call, in-progress rcu-walk path lookup will fail. This
301 * should be called after unhashing, and after changing d_inode (if
302 * the dentry has not already been unhashed).
303 */
304static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
305{
306 assert_spin_locked(&dentry->d_lock);
307 /* Go through a barrier */
308 write_seqcount_barrier(&dentry->d_seq);
309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311/*
312 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100313 * d_iput() operation if defined. Dentry has no refcount
314 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800316static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200317 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100318 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
320 struct inode *inode = dentry->d_inode;
321 if (inode) {
322 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400323 hlist_del_init(&dentry->d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100325 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700326 if (!inode->i_nlink)
327 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 if (dentry->d_op && dentry->d_op->d_iput)
329 dentry->d_op->d_iput(dentry, inode);
330 else
331 iput(inode);
332 } else {
333 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 }
335}
336
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700337/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100338 * Release the dentry's inode, using the filesystem
339 * d_iput() operation if defined. dentry remains in-use.
340 */
341static void dentry_unlink_inode(struct dentry * dentry)
342 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100343 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100344{
345 struct inode *inode = dentry->d_inode;
346 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400347 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100348 dentry_rcuwalk_barrier(dentry);
349 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100350 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100351 if (!inode->i_nlink)
352 fsnotify_inoderemove(inode);
353 if (dentry->d_op && dentry->d_op->d_iput)
354 dentry->d_op->d_iput(dentry, inode);
355 else
356 iput(inode);
357}
358
359/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400360 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
361 * is in use - which includes both the "real" per-superblock
362 * LRU list _and_ the DCACHE_SHRINK_LIST use.
363 *
364 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
365 * on the shrink list (ie not on the superblock LRU list).
366 *
367 * The per-cpu "nr_dentry_unused" counters are updated with
368 * the DCACHE_LRU_LIST bit.
369 *
370 * These helper functions make sure we always follow the
371 * rules. d_lock must be held by the caller.
372 */
373#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
374static void d_lru_add(struct dentry *dentry)
375{
376 D_FLAG_VERIFY(dentry, 0);
377 dentry->d_flags |= DCACHE_LRU_LIST;
378 this_cpu_inc(nr_dentry_unused);
379 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
380}
381
382static void d_lru_del(struct dentry *dentry)
383{
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags &= ~DCACHE_LRU_LIST;
386 this_cpu_dec(nr_dentry_unused);
387 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
388}
389
390static void d_shrink_del(struct dentry *dentry)
391{
392 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
393 list_del_init(&dentry->d_lru);
394 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
395 this_cpu_dec(nr_dentry_unused);
396}
397
398static void d_shrink_add(struct dentry *dentry, struct list_head *list)
399{
400 D_FLAG_VERIFY(dentry, 0);
401 list_add(&dentry->d_lru, list);
402 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
403 this_cpu_inc(nr_dentry_unused);
404}
405
406/*
407 * These can only be called under the global LRU lock, ie during the
408 * callback for freeing the LRU list. "isolate" removes it from the
409 * LRU lists entirely, while shrink_move moves it to the indicated
410 * private list.
411 */
412static void d_lru_isolate(struct dentry *dentry)
413{
414 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
415 dentry->d_flags &= ~DCACHE_LRU_LIST;
416 this_cpu_dec(nr_dentry_unused);
417 list_del_init(&dentry->d_lru);
418}
419
420static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
421{
422 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
423 dentry->d_flags |= DCACHE_SHRINK_LIST;
424 list_move_tail(&dentry->d_lru, list);
425}
426
427/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000428 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700429 */
430static void dentry_lru_add(struct dentry *dentry)
431{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400432 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
433 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700434}
435
Sage Weilf0023bc2011-10-28 10:02:42 -0700436/*
437 * Remove a dentry with references from the LRU.
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000438 *
439 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
440 * lose our last reference through the parent walk. In this case, we need to
441 * remove ourselves from the shrink list, not the LRU.
Sage Weilf0023bc2011-10-28 10:02:42 -0700442 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700443static void dentry_lru_del(struct dentry *dentry)
444{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400445 if (dentry->d_flags & DCACHE_LRU_LIST) {
446 if (dentry->d_flags & DCACHE_SHRINK_LIST)
447 return d_shrink_del(dentry);
448 d_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700449 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700450}
451
Miklos Szeredid52b9082007-05-08 00:23:46 -0700452/**
453 * d_kill - kill dentry and return parent
454 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800455 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700456 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200457 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700458 *
459 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100460 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100461 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
462 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700463 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100464static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200465 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100466 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100467 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700468{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700469 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400470 /*
471 * Inform try_to_ascend() that we are no longer attached to the
472 * dentry tree
473 */
Miklos Szeredib161dfa62012-09-17 22:31:38 +0200474 dentry->d_flags |= DCACHE_DENTRY_KILLED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100475 if (parent)
476 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700477 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100478 /*
479 * dentry_iput drops the locks, at which point nobody (except
480 * transient RCU lookups) can reach this dentry.
481 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700482 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900483 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700484}
485
David Howellsc6627c62011-06-07 14:09:20 +0100486/*
487 * Unhash a dentry without inserting an RCU walk barrier or checking that
488 * dentry->d_lock is locked. The caller must take care of that, if
489 * appropriate.
490 */
491static void __d_shrink(struct dentry *dentry)
492{
493 if (!d_unhashed(dentry)) {
494 struct hlist_bl_head *b;
495 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
496 b = &dentry->d_sb->s_anon;
497 else
498 b = d_hash(dentry->d_parent, dentry->d_name.hash);
499
500 hlist_bl_lock(b);
501 __hlist_bl_del(&dentry->d_hash);
502 dentry->d_hash.pprev = NULL;
503 hlist_bl_unlock(b);
504 }
505}
506
Nick Piggin789680d2011-01-07 17:49:30 +1100507/**
508 * d_drop - drop a dentry
509 * @dentry: dentry to drop
510 *
511 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
512 * be found through a VFS lookup any more. Note that this is different from
513 * deleting the dentry - d_delete will try to mark the dentry negative if
514 * possible, giving a successful _negative_ lookup, while d_drop will
515 * just make the cache lookup fail.
516 *
517 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
518 * reason (NFS timeouts or autofs deletes).
519 *
520 * __d_drop requires dentry->d_lock.
521 */
522void __d_drop(struct dentry *dentry)
523{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700524 if (!d_unhashed(dentry)) {
David Howellsc6627c62011-06-07 14:09:20 +0100525 __d_shrink(dentry);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700526 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100527 }
528}
529EXPORT_SYMBOL(__d_drop);
530
531void d_drop(struct dentry *dentry)
532{
Nick Piggin789680d2011-01-07 17:49:30 +1100533 spin_lock(&dentry->d_lock);
534 __d_drop(dentry);
535 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100536}
537EXPORT_SYMBOL(d_drop);
538
Nick Piggin77812a12011-01-07 17:49:48 +1100539/*
540 * Finish off a dentry we've decided to kill.
541 * dentry->d_lock must be held, returns with it unlocked.
542 * If ref is non-zero, then decrement the refcount too.
543 * Returns dentry requiring refcount drop, or NULL if we're done.
544 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000545static inline struct dentry *
546dentry_kill(struct dentry *dentry, int unlock_on_failure)
Nick Piggin77812a12011-01-07 17:49:48 +1100547 __releases(dentry->d_lock)
548{
Nick Piggin873feea2011-01-07 17:50:06 +1100549 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100550 struct dentry *parent;
551
Nick Piggin873feea2011-01-07 17:50:06 +1100552 inode = dentry->d_inode;
553 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100554relock:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000555 if (unlock_on_failure) {
556 spin_unlock(&dentry->d_lock);
557 cpu_relax();
558 }
Nick Piggin77812a12011-01-07 17:49:48 +1100559 return dentry; /* try again with same dentry */
560 }
561 if (IS_ROOT(dentry))
562 parent = NULL;
563 else
564 parent = dentry->d_parent;
565 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100566 if (inode)
567 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100568 goto relock;
569 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100570
Linus Torvalds0d984392013-09-08 13:46:52 -0700571 /*
572 * The dentry is now unrecoverably dead to the world.
573 */
574 lockref_mark_dead(&dentry->d_lockref);
575
Sage Weilf0023bc2011-10-28 10:02:42 -0700576 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700577 * inform the fs via d_prune that this dentry is about to be
578 * unhashed and destroyed.
579 */
Yan, Zheng590fb512013-08-13 15:42:02 +0800580 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +0800581 dentry->d_op->d_prune(dentry);
582
583 dentry_lru_del(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100584 /* if it was on the hash then remove it */
585 __d_drop(dentry);
586 return d_kill(dentry, parent);
587}
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589/*
590 * This is dput
591 *
592 * This is complicated by the fact that we do not want to put
593 * dentries that are no longer on any hash chain on the unused
594 * list: we'd much rather just get rid of them immediately.
595 *
596 * However, that implies that we have to traverse the dentry
597 * tree upwards to the parents which might _also_ now be
598 * scheduled for deletion (it may have been only waiting for
599 * its last child to go away).
600 *
601 * This tail recursion is done by hand as we don't want to depend
602 * on the compiler to always get this right (gcc generally doesn't).
603 * Real recursion would eat up our stack space.
604 */
605
606/*
607 * dput - release a dentry
608 * @dentry: dentry to release
609 *
610 * Release a dentry. This will drop the usage count and if appropriate
611 * call the dentry unlink method as well as removing it from the queues and
612 * releasing its resources. If the parent dentries were scheduled for release
613 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615void dput(struct dentry *dentry)
616{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700617 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 return;
619
620repeat:
Waiman Long98474232013-08-28 18:24:59 -0700621 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700624 /* Unreachable? Get rid of it */
625 if (unlikely(d_unhashed(dentry)))
626 goto kill_it;
627
628 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100630 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 }
Nick Piggin265ac902010-10-10 05:36:24 -0400632
Jeff Layton39e3c952012-11-28 11:30:53 -0500633 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400634 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400635
Waiman Long98474232013-08-28 18:24:59 -0700636 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100637 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 return;
639
Miklos Szeredid52b9082007-05-08 00:23:46 -0700640kill_it:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000641 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700642 if (dentry)
643 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700645EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647/**
648 * d_invalidate - invalidate a dentry
649 * @dentry: dentry to invalidate
650 *
651 * Try to invalidate the dentry if it turns out to be
652 * possible. If there are other dentries that can be
653 * reached through this one we can't delete it and we
654 * return -EBUSY. On success we return 0.
655 *
656 * no dcache lock.
657 */
658
659int d_invalidate(struct dentry * dentry)
660{
661 /*
662 * If it's already been dropped, return OK.
663 */
Nick Pigginda502952011-01-07 17:49:33 +1100664 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100666 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 return 0;
668 }
669 /*
670 * Check whether to do a partial shrink_dcache
671 * to get rid of unused child entries.
672 */
673 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100674 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100676 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
678
679 /*
680 * Somebody else still using it?
681 *
682 * If it's a directory, we can't drop it
683 * for fear of somebody re-populating it
684 * with children (even though dropping it
685 * would make it unreachable from the root,
686 * we might still populate it if it was a
687 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000688 * We also need to leave mountpoints alone,
689 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 */
Waiman Long98474232013-08-28 18:24:59 -0700691 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000692 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 return -EBUSY;
695 }
696 }
697
698 __d_drop(dentry);
699 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return 0;
701}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700702EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100704/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100705static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706{
Waiman Long98474232013-08-28 18:24:59 -0700707 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
Nick Piggindc0474b2011-01-07 17:49:43 +1100710static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100711{
Waiman Long98474232013-08-28 18:24:59 -0700712 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100713}
714
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100715struct dentry *dget_parent(struct dentry *dentry)
716{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700717 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100718 struct dentry *ret;
719
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700720 /*
721 * Do optimistic parent lookup without any
722 * locking.
723 */
724 rcu_read_lock();
725 ret = ACCESS_ONCE(dentry->d_parent);
726 gotref = lockref_get_not_zero(&ret->d_lockref);
727 rcu_read_unlock();
728 if (likely(gotref)) {
729 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
730 return ret;
731 dput(ret);
732 }
733
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100734repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100735 /*
736 * Don't need rcu_dereference because we re-check it was correct under
737 * the lock.
738 */
739 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100740 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100741 spin_lock(&ret->d_lock);
742 if (unlikely(ret != dentry->d_parent)) {
743 spin_unlock(&ret->d_lock);
744 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100745 goto repeat;
746 }
Nick Piggina734eb42011-01-07 17:49:44 +1100747 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700748 BUG_ON(!ret->d_lockref.count);
749 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100750 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100751 return ret;
752}
753EXPORT_SYMBOL(dget_parent);
754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755/**
756 * d_find_alias - grab a hashed alias of inode
757 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700758 * @want_discon: flag, used by d_splice_alias, to request
759 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 *
761 * If inode has a hashed alias, or is a directory and has any alias,
762 * acquire the reference to alias and return it. Otherwise return NULL.
763 * Notice that if inode is a directory there can be only one alias and
764 * it can be unhashed only if it has no children, or if it is the root
765 * of a filesystem.
766 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700767 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700768 * any other hashed alias over that one unless @want_discon is set,
769 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700771static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772{
Nick Pigginda502952011-01-07 17:49:33 +1100773 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Nick Pigginda502952011-01-07 17:49:33 +1100775again:
776 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800777 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100778 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700780 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100781 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700783 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100784 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100785 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 return alias;
787 }
788 }
Nick Pigginda502952011-01-07 17:49:33 +1100789 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
Nick Pigginda502952011-01-07 17:49:33 +1100791 if (discon_alias) {
792 alias = discon_alias;
793 spin_lock(&alias->d_lock);
794 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
795 if (IS_ROOT(alias) &&
796 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100797 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100798 spin_unlock(&alias->d_lock);
799 return alias;
800 }
801 }
802 spin_unlock(&alias->d_lock);
803 goto again;
804 }
805 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
Nick Pigginda502952011-01-07 17:49:33 +1100808struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809{
David Howells214fda12006-03-25 03:06:36 -0800810 struct dentry *de = NULL;
811
Al Virob3d9b7a2012-06-09 13:51:19 -0400812 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100813 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700814 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100815 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return de;
818}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700819EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821/*
822 * Try to kill dentries associated with this inode.
823 * WARNING: you must own a reference to inode.
824 */
825void d_prune_aliases(struct inode *inode)
826{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700827 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100829 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800830 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700832 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800833 /*
834 * inform the fs via d_prune that this dentry
835 * is about to be unhashed and destroyed.
836 */
837 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
838 !d_unhashed(dentry))
839 dentry->d_op->d_prune(dentry);
840
Nick Piggindc0474b2011-01-07 17:49:43 +1100841 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 __d_drop(dentry);
843 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100844 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 dput(dentry);
846 goto restart;
847 }
848 spin_unlock(&dentry->d_lock);
849 }
Nick Piggin873feea2011-01-07 17:50:06 +1100850 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700852EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854/*
Nick Piggin77812a12011-01-07 17:49:48 +1100855 * Try to throw away a dentry - free the inode, dput the parent.
856 * Requires dentry->d_lock is held, and dentry->d_count == 0.
857 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700858 *
Nick Piggin77812a12011-01-07 17:49:48 +1100859 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000861static struct dentry * try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200862 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Nick Piggin77812a12011-01-07 17:49:48 +1100864 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700865
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000866 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700867 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100868 * If dentry_kill returns NULL, we have nothing more to do.
869 * if it returns the same dentry, trylocks failed. In either
870 * case, just loop again.
871 *
872 * Otherwise, we need to prune ancestors too. This is necessary
873 * to prevent quadratic behavior of shrink_dcache_parent(), but
874 * is also expected to be beneficial in reducing dentry cache
875 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700876 */
Nick Piggin77812a12011-01-07 17:49:48 +1100877 if (!parent)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000878 return NULL;
Nick Piggin77812a12011-01-07 17:49:48 +1100879 if (parent == dentry)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000880 return dentry;
Nick Piggin77812a12011-01-07 17:49:48 +1100881
882 /* Prune ancestors. */
883 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700884 while (dentry) {
Waiman Long98474232013-08-28 18:24:59 -0700885 if (lockref_put_or_lock(&dentry->d_lockref))
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000886 return NULL;
887 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700888 }
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000889 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890}
891
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400892static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700894 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700895
Nick Pigginec336792011-01-07 17:49:47 +1100896 rcu_read_lock();
897 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100898 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
899 if (&dentry->d_lru == list)
900 break; /* empty */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400901
902 /*
903 * Get the dentry lock, and re-verify that the dentry is
904 * this on the shrinking list. If it is, we know that
905 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
906 */
Nick Pigginec336792011-01-07 17:49:47 +1100907 spin_lock(&dentry->d_lock);
908 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
909 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100910 continue;
911 }
912
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000914 * The dispose list is isolated and dentries are not accounted
915 * to the LRU here, so we can simply remove it from the list
916 * here regardless of whether it is referenced or not.
917 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400918 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000919
920 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000922 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 */
Waiman Long98474232013-08-28 18:24:59 -0700924 if (dentry->d_lockref.count) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700925 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 continue;
927 }
Nick Pigginec336792011-01-07 17:49:47 +1100928 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100929
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400930 /*
931 * If 'try_to_prune()' returns a dentry, it will
932 * be the same one we passed in, and d_lock will
933 * have been held the whole time, so it will not
934 * have been added to any other lists. We failed
935 * to get the inode lock.
936 *
937 * We just add it back to the shrink list.
938 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000939 dentry = try_prune_one_dentry(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100940
Nick Pigginec336792011-01-07 17:49:47 +1100941 rcu_read_lock();
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000942 if (dentry) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400943 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000944 spin_unlock(&dentry->d_lock);
945 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
Nick Pigginec336792011-01-07 17:49:47 +1100947 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400948}
949
Dave Chinnerf6041562013-08-28 10:18:00 +1000950static enum lru_status
951dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
952{
953 struct list_head *freeable = arg;
954 struct dentry *dentry = container_of(item, struct dentry, d_lru);
955
956
957 /*
958 * we are inverting the lru lock/dentry->d_lock here,
959 * so use a trylock. If we fail to get the lock, just skip
960 * it
961 */
962 if (!spin_trylock(&dentry->d_lock))
963 return LRU_SKIP;
964
965 /*
966 * Referenced dentries are still in use. If they have active
967 * counts, just remove them from the LRU. Otherwise give them
968 * another pass through the LRU.
969 */
970 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400971 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000972 spin_unlock(&dentry->d_lock);
973 return LRU_REMOVED;
974 }
975
976 if (dentry->d_flags & DCACHE_REFERENCED) {
977 dentry->d_flags &= ~DCACHE_REFERENCED;
978 spin_unlock(&dentry->d_lock);
979
980 /*
981 * The list move itself will be made by the common LRU code. At
982 * this point, we've dropped the dentry->d_lock but keep the
983 * lru lock. This is safe to do, since every list movement is
984 * protected by the lru lock even if both locks are held.
985 *
986 * This is guaranteed by the fact that all LRU management
987 * functions are intermediated by the LRU API calls like
988 * list_lru_add and list_lru_del. List movement in this file
989 * only ever occur through this functions or through callbacks
990 * like this one, that are called from the LRU API.
991 *
992 * The only exceptions to this are functions like
993 * shrink_dentry_list, and code that first checks for the
994 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
995 * operating only with stack provided lists after they are
996 * properly isolated from the main list. It is thus, always a
997 * local access.
998 */
999 return LRU_ROTATE;
1000 }
1001
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001002 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001003 spin_unlock(&dentry->d_lock);
1004
1005 return LRU_REMOVED;
1006}
1007
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001008/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001009 * prune_dcache_sb - shrink the dcache
1010 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +10001011 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +10001012 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001013 *
Dave Chinnerf6041562013-08-28 10:18:00 +10001014 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001015 * done when we need more memory an called from the superblock shrinker
1016 * function.
1017 *
1018 * This function may fail to free any resources if all the dentries are in
1019 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001020 */
Dave Chinner9b17c622013-08-28 10:18:05 +10001021long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
1022 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001023{
Dave Chinnerf6041562013-08-28 10:18:00 +10001024 LIST_HEAD(dispose);
1025 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001026
Dave Chinner9b17c622013-08-28 10:18:05 +10001027 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
1028 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +10001029 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001030 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
1032
Glauber Costa4e717f52013-08-28 10:18:03 +10001033static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1034 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001035{
Glauber Costa4e717f52013-08-28 10:18:03 +10001036 struct list_head *freeable = arg;
1037 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001038
Glauber Costa4e717f52013-08-28 10:18:03 +10001039 /*
1040 * we are inverting the lru lock/dentry->d_lock here,
1041 * so use a trylock. If we fail to get the lock, just skip
1042 * it
1043 */
1044 if (!spin_trylock(&dentry->d_lock))
1045 return LRU_SKIP;
1046
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001047 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001048 spin_unlock(&dentry->d_lock);
1049
1050 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001051}
1052
Glauber Costa4e717f52013-08-28 10:18:03 +10001053
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001054/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 * shrink_dcache_sb - shrink dcache for a superblock
1056 * @sb: superblock
1057 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001058 * Shrink the dcache for the specified super block. This is used to free
1059 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001061void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
Glauber Costa4e717f52013-08-28 10:18:03 +10001063 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001064
Glauber Costa4e717f52013-08-28 10:18:03 +10001065 do {
1066 LIST_HEAD(dispose);
1067
1068 freed = list_lru_walk(&sb->s_dentry_lru,
1069 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1070
1071 this_cpu_sub(nr_dentry_unused, freed);
1072 shrink_dentry_list(&dispose);
1073 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001075EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077/*
David Howellsc636ebd2006-10-11 01:22:19 -07001078 * destroy a single subtree of dentries for unmount
1079 * - see the comments on shrink_dcache_for_umount() for a description of the
1080 * locking
1081 */
1082static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
1083{
1084 struct dentry *parent;
1085
1086 BUG_ON(!IS_ROOT(dentry));
1087
David Howellsc636ebd2006-10-11 01:22:19 -07001088 for (;;) {
1089 /* descend to the first leaf in the current subtree */
David Howells43c1c9c2011-06-07 14:09:30 +01001090 while (!list_empty(&dentry->d_subdirs))
David Howellsc636ebd2006-10-11 01:22:19 -07001091 dentry = list_entry(dentry->d_subdirs.next,
1092 struct dentry, d_u.d_child);
David Howellsc636ebd2006-10-11 01:22:19 -07001093
1094 /* consume the dentries from this leaf up through its parents
1095 * until we find one with children or run out altogether */
1096 do {
1097 struct inode *inode;
1098
Sage Weilf0023bc2011-10-28 10:02:42 -07001099 /*
Yan, Zheng61572bb2013-04-15 14:13:21 +08001100 * inform the fs that this dentry is about to be
Sage Weilf0023bc2011-10-28 10:02:42 -07001101 * unhashed and destroyed.
1102 */
Yan, Zheng590fb512013-08-13 15:42:02 +08001103 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
1104 !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +08001105 dentry->d_op->d_prune(dentry);
1106
1107 dentry_lru_del(dentry);
David Howells43c1c9c2011-06-07 14:09:30 +01001108 __d_shrink(dentry);
1109
Waiman Long98474232013-08-28 18:24:59 -07001110 if (dentry->d_lockref.count != 0) {
David Howellsc636ebd2006-10-11 01:22:19 -07001111 printk(KERN_ERR
1112 "BUG: Dentry %p{i=%lx,n=%s}"
1113 " still in use (%d)"
1114 " [unmount of %s %s]\n",
1115 dentry,
1116 dentry->d_inode ?
1117 dentry->d_inode->i_ino : 0UL,
1118 dentry->d_name.name,
Waiman Long98474232013-08-28 18:24:59 -07001119 dentry->d_lockref.count,
David Howellsc636ebd2006-10-11 01:22:19 -07001120 dentry->d_sb->s_type->name,
1121 dentry->d_sb->s_id);
1122 BUG();
1123 }
1124
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001125 if (IS_ROOT(dentry)) {
David Howellsc636ebd2006-10-11 01:22:19 -07001126 parent = NULL;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001127 list_del(&dentry->d_u.d_child);
1128 } else {
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001129 parent = dentry->d_parent;
Waiman Long98474232013-08-28 18:24:59 -07001130 parent->d_lockref.count--;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001131 list_del(&dentry->d_u.d_child);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09001132 }
David Howellsc636ebd2006-10-11 01:22:19 -07001133
David Howellsc636ebd2006-10-11 01:22:19 -07001134 inode = dentry->d_inode;
1135 if (inode) {
1136 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001137 hlist_del_init(&dentry->d_alias);
David Howellsc636ebd2006-10-11 01:22:19 -07001138 if (dentry->d_op && dentry->d_op->d_iput)
1139 dentry->d_op->d_iput(dentry, inode);
1140 else
1141 iput(inode);
1142 }
1143
1144 d_free(dentry);
1145
1146 /* finished when we fall off the top of the tree,
1147 * otherwise we ascend to the parent and move to the
1148 * next sibling if there is one */
1149 if (!parent)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001150 return;
David Howellsc636ebd2006-10-11 01:22:19 -07001151 dentry = parent;
David Howellsc636ebd2006-10-11 01:22:19 -07001152 } while (list_empty(&dentry->d_subdirs));
1153
1154 dentry = list_entry(dentry->d_subdirs.next,
1155 struct dentry, d_u.d_child);
1156 }
1157}
1158
1159/*
1160 * destroy the dentries attached to a superblock on unmounting
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001161 * - we don't need to use dentry->d_lock because:
David Howellsc636ebd2006-10-11 01:22:19 -07001162 * - the superblock is detached from all mountings and open files, so the
1163 * dentry trees will not be rearranged by the VFS
1164 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1165 * any dentries belonging to this superblock that it comes across
1166 * - the filesystem itself is no longer permitted to rearrange the dentries
1167 * in this superblock
1168 */
1169void shrink_dcache_for_umount(struct super_block *sb)
1170{
1171 struct dentry *dentry;
1172
1173 if (down_read_trylock(&sb->s_umount))
1174 BUG();
1175
1176 dentry = sb->s_root;
1177 sb->s_root = NULL;
Waiman Long98474232013-08-28 18:24:59 -07001178 dentry->d_lockref.count--;
David Howellsc636ebd2006-10-11 01:22:19 -07001179 shrink_dcache_for_umount_subtree(dentry);
1180
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001181 while (!hlist_bl_empty(&sb->s_anon)) {
1182 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
David Howellsc636ebd2006-10-11 01:22:19 -07001183 shrink_dcache_for_umount_subtree(dentry);
1184 }
1185}
1186
1187/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001188 * This tries to ascend one level of parenthood, but
1189 * we can race with renaming, so we need to re-check
1190 * the parenthood after dropping the lock and check
1191 * that the sequence number still matches.
1192 */
Al Viro48f5ec22013-09-09 15:22:25 -04001193static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001194{
1195 struct dentry *new = old->d_parent;
1196
1197 rcu_read_lock();
1198 spin_unlock(&old->d_lock);
1199 spin_lock(&new->d_lock);
1200
1201 /*
1202 * might go back up the wrong parent if we have had a rename
1203 * or deletion
1204 */
1205 if (new != old->d_parent ||
Miklos Szeredib161dfa62012-09-17 22:31:38 +02001206 (old->d_flags & DCACHE_DENTRY_KILLED) ||
Al Viro48f5ec22013-09-09 15:22:25 -04001207 need_seqretry(&rename_lock, seq)) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001208 spin_unlock(&new->d_lock);
1209 new = NULL;
1210 }
1211 rcu_read_unlock();
1212 return new;
1213}
1214
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001216 * enum d_walk_ret - action to talke during tree walk
1217 * @D_WALK_CONTINUE: contrinue walk
1218 * @D_WALK_QUIT: quit walk
1219 * @D_WALK_NORETRY: quit when retry is needed
1220 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001222enum d_walk_ret {
1223 D_WALK_CONTINUE,
1224 D_WALK_QUIT,
1225 D_WALK_NORETRY,
1226 D_WALK_SKIP,
1227};
1228
1229/**
1230 * d_walk - walk the dentry tree
1231 * @parent: start of walk
1232 * @data: data passed to @enter() and @finish()
1233 * @enter: callback when first entering the dentry
1234 * @finish: callback when successfully finished the walk
1235 *
1236 * The @enter() and @finish() callbacks are called with d_lock held.
1237 */
1238static void d_walk(struct dentry *parent, void *data,
1239 enum d_walk_ret (*enter)(void *, struct dentry *),
1240 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
Nick Piggin949854d2011-01-07 17:49:37 +11001242 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001244 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001245 enum d_walk_ret ret;
1246 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001247
Nick Piggin58db63d2011-01-07 17:49:39 +11001248again:
Al Viro48f5ec22013-09-09 15:22:25 -04001249 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001250 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001251 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001252
1253 ret = enter(data, this_parent);
1254 switch (ret) {
1255 case D_WALK_CONTINUE:
1256 break;
1257 case D_WALK_QUIT:
1258 case D_WALK_SKIP:
1259 goto out_unlock;
1260 case D_WALK_NORETRY:
1261 retry = false;
1262 break;
1263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264repeat:
1265 next = this_parent->d_subdirs.next;
1266resume:
1267 while (next != &this_parent->d_subdirs) {
1268 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001269 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001271
1272 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001273
1274 ret = enter(data, dentry);
1275 switch (ret) {
1276 case D_WALK_CONTINUE:
1277 break;
1278 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001279 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001280 goto out_unlock;
1281 case D_WALK_NORETRY:
1282 retry = false;
1283 break;
1284 case D_WALK_SKIP:
1285 spin_unlock(&dentry->d_lock);
1286 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001287 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001290 spin_unlock(&this_parent->d_lock);
1291 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001293 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 goto repeat;
1295 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001296 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 }
1298 /*
1299 * All done at this level ... ascend and resume the search.
1300 */
1301 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001302 struct dentry *child = this_parent;
Al Viro48f5ec22013-09-09 15:22:25 -04001303 this_parent = try_to_ascend(this_parent, seq);
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001304 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001305 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001306 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 goto resume;
1308 }
Al Viro48f5ec22013-09-09 15:22:25 -04001309 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001310 spin_unlock(&this_parent->d_lock);
1311 goto rename_retry;
1312 }
1313 if (finish)
1314 finish(data);
1315
1316out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001317 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001318 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001319 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001320
1321rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001322 if (!retry)
1323 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001324 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001325 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001327
1328/*
1329 * Search for at least 1 mount point in the dentry's subdirs.
1330 * We descend to the next level whenever the d_subdirs
1331 * list is non-empty and continue searching.
1332 */
1333
1334/**
1335 * have_submounts - check for mounts over a dentry
1336 * @parent: dentry to check.
1337 *
1338 * Return true if the parent or its subdirectories contain
1339 * a mount point
1340 */
1341
1342static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1343{
1344 int *ret = data;
1345 if (d_mountpoint(dentry)) {
1346 *ret = 1;
1347 return D_WALK_QUIT;
1348 }
1349 return D_WALK_CONTINUE;
1350}
1351
1352int have_submounts(struct dentry *parent)
1353{
1354 int ret = 0;
1355
1356 d_walk(parent, &ret, check_mount, NULL);
1357
1358 return ret;
1359}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001360EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
1362/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001363 * Called by mount code to set a mountpoint and check if the mountpoint is
1364 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1365 * subtree can become unreachable).
1366 *
1367 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1368 * this reason take rename_lock and d_lock on dentry and ancestors.
1369 */
1370int d_set_mounted(struct dentry *dentry)
1371{
1372 struct dentry *p;
1373 int ret = -ENOENT;
1374 write_seqlock(&rename_lock);
1375 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1376 /* Need exclusion wrt. check_submounts_and_drop() */
1377 spin_lock(&p->d_lock);
1378 if (unlikely(d_unhashed(p))) {
1379 spin_unlock(&p->d_lock);
1380 goto out;
1381 }
1382 spin_unlock(&p->d_lock);
1383 }
1384 spin_lock(&dentry->d_lock);
1385 if (!d_unlinked(dentry)) {
1386 dentry->d_flags |= DCACHE_MOUNTED;
1387 ret = 0;
1388 }
1389 spin_unlock(&dentry->d_lock);
1390out:
1391 write_sequnlock(&rename_lock);
1392 return ret;
1393}
1394
1395/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001396 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 * and move any unused dentries to the end of the unused
1398 * list for prune_dcache(). We descend to the next level
1399 * whenever the d_subdirs list is non-empty and continue
1400 * searching.
1401 *
1402 * It returns zero iff there are no unused children,
1403 * otherwise it returns the number of children moved to
1404 * the end of the unused list. This may not be the total
1405 * number of unused children, because select_parent can
1406 * drop the lock and return early due to latency
1407 * constraints.
1408 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001409
1410struct select_data {
1411 struct dentry *start;
1412 struct list_head dispose;
1413 int found;
1414};
1415
1416static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001418 struct select_data *data = _data;
1419 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001421 if (data->start == dentry)
1422 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001424 /*
1425 * move only zero ref count dentries to the dispose list.
1426 *
1427 * Those which are presently on the shrink list, being processed
1428 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1429 * loop in shrink_dcache_parent() might not make any progress
1430 * and loop forever.
1431 */
1432 if (dentry->d_lockref.count) {
1433 dentry_lru_del(dentry);
1434 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001435 /*
1436 * We can't use d_lru_shrink_move() because we
1437 * need to get the global LRU lock and do the
Linus Torvalds05a82522013-09-15 07:11:01 -04001438 * LRU accounting.
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001439 */
1440 d_lru_del(dentry);
1441 d_shrink_add(dentry, &data->dispose);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001442 data->found++;
1443 ret = D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 }
1445 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001446 * We can return to the caller if we have found some (this
1447 * ensures forward progress). We'll be coming back to find
1448 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001450 if (data->found && need_resched())
1451 ret = D_WALK_QUIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001453 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454}
1455
1456/**
1457 * shrink_dcache_parent - prune dcache
1458 * @parent: parent of entries to prune
1459 *
1460 * Prune the dcache to remove unused children of the parent dentry.
1461 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001462void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001464 for (;;) {
1465 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001467 INIT_LIST_HEAD(&data.dispose);
1468 data.start = parent;
1469 data.found = 0;
1470
1471 d_walk(parent, &data, select_collect, NULL);
1472 if (!data.found)
1473 break;
1474
1475 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001476 cond_resched();
1477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001479EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001481static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1482{
1483 struct select_data *data = _data;
1484
1485 if (d_mountpoint(dentry)) {
1486 data->found = -EBUSY;
1487 return D_WALK_QUIT;
1488 }
1489
1490 return select_collect(_data, dentry);
1491}
1492
1493static void check_and_drop(void *_data)
1494{
1495 struct select_data *data = _data;
1496
1497 if (d_mountpoint(data->start))
1498 data->found = -EBUSY;
1499 if (!data->found)
1500 __d_drop(data->start);
1501}
1502
1503/**
1504 * check_submounts_and_drop - prune dcache, check for submounts and drop
1505 *
1506 * All done as a single atomic operation relative to has_unlinked_ancestor().
1507 * Returns 0 if successfully unhashed @parent. If there were submounts then
1508 * return -EBUSY.
1509 *
1510 * @dentry: dentry to prune and drop
1511 */
1512int check_submounts_and_drop(struct dentry *dentry)
1513{
1514 int ret = 0;
1515
1516 /* Negative dentries can be dropped without further checks */
1517 if (!dentry->d_inode) {
1518 d_drop(dentry);
1519 goto out;
1520 }
1521
1522 for (;;) {
1523 struct select_data data;
1524
1525 INIT_LIST_HEAD(&data.dispose);
1526 data.start = dentry;
1527 data.found = 0;
1528
1529 d_walk(dentry, &data, check_and_collect, check_and_drop);
1530 ret = data.found;
1531
1532 if (!list_empty(&data.dispose))
1533 shrink_dentry_list(&data.dispose);
1534
1535 if (ret <= 0)
1536 break;
1537
1538 cond_resched();
1539 }
1540
1541out:
1542 return ret;
1543}
1544EXPORT_SYMBOL(check_submounts_and_drop);
1545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546/**
Al Viroa4464db2011-07-07 15:03:58 -04001547 * __d_alloc - allocate a dcache entry
1548 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 * @name: qstr of the name
1550 *
1551 * Allocates a dentry. It returns %NULL if there is insufficient memory
1552 * available. On a success the dentry is returned. The name passed in is
1553 * copied and the copy passed in may be reused after this call.
1554 */
1555
Al Viroa4464db2011-07-07 15:03:58 -04001556struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
1558 struct dentry *dentry;
1559 char *dname;
1560
Mel Gormane12ba742007-10-16 01:25:52 -07001561 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (!dentry)
1563 return NULL;
1564
Linus Torvalds6326c712012-05-21 16:14:04 -07001565 /*
1566 * We guarantee that the inline name is always NUL-terminated.
1567 * This way the memcpy() done by the name switching in rename
1568 * will still always have a NUL at the end, even if we might
1569 * be overwriting an internal NUL character
1570 */
1571 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (name->len > DNAME_INLINE_LEN-1) {
1573 dname = kmalloc(name->len + 1, GFP_KERNEL);
1574 if (!dname) {
1575 kmem_cache_free(dentry_cache, dentry);
1576 return NULL;
1577 }
1578 } else {
1579 dname = dentry->d_iname;
1580 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 dentry->d_name.len = name->len;
1583 dentry->d_name.hash = name->hash;
1584 memcpy(dname, name->name, name->len);
1585 dname[name->len] = 0;
1586
Linus Torvalds6326c712012-05-21 16:14:04 -07001587 /* Make sure we always see the terminating NUL character */
1588 smp_wmb();
1589 dentry->d_name.name = dname;
1590
Waiman Long98474232013-08-28 18:24:59 -07001591 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001592 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001594 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001596 dentry->d_parent = dentry;
1597 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 dentry->d_op = NULL;
1599 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001600 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 INIT_LIST_HEAD(&dentry->d_lru);
1602 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001603 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001604 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001605 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Nick Piggin3e880fb2011-01-07 17:49:19 +11001607 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 return dentry;
1610}
Al Viroa4464db2011-07-07 15:03:58 -04001611
1612/**
1613 * d_alloc - allocate a dcache entry
1614 * @parent: parent of entry to allocate
1615 * @name: qstr of the name
1616 *
1617 * Allocates a dentry. It returns %NULL if there is insufficient memory
1618 * available. On a success the dentry is returned. The name passed in is
1619 * copied and the copy passed in may be reused after this call.
1620 */
1621struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1622{
1623 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1624 if (!dentry)
1625 return NULL;
1626
1627 spin_lock(&parent->d_lock);
1628 /*
1629 * don't need child lock because it is not subject
1630 * to concurrency here
1631 */
1632 __dget_dlock(parent);
1633 dentry->d_parent = parent;
1634 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1635 spin_unlock(&parent->d_lock);
1636
1637 return dentry;
1638}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001639EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Nick Piggin4b936882011-01-07 17:50:07 +11001641struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1642{
Al Viroa4464db2011-07-07 15:03:58 -04001643 struct dentry *dentry = __d_alloc(sb, name);
1644 if (dentry)
Nick Piggin4b936882011-01-07 17:50:07 +11001645 dentry->d_flags |= DCACHE_DISCONNECTED;
Nick Piggin4b936882011-01-07 17:50:07 +11001646 return dentry;
1647}
1648EXPORT_SYMBOL(d_alloc_pseudo);
1649
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1651{
1652 struct qstr q;
1653
1654 q.name = name;
1655 q.len = strlen(name);
1656 q.hash = full_name_hash(q.name, q.len);
1657 return d_alloc(parent, &q);
1658}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001659EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
Nick Pigginfb045ad2011-01-07 17:49:55 +11001661void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1662{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001663 WARN_ON_ONCE(dentry->d_op);
1664 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001665 DCACHE_OP_COMPARE |
1666 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001667 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001668 DCACHE_OP_DELETE ));
1669 dentry->d_op = op;
1670 if (!op)
1671 return;
1672 if (op->d_hash)
1673 dentry->d_flags |= DCACHE_OP_HASH;
1674 if (op->d_compare)
1675 dentry->d_flags |= DCACHE_OP_COMPARE;
1676 if (op->d_revalidate)
1677 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001678 if (op->d_weak_revalidate)
1679 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001680 if (op->d_delete)
1681 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001682 if (op->d_prune)
1683 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001684
1685}
1686EXPORT_SYMBOL(d_set_d_op);
1687
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001688static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1689{
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001690 spin_lock(&dentry->d_lock);
David Howells9875cf82011-01-14 18:45:21 +00001691 if (inode) {
1692 if (unlikely(IS_AUTOMOUNT(inode)))
1693 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
Al Virob3d9b7a2012-06-09 13:51:19 -04001694 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
David Howells9875cf82011-01-14 18:45:21 +00001695 }
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001696 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001697 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001698 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001699 fsnotify_d_instantiate(dentry, inode);
1700}
1701
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702/**
1703 * d_instantiate - fill in inode information for a dentry
1704 * @entry: dentry to complete
1705 * @inode: inode to attach to this dentry
1706 *
1707 * Fill in inode information in the entry.
1708 *
1709 * This turns negative dentries into productive full members
1710 * of society.
1711 *
1712 * NOTE! This assumes that the inode count has been incremented
1713 * (or otherwise set) by the caller to indicate that it is now
1714 * in use by the dcache.
1715 */
1716
1717void d_instantiate(struct dentry *entry, struct inode * inode)
1718{
Al Virob3d9b7a2012-06-09 13:51:19 -04001719 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001720 if (inode)
1721 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001722 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001723 if (inode)
1724 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 security_d_instantiate(entry, inode);
1726}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001727EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729/**
1730 * d_instantiate_unique - instantiate a non-aliased dentry
1731 * @entry: dentry to instantiate
1732 * @inode: inode to attach to this dentry
1733 *
1734 * Fill in inode information in the entry. On success, it returns NULL.
1735 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001736 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 *
1738 * Note that in order to avoid conflicts with rename() etc, the caller
1739 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001740 *
1741 * This also assumes that the inode count has been incremented
1742 * (or otherwise set) by the caller to indicate that it is now
1743 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 */
David Howells770bfad2006-08-22 20:06:07 -04001745static struct dentry *__d_instantiate_unique(struct dentry *entry,
1746 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747{
1748 struct dentry *alias;
1749 int len = entry->d_name.len;
1750 const char *name = entry->d_name.name;
1751 unsigned int hash = entry->d_name.hash;
1752
David Howells770bfad2006-08-22 20:06:07 -04001753 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001754 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001755 return NULL;
1756 }
1757
Sasha Levinb67bfe02013-02-27 17:06:00 -08001758 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001759 /*
1760 * Don't need alias->d_lock here, because aliases with
1761 * d_parent == entry->d_parent are not subject to name or
1762 * parent changes, because the parent inode i_mutex is held.
1763 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001764 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 continue;
1766 if (alias->d_parent != entry->d_parent)
1767 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001768 if (alias->d_name.len != len)
1769 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001770 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001772 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 return alias;
1774 }
David Howells770bfad2006-08-22 20:06:07 -04001775
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001776 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 return NULL;
1778}
David Howells770bfad2006-08-22 20:06:07 -04001779
1780struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1781{
1782 struct dentry *result;
1783
Al Virob3d9b7a2012-06-09 13:51:19 -04001784 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001785
Nick Piggin873feea2011-01-07 17:50:06 +11001786 if (inode)
1787 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001788 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001789 if (inode)
1790 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001791
1792 if (!result) {
1793 security_d_instantiate(entry, inode);
1794 return NULL;
1795 }
1796
1797 BUG_ON(!d_unhashed(result));
1798 iput(inode);
1799 return result;
1800}
1801
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802EXPORT_SYMBOL(d_instantiate_unique);
1803
Al Viroadc0e912012-01-08 16:49:21 -05001804struct dentry *d_make_root(struct inode *root_inode)
1805{
1806 struct dentry *res = NULL;
1807
1808 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001809 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001810
1811 res = __d_alloc(root_inode->i_sb, &name);
1812 if (res)
1813 d_instantiate(res, root_inode);
1814 else
1815 iput(root_inode);
1816 }
1817 return res;
1818}
1819EXPORT_SYMBOL(d_make_root);
1820
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001821static struct dentry * __d_find_any_alias(struct inode *inode)
1822{
1823 struct dentry *alias;
1824
Al Virob3d9b7a2012-06-09 13:51:19 -04001825 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001826 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001827 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001828 __dget(alias);
1829 return alias;
1830}
1831
Sage Weil46f72b32012-01-10 09:04:37 -08001832/**
1833 * d_find_any_alias - find any alias for a given inode
1834 * @inode: inode to find an alias for
1835 *
1836 * If any aliases exist for the given inode, take and return a
1837 * reference for one of them. If no aliases exist, return %NULL.
1838 */
1839struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001840{
1841 struct dentry *de;
1842
1843 spin_lock(&inode->i_lock);
1844 de = __d_find_any_alias(inode);
1845 spin_unlock(&inode->i_lock);
1846 return de;
1847}
Sage Weil46f72b32012-01-10 09:04:37 -08001848EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001851 * d_obtain_alias - find or allocate a dentry for a given inode
1852 * @inode: inode to allocate the dentry for
1853 *
1854 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1855 * similar open by handle operations. The returned dentry may be anonymous,
1856 * or may have a full name (if the inode was already in the cache).
1857 *
1858 * When called on a directory inode, we must ensure that the inode only ever
1859 * has one dentry. If a dentry is found, that is returned instead of
1860 * allocating a new one.
1861 *
1862 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001863 * to the dentry. In case of an error the reference on the inode is released.
1864 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1865 * be passed in and will be the error will be propagate to the return value,
1866 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001867 */
1868struct dentry *d_obtain_alias(struct inode *inode)
1869{
NeilBrownb911a6b2012-11-08 16:09:37 -08001870 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001871 struct dentry *tmp;
1872 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001873
1874 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001875 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001876 if (IS_ERR(inode))
1877 return ERR_CAST(inode);
1878
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001879 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001880 if (res)
1881 goto out_iput;
1882
Al Viroa4464db2011-07-07 15:03:58 -04001883 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001884 if (!tmp) {
1885 res = ERR_PTR(-ENOMEM);
1886 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001887 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001888
Nick Piggin873feea2011-01-07 17:50:06 +11001889 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001890 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001891 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001892 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001893 dput(tmp);
1894 goto out_iput;
1895 }
1896
1897 /* attach a disconnected dentry */
1898 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001899 tmp->d_inode = inode;
1900 tmp->d_flags |= DCACHE_DISCONNECTED;
Al Virob3d9b7a2012-06-09 13:51:19 -04001901 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001902 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001903 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001904 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001905 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001906 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001907 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001908
Christoph Hellwig9308a612008-08-11 15:49:12 +02001909 return tmp;
1910
1911 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001912 if (res && !IS_ERR(res))
1913 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001914 iput(inode);
1915 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001916}
Benny Halevyadc48722009-02-27 14:02:59 -08001917EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
1919/**
1920 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1921 * @inode: the inode which may have a disconnected dentry
1922 * @dentry: a negative dentry which we want to point to the inode.
1923 *
1924 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1925 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1926 * and return it, else simply d_add the inode to the dentry and return NULL.
1927 *
1928 * This is needed in the lookup routine of any filesystem that is exportable
1929 * (via knfsd) so that we can build dcache paths to directories effectively.
1930 *
1931 * If a dentry was found and moved, then it is returned. Otherwise NULL
1932 * is returned. This matches the expected return value of ->lookup.
1933 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001934 * Cluster filesystems may call this function with a negative, hashed dentry.
1935 * In that case, we know that the inode will be a regular file, and also this
1936 * will only occur during atomic_open. So we need to check for the dentry
1937 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 */
1939struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1940{
1941 struct dentry *new = NULL;
1942
Al Viroa9049372011-07-08 21:20:11 -04001943 if (IS_ERR(inode))
1944 return ERR_CAST(inode);
1945
NeilBrown21c0d8f2006-10-04 02:16:16 -07001946 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001947 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001948 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001950 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001951 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 d_move(new, dentry);
1954 iput(inode);
1955 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001956 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001957 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001958 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 security_d_instantiate(dentry, inode);
1960 d_rehash(dentry);
1961 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001962 } else {
1963 d_instantiate(dentry, inode);
1964 if (d_unhashed(dentry))
1965 d_rehash(dentry);
1966 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 return new;
1968}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001969EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Barry Naujok94035402008-05-21 16:50:46 +10001971/**
1972 * d_add_ci - lookup or allocate new dentry with case-exact name
1973 * @inode: the inode case-insensitive lookup has found
1974 * @dentry: the negative dentry that was passed to the parent's lookup func
1975 * @name: the case-exact name to be associated with the returned dentry
1976 *
1977 * This is to avoid filling the dcache with case-insensitive names to the
1978 * same inode, only the actual correct case is stored in the dcache for
1979 * case-insensitive filesystems.
1980 *
1981 * For a case-insensitive lookup match and if the the case-exact dentry
1982 * already exists in in the dcache, use it and return it.
1983 *
1984 * If no entry exists with the exact case name, allocate new dentry with
1985 * the exact case, and return the spliced entry.
1986 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02001987struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10001988 struct qstr *name)
1989{
Barry Naujok94035402008-05-21 16:50:46 +10001990 struct dentry *found;
1991 struct dentry *new;
1992
Christoph Hellwigb6520c82009-01-05 19:10:37 +01001993 /*
1994 * First check if a dentry matching the name already exists,
1995 * if not go ahead and create it now.
1996 */
Barry Naujok94035402008-05-21 16:50:46 +10001997 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05001998 if (unlikely(IS_ERR(found)))
1999 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10002000 if (!found) {
2001 new = d_alloc(dentry->d_parent, name);
2002 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002003 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10002004 goto err_out;
2005 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002006
Barry Naujok94035402008-05-21 16:50:46 +10002007 found = d_splice_alias(inode, new);
2008 if (found) {
2009 dput(new);
2010 return found;
2011 }
2012 return new;
2013 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002014
2015 /*
2016 * If a matching dentry exists, and it's not negative use it.
2017 *
2018 * Decrement the reference count to balance the iget() done
2019 * earlier on.
2020 */
Barry Naujok94035402008-05-21 16:50:46 +10002021 if (found->d_inode) {
2022 if (unlikely(found->d_inode != inode)) {
2023 /* This can't happen because bad inodes are unhashed. */
2024 BUG_ON(!is_bad_inode(inode));
2025 BUG_ON(!is_bad_inode(found->d_inode));
2026 }
Barry Naujok94035402008-05-21 16:50:46 +10002027 iput(inode);
2028 return found;
2029 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002030
Barry Naujok94035402008-05-21 16:50:46 +10002031 /*
2032 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002033 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10002034 */
Al Viro4513d892011-07-17 10:52:14 -04002035 new = d_splice_alias(inode, found);
2036 if (new) {
2037 dput(found);
2038 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10002039 }
Al Viro4513d892011-07-17 10:52:14 -04002040 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002041
2042err_out:
2043 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002044 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002045}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002046EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002048/*
2049 * Do the slow-case of the dentry name compare.
2050 *
2051 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002052 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002053 * filesystem can rely on them, and can use the 'name' and
2054 * 'len' information without worrying about walking off the
2055 * end of memory etc.
2056 *
2057 * Thus the read_seqcount_retry() and the "duplicate" info
2058 * in arguments (the low-level filesystem should not look
2059 * at the dentry inode or name contents directly, since
2060 * rename can change them while we're in RCU mode).
2061 */
2062enum slow_d_compare {
2063 D_COMP_OK,
2064 D_COMP_NOMATCH,
2065 D_COMP_SEQRETRY,
2066};
2067
2068static noinline enum slow_d_compare slow_dentry_cmp(
2069 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002070 struct dentry *dentry,
2071 unsigned int seq,
2072 const struct qstr *name)
2073{
2074 int tlen = dentry->d_name.len;
2075 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002076
2077 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2078 cpu_relax();
2079 return D_COMP_SEQRETRY;
2080 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002081 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002082 return D_COMP_NOMATCH;
2083 return D_COMP_OK;
2084}
2085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002087 * __d_lookup_rcu - search for a dentry (racy, store-free)
2088 * @parent: parent dentry
2089 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002090 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002091 * Returns: dentry, or NULL
2092 *
2093 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2094 * resolution (store-free path walking) design described in
2095 * Documentation/filesystems/path-lookup.txt.
2096 *
2097 * This is not to be used outside core vfs.
2098 *
2099 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2100 * held, and rcu_read_lock held. The returned dentry must not be stored into
2101 * without taking d_lock and checking d_seq sequence count against @seq
2102 * returned here.
2103 *
Linus Torvalds15570082013-09-02 11:38:06 -07002104 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002105 * function.
2106 *
2107 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2108 * the returned dentry, so long as its parent's seqlock is checked after the
2109 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2110 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002111 *
2112 * NOTE! The caller *has* to check the resulting dentry against the sequence
2113 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002114 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002115struct dentry *__d_lookup_rcu(const struct dentry *parent,
2116 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002117 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002118{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002119 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002120 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002121 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002122 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002123 struct dentry *dentry;
2124
2125 /*
2126 * Note: There is significant duplication with __d_lookup_rcu which is
2127 * required to prevent single threaded performance regressions
2128 * especially on architectures where smp_rmb (in seqcounts) are costly.
2129 * Keep the two functions in sync.
2130 */
2131
2132 /*
2133 * The hash list is protected using RCU.
2134 *
2135 * Carefully use d_seq when comparing a candidate dentry, to avoid
2136 * races with d_move().
2137 *
2138 * It is possible that concurrent renames can mess up our list
2139 * walk here and result in missing our dentry, resulting in the
2140 * false-negative result. d_lookup() protects against concurrent
2141 * renames using rename_lock seqlock.
2142 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002143 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002144 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002145 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002146 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002147
Nick Piggin31e6b012011-01-07 17:49:52 +11002148seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002149 /*
2150 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002151 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002152 *
2153 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002154 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002155 *
2156 * NOTE! We do a "raw" seqcount_begin here. That means that
2157 * we don't wait for the sequence count to stabilize if it
2158 * is in the middle of a sequence change. If we do the slow
2159 * dentry compare, we will do seqretries until it is stable,
2160 * and if we end up with a successful lookup, we actually
2161 * want to exit RCU lookup anyway.
2162 */
2163 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002164 if (dentry->d_parent != parent)
2165 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002166 if (d_unhashed(dentry))
2167 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002168
2169 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002170 if (dentry->d_name.hash != hashlen_hash(hashlen))
2171 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002172 *seqp = seq;
2173 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002174 case D_COMP_OK:
2175 return dentry;
2176 case D_COMP_NOMATCH:
2177 continue;
2178 default:
2179 goto seqretry;
2180 }
2181 }
2182
Linus Torvalds26fe5752012-05-10 13:14:12 -07002183 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002184 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002185 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002186 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002187 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002188 }
2189 return NULL;
2190}
2191
2192/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 * d_lookup - search for a dentry
2194 * @parent: parent dentry
2195 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002196 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002198 * d_lookup searches the children of the parent dentry for the name in
2199 * question. If the dentry is found its reference count is incremented and the
2200 * dentry is returned. The caller must use dput to free the entry when it has
2201 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 */
Al Viroda2d8452013-01-24 18:29:34 -05002203struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204{
Nick Piggin31e6b012011-01-07 17:49:52 +11002205 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002206 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
2208 do {
2209 seq = read_seqbegin(&rename_lock);
2210 dentry = __d_lookup(parent, name);
2211 if (dentry)
2212 break;
2213 } while (read_seqretry(&rename_lock, seq));
2214 return dentry;
2215}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002216EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
Nick Piggin31e6b012011-01-07 17:49:52 +11002218/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002219 * __d_lookup - search for a dentry (racy)
2220 * @parent: parent dentry
2221 * @name: qstr of name we wish to find
2222 * Returns: dentry, or NULL
2223 *
2224 * __d_lookup is like d_lookup, however it may (rarely) return a
2225 * false-negative result due to unrelated rename activity.
2226 *
2227 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2228 * however it must be used carefully, eg. with a following d_lookup in
2229 * the case of failure.
2230 *
2231 * __d_lookup callers must be commented.
2232 */
Al Viroa713ca22013-01-24 18:27:00 -05002233struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234{
2235 unsigned int len = name->len;
2236 unsigned int hash = name->hash;
2237 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002238 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002239 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002240 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002241 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
Nick Pigginb04f7842010-08-18 04:37:34 +10002243 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002244 * Note: There is significant duplication with __d_lookup_rcu which is
2245 * required to prevent single threaded performance regressions
2246 * especially on architectures where smp_rmb (in seqcounts) are costly.
2247 * Keep the two functions in sync.
2248 */
2249
2250 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002251 * The hash list is protected using RCU.
2252 *
2253 * Take d_lock when comparing a candidate dentry, to avoid races
2254 * with d_move().
2255 *
2256 * It is possible that concurrent renames can mess up our list
2257 * walk here and result in missing our dentry, resulting in the
2258 * false-negative result. d_lookup() protects against concurrent
2259 * renames using rename_lock seqlock.
2260 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002261 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002262 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263 rcu_read_lock();
2264
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002265 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 if (dentry->d_name.hash != hash)
2268 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
2270 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 if (dentry->d_parent != parent)
2272 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002273 if (d_unhashed(dentry))
2274 goto next;
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 /*
2277 * It is safe to compare names since d_move() cannot
2278 * change the qstr (protected by d_lock).
2279 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002280 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002281 int tlen = dentry->d_name.len;
2282 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002283 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 goto next;
2285 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002286 if (dentry->d_name.len != len)
2287 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002288 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 goto next;
2290 }
2291
Waiman Long98474232013-08-28 18:24:59 -07002292 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002293 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 spin_unlock(&dentry->d_lock);
2295 break;
2296next:
2297 spin_unlock(&dentry->d_lock);
2298 }
2299 rcu_read_unlock();
2300
2301 return found;
2302}
2303
2304/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002305 * d_hash_and_lookup - hash the qstr then search for a dentry
2306 * @dir: Directory to search in
2307 * @name: qstr of name we wish to find
2308 *
Al Viro4f522a22013-02-11 23:20:37 -05002309 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002310 */
2311struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2312{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002313 /*
2314 * Check for a fs-specific hash function. Note that we must
2315 * calculate the standard hash first, as the d_op->d_hash()
2316 * routine may choose to leave the hash value unchanged.
2317 */
2318 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002319 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002320 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002321 if (unlikely(err < 0))
2322 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002323 }
Al Viro4f522a22013-02-11 23:20:37 -05002324 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002325}
Al Viro4f522a22013-02-11 23:20:37 -05002326EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002327
2328/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002329 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002331 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 *
2333 * An insecure source has sent us a dentry, here we verify it and dget() it.
2334 * This is used by ncpfs in its readdir implementation.
2335 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002336 *
2337 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002339int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340{
Nick Piggin786a5e12011-01-07 17:49:16 +11002341 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002342
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002343 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002344 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2345 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002346 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002347 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002348 spin_unlock(&dentry->d_lock);
2349 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 return 1;
2351 }
2352 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002353 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002354
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 return 0;
2356}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002357EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358
2359/*
2360 * When a file is deleted, we have two options:
2361 * - turn this dentry into a negative dentry
2362 * - unhash this dentry and free it.
2363 *
2364 * Usually, we want to just turn this into
2365 * a negative dentry, but if anybody else is
2366 * currently using the dentry or the inode
2367 * we can't do that and we fall back on removing
2368 * it from the hash queues and waiting for
2369 * it to be deleted later when it has no users
2370 */
2371
2372/**
2373 * d_delete - delete a dentry
2374 * @dentry: The dentry to delete
2375 *
2376 * Turn the dentry into a negative dentry if possible, otherwise
2377 * remove it from the hash queues so it can be deleted later
2378 */
2379
2380void d_delete(struct dentry * dentry)
2381{
Nick Piggin873feea2011-01-07 17:50:06 +11002382 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002383 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 /*
2385 * Are we the only user?
2386 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002387again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002389 inode = dentry->d_inode;
2390 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002391 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002392 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002393 spin_unlock(&dentry->d_lock);
2394 cpu_relax();
2395 goto again;
2396 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002397 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002398 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002399 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400 return;
2401 }
2402
2403 if (!d_unhashed(dentry))
2404 __d_drop(dentry);
2405
2406 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002407
2408 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002410EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002412static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002414 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002415 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002416 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002417 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002418 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419}
2420
David Howells770bfad2006-08-22 20:06:07 -04002421static void _d_rehash(struct dentry * entry)
2422{
2423 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2424}
2425
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426/**
2427 * d_rehash - add an entry back to the hash
2428 * @entry: dentry to add to the hash
2429 *
2430 * Adds a dentry to the hash according to its name.
2431 */
2432
2433void d_rehash(struct dentry * entry)
2434{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002436 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002439EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002441/**
2442 * dentry_update_name_case - update case insensitive dentry with a new name
2443 * @dentry: dentry to be updated
2444 * @name: new name
2445 *
2446 * Update a case insensitive dentry with new case of name.
2447 *
2448 * dentry must have been returned by d_lookup with name @name. Old and new
2449 * name lengths must match (ie. no d_compare which allows mismatched name
2450 * lengths).
2451 *
2452 * Parent inode i_mutex must be held over d_lookup and into this call (to
2453 * keep renames and concurrent inserts, and readdir(2) away).
2454 */
2455void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2456{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002457 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002458 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2459
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002460 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002461 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002462 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002463 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002464 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002465}
2466EXPORT_SYMBOL(dentry_update_name_case);
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468static void switch_names(struct dentry *dentry, struct dentry *target)
2469{
2470 if (dname_external(target)) {
2471 if (dname_external(dentry)) {
2472 /*
2473 * Both external: swap the pointers
2474 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002475 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 } else {
2477 /*
2478 * dentry:internal, target:external. Steal target's
2479 * storage and make target internal.
2480 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002481 memcpy(target->d_iname, dentry->d_name.name,
2482 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 dentry->d_name.name = target->d_name.name;
2484 target->d_name.name = target->d_iname;
2485 }
2486 } else {
2487 if (dname_external(dentry)) {
2488 /*
2489 * dentry:external, target:internal. Give dentry's
2490 * storage to target and make dentry internal
2491 */
2492 memcpy(dentry->d_iname, target->d_name.name,
2493 target->d_name.len + 1);
2494 target->d_name.name = dentry->d_name.name;
2495 dentry->d_name.name = dentry->d_iname;
2496 } else {
2497 /*
2498 * Both are internal. Just copy target to dentry
2499 */
2500 memcpy(dentry->d_iname, target->d_name.name,
2501 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002502 dentry->d_name.len = target->d_name.len;
2503 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 }
2505 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002506 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507}
2508
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002509static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2510{
2511 /*
2512 * XXXX: do we really need to take target->d_lock?
2513 */
2514 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2515 spin_lock(&target->d_parent->d_lock);
2516 else {
2517 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2518 spin_lock(&dentry->d_parent->d_lock);
2519 spin_lock_nested(&target->d_parent->d_lock,
2520 DENTRY_D_LOCK_NESTED);
2521 } else {
2522 spin_lock(&target->d_parent->d_lock);
2523 spin_lock_nested(&dentry->d_parent->d_lock,
2524 DENTRY_D_LOCK_NESTED);
2525 }
2526 }
2527 if (target < dentry) {
2528 spin_lock_nested(&target->d_lock, 2);
2529 spin_lock_nested(&dentry->d_lock, 3);
2530 } else {
2531 spin_lock_nested(&dentry->d_lock, 2);
2532 spin_lock_nested(&target->d_lock, 3);
2533 }
2534}
2535
2536static void dentry_unlock_parents_for_move(struct dentry *dentry,
2537 struct dentry *target)
2538{
2539 if (target->d_parent != dentry->d_parent)
2540 spin_unlock(&dentry->d_parent->d_lock);
2541 if (target->d_parent != target)
2542 spin_unlock(&target->d_parent->d_lock);
2543}
2544
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002546 * When switching names, the actual string doesn't strictly have to
2547 * be preserved in the target - because we're dropping the target
2548 * anyway. As such, we can just do a simple memcpy() to copy over
2549 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002551 * Note that we have to be a lot more careful about getting the hash
2552 * switched - we have to switch the hash value properly even if it
2553 * then no longer matches the actual (corrupted) string of the target.
2554 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002556/*
Al Viro18367502011-07-12 21:42:24 -04002557 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 * @dentry: entry to move
2559 * @target: new dentry
2560 *
2561 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002562 * dcache entries should not be moved in this way. Caller must hold
2563 * rename_lock, the i_mutex of the source and target directories,
2564 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 */
Al Viro18367502011-07-12 21:42:24 -04002566static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 if (!dentry->d_inode)
2569 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2570
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002571 BUG_ON(d_ancestor(dentry, target));
2572 BUG_ON(d_ancestor(target, dentry));
2573
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002574 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575
Nick Piggin31e6b012011-01-07 17:49:52 +11002576 write_seqcount_begin(&dentry->d_seq);
2577 write_seqcount_begin(&target->d_seq);
2578
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002579 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2580
2581 /*
2582 * Move the dentry to the target hash queue. Don't bother checking
2583 * for the same hash queue because of how unlikely it is.
2584 */
2585 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002586 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587
2588 /* Unhash the target: dput() will then get rid of it */
2589 __d_drop(target);
2590
Eric Dumazet5160ee62006-01-08 01:03:32 -08002591 list_del(&dentry->d_u.d_child);
2592 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593
2594 /* Switch the names.. */
2595 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002596 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597
2598 /* ... and switch the parents */
2599 if (IS_ROOT(dentry)) {
2600 dentry->d_parent = target->d_parent;
2601 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002602 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002604 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605
2606 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002607 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 }
2609
Eric Dumazet5160ee62006-01-08 01:03:32 -08002610 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002611
Nick Piggin31e6b012011-01-07 17:49:52 +11002612 write_seqcount_end(&target->d_seq);
2613 write_seqcount_end(&dentry->d_seq);
2614
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002615 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002617 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002619}
2620
2621/*
2622 * d_move - move a dentry
2623 * @dentry: entry to move
2624 * @target: new dentry
2625 *
2626 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002627 * dcache entries should not be moved in this way. See the locking
2628 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002629 */
2630void d_move(struct dentry *dentry, struct dentry *target)
2631{
2632 write_seqlock(&rename_lock);
2633 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002634 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002635}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002636EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002638/**
2639 * d_ancestor - search for an ancestor
2640 * @p1: ancestor dentry
2641 * @p2: child dentry
2642 *
2643 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2644 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002645 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002646struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002647{
2648 struct dentry *p;
2649
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002650 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002651 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002652 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002653 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002654 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002655}
2656
2657/*
2658 * This helper attempts to cope with remotely renamed directories
2659 *
2660 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002661 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002662 *
2663 * Note: If ever the locking in lock_rename() changes, then please
2664 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002665 */
Nick Piggin873feea2011-01-07 17:50:06 +11002666static struct dentry *__d_unalias(struct inode *inode,
2667 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002668{
2669 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002670 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002671
2672 /* If alias and dentry share a parent, then no extra locks required */
2673 if (alias->d_parent == dentry->d_parent)
2674 goto out_unalias;
2675
Trond Myklebust9eaef272006-10-21 10:24:20 -07002676 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002677 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2678 goto out_err;
2679 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2680 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2681 goto out_err;
2682 m2 = &alias->d_parent->d_inode->i_mutex;
2683out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002684 if (likely(!d_mountpoint(alias))) {
2685 __d_move(alias, dentry);
2686 ret = alias;
2687 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002688out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002689 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002690 if (m2)
2691 mutex_unlock(m2);
2692 if (m1)
2693 mutex_unlock(m1);
2694 return ret;
2695}
2696
2697/*
David Howells770bfad2006-08-22 20:06:07 -04002698 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2699 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002700 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002701 */
2702static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2703{
Al Viro740da422013-01-30 10:13:38 -05002704 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002705
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002706 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002707
Nick Piggin31e6b012011-01-07 17:49:52 +11002708 write_seqcount_begin(&dentry->d_seq);
2709 write_seqcount_begin(&anon->d_seq);
2710
David Howells770bfad2006-08-22 20:06:07 -04002711 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002712
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002713 switch_names(dentry, anon);
2714 swap(dentry->d_name.hash, anon->d_name.hash);
2715
Al Viro740da422013-01-30 10:13:38 -05002716 dentry->d_parent = dentry;
2717 list_del_init(&dentry->d_u.d_child);
2718 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002719 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002720
Nick Piggin31e6b012011-01-07 17:49:52 +11002721 write_seqcount_end(&dentry->d_seq);
2722 write_seqcount_end(&anon->d_seq);
2723
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002724 dentry_unlock_parents_for_move(anon, dentry);
2725 spin_unlock(&dentry->d_lock);
2726
2727 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002728 anon->d_flags &= ~DCACHE_DISCONNECTED;
2729}
2730
2731/**
2732 * d_materialise_unique - introduce an inode into the tree
2733 * @dentry: candidate dentry
2734 * @inode: inode to bind to the dentry, to which aliases may be attached
2735 *
2736 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002737 * root directory alias in its place if there is one. Caller must hold the
2738 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002739 */
2740struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2741{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002742 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002743
2744 BUG_ON(!d_unhashed(dentry));
2745
David Howells770bfad2006-08-22 20:06:07 -04002746 if (!inode) {
2747 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002748 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002749 d_rehash(actual);
2750 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002751 }
2752
Nick Piggin873feea2011-01-07 17:50:06 +11002753 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002754
Trond Myklebust9eaef272006-10-21 10:24:20 -07002755 if (S_ISDIR(inode->i_mode)) {
2756 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002757
Trond Myklebust9eaef272006-10-21 10:24:20 -07002758 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002759 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002760 if (alias) {
2761 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002762 write_seqlock(&rename_lock);
2763
2764 if (d_ancestor(alias, dentry)) {
2765 /* Check for loops */
2766 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002767 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002768 } else if (IS_ROOT(alias)) {
2769 /* Is this an anonymous mountpoint that we
2770 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002771 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002772 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002773 __d_drop(alias);
2774 goto found;
Al Viro18367502011-07-12 21:42:24 -04002775 } else {
2776 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002777 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002778 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002779 }
Al Viro18367502011-07-12 21:42:24 -04002780 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002781 if (IS_ERR(actual)) {
2782 if (PTR_ERR(actual) == -ELOOP)
2783 pr_warn_ratelimited(
2784 "VFS: Lookup of '%s' in %s %s"
2785 " would have caused loop\n",
2786 dentry->d_name.name,
2787 inode->i_sb->s_type->name,
2788 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002789 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002790 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002791 goto out_nolock;
2792 }
David Howells770bfad2006-08-22 20:06:07 -04002793 }
2794
2795 /* Add a unique reference */
2796 actual = __d_instantiate_unique(dentry, inode);
2797 if (!actual)
2798 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002799 else
2800 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002801
David Howells770bfad2006-08-22 20:06:07 -04002802 spin_lock(&actual->d_lock);
2803found:
2804 _d_rehash(actual);
2805 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002806 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002807out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002808 if (actual == dentry) {
2809 security_d_instantiate(dentry, inode);
2810 return NULL;
2811 }
2812
2813 iput(inode);
2814 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002815}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002816EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002817
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002818static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002819{
2820 *buflen -= namelen;
2821 if (*buflen < 0)
2822 return -ENAMETOOLONG;
2823 *buffer -= namelen;
2824 memcpy(*buffer, str, namelen);
2825 return 0;
2826}
2827
Waiman Long232d2d62013-09-09 12:18:13 -04002828/**
2829 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002830 * @buffer: buffer pointer
2831 * @buflen: allocated length of the buffer
2832 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002833 *
2834 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2835 * make sure that either the old or the new name pointer and length are
2836 * fetched. However, there may be mismatch between length and pointer.
2837 * The length cannot be trusted, we need to copy it byte-by-byte until
2838 * the length is reached or a null byte is found. It also prepends "/" at
2839 * the beginning of the name. The sequence number check at the caller will
2840 * retry it again when a d_move() does happen. So any garbage in the buffer
2841 * due to mismatched pointer and length will be discarded.
2842 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002843static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2844{
Waiman Long232d2d62013-09-09 12:18:13 -04002845 const char *dname = ACCESS_ONCE(name->name);
2846 u32 dlen = ACCESS_ONCE(name->len);
2847 char *p;
2848
2849 if (*buflen < dlen + 1)
2850 return -ENAMETOOLONG;
2851 *buflen -= dlen + 1;
2852 p = *buffer -= dlen + 1;
2853 *p++ = '/';
2854 while (dlen--) {
2855 char c = *dname++;
2856 if (!c)
2857 break;
2858 *p++ = c;
2859 }
2860 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002861}
2862
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002864 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002865 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002866 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002867 * @buffer: pointer to the end of the buffer
2868 * @buflen: pointer to buffer length
2869 *
Waiman Long18129972013-09-12 10:55:35 -04002870 * The function will first try to write out the pathname without taking any
2871 * lock other than the RCU read lock to make sure that dentries won't go away.
2872 * It only checks the sequence number of the global rename_lock as any change
2873 * in the dentry's d_seq will be preceded by changes in the rename_lock
2874 * sequence number. If the sequence number had been changed, it will restart
2875 * the whole pathname back-tracing sequence again by taking the rename_lock.
2876 * In this case, there is no need to take the RCU read lock as the recursive
2877 * parent pointer references will keep the dentry chain alive as long as no
2878 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002879 */
Al Viro02125a82011-12-05 08:43:34 -05002880static int prepend_path(const struct path *path,
2881 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002882 char **buffer, int *buflen)
2883{
2884 struct dentry *dentry = path->dentry;
2885 struct vfsmount *vfsmnt = path->mnt;
Al Viro0714a532011-11-24 22:19:58 -05002886 struct mount *mnt = real_mount(vfsmnt);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002887 int error = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002888 unsigned seq = 0;
2889 char *bptr;
2890 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002891
Al Viro48f5ec22013-09-09 15:22:25 -04002892 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002893restart:
2894 bptr = *buffer;
2895 blen = *buflen;
2896 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002897 while (dentry != root->dentry || vfsmnt != root->mnt) {
2898 struct dentry * parent;
2899
2900 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2901 /* Global root? */
Waiman Long232d2d62013-09-09 12:18:13 -04002902 if (mnt_has_parent(mnt)) {
2903 dentry = mnt->mnt_mountpoint;
2904 mnt = mnt->mnt_parent;
2905 vfsmnt = &mnt->mnt;
2906 continue;
2907 }
2908 /*
2909 * Filesystems needing to implement special "root names"
2910 * should do so with ->d_dname()
2911 */
2912 if (IS_ROOT(dentry) &&
2913 (dentry->d_name.len != 1 ||
2914 dentry->d_name.name[0] != '/')) {
2915 WARN(1, "Root dentry has weird name <%.*s>\n",
2916 (int) dentry->d_name.len,
2917 dentry->d_name.name);
2918 }
2919 if (!error)
2920 error = is_mounted(vfsmnt) ? 1 : 2;
2921 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002922 }
2923 parent = dentry->d_parent;
2924 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002925 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002926 if (error)
2927 break;
2928
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002929 dentry = parent;
2930 }
Al Viro48f5ec22013-09-09 15:22:25 -04002931 if (!(seq & 1))
2932 rcu_read_unlock();
2933 if (need_seqretry(&rename_lock, seq)) {
2934 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002935 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002936 }
Al Viro48f5ec22013-09-09 15:22:25 -04002937 done_seqretry(&rename_lock, seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002938
Waiman Long232d2d62013-09-09 12:18:13 -04002939 if (error >= 0 && bptr == *buffer) {
2940 if (--blen < 0)
2941 error = -ENAMETOOLONG;
2942 else
2943 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002944 }
Waiman Long232d2d62013-09-09 12:18:13 -04002945 *buffer = bptr;
2946 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002947 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002948}
2949
2950/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002951 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002952 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002953 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002954 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 * @buflen: buffer length
2956 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002957 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002959 * Returns a pointer into the buffer or an error code if the
2960 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002961 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002962 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002963 *
Al Viro02125a82011-12-05 08:43:34 -05002964 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 */
Al Viro02125a82011-12-05 08:43:34 -05002966char *__d_path(const struct path *path,
2967 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002968 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002970 char *res = buf + buflen;
2971 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002973 prepend(&res, &buflen, "\0", 1);
Al Viro7ea600b2013-03-26 18:25:57 -04002974 br_read_lock(&vfsmount_lock);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002975 error = prepend_path(path, root, &res, &buflen);
Al Viro7ea600b2013-03-26 18:25:57 -04002976 br_read_unlock(&vfsmount_lock);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002977
Al Viro02125a82011-12-05 08:43:34 -05002978 if (error < 0)
2979 return ERR_PTR(error);
2980 if (error > 0)
2981 return NULL;
2982 return res;
2983}
2984
2985char *d_absolute_path(const struct path *path,
2986 char *buf, int buflen)
2987{
2988 struct path root = {};
2989 char *res = buf + buflen;
2990 int error;
2991
2992 prepend(&res, &buflen, "\0", 1);
Al Viro7ea600b2013-03-26 18:25:57 -04002993 br_read_lock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002994 error = prepend_path(path, &root, &res, &buflen);
Al Viro7ea600b2013-03-26 18:25:57 -04002995 br_read_unlock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05002996
2997 if (error > 1)
2998 error = -EINVAL;
2999 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003000 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003001 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002}
3003
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003004/*
3005 * same as __d_path but appends "(deleted)" for unlinked files.
3006 */
Al Viro02125a82011-12-05 08:43:34 -05003007static int path_with_deleted(const struct path *path,
3008 const struct path *root,
3009 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003010{
3011 prepend(buf, buflen, "\0", 1);
3012 if (d_unlinked(path->dentry)) {
3013 int error = prepend(buf, buflen, " (deleted)", 10);
3014 if (error)
3015 return error;
3016 }
3017
3018 return prepend_path(path, root, buf, buflen);
3019}
3020
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003021static int prepend_unreachable(char **buffer, int *buflen)
3022{
3023 return prepend(buffer, buflen, "(unreachable)", 13);
3024}
3025
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003026static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3027{
3028 unsigned seq;
3029
3030 do {
3031 seq = read_seqcount_begin(&fs->seq);
3032 *root = fs->root;
3033 } while (read_seqcount_retry(&fs->seq, seq));
3034}
3035
Jan Bluncka03a8a702008-02-14 19:38:32 -08003036/**
3037 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003038 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003039 * @buf: buffer to return value in
3040 * @buflen: buffer length
3041 *
3042 * Convert a dentry into an ASCII path name. If the entry has been deleted
3043 * the string " (deleted)" is appended. Note that this is ambiguous.
3044 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003045 * Returns a pointer into the buffer or an error code if the path was
3046 * too long. Note: Callers should use the returned pointer, not the passed
3047 * in buffer, to use the name! The implementation often starts at an offset
3048 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003049 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003050 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003051 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003052char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003054 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003055 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003056 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003058 /*
3059 * We have various synthetic filesystems that never get mounted. On
3060 * these filesystems dentries are never used for lookup purposes, and
3061 * thus don't need to be hashed. They also don't need a name until a
3062 * user wants to identify the object in /proc/pid/fd/. The little hack
3063 * below allows us to generate a name for these objects on demand:
3064 */
Jan Blunckcf28b482008-02-14 19:38:44 -08003065 if (path->dentry->d_op && path->dentry->d_op->d_dname)
3066 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003067
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003068 rcu_read_lock();
3069 get_fs_root_rcu(current->fs, &root);
Al Viro7ea600b2013-03-26 18:25:57 -04003070 br_read_lock(&vfsmount_lock);
Al Viro02125a82011-12-05 08:43:34 -05003071 error = path_with_deleted(path, &root, &res, &buflen);
Al Viro7ea600b2013-03-26 18:25:57 -04003072 br_read_unlock(&vfsmount_lock);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003073 rcu_read_unlock();
3074
Al Viro02125a82011-12-05 08:43:34 -05003075 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003076 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077 return res;
3078}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003079EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
3081/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003082 * Helper function for dentry_operations.d_dname() members
3083 */
3084char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3085 const char *fmt, ...)
3086{
3087 va_list args;
3088 char temp[64];
3089 int sz;
3090
3091 va_start(args, fmt);
3092 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3093 va_end(args);
3094
3095 if (sz > sizeof(temp) || sz > buflen)
3096 return ERR_PTR(-ENAMETOOLONG);
3097
3098 buffer += buflen - sz;
3099 return memcpy(buffer, temp, sz);
3100}
3101
Al Viro118b2302013-08-24 12:08:17 -04003102char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3103{
3104 char *end = buffer + buflen;
3105 /* these dentries are never renamed, so d_lock is not needed */
3106 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003107 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003108 prepend(&end, &buflen, "/", 1))
3109 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003110 return end;
Al Viro118b2302013-08-24 12:08:17 -04003111}
3112
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003113/*
Ram Pai6092d042008-03-27 13:06:20 +01003114 * Write full pathname from the root of the filesystem into the buffer.
3115 */
Nick Pigginec2447c2011-01-07 17:49:29 +11003116static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003117{
Waiman Long232d2d62013-09-09 12:18:13 -04003118 char *end, *retval;
3119 int len, seq = 0;
3120 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003121
Al Viro48f5ec22013-09-09 15:22:25 -04003122 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003123restart:
3124 end = buf + buflen;
3125 len = buflen;
3126 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003127 if (buflen < 1)
3128 goto Elong;
3129 /* Get '/' right */
3130 retval = end-1;
3131 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003132 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003133 while (!IS_ROOT(dentry)) {
3134 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11003135 int error;
Ram Pai6092d042008-03-27 13:06:20 +01003136
Ram Pai6092d042008-03-27 13:06:20 +01003137 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003138 error = prepend_name(&end, &len, &dentry->d_name);
3139 if (error)
3140 break;
Ram Pai6092d042008-03-27 13:06:20 +01003141
3142 retval = end;
3143 dentry = parent;
3144 }
Al Viro48f5ec22013-09-09 15:22:25 -04003145 if (!(seq & 1))
3146 rcu_read_unlock();
3147 if (need_seqretry(&rename_lock, seq)) {
3148 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003149 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003150 }
3151 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003152 if (error)
3153 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003154 return retval;
3155Elong:
3156 return ERR_PTR(-ENAMETOOLONG);
3157}
Nick Pigginec2447c2011-01-07 17:49:29 +11003158
3159char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3160{
Waiman Long232d2d62013-09-09 12:18:13 -04003161 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003162}
3163EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003164
3165char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3166{
3167 char *p = NULL;
3168 char *retval;
3169
Al Viroc1031352010-06-06 22:31:14 -04003170 if (d_unlinked(dentry)) {
3171 p = buf + buflen;
3172 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3173 goto Elong;
3174 buflen++;
3175 }
3176 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003177 if (!IS_ERR(retval) && p)
3178 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003179 return retval;
3180Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003181 return ERR_PTR(-ENAMETOOLONG);
3182}
3183
Linus Torvalds8b19e342013-09-12 10:35:47 -07003184static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3185 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003186{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003187 unsigned seq;
3188
3189 do {
3190 seq = read_seqcount_begin(&fs->seq);
3191 *root = fs->root;
3192 *pwd = fs->pwd;
3193 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003194}
3195
Ram Pai6092d042008-03-27 13:06:20 +01003196/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 * NOTE! The user-level library version returns a
3198 * character pointer. The kernel system call just
3199 * returns the length of the buffer filled (which
3200 * includes the ending '\0' character), or a negative
3201 * error value. So libc would do something like
3202 *
3203 * char *getcwd(char * buf, size_t size)
3204 * {
3205 * int retval;
3206 *
3207 * retval = sys_getcwd(buf, size);
3208 * if (retval >= 0)
3209 * return buf;
3210 * errno = -retval;
3211 * return NULL;
3212 * }
3213 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003214SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215{
Linus Torvalds552ce542007-02-13 12:08:18 -08003216 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003217 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003218 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003219
3220 if (!page)
3221 return -ENOMEM;
3222
Linus Torvalds8b19e342013-09-12 10:35:47 -07003223 rcu_read_lock();
3224 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003225
Linus Torvalds552ce542007-02-13 12:08:18 -08003226 error = -ENOENT;
Al Viro7ea600b2013-03-26 18:25:57 -04003227 br_read_lock(&vfsmount_lock);
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003228 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003229 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003230 char *cwd = page + PATH_MAX;
3231 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003233 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003234 error = prepend_path(&pwd, &root, &cwd, &buflen);
Al Viro7ea600b2013-03-26 18:25:57 -04003235 br_read_unlock(&vfsmount_lock);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003236 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003237
Al Viro02125a82011-12-05 08:43:34 -05003238 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003239 goto out;
3240
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003241 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003242 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003243 error = prepend_unreachable(&cwd, &buflen);
3244 if (error)
3245 goto out;
3246 }
3247
Linus Torvalds552ce542007-02-13 12:08:18 -08003248 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003249 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003250 if (len <= size) {
3251 error = len;
3252 if (copy_to_user(buf, cwd, len))
3253 error = -EFAULT;
3254 }
Nick Piggin949854d2011-01-07 17:49:37 +11003255 } else {
Al Viro7ea600b2013-03-26 18:25:57 -04003256 br_read_unlock(&vfsmount_lock);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003257 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259
3260out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003261 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 return error;
3263}
3264
3265/*
3266 * Test whether new_dentry is a subdirectory of old_dentry.
3267 *
3268 * Trivially implemented using the dcache structure
3269 */
3270
3271/**
3272 * is_subdir - is new dentry a subdirectory of old_dentry
3273 * @new_dentry: new dentry
3274 * @old_dentry: old dentry
3275 *
3276 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3277 * Returns 0 otherwise.
3278 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3279 */
3280
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003281int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003282{
3283 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003284 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003285
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003286 if (new_dentry == old_dentry)
3287 return 1;
3288
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003289 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003290 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003292 /*
3293 * Need rcu_readlock to protect against the d_parent trashing
3294 * due to d_move
3295 */
3296 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003297 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003299 else
3300 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003301 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303
3304 return result;
3305}
3306
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003307static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003309 struct dentry *root = data;
3310 if (dentry != root) {
3311 if (d_unhashed(dentry) || !dentry->d_inode)
3312 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003314 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3315 dentry->d_flags |= DCACHE_GENOCIDE;
3316 dentry->d_lockref.count--;
3317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003319 return D_WALK_CONTINUE;
3320}
Nick Piggin58db63d2011-01-07 17:49:39 +11003321
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003322void d_genocide(struct dentry *parent)
3323{
3324 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325}
3326
Al Viro60545d02013-06-07 01:20:27 -04003327void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328{
Al Viro60545d02013-06-07 01:20:27 -04003329 inode_dec_link_count(inode);
3330 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3331 !hlist_unhashed(&dentry->d_alias) ||
3332 !d_unlinked(dentry));
3333 spin_lock(&dentry->d_parent->d_lock);
3334 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3335 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3336 (unsigned long long)inode->i_ino);
3337 spin_unlock(&dentry->d_lock);
3338 spin_unlock(&dentry->d_parent->d_lock);
3339 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340}
Al Viro60545d02013-06-07 01:20:27 -04003341EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
3343static __initdata unsigned long dhash_entries;
3344static int __init set_dhash_entries(char *str)
3345{
3346 if (!str)
3347 return 0;
3348 dhash_entries = simple_strtoul(str, &str, 0);
3349 return 1;
3350}
3351__setup("dhash_entries=", set_dhash_entries);
3352
3353static void __init dcache_init_early(void)
3354{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003355 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356
3357 /* If hashes are distributed across NUMA nodes, defer
3358 * hash allocation until vmalloc space is available.
3359 */
3360 if (hashdist)
3361 return;
3362
3363 dentry_hashtable =
3364 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003365 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 dhash_entries,
3367 13,
3368 HASH_EARLY,
3369 &d_hash_shift,
3370 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003371 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372 0);
3373
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003374 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003375 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003376}
3377
Denis Cheng74bf17c2007-10-16 23:26:30 -07003378static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003380 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
3382 /*
3383 * A constructor could be added for stable state like the lists,
3384 * but it is probably not worth it because of the cache nature
3385 * of the dcache.
3386 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003387 dentry_cache = KMEM_CACHE(dentry,
3388 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389
3390 /* Hash may have been set up in dcache_init_early */
3391 if (!hashdist)
3392 return;
3393
3394 dentry_hashtable =
3395 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003396 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397 dhash_entries,
3398 13,
3399 0,
3400 &d_hash_shift,
3401 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003402 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 0);
3404
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003405 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003406 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407}
3408
3409/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003410struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003411EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413EXPORT_SYMBOL(d_genocide);
3414
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415void __init vfs_caches_init_early(void)
3416{
3417 dcache_init_early();
3418 inode_init_early();
3419}
3420
3421void __init vfs_caches_init(unsigned long mempages)
3422{
3423 unsigned long reserve;
3424
3425 /* Base hash sizes on available memory, with a reserve equal to
3426 150% of current kernel size */
3427
3428 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3429 mempages -= reserve;
3430
3431 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003432 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433
Denis Cheng74bf17c2007-10-16 23:26:30 -07003434 dcache_init();
3435 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003437 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 bdev_cache_init();
3439 chrdev_init();
3440}