blob: 0a38ef8d7f0088579089d101c99a0e12193f22f5 [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;
David Howellsb18825a2013-09-12 19:22:53 +0100346 __d_clear_type(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100347 dentry->d_inode = NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -0400348 hlist_del_init(&dentry->d_alias);
Nick Piggin31e6b012011-01-07 17:49:52 +1100349 dentry_rcuwalk_barrier(dentry);
350 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100351 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100352 if (!inode->i_nlink)
353 fsnotify_inoderemove(inode);
354 if (dentry->d_op && dentry->d_op->d_iput)
355 dentry->d_op->d_iput(dentry, inode);
356 else
357 iput(inode);
358}
359
360/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400361 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
362 * is in use - which includes both the "real" per-superblock
363 * LRU list _and_ the DCACHE_SHRINK_LIST use.
364 *
365 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
366 * on the shrink list (ie not on the superblock LRU list).
367 *
368 * The per-cpu "nr_dentry_unused" counters are updated with
369 * the DCACHE_LRU_LIST bit.
370 *
371 * These helper functions make sure we always follow the
372 * rules. d_lock must be held by the caller.
373 */
374#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
375static void d_lru_add(struct dentry *dentry)
376{
377 D_FLAG_VERIFY(dentry, 0);
378 dentry->d_flags |= DCACHE_LRU_LIST;
379 this_cpu_inc(nr_dentry_unused);
380 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
381}
382
383static void d_lru_del(struct dentry *dentry)
384{
385 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
386 dentry->d_flags &= ~DCACHE_LRU_LIST;
387 this_cpu_dec(nr_dentry_unused);
388 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
389}
390
391static void d_shrink_del(struct dentry *dentry)
392{
393 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
394 list_del_init(&dentry->d_lru);
395 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
396 this_cpu_dec(nr_dentry_unused);
397}
398
399static void d_shrink_add(struct dentry *dentry, struct list_head *list)
400{
401 D_FLAG_VERIFY(dentry, 0);
402 list_add(&dentry->d_lru, list);
403 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
404 this_cpu_inc(nr_dentry_unused);
405}
406
407/*
408 * These can only be called under the global LRU lock, ie during the
409 * callback for freeing the LRU list. "isolate" removes it from the
410 * LRU lists entirely, while shrink_move moves it to the indicated
411 * private list.
412 */
413static void d_lru_isolate(struct dentry *dentry)
414{
415 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
416 dentry->d_flags &= ~DCACHE_LRU_LIST;
417 this_cpu_dec(nr_dentry_unused);
418 list_del_init(&dentry->d_lru);
419}
420
421static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
422{
423 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
424 dentry->d_flags |= DCACHE_SHRINK_LIST;
425 list_move_tail(&dentry->d_lru, list);
426}
427
428/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000429 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700430 */
431static void dentry_lru_add(struct dentry *dentry)
432{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400433 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
434 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700435}
436
Sage Weilf0023bc2011-10-28 10:02:42 -0700437/*
438 * Remove a dentry with references from the LRU.
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000439 *
440 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
441 * lose our last reference through the parent walk. In this case, we need to
442 * remove ourselves from the shrink list, not the LRU.
Sage Weilf0023bc2011-10-28 10:02:42 -0700443 */
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700444static void dentry_lru_del(struct dentry *dentry)
445{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400446 if (dentry->d_flags & DCACHE_LRU_LIST) {
447 if (dentry->d_flags & DCACHE_SHRINK_LIST)
448 return d_shrink_del(dentry);
449 d_lru_del(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700450 }
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700451}
452
Miklos Szeredid52b9082007-05-08 00:23:46 -0700453/**
454 * d_kill - kill dentry and return parent
455 * @dentry: dentry to kill
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800456 * @parent: parent dentry
Miklos Szeredid52b9082007-05-08 00:23:46 -0700457 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200458 * The dentry must already be unhashed and removed from the LRU.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700459 *
460 * If this is the root of the dentry tree, return NULL.
Nick Piggin23044502011-01-07 17:49:31 +1100461 *
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100462 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
463 * d_kill.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700464 */
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100465static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200466 __releases(dentry->d_lock)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100467 __releases(parent->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100468 __releases(dentry->d_inode->i_lock)
Miklos Szeredid52b9082007-05-08 00:23:46 -0700469{
Miklos Szeredid52b9082007-05-08 00:23:46 -0700470 list_del(&dentry->d_u.d_child);
Trond Myklebustc83ce982011-03-15 13:36:43 -0400471 /*
472 * Inform try_to_ascend() that we are no longer attached to the
473 * dentry tree
474 */
Miklos Szeredib161dfa62012-09-17 22:31:38 +0200475 dentry->d_flags |= DCACHE_DENTRY_KILLED;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +1100476 if (parent)
477 spin_unlock(&parent->d_lock);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700478 dentry_iput(dentry);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100479 /*
480 * dentry_iput drops the locks, at which point nobody (except
481 * transient RCU lookups) can reach this dentry.
482 */
Miklos Szeredid52b9082007-05-08 00:23:46 -0700483 d_free(dentry);
OGAWA Hirofumi871c0062008-10-16 07:50:27 +0900484 return parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700485}
486
Nick Piggin789680d2011-01-07 17:49:30 +1100487/**
488 * d_drop - drop a dentry
489 * @dentry: dentry to drop
490 *
491 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
492 * be found through a VFS lookup any more. Note that this is different from
493 * deleting the dentry - d_delete will try to mark the dentry negative if
494 * possible, giving a successful _negative_ lookup, while d_drop will
495 * just make the cache lookup fail.
496 *
497 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
498 * reason (NFS timeouts or autofs deletes).
499 *
500 * __d_drop requires dentry->d_lock.
501 */
502void __d_drop(struct dentry *dentry)
503{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700504 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400505 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400506 /*
507 * Hashed dentries are normally on the dentry hashtable,
508 * with the exception of those newly allocated by
509 * d_obtain_alias, which are always IS_ROOT:
510 */
511 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400512 b = &dentry->d_sb->s_anon;
513 else
514 b = d_hash(dentry->d_parent, dentry->d_name.hash);
515
516 hlist_bl_lock(b);
517 __hlist_bl_del(&dentry->d_hash);
518 dentry->d_hash.pprev = NULL;
519 hlist_bl_unlock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -0700520 dentry_rcuwalk_barrier(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100521 }
522}
523EXPORT_SYMBOL(__d_drop);
524
525void d_drop(struct dentry *dentry)
526{
Nick Piggin789680d2011-01-07 17:49:30 +1100527 spin_lock(&dentry->d_lock);
528 __d_drop(dentry);
529 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100530}
531EXPORT_SYMBOL(d_drop);
532
Nick Piggin77812a12011-01-07 17:49:48 +1100533/*
534 * Finish off a dentry we've decided to kill.
535 * dentry->d_lock must be held, returns with it unlocked.
536 * If ref is non-zero, then decrement the refcount too.
537 * Returns dentry requiring refcount drop, or NULL if we're done.
538 */
Linus Torvalds358eec12013-10-31 15:43:02 -0700539static struct dentry *
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000540dentry_kill(struct dentry *dentry, int unlock_on_failure)
Nick Piggin77812a12011-01-07 17:49:48 +1100541 __releases(dentry->d_lock)
542{
Nick Piggin873feea2011-01-07 17:50:06 +1100543 struct inode *inode;
Nick Piggin77812a12011-01-07 17:49:48 +1100544 struct dentry *parent;
545
Nick Piggin873feea2011-01-07 17:50:06 +1100546 inode = dentry->d_inode;
547 if (inode && !spin_trylock(&inode->i_lock)) {
Nick Piggin77812a12011-01-07 17:49:48 +1100548relock:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000549 if (unlock_on_failure) {
550 spin_unlock(&dentry->d_lock);
551 cpu_relax();
552 }
Nick Piggin77812a12011-01-07 17:49:48 +1100553 return dentry; /* try again with same dentry */
554 }
555 if (IS_ROOT(dentry))
556 parent = NULL;
557 else
558 parent = dentry->d_parent;
559 if (parent && !spin_trylock(&parent->d_lock)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100560 if (inode)
561 spin_unlock(&inode->i_lock);
Nick Piggin77812a12011-01-07 17:49:48 +1100562 goto relock;
563 }
Nick Piggin31e6b012011-01-07 17:49:52 +1100564
Linus Torvalds0d984392013-09-08 13:46:52 -0700565 /*
566 * The dentry is now unrecoverably dead to the world.
567 */
568 lockref_mark_dead(&dentry->d_lockref);
569
Sage Weilf0023bc2011-10-28 10:02:42 -0700570 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700571 * inform the fs via d_prune that this dentry is about to be
572 * unhashed and destroyed.
573 */
Yan, Zheng590fb512013-08-13 15:42:02 +0800574 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
Yan, Zheng61572bb2013-04-15 14:13:21 +0800575 dentry->d_op->d_prune(dentry);
576
577 dentry_lru_del(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100578 /* if it was on the hash then remove it */
579 __d_drop(dentry);
580 return d_kill(dentry, parent);
581}
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583/*
584 * This is dput
585 *
586 * This is complicated by the fact that we do not want to put
587 * dentries that are no longer on any hash chain on the unused
588 * list: we'd much rather just get rid of them immediately.
589 *
590 * However, that implies that we have to traverse the dentry
591 * tree upwards to the parents which might _also_ now be
592 * scheduled for deletion (it may have been only waiting for
593 * its last child to go away).
594 *
595 * This tail recursion is done by hand as we don't want to depend
596 * on the compiler to always get this right (gcc generally doesn't).
597 * Real recursion would eat up our stack space.
598 */
599
600/*
601 * dput - release a dentry
602 * @dentry: dentry to release
603 *
604 * Release a dentry. This will drop the usage count and if appropriate
605 * call the dentry unlink method as well as removing it from the queues and
606 * releasing its resources. If the parent dentries were scheduled for release
607 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609void dput(struct dentry *dentry)
610{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700611 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 return;
613
614repeat:
Waiman Long98474232013-08-28 18:24:59 -0700615 if (lockref_put_or_lock(&dentry->d_lockref))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700618 /* Unreachable? Get rid of it */
619 if (unlikely(d_unhashed(dentry)))
620 goto kill_it;
621
622 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100624 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
Nick Piggin265ac902010-10-10 05:36:24 -0400626
Linus Torvalds358eec12013-10-31 15:43:02 -0700627 if (!(dentry->d_flags & DCACHE_REFERENCED))
628 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400629 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400630
Waiman Long98474232013-08-28 18:24:59 -0700631 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100632 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return;
634
Miklos Szeredid52b9082007-05-08 00:23:46 -0700635kill_it:
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000636 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700637 if (dentry)
638 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700640EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642/**
643 * d_invalidate - invalidate a dentry
644 * @dentry: dentry to invalidate
645 *
646 * Try to invalidate the dentry if it turns out to be
647 * possible. If there are other dentries that can be
648 * reached through this one we can't delete it and we
649 * return -EBUSY. On success we return 0.
650 *
651 * no dcache lock.
652 */
653
654int d_invalidate(struct dentry * dentry)
655{
656 /*
657 * If it's already been dropped, return OK.
658 */
Nick Pigginda502952011-01-07 17:49:33 +1100659 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 if (d_unhashed(dentry)) {
Nick Pigginda502952011-01-07 17:49:33 +1100661 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return 0;
663 }
664 /*
665 * Check whether to do a partial shrink_dcache
666 * to get rid of unused child entries.
667 */
668 if (!list_empty(&dentry->d_subdirs)) {
Nick Pigginda502952011-01-07 17:49:33 +1100669 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 shrink_dcache_parent(dentry);
Nick Pigginda502952011-01-07 17:49:33 +1100671 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673
674 /*
675 * Somebody else still using it?
676 *
677 * If it's a directory, we can't drop it
678 * for fear of somebody re-populating it
679 * with children (even though dropping it
680 * would make it unreachable from the root,
681 * we might still populate it if it was a
682 * working directory or similar).
Al Viro50e69632011-11-07 16:39:57 +0000683 * We also need to leave mountpoints alone,
684 * directory or not.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 */
Waiman Long98474232013-08-28 18:24:59 -0700686 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
Al Viro50e69632011-11-07 16:39:57 +0000687 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 return -EBUSY;
690 }
691 }
692
693 __d_drop(dentry);
694 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return 0;
696}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700697EXPORT_SYMBOL(d_invalidate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100699/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100700static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
Waiman Long98474232013-08-28 18:24:59 -0700702 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703}
704
Nick Piggindc0474b2011-01-07 17:49:43 +1100705static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100706{
Waiman Long98474232013-08-28 18:24:59 -0700707 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100708}
709
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100710struct dentry *dget_parent(struct dentry *dentry)
711{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700712 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100713 struct dentry *ret;
714
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700715 /*
716 * Do optimistic parent lookup without any
717 * locking.
718 */
719 rcu_read_lock();
720 ret = ACCESS_ONCE(dentry->d_parent);
721 gotref = lockref_get_not_zero(&ret->d_lockref);
722 rcu_read_unlock();
723 if (likely(gotref)) {
724 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
725 return ret;
726 dput(ret);
727 }
728
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100729repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100730 /*
731 * Don't need rcu_dereference because we re-check it was correct under
732 * the lock.
733 */
734 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100735 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100736 spin_lock(&ret->d_lock);
737 if (unlikely(ret != dentry->d_parent)) {
738 spin_unlock(&ret->d_lock);
739 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100740 goto repeat;
741 }
Nick Piggina734eb42011-01-07 17:49:44 +1100742 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700743 BUG_ON(!ret->d_lockref.count);
744 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100745 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100746 return ret;
747}
748EXPORT_SYMBOL(dget_parent);
749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750/**
751 * d_find_alias - grab a hashed alias of inode
752 * @inode: inode in question
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700753 * @want_discon: flag, used by d_splice_alias, to request
754 * that only a DISCONNECTED alias be returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 *
756 * If inode has a hashed alias, or is a directory and has any alias,
757 * acquire the reference to alias and return it. Otherwise return NULL.
758 * Notice that if inode is a directory there can be only one alias and
759 * it can be unhashed only if it has no children, or if it is the root
760 * of a filesystem.
761 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700762 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700763 * any other hashed alias over that one unless @want_discon is set,
764 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 */
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700766static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767{
Nick Pigginda502952011-01-07 17:49:33 +1100768 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Nick Pigginda502952011-01-07 17:49:33 +1100770again:
771 discon_alias = NULL;
Sasha Levinb67bfe02013-02-27 17:06:00 -0800772 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100773 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700775 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100776 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 discon_alias = alias;
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700778 } else if (!want_discon) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100779 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100780 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 return alias;
782 }
783 }
Nick Pigginda502952011-01-07 17:49:33 +1100784 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
Nick Pigginda502952011-01-07 17:49:33 +1100786 if (discon_alias) {
787 alias = discon_alias;
788 spin_lock(&alias->d_lock);
789 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
790 if (IS_ROOT(alias) &&
791 (alias->d_flags & DCACHE_DISCONNECTED)) {
Nick Piggindc0474b2011-01-07 17:49:43 +1100792 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100793 spin_unlock(&alias->d_lock);
794 return alias;
795 }
796 }
797 spin_unlock(&alias->d_lock);
798 goto again;
799 }
800 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801}
802
Nick Pigginda502952011-01-07 17:49:33 +1100803struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804{
David Howells214fda12006-03-25 03:06:36 -0800805 struct dentry *de = NULL;
806
Al Virob3d9b7a2012-06-09 13:51:19 -0400807 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100808 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -0700809 de = __d_find_alias(inode, 0);
Nick Piggin873feea2011-01-07 17:50:06 +1100810 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return de;
813}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700814EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816/*
817 * Try to kill dentries associated with this inode.
818 * WARNING: you must own a reference to inode.
819 */
820void d_prune_aliases(struct inode *inode)
821{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700822 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100824 spin_lock(&inode->i_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800825 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700827 if (!dentry->d_lockref.count) {
Yan, Zheng590fb512013-08-13 15:42:02 +0800828 /*
829 * inform the fs via d_prune that this dentry
830 * is about to be unhashed and destroyed.
831 */
832 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
833 !d_unhashed(dentry))
834 dentry->d_op->d_prune(dentry);
835
Nick Piggindc0474b2011-01-07 17:49:43 +1100836 __dget_dlock(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 __d_drop(dentry);
838 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100839 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 dput(dentry);
841 goto restart;
842 }
843 spin_unlock(&dentry->d_lock);
844 }
Nick Piggin873feea2011-01-07 17:50:06 +1100845 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700847EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849/*
Nick Piggin77812a12011-01-07 17:49:48 +1100850 * Try to throw away a dentry - free the inode, dput the parent.
851 * Requires dentry->d_lock is held, and dentry->d_count == 0.
852 * Releases dentry->d_lock.
Andrew Mortond702ccb2006-06-22 14:47:31 -0700853 *
Nick Piggin77812a12011-01-07 17:49:48 +1100854 * This may fail if locks cannot be acquired no problem, just try again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000856static struct dentry * try_prune_one_dentry(struct dentry *dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200857 __releases(dentry->d_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Nick Piggin77812a12011-01-07 17:49:48 +1100859 struct dentry *parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700860
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000861 parent = dentry_kill(dentry, 0);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700862 /*
Nick Piggin77812a12011-01-07 17:49:48 +1100863 * If dentry_kill returns NULL, we have nothing more to do.
864 * if it returns the same dentry, trylocks failed. In either
865 * case, just loop again.
866 *
867 * Otherwise, we need to prune ancestors too. This is necessary
868 * to prevent quadratic behavior of shrink_dcache_parent(), but
869 * is also expected to be beneficial in reducing dentry cache
870 * fragmentation.
Miklos Szeredid52b9082007-05-08 00:23:46 -0700871 */
Nick Piggin77812a12011-01-07 17:49:48 +1100872 if (!parent)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000873 return NULL;
Nick Piggin77812a12011-01-07 17:49:48 +1100874 if (parent == dentry)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000875 return dentry;
Nick Piggin77812a12011-01-07 17:49:48 +1100876
877 /* Prune ancestors. */
878 dentry = parent;
Miklos Szeredid52b9082007-05-08 00:23:46 -0700879 while (dentry) {
Waiman Long98474232013-08-28 18:24:59 -0700880 if (lockref_put_or_lock(&dentry->d_lockref))
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000881 return NULL;
882 dentry = dentry_kill(dentry, 1);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700883 }
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000884 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400887static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888{
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700889 struct dentry *dentry;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700890
Nick Pigginec336792011-01-07 17:49:47 +1100891 rcu_read_lock();
892 for (;;) {
Nick Pigginec336792011-01-07 17:49:47 +1100893 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
894 if (&dentry->d_lru == list)
895 break; /* empty */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400896
897 /*
898 * Get the dentry lock, and re-verify that the dentry is
899 * this on the shrinking list. If it is, we know that
900 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
901 */
Nick Pigginec336792011-01-07 17:49:47 +1100902 spin_lock(&dentry->d_lock);
903 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
904 spin_unlock(&dentry->d_lock);
Nick Piggin23044502011-01-07 17:49:31 +1100905 continue;
906 }
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000909 * The dispose list is isolated and dentries are not accounted
910 * to the LRU here, so we can simply remove it from the list
911 * here regardless of whether it is referenced or not.
912 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400913 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000914
915 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000917 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 */
Waiman Long98474232013-08-28 18:24:59 -0700919 if (dentry->d_lockref.count) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700920 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 continue;
922 }
Nick Pigginec336792011-01-07 17:49:47 +1100923 rcu_read_unlock();
Nick Piggin77812a12011-01-07 17:49:48 +1100924
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400925 /*
926 * If 'try_to_prune()' returns a dentry, it will
927 * be the same one we passed in, and d_lock will
928 * have been held the whole time, so it will not
929 * have been added to any other lists. We failed
930 * to get the inode lock.
931 *
932 * We just add it back to the shrink list.
933 */
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000934 dentry = try_prune_one_dentry(dentry);
Nick Piggin77812a12011-01-07 17:49:48 +1100935
Nick Pigginec336792011-01-07 17:49:47 +1100936 rcu_read_lock();
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000937 if (dentry) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400938 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000939 spin_unlock(&dentry->d_lock);
940 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 }
Nick Pigginec336792011-01-07 17:49:47 +1100942 rcu_read_unlock();
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400943}
944
Dave Chinnerf6041562013-08-28 10:18:00 +1000945static enum lru_status
946dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
947{
948 struct list_head *freeable = arg;
949 struct dentry *dentry = container_of(item, struct dentry, d_lru);
950
951
952 /*
953 * we are inverting the lru lock/dentry->d_lock here,
954 * so use a trylock. If we fail to get the lock, just skip
955 * it
956 */
957 if (!spin_trylock(&dentry->d_lock))
958 return LRU_SKIP;
959
960 /*
961 * Referenced dentries are still in use. If they have active
962 * counts, just remove them from the LRU. Otherwise give them
963 * another pass through the LRU.
964 */
965 if (dentry->d_lockref.count) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400966 d_lru_isolate(dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +1000967 spin_unlock(&dentry->d_lock);
968 return LRU_REMOVED;
969 }
970
971 if (dentry->d_flags & DCACHE_REFERENCED) {
972 dentry->d_flags &= ~DCACHE_REFERENCED;
973 spin_unlock(&dentry->d_lock);
974
975 /*
976 * The list move itself will be made by the common LRU code. At
977 * this point, we've dropped the dentry->d_lock but keep the
978 * lru lock. This is safe to do, since every list movement is
979 * protected by the lru lock even if both locks are held.
980 *
981 * This is guaranteed by the fact that all LRU management
982 * functions are intermediated by the LRU API calls like
983 * list_lru_add and list_lru_del. List movement in this file
984 * only ever occur through this functions or through callbacks
985 * like this one, that are called from the LRU API.
986 *
987 * The only exceptions to this are functions like
988 * shrink_dentry_list, and code that first checks for the
989 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
990 * operating only with stack provided lists after they are
991 * properly isolated from the main list. It is thus, always a
992 * local access.
993 */
994 return LRU_ROTATE;
995 }
996
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400997 d_lru_shrink_move(dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +1000998 spin_unlock(&dentry->d_lock);
999
1000 return LRU_REMOVED;
1001}
1002
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001003/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001004 * prune_dcache_sb - shrink the dcache
1005 * @sb: superblock
Dave Chinnerf6041562013-08-28 10:18:00 +10001006 * @nr_to_scan : number of entries to try to free
Dave Chinner9b17c622013-08-28 10:18:05 +10001007 * @nid: which node to scan for freeable entities
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001008 *
Dave Chinnerf6041562013-08-28 10:18:00 +10001009 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001010 * done when we need more memory an called from the superblock shrinker
1011 * function.
1012 *
1013 * This function may fail to free any resources if all the dentries are in
1014 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001015 */
Dave Chinner9b17c622013-08-28 10:18:05 +10001016long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
1017 int nid)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001018{
Dave Chinnerf6041562013-08-28 10:18:00 +10001019 LIST_HEAD(dispose);
1020 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001021
Dave Chinner9b17c622013-08-28 10:18:05 +10001022 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
1023 &dispose, &nr_to_scan);
Dave Chinnerf6041562013-08-28 10:18:00 +10001024 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001025 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026}
1027
Glauber Costa4e717f52013-08-28 10:18:03 +10001028static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1029 spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001030{
Glauber Costa4e717f52013-08-28 10:18:03 +10001031 struct list_head *freeable = arg;
1032 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001033
Glauber Costa4e717f52013-08-28 10:18:03 +10001034 /*
1035 * we are inverting the lru lock/dentry->d_lock here,
1036 * so use a trylock. If we fail to get the lock, just skip
1037 * it
1038 */
1039 if (!spin_trylock(&dentry->d_lock))
1040 return LRU_SKIP;
1041
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001042 d_lru_shrink_move(dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001043 spin_unlock(&dentry->d_lock);
1044
1045 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001046}
1047
Glauber Costa4e717f52013-08-28 10:18:03 +10001048
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001049/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 * shrink_dcache_sb - shrink dcache for a superblock
1051 * @sb: superblock
1052 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001053 * Shrink the dcache for the specified super block. This is used to free
1054 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001056void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057{
Glauber Costa4e717f52013-08-28 10:18:03 +10001058 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001059
Glauber Costa4e717f52013-08-28 10:18:03 +10001060 do {
1061 LIST_HEAD(dispose);
1062
1063 freed = list_lru_walk(&sb->s_dentry_lru,
1064 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1065
1066 this_cpu_sub(nr_dentry_unused, freed);
1067 shrink_dentry_list(&dispose);
1068 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001070EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
1072/*
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001073 * This tries to ascend one level of parenthood, but
1074 * we can race with renaming, so we need to re-check
1075 * the parenthood after dropping the lock and check
1076 * that the sequence number still matches.
1077 */
Al Viro48f5ec22013-09-09 15:22:25 -04001078static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001079{
1080 struct dentry *new = old->d_parent;
1081
1082 rcu_read_lock();
1083 spin_unlock(&old->d_lock);
1084 spin_lock(&new->d_lock);
1085
1086 /*
1087 * might go back up the wrong parent if we have had a rename
1088 * or deletion
1089 */
1090 if (new != old->d_parent ||
Miklos Szeredib161dfa62012-09-17 22:31:38 +02001091 (old->d_flags & DCACHE_DENTRY_KILLED) ||
Al Viro48f5ec22013-09-09 15:22:25 -04001092 need_seqretry(&rename_lock, seq)) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001093 spin_unlock(&new->d_lock);
1094 new = NULL;
1095 }
1096 rcu_read_unlock();
1097 return new;
1098}
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001101 * enum d_walk_ret - action to talke during tree walk
1102 * @D_WALK_CONTINUE: contrinue walk
1103 * @D_WALK_QUIT: quit walk
1104 * @D_WALK_NORETRY: quit when retry is needed
1105 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001107enum d_walk_ret {
1108 D_WALK_CONTINUE,
1109 D_WALK_QUIT,
1110 D_WALK_NORETRY,
1111 D_WALK_SKIP,
1112};
1113
1114/**
1115 * d_walk - walk the dentry tree
1116 * @parent: start of walk
1117 * @data: data passed to @enter() and @finish()
1118 * @enter: callback when first entering the dentry
1119 * @finish: callback when successfully finished the walk
1120 *
1121 * The @enter() and @finish() callbacks are called with d_lock held.
1122 */
1123static void d_walk(struct dentry *parent, void *data,
1124 enum d_walk_ret (*enter)(void *, struct dentry *),
1125 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126{
Nick Piggin949854d2011-01-07 17:49:37 +11001127 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001129 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001130 enum d_walk_ret ret;
1131 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001132
Nick Piggin58db63d2011-01-07 17:49:39 +11001133again:
Al Viro48f5ec22013-09-09 15:22:25 -04001134 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001135 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001136 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001137
1138 ret = enter(data, this_parent);
1139 switch (ret) {
1140 case D_WALK_CONTINUE:
1141 break;
1142 case D_WALK_QUIT:
1143 case D_WALK_SKIP:
1144 goto out_unlock;
1145 case D_WALK_NORETRY:
1146 retry = false;
1147 break;
1148 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149repeat:
1150 next = this_parent->d_subdirs.next;
1151resume:
1152 while (next != &this_parent->d_subdirs) {
1153 struct list_head *tmp = next;
Eric Dumazet5160ee62006-01-08 01:03:32 -08001154 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001156
1157 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001158
1159 ret = enter(data, dentry);
1160 switch (ret) {
1161 case D_WALK_CONTINUE:
1162 break;
1163 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001164 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001165 goto out_unlock;
1166 case D_WALK_NORETRY:
1167 retry = false;
1168 break;
1169 case D_WALK_SKIP:
1170 spin_unlock(&dentry->d_lock);
1171 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001172 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001173
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001175 spin_unlock(&this_parent->d_lock);
1176 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001178 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 goto repeat;
1180 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001181 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
1183 /*
1184 * All done at this level ... ascend and resume the search.
1185 */
1186 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001187 struct dentry *child = this_parent;
Al Viro48f5ec22013-09-09 15:22:25 -04001188 this_parent = try_to_ascend(this_parent, seq);
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001189 if (!this_parent)
Nick Piggin949854d2011-01-07 17:49:37 +11001190 goto rename_retry;
Nick Piggin949854d2011-01-07 17:49:37 +11001191 next = child->d_u.d_child.next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 goto resume;
1193 }
Al Viro48f5ec22013-09-09 15:22:25 -04001194 if (need_seqretry(&rename_lock, seq)) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001195 spin_unlock(&this_parent->d_lock);
1196 goto rename_retry;
1197 }
1198 if (finish)
1199 finish(data);
1200
1201out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001202 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001203 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001204 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001205
1206rename_retry:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001207 if (!retry)
1208 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001209 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001210 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001212
1213/*
1214 * Search for at least 1 mount point in the dentry's subdirs.
1215 * We descend to the next level whenever the d_subdirs
1216 * list is non-empty and continue searching.
1217 */
1218
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001219static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1220{
1221 int *ret = data;
1222 if (d_mountpoint(dentry)) {
1223 *ret = 1;
1224 return D_WALK_QUIT;
1225 }
1226 return D_WALK_CONTINUE;
1227}
1228
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001229/**
1230 * have_submounts - check for mounts over a dentry
1231 * @parent: dentry to check.
1232 *
1233 * Return true if the parent or its subdirectories contain
1234 * a mount point
1235 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001236int have_submounts(struct dentry *parent)
1237{
1238 int ret = 0;
1239
1240 d_walk(parent, &ret, check_mount, NULL);
1241
1242 return ret;
1243}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001244EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
1246/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001247 * Called by mount code to set a mountpoint and check if the mountpoint is
1248 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1249 * subtree can become unreachable).
1250 *
1251 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1252 * this reason take rename_lock and d_lock on dentry and ancestors.
1253 */
1254int d_set_mounted(struct dentry *dentry)
1255{
1256 struct dentry *p;
1257 int ret = -ENOENT;
1258 write_seqlock(&rename_lock);
1259 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1260 /* Need exclusion wrt. check_submounts_and_drop() */
1261 spin_lock(&p->d_lock);
1262 if (unlikely(d_unhashed(p))) {
1263 spin_unlock(&p->d_lock);
1264 goto out;
1265 }
1266 spin_unlock(&p->d_lock);
1267 }
1268 spin_lock(&dentry->d_lock);
1269 if (!d_unlinked(dentry)) {
1270 dentry->d_flags |= DCACHE_MOUNTED;
1271 ret = 0;
1272 }
1273 spin_unlock(&dentry->d_lock);
1274out:
1275 write_sequnlock(&rename_lock);
1276 return ret;
1277}
1278
1279/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001280 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 * and move any unused dentries to the end of the unused
1282 * list for prune_dcache(). We descend to the next level
1283 * whenever the d_subdirs list is non-empty and continue
1284 * searching.
1285 *
1286 * It returns zero iff there are no unused children,
1287 * otherwise it returns the number of children moved to
1288 * the end of the unused list. This may not be the total
1289 * number of unused children, because select_parent can
1290 * drop the lock and return early due to latency
1291 * constraints.
1292 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001293
1294struct select_data {
1295 struct dentry *start;
1296 struct list_head dispose;
1297 int found;
1298};
1299
1300static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001302 struct select_data *data = _data;
1303 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001305 if (data->start == dentry)
1306 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001308 /*
1309 * move only zero ref count dentries to the dispose list.
1310 *
1311 * Those which are presently on the shrink list, being processed
1312 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1313 * loop in shrink_dcache_parent() might not make any progress
1314 * and loop forever.
1315 */
1316 if (dentry->d_lockref.count) {
1317 dentry_lru_del(dentry);
1318 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001319 /*
1320 * We can't use d_lru_shrink_move() because we
1321 * need to get the global LRU lock and do the
Linus Torvalds05a82522013-09-15 07:11:01 -04001322 * LRU accounting.
Linus Torvalds89dc77b2013-09-13 22:55:10 -04001323 */
1324 d_lru_del(dentry);
1325 d_shrink_add(dentry, &data->dispose);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001326 data->found++;
1327 ret = D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 }
1329 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001330 * We can return to the caller if we have found some (this
1331 * ensures forward progress). We'll be coming back to find
1332 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001334 if (data->found && need_resched())
1335 ret = D_WALK_QUIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001337 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338}
1339
1340/**
1341 * shrink_dcache_parent - prune dcache
1342 * @parent: parent of entries to prune
1343 *
1344 * Prune the dcache to remove unused children of the parent dentry.
1345 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001346void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001348 for (;;) {
1349 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001351 INIT_LIST_HEAD(&data.dispose);
1352 data.start = parent;
1353 data.found = 0;
1354
1355 d_walk(parent, &data, select_collect, NULL);
1356 if (!data.found)
1357 break;
1358
1359 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001360 cond_resched();
1361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001363EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Al Viro42c32602013-11-08 12:31:16 -05001365static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1366{
1367 struct select_data *data = _data;
1368 enum d_walk_ret ret = D_WALK_CONTINUE;
1369
1370 if (dentry->d_lockref.count) {
1371 dentry_lru_del(dentry);
1372 if (likely(!list_empty(&dentry->d_subdirs)))
1373 goto out;
1374 if (dentry == data->start && dentry->d_lockref.count == 1)
1375 goto out;
1376 printk(KERN_ERR
1377 "BUG: Dentry %p{i=%lx,n=%s}"
1378 " still in use (%d)"
1379 " [unmount of %s %s]\n",
1380 dentry,
1381 dentry->d_inode ?
1382 dentry->d_inode->i_ino : 0UL,
1383 dentry->d_name.name,
1384 dentry->d_lockref.count,
1385 dentry->d_sb->s_type->name,
1386 dentry->d_sb->s_id);
1387 BUG();
1388 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1389 /*
1390 * We can't use d_lru_shrink_move() because we
1391 * need to get the global LRU lock and do the
1392 * LRU accounting.
1393 */
1394 if (dentry->d_flags & DCACHE_LRU_LIST)
1395 d_lru_del(dentry);
1396 d_shrink_add(dentry, &data->dispose);
1397 data->found++;
1398 ret = D_WALK_NORETRY;
1399 }
1400out:
1401 if (data->found && need_resched())
1402 ret = D_WALK_QUIT;
1403 return ret;
1404}
1405
1406/*
1407 * destroy the dentries attached to a superblock on unmounting
1408 */
1409void shrink_dcache_for_umount(struct super_block *sb)
1410{
1411 struct dentry *dentry;
1412
1413 if (down_read_trylock(&sb->s_umount))
1414 BUG();
1415
1416 dentry = sb->s_root;
1417 sb->s_root = NULL;
1418 for (;;) {
1419 struct select_data data;
1420
1421 INIT_LIST_HEAD(&data.dispose);
1422 data.start = dentry;
1423 data.found = 0;
1424
1425 d_walk(dentry, &data, umount_collect, NULL);
1426 if (!data.found)
1427 break;
1428
1429 shrink_dentry_list(&data.dispose);
1430 cond_resched();
1431 }
1432 d_drop(dentry);
1433 dput(dentry);
1434
1435 while (!hlist_bl_empty(&sb->s_anon)) {
1436 struct select_data data;
1437 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1438
1439 INIT_LIST_HEAD(&data.dispose);
1440 data.start = NULL;
1441 data.found = 0;
1442
1443 d_walk(dentry, &data, umount_collect, NULL);
1444 if (data.found)
1445 shrink_dentry_list(&data.dispose);
1446 cond_resched();
1447 }
1448}
1449
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001450static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1451{
1452 struct select_data *data = _data;
1453
1454 if (d_mountpoint(dentry)) {
1455 data->found = -EBUSY;
1456 return D_WALK_QUIT;
1457 }
1458
1459 return select_collect(_data, dentry);
1460}
1461
1462static void check_and_drop(void *_data)
1463{
1464 struct select_data *data = _data;
1465
1466 if (d_mountpoint(data->start))
1467 data->found = -EBUSY;
1468 if (!data->found)
1469 __d_drop(data->start);
1470}
1471
1472/**
1473 * check_submounts_and_drop - prune dcache, check for submounts and drop
1474 *
1475 * All done as a single atomic operation relative to has_unlinked_ancestor().
1476 * Returns 0 if successfully unhashed @parent. If there were submounts then
1477 * return -EBUSY.
1478 *
1479 * @dentry: dentry to prune and drop
1480 */
1481int check_submounts_and_drop(struct dentry *dentry)
1482{
1483 int ret = 0;
1484
1485 /* Negative dentries can be dropped without further checks */
1486 if (!dentry->d_inode) {
1487 d_drop(dentry);
1488 goto out;
1489 }
1490
1491 for (;;) {
1492 struct select_data data;
1493
1494 INIT_LIST_HEAD(&data.dispose);
1495 data.start = dentry;
1496 data.found = 0;
1497
1498 d_walk(dentry, &data, check_and_collect, check_and_drop);
1499 ret = data.found;
1500
1501 if (!list_empty(&data.dispose))
1502 shrink_dentry_list(&data.dispose);
1503
1504 if (ret <= 0)
1505 break;
1506
1507 cond_resched();
1508 }
1509
1510out:
1511 return ret;
1512}
1513EXPORT_SYMBOL(check_submounts_and_drop);
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515/**
Al Viroa4464db2011-07-07 15:03:58 -04001516 * __d_alloc - allocate a dcache entry
1517 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 * @name: qstr of the name
1519 *
1520 * Allocates a dentry. It returns %NULL if there is insufficient memory
1521 * available. On a success the dentry is returned. The name passed in is
1522 * copied and the copy passed in may be reused after this call.
1523 */
1524
Al Viroa4464db2011-07-07 15:03:58 -04001525struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
1527 struct dentry *dentry;
1528 char *dname;
1529
Mel Gormane12ba742007-10-16 01:25:52 -07001530 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 if (!dentry)
1532 return NULL;
1533
Linus Torvalds6326c712012-05-21 16:14:04 -07001534 /*
1535 * We guarantee that the inline name is always NUL-terminated.
1536 * This way the memcpy() done by the name switching in rename
1537 * will still always have a NUL at the end, even if we might
1538 * be overwriting an internal NUL character
1539 */
1540 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 if (name->len > DNAME_INLINE_LEN-1) {
1542 dname = kmalloc(name->len + 1, GFP_KERNEL);
1543 if (!dname) {
1544 kmem_cache_free(dentry_cache, dentry);
1545 return NULL;
1546 }
1547 } else {
1548 dname = dentry->d_iname;
1549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
1551 dentry->d_name.len = name->len;
1552 dentry->d_name.hash = name->hash;
1553 memcpy(dname, name->name, name->len);
1554 dname[name->len] = 0;
1555
Linus Torvalds6326c712012-05-21 16:14:04 -07001556 /* Make sure we always see the terminating NUL character */
1557 smp_wmb();
1558 dentry->d_name.name = dname;
1559
Waiman Long98474232013-08-28 18:24:59 -07001560 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001561 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001563 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001565 dentry->d_parent = dentry;
1566 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 dentry->d_op = NULL;
1568 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001569 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 INIT_LIST_HEAD(&dentry->d_lru);
1571 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Virob3d9b7a2012-06-09 13:51:19 -04001572 INIT_HLIST_NODE(&dentry->d_alias);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001573 INIT_LIST_HEAD(&dentry->d_u.d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001574 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575
Nick Piggin3e880fb2011-01-07 17:49:19 +11001576 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return dentry;
1579}
Al Viroa4464db2011-07-07 15:03:58 -04001580
1581/**
1582 * d_alloc - allocate a dcache entry
1583 * @parent: parent of entry to allocate
1584 * @name: qstr of the name
1585 *
1586 * Allocates a dentry. It returns %NULL if there is insufficient memory
1587 * available. On a success the dentry is returned. The name passed in is
1588 * copied and the copy passed in may be reused after this call.
1589 */
1590struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1591{
1592 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1593 if (!dentry)
1594 return NULL;
1595
1596 spin_lock(&parent->d_lock);
1597 /*
1598 * don't need child lock because it is not subject
1599 * to concurrency here
1600 */
1601 __dget_dlock(parent);
1602 dentry->d_parent = parent;
1603 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1604 spin_unlock(&parent->d_lock);
1605
1606 return dentry;
1607}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001608EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001610/**
1611 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1612 * @sb: the superblock
1613 * @name: qstr of the name
1614 *
1615 * For a filesystem that just pins its dentries in memory and never
1616 * performs lookups at all, return an unhashed IS_ROOT dentry.
1617 */
Nick Piggin4b936882011-01-07 17:50:07 +11001618struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1619{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001620 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001621}
1622EXPORT_SYMBOL(d_alloc_pseudo);
1623
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1625{
1626 struct qstr q;
1627
1628 q.name = name;
1629 q.len = strlen(name);
1630 q.hash = full_name_hash(q.name, q.len);
1631 return d_alloc(parent, &q);
1632}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001633EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
Nick Pigginfb045ad2011-01-07 17:49:55 +11001635void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1636{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001637 WARN_ON_ONCE(dentry->d_op);
1638 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001639 DCACHE_OP_COMPARE |
1640 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001641 DCACHE_OP_WEAK_REVALIDATE |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001642 DCACHE_OP_DELETE ));
1643 dentry->d_op = op;
1644 if (!op)
1645 return;
1646 if (op->d_hash)
1647 dentry->d_flags |= DCACHE_OP_HASH;
1648 if (op->d_compare)
1649 dentry->d_flags |= DCACHE_OP_COMPARE;
1650 if (op->d_revalidate)
1651 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001652 if (op->d_weak_revalidate)
1653 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001654 if (op->d_delete)
1655 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001656 if (op->d_prune)
1657 dentry->d_flags |= DCACHE_OP_PRUNE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001658
1659}
1660EXPORT_SYMBOL(d_set_d_op);
1661
David Howellsb18825a2013-09-12 19:22:53 +01001662static unsigned d_flags_for_inode(struct inode *inode)
1663{
1664 unsigned add_flags = DCACHE_FILE_TYPE;
1665
1666 if (!inode)
1667 return DCACHE_MISS_TYPE;
1668
1669 if (S_ISDIR(inode->i_mode)) {
1670 add_flags = DCACHE_DIRECTORY_TYPE;
1671 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1672 if (unlikely(!inode->i_op->lookup))
1673 add_flags = DCACHE_AUTODIR_TYPE;
1674 else
1675 inode->i_opflags |= IOP_LOOKUP;
1676 }
1677 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1678 if (unlikely(inode->i_op->follow_link))
1679 add_flags = DCACHE_SYMLINK_TYPE;
1680 else
1681 inode->i_opflags |= IOP_NOFOLLOW;
1682 }
1683
1684 if (unlikely(IS_AUTOMOUNT(inode)))
1685 add_flags |= DCACHE_NEED_AUTOMOUNT;
1686 return add_flags;
1687}
1688
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001689static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1690{
David Howellsb18825a2013-09-12 19:22:53 +01001691 unsigned add_flags = d_flags_for_inode(inode);
1692
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001693 spin_lock(&dentry->d_lock);
David Howellsb18825a2013-09-12 19:22:53 +01001694 dentry->d_flags &= ~DCACHE_ENTRY_TYPE;
1695 dentry->d_flags |= add_flags;
1696 if (inode)
Al Virob3d9b7a2012-06-09 13:51:19 -04001697 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001698 dentry->d_inode = inode;
Nick Piggin31e6b012011-01-07 17:49:52 +11001699 dentry_rcuwalk_barrier(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001700 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001701 fsnotify_d_instantiate(dentry, inode);
1702}
1703
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704/**
1705 * d_instantiate - fill in inode information for a dentry
1706 * @entry: dentry to complete
1707 * @inode: inode to attach to this dentry
1708 *
1709 * Fill in inode information in the entry.
1710 *
1711 * This turns negative dentries into productive full members
1712 * of society.
1713 *
1714 * NOTE! This assumes that the inode count has been incremented
1715 * (or otherwise set) by the caller to indicate that it is now
1716 * in use by the dcache.
1717 */
1718
1719void d_instantiate(struct dentry *entry, struct inode * inode)
1720{
Al Virob3d9b7a2012-06-09 13:51:19 -04001721 BUG_ON(!hlist_unhashed(&entry->d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001722 if (inode)
1723 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001724 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001725 if (inode)
1726 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 security_d_instantiate(entry, inode);
1728}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001729EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
1731/**
1732 * d_instantiate_unique - instantiate a non-aliased dentry
1733 * @entry: dentry to instantiate
1734 * @inode: inode to attach to this dentry
1735 *
1736 * Fill in inode information in the entry. On success, it returns NULL.
1737 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001738 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 *
1740 * Note that in order to avoid conflicts with rename() etc, the caller
1741 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001742 *
1743 * This also assumes that the inode count has been incremented
1744 * (or otherwise set) by the caller to indicate that it is now
1745 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 */
David Howells770bfad2006-08-22 20:06:07 -04001747static struct dentry *__d_instantiate_unique(struct dentry *entry,
1748 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749{
1750 struct dentry *alias;
1751 int len = entry->d_name.len;
1752 const char *name = entry->d_name.name;
1753 unsigned int hash = entry->d_name.hash;
1754
David Howells770bfad2006-08-22 20:06:07 -04001755 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001756 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001757 return NULL;
1758 }
1759
Sasha Levinb67bfe02013-02-27 17:06:00 -08001760 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001761 /*
1762 * Don't need alias->d_lock here, because aliases with
1763 * d_parent == entry->d_parent are not subject to name or
1764 * parent changes, because the parent inode i_mutex is held.
1765 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001766 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 continue;
1768 if (alias->d_parent != entry->d_parent)
1769 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001770 if (alias->d_name.len != len)
1771 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001772 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001774 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 return alias;
1776 }
David Howells770bfad2006-08-22 20:06:07 -04001777
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001778 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return NULL;
1780}
David Howells770bfad2006-08-22 20:06:07 -04001781
1782struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1783{
1784 struct dentry *result;
1785
Al Virob3d9b7a2012-06-09 13:51:19 -04001786 BUG_ON(!hlist_unhashed(&entry->d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001787
Nick Piggin873feea2011-01-07 17:50:06 +11001788 if (inode)
1789 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001790 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001791 if (inode)
1792 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001793
1794 if (!result) {
1795 security_d_instantiate(entry, inode);
1796 return NULL;
1797 }
1798
1799 BUG_ON(!d_unhashed(result));
1800 iput(inode);
1801 return result;
1802}
1803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804EXPORT_SYMBOL(d_instantiate_unique);
1805
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001806/**
1807 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1808 * @entry: dentry to complete
1809 * @inode: inode to attach to this dentry
1810 *
1811 * Fill in inode information in the entry. If a directory alias is found, then
1812 * return an error (and drop inode). Together with d_materialise_unique() this
1813 * guarantees that a directory inode may never have more than one alias.
1814 */
1815int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1816{
1817 BUG_ON(!hlist_unhashed(&entry->d_alias));
1818
1819 spin_lock(&inode->i_lock);
1820 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1821 spin_unlock(&inode->i_lock);
1822 iput(inode);
1823 return -EBUSY;
1824 }
1825 __d_instantiate(entry, inode);
1826 spin_unlock(&inode->i_lock);
1827 security_d_instantiate(entry, inode);
1828
1829 return 0;
1830}
1831EXPORT_SYMBOL(d_instantiate_no_diralias);
1832
Al Viroadc0e912012-01-08 16:49:21 -05001833struct dentry *d_make_root(struct inode *root_inode)
1834{
1835 struct dentry *res = NULL;
1836
1837 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001838 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001839
1840 res = __d_alloc(root_inode->i_sb, &name);
1841 if (res)
1842 d_instantiate(res, root_inode);
1843 else
1844 iput(root_inode);
1845 }
1846 return res;
1847}
1848EXPORT_SYMBOL(d_make_root);
1849
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001850static struct dentry * __d_find_any_alias(struct inode *inode)
1851{
1852 struct dentry *alias;
1853
Al Virob3d9b7a2012-06-09 13:51:19 -04001854 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001855 return NULL;
Al Virob3d9b7a2012-06-09 13:51:19 -04001856 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001857 __dget(alias);
1858 return alias;
1859}
1860
Sage Weil46f72b32012-01-10 09:04:37 -08001861/**
1862 * d_find_any_alias - find any alias for a given inode
1863 * @inode: inode to find an alias for
1864 *
1865 * If any aliases exist for the given inode, take and return a
1866 * reference for one of them. If no aliases exist, return %NULL.
1867 */
1868struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001869{
1870 struct dentry *de;
1871
1872 spin_lock(&inode->i_lock);
1873 de = __d_find_any_alias(inode);
1874 spin_unlock(&inode->i_lock);
1875 return de;
1876}
Sage Weil46f72b32012-01-10 09:04:37 -08001877EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879/**
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001880 * d_obtain_alias - find or allocate a dentry for a given inode
1881 * @inode: inode to allocate the dentry for
1882 *
1883 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1884 * similar open by handle operations. The returned dentry may be anonymous,
1885 * or may have a full name (if the inode was already in the cache).
1886 *
1887 * When called on a directory inode, we must ensure that the inode only ever
1888 * has one dentry. If a dentry is found, that is returned instead of
1889 * allocating a new one.
1890 *
1891 * On successful return, the reference to the inode has been transferred
Christoph Hellwig44003722008-08-11 15:49:04 +02001892 * to the dentry. In case of an error the reference on the inode is released.
1893 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1894 * be passed in and will be the error will be propagate to the return value,
1895 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001896 */
1897struct dentry *d_obtain_alias(struct inode *inode)
1898{
NeilBrownb911a6b2012-11-08 16:09:37 -08001899 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001900 struct dentry *tmp;
1901 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001902 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001903
1904 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001905 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001906 if (IS_ERR(inode))
1907 return ERR_CAST(inode);
1908
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001909 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001910 if (res)
1911 goto out_iput;
1912
Al Viroa4464db2011-07-07 15:03:58 -04001913 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001914 if (!tmp) {
1915 res = ERR_PTR(-ENOMEM);
1916 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001917 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001918
Nick Piggin873feea2011-01-07 17:50:06 +11001919 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001920 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001921 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001922 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001923 dput(tmp);
1924 goto out_iput;
1925 }
1926
1927 /* attach a disconnected dentry */
David Howellsb18825a2013-09-12 19:22:53 +01001928 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1929
Christoph Hellwig9308a612008-08-11 15:49:12 +02001930 spin_lock(&tmp->d_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001931 tmp->d_inode = inode;
David Howellsb18825a2013-09-12 19:22:53 +01001932 tmp->d_flags |= add_flags;
Al Virob3d9b7a2012-06-09 13:51:19 -04001933 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001934 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001935 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001936 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001937 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001938 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001939 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001940
Christoph Hellwig9308a612008-08-11 15:49:12 +02001941 return tmp;
1942
1943 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001944 if (res && !IS_ERR(res))
1945 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001946 iput(inode);
1947 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001948}
Benny Halevyadc48722009-02-27 14:02:59 -08001949EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951/**
1952 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1953 * @inode: the inode which may have a disconnected dentry
1954 * @dentry: a negative dentry which we want to point to the inode.
1955 *
1956 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1957 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1958 * and return it, else simply d_add the inode to the dentry and return NULL.
1959 *
1960 * This is needed in the lookup routine of any filesystem that is exportable
1961 * (via knfsd) so that we can build dcache paths to directories effectively.
1962 *
1963 * If a dentry was found and moved, then it is returned. Otherwise NULL
1964 * is returned. This matches the expected return value of ->lookup.
1965 *
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001966 * Cluster filesystems may call this function with a negative, hashed dentry.
1967 * In that case, we know that the inode will be a regular file, and also this
1968 * will only occur during atomic_open. So we need to check for the dentry
1969 * being already hashed only in the final case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 */
1971struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1972{
1973 struct dentry *new = NULL;
1974
Al Viroa9049372011-07-08 21:20:11 -04001975 if (IS_ERR(inode))
1976 return ERR_CAST(inode);
1977
NeilBrown21c0d8f2006-10-04 02:16:16 -07001978 if (inode && S_ISDIR(inode->i_mode)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001979 spin_lock(&inode->i_lock);
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001980 new = __d_find_alias(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (new) {
Linus Torvalds32ba9c32012-06-08 10:34:03 -07001982 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
Nick Piggin873feea2011-01-07 17:50:06 +11001983 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 security_d_instantiate(new, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 d_move(new, dentry);
1986 iput(inode);
1987 } else {
Nick Piggin873feea2011-01-07 17:50:06 +11001988 /* already taking inode->i_lock, so d_add() by hand */
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001989 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001990 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 security_d_instantiate(dentry, inode);
1992 d_rehash(dentry);
1993 }
Steven Whitehouse6d4ade92013-06-14 11:17:15 +01001994 } else {
1995 d_instantiate(dentry, inode);
1996 if (d_unhashed(dentry))
1997 d_rehash(dentry);
1998 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 return new;
2000}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002001EXPORT_SYMBOL(d_splice_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Barry Naujok94035402008-05-21 16:50:46 +10002003/**
2004 * d_add_ci - lookup or allocate new dentry with case-exact name
2005 * @inode: the inode case-insensitive lookup has found
2006 * @dentry: the negative dentry that was passed to the parent's lookup func
2007 * @name: the case-exact name to be associated with the returned dentry
2008 *
2009 * This is to avoid filling the dcache with case-insensitive names to the
2010 * same inode, only the actual correct case is stored in the dcache for
2011 * case-insensitive filesystems.
2012 *
2013 * For a case-insensitive lookup match and if the the case-exact dentry
2014 * already exists in in the dcache, use it and return it.
2015 *
2016 * If no entry exists with the exact case name, allocate new dentry with
2017 * the exact case, and return the spliced entry.
2018 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002019struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002020 struct qstr *name)
2021{
Barry Naujok94035402008-05-21 16:50:46 +10002022 struct dentry *found;
2023 struct dentry *new;
2024
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002025 /*
2026 * First check if a dentry matching the name already exists,
2027 * if not go ahead and create it now.
2028 */
Barry Naujok94035402008-05-21 16:50:46 +10002029 found = d_hash_and_lookup(dentry->d_parent, name);
Al Viro4f522a22013-02-11 23:20:37 -05002030 if (unlikely(IS_ERR(found)))
2031 goto err_out;
Barry Naujok94035402008-05-21 16:50:46 +10002032 if (!found) {
2033 new = d_alloc(dentry->d_parent, name);
2034 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002035 found = ERR_PTR(-ENOMEM);
Barry Naujok94035402008-05-21 16:50:46 +10002036 goto err_out;
2037 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002038
Barry Naujok94035402008-05-21 16:50:46 +10002039 found = d_splice_alias(inode, new);
2040 if (found) {
2041 dput(new);
2042 return found;
2043 }
2044 return new;
2045 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002046
2047 /*
2048 * If a matching dentry exists, and it's not negative use it.
2049 *
2050 * Decrement the reference count to balance the iget() done
2051 * earlier on.
2052 */
Barry Naujok94035402008-05-21 16:50:46 +10002053 if (found->d_inode) {
2054 if (unlikely(found->d_inode != inode)) {
2055 /* This can't happen because bad inodes are unhashed. */
2056 BUG_ON(!is_bad_inode(inode));
2057 BUG_ON(!is_bad_inode(found->d_inode));
2058 }
Barry Naujok94035402008-05-21 16:50:46 +10002059 iput(inode);
2060 return found;
2061 }
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002062
Barry Naujok94035402008-05-21 16:50:46 +10002063 /*
2064 * Negative dentry: instantiate it unless the inode is a directory and
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002065 * already has a dentry.
Barry Naujok94035402008-05-21 16:50:46 +10002066 */
Al Viro4513d892011-07-17 10:52:14 -04002067 new = d_splice_alias(inode, found);
2068 if (new) {
2069 dput(found);
2070 found = new;
Barry Naujok94035402008-05-21 16:50:46 +10002071 }
Al Viro4513d892011-07-17 10:52:14 -04002072 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002073
2074err_out:
2075 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002076 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002077}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002078EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002080/*
2081 * Do the slow-case of the dentry name compare.
2082 *
2083 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002084 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002085 * filesystem can rely on them, and can use the 'name' and
2086 * 'len' information without worrying about walking off the
2087 * end of memory etc.
2088 *
2089 * Thus the read_seqcount_retry() and the "duplicate" info
2090 * in arguments (the low-level filesystem should not look
2091 * at the dentry inode or name contents directly, since
2092 * rename can change them while we're in RCU mode).
2093 */
2094enum slow_d_compare {
2095 D_COMP_OK,
2096 D_COMP_NOMATCH,
2097 D_COMP_SEQRETRY,
2098};
2099
2100static noinline enum slow_d_compare slow_dentry_cmp(
2101 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002102 struct dentry *dentry,
2103 unsigned int seq,
2104 const struct qstr *name)
2105{
2106 int tlen = dentry->d_name.len;
2107 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002108
2109 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2110 cpu_relax();
2111 return D_COMP_SEQRETRY;
2112 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002113 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002114 return D_COMP_NOMATCH;
2115 return D_COMP_OK;
2116}
2117
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002119 * __d_lookup_rcu - search for a dentry (racy, store-free)
2120 * @parent: parent dentry
2121 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002122 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002123 * Returns: dentry, or NULL
2124 *
2125 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2126 * resolution (store-free path walking) design described in
2127 * Documentation/filesystems/path-lookup.txt.
2128 *
2129 * This is not to be used outside core vfs.
2130 *
2131 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2132 * held, and rcu_read_lock held. The returned dentry must not be stored into
2133 * without taking d_lock and checking d_seq sequence count against @seq
2134 * returned here.
2135 *
Linus Torvalds15570082013-09-02 11:38:06 -07002136 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002137 * function.
2138 *
2139 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2140 * the returned dentry, so long as its parent's seqlock is checked after the
2141 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2142 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002143 *
2144 * NOTE! The caller *has* to check the resulting dentry against the sequence
2145 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002146 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002147struct dentry *__d_lookup_rcu(const struct dentry *parent,
2148 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002149 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002150{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002151 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002152 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002153 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002154 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002155 struct dentry *dentry;
2156
2157 /*
2158 * Note: There is significant duplication with __d_lookup_rcu which is
2159 * required to prevent single threaded performance regressions
2160 * especially on architectures where smp_rmb (in seqcounts) are costly.
2161 * Keep the two functions in sync.
2162 */
2163
2164 /*
2165 * The hash list is protected using RCU.
2166 *
2167 * Carefully use d_seq when comparing a candidate dentry, to avoid
2168 * races with d_move().
2169 *
2170 * It is possible that concurrent renames can mess up our list
2171 * walk here and result in missing our dentry, resulting in the
2172 * false-negative result. d_lookup() protects against concurrent
2173 * renames using rename_lock seqlock.
2174 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002175 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002176 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002177 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002178 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002179
Nick Piggin31e6b012011-01-07 17:49:52 +11002180seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002181 /*
2182 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002183 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002184 *
2185 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002186 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002187 *
2188 * NOTE! We do a "raw" seqcount_begin here. That means that
2189 * we don't wait for the sequence count to stabilize if it
2190 * is in the middle of a sequence change. If we do the slow
2191 * dentry compare, we will do seqretries until it is stable,
2192 * and if we end up with a successful lookup, we actually
2193 * want to exit RCU lookup anyway.
2194 */
2195 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002196 if (dentry->d_parent != parent)
2197 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002198 if (d_unhashed(dentry))
2199 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002200
2201 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002202 if (dentry->d_name.hash != hashlen_hash(hashlen))
2203 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002204 *seqp = seq;
2205 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002206 case D_COMP_OK:
2207 return dentry;
2208 case D_COMP_NOMATCH:
2209 continue;
2210 default:
2211 goto seqretry;
2212 }
2213 }
2214
Linus Torvalds26fe5752012-05-10 13:14:12 -07002215 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002216 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002217 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002218 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002219 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002220 }
2221 return NULL;
2222}
2223
2224/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 * d_lookup - search for a dentry
2226 * @parent: parent dentry
2227 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002228 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002230 * d_lookup searches the children of the parent dentry for the name in
2231 * question. If the dentry is found its reference count is incremented and the
2232 * dentry is returned. The caller must use dput to free the entry when it has
2233 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 */
Al Viroda2d8452013-01-24 18:29:34 -05002235struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236{
Nick Piggin31e6b012011-01-07 17:49:52 +11002237 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002238 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239
2240 do {
2241 seq = read_seqbegin(&rename_lock);
2242 dentry = __d_lookup(parent, name);
2243 if (dentry)
2244 break;
2245 } while (read_seqretry(&rename_lock, seq));
2246 return dentry;
2247}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002248EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249
Nick Piggin31e6b012011-01-07 17:49:52 +11002250/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002251 * __d_lookup - search for a dentry (racy)
2252 * @parent: parent dentry
2253 * @name: qstr of name we wish to find
2254 * Returns: dentry, or NULL
2255 *
2256 * __d_lookup is like d_lookup, however it may (rarely) return a
2257 * false-negative result due to unrelated rename activity.
2258 *
2259 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2260 * however it must be used carefully, eg. with a following d_lookup in
2261 * the case of failure.
2262 *
2263 * __d_lookup callers must be commented.
2264 */
Al Viroa713ca22013-01-24 18:27:00 -05002265struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266{
2267 unsigned int len = name->len;
2268 unsigned int hash = name->hash;
2269 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002270 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002271 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002272 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002273 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
Nick Pigginb04f7842010-08-18 04:37:34 +10002275 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002276 * Note: There is significant duplication with __d_lookup_rcu which is
2277 * required to prevent single threaded performance regressions
2278 * especially on architectures where smp_rmb (in seqcounts) are costly.
2279 * Keep the two functions in sync.
2280 */
2281
2282 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002283 * The hash list is protected using RCU.
2284 *
2285 * Take d_lock when comparing a candidate dentry, to avoid races
2286 * with d_move().
2287 *
2288 * It is possible that concurrent renames can mess up our list
2289 * walk here and result in missing our dentry, resulting in the
2290 * false-negative result. d_lookup() protects against concurrent
2291 * renames using rename_lock seqlock.
2292 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002293 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002294 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 rcu_read_lock();
2296
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002297 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 if (dentry->d_name.hash != hash)
2300 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
2302 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 if (dentry->d_parent != parent)
2304 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002305 if (d_unhashed(dentry))
2306 goto next;
2307
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 /*
2309 * It is safe to compare names since d_move() cannot
2310 * change the qstr (protected by d_lock).
2311 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002312 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002313 int tlen = dentry->d_name.len;
2314 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002315 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 goto next;
2317 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002318 if (dentry->d_name.len != len)
2319 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002320 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 goto next;
2322 }
2323
Waiman Long98474232013-08-28 18:24:59 -07002324 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002325 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 spin_unlock(&dentry->d_lock);
2327 break;
2328next:
2329 spin_unlock(&dentry->d_lock);
2330 }
2331 rcu_read_unlock();
2332
2333 return found;
2334}
2335
2336/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002337 * d_hash_and_lookup - hash the qstr then search for a dentry
2338 * @dir: Directory to search in
2339 * @name: qstr of name we wish to find
2340 *
Al Viro4f522a22013-02-11 23:20:37 -05002341 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002342 */
2343struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2344{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002345 /*
2346 * Check for a fs-specific hash function. Note that we must
2347 * calculate the standard hash first, as the d_op->d_hash()
2348 * routine may choose to leave the hash value unchanged.
2349 */
2350 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002351 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002352 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002353 if (unlikely(err < 0))
2354 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002355 }
Al Viro4f522a22013-02-11 23:20:37 -05002356 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002357}
Al Viro4f522a22013-02-11 23:20:37 -05002358EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002359
2360/**
Nick Piggin786a5e12011-01-07 17:49:16 +11002361 * d_validate - verify dentry provided from insecure source (deprecated)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 * @dentry: The dentry alleged to be valid child of @dparent
Randy Dunlapff5fdb62011-01-22 20:16:06 -08002363 * @dparent: The parent dentry (known to be valid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 *
2365 * An insecure source has sent us a dentry, here we verify it and dget() it.
2366 * This is used by ncpfs in its readdir implementation.
2367 * Zero is returned in the dentry is invalid.
Nick Piggin786a5e12011-01-07 17:49:16 +11002368 *
2369 * This function is slow for big directories, and deprecated, do not use it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 */
Nick Piggind3a23e12011-01-05 20:01:21 +11002371int d_validate(struct dentry *dentry, struct dentry *dparent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372{
Nick Piggin786a5e12011-01-07 17:49:16 +11002373 struct dentry *child;
Nick Piggind3a23e12011-01-05 20:01:21 +11002374
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002375 spin_lock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002376 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2377 if (dentry == child) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002378 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Nick Piggindc0474b2011-01-07 17:49:43 +11002379 __dget_dlock(dentry);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002380 spin_unlock(&dentry->d_lock);
2381 spin_unlock(&dparent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 return 1;
2383 }
2384 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002385 spin_unlock(&dparent->d_lock);
Nick Piggin786a5e12011-01-07 17:49:16 +11002386
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 return 0;
2388}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002389EXPORT_SYMBOL(d_validate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390
2391/*
2392 * When a file is deleted, we have two options:
2393 * - turn this dentry into a negative dentry
2394 * - unhash this dentry and free it.
2395 *
2396 * Usually, we want to just turn this into
2397 * a negative dentry, but if anybody else is
2398 * currently using the dentry or the inode
2399 * we can't do that and we fall back on removing
2400 * it from the hash queues and waiting for
2401 * it to be deleted later when it has no users
2402 */
2403
2404/**
2405 * d_delete - delete a dentry
2406 * @dentry: The dentry to delete
2407 *
2408 * Turn the dentry into a negative dentry if possible, otherwise
2409 * remove it from the hash queues so it can be deleted later
2410 */
2411
2412void d_delete(struct dentry * dentry)
2413{
Nick Piggin873feea2011-01-07 17:50:06 +11002414 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002415 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 /*
2417 * Are we the only user?
2418 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002419again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002421 inode = dentry->d_inode;
2422 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002423 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002424 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002425 spin_unlock(&dentry->d_lock);
2426 cpu_relax();
2427 goto again;
2428 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002429 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002430 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002431 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 return;
2433 }
2434
2435 if (!d_unhashed(dentry))
2436 __d_drop(dentry);
2437
2438 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002439
2440 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002442EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002444static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002446 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002447 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002448 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002449 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002450 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451}
2452
David Howells770bfad2006-08-22 20:06:07 -04002453static void _d_rehash(struct dentry * entry)
2454{
2455 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2456}
2457
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458/**
2459 * d_rehash - add an entry back to the hash
2460 * @entry: dentry to add to the hash
2461 *
2462 * Adds a dentry to the hash according to its name.
2463 */
2464
2465void d_rehash(struct dentry * entry)
2466{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002468 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002471EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002473/**
2474 * dentry_update_name_case - update case insensitive dentry with a new name
2475 * @dentry: dentry to be updated
2476 * @name: new name
2477 *
2478 * Update a case insensitive dentry with new case of name.
2479 *
2480 * dentry must have been returned by d_lookup with name @name. Old and new
2481 * name lengths must match (ie. no d_compare which allows mismatched name
2482 * lengths).
2483 *
2484 * Parent inode i_mutex must be held over d_lookup and into this call (to
2485 * keep renames and concurrent inserts, and readdir(2) away).
2486 */
2487void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2488{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002489 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002490 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2491
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002492 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002493 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002494 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002495 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002496 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002497}
2498EXPORT_SYMBOL(dentry_update_name_case);
2499
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500static void switch_names(struct dentry *dentry, struct dentry *target)
2501{
2502 if (dname_external(target)) {
2503 if (dname_external(dentry)) {
2504 /*
2505 * Both external: swap the pointers
2506 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002507 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 } else {
2509 /*
2510 * dentry:internal, target:external. Steal target's
2511 * storage and make target internal.
2512 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002513 memcpy(target->d_iname, dentry->d_name.name,
2514 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 dentry->d_name.name = target->d_name.name;
2516 target->d_name.name = target->d_iname;
2517 }
2518 } else {
2519 if (dname_external(dentry)) {
2520 /*
2521 * dentry:external, target:internal. Give dentry's
2522 * storage to target and make dentry internal
2523 */
2524 memcpy(dentry->d_iname, target->d_name.name,
2525 target->d_name.len + 1);
2526 target->d_name.name = dentry->d_name.name;
2527 dentry->d_name.name = dentry->d_iname;
2528 } else {
2529 /*
2530 * Both are internal. Just copy target to dentry
2531 */
2532 memcpy(dentry->d_iname, target->d_name.name,
2533 target->d_name.len + 1);
Al Virodc711ca2008-11-03 15:03:50 -05002534 dentry->d_name.len = target->d_name.len;
2535 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 }
2537 }
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002538 swap(dentry->d_name.len, target->d_name.len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539}
2540
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002541static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2542{
2543 /*
2544 * XXXX: do we really need to take target->d_lock?
2545 */
2546 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2547 spin_lock(&target->d_parent->d_lock);
2548 else {
2549 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2550 spin_lock(&dentry->d_parent->d_lock);
2551 spin_lock_nested(&target->d_parent->d_lock,
2552 DENTRY_D_LOCK_NESTED);
2553 } else {
2554 spin_lock(&target->d_parent->d_lock);
2555 spin_lock_nested(&dentry->d_parent->d_lock,
2556 DENTRY_D_LOCK_NESTED);
2557 }
2558 }
2559 if (target < dentry) {
2560 spin_lock_nested(&target->d_lock, 2);
2561 spin_lock_nested(&dentry->d_lock, 3);
2562 } else {
2563 spin_lock_nested(&dentry->d_lock, 2);
2564 spin_lock_nested(&target->d_lock, 3);
2565 }
2566}
2567
2568static void dentry_unlock_parents_for_move(struct dentry *dentry,
2569 struct dentry *target)
2570{
2571 if (target->d_parent != dentry->d_parent)
2572 spin_unlock(&dentry->d_parent->d_lock);
2573 if (target->d_parent != target)
2574 spin_unlock(&target->d_parent->d_lock);
2575}
2576
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002578 * When switching names, the actual string doesn't strictly have to
2579 * be preserved in the target - because we're dropping the target
2580 * anyway. As such, we can just do a simple memcpy() to copy over
2581 * the new name before we switch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 *
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002583 * Note that we have to be a lot more careful about getting the hash
2584 * switched - we have to switch the hash value properly even if it
2585 * then no longer matches the actual (corrupted) string of the target.
2586 * The hash value has to match the hash queue that the dentry is on..
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002588/*
Al Viro18367502011-07-12 21:42:24 -04002589 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 * @dentry: entry to move
2591 * @target: new dentry
2592 *
2593 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002594 * dcache entries should not be moved in this way. Caller must hold
2595 * rename_lock, the i_mutex of the source and target directories,
2596 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 */
Al Viro18367502011-07-12 21:42:24 -04002598static void __d_move(struct dentry * dentry, struct dentry * target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 if (!dentry->d_inode)
2601 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2602
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002603 BUG_ON(d_ancestor(dentry, target));
2604 BUG_ON(d_ancestor(target, dentry));
2605
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002606 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Nick Piggin31e6b012011-01-07 17:49:52 +11002608 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002609 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002610
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002611 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2612
2613 /*
2614 * Move the dentry to the target hash queue. Don't bother checking
2615 * for the same hash queue because of how unlikely it is.
2616 */
2617 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002618 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
2620 /* Unhash the target: dput() will then get rid of it */
2621 __d_drop(target);
2622
Eric Dumazet5160ee62006-01-08 01:03:32 -08002623 list_del(&dentry->d_u.d_child);
2624 list_del(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625
2626 /* Switch the names.. */
2627 switch_names(dentry, target);
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002628 swap(dentry->d_name.hash, target->d_name.hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
2630 /* ... and switch the parents */
2631 if (IS_ROOT(dentry)) {
2632 dentry->d_parent = target->d_parent;
2633 target->d_parent = target;
Eric Dumazet5160ee62006-01-08 01:03:32 -08002634 INIT_LIST_HEAD(&target->d_u.d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 } else {
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002636 swap(dentry->d_parent, target->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637
2638 /* And add them back to the (new) parent lists */
Eric Dumazet5160ee62006-01-08 01:03:32 -08002639 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 }
2641
Eric Dumazet5160ee62006-01-08 01:03:32 -08002642 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002643
Nick Piggin31e6b012011-01-07 17:49:52 +11002644 write_seqcount_end(&target->d_seq);
2645 write_seqcount_end(&dentry->d_seq);
2646
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002647 dentry_unlock_parents_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 spin_unlock(&target->d_lock);
Nick Pigginc32ccd82006-03-25 03:07:09 -08002649 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002651}
2652
2653/*
2654 * d_move - move a dentry
2655 * @dentry: entry to move
2656 * @target: new dentry
2657 *
2658 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002659 * dcache entries should not be moved in this way. See the locking
2660 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002661 */
2662void d_move(struct dentry *dentry, struct dentry *target)
2663{
2664 write_seqlock(&rename_lock);
2665 __d_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002667}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002668EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002670/**
2671 * d_ancestor - search for an ancestor
2672 * @p1: ancestor dentry
2673 * @p2: child dentry
2674 *
2675 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2676 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002677 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002678struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002679{
2680 struct dentry *p;
2681
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002682 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002683 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002684 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002685 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002686 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002687}
2688
2689/*
2690 * This helper attempts to cope with remotely renamed directories
2691 *
2692 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002693 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002694 *
2695 * Note: If ever the locking in lock_rename() changes, then please
2696 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002697 */
Nick Piggin873feea2011-01-07 17:50:06 +11002698static struct dentry *__d_unalias(struct inode *inode,
2699 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002700{
2701 struct mutex *m1 = NULL, *m2 = NULL;
Al Viroee3efa92012-06-08 15:59:33 -04002702 struct dentry *ret = ERR_PTR(-EBUSY);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002703
2704 /* If alias and dentry share a parent, then no extra locks required */
2705 if (alias->d_parent == dentry->d_parent)
2706 goto out_unalias;
2707
Trond Myklebust9eaef272006-10-21 10:24:20 -07002708 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002709 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2710 goto out_err;
2711 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2712 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2713 goto out_err;
2714 m2 = &alias->d_parent->d_inode->i_mutex;
2715out_unalias:
Al Viroee3efa92012-06-08 15:59:33 -04002716 if (likely(!d_mountpoint(alias))) {
2717 __d_move(alias, dentry);
2718 ret = alias;
2719 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002720out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002721 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002722 if (m2)
2723 mutex_unlock(m2);
2724 if (m1)
2725 mutex_unlock(m1);
2726 return ret;
2727}
2728
2729/*
David Howells770bfad2006-08-22 20:06:07 -04002730 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2731 * named dentry in place of the dentry to be replaced.
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002732 * returns with anon->d_lock held!
David Howells770bfad2006-08-22 20:06:07 -04002733 */
2734static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2735{
Al Viro740da422013-01-30 10:13:38 -05002736 struct dentry *dparent;
David Howells770bfad2006-08-22 20:06:07 -04002737
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002738 dentry_lock_for_move(anon, dentry);
David Howells770bfad2006-08-22 20:06:07 -04002739
Nick Piggin31e6b012011-01-07 17:49:52 +11002740 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002741 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002742
David Howells770bfad2006-08-22 20:06:07 -04002743 dparent = dentry->d_parent;
David Howells770bfad2006-08-22 20:06:07 -04002744
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002745 switch_names(dentry, anon);
2746 swap(dentry->d_name.hash, anon->d_name.hash);
2747
Al Viro740da422013-01-30 10:13:38 -05002748 dentry->d_parent = dentry;
2749 list_del_init(&dentry->d_u.d_child);
2750 anon->d_parent = dparent;
Wei Yongjun9ed53b12013-03-12 00:10:50 +08002751 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
David Howells770bfad2006-08-22 20:06:07 -04002752
Nick Piggin31e6b012011-01-07 17:49:52 +11002753 write_seqcount_end(&dentry->d_seq);
2754 write_seqcount_end(&anon->d_seq);
2755
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002756 dentry_unlock_parents_for_move(anon, dentry);
2757 spin_unlock(&dentry->d_lock);
2758
2759 /* anon->d_lock still locked, returns locked */
David Howells770bfad2006-08-22 20:06:07 -04002760}
2761
2762/**
2763 * d_materialise_unique - introduce an inode into the tree
2764 * @dentry: candidate dentry
2765 * @inode: inode to bind to the dentry, to which aliases may be attached
2766 *
2767 * Introduces an dentry into the tree, substituting an extant disconnected
Jeff Laytonc46c8872011-07-26 13:33:16 -04002768 * root directory alias in its place if there is one. Caller must hold the
2769 * i_mutex of the parent directory.
David Howells770bfad2006-08-22 20:06:07 -04002770 */
2771struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2772{
Trond Myklebust9eaef272006-10-21 10:24:20 -07002773 struct dentry *actual;
David Howells770bfad2006-08-22 20:06:07 -04002774
2775 BUG_ON(!d_unhashed(dentry));
2776
David Howells770bfad2006-08-22 20:06:07 -04002777 if (!inode) {
2778 actual = dentry;
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002779 __d_instantiate(dentry, NULL);
Nick Piggin357f8e62011-01-07 17:49:42 +11002780 d_rehash(actual);
2781 goto out_nolock;
David Howells770bfad2006-08-22 20:06:07 -04002782 }
2783
Nick Piggin873feea2011-01-07 17:50:06 +11002784 spin_lock(&inode->i_lock);
Nick Piggin357f8e62011-01-07 17:49:42 +11002785
Trond Myklebust9eaef272006-10-21 10:24:20 -07002786 if (S_ISDIR(inode->i_mode)) {
2787 struct dentry *alias;
David Howells770bfad2006-08-22 20:06:07 -04002788
Trond Myklebust9eaef272006-10-21 10:24:20 -07002789 /* Does an aliased dentry already exist? */
Linus Torvalds32ba9c32012-06-08 10:34:03 -07002790 alias = __d_find_alias(inode, 0);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002791 if (alias) {
2792 actual = alias;
Al Viro18367502011-07-12 21:42:24 -04002793 write_seqlock(&rename_lock);
2794
2795 if (d_ancestor(alias, dentry)) {
2796 /* Check for loops */
2797 actual = ERR_PTR(-ELOOP);
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002798 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002799 } else if (IS_ROOT(alias)) {
2800 /* Is this an anonymous mountpoint that we
2801 * could splice into our tree? */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002802 __d_materialise_dentry(dentry, alias);
Al Viro18367502011-07-12 21:42:24 -04002803 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002804 __d_drop(alias);
2805 goto found;
Al Viro18367502011-07-12 21:42:24 -04002806 } else {
2807 /* Nope, but we must(!) avoid directory
Michel Lespinasseb18dafc2012-03-26 17:32:44 -07002808 * aliasing. This drops inode->i_lock */
Al Viro18367502011-07-12 21:42:24 -04002809 actual = __d_unalias(inode, dentry, alias);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002810 }
Al Viro18367502011-07-12 21:42:24 -04002811 write_sequnlock(&rename_lock);
David Howellsdd179942011-08-16 15:31:30 +01002812 if (IS_ERR(actual)) {
2813 if (PTR_ERR(actual) == -ELOOP)
2814 pr_warn_ratelimited(
2815 "VFS: Lookup of '%s' in %s %s"
2816 " would have caused loop\n",
2817 dentry->d_name.name,
2818 inode->i_sb->s_type->name,
2819 inode->i_sb->s_id);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002820 dput(alias);
David Howellsdd179942011-08-16 15:31:30 +01002821 }
Trond Myklebust9eaef272006-10-21 10:24:20 -07002822 goto out_nolock;
2823 }
David Howells770bfad2006-08-22 20:06:07 -04002824 }
2825
2826 /* Add a unique reference */
2827 actual = __d_instantiate_unique(dentry, inode);
2828 if (!actual)
2829 actual = dentry;
Nick Piggin357f8e62011-01-07 17:49:42 +11002830 else
2831 BUG_ON(!d_unhashed(actual));
David Howells770bfad2006-08-22 20:06:07 -04002832
David Howells770bfad2006-08-22 20:06:07 -04002833 spin_lock(&actual->d_lock);
2834found:
2835 _d_rehash(actual);
2836 spin_unlock(&actual->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002837 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002838out_nolock:
David Howells770bfad2006-08-22 20:06:07 -04002839 if (actual == dentry) {
2840 security_d_instantiate(dentry, inode);
2841 return NULL;
2842 }
2843
2844 iput(inode);
2845 return actual;
David Howells770bfad2006-08-22 20:06:07 -04002846}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002847EXPORT_SYMBOL_GPL(d_materialise_unique);
David Howells770bfad2006-08-22 20:06:07 -04002848
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002849static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002850{
2851 *buflen -= namelen;
2852 if (*buflen < 0)
2853 return -ENAMETOOLONG;
2854 *buffer -= namelen;
2855 memcpy(*buffer, str, namelen);
2856 return 0;
2857}
2858
Waiman Long232d2d62013-09-09 12:18:13 -04002859/**
2860 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002861 * @buffer: buffer pointer
2862 * @buflen: allocated length of the buffer
2863 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002864 *
2865 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2866 * make sure that either the old or the new name pointer and length are
2867 * fetched. However, there may be mismatch between length and pointer.
2868 * The length cannot be trusted, we need to copy it byte-by-byte until
2869 * the length is reached or a null byte is found. It also prepends "/" at
2870 * the beginning of the name. The sequence number check at the caller will
2871 * retry it again when a d_move() does happen. So any garbage in the buffer
2872 * due to mismatched pointer and length will be discarded.
2873 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002874static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2875{
Waiman Long232d2d62013-09-09 12:18:13 -04002876 const char *dname = ACCESS_ONCE(name->name);
2877 u32 dlen = ACCESS_ONCE(name->len);
2878 char *p;
2879
2880 if (*buflen < dlen + 1)
2881 return -ENAMETOOLONG;
2882 *buflen -= dlen + 1;
2883 p = *buffer -= dlen + 1;
2884 *p++ = '/';
2885 while (dlen--) {
2886 char c = *dname++;
2887 if (!c)
2888 break;
2889 *p++ = c;
2890 }
2891 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002892}
2893
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002895 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002896 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002897 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002898 * @buffer: pointer to the end of the buffer
2899 * @buflen: pointer to buffer length
2900 *
Waiman Long18129972013-09-12 10:55:35 -04002901 * The function will first try to write out the pathname without taking any
2902 * lock other than the RCU read lock to make sure that dentries won't go away.
2903 * It only checks the sequence number of the global rename_lock as any change
2904 * in the dentry's d_seq will be preceded by changes in the rename_lock
2905 * sequence number. If the sequence number had been changed, it will restart
2906 * the whole pathname back-tracing sequence again by taking the rename_lock.
2907 * In this case, there is no need to take the RCU read lock as the recursive
2908 * parent pointer references will keep the dentry chain alive as long as no
2909 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002910 */
Al Viro02125a82011-12-05 08:43:34 -05002911static int prepend_path(const struct path *path,
2912 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002913 char **buffer, int *buflen)
2914{
Al Viroede4ceb2013-11-13 07:45:40 -05002915 struct dentry *dentry;
2916 struct vfsmount *vfsmnt;
2917 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002918 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002919 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002920 char *bptr;
2921 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002922
Al Viro48f5ec22013-09-09 15:22:25 -04002923 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002924restart_mnt:
2925 read_seqbegin_or_lock(&mount_lock, &m_seq);
2926 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002927 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002928restart:
2929 bptr = *buffer;
2930 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002931 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002932 dentry = path->dentry;
2933 vfsmnt = path->mnt;
2934 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002935 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002936 while (dentry != root->dentry || vfsmnt != root->mnt) {
2937 struct dentry * parent;
2938
2939 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002940 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002941 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002942 if (mnt != parent) {
2943 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2944 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002945 vfsmnt = &mnt->mnt;
2946 continue;
2947 }
2948 /*
2949 * Filesystems needing to implement special "root names"
2950 * should do so with ->d_dname()
2951 */
2952 if (IS_ROOT(dentry) &&
2953 (dentry->d_name.len != 1 ||
2954 dentry->d_name.name[0] != '/')) {
2955 WARN(1, "Root dentry has weird name <%.*s>\n",
2956 (int) dentry->d_name.len,
2957 dentry->d_name.name);
2958 }
2959 if (!error)
2960 error = is_mounted(vfsmnt) ? 1 : 2;
2961 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002962 }
2963 parent = dentry->d_parent;
2964 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002965 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002966 if (error)
2967 break;
2968
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002969 dentry = parent;
2970 }
Al Viro48f5ec22013-09-09 15:22:25 -04002971 if (!(seq & 1))
2972 rcu_read_unlock();
2973 if (need_seqretry(&rename_lock, seq)) {
2974 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002975 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002976 }
Al Viro48f5ec22013-09-09 15:22:25 -04002977 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002978
2979 if (!(m_seq & 1))
2980 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002981 if (need_seqretry(&mount_lock, m_seq)) {
2982 m_seq = 1;
2983 goto restart_mnt;
2984 }
2985 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002986
Waiman Long232d2d62013-09-09 12:18:13 -04002987 if (error >= 0 && bptr == *buffer) {
2988 if (--blen < 0)
2989 error = -ENAMETOOLONG;
2990 else
2991 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002992 }
Waiman Long232d2d62013-09-09 12:18:13 -04002993 *buffer = bptr;
2994 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002995 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002996}
2997
2998/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002999 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003000 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05003001 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07003002 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 * @buflen: buffer length
3004 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003005 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003007 * Returns a pointer into the buffer or an error code if the
3008 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08003009 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04003010 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01003011 *
Al Viro02125a82011-12-05 08:43:34 -05003012 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 */
Al Viro02125a82011-12-05 08:43:34 -05003014char *__d_path(const struct path *path,
3015 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003016 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003018 char *res = buf + buflen;
3019 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003021 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003022 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003023
Al Viro02125a82011-12-05 08:43:34 -05003024 if (error < 0)
3025 return ERR_PTR(error);
3026 if (error > 0)
3027 return NULL;
3028 return res;
3029}
3030
3031char *d_absolute_path(const struct path *path,
3032 char *buf, int buflen)
3033{
3034 struct path root = {};
3035 char *res = buf + buflen;
3036 int error;
3037
3038 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003039 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003040
3041 if (error > 1)
3042 error = -EINVAL;
3043 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003044 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003045 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046}
3047
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003048/*
3049 * same as __d_path but appends "(deleted)" for unlinked files.
3050 */
Al Viro02125a82011-12-05 08:43:34 -05003051static int path_with_deleted(const struct path *path,
3052 const struct path *root,
3053 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003054{
3055 prepend(buf, buflen, "\0", 1);
3056 if (d_unlinked(path->dentry)) {
3057 int error = prepend(buf, buflen, " (deleted)", 10);
3058 if (error)
3059 return error;
3060 }
3061
3062 return prepend_path(path, root, buf, buflen);
3063}
3064
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003065static int prepend_unreachable(char **buffer, int *buflen)
3066{
3067 return prepend(buffer, buflen, "(unreachable)", 13);
3068}
3069
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003070static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3071{
3072 unsigned seq;
3073
3074 do {
3075 seq = read_seqcount_begin(&fs->seq);
3076 *root = fs->root;
3077 } while (read_seqcount_retry(&fs->seq, seq));
3078}
3079
Jan Bluncka03a8a702008-02-14 19:38:32 -08003080/**
3081 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003082 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003083 * @buf: buffer to return value in
3084 * @buflen: buffer length
3085 *
3086 * Convert a dentry into an ASCII path name. If the entry has been deleted
3087 * the string " (deleted)" is appended. Note that this is ambiguous.
3088 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003089 * Returns a pointer into the buffer or an error code if the path was
3090 * too long. Note: Callers should use the returned pointer, not the passed
3091 * in buffer, to use the name! The implementation often starts at an offset
3092 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003093 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003094 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003095 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003096char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003098 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003099 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003100 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003102 /*
3103 * We have various synthetic filesystems that never get mounted. On
3104 * these filesystems dentries are never used for lookup purposes, and
3105 * thus don't need to be hashed. They also don't need a name until a
3106 * user wants to identify the object in /proc/pid/fd/. The little hack
3107 * below allows us to generate a name for these objects on demand:
3108 */
Jan Blunckcf28b482008-02-14 19:38:44 -08003109 if (path->dentry->d_op && path->dentry->d_op->d_dname)
3110 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003111
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003112 rcu_read_lock();
3113 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003114 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003115 rcu_read_unlock();
3116
Al Viro02125a82011-12-05 08:43:34 -05003117 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003118 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 return res;
3120}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003121EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122
3123/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003124 * Helper function for dentry_operations.d_dname() members
3125 */
3126char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3127 const char *fmt, ...)
3128{
3129 va_list args;
3130 char temp[64];
3131 int sz;
3132
3133 va_start(args, fmt);
3134 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3135 va_end(args);
3136
3137 if (sz > sizeof(temp) || sz > buflen)
3138 return ERR_PTR(-ENAMETOOLONG);
3139
3140 buffer += buflen - sz;
3141 return memcpy(buffer, temp, sz);
3142}
3143
Al Viro118b2302013-08-24 12:08:17 -04003144char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3145{
3146 char *end = buffer + buflen;
3147 /* these dentries are never renamed, so d_lock is not needed */
3148 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003149 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003150 prepend(&end, &buflen, "/", 1))
3151 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003152 return end;
Al Viro118b2302013-08-24 12:08:17 -04003153}
3154
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003155/*
Ram Pai6092d042008-03-27 13:06:20 +01003156 * Write full pathname from the root of the filesystem into the buffer.
3157 */
Nick Pigginec2447c2011-01-07 17:49:29 +11003158static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003159{
Waiman Long232d2d62013-09-09 12:18:13 -04003160 char *end, *retval;
3161 int len, seq = 0;
3162 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003163
Al Viro48f5ec22013-09-09 15:22:25 -04003164 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003165restart:
3166 end = buf + buflen;
3167 len = buflen;
3168 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003169 if (buflen < 1)
3170 goto Elong;
3171 /* Get '/' right */
3172 retval = end-1;
3173 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003174 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003175 while (!IS_ROOT(dentry)) {
3176 struct dentry *parent = dentry->d_parent;
Nick Piggin9abca362011-01-07 17:49:36 +11003177 int error;
Ram Pai6092d042008-03-27 13:06:20 +01003178
Ram Pai6092d042008-03-27 13:06:20 +01003179 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003180 error = prepend_name(&end, &len, &dentry->d_name);
3181 if (error)
3182 break;
Ram Pai6092d042008-03-27 13:06:20 +01003183
3184 retval = end;
3185 dentry = parent;
3186 }
Al Viro48f5ec22013-09-09 15:22:25 -04003187 if (!(seq & 1))
3188 rcu_read_unlock();
3189 if (need_seqretry(&rename_lock, seq)) {
3190 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003191 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003192 }
3193 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003194 if (error)
3195 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003196 return retval;
3197Elong:
3198 return ERR_PTR(-ENAMETOOLONG);
3199}
Nick Pigginec2447c2011-01-07 17:49:29 +11003200
3201char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3202{
Waiman Long232d2d62013-09-09 12:18:13 -04003203 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003204}
3205EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003206
3207char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3208{
3209 char *p = NULL;
3210 char *retval;
3211
Al Viroc1031352010-06-06 22:31:14 -04003212 if (d_unlinked(dentry)) {
3213 p = buf + buflen;
3214 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3215 goto Elong;
3216 buflen++;
3217 }
3218 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003219 if (!IS_ERR(retval) && p)
3220 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003221 return retval;
3222Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003223 return ERR_PTR(-ENAMETOOLONG);
3224}
3225
Linus Torvalds8b19e342013-09-12 10:35:47 -07003226static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3227 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003228{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003229 unsigned seq;
3230
3231 do {
3232 seq = read_seqcount_begin(&fs->seq);
3233 *root = fs->root;
3234 *pwd = fs->pwd;
3235 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003236}
3237
Ram Pai6092d042008-03-27 13:06:20 +01003238/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 * NOTE! The user-level library version returns a
3240 * character pointer. The kernel system call just
3241 * returns the length of the buffer filled (which
3242 * includes the ending '\0' character), or a negative
3243 * error value. So libc would do something like
3244 *
3245 * char *getcwd(char * buf, size_t size)
3246 * {
3247 * int retval;
3248 *
3249 * retval = sys_getcwd(buf, size);
3250 * if (retval >= 0)
3251 * return buf;
3252 * errno = -retval;
3253 * return NULL;
3254 * }
3255 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003256SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257{
Linus Torvalds552ce542007-02-13 12:08:18 -08003258 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003259 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003260 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261
3262 if (!page)
3263 return -ENOMEM;
3264
Linus Torvalds8b19e342013-09-12 10:35:47 -07003265 rcu_read_lock();
3266 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267
Linus Torvalds552ce542007-02-13 12:08:18 -08003268 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003269 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003270 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003271 char *cwd = page + PATH_MAX;
3272 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003274 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003275 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003276 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003277
Al Viro02125a82011-12-05 08:43:34 -05003278 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003279 goto out;
3280
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003281 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003282 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003283 error = prepend_unreachable(&cwd, &buflen);
3284 if (error)
3285 goto out;
3286 }
3287
Linus Torvalds552ce542007-02-13 12:08:18 -08003288 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003289 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003290 if (len <= size) {
3291 error = len;
3292 if (copy_to_user(buf, cwd, len))
3293 error = -EFAULT;
3294 }
Nick Piggin949854d2011-01-07 17:49:37 +11003295 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003296 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298
3299out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003300 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 return error;
3302}
3303
3304/*
3305 * Test whether new_dentry is a subdirectory of old_dentry.
3306 *
3307 * Trivially implemented using the dcache structure
3308 */
3309
3310/**
3311 * is_subdir - is new dentry a subdirectory of old_dentry
3312 * @new_dentry: new dentry
3313 * @old_dentry: old dentry
3314 *
3315 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3316 * Returns 0 otherwise.
3317 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3318 */
3319
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003320int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321{
3322 int result;
Nick Piggin949854d2011-01-07 17:49:37 +11003323 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003325 if (new_dentry == old_dentry)
3326 return 1;
3327
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003328 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003331 /*
3332 * Need rcu_readlock to protect against the d_parent trashing
3333 * due to d_move
3334 */
3335 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003336 if (d_ancestor(old_dentry, new_dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003338 else
3339 result = 0;
Nick Piggin949854d2011-01-07 17:49:37 +11003340 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
3343 return result;
3344}
3345
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003346static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003348 struct dentry *root = data;
3349 if (dentry != root) {
3350 if (d_unhashed(dentry) || !dentry->d_inode)
3351 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003353 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3354 dentry->d_flags |= DCACHE_GENOCIDE;
3355 dentry->d_lockref.count--;
3356 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003358 return D_WALK_CONTINUE;
3359}
Nick Piggin58db63d2011-01-07 17:49:39 +11003360
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003361void d_genocide(struct dentry *parent)
3362{
3363 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364}
3365
Al Viro60545d02013-06-07 01:20:27 -04003366void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367{
Al Viro60545d02013-06-07 01:20:27 -04003368 inode_dec_link_count(inode);
3369 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3370 !hlist_unhashed(&dentry->d_alias) ||
3371 !d_unlinked(dentry));
3372 spin_lock(&dentry->d_parent->d_lock);
3373 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3374 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3375 (unsigned long long)inode->i_ino);
3376 spin_unlock(&dentry->d_lock);
3377 spin_unlock(&dentry->d_parent->d_lock);
3378 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379}
Al Viro60545d02013-06-07 01:20:27 -04003380EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381
3382static __initdata unsigned long dhash_entries;
3383static int __init set_dhash_entries(char *str)
3384{
3385 if (!str)
3386 return 0;
3387 dhash_entries = simple_strtoul(str, &str, 0);
3388 return 1;
3389}
3390__setup("dhash_entries=", set_dhash_entries);
3391
3392static void __init dcache_init_early(void)
3393{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003394 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395
3396 /* If hashes are distributed across NUMA nodes, defer
3397 * hash allocation until vmalloc space is available.
3398 */
3399 if (hashdist)
3400 return;
3401
3402 dentry_hashtable =
3403 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003404 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 dhash_entries,
3406 13,
3407 HASH_EARLY,
3408 &d_hash_shift,
3409 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003410 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411 0);
3412
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003413 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003414 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415}
3416
Denis Cheng74bf17c2007-10-16 23:26:30 -07003417static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003419 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420
3421 /*
3422 * A constructor could be added for stable state like the lists,
3423 * but it is probably not worth it because of the cache nature
3424 * of the dcache.
3425 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003426 dentry_cache = KMEM_CACHE(dentry,
3427 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428
3429 /* Hash may have been set up in dcache_init_early */
3430 if (!hashdist)
3431 return;
3432
3433 dentry_hashtable =
3434 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003435 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 dhash_entries,
3437 13,
3438 0,
3439 &d_hash_shift,
3440 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003441 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442 0);
3443
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003444 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003445 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446}
3447
3448/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003449struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003450EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003451
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452EXPORT_SYMBOL(d_genocide);
3453
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454void __init vfs_caches_init_early(void)
3455{
3456 dcache_init_early();
3457 inode_init_early();
3458}
3459
3460void __init vfs_caches_init(unsigned long mempages)
3461{
3462 unsigned long reserve;
3463
3464 /* Base hash sizes on available memory, with a reserve equal to
3465 150% of current kernel size */
3466
3467 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3468 mempages -= reserve;
3469
3470 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003471 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472
Denis Cheng74bf17c2007-10-16 23:26:30 -07003473 dcache_init();
3474 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003476 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003477 bdev_cache_init();
3478 chrdev_init();
3479}