blob: 2398f9f9433752aa53dbd3af4cfe0f82aba22d99 [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>
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -080041#include <linux/kasan.h>
42
David Howells07f3f052006-09-30 20:52:18 +020043#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050044#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Nick Piggin789680d2011-01-07 17:49:30 +110046/*
47 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110048 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040049 * - i_dentry, d_u.d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110050 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100054 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110055 * - the dcache lru lists and counters
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110060 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110061 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110062 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
Al Viro946e51f2014-10-26 19:19:16 -040064 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110065 *
66 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110067 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110068 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100069 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110070 * dcache_hash_bucket lock
71 * s_anon lock
Nick Piggin789680d2011-01-07 17:49:30 +110072 *
Nick Pigginda502952011-01-07 17:49:33 +110073 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
78 *
79 * If no ancestor relationship:
Nick Piggin789680d2011-01-07 17:49:30 +110080 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
83 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080084int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
Al Viro74c3cbe2007-07-22 08:04:18 -040087__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Nick Piggin949854d2011-01-07 17:49:37 +110089EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Christoph Lametere18b8902006-12-06 20:33:20 -080091static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Linus Torvalds1da177e2005-04-16 15:20:36 -070093/*
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
97 *
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
100 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800102static unsigned int d_hash_mask __read_mostly;
103static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700105static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106
Linus Torvalds8966be92012-03-02 14:23:30 -0800107static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700108 unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100109{
Linus Torvalds6d7d1a02012-03-19 16:19:53 -0700110 hash += (unsigned long) parent / L1_CACHE_BYTES;
Linus Torvalds99d263d2014-09-13 11:30:10 -0700111 return dentry_hashtable + hash_32(hash, d_hash_shift);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100112}
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/* Statistics gathering. */
115struct dentry_stat_t dentry_stat = {
116 .age_limit = 45,
117};
118
Glauber Costa3942c072013-08-28 10:17:53 +1000119static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000120static DEFINE_PER_CPU(long, nr_dentry_unused);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400121
122#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Dave Chinner62d36c72013-08-28 10:17:54 +1000123
124/*
125 * Here we resort to our own counters instead of using generic per-cpu counters
126 * for consistency with what the vfs inode code does. We are expected to harvest
127 * better code and performance by having our own specialized counters.
128 *
129 * Please note that the loop is done over all possible CPUs, not over all online
130 * CPUs. The reason for this is that we don't want to play games with CPUs going
131 * on and off. If one of them goes off, we will just keep their counters.
132 *
133 * glommer: See cffbc8a for details, and if you ever intend to change this,
134 * please update all vfs counters to match.
135 */
Glauber Costa3942c072013-08-28 10:17:53 +1000136static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100137{
138 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000139 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100140 for_each_possible_cpu(i)
141 sum += per_cpu(nr_dentry, i);
142 return sum < 0 ? 0 : sum;
143}
144
Dave Chinner62d36c72013-08-28 10:17:54 +1000145static long get_nr_dentry_unused(void)
146{
147 int i;
148 long sum = 0;
149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry_unused, i);
151 return sum < 0 ? 0 : sum;
152}
153
Joe Perches1f7e0612014-06-06 14:38:05 -0700154int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400155 size_t *lenp, loff_t *ppos)
156{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100157 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000158 dentry_stat.nr_unused = get_nr_dentry_unused();
Glauber Costa3942c072013-08-28 10:17:53 +1000159 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400160}
161#endif
162
Linus Torvalds5483f182012-03-04 15:51:42 -0800163/*
164 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165 * The strings are both count bytes long, and count is non-zero.
166 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700167#ifdef CONFIG_DCACHE_WORD_ACCESS
168
169#include <asm/word-at-a-time.h>
170/*
171 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172 * aligned allocation for this particular component. We don't
173 * strictly need the load_unaligned_zeropad() safety, but it
174 * doesn't hurt either.
175 *
176 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177 * need the careful unaligned handling.
178 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700179static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800180{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800181 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800182
183 for (;;) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -0700184 a = *(unsigned long *)cs;
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700185 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800186 if (tcount < sizeof(unsigned long))
187 break;
188 if (unlikely(a != b))
189 return 1;
190 cs += sizeof(unsigned long);
191 ct += sizeof(unsigned long);
192 tcount -= sizeof(unsigned long);
193 if (!tcount)
194 return 0;
195 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000196 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800197 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700198}
199
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800200#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700201
Linus Torvalds94753db52012-05-10 12:19:19 -0700202static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700203{
Linus Torvalds5483f182012-03-04 15:51:42 -0800204 do {
205 if (*cs != *ct)
206 return 1;
207 cs++;
208 ct++;
209 tcount--;
210 } while (tcount);
211 return 0;
212}
213
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700214#endif
215
Linus Torvalds94753db52012-05-10 12:19:19 -0700216static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217{
Linus Torvalds6326c712012-05-21 16:14:04 -0700218 const unsigned char *cs;
Linus Torvalds94753db52012-05-10 12:19:19 -0700219 /*
220 * Be careful about RCU walk racing with rename:
221 * use ACCESS_ONCE to fetch the name pointer.
222 *
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
230 *
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
234 */
Linus Torvalds6326c712012-05-21 16:14:04 -0700235 cs = ACCESS_ONCE(dentry->d_name.name);
236 smp_read_barrier_depends();
237 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700238}
239
Al Viro8d85b482014-09-29 14:54:27 -0400240struct external_name {
241 union {
242 atomic_t count;
243 struct rcu_head head;
244 } u;
245 unsigned char name[];
246};
247
248static inline struct external_name *external_name(struct dentry *dentry)
249{
250 return container_of(dentry->d_name.name, struct external_name, name[0]);
251}
252
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400253static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400255 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
256
Al Viro8d85b482014-09-29 14:54:27 -0400257 kmem_cache_free(dentry_cache, dentry);
258}
259
260static void __d_free_external(struct rcu_head *head)
261{
262 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Al Viro8d85b482014-09-29 14:54:27 -0400263 kfree(external_name(dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 kmem_cache_free(dentry_cache, dentry);
265}
266
Al Viro810bb172014-10-12 12:45:37 -0400267static inline int dname_external(const struct dentry *dentry)
268{
269 return dentry->d_name.name != dentry->d_iname;
270}
271
David Howells4bf46a22015-03-05 14:09:22 +0000272static inline void __d_set_inode_and_type(struct dentry *dentry,
273 struct inode *inode,
274 unsigned type_flags)
275{
276 unsigned flags;
277
278 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000279 flags = READ_ONCE(dentry->d_flags);
280 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
281 flags |= type_flags;
282 WRITE_ONCE(dentry->d_flags, flags);
283}
284
David Howells4bf46a22015-03-05 14:09:22 +0000285static inline void __d_clear_type_and_inode(struct dentry *dentry)
286{
287 unsigned flags = READ_ONCE(dentry->d_flags);
288
289 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
290 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000291 dentry->d_inode = NULL;
292}
293
Al Virob4f03542014-04-29 23:40:14 -0400294static void dentry_free(struct dentry *dentry)
295{
Al Viro946e51f2014-10-26 19:19:16 -0400296 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400297 if (unlikely(dname_external(dentry))) {
298 struct external_name *p = external_name(dentry);
299 if (likely(atomic_dec_and_test(&p->u.count))) {
300 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
301 return;
302 }
303 }
Al Virob4f03542014-04-29 23:40:14 -0400304 /* if dentry was never visible to RCU, immediate free is OK */
305 if (!(dentry->d_flags & DCACHE_RCUACCESS))
306 __d_free(&dentry->d_u.d_rcu);
307 else
308 call_rcu(&dentry->d_u.d_rcu, __d_free);
309}
310
Nick Piggin31e6b012011-01-07 17:49:52 +1100311/**
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200312 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800313 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100314 * After this call, in-progress rcu-walk path lookup will fail. This
315 * should be called after unhashing, and after changing d_inode (if
316 * the dentry has not already been unhashed).
317 */
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200318static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100319{
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200320 lockdep_assert_held(&dentry->d_lock);
321 /* Go through am invalidation barrier */
322 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100323}
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325/*
326 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100327 * d_iput() operation if defined. Dentry has no refcount
328 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800330static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200331 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100332 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
334 struct inode *inode = dentry->d_inode;
335 if (inode) {
David Howells4bf46a22015-03-05 14:09:22 +0000336 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400337 hlist_del_init(&dentry->d_u.d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100339 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700340 if (!inode->i_nlink)
341 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (dentry->d_op && dentry->d_op->d_iput)
343 dentry->d_op->d_iput(dentry, inode);
344 else
345 iput(inode);
346 } else {
347 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 }
349}
350
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700351/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100352 * Release the dentry's inode, using the filesystem
353 * d_iput() operation if defined. dentry remains in-use.
354 */
355static void dentry_unlink_inode(struct dentry * dentry)
356 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100357 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100358{
359 struct inode *inode = dentry->d_inode;
Al Viroa528aca2016-02-29 12:12:46 -0500360
361 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000362 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400363 hlist_del_init(&dentry->d_u.d_alias);
Al Viroa528aca2016-02-29 12:12:46 -0500364 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100365 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100366 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100367 if (!inode->i_nlink)
368 fsnotify_inoderemove(inode);
369 if (dentry->d_op && dentry->d_op->d_iput)
370 dentry->d_op->d_iput(dentry, inode);
371 else
372 iput(inode);
373}
374
375/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400376 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
377 * is in use - which includes both the "real" per-superblock
378 * LRU list _and_ the DCACHE_SHRINK_LIST use.
379 *
380 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
381 * on the shrink list (ie not on the superblock LRU list).
382 *
383 * The per-cpu "nr_dentry_unused" counters are updated with
384 * the DCACHE_LRU_LIST bit.
385 *
386 * These helper functions make sure we always follow the
387 * rules. d_lock must be held by the caller.
388 */
389#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
390static void d_lru_add(struct dentry *dentry)
391{
392 D_FLAG_VERIFY(dentry, 0);
393 dentry->d_flags |= DCACHE_LRU_LIST;
394 this_cpu_inc(nr_dentry_unused);
395 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
396}
397
398static void d_lru_del(struct dentry *dentry)
399{
400 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
401 dentry->d_flags &= ~DCACHE_LRU_LIST;
402 this_cpu_dec(nr_dentry_unused);
403 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
404}
405
406static void d_shrink_del(struct dentry *dentry)
407{
408 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
409 list_del_init(&dentry->d_lru);
410 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
411 this_cpu_dec(nr_dentry_unused);
412}
413
414static void d_shrink_add(struct dentry *dentry, struct list_head *list)
415{
416 D_FLAG_VERIFY(dentry, 0);
417 list_add(&dentry->d_lru, list);
418 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
419 this_cpu_inc(nr_dentry_unused);
420}
421
422/*
423 * These can only be called under the global LRU lock, ie during the
424 * callback for freeing the LRU list. "isolate" removes it from the
425 * LRU lists entirely, while shrink_move moves it to the indicated
426 * private list.
427 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800428static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400429{
430 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
431 dentry->d_flags &= ~DCACHE_LRU_LIST;
432 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800433 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400434}
435
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800436static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
437 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400438{
439 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
440 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800441 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400442}
443
444/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000445 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700446 */
447static void dentry_lru_add(struct dentry *dentry)
448{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400449 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
450 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700451}
452
Miklos Szeredid52b9082007-05-08 00:23:46 -0700453/**
Nick Piggin789680d2011-01-07 17:49:30 +1100454 * d_drop - drop a dentry
455 * @dentry: dentry to drop
456 *
457 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
458 * be found through a VFS lookup any more. Note that this is different from
459 * deleting the dentry - d_delete will try to mark the dentry negative if
460 * possible, giving a successful _negative_ lookup, while d_drop will
461 * just make the cache lookup fail.
462 *
463 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
464 * reason (NFS timeouts or autofs deletes).
465 *
466 * __d_drop requires dentry->d_lock.
467 */
468void __d_drop(struct dentry *dentry)
469{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700470 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400471 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400472 /*
473 * Hashed dentries are normally on the dentry hashtable,
474 * with the exception of those newly allocated by
475 * d_obtain_alias, which are always IS_ROOT:
476 */
477 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400478 b = &dentry->d_sb->s_anon;
479 else
480 b = d_hash(dentry->d_parent, dentry->d_name.hash);
481
482 hlist_bl_lock(b);
483 __hlist_bl_del(&dentry->d_hash);
484 dentry->d_hash.pprev = NULL;
485 hlist_bl_unlock(b);
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200486 dentry_rcuwalk_invalidate(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100487 }
488}
489EXPORT_SYMBOL(__d_drop);
490
491void d_drop(struct dentry *dentry)
492{
Nick Piggin789680d2011-01-07 17:49:30 +1100493 spin_lock(&dentry->d_lock);
494 __d_drop(dentry);
495 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100496}
497EXPORT_SYMBOL(d_drop);
498
Al Viroe55fd012014-05-28 13:51:12 -0400499static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100500{
Al Viro41edf272014-05-01 10:30:00 -0400501 struct dentry *parent = NULL;
502 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400503 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100504 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100505
Linus Torvalds0d984392013-09-08 13:46:52 -0700506 /*
507 * The dentry is now unrecoverably dead to the world.
508 */
509 lockref_mark_dead(&dentry->d_lockref);
510
Sage Weilf0023bc2011-10-28 10:02:42 -0700511 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700512 * inform the fs via d_prune that this dentry is about to be
513 * unhashed and destroyed.
514 */
Al Viro29266202014-05-30 11:39:02 -0400515 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800516 dentry->d_op->d_prune(dentry);
517
Al Viro01b60352014-04-29 23:42:52 -0400518 if (dentry->d_flags & DCACHE_LRU_LIST) {
519 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
520 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400521 }
Nick Piggin77812a12011-01-07 17:49:48 +1100522 /* if it was on the hash then remove it */
523 __d_drop(dentry);
Al Viroca5358e2014-10-26 19:31:10 -0400524 __list_del_entry(&dentry->d_child);
Al Viro03b3b882014-04-29 15:45:28 -0400525 /*
526 * Inform d_walk() that we are no longer attached to the
527 * dentry tree
528 */
529 dentry->d_flags |= DCACHE_DENTRY_KILLED;
530 if (parent)
531 spin_unlock(&parent->d_lock);
532 dentry_iput(dentry);
533 /*
534 * dentry_iput drops the locks, at which point nobody (except
535 * transient RCU lookups) can reach this dentry.
536 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800537 BUG_ON(dentry->d_lockref.count > 0);
Al Viro03b3b882014-04-29 15:45:28 -0400538 this_cpu_dec(nr_dentry);
539 if (dentry->d_op && dentry->d_op->d_release)
540 dentry->d_op->d_release(dentry);
541
Al Viro41edf272014-05-01 10:30:00 -0400542 spin_lock(&dentry->d_lock);
543 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
544 dentry->d_flags |= DCACHE_MAY_FREE;
545 can_free = false;
546 }
547 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400548 if (likely(can_free))
549 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400550}
551
552/*
553 * Finish off a dentry we've decided to kill.
554 * dentry->d_lock must be held, returns with it unlocked.
555 * If ref is non-zero, then decrement the refcount too.
556 * Returns dentry requiring refcount drop, or NULL if we're done.
557 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400558static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400559 __releases(dentry->d_lock)
560{
561 struct inode *inode = dentry->d_inode;
562 struct dentry *parent = NULL;
563
564 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
565 goto failed;
566
567 if (!IS_ROOT(dentry)) {
568 parent = dentry->d_parent;
569 if (unlikely(!spin_trylock(&parent->d_lock))) {
570 if (inode)
571 spin_unlock(&inode->i_lock);
572 goto failed;
573 }
574 }
575
576 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400577 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400578
579failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400580 spin_unlock(&dentry->d_lock);
581 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400582 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100583}
584
Al Viro046b9612014-05-29 08:54:52 -0400585static inline struct dentry *lock_parent(struct dentry *dentry)
586{
587 struct dentry *parent = dentry->d_parent;
588 if (IS_ROOT(dentry))
589 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800590 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400591 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400592 if (likely(spin_trylock(&parent->d_lock)))
593 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400594 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400595 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400596again:
597 parent = ACCESS_ONCE(dentry->d_parent);
598 spin_lock(&parent->d_lock);
599 /*
600 * We can't blindly lock dentry until we are sure
601 * that we won't violate the locking order.
602 * Any changes of dentry->d_parent must have
603 * been done with parent->d_lock held, so
604 * spin_lock() above is enough of a barrier
605 * for checking if it's still our child.
606 */
607 if (unlikely(parent != dentry->d_parent)) {
608 spin_unlock(&parent->d_lock);
609 goto again;
610 }
611 rcu_read_unlock();
612 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700613 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400614 else
615 parent = NULL;
616 return parent;
617}
618
Linus Torvalds360f5472015-01-09 15:19:03 -0800619/*
620 * Try to do a lockless dput(), and return whether that was successful.
621 *
622 * If unsuccessful, we return false, having already taken the dentry lock.
623 *
624 * The caller needs to hold the RCU read lock, so that the dentry is
625 * guaranteed to stay around even if the refcount goes down to zero!
626 */
627static inline bool fast_dput(struct dentry *dentry)
628{
629 int ret;
630 unsigned int d_flags;
631
632 /*
633 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100634 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800635 */
636 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
637 return lockref_put_or_lock(&dentry->d_lockref);
638
639 /*
640 * .. otherwise, we can try to just decrement the
641 * lockref optimistically.
642 */
643 ret = lockref_put_return(&dentry->d_lockref);
644
645 /*
646 * If the lockref_put_return() failed due to the lock being held
647 * by somebody else, the fast path has failed. We will need to
648 * get the lock, and then check the count again.
649 */
650 if (unlikely(ret < 0)) {
651 spin_lock(&dentry->d_lock);
652 if (dentry->d_lockref.count > 1) {
653 dentry->d_lockref.count--;
654 spin_unlock(&dentry->d_lock);
655 return 1;
656 }
657 return 0;
658 }
659
660 /*
661 * If we weren't the last ref, we're done.
662 */
663 if (ret)
664 return 1;
665
666 /*
667 * Careful, careful. The reference count went down
668 * to zero, but we don't hold the dentry lock, so
669 * somebody else could get it again, and do another
670 * dput(), and we need to not race with that.
671 *
672 * However, there is a very special and common case
673 * where we don't care, because there is nothing to
674 * do: the dentry is still hashed, it does not have
675 * a 'delete' op, and it's referenced and already on
676 * the LRU list.
677 *
678 * NOTE! Since we aren't locked, these values are
679 * not "stable". However, it is sufficient that at
680 * some point after we dropped the reference the
681 * dentry was hashed and the flags had the proper
682 * value. Other dentry users may have re-gotten
683 * a reference to the dentry and change that, but
684 * our work is done - we can leave the dentry
685 * around with a zero refcount.
686 */
687 smp_rmb();
688 d_flags = ACCESS_ONCE(dentry->d_flags);
Al Viro75a6f822015-07-08 02:42:38 +0100689 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800690
691 /* Nothing to do? Dropping the reference was all we needed? */
692 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
693 return 1;
694
695 /*
696 * Not the fast normal case? Get the lock. We've already decremented
697 * the refcount, but we'll need to re-check the situation after
698 * getting the lock.
699 */
700 spin_lock(&dentry->d_lock);
701
702 /*
703 * Did somebody else grab a reference to it in the meantime, and
704 * we're no longer the last user after all? Alternatively, somebody
705 * else could have killed it and marked it dead. Either way, we
706 * don't need to do anything else.
707 */
708 if (dentry->d_lockref.count) {
709 spin_unlock(&dentry->d_lock);
710 return 1;
711 }
712
713 /*
714 * Re-get the reference we optimistically dropped. We hold the
715 * lock, and we just tested that it was zero, so we can just
716 * set it to 1.
717 */
718 dentry->d_lockref.count = 1;
719 return 0;
720}
721
722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/*
724 * This is dput
725 *
726 * This is complicated by the fact that we do not want to put
727 * dentries that are no longer on any hash chain on the unused
728 * list: we'd much rather just get rid of them immediately.
729 *
730 * However, that implies that we have to traverse the dentry
731 * tree upwards to the parents which might _also_ now be
732 * scheduled for deletion (it may have been only waiting for
733 * its last child to go away).
734 *
735 * This tail recursion is done by hand as we don't want to depend
736 * on the compiler to always get this right (gcc generally doesn't).
737 * Real recursion would eat up our stack space.
738 */
739
740/*
741 * dput - release a dentry
742 * @dentry: dentry to release
743 *
744 * Release a dentry. This will drop the usage count and if appropriate
745 * call the dentry unlink method as well as removing it from the queues and
746 * releasing its resources. If the parent dentries were scheduled for release
747 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749void dput(struct dentry *dentry)
750{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700751 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 return;
753
754repeat:
Linus Torvalds360f5472015-01-09 15:19:03 -0800755 rcu_read_lock();
756 if (likely(fast_dput(dentry))) {
757 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800759 }
760
761 /* Slow case: now with the dentry lock held */
762 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700764 /* Unreachable? Get rid of it */
765 if (unlikely(d_unhashed(dentry)))
766 goto kill_it;
767
Al Viro75a6f822015-07-08 02:42:38 +0100768 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
769 goto kill_it;
770
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700771 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100773 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 }
Nick Piggin265ac902010-10-10 05:36:24 -0400775
Linus Torvalds358eec12013-10-31 15:43:02 -0700776 if (!(dentry->d_flags & DCACHE_REFERENCED))
777 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400778 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400779
Waiman Long98474232013-08-28 18:24:59 -0700780 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100781 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return;
783
Miklos Szeredid52b9082007-05-08 00:23:46 -0700784kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400785 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700786 if (dentry)
787 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700789EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100792/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100793static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
Waiman Long98474232013-08-28 18:24:59 -0700795 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
797
Nick Piggindc0474b2011-01-07 17:49:43 +1100798static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100799{
Waiman Long98474232013-08-28 18:24:59 -0700800 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100801}
802
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100803struct dentry *dget_parent(struct dentry *dentry)
804{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700805 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100806 struct dentry *ret;
807
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700808 /*
809 * Do optimistic parent lookup without any
810 * locking.
811 */
812 rcu_read_lock();
813 ret = ACCESS_ONCE(dentry->d_parent);
814 gotref = lockref_get_not_zero(&ret->d_lockref);
815 rcu_read_unlock();
816 if (likely(gotref)) {
817 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
818 return ret;
819 dput(ret);
820 }
821
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100822repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100823 /*
824 * Don't need rcu_dereference because we re-check it was correct under
825 * the lock.
826 */
827 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100828 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100829 spin_lock(&ret->d_lock);
830 if (unlikely(ret != dentry->d_parent)) {
831 spin_unlock(&ret->d_lock);
832 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100833 goto repeat;
834 }
Nick Piggina734eb42011-01-07 17:49:44 +1100835 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700836 BUG_ON(!ret->d_lockref.count);
837 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100838 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100839 return ret;
840}
841EXPORT_SYMBOL(dget_parent);
842
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843/**
844 * d_find_alias - grab a hashed alias of inode
845 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 *
847 * If inode has a hashed alias, or is a directory and has any alias,
848 * acquire the reference to alias and return it. Otherwise return NULL.
849 * Notice that if inode is a directory there can be only one alias and
850 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800851 * of a filesystem, or if the directory was renamed and d_revalidate
852 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700854 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500855 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500857static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
Nick Pigginda502952011-01-07 17:49:33 +1100859 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860
Nick Pigginda502952011-01-07 17:49:33 +1100861again:
862 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400863 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100864 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700866 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100867 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500869 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100870 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100871 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 return alias;
873 }
874 }
Nick Pigginda502952011-01-07 17:49:33 +1100875 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 }
Nick Pigginda502952011-01-07 17:49:33 +1100877 if (discon_alias) {
878 alias = discon_alias;
879 spin_lock(&alias->d_lock);
880 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500881 __dget_dlock(alias);
882 spin_unlock(&alias->d_lock);
883 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100884 }
885 spin_unlock(&alias->d_lock);
886 goto again;
887 }
888 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889}
890
Nick Pigginda502952011-01-07 17:49:33 +1100891struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892{
David Howells214fda12006-03-25 03:06:36 -0800893 struct dentry *de = NULL;
894
Al Virob3d9b7a2012-06-09 13:51:19 -0400895 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100896 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500897 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100898 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return de;
901}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700902EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
904/*
905 * Try to kill dentries associated with this inode.
906 * WARNING: you must own a reference to inode.
907 */
908void d_prune_aliases(struct inode *inode)
909{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700910 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100912 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400913 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700915 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400916 struct dentry *parent = lock_parent(dentry);
917 if (likely(!dentry->d_lockref.count)) {
918 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800919 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400920 goto restart;
921 }
922 if (parent)
923 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 }
925 spin_unlock(&dentry->d_lock);
926 }
Nick Piggin873feea2011-01-07 17:50:06 +1100927 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700929EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400931static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932{
Al Viro5c47e6d2014-04-29 16:13:18 -0400933 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700934
Miklos Szeredi60942f22014-05-02 15:38:39 -0400935 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400936 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400937 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100938 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400939 parent = lock_parent(dentry);
940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000942 * The dispose list is isolated and dentries are not accounted
943 * to the LRU here, so we can simply remove it from the list
944 * here regardless of whether it is referenced or not.
945 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400946 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000947
948 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000950 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800952 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700953 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400954 if (parent)
955 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 continue;
957 }
Nick Piggin77812a12011-01-07 17:49:48 +1100958
Al Viro64fd72e2014-05-28 09:48:44 -0400959
960 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
961 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
962 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400963 if (parent)
964 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400965 if (can_free)
966 dentry_free(dentry);
967 continue;
968 }
969
Al Viroff2fde92014-05-28 13:59:13 -0400970 inode = dentry->d_inode;
971 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400972 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000973 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400974 if (parent)
975 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400976 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000977 }
Al Viroff2fde92014-05-28 13:59:13 -0400978
Al Viroff2fde92014-05-28 13:59:13 -0400979 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400980
Al Viro5c47e6d2014-04-29 16:13:18 -0400981 /*
982 * We need to prune ancestors too. This is necessary to prevent
983 * quadratic behavior of shrink_dcache_parent(), but is also
984 * expected to be beneficial in reducing dentry cache
985 * fragmentation.
986 */
987 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400988 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
989 parent = lock_parent(dentry);
990 if (dentry->d_lockref.count != 1) {
991 dentry->d_lockref.count--;
992 spin_unlock(&dentry->d_lock);
993 if (parent)
994 spin_unlock(&parent->d_lock);
995 break;
996 }
997 inode = dentry->d_inode; /* can't be NULL */
998 if (unlikely(!spin_trylock(&inode->i_lock))) {
999 spin_unlock(&dentry->d_lock);
1000 if (parent)
1001 spin_unlock(&parent->d_lock);
1002 cpu_relax();
1003 continue;
1004 }
1005 __dentry_kill(dentry);
1006 dentry = parent;
1007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001009}
1010
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001011static enum lru_status dentry_lru_isolate(struct list_head *item,
1012 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001013{
1014 struct list_head *freeable = arg;
1015 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1016
1017
1018 /*
1019 * we are inverting the lru lock/dentry->d_lock here,
1020 * so use a trylock. If we fail to get the lock, just skip
1021 * it
1022 */
1023 if (!spin_trylock(&dentry->d_lock))
1024 return LRU_SKIP;
1025
1026 /*
1027 * Referenced dentries are still in use. If they have active
1028 * counts, just remove them from the LRU. Otherwise give them
1029 * another pass through the LRU.
1030 */
1031 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001032 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001033 spin_unlock(&dentry->d_lock);
1034 return LRU_REMOVED;
1035 }
1036
1037 if (dentry->d_flags & DCACHE_REFERENCED) {
1038 dentry->d_flags &= ~DCACHE_REFERENCED;
1039 spin_unlock(&dentry->d_lock);
1040
1041 /*
1042 * The list move itself will be made by the common LRU code. At
1043 * this point, we've dropped the dentry->d_lock but keep the
1044 * lru lock. This is safe to do, since every list movement is
1045 * protected by the lru lock even if both locks are held.
1046 *
1047 * This is guaranteed by the fact that all LRU management
1048 * functions are intermediated by the LRU API calls like
1049 * list_lru_add and list_lru_del. List movement in this file
1050 * only ever occur through this functions or through callbacks
1051 * like this one, that are called from the LRU API.
1052 *
1053 * The only exceptions to this are functions like
1054 * shrink_dentry_list, and code that first checks for the
1055 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1056 * operating only with stack provided lists after they are
1057 * properly isolated from the main list. It is thus, always a
1058 * local access.
1059 */
1060 return LRU_ROTATE;
1061 }
1062
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001063 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001064 spin_unlock(&dentry->d_lock);
1065
1066 return LRU_REMOVED;
1067}
1068
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001069/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001070 * prune_dcache_sb - shrink the dcache
1071 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001072 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001073 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001074 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1075 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001076 * function.
1077 *
1078 * This function may fail to free any resources if all the dentries are in
1079 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001080 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001081long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001082{
Dave Chinnerf6041562013-08-28 10:18:00 +10001083 LIST_HEAD(dispose);
1084 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001085
Vladimir Davydov503c3582015-02-12 14:58:47 -08001086 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1087 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001088 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001089 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090}
1091
Glauber Costa4e717f52013-08-28 10:18:03 +10001092static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001093 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001094{
Glauber Costa4e717f52013-08-28 10:18:03 +10001095 struct list_head *freeable = arg;
1096 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001097
Glauber Costa4e717f52013-08-28 10:18:03 +10001098 /*
1099 * we are inverting the lru lock/dentry->d_lock here,
1100 * so use a trylock. If we fail to get the lock, just skip
1101 * it
1102 */
1103 if (!spin_trylock(&dentry->d_lock))
1104 return LRU_SKIP;
1105
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001106 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001107 spin_unlock(&dentry->d_lock);
1108
1109 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001110}
1111
Glauber Costa4e717f52013-08-28 10:18:03 +10001112
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001113/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 * shrink_dcache_sb - shrink dcache for a superblock
1115 * @sb: superblock
1116 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001117 * Shrink the dcache for the specified super block. This is used to free
1118 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001120void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121{
Glauber Costa4e717f52013-08-28 10:18:03 +10001122 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001123
Glauber Costa4e717f52013-08-28 10:18:03 +10001124 do {
1125 LIST_HEAD(dispose);
1126
1127 freed = list_lru_walk(&sb->s_dentry_lru,
1128 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1129
1130 this_cpu_sub(nr_dentry_unused, freed);
1131 shrink_dentry_list(&dispose);
1132 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001134EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001137 * enum d_walk_ret - action to talke during tree walk
1138 * @D_WALK_CONTINUE: contrinue walk
1139 * @D_WALK_QUIT: quit walk
1140 * @D_WALK_NORETRY: quit when retry is needed
1141 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001143enum d_walk_ret {
1144 D_WALK_CONTINUE,
1145 D_WALK_QUIT,
1146 D_WALK_NORETRY,
1147 D_WALK_SKIP,
1148};
1149
1150/**
1151 * d_walk - walk the dentry tree
1152 * @parent: start of walk
1153 * @data: data passed to @enter() and @finish()
1154 * @enter: callback when first entering the dentry
1155 * @finish: callback when successfully finished the walk
1156 *
1157 * The @enter() and @finish() callbacks are called with d_lock held.
1158 */
1159static void d_walk(struct dentry *parent, void *data,
1160 enum d_walk_ret (*enter)(void *, struct dentry *),
1161 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162{
Nick Piggin949854d2011-01-07 17:49:37 +11001163 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001165 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001166 enum d_walk_ret ret;
1167 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001168
Nick Piggin58db63d2011-01-07 17:49:39 +11001169again:
Al Viro48f5ec22013-09-09 15:22:25 -04001170 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001171 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001172 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001173
1174 ret = enter(data, this_parent);
1175 switch (ret) {
1176 case D_WALK_CONTINUE:
1177 break;
1178 case D_WALK_QUIT:
1179 case D_WALK_SKIP:
1180 goto out_unlock;
1181 case D_WALK_NORETRY:
1182 retry = false;
1183 break;
1184 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185repeat:
1186 next = this_parent->d_subdirs.next;
1187resume:
1188 while (next != &this_parent->d_subdirs) {
1189 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001190 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001192
1193 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001194
1195 ret = enter(data, dentry);
1196 switch (ret) {
1197 case D_WALK_CONTINUE:
1198 break;
1199 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001200 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001201 goto out_unlock;
1202 case D_WALK_NORETRY:
1203 retry = false;
1204 break;
1205 case D_WALK_SKIP:
1206 spin_unlock(&dentry->d_lock);
1207 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001208 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001211 spin_unlock(&this_parent->d_lock);
1212 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001214 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 goto repeat;
1216 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001217 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
1219 /*
1220 * All done at this level ... ascend and resume the search.
1221 */
Al Viroca5358e2014-10-26 19:31:10 -04001222 rcu_read_lock();
1223ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001225 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001226 this_parent = child->d_parent;
1227
Al Viro31dec132013-10-25 17:04:27 -04001228 spin_unlock(&child->d_lock);
1229 spin_lock(&this_parent->d_lock);
1230
Al Viroca5358e2014-10-26 19:31:10 -04001231 /* might go back up the wrong parent if we have had a rename. */
1232 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001233 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001234 /* go into the first sibling still alive */
1235 do {
1236 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001237 if (next == &this_parent->d_subdirs)
1238 goto ascend;
1239 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001240 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001241 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 goto resume;
1243 }
Al Viroca5358e2014-10-26 19:31:10 -04001244 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001245 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001246 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001247 if (finish)
1248 finish(data);
1249
1250out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001251 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001252 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001253 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001254
1255rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001256 spin_unlock(&this_parent->d_lock);
1257 rcu_read_unlock();
1258 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001259 if (!retry)
1260 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001261 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001262 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001264
1265/*
1266 * Search for at least 1 mount point in the dentry's subdirs.
1267 * We descend to the next level whenever the d_subdirs
1268 * list is non-empty and continue searching.
1269 */
1270
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001271static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1272{
1273 int *ret = data;
1274 if (d_mountpoint(dentry)) {
1275 *ret = 1;
1276 return D_WALK_QUIT;
1277 }
1278 return D_WALK_CONTINUE;
1279}
1280
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001281/**
1282 * have_submounts - check for mounts over a dentry
1283 * @parent: dentry to check.
1284 *
1285 * Return true if the parent or its subdirectories contain
1286 * a mount point
1287 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001288int have_submounts(struct dentry *parent)
1289{
1290 int ret = 0;
1291
1292 d_walk(parent, &ret, check_mount, NULL);
1293
1294 return ret;
1295}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001296EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001299 * Called by mount code to set a mountpoint and check if the mountpoint is
1300 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1301 * subtree can become unreachable).
1302 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001303 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001304 * this reason take rename_lock and d_lock on dentry and ancestors.
1305 */
1306int d_set_mounted(struct dentry *dentry)
1307{
1308 struct dentry *p;
1309 int ret = -ENOENT;
1310 write_seqlock(&rename_lock);
1311 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001312 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001313 spin_lock(&p->d_lock);
1314 if (unlikely(d_unhashed(p))) {
1315 spin_unlock(&p->d_lock);
1316 goto out;
1317 }
1318 spin_unlock(&p->d_lock);
1319 }
1320 spin_lock(&dentry->d_lock);
1321 if (!d_unlinked(dentry)) {
1322 dentry->d_flags |= DCACHE_MOUNTED;
1323 ret = 0;
1324 }
1325 spin_unlock(&dentry->d_lock);
1326out:
1327 write_sequnlock(&rename_lock);
1328 return ret;
1329}
1330
1331/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001332 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 * and move any unused dentries to the end of the unused
1334 * list for prune_dcache(). We descend to the next level
1335 * whenever the d_subdirs list is non-empty and continue
1336 * searching.
1337 *
1338 * It returns zero iff there are no unused children,
1339 * otherwise it returns the number of children moved to
1340 * the end of the unused list. This may not be the total
1341 * number of unused children, because select_parent can
1342 * drop the lock and return early due to latency
1343 * constraints.
1344 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001345
1346struct select_data {
1347 struct dentry *start;
1348 struct list_head dispose;
1349 int found;
1350};
1351
1352static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001354 struct select_data *data = _data;
1355 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001357 if (data->start == dentry)
1358 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Al Virofe915222014-05-03 00:02:25 -04001360 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001361 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001362 } else {
1363 if (dentry->d_flags & DCACHE_LRU_LIST)
1364 d_lru_del(dentry);
1365 if (!dentry->d_lockref.count) {
1366 d_shrink_add(dentry, &data->dispose);
1367 data->found++;
1368 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 }
1370 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001371 * We can return to the caller if we have found some (this
1372 * ensures forward progress). We'll be coming back to find
1373 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 */
Al Virofe915222014-05-03 00:02:25 -04001375 if (!list_empty(&data->dispose))
1376 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001378 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379}
1380
1381/**
1382 * shrink_dcache_parent - prune dcache
1383 * @parent: parent of entries to prune
1384 *
1385 * Prune the dcache to remove unused children of the parent dentry.
1386 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001387void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001389 for (;;) {
1390 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001392 INIT_LIST_HEAD(&data.dispose);
1393 data.start = parent;
1394 data.found = 0;
1395
1396 d_walk(parent, &data, select_collect, NULL);
1397 if (!data.found)
1398 break;
1399
1400 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001401 cond_resched();
1402 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001404EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Al Viro9c8c10e2014-05-02 20:36:10 -04001406static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001407{
Al Viro9c8c10e2014-05-02 20:36:10 -04001408 /* it has busy descendents; complain about those instead */
1409 if (!list_empty(&dentry->d_subdirs))
1410 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001411
Al Viro9c8c10e2014-05-02 20:36:10 -04001412 /* root with refcount 1 is fine */
1413 if (dentry == _data && dentry->d_lockref.count == 1)
1414 return D_WALK_CONTINUE;
1415
1416 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1417 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001418 dentry,
1419 dentry->d_inode ?
1420 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001421 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001422 dentry->d_lockref.count,
1423 dentry->d_sb->s_type->name,
1424 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001425 WARN_ON(1);
1426 return D_WALK_CONTINUE;
1427}
1428
1429static void do_one_tree(struct dentry *dentry)
1430{
1431 shrink_dcache_parent(dentry);
1432 d_walk(dentry, dentry, umount_check, NULL);
1433 d_drop(dentry);
1434 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001435}
1436
1437/*
1438 * destroy the dentries attached to a superblock on unmounting
1439 */
1440void shrink_dcache_for_umount(struct super_block *sb)
1441{
1442 struct dentry *dentry;
1443
Al Viro9c8c10e2014-05-02 20:36:10 -04001444 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001445
1446 dentry = sb->s_root;
1447 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001448 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001449
1450 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001451 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1452 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001453 }
1454}
1455
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001456struct detach_data {
1457 struct select_data select;
1458 struct dentry *mountpoint;
1459};
1460static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001461{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001462 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001463
1464 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001465 __dget_dlock(dentry);
1466 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001467 return D_WALK_QUIT;
1468 }
1469
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001470 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001471}
1472
1473static void check_and_drop(void *_data)
1474{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001475 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001476
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001477 if (!data->mountpoint && !data->select.found)
1478 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001479}
1480
1481/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001482 * d_invalidate - detach submounts, prune dcache, and drop
1483 * @dentry: dentry to invalidate (aka detach, prune and drop)
1484 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001485 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001486 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001487 * The final d_drop is done as an atomic operation relative to
1488 * rename_lock ensuring there are no races with d_set_mounted. This
1489 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001490 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001491void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001492{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001493 /*
1494 * If it's already been dropped, return OK.
1495 */
1496 spin_lock(&dentry->d_lock);
1497 if (d_unhashed(dentry)) {
1498 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001499 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001500 }
1501 spin_unlock(&dentry->d_lock);
1502
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001503 /* Negative dentries can be dropped without further checks */
1504 if (!dentry->d_inode) {
1505 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001506 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001507 }
1508
1509 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001510 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001511
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001512 data.mountpoint = NULL;
1513 INIT_LIST_HEAD(&data.select.dispose);
1514 data.select.start = dentry;
1515 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001516
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001517 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001518
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001519 if (data.select.found)
1520 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001521
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001522 if (data.mountpoint) {
1523 detach_mounts(data.mountpoint);
1524 dput(data.mountpoint);
1525 }
1526
1527 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001528 break;
1529
1530 cond_resched();
1531 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001532}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001533EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535/**
Al Viroa4464db2011-07-07 15:03:58 -04001536 * __d_alloc - allocate a dcache entry
1537 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 * @name: qstr of the name
1539 *
1540 * Allocates a dentry. It returns %NULL if there is insufficient memory
1541 * available. On a success the dentry is returned. The name passed in is
1542 * copied and the copy passed in may be reused after this call.
1543 */
1544
Al Viroa4464db2011-07-07 15:03:58 -04001545struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
1547 struct dentry *dentry;
1548 char *dname;
1549
Mel Gormane12ba742007-10-16 01:25:52 -07001550 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 if (!dentry)
1552 return NULL;
1553
Linus Torvalds6326c712012-05-21 16:14:04 -07001554 /*
1555 * We guarantee that the inline name is always NUL-terminated.
1556 * This way the memcpy() done by the name switching in rename
1557 * will still always have a NUL at the end, even if we might
1558 * be overwriting an internal NUL character
1559 */
1560 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001562 size_t size = offsetof(struct external_name, name[1]);
Vladimir Davydov5d097052016-01-14 15:18:21 -08001563 struct external_name *p = kmalloc(size + name->len,
1564 GFP_KERNEL_ACCOUNT);
Al Viro8d85b482014-09-29 14:54:27 -04001565 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 kmem_cache_free(dentry_cache, dentry);
1567 return NULL;
1568 }
Al Viro8d85b482014-09-29 14:54:27 -04001569 atomic_set(&p->u.count, 1);
1570 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001571 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1572 kasan_unpoison_shadow(dname,
1573 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 } else {
1575 dname = dentry->d_iname;
1576 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
1578 dentry->d_name.len = name->len;
1579 dentry->d_name.hash = name->hash;
1580 memcpy(dname, name->name, name->len);
1581 dname[name->len] = 0;
1582
Linus Torvalds6326c712012-05-21 16:14:04 -07001583 /* Make sure we always see the terminating NUL character */
1584 smp_wmb();
1585 dentry->d_name.name = dname;
1586
Waiman Long98474232013-08-28 18:24:59 -07001587 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001588 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001590 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001592 dentry->d_parent = dentry;
1593 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 dentry->d_op = NULL;
1595 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001596 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 INIT_LIST_HEAD(&dentry->d_lru);
1598 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001599 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1600 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001601 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602
Nick Piggin3e880fb2011-01-07 17:49:19 +11001603 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 return dentry;
1606}
Al Viroa4464db2011-07-07 15:03:58 -04001607
1608/**
1609 * d_alloc - allocate a dcache entry
1610 * @parent: parent of entry to allocate
1611 * @name: qstr of the name
1612 *
1613 * Allocates a dentry. It returns %NULL if there is insufficient memory
1614 * available. On a success the dentry is returned. The name passed in is
1615 * copied and the copy passed in may be reused after this call.
1616 */
1617struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1618{
1619 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1620 if (!dentry)
1621 return NULL;
1622
1623 spin_lock(&parent->d_lock);
1624 /*
1625 * don't need child lock because it is not subject
1626 * to concurrency here
1627 */
1628 __dget_dlock(parent);
1629 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001630 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001631 spin_unlock(&parent->d_lock);
1632
1633 return dentry;
1634}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001635EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001637/**
1638 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1639 * @sb: the superblock
1640 * @name: qstr of the name
1641 *
1642 * For a filesystem that just pins its dentries in memory and never
1643 * performs lookups at all, return an unhashed IS_ROOT dentry.
1644 */
Nick Piggin4b936882011-01-07 17:50:07 +11001645struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1646{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001647 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001648}
1649EXPORT_SYMBOL(d_alloc_pseudo);
1650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1652{
1653 struct qstr q;
1654
1655 q.name = name;
1656 q.len = strlen(name);
1657 q.hash = full_name_hash(q.name, q.len);
1658 return d_alloc(parent, &q);
1659}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001660EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Nick Pigginfb045ad2011-01-07 17:49:55 +11001662void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1663{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001664 WARN_ON_ONCE(dentry->d_op);
1665 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001666 DCACHE_OP_COMPARE |
1667 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001668 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001669 DCACHE_OP_DELETE |
1670 DCACHE_OP_SELECT_INODE));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001671 dentry->d_op = op;
1672 if (!op)
1673 return;
1674 if (op->d_hash)
1675 dentry->d_flags |= DCACHE_OP_HASH;
1676 if (op->d_compare)
1677 dentry->d_flags |= DCACHE_OP_COMPARE;
1678 if (op->d_revalidate)
1679 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001680 if (op->d_weak_revalidate)
1681 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001682 if (op->d_delete)
1683 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001684 if (op->d_prune)
1685 dentry->d_flags |= DCACHE_OP_PRUNE;
David Howells4bacc9c2015-06-18 14:32:31 +01001686 if (op->d_select_inode)
1687 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001688
1689}
1690EXPORT_SYMBOL(d_set_d_op);
1691
David Howellsdf1a0852015-01-29 12:02:28 +00001692
1693/*
1694 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1695 * @dentry - The dentry to mark
1696 *
1697 * Mark a dentry as falling through to the lower layer (as set with
1698 * d_pin_lower()). This flag may be recorded on the medium.
1699 */
1700void d_set_fallthru(struct dentry *dentry)
1701{
1702 spin_lock(&dentry->d_lock);
1703 dentry->d_flags |= DCACHE_FALLTHRU;
1704 spin_unlock(&dentry->d_lock);
1705}
1706EXPORT_SYMBOL(d_set_fallthru);
1707
David Howellsb18825a2013-09-12 19:22:53 +01001708static unsigned d_flags_for_inode(struct inode *inode)
1709{
David Howells44bdb5e2015-01-29 12:02:29 +00001710 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001711
1712 if (!inode)
1713 return DCACHE_MISS_TYPE;
1714
1715 if (S_ISDIR(inode->i_mode)) {
1716 add_flags = DCACHE_DIRECTORY_TYPE;
1717 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1718 if (unlikely(!inode->i_op->lookup))
1719 add_flags = DCACHE_AUTODIR_TYPE;
1720 else
1721 inode->i_opflags |= IOP_LOOKUP;
1722 }
David Howells44bdb5e2015-01-29 12:02:29 +00001723 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001724 }
1725
David Howells44bdb5e2015-01-29 12:02:29 +00001726 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001727 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001728 add_flags = DCACHE_SYMLINK_TYPE;
1729 goto type_determined;
1730 }
1731 inode->i_opflags |= IOP_NOFOLLOW;
1732 }
1733
1734 if (unlikely(!S_ISREG(inode->i_mode)))
1735 add_flags = DCACHE_SPECIAL_TYPE;
1736
1737type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001738 if (unlikely(IS_AUTOMOUNT(inode)))
1739 add_flags |= DCACHE_NEED_AUTOMOUNT;
1740 return add_flags;
1741}
1742
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001743static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1744{
David Howellsb18825a2013-09-12 19:22:53 +01001745 unsigned add_flags = d_flags_for_inode(inode);
1746
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001747 spin_lock(&dentry->d_lock);
David Howellsb18825a2013-09-12 19:22:53 +01001748 if (inode)
Al Viro946e51f2014-10-26 19:19:16 -04001749 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05001750 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00001751 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05001752 raw_write_seqcount_end(&dentry->d_seq);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001753 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001754 fsnotify_d_instantiate(dentry, inode);
1755}
1756
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757/**
1758 * d_instantiate - fill in inode information for a dentry
1759 * @entry: dentry to complete
1760 * @inode: inode to attach to this dentry
1761 *
1762 * Fill in inode information in the entry.
1763 *
1764 * This turns negative dentries into productive full members
1765 * of society.
1766 *
1767 * NOTE! This assumes that the inode count has been incremented
1768 * (or otherwise set) by the caller to indicate that it is now
1769 * in use by the dcache.
1770 */
1771
1772void d_instantiate(struct dentry *entry, struct inode * inode)
1773{
Al Viro946e51f2014-10-26 19:19:16 -04001774 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001775 if (inode)
1776 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001777 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001778 if (inode)
1779 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 security_d_instantiate(entry, inode);
1781}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001782EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
1784/**
1785 * d_instantiate_unique - instantiate a non-aliased dentry
1786 * @entry: dentry to instantiate
1787 * @inode: inode to attach to this dentry
1788 *
1789 * Fill in inode information in the entry. On success, it returns NULL.
1790 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001791 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 *
1793 * Note that in order to avoid conflicts with rename() etc, the caller
1794 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001795 *
1796 * This also assumes that the inode count has been incremented
1797 * (or otherwise set) by the caller to indicate that it is now
1798 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 */
David Howells770bfad2006-08-22 20:06:07 -04001800static struct dentry *__d_instantiate_unique(struct dentry *entry,
1801 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802{
1803 struct dentry *alias;
1804 int len = entry->d_name.len;
1805 const char *name = entry->d_name.name;
1806 unsigned int hash = entry->d_name.hash;
1807
David Howells770bfad2006-08-22 20:06:07 -04001808 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001809 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001810 return NULL;
1811 }
1812
Al Viro946e51f2014-10-26 19:19:16 -04001813 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001814 /*
1815 * Don't need alias->d_lock here, because aliases with
1816 * d_parent == entry->d_parent are not subject to name or
1817 * parent changes, because the parent inode i_mutex is held.
1818 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001819 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 continue;
1821 if (alias->d_parent != entry->d_parent)
1822 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001823 if (alias->d_name.len != len)
1824 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001825 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001827 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return alias;
1829 }
David Howells770bfad2006-08-22 20:06:07 -04001830
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001831 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 return NULL;
1833}
David Howells770bfad2006-08-22 20:06:07 -04001834
1835struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1836{
1837 struct dentry *result;
1838
Al Viro946e51f2014-10-26 19:19:16 -04001839 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001840
Nick Piggin873feea2011-01-07 17:50:06 +11001841 if (inode)
1842 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001843 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001844 if (inode)
1845 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001846
1847 if (!result) {
1848 security_d_instantiate(entry, inode);
1849 return NULL;
1850 }
1851
1852 BUG_ON(!d_unhashed(result));
1853 iput(inode);
1854 return result;
1855}
1856
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857EXPORT_SYMBOL(d_instantiate_unique);
1858
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001859/**
1860 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1861 * @entry: dentry to complete
1862 * @inode: inode to attach to this dentry
1863 *
1864 * Fill in inode information in the entry. If a directory alias is found, then
1865 * return an error (and drop inode). Together with d_materialise_unique() this
1866 * guarantees that a directory inode may never have more than one alias.
1867 */
1868int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1869{
Al Viro946e51f2014-10-26 19:19:16 -04001870 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001871
1872 spin_lock(&inode->i_lock);
1873 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1874 spin_unlock(&inode->i_lock);
1875 iput(inode);
1876 return -EBUSY;
1877 }
1878 __d_instantiate(entry, inode);
1879 spin_unlock(&inode->i_lock);
1880 security_d_instantiate(entry, inode);
1881
1882 return 0;
1883}
1884EXPORT_SYMBOL(d_instantiate_no_diralias);
1885
Al Viroadc0e912012-01-08 16:49:21 -05001886struct dentry *d_make_root(struct inode *root_inode)
1887{
1888 struct dentry *res = NULL;
1889
1890 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001891 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001892
1893 res = __d_alloc(root_inode->i_sb, &name);
1894 if (res)
1895 d_instantiate(res, root_inode);
1896 else
1897 iput(root_inode);
1898 }
1899 return res;
1900}
1901EXPORT_SYMBOL(d_make_root);
1902
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001903static struct dentry * __d_find_any_alias(struct inode *inode)
1904{
1905 struct dentry *alias;
1906
Al Virob3d9b7a2012-06-09 13:51:19 -04001907 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001908 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001909 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001910 __dget(alias);
1911 return alias;
1912}
1913
Sage Weil46f72b32012-01-10 09:04:37 -08001914/**
1915 * d_find_any_alias - find any alias for a given inode
1916 * @inode: inode to find an alias for
1917 *
1918 * If any aliases exist for the given inode, take and return a
1919 * reference for one of them. If no aliases exist, return %NULL.
1920 */
1921struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001922{
1923 struct dentry *de;
1924
1925 spin_lock(&inode->i_lock);
1926 de = __d_find_any_alias(inode);
1927 spin_unlock(&inode->i_lock);
1928 return de;
1929}
Sage Weil46f72b32012-01-10 09:04:37 -08001930EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001931
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001932static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001933{
NeilBrownb911a6b2012-11-08 16:09:37 -08001934 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001935 struct dentry *tmp;
1936 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001937 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001938
1939 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001940 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001941 if (IS_ERR(inode))
1942 return ERR_CAST(inode);
1943
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001944 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001945 if (res)
1946 goto out_iput;
1947
Al Viroa4464db2011-07-07 15:03:58 -04001948 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001949 if (!tmp) {
1950 res = ERR_PTR(-ENOMEM);
1951 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001952 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001953
Nick Piggin873feea2011-01-07 17:50:06 +11001954 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001955 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001956 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001957 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001958 dput(tmp);
1959 goto out_iput;
1960 }
1961
1962 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001963 add_flags = d_flags_for_inode(inode);
1964
1965 if (disconnected)
1966 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001967
Christoph Hellwig9308a612008-08-11 15:49:12 +02001968 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001969 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001970 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001971 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001972 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001973 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001974 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001975 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001976 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001977
Christoph Hellwig9308a612008-08-11 15:49:12 +02001978 return tmp;
1979
1980 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001981 if (res && !IS_ERR(res))
1982 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001983 iput(inode);
1984 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001985}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001986
1987/**
1988 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1989 * @inode: inode to allocate the dentry for
1990 *
1991 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1992 * similar open by handle operations. The returned dentry may be anonymous,
1993 * or may have a full name (if the inode was already in the cache).
1994 *
1995 * When called on a directory inode, we must ensure that the inode only ever
1996 * has one dentry. If a dentry is found, that is returned instead of
1997 * allocating a new one.
1998 *
1999 * On successful return, the reference to the inode has been transferred
2000 * to the dentry. In case of an error the reference on the inode is released.
2001 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2002 * be passed in and the error will be propagated to the return value,
2003 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2004 */
2005struct dentry *d_obtain_alias(struct inode *inode)
2006{
2007 return __d_obtain_alias(inode, 1);
2008}
Benny Halevyadc48722009-02-27 14:02:59 -08002009EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
2011/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002012 * d_obtain_root - find or allocate a dentry for a given inode
2013 * @inode: inode to allocate the dentry for
2014 *
2015 * Obtain an IS_ROOT dentry for the root of a filesystem.
2016 *
2017 * We must ensure that directory inodes only ever have one dentry. If a
2018 * dentry is found, that is returned instead of allocating a new one.
2019 *
2020 * On successful return, the reference to the inode has been transferred
2021 * to the dentry. In case of an error the reference on the inode is
2022 * released. A %NULL or IS_ERR inode may be passed in and will be the
2023 * error will be propagate to the return value, with a %NULL @inode
2024 * replaced by ERR_PTR(-ESTALE).
2025 */
2026struct dentry *d_obtain_root(struct inode *inode)
2027{
2028 return __d_obtain_alias(inode, 0);
2029}
2030EXPORT_SYMBOL(d_obtain_root);
2031
2032/**
Barry Naujok94035402008-05-21 16:50:46 +10002033 * d_add_ci - lookup or allocate new dentry with case-exact name
2034 * @inode: the inode case-insensitive lookup has found
2035 * @dentry: the negative dentry that was passed to the parent's lookup func
2036 * @name: the case-exact name to be associated with the returned dentry
2037 *
2038 * This is to avoid filling the dcache with case-insensitive names to the
2039 * same inode, only the actual correct case is stored in the dcache for
2040 * case-insensitive filesystems.
2041 *
2042 * For a case-insensitive lookup match and if the the case-exact dentry
2043 * already exists in in the dcache, use it and return it.
2044 *
2045 * If no entry exists with the exact case name, allocate new dentry with
2046 * the exact case, and return the spliced entry.
2047 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002048struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002049 struct qstr *name)
2050{
Barry Naujok94035402008-05-21 16:50:46 +10002051 struct dentry *found;
2052 struct dentry *new;
2053
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002054 /*
2055 * First check if a dentry matching the name already exists,
2056 * if not go ahead and create it now.
2057 */
Barry Naujok94035402008-05-21 16:50:46 +10002058 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10002059 if (!found) {
2060 new = d_alloc(dentry->d_parent, name);
2061 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002062 found = ERR_PTR(-ENOMEM);
Al Viro427c77d2014-10-12 18:46:26 -04002063 } else {
2064 found = d_splice_alias(inode, new);
2065 if (found) {
2066 dput(new);
2067 return found;
2068 }
2069 return new;
Barry Naujok94035402008-05-21 16:50:46 +10002070 }
Barry Naujok94035402008-05-21 16:50:46 +10002071 }
Barry Naujok94035402008-05-21 16:50:46 +10002072 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002073 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002074}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002075EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002077/*
2078 * Do the slow-case of the dentry name compare.
2079 *
2080 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002081 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002082 * filesystem can rely on them, and can use the 'name' and
2083 * 'len' information without worrying about walking off the
2084 * end of memory etc.
2085 *
2086 * Thus the read_seqcount_retry() and the "duplicate" info
2087 * in arguments (the low-level filesystem should not look
2088 * at the dentry inode or name contents directly, since
2089 * rename can change them while we're in RCU mode).
2090 */
2091enum slow_d_compare {
2092 D_COMP_OK,
2093 D_COMP_NOMATCH,
2094 D_COMP_SEQRETRY,
2095};
2096
2097static noinline enum slow_d_compare slow_dentry_cmp(
2098 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002099 struct dentry *dentry,
2100 unsigned int seq,
2101 const struct qstr *name)
2102{
2103 int tlen = dentry->d_name.len;
2104 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002105
2106 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2107 cpu_relax();
2108 return D_COMP_SEQRETRY;
2109 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002110 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002111 return D_COMP_NOMATCH;
2112 return D_COMP_OK;
2113}
2114
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002116 * __d_lookup_rcu - search for a dentry (racy, store-free)
2117 * @parent: parent dentry
2118 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002119 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002120 * Returns: dentry, or NULL
2121 *
2122 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2123 * resolution (store-free path walking) design described in
2124 * Documentation/filesystems/path-lookup.txt.
2125 *
2126 * This is not to be used outside core vfs.
2127 *
2128 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2129 * held, and rcu_read_lock held. The returned dentry must not be stored into
2130 * without taking d_lock and checking d_seq sequence count against @seq
2131 * returned here.
2132 *
Linus Torvalds15570082013-09-02 11:38:06 -07002133 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002134 * function.
2135 *
2136 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2137 * the returned dentry, so long as its parent's seqlock is checked after the
2138 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2139 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002140 *
2141 * NOTE! The caller *has* to check the resulting dentry against the sequence
2142 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002143 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002144struct dentry *__d_lookup_rcu(const struct dentry *parent,
2145 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002146 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002147{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002148 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002149 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002150 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002151 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002152 struct dentry *dentry;
2153
2154 /*
2155 * Note: There is significant duplication with __d_lookup_rcu which is
2156 * required to prevent single threaded performance regressions
2157 * especially on architectures where smp_rmb (in seqcounts) are costly.
2158 * Keep the two functions in sync.
2159 */
2160
2161 /*
2162 * The hash list is protected using RCU.
2163 *
2164 * Carefully use d_seq when comparing a candidate dentry, to avoid
2165 * races with d_move().
2166 *
2167 * It is possible that concurrent renames can mess up our list
2168 * walk here and result in missing our dentry, resulting in the
2169 * false-negative result. d_lookup() protects against concurrent
2170 * renames using rename_lock seqlock.
2171 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002172 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002173 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002174 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002175 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002176
Nick Piggin31e6b012011-01-07 17:49:52 +11002177seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002178 /*
2179 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002180 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002181 *
2182 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002183 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002184 *
2185 * NOTE! We do a "raw" seqcount_begin here. That means that
2186 * we don't wait for the sequence count to stabilize if it
2187 * is in the middle of a sequence change. If we do the slow
2188 * dentry compare, we will do seqretries until it is stable,
2189 * and if we end up with a successful lookup, we actually
2190 * want to exit RCU lookup anyway.
2191 */
2192 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002193 if (dentry->d_parent != parent)
2194 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002195 if (d_unhashed(dentry))
2196 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002197
2198 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002199 if (dentry->d_name.hash != hashlen_hash(hashlen))
2200 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002201 *seqp = seq;
2202 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002203 case D_COMP_OK:
2204 return dentry;
2205 case D_COMP_NOMATCH:
2206 continue;
2207 default:
2208 goto seqretry;
2209 }
2210 }
2211
Linus Torvalds26fe5752012-05-10 13:14:12 -07002212 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002213 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002214 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002215 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002216 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002217 }
2218 return NULL;
2219}
2220
2221/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 * d_lookup - search for a dentry
2223 * @parent: parent dentry
2224 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002225 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002227 * d_lookup searches the children of the parent dentry for the name in
2228 * question. If the dentry is found its reference count is incremented and the
2229 * dentry is returned. The caller must use dput to free the entry when it has
2230 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 */
Al Viroda2d8452013-01-24 18:29:34 -05002232struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233{
Nick Piggin31e6b012011-01-07 17:49:52 +11002234 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002235 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
Daeseok Younb8314f92014-08-11 11:46:53 +09002237 do {
2238 seq = read_seqbegin(&rename_lock);
2239 dentry = __d_lookup(parent, name);
2240 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 break;
2242 } while (read_seqretry(&rename_lock, seq));
2243 return dentry;
2244}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002245EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Nick Piggin31e6b012011-01-07 17:49:52 +11002247/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002248 * __d_lookup - search for a dentry (racy)
2249 * @parent: parent dentry
2250 * @name: qstr of name we wish to find
2251 * Returns: dentry, or NULL
2252 *
2253 * __d_lookup is like d_lookup, however it may (rarely) return a
2254 * false-negative result due to unrelated rename activity.
2255 *
2256 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2257 * however it must be used carefully, eg. with a following d_lookup in
2258 * the case of failure.
2259 *
2260 * __d_lookup callers must be commented.
2261 */
Al Viroa713ca22013-01-24 18:27:00 -05002262struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263{
2264 unsigned int len = name->len;
2265 unsigned int hash = name->hash;
2266 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002267 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002268 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002269 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002270 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
Nick Pigginb04f7842010-08-18 04:37:34 +10002272 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002273 * Note: There is significant duplication with __d_lookup_rcu which is
2274 * required to prevent single threaded performance regressions
2275 * especially on architectures where smp_rmb (in seqcounts) are costly.
2276 * Keep the two functions in sync.
2277 */
2278
2279 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002280 * The hash list is protected using RCU.
2281 *
2282 * Take d_lock when comparing a candidate dentry, to avoid races
2283 * with d_move().
2284 *
2285 * It is possible that concurrent renames can mess up our list
2286 * walk here and result in missing our dentry, resulting in the
2287 * false-negative result. d_lookup() protects against concurrent
2288 * renames using rename_lock seqlock.
2289 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002290 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002291 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 rcu_read_lock();
2293
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002294 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 if (dentry->d_name.hash != hash)
2297 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
2299 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 if (dentry->d_parent != parent)
2301 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002302 if (d_unhashed(dentry))
2303 goto next;
2304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 /*
2306 * It is safe to compare names since d_move() cannot
2307 * change the qstr (protected by d_lock).
2308 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002309 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002310 int tlen = dentry->d_name.len;
2311 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002312 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 goto next;
2314 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002315 if (dentry->d_name.len != len)
2316 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002317 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 goto next;
2319 }
2320
Waiman Long98474232013-08-28 18:24:59 -07002321 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002322 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 spin_unlock(&dentry->d_lock);
2324 break;
2325next:
2326 spin_unlock(&dentry->d_lock);
2327 }
2328 rcu_read_unlock();
2329
2330 return found;
2331}
2332
2333/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002334 * d_hash_and_lookup - hash the qstr then search for a dentry
2335 * @dir: Directory to search in
2336 * @name: qstr of name we wish to find
2337 *
Al Viro4f522a22013-02-11 23:20:37 -05002338 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002339 */
2340struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2341{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002342 /*
2343 * Check for a fs-specific hash function. Note that we must
2344 * calculate the standard hash first, as the d_op->d_hash()
2345 * routine may choose to leave the hash value unchanged.
2346 */
2347 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002348 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002349 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002350 if (unlikely(err < 0))
2351 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002352 }
Al Viro4f522a22013-02-11 23:20:37 -05002353 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002354}
Al Viro4f522a22013-02-11 23:20:37 -05002355EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002356
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357/*
2358 * When a file is deleted, we have two options:
2359 * - turn this dentry into a negative dentry
2360 * - unhash this dentry and free it.
2361 *
2362 * Usually, we want to just turn this into
2363 * a negative dentry, but if anybody else is
2364 * currently using the dentry or the inode
2365 * we can't do that and we fall back on removing
2366 * it from the hash queues and waiting for
2367 * it to be deleted later when it has no users
2368 */
2369
2370/**
2371 * d_delete - delete a dentry
2372 * @dentry: The dentry to delete
2373 *
2374 * Turn the dentry into a negative dentry if possible, otherwise
2375 * remove it from the hash queues so it can be deleted later
2376 */
2377
2378void d_delete(struct dentry * dentry)
2379{
Nick Piggin873feea2011-01-07 17:50:06 +11002380 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002381 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 /*
2383 * Are we the only user?
2384 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002385again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002387 inode = dentry->d_inode;
2388 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002389 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002390 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002391 spin_unlock(&dentry->d_lock);
2392 cpu_relax();
2393 goto again;
2394 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002395 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002396 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002397 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 return;
2399 }
2400
2401 if (!d_unhashed(dentry))
2402 __d_drop(dentry);
2403
2404 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002405
2406 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002408EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002410static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002412 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002413 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002414 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002415 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002416 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417}
2418
David Howells770bfad2006-08-22 20:06:07 -04002419static void _d_rehash(struct dentry * entry)
2420{
2421 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2422}
2423
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424/**
2425 * d_rehash - add an entry back to the hash
2426 * @entry: dentry to add to the hash
2427 *
2428 * Adds a dentry to the hash according to its name.
2429 */
2430
2431void d_rehash(struct dentry * entry)
2432{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002434 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002437EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002439/**
2440 * dentry_update_name_case - update case insensitive dentry with a new name
2441 * @dentry: dentry to be updated
2442 * @name: new name
2443 *
2444 * Update a case insensitive dentry with new case of name.
2445 *
2446 * dentry must have been returned by d_lookup with name @name. Old and new
2447 * name lengths must match (ie. no d_compare which allows mismatched name
2448 * lengths).
2449 *
2450 * Parent inode i_mutex must be held over d_lookup and into this call (to
2451 * keep renames and concurrent inserts, and readdir(2) away).
2452 */
2453void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2454{
Al Viro59551022016-01-22 15:40:57 -05002455 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002456 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2457
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002458 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002459 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002460 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002461 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002462 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002463}
2464EXPORT_SYMBOL(dentry_update_name_case);
2465
Al Viro8d85b482014-09-29 14:54:27 -04002466static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467{
Al Viro8d85b482014-09-29 14:54:27 -04002468 if (unlikely(dname_external(target))) {
2469 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 /*
2471 * Both external: swap the pointers
2472 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002473 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 } else {
2475 /*
2476 * dentry:internal, target:external. Steal target's
2477 * storage and make target internal.
2478 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002479 memcpy(target->d_iname, dentry->d_name.name,
2480 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 dentry->d_name.name = target->d_name.name;
2482 target->d_name.name = target->d_iname;
2483 }
2484 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002485 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 /*
2487 * dentry:external, target:internal. Give dentry's
2488 * storage to target and make dentry internal
2489 */
2490 memcpy(dentry->d_iname, target->d_name.name,
2491 target->d_name.len + 1);
2492 target->d_name.name = dentry->d_name.name;
2493 dentry->d_name.name = dentry->d_iname;
2494 } else {
2495 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002496 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002498 unsigned int i;
2499 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002500 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2501 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002502 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2503 swap(((long *) &dentry->d_iname)[i],
2504 ((long *) &target->d_iname)[i]);
2505 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 }
2507 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002508 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509}
2510
Al Viro8d85b482014-09-29 14:54:27 -04002511static void copy_name(struct dentry *dentry, struct dentry *target)
2512{
2513 struct external_name *old_name = NULL;
2514 if (unlikely(dname_external(dentry)))
2515 old_name = external_name(dentry);
2516 if (unlikely(dname_external(target))) {
2517 atomic_inc(&external_name(target)->u.count);
2518 dentry->d_name = target->d_name;
2519 } else {
2520 memcpy(dentry->d_iname, target->d_name.name,
2521 target->d_name.len + 1);
2522 dentry->d_name.name = dentry->d_iname;
2523 dentry->d_name.hash_len = target->d_name.hash_len;
2524 }
2525 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2526 kfree_rcu(old_name, u.head);
2527}
2528
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002529static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2530{
2531 /*
2532 * XXXX: do we really need to take target->d_lock?
2533 */
2534 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2535 spin_lock(&target->d_parent->d_lock);
2536 else {
2537 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2538 spin_lock(&dentry->d_parent->d_lock);
2539 spin_lock_nested(&target->d_parent->d_lock,
2540 DENTRY_D_LOCK_NESTED);
2541 } else {
2542 spin_lock(&target->d_parent->d_lock);
2543 spin_lock_nested(&dentry->d_parent->d_lock,
2544 DENTRY_D_LOCK_NESTED);
2545 }
2546 }
2547 if (target < dentry) {
2548 spin_lock_nested(&target->d_lock, 2);
2549 spin_lock_nested(&dentry->d_lock, 3);
2550 } else {
2551 spin_lock_nested(&dentry->d_lock, 2);
2552 spin_lock_nested(&target->d_lock, 3);
2553 }
2554}
2555
Al Viro986c0192014-09-26 23:11:15 -04002556static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002557{
2558 if (target->d_parent != dentry->d_parent)
2559 spin_unlock(&dentry->d_parent->d_lock);
2560 if (target->d_parent != target)
2561 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002562 spin_unlock(&target->d_lock);
2563 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002564}
2565
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002567 * When switching names, the actual string doesn't strictly have to
2568 * be preserved in the target - because we're dropping the target
2569 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002570 * the new name before we switch, unless we are going to rehash
2571 * it. Note that if we *do* unhash the target, we are not allowed
2572 * to rehash it without giving it a new name/hash key - whether
2573 * we swap or overwrite the names here, resulting name won't match
2574 * the reality in filesystem; it's only there for d_path() purposes.
2575 * Note that all of this is happening under rename_lock, so the
2576 * any hash lookup seeing it in the middle of manipulations will
2577 * be discarded anyway. So we do not care what happens to the hash
2578 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002580/*
Al Viro18367502011-07-12 21:42:24 -04002581 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 * @dentry: entry to move
2583 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002584 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 *
2586 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002587 * dcache entries should not be moved in this way. Caller must hold
2588 * rename_lock, the i_mutex of the source and target directories,
2589 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002591static void __d_move(struct dentry *dentry, struct dentry *target,
2592 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 if (!dentry->d_inode)
2595 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2596
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002597 BUG_ON(d_ancestor(dentry, target));
2598 BUG_ON(d_ancestor(target, dentry));
2599
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002600 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601
Nick Piggin31e6b012011-01-07 17:49:52 +11002602 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002603 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002604
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002605 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2606
2607 /*
2608 * Move the dentry to the target hash queue. Don't bother checking
2609 * for the same hash queue because of how unlikely it is.
2610 */
2611 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002612 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
Miklos Szeredida1ce062014-04-01 17:08:43 +02002614 /*
2615 * Unhash the target (d_delete() is not usable here). If exchanging
2616 * the two dentries, then rehash onto the other's hash queue.
2617 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002619 if (exchange) {
2620 __d_rehash(target,
2621 d_hash(dentry->d_parent, dentry->d_name.hash));
2622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002625 if (exchange)
2626 swap_names(dentry, target);
2627 else
2628 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
Al Viro63cf4272014-09-26 23:06:14 -04002630 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002632 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 dentry->d_parent = target->d_parent;
2634 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002635 list_del_init(&target->d_child);
2636 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002638 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002639 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002640 list_move(&target->d_child, &target->d_parent->d_subdirs);
2641 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002642 if (exchange)
2643 fsnotify_d_move(target);
2644 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 }
2646
Nick Piggin31e6b012011-01-07 17:49:52 +11002647 write_seqcount_end(&target->d_seq);
2648 write_seqcount_end(&dentry->d_seq);
2649
Al Viro986c0192014-09-26 23:11:15 -04002650 dentry_unlock_for_move(dentry, target);
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);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002665 __d_move(dentry, target, false);
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
Miklos Szeredida1ce062014-04-01 17:08:43 +02002670/*
2671 * d_exchange - exchange two dentries
2672 * @dentry1: first dentry
2673 * @dentry2: second dentry
2674 */
2675void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2676{
2677 write_seqlock(&rename_lock);
2678
2679 WARN_ON(!dentry1->d_inode);
2680 WARN_ON(!dentry2->d_inode);
2681 WARN_ON(IS_ROOT(dentry1));
2682 WARN_ON(IS_ROOT(dentry2));
2683
2684 __d_move(dentry1, dentry2, true);
2685
2686 write_sequnlock(&rename_lock);
2687}
2688
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002689/**
2690 * d_ancestor - search for an ancestor
2691 * @p1: ancestor dentry
2692 * @p2: child dentry
2693 *
2694 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2695 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002696 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002697struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002698{
2699 struct dentry *p;
2700
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002701 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002702 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002703 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002704 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002705 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002706}
2707
2708/*
2709 * This helper attempts to cope with remotely renamed directories
2710 *
2711 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002712 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002713 *
2714 * Note: If ever the locking in lock_rename() changes, then please
2715 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002716 */
Al Virob5ae6b12014-10-12 22:16:02 -04002717static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002718 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002719{
2720 struct mutex *m1 = NULL, *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002721 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002722
2723 /* If alias and dentry share a parent, then no extra locks required */
2724 if (alias->d_parent == dentry->d_parent)
2725 goto out_unalias;
2726
Trond Myklebust9eaef272006-10-21 10:24:20 -07002727 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002728 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2729 goto out_err;
2730 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro59551022016-01-22 15:40:57 -05002731 if (!inode_trylock(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07002732 goto out_err;
2733 m2 = &alias->d_parent->d_inode->i_mutex;
2734out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002735 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002736 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002737out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002738 if (m2)
2739 mutex_unlock(m2);
2740 if (m1)
2741 mutex_unlock(m1);
2742 return ret;
2743}
2744
David Howells770bfad2006-08-22 20:06:07 -04002745/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002746 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2747 * @inode: the inode which may have a disconnected dentry
2748 * @dentry: a negative dentry which we want to point to the inode.
2749 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002750 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2751 * place of the given dentry and return it, else simply d_add the inode
2752 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002753 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002754 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2755 * we should error out: directories can't have multiple aliases.
2756 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002757 * This is needed in the lookup routine of any filesystem that is exportable
2758 * (via knfsd) so that we can build dcache paths to directories effectively.
2759 *
2760 * If a dentry was found and moved, then it is returned. Otherwise NULL
2761 * is returned. This matches the expected return value of ->lookup.
2762 *
2763 * Cluster filesystems may call this function with a negative, hashed dentry.
2764 * In that case, we know that the inode will be a regular file, and also this
2765 * will only occur during atomic_open. So we need to check for the dentry
2766 * being already hashed only in the final case.
2767 */
2768struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2769{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002770 if (IS_ERR(inode))
2771 return ERR_CAST(inode);
2772
David Howells770bfad2006-08-22 20:06:07 -04002773 BUG_ON(!d_unhashed(dentry));
2774
David Howells770bfad2006-08-22 20:06:07 -04002775 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002776 __d_instantiate(dentry, NULL);
Al Virob5ae6b12014-10-12 22:16:02 -04002777 goto out;
David Howells770bfad2006-08-22 20:06:07 -04002778 }
Nick Piggin873feea2011-01-07 17:50:06 +11002779 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002780 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002781 struct dentry *new = __d_find_any_alias(inode);
2782 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002783 /* The reference to new ensures it remains an alias */
2784 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002785 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002786 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002787 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002788 dput(new);
2789 new = ERR_PTR(-ELOOP);
2790 pr_warn_ratelimited(
2791 "VFS: Lookup of '%s' in %s %s"
2792 " would have caused loop\n",
2793 dentry->d_name.name,
2794 inode->i_sb->s_type->name,
2795 inode->i_sb->s_id);
2796 } else if (!IS_ROOT(new)) {
2797 int err = __d_unalias(inode, dentry, new);
2798 write_sequnlock(&rename_lock);
2799 if (err) {
2800 dput(new);
2801 new = ERR_PTR(err);
2802 }
Al Viro18367502011-07-12 21:42:24 -04002803 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002804 __d_move(new, dentry, false);
2805 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002806 security_d_instantiate(new, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002807 }
Al Virob5ae6b12014-10-12 22:16:02 -04002808 iput(inode);
2809 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002810 }
David Howells770bfad2006-08-22 20:06:07 -04002811 }
Al Virob5ae6b12014-10-12 22:16:02 -04002812 /* already taking inode->i_lock, so d_add() by hand */
2813 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002814 spin_unlock(&inode->i_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002815out:
2816 security_d_instantiate(dentry, inode);
2817 d_rehash(dentry);
2818 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002819}
Al Virob5ae6b12014-10-12 22:16:02 -04002820EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002821
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002822static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002823{
2824 *buflen -= namelen;
2825 if (*buflen < 0)
2826 return -ENAMETOOLONG;
2827 *buffer -= namelen;
2828 memcpy(*buffer, str, namelen);
2829 return 0;
2830}
2831
Waiman Long232d2d62013-09-09 12:18:13 -04002832/**
2833 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002834 * @buffer: buffer pointer
2835 * @buflen: allocated length of the buffer
2836 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002837 *
2838 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2839 * make sure that either the old or the new name pointer and length are
2840 * fetched. However, there may be mismatch between length and pointer.
2841 * The length cannot be trusted, we need to copy it byte-by-byte until
2842 * the length is reached or a null byte is found. It also prepends "/" at
2843 * the beginning of the name. The sequence number check at the caller will
2844 * retry it again when a d_move() does happen. So any garbage in the buffer
2845 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002846 *
2847 * Data dependency barrier is needed to make sure that we see that terminating
2848 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002849 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002850static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2851{
Waiman Long232d2d62013-09-09 12:18:13 -04002852 const char *dname = ACCESS_ONCE(name->name);
2853 u32 dlen = ACCESS_ONCE(name->len);
2854 char *p;
2855
Al Viro6d13f692014-09-29 14:46:30 -04002856 smp_read_barrier_depends();
2857
Waiman Long232d2d62013-09-09 12:18:13 -04002858 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002859 if (*buflen < 0)
2860 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002861 p = *buffer -= dlen + 1;
2862 *p++ = '/';
2863 while (dlen--) {
2864 char c = *dname++;
2865 if (!c)
2866 break;
2867 *p++ = c;
2868 }
2869 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002870}
2871
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002873 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002874 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002875 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002876 * @buffer: pointer to the end of the buffer
2877 * @buflen: pointer to buffer length
2878 *
Waiman Long18129972013-09-12 10:55:35 -04002879 * The function will first try to write out the pathname without taking any
2880 * lock other than the RCU read lock to make sure that dentries won't go away.
2881 * It only checks the sequence number of the global rename_lock as any change
2882 * in the dentry's d_seq will be preceded by changes in the rename_lock
2883 * sequence number. If the sequence number had been changed, it will restart
2884 * the whole pathname back-tracing sequence again by taking the rename_lock.
2885 * In this case, there is no need to take the RCU read lock as the recursive
2886 * parent pointer references will keep the dentry chain alive as long as no
2887 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002888 */
Al Viro02125a82011-12-05 08:43:34 -05002889static int prepend_path(const struct path *path,
2890 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002891 char **buffer, int *buflen)
2892{
Al Viroede4ceb2013-11-13 07:45:40 -05002893 struct dentry *dentry;
2894 struct vfsmount *vfsmnt;
2895 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002896 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002897 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002898 char *bptr;
2899 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002900
Al Viro48f5ec22013-09-09 15:22:25 -04002901 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002902restart_mnt:
2903 read_seqbegin_or_lock(&mount_lock, &m_seq);
2904 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002905 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002906restart:
2907 bptr = *buffer;
2908 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002909 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002910 dentry = path->dentry;
2911 vfsmnt = path->mnt;
2912 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002913 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002914 while (dentry != root->dentry || vfsmnt != root->mnt) {
2915 struct dentry * parent;
2916
2917 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002918 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05002919 /* Escaped? */
2920 if (dentry != vfsmnt->mnt_root) {
2921 bptr = *buffer;
2922 blen = *buflen;
2923 error = 3;
2924 break;
2925 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002926 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002927 if (mnt != parent) {
2928 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2929 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002930 vfsmnt = &mnt->mnt;
2931 continue;
2932 }
Waiman Long232d2d62013-09-09 12:18:13 -04002933 if (!error)
2934 error = is_mounted(vfsmnt) ? 1 : 2;
2935 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002936 }
2937 parent = dentry->d_parent;
2938 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002939 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002940 if (error)
2941 break;
2942
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002943 dentry = parent;
2944 }
Al Viro48f5ec22013-09-09 15:22:25 -04002945 if (!(seq & 1))
2946 rcu_read_unlock();
2947 if (need_seqretry(&rename_lock, seq)) {
2948 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002949 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002950 }
Al Viro48f5ec22013-09-09 15:22:25 -04002951 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002952
2953 if (!(m_seq & 1))
2954 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002955 if (need_seqretry(&mount_lock, m_seq)) {
2956 m_seq = 1;
2957 goto restart_mnt;
2958 }
2959 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002960
Waiman Long232d2d62013-09-09 12:18:13 -04002961 if (error >= 0 && bptr == *buffer) {
2962 if (--blen < 0)
2963 error = -ENAMETOOLONG;
2964 else
2965 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002966 }
Waiman Long232d2d62013-09-09 12:18:13 -04002967 *buffer = bptr;
2968 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002969 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002970}
2971
2972/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002973 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002974 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002975 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002976 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 * @buflen: buffer length
2978 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002979 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002981 * Returns a pointer into the buffer or an error code if the
2982 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002983 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002984 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002985 *
Al Viro02125a82011-12-05 08:43:34 -05002986 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 */
Al Viro02125a82011-12-05 08:43:34 -05002988char *__d_path(const struct path *path,
2989 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002990 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002992 char *res = buf + buflen;
2993 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002995 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002996 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04002997
Al Viro02125a82011-12-05 08:43:34 -05002998 if (error < 0)
2999 return ERR_PTR(error);
3000 if (error > 0)
3001 return NULL;
3002 return res;
3003}
3004
3005char *d_absolute_path(const struct path *path,
3006 char *buf, int buflen)
3007{
3008 struct path root = {};
3009 char *res = buf + buflen;
3010 int error;
3011
3012 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003013 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003014
3015 if (error > 1)
3016 error = -EINVAL;
3017 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003018 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003019 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020}
3021
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003022/*
3023 * same as __d_path but appends "(deleted)" for unlinked files.
3024 */
Al Viro02125a82011-12-05 08:43:34 -05003025static int path_with_deleted(const struct path *path,
3026 const struct path *root,
3027 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003028{
3029 prepend(buf, buflen, "\0", 1);
3030 if (d_unlinked(path->dentry)) {
3031 int error = prepend(buf, buflen, " (deleted)", 10);
3032 if (error)
3033 return error;
3034 }
3035
3036 return prepend_path(path, root, buf, buflen);
3037}
3038
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003039static int prepend_unreachable(char **buffer, int *buflen)
3040{
3041 return prepend(buffer, buflen, "(unreachable)", 13);
3042}
3043
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003044static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3045{
3046 unsigned seq;
3047
3048 do {
3049 seq = read_seqcount_begin(&fs->seq);
3050 *root = fs->root;
3051 } while (read_seqcount_retry(&fs->seq, seq));
3052}
3053
Jan Bluncka03a8a702008-02-14 19:38:32 -08003054/**
3055 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003056 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003057 * @buf: buffer to return value in
3058 * @buflen: buffer length
3059 *
3060 * Convert a dentry into an ASCII path name. If the entry has been deleted
3061 * the string " (deleted)" is appended. Note that this is ambiguous.
3062 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003063 * Returns a pointer into the buffer or an error code if the path was
3064 * too long. Note: Callers should use the returned pointer, not the passed
3065 * in buffer, to use the name! The implementation often starts at an offset
3066 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003067 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003068 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003069 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003070char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003072 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003073 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003074 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003076 /*
3077 * We have various synthetic filesystems that never get mounted. On
3078 * these filesystems dentries are never used for lookup purposes, and
3079 * thus don't need to be hashed. They also don't need a name until a
3080 * user wants to identify the object in /proc/pid/fd/. The little hack
3081 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003082 *
3083 * Some pseudo inodes are mountable. When they are mounted
3084 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3085 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003086 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003087 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3088 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003089 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003090
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003091 rcu_read_lock();
3092 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003093 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003094 rcu_read_unlock();
3095
Al Viro02125a82011-12-05 08:43:34 -05003096 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003097 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 return res;
3099}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003100EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101
3102/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003103 * Helper function for dentry_operations.d_dname() members
3104 */
3105char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3106 const char *fmt, ...)
3107{
3108 va_list args;
3109 char temp[64];
3110 int sz;
3111
3112 va_start(args, fmt);
3113 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3114 va_end(args);
3115
3116 if (sz > sizeof(temp) || sz > buflen)
3117 return ERR_PTR(-ENAMETOOLONG);
3118
3119 buffer += buflen - sz;
3120 return memcpy(buffer, temp, sz);
3121}
3122
Al Viro118b2302013-08-24 12:08:17 -04003123char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3124{
3125 char *end = buffer + buflen;
3126 /* these dentries are never renamed, so d_lock is not needed */
3127 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003128 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003129 prepend(&end, &buflen, "/", 1))
3130 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003131 return end;
Al Viro118b2302013-08-24 12:08:17 -04003132}
David Herrmann31bbe162014-01-03 14:09:47 +01003133EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003134
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003135/*
Ram Pai6092d042008-03-27 13:06:20 +01003136 * Write full pathname from the root of the filesystem into the buffer.
3137 */
Al Virof6500802014-01-26 12:37:55 -05003138static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003139{
Al Virof6500802014-01-26 12:37:55 -05003140 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003141 char *end, *retval;
3142 int len, seq = 0;
3143 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003144
Al Virof6500802014-01-26 12:37:55 -05003145 if (buflen < 2)
3146 goto Elong;
3147
Al Viro48f5ec22013-09-09 15:22:25 -04003148 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003149restart:
Al Virof6500802014-01-26 12:37:55 -05003150 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003151 end = buf + buflen;
3152 len = buflen;
3153 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003154 /* Get '/' right */
3155 retval = end-1;
3156 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003157 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003158 while (!IS_ROOT(dentry)) {
3159 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003160
Ram Pai6092d042008-03-27 13:06:20 +01003161 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003162 error = prepend_name(&end, &len, &dentry->d_name);
3163 if (error)
3164 break;
Ram Pai6092d042008-03-27 13:06:20 +01003165
3166 retval = end;
3167 dentry = parent;
3168 }
Al Viro48f5ec22013-09-09 15:22:25 -04003169 if (!(seq & 1))
3170 rcu_read_unlock();
3171 if (need_seqretry(&rename_lock, seq)) {
3172 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003173 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003174 }
3175 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003176 if (error)
3177 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003178 return retval;
3179Elong:
3180 return ERR_PTR(-ENAMETOOLONG);
3181}
Nick Pigginec2447c2011-01-07 17:49:29 +11003182
3183char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3184{
Waiman Long232d2d62013-09-09 12:18:13 -04003185 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003186}
3187EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003188
3189char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3190{
3191 char *p = NULL;
3192 char *retval;
3193
Al Viroc1031352010-06-06 22:31:14 -04003194 if (d_unlinked(dentry)) {
3195 p = buf + buflen;
3196 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3197 goto Elong;
3198 buflen++;
3199 }
3200 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003201 if (!IS_ERR(retval) && p)
3202 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003203 return retval;
3204Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003205 return ERR_PTR(-ENAMETOOLONG);
3206}
3207
Linus Torvalds8b19e342013-09-12 10:35:47 -07003208static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3209 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003210{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003211 unsigned seq;
3212
3213 do {
3214 seq = read_seqcount_begin(&fs->seq);
3215 *root = fs->root;
3216 *pwd = fs->pwd;
3217 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003218}
3219
Ram Pai6092d042008-03-27 13:06:20 +01003220/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221 * NOTE! The user-level library version returns a
3222 * character pointer. The kernel system call just
3223 * returns the length of the buffer filled (which
3224 * includes the ending '\0' character), or a negative
3225 * error value. So libc would do something like
3226 *
3227 * char *getcwd(char * buf, size_t size)
3228 * {
3229 * int retval;
3230 *
3231 * retval = sys_getcwd(buf, size);
3232 * if (retval >= 0)
3233 * return buf;
3234 * errno = -retval;
3235 * return NULL;
3236 * }
3237 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003238SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239{
Linus Torvalds552ce542007-02-13 12:08:18 -08003240 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003241 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003242 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243
3244 if (!page)
3245 return -ENOMEM;
3246
Linus Torvalds8b19e342013-09-12 10:35:47 -07003247 rcu_read_lock();
3248 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249
Linus Torvalds552ce542007-02-13 12:08:18 -08003250 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003251 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003252 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003253 char *cwd = page + PATH_MAX;
3254 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003256 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003257 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003258 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003259
Al Viro02125a82011-12-05 08:43:34 -05003260 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003261 goto out;
3262
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003263 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003264 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003265 error = prepend_unreachable(&cwd, &buflen);
3266 if (error)
3267 goto out;
3268 }
3269
Linus Torvalds552ce542007-02-13 12:08:18 -08003270 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003271 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003272 if (len <= size) {
3273 error = len;
3274 if (copy_to_user(buf, cwd, len))
3275 error = -EFAULT;
3276 }
Nick Piggin949854d2011-01-07 17:49:37 +11003277 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003278 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003279 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280
3281out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003282 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283 return error;
3284}
3285
3286/*
3287 * Test whether new_dentry is a subdirectory of old_dentry.
3288 *
3289 * Trivially implemented using the dcache structure
3290 */
3291
3292/**
3293 * is_subdir - is new dentry a subdirectory of old_dentry
3294 * @new_dentry: new dentry
3295 * @old_dentry: old dentry
3296 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003297 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3298 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3300 */
3301
Yaowei Baia6e57872015-11-17 14:40:11 +08003302bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303{
Yaowei Baia6e57872015-11-17 14:40:11 +08003304 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003305 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003307 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003308 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003309
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003310 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003313 /*
3314 * Need rcu_readlock to protect against the d_parent trashing
3315 * due to d_move
3316 */
3317 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003318 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003319 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003320 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003321 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003322 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003323 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324
3325 return result;
3326}
3327
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003328static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003329{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003330 struct dentry *root = data;
3331 if (dentry != root) {
3332 if (d_unhashed(dentry) || !dentry->d_inode)
3333 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003335 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3336 dentry->d_flags |= DCACHE_GENOCIDE;
3337 dentry->d_lockref.count--;
3338 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003340 return D_WALK_CONTINUE;
3341}
Nick Piggin58db63d2011-01-07 17:49:39 +11003342
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003343void d_genocide(struct dentry *parent)
3344{
3345 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346}
3347
Al Viro60545d02013-06-07 01:20:27 -04003348void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349{
Al Viro60545d02013-06-07 01:20:27 -04003350 inode_dec_link_count(inode);
3351 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003352 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003353 !d_unlinked(dentry));
3354 spin_lock(&dentry->d_parent->d_lock);
3355 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3356 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3357 (unsigned long long)inode->i_ino);
3358 spin_unlock(&dentry->d_lock);
3359 spin_unlock(&dentry->d_parent->d_lock);
3360 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003361}
Al Viro60545d02013-06-07 01:20:27 -04003362EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363
3364static __initdata unsigned long dhash_entries;
3365static int __init set_dhash_entries(char *str)
3366{
3367 if (!str)
3368 return 0;
3369 dhash_entries = simple_strtoul(str, &str, 0);
3370 return 1;
3371}
3372__setup("dhash_entries=", set_dhash_entries);
3373
3374static void __init dcache_init_early(void)
3375{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003376 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377
3378 /* If hashes are distributed across NUMA nodes, defer
3379 * hash allocation until vmalloc space is available.
3380 */
3381 if (hashdist)
3382 return;
3383
3384 dentry_hashtable =
3385 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003386 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 dhash_entries,
3388 13,
3389 HASH_EARLY,
3390 &d_hash_shift,
3391 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003392 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393 0);
3394
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003395 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003396 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397}
3398
Denis Cheng74bf17c2007-10-16 23:26:30 -07003399static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003401 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402
3403 /*
3404 * A constructor could be added for stable state like the lists,
3405 * but it is probably not worth it because of the cache nature
3406 * of the dcache.
3407 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003408 dentry_cache = KMEM_CACHE(dentry,
Vladimir Davydov5d097052016-01-14 15:18:21 -08003409 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410
3411 /* Hash may have been set up in dcache_init_early */
3412 if (!hashdist)
3413 return;
3414
3415 dentry_hashtable =
3416 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003417 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 dhash_entries,
3419 13,
3420 0,
3421 &d_hash_shift,
3422 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003423 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 0);
3425
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003426 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003427 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003428}
3429
3430/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003431struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003432EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433
Linus Torvalds1da177e2005-04-16 15:20:36 -07003434EXPORT_SYMBOL(d_genocide);
3435
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436void __init vfs_caches_init_early(void)
3437{
3438 dcache_init_early();
3439 inode_init_early();
3440}
3441
Mel Gorman4248b0d2015-08-06 15:46:20 -07003442void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003445 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446
Denis Cheng74bf17c2007-10-16 23:26:30 -07003447 dcache_init();
3448 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003449 files_init();
3450 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003451 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452 bdev_cache_init();
3453 chrdev_init();
3454}