blob: 7a3f3e5f9ceabfc4cad41a5d5478258f7e817dc1 [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 +0000272/*
273 * Make sure other CPUs see the inode attached before the type is set.
274 */
275static inline void __d_set_inode_and_type(struct dentry *dentry,
276 struct inode *inode,
277 unsigned type_flags)
278{
279 unsigned flags;
280
281 dentry->d_inode = inode;
282 smp_wmb();
283 flags = READ_ONCE(dentry->d_flags);
284 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
285 flags |= type_flags;
286 WRITE_ONCE(dentry->d_flags, flags);
287}
288
289/*
290 * Ideally, we want to make sure that other CPUs see the flags cleared before
291 * the inode is detached, but this is really a violation of RCU principles
292 * since the ordering suggests we should always set inode before flags.
293 *
294 * We should instead replace or discard the entire dentry - but that sucks
295 * performancewise on mass deletion/rename.
296 */
297static inline void __d_clear_type_and_inode(struct dentry *dentry)
298{
299 unsigned flags = READ_ONCE(dentry->d_flags);
300
301 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
302 WRITE_ONCE(dentry->d_flags, flags);
303 smp_wmb();
304 dentry->d_inode = NULL;
305}
306
Al Virob4f03542014-04-29 23:40:14 -0400307static void dentry_free(struct dentry *dentry)
308{
Al Viro946e51f2014-10-26 19:19:16 -0400309 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400310 if (unlikely(dname_external(dentry))) {
311 struct external_name *p = external_name(dentry);
312 if (likely(atomic_dec_and_test(&p->u.count))) {
313 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
314 return;
315 }
316 }
Al Virob4f03542014-04-29 23:40:14 -0400317 /* if dentry was never visible to RCU, immediate free is OK */
318 if (!(dentry->d_flags & DCACHE_RCUACCESS))
319 __d_free(&dentry->d_u.d_rcu);
320 else
321 call_rcu(&dentry->d_u.d_rcu, __d_free);
322}
323
Nick Piggin31e6b012011-01-07 17:49:52 +1100324/**
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200325 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
Randy Dunlapff5fdb62011-01-22 20:16:06 -0800326 * @dentry: the target dentry
Nick Piggin31e6b012011-01-07 17:49:52 +1100327 * After this call, in-progress rcu-walk path lookup will fail. This
328 * should be called after unhashing, and after changing d_inode (if
329 * the dentry has not already been unhashed).
330 */
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200331static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
Nick Piggin31e6b012011-01-07 17:49:52 +1100332{
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200333 lockdep_assert_held(&dentry->d_lock);
334 /* Go through am invalidation barrier */
335 write_seqcount_invalidate(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100336}
337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338/*
339 * Release the dentry's inode, using the filesystem
Nick Piggin31e6b012011-01-07 17:49:52 +1100340 * d_iput() operation if defined. Dentry has no refcount
341 * and is unhashed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800343static void dentry_iput(struct dentry * dentry)
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +0200344 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100345 __releases(dentry->d_inode->i_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
347 struct inode *inode = dentry->d_inode;
348 if (inode) {
David Howells4bf46a22015-03-05 14:09:22 +0000349 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400350 hlist_del_init(&dentry->d_u.d_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100352 spin_unlock(&inode->i_lock);
Linus Torvaldsf805fbd2005-09-19 19:54:29 -0700353 if (!inode->i_nlink)
354 fsnotify_inoderemove(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 if (dentry->d_op && dentry->d_op->d_iput)
356 dentry->d_op->d_iput(dentry, inode);
357 else
358 iput(inode);
359 } else {
360 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 }
362}
363
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700364/*
Nick Piggin31e6b012011-01-07 17:49:52 +1100365 * Release the dentry's inode, using the filesystem
366 * d_iput() operation if defined. dentry remains in-use.
367 */
368static void dentry_unlink_inode(struct dentry * dentry)
369 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100370 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100371{
372 struct inode *inode = dentry->d_inode;
David Howells4bf46a22015-03-05 14:09:22 +0000373 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400374 hlist_del_init(&dentry->d_u.d_alias);
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200375 dentry_rcuwalk_invalidate(dentry);
Nick Piggin31e6b012011-01-07 17:49:52 +1100376 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100377 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100378 if (!inode->i_nlink)
379 fsnotify_inoderemove(inode);
380 if (dentry->d_op && dentry->d_op->d_iput)
381 dentry->d_op->d_iput(dentry, inode);
382 else
383 iput(inode);
384}
385
386/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400387 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
388 * is in use - which includes both the "real" per-superblock
389 * LRU list _and_ the DCACHE_SHRINK_LIST use.
390 *
391 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
392 * on the shrink list (ie not on the superblock LRU list).
393 *
394 * The per-cpu "nr_dentry_unused" counters are updated with
395 * the DCACHE_LRU_LIST bit.
396 *
397 * These helper functions make sure we always follow the
398 * rules. d_lock must be held by the caller.
399 */
400#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
401static void d_lru_add(struct dentry *dentry)
402{
403 D_FLAG_VERIFY(dentry, 0);
404 dentry->d_flags |= DCACHE_LRU_LIST;
405 this_cpu_inc(nr_dentry_unused);
406 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407}
408
409static void d_lru_del(struct dentry *dentry)
410{
411 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
412 dentry->d_flags &= ~DCACHE_LRU_LIST;
413 this_cpu_dec(nr_dentry_unused);
414 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
415}
416
417static void d_shrink_del(struct dentry *dentry)
418{
419 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
420 list_del_init(&dentry->d_lru);
421 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
422 this_cpu_dec(nr_dentry_unused);
423}
424
425static void d_shrink_add(struct dentry *dentry, struct list_head *list)
426{
427 D_FLAG_VERIFY(dentry, 0);
428 list_add(&dentry->d_lru, list);
429 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
430 this_cpu_inc(nr_dentry_unused);
431}
432
433/*
434 * These can only be called under the global LRU lock, ie during the
435 * callback for freeing the LRU list. "isolate" removes it from the
436 * LRU lists entirely, while shrink_move moves it to the indicated
437 * private list.
438 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800439static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400440{
441 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
442 dentry->d_flags &= ~DCACHE_LRU_LIST;
443 this_cpu_dec(nr_dentry_unused);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800444 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400445}
446
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800447static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
448 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400449{
450 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
451 dentry->d_flags |= DCACHE_SHRINK_LIST;
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800452 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400453}
454
455/*
Dave Chinnerf6041562013-08-28 10:18:00 +1000456 * dentry_lru_(add|del)_list) must be called with d_lock held.
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700457 */
458static void dentry_lru_add(struct dentry *dentry)
459{
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400460 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
461 d_lru_add(dentry);
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700462}
463
Miklos Szeredid52b9082007-05-08 00:23:46 -0700464/**
Nick Piggin789680d2011-01-07 17:49:30 +1100465 * d_drop - drop a dentry
466 * @dentry: dentry to drop
467 *
468 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
469 * be found through a VFS lookup any more. Note that this is different from
470 * deleting the dentry - d_delete will try to mark the dentry negative if
471 * possible, giving a successful _negative_ lookup, while d_drop will
472 * just make the cache lookup fail.
473 *
474 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
475 * reason (NFS timeouts or autofs deletes).
476 *
477 * __d_drop requires dentry->d_lock.
478 */
479void __d_drop(struct dentry *dentry)
480{
Linus Torvaldsdea36672011-04-24 07:58:46 -0700481 if (!d_unhashed(dentry)) {
Al Virob61625d2013-10-04 11:09:01 -0400482 struct hlist_bl_head *b;
J. Bruce Fields7632e462012-06-28 12:10:55 -0400483 /*
484 * Hashed dentries are normally on the dentry hashtable,
485 * with the exception of those newly allocated by
486 * d_obtain_alias, which are always IS_ROOT:
487 */
488 if (unlikely(IS_ROOT(dentry)))
Al Virob61625d2013-10-04 11:09:01 -0400489 b = &dentry->d_sb->s_anon;
490 else
491 b = d_hash(dentry->d_parent, dentry->d_name.hash);
492
493 hlist_bl_lock(b);
494 __hlist_bl_del(&dentry->d_hash);
495 dentry->d_hash.pprev = NULL;
496 hlist_bl_unlock(b);
Peter Zijlstraa7c6f572015-06-11 14:46:46 +0200497 dentry_rcuwalk_invalidate(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +1100498 }
499}
500EXPORT_SYMBOL(__d_drop);
501
502void d_drop(struct dentry *dentry)
503{
Nick Piggin789680d2011-01-07 17:49:30 +1100504 spin_lock(&dentry->d_lock);
505 __d_drop(dentry);
506 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100507}
508EXPORT_SYMBOL(d_drop);
509
Al Viroe55fd012014-05-28 13:51:12 -0400510static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100511{
Al Viro41edf272014-05-01 10:30:00 -0400512 struct dentry *parent = NULL;
513 bool can_free = true;
Al Viro41edf272014-05-01 10:30:00 -0400514 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100515 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100516
Linus Torvalds0d984392013-09-08 13:46:52 -0700517 /*
518 * The dentry is now unrecoverably dead to the world.
519 */
520 lockref_mark_dead(&dentry->d_lockref);
521
Sage Weilf0023bc2011-10-28 10:02:42 -0700522 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700523 * inform the fs via d_prune that this dentry is about to be
524 * unhashed and destroyed.
525 */
Al Viro29266202014-05-30 11:39:02 -0400526 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800527 dentry->d_op->d_prune(dentry);
528
Al Viro01b60352014-04-29 23:42:52 -0400529 if (dentry->d_flags & DCACHE_LRU_LIST) {
530 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
531 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400532 }
Nick Piggin77812a12011-01-07 17:49:48 +1100533 /* if it was on the hash then remove it */
534 __d_drop(dentry);
Al Viroca5358e2014-10-26 19:31:10 -0400535 __list_del_entry(&dentry->d_child);
Al Viro03b3b882014-04-29 15:45:28 -0400536 /*
537 * Inform d_walk() that we are no longer attached to the
538 * dentry tree
539 */
540 dentry->d_flags |= DCACHE_DENTRY_KILLED;
541 if (parent)
542 spin_unlock(&parent->d_lock);
543 dentry_iput(dentry);
544 /*
545 * dentry_iput drops the locks, at which point nobody (except
546 * transient RCU lookups) can reach this dentry.
547 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800548 BUG_ON(dentry->d_lockref.count > 0);
Al Viro03b3b882014-04-29 15:45:28 -0400549 this_cpu_dec(nr_dentry);
550 if (dentry->d_op && dentry->d_op->d_release)
551 dentry->d_op->d_release(dentry);
552
Al Viro41edf272014-05-01 10:30:00 -0400553 spin_lock(&dentry->d_lock);
554 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
555 dentry->d_flags |= DCACHE_MAY_FREE;
556 can_free = false;
557 }
558 spin_unlock(&dentry->d_lock);
Al Viro41edf272014-05-01 10:30:00 -0400559 if (likely(can_free))
560 dentry_free(dentry);
Al Viroe55fd012014-05-28 13:51:12 -0400561}
562
563/*
564 * Finish off a dentry we've decided to kill.
565 * dentry->d_lock must be held, returns with it unlocked.
566 * If ref is non-zero, then decrement the refcount too.
567 * Returns dentry requiring refcount drop, or NULL if we're done.
568 */
Al Viro8cbf74d2014-05-29 09:18:26 -0400569static struct dentry *dentry_kill(struct dentry *dentry)
Al Viroe55fd012014-05-28 13:51:12 -0400570 __releases(dentry->d_lock)
571{
572 struct inode *inode = dentry->d_inode;
573 struct dentry *parent = NULL;
574
575 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
576 goto failed;
577
578 if (!IS_ROOT(dentry)) {
579 parent = dentry->d_parent;
580 if (unlikely(!spin_trylock(&parent->d_lock))) {
581 if (inode)
582 spin_unlock(&inode->i_lock);
583 goto failed;
584 }
585 }
586
587 __dentry_kill(dentry);
Al Viro03b3b882014-04-29 15:45:28 -0400588 return parent;
Al Viroe55fd012014-05-28 13:51:12 -0400589
590failed:
Al Viro8cbf74d2014-05-29 09:18:26 -0400591 spin_unlock(&dentry->d_lock);
592 cpu_relax();
Al Viroe55fd012014-05-28 13:51:12 -0400593 return dentry; /* try again with same dentry */
Nick Piggin77812a12011-01-07 17:49:48 +1100594}
595
Al Viro046b9612014-05-29 08:54:52 -0400596static inline struct dentry *lock_parent(struct dentry *dentry)
597{
598 struct dentry *parent = dentry->d_parent;
599 if (IS_ROOT(dentry))
600 return NULL;
Linus Torvalds360f5472015-01-09 15:19:03 -0800601 if (unlikely(dentry->d_lockref.count < 0))
Al Viroc2338f22014-06-12 00:29:13 -0400602 return NULL;
Al Viro046b9612014-05-29 08:54:52 -0400603 if (likely(spin_trylock(&parent->d_lock)))
604 return parent;
Al Viro046b9612014-05-29 08:54:52 -0400605 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400606 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400607again:
608 parent = ACCESS_ONCE(dentry->d_parent);
609 spin_lock(&parent->d_lock);
610 /*
611 * We can't blindly lock dentry until we are sure
612 * that we won't violate the locking order.
613 * Any changes of dentry->d_parent must have
614 * been done with parent->d_lock held, so
615 * spin_lock() above is enough of a barrier
616 * for checking if it's still our child.
617 */
618 if (unlikely(parent != dentry->d_parent)) {
619 spin_unlock(&parent->d_lock);
620 goto again;
621 }
622 rcu_read_unlock();
623 if (parent != dentry)
Linus Torvalds9f126002014-05-31 09:13:21 -0700624 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Al Viro046b9612014-05-29 08:54:52 -0400625 else
626 parent = NULL;
627 return parent;
628}
629
Linus Torvalds360f5472015-01-09 15:19:03 -0800630/*
631 * Try to do a lockless dput(), and return whether that was successful.
632 *
633 * If unsuccessful, we return false, having already taken the dentry lock.
634 *
635 * The caller needs to hold the RCU read lock, so that the dentry is
636 * guaranteed to stay around even if the refcount goes down to zero!
637 */
638static inline bool fast_dput(struct dentry *dentry)
639{
640 int ret;
641 unsigned int d_flags;
642
643 /*
644 * If we have a d_op->d_delete() operation, we sould not
645 * let the dentry count go to zero, so use "put__or_lock".
646 */
647 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
648 return lockref_put_or_lock(&dentry->d_lockref);
649
650 /*
651 * .. otherwise, we can try to just decrement the
652 * lockref optimistically.
653 */
654 ret = lockref_put_return(&dentry->d_lockref);
655
656 /*
657 * If the lockref_put_return() failed due to the lock being held
658 * by somebody else, the fast path has failed. We will need to
659 * get the lock, and then check the count again.
660 */
661 if (unlikely(ret < 0)) {
662 spin_lock(&dentry->d_lock);
663 if (dentry->d_lockref.count > 1) {
664 dentry->d_lockref.count--;
665 spin_unlock(&dentry->d_lock);
666 return 1;
667 }
668 return 0;
669 }
670
671 /*
672 * If we weren't the last ref, we're done.
673 */
674 if (ret)
675 return 1;
676
677 /*
678 * Careful, careful. The reference count went down
679 * to zero, but we don't hold the dentry lock, so
680 * somebody else could get it again, and do another
681 * dput(), and we need to not race with that.
682 *
683 * However, there is a very special and common case
684 * where we don't care, because there is nothing to
685 * do: the dentry is still hashed, it does not have
686 * a 'delete' op, and it's referenced and already on
687 * the LRU list.
688 *
689 * NOTE! Since we aren't locked, these values are
690 * not "stable". However, it is sufficient that at
691 * some point after we dropped the reference the
692 * dentry was hashed and the flags had the proper
693 * value. Other dentry users may have re-gotten
694 * a reference to the dentry and change that, but
695 * our work is done - we can leave the dentry
696 * around with a zero refcount.
697 */
698 smp_rmb();
699 d_flags = ACCESS_ONCE(dentry->d_flags);
700 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST;
701
702 /* Nothing to do? Dropping the reference was all we needed? */
703 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
704 return 1;
705
706 /*
707 * Not the fast normal case? Get the lock. We've already decremented
708 * the refcount, but we'll need to re-check the situation after
709 * getting the lock.
710 */
711 spin_lock(&dentry->d_lock);
712
713 /*
714 * Did somebody else grab a reference to it in the meantime, and
715 * we're no longer the last user after all? Alternatively, somebody
716 * else could have killed it and marked it dead. Either way, we
717 * don't need to do anything else.
718 */
719 if (dentry->d_lockref.count) {
720 spin_unlock(&dentry->d_lock);
721 return 1;
722 }
723
724 /*
725 * Re-get the reference we optimistically dropped. We hold the
726 * lock, and we just tested that it was zero, so we can just
727 * set it to 1.
728 */
729 dentry->d_lockref.count = 1;
730 return 0;
731}
732
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734/*
735 * This is dput
736 *
737 * This is complicated by the fact that we do not want to put
738 * dentries that are no longer on any hash chain on the unused
739 * list: we'd much rather just get rid of them immediately.
740 *
741 * However, that implies that we have to traverse the dentry
742 * tree upwards to the parents which might _also_ now be
743 * scheduled for deletion (it may have been only waiting for
744 * its last child to go away).
745 *
746 * This tail recursion is done by hand as we don't want to depend
747 * on the compiler to always get this right (gcc generally doesn't).
748 * Real recursion would eat up our stack space.
749 */
750
751/*
752 * dput - release a dentry
753 * @dentry: dentry to release
754 *
755 * Release a dentry. This will drop the usage count and if appropriate
756 * call the dentry unlink method as well as removing it from the queues and
757 * releasing its resources. If the parent dentries were scheduled for release
758 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760void dput(struct dentry *dentry)
761{
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700762 if (unlikely(!dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return;
764
765repeat:
Linus Torvalds360f5472015-01-09 15:19:03 -0800766 rcu_read_lock();
767 if (likely(fast_dput(dentry))) {
768 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return;
Linus Torvalds360f5472015-01-09 15:19:03 -0800770 }
771
772 /* Slow case: now with the dentry lock held */
773 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700775 /* Unreachable? Get rid of it */
776 if (unlikely(d_unhashed(dentry)))
777 goto kill_it;
778
779 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100781 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 }
Nick Piggin265ac902010-10-10 05:36:24 -0400783
Linus Torvalds358eec12013-10-31 15:43:02 -0700784 if (!(dentry->d_flags & DCACHE_REFERENCED))
785 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400786 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400787
Waiman Long98474232013-08-28 18:24:59 -0700788 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100789 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 return;
791
Miklos Szeredid52b9082007-05-08 00:23:46 -0700792kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400793 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700794 if (dentry)
795 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700797EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100800/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100801static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802{
Waiman Long98474232013-08-28 18:24:59 -0700803 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804}
805
Nick Piggindc0474b2011-01-07 17:49:43 +1100806static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100807{
Waiman Long98474232013-08-28 18:24:59 -0700808 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100809}
810
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100811struct dentry *dget_parent(struct dentry *dentry)
812{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700813 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100814 struct dentry *ret;
815
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700816 /*
817 * Do optimistic parent lookup without any
818 * locking.
819 */
820 rcu_read_lock();
821 ret = ACCESS_ONCE(dentry->d_parent);
822 gotref = lockref_get_not_zero(&ret->d_lockref);
823 rcu_read_unlock();
824 if (likely(gotref)) {
825 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
826 return ret;
827 dput(ret);
828 }
829
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100830repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100831 /*
832 * Don't need rcu_dereference because we re-check it was correct under
833 * the lock.
834 */
835 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100836 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100837 spin_lock(&ret->d_lock);
838 if (unlikely(ret != dentry->d_parent)) {
839 spin_unlock(&ret->d_lock);
840 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100841 goto repeat;
842 }
Nick Piggina734eb42011-01-07 17:49:44 +1100843 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700844 BUG_ON(!ret->d_lockref.count);
845 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100846 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100847 return ret;
848}
849EXPORT_SYMBOL(dget_parent);
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851/**
852 * d_find_alias - grab a hashed alias of inode
853 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 *
855 * If inode has a hashed alias, or is a directory and has any alias,
856 * acquire the reference to alias and return it. Otherwise return NULL.
857 * Notice that if inode is a directory there can be only one alias and
858 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800859 * of a filesystem, or if the directory was renamed and d_revalidate
860 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700862 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500863 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500865static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866{
Nick Pigginda502952011-01-07 17:49:33 +1100867 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
Nick Pigginda502952011-01-07 17:49:33 +1100869again:
870 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400871 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100872 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700874 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100875 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500877 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100878 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100879 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return alias;
881 }
882 }
Nick Pigginda502952011-01-07 17:49:33 +1100883 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 }
Nick Pigginda502952011-01-07 17:49:33 +1100885 if (discon_alias) {
886 alias = discon_alias;
887 spin_lock(&alias->d_lock);
888 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500889 __dget_dlock(alias);
890 spin_unlock(&alias->d_lock);
891 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100892 }
893 spin_unlock(&alias->d_lock);
894 goto again;
895 }
896 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Nick Pigginda502952011-01-07 17:49:33 +1100899struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
David Howells214fda12006-03-25 03:06:36 -0800901 struct dentry *de = NULL;
902
Al Virob3d9b7a2012-06-09 13:51:19 -0400903 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100904 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500905 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100906 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800907 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return de;
909}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700910EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912/*
913 * Try to kill dentries associated with this inode.
914 * WARNING: you must own a reference to inode.
915 */
916void d_prune_aliases(struct inode *inode)
917{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700918 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100920 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400921 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700923 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400924 struct dentry *parent = lock_parent(dentry);
925 if (likely(!dentry->d_lockref.count)) {
926 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800927 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400928 goto restart;
929 }
930 if (parent)
931 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 }
933 spin_unlock(&dentry->d_lock);
934 }
Nick Piggin873feea2011-01-07 17:50:06 +1100935 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700937EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400939static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Al Viro5c47e6d2014-04-29 16:13:18 -0400941 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700942
Miklos Szeredi60942f22014-05-02 15:38:39 -0400943 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400944 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400945 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100946 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400947 parent = lock_parent(dentry);
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000950 * The dispose list is isolated and dentries are not accounted
951 * to the LRU here, so we can simply remove it from the list
952 * here regardless of whether it is referenced or not.
953 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400954 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000955
956 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000958 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800960 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700961 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400962 if (parent)
963 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 continue;
965 }
Nick Piggin77812a12011-01-07 17:49:48 +1100966
Al Viro64fd72e2014-05-28 09:48:44 -0400967
968 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
969 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
970 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400971 if (parent)
972 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400973 if (can_free)
974 dentry_free(dentry);
975 continue;
976 }
977
Al Viroff2fde92014-05-28 13:59:13 -0400978 inode = dentry->d_inode;
979 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400980 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000981 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400982 if (parent)
983 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400984 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000985 }
Al Viroff2fde92014-05-28 13:59:13 -0400986
Al Viroff2fde92014-05-28 13:59:13 -0400987 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400988
Al Viro5c47e6d2014-04-29 16:13:18 -0400989 /*
990 * We need to prune ancestors too. This is necessary to prevent
991 * quadratic behavior of shrink_dcache_parent(), but is also
992 * expected to be beneficial in reducing dentry cache
993 * fragmentation.
994 */
995 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400996 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
997 parent = lock_parent(dentry);
998 if (dentry->d_lockref.count != 1) {
999 dentry->d_lockref.count--;
1000 spin_unlock(&dentry->d_lock);
1001 if (parent)
1002 spin_unlock(&parent->d_lock);
1003 break;
1004 }
1005 inode = dentry->d_inode; /* can't be NULL */
1006 if (unlikely(!spin_trylock(&inode->i_lock))) {
1007 spin_unlock(&dentry->d_lock);
1008 if (parent)
1009 spin_unlock(&parent->d_lock);
1010 cpu_relax();
1011 continue;
1012 }
1013 __dentry_kill(dentry);
1014 dentry = parent;
1015 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001017}
1018
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001019static enum lru_status dentry_lru_isolate(struct list_head *item,
1020 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001021{
1022 struct list_head *freeable = arg;
1023 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1024
1025
1026 /*
1027 * we are inverting the lru lock/dentry->d_lock here,
1028 * so use a trylock. If we fail to get the lock, just skip
1029 * it
1030 */
1031 if (!spin_trylock(&dentry->d_lock))
1032 return LRU_SKIP;
1033
1034 /*
1035 * Referenced dentries are still in use. If they have active
1036 * counts, just remove them from the LRU. Otherwise give them
1037 * another pass through the LRU.
1038 */
1039 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001040 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001041 spin_unlock(&dentry->d_lock);
1042 return LRU_REMOVED;
1043 }
1044
1045 if (dentry->d_flags & DCACHE_REFERENCED) {
1046 dentry->d_flags &= ~DCACHE_REFERENCED;
1047 spin_unlock(&dentry->d_lock);
1048
1049 /*
1050 * The list move itself will be made by the common LRU code. At
1051 * this point, we've dropped the dentry->d_lock but keep the
1052 * lru lock. This is safe to do, since every list movement is
1053 * protected by the lru lock even if both locks are held.
1054 *
1055 * This is guaranteed by the fact that all LRU management
1056 * functions are intermediated by the LRU API calls like
1057 * list_lru_add and list_lru_del. List movement in this file
1058 * only ever occur through this functions or through callbacks
1059 * like this one, that are called from the LRU API.
1060 *
1061 * The only exceptions to this are functions like
1062 * shrink_dentry_list, and code that first checks for the
1063 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1064 * operating only with stack provided lists after they are
1065 * properly isolated from the main list. It is thus, always a
1066 * local access.
1067 */
1068 return LRU_ROTATE;
1069 }
1070
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001071 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001072 spin_unlock(&dentry->d_lock);
1073
1074 return LRU_REMOVED;
1075}
1076
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001077/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001078 * prune_dcache_sb - shrink the dcache
1079 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001080 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001081 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001082 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1083 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001084 * function.
1085 *
1086 * This function may fail to free any resources if all the dentries are in
1087 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001088 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001089long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001090{
Dave Chinnerf6041562013-08-28 10:18:00 +10001091 LIST_HEAD(dispose);
1092 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001093
Vladimir Davydov503c3582015-02-12 14:58:47 -08001094 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1095 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001096 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001097 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098}
1099
Glauber Costa4e717f52013-08-28 10:18:03 +10001100static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001101 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001102{
Glauber Costa4e717f52013-08-28 10:18:03 +10001103 struct list_head *freeable = arg;
1104 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001105
Glauber Costa4e717f52013-08-28 10:18:03 +10001106 /*
1107 * we are inverting the lru lock/dentry->d_lock here,
1108 * so use a trylock. If we fail to get the lock, just skip
1109 * it
1110 */
1111 if (!spin_trylock(&dentry->d_lock))
1112 return LRU_SKIP;
1113
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001114 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001115 spin_unlock(&dentry->d_lock);
1116
1117 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001118}
1119
Glauber Costa4e717f52013-08-28 10:18:03 +10001120
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001121/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 * shrink_dcache_sb - shrink dcache for a superblock
1123 * @sb: superblock
1124 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001125 * Shrink the dcache for the specified super block. This is used to free
1126 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001128void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
Glauber Costa4e717f52013-08-28 10:18:03 +10001130 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001131
Glauber Costa4e717f52013-08-28 10:18:03 +10001132 do {
1133 LIST_HEAD(dispose);
1134
1135 freed = list_lru_walk(&sb->s_dentry_lru,
1136 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1137
1138 this_cpu_sub(nr_dentry_unused, freed);
1139 shrink_dentry_list(&dispose);
1140 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001142EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001145 * enum d_walk_ret - action to talke during tree walk
1146 * @D_WALK_CONTINUE: contrinue walk
1147 * @D_WALK_QUIT: quit walk
1148 * @D_WALK_NORETRY: quit when retry is needed
1149 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001151enum d_walk_ret {
1152 D_WALK_CONTINUE,
1153 D_WALK_QUIT,
1154 D_WALK_NORETRY,
1155 D_WALK_SKIP,
1156};
1157
1158/**
1159 * d_walk - walk the dentry tree
1160 * @parent: start of walk
1161 * @data: data passed to @enter() and @finish()
1162 * @enter: callback when first entering the dentry
1163 * @finish: callback when successfully finished the walk
1164 *
1165 * The @enter() and @finish() callbacks are called with d_lock held.
1166 */
1167static void d_walk(struct dentry *parent, void *data,
1168 enum d_walk_ret (*enter)(void *, struct dentry *),
1169 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
Nick Piggin949854d2011-01-07 17:49:37 +11001171 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001173 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001174 enum d_walk_ret ret;
1175 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001176
Nick Piggin58db63d2011-01-07 17:49:39 +11001177again:
Al Viro48f5ec22013-09-09 15:22:25 -04001178 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001179 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001180 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001181
1182 ret = enter(data, this_parent);
1183 switch (ret) {
1184 case D_WALK_CONTINUE:
1185 break;
1186 case D_WALK_QUIT:
1187 case D_WALK_SKIP:
1188 goto out_unlock;
1189 case D_WALK_NORETRY:
1190 retry = false;
1191 break;
1192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193repeat:
1194 next = this_parent->d_subdirs.next;
1195resume:
1196 while (next != &this_parent->d_subdirs) {
1197 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001198 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001200
1201 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001202
1203 ret = enter(data, dentry);
1204 switch (ret) {
1205 case D_WALK_CONTINUE:
1206 break;
1207 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001208 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001209 goto out_unlock;
1210 case D_WALK_NORETRY:
1211 retry = false;
1212 break;
1213 case D_WALK_SKIP:
1214 spin_unlock(&dentry->d_lock);
1215 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001216 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001217
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001219 spin_unlock(&this_parent->d_lock);
1220 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001222 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 goto repeat;
1224 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001225 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
1227 /*
1228 * All done at this level ... ascend and resume the search.
1229 */
Al Viroca5358e2014-10-26 19:31:10 -04001230 rcu_read_lock();
1231ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001233 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001234 this_parent = child->d_parent;
1235
Al Viro31dec132013-10-25 17:04:27 -04001236 spin_unlock(&child->d_lock);
1237 spin_lock(&this_parent->d_lock);
1238
Al Viroca5358e2014-10-26 19:31:10 -04001239 /* might go back up the wrong parent if we have had a rename. */
1240 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001241 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001242 /* go into the first sibling still alive */
1243 do {
1244 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001245 if (next == &this_parent->d_subdirs)
1246 goto ascend;
1247 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001248 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001249 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 goto resume;
1251 }
Al Viroca5358e2014-10-26 19:31:10 -04001252 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001253 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001254 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001255 if (finish)
1256 finish(data);
1257
1258out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001259 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001260 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001261 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001262
1263rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001264 spin_unlock(&this_parent->d_lock);
1265 rcu_read_unlock();
1266 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001267 if (!retry)
1268 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001269 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001270 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001272
1273/*
1274 * Search for at least 1 mount point in the dentry's subdirs.
1275 * We descend to the next level whenever the d_subdirs
1276 * list is non-empty and continue searching.
1277 */
1278
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001279static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1280{
1281 int *ret = data;
1282 if (d_mountpoint(dentry)) {
1283 *ret = 1;
1284 return D_WALK_QUIT;
1285 }
1286 return D_WALK_CONTINUE;
1287}
1288
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001289/**
1290 * have_submounts - check for mounts over a dentry
1291 * @parent: dentry to check.
1292 *
1293 * Return true if the parent or its subdirectories contain
1294 * a mount point
1295 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001296int have_submounts(struct dentry *parent)
1297{
1298 int ret = 0;
1299
1300 d_walk(parent, &ret, check_mount, NULL);
1301
1302 return ret;
1303}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001304EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001307 * Called by mount code to set a mountpoint and check if the mountpoint is
1308 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1309 * subtree can become unreachable).
1310 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001311 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001312 * this reason take rename_lock and d_lock on dentry and ancestors.
1313 */
1314int d_set_mounted(struct dentry *dentry)
1315{
1316 struct dentry *p;
1317 int ret = -ENOENT;
1318 write_seqlock(&rename_lock);
1319 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001320 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001321 spin_lock(&p->d_lock);
1322 if (unlikely(d_unhashed(p))) {
1323 spin_unlock(&p->d_lock);
1324 goto out;
1325 }
1326 spin_unlock(&p->d_lock);
1327 }
1328 spin_lock(&dentry->d_lock);
1329 if (!d_unlinked(dentry)) {
1330 dentry->d_flags |= DCACHE_MOUNTED;
1331 ret = 0;
1332 }
1333 spin_unlock(&dentry->d_lock);
1334out:
1335 write_sequnlock(&rename_lock);
1336 return ret;
1337}
1338
1339/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001340 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 * and move any unused dentries to the end of the unused
1342 * list for prune_dcache(). We descend to the next level
1343 * whenever the d_subdirs list is non-empty and continue
1344 * searching.
1345 *
1346 * It returns zero iff there are no unused children,
1347 * otherwise it returns the number of children moved to
1348 * the end of the unused list. This may not be the total
1349 * number of unused children, because select_parent can
1350 * drop the lock and return early due to latency
1351 * constraints.
1352 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001353
1354struct select_data {
1355 struct dentry *start;
1356 struct list_head dispose;
1357 int found;
1358};
1359
1360static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001362 struct select_data *data = _data;
1363 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001365 if (data->start == dentry)
1366 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Al Virofe915222014-05-03 00:02:25 -04001368 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001369 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001370 } else {
1371 if (dentry->d_flags & DCACHE_LRU_LIST)
1372 d_lru_del(dentry);
1373 if (!dentry->d_lockref.count) {
1374 d_shrink_add(dentry, &data->dispose);
1375 data->found++;
1376 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 }
1378 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001379 * We can return to the caller if we have found some (this
1380 * ensures forward progress). We'll be coming back to find
1381 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 */
Al Virofe915222014-05-03 00:02:25 -04001383 if (!list_empty(&data->dispose))
1384 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001386 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
1389/**
1390 * shrink_dcache_parent - prune dcache
1391 * @parent: parent of entries to prune
1392 *
1393 * Prune the dcache to remove unused children of the parent dentry.
1394 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001395void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001397 for (;;) {
1398 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001400 INIT_LIST_HEAD(&data.dispose);
1401 data.start = parent;
1402 data.found = 0;
1403
1404 d_walk(parent, &data, select_collect, NULL);
1405 if (!data.found)
1406 break;
1407
1408 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001409 cond_resched();
1410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001412EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
Al Viro9c8c10e2014-05-02 20:36:10 -04001414static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001415{
Al Viro9c8c10e2014-05-02 20:36:10 -04001416 /* it has busy descendents; complain about those instead */
1417 if (!list_empty(&dentry->d_subdirs))
1418 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001419
Al Viro9c8c10e2014-05-02 20:36:10 -04001420 /* root with refcount 1 is fine */
1421 if (dentry == _data && dentry->d_lockref.count == 1)
1422 return D_WALK_CONTINUE;
1423
1424 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1425 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001426 dentry,
1427 dentry->d_inode ?
1428 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001429 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001430 dentry->d_lockref.count,
1431 dentry->d_sb->s_type->name,
1432 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001433 WARN_ON(1);
1434 return D_WALK_CONTINUE;
1435}
1436
1437static void do_one_tree(struct dentry *dentry)
1438{
1439 shrink_dcache_parent(dentry);
1440 d_walk(dentry, dentry, umount_check, NULL);
1441 d_drop(dentry);
1442 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001443}
1444
1445/*
1446 * destroy the dentries attached to a superblock on unmounting
1447 */
1448void shrink_dcache_for_umount(struct super_block *sb)
1449{
1450 struct dentry *dentry;
1451
Al Viro9c8c10e2014-05-02 20:36:10 -04001452 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001453
1454 dentry = sb->s_root;
1455 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001456 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001457
1458 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001459 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1460 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001461 }
1462}
1463
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001464struct detach_data {
1465 struct select_data select;
1466 struct dentry *mountpoint;
1467};
1468static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001469{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001470 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001471
1472 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001473 __dget_dlock(dentry);
1474 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001475 return D_WALK_QUIT;
1476 }
1477
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001478 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001479}
1480
1481static void check_and_drop(void *_data)
1482{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001483 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001484
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001485 if (!data->mountpoint && !data->select.found)
1486 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001487}
1488
1489/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001490 * d_invalidate - detach submounts, prune dcache, and drop
1491 * @dentry: dentry to invalidate (aka detach, prune and drop)
1492 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001493 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001494 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001495 * The final d_drop is done as an atomic operation relative to
1496 * rename_lock ensuring there are no races with d_set_mounted. This
1497 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001498 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001499void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001500{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001501 /*
1502 * If it's already been dropped, return OK.
1503 */
1504 spin_lock(&dentry->d_lock);
1505 if (d_unhashed(dentry)) {
1506 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001507 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001508 }
1509 spin_unlock(&dentry->d_lock);
1510
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001511 /* Negative dentries can be dropped without further checks */
1512 if (!dentry->d_inode) {
1513 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001514 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001515 }
1516
1517 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001518 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001519
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001520 data.mountpoint = NULL;
1521 INIT_LIST_HEAD(&data.select.dispose);
1522 data.select.start = dentry;
1523 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001524
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001525 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001526
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001527 if (data.select.found)
1528 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001529
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001530 if (data.mountpoint) {
1531 detach_mounts(data.mountpoint);
1532 dput(data.mountpoint);
1533 }
1534
1535 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001536 break;
1537
1538 cond_resched();
1539 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001540}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001541EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543/**
Al Viroa4464db2011-07-07 15:03:58 -04001544 * __d_alloc - allocate a dcache entry
1545 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 * @name: qstr of the name
1547 *
1548 * Allocates a dentry. It returns %NULL if there is insufficient memory
1549 * available. On a success the dentry is returned. The name passed in is
1550 * copied and the copy passed in may be reused after this call.
1551 */
1552
Al Viroa4464db2011-07-07 15:03:58 -04001553struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
1555 struct dentry *dentry;
1556 char *dname;
1557
Mel Gormane12ba742007-10-16 01:25:52 -07001558 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 if (!dentry)
1560 return NULL;
1561
Linus Torvalds6326c712012-05-21 16:14:04 -07001562 /*
1563 * We guarantee that the inline name is always NUL-terminated.
1564 * This way the memcpy() done by the name switching in rename
1565 * will still always have a NUL at the end, even if we might
1566 * be overwriting an internal NUL character
1567 */
1568 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001570 size_t size = offsetof(struct external_name, name[1]);
1571 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1572 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 kmem_cache_free(dentry_cache, dentry);
1574 return NULL;
1575 }
Al Viro8d85b482014-09-29 14:54:27 -04001576 atomic_set(&p->u.count, 1);
1577 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001578 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1579 kasan_unpoison_shadow(dname,
1580 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 } else {
1582 dname = dentry->d_iname;
1583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584
1585 dentry->d_name.len = name->len;
1586 dentry->d_name.hash = name->hash;
1587 memcpy(dname, name->name, name->len);
1588 dname[name->len] = 0;
1589
Linus Torvalds6326c712012-05-21 16:14:04 -07001590 /* Make sure we always see the terminating NUL character */
1591 smp_wmb();
1592 dentry->d_name.name = dname;
1593
Waiman Long98474232013-08-28 18:24:59 -07001594 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001595 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001597 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001599 dentry->d_parent = dentry;
1600 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 dentry->d_op = NULL;
1602 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001603 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 INIT_LIST_HEAD(&dentry->d_lru);
1605 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001606 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1607 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001608 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Nick Piggin3e880fb2011-01-07 17:49:19 +11001610 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 return dentry;
1613}
Al Viroa4464db2011-07-07 15:03:58 -04001614
1615/**
1616 * d_alloc - allocate a dcache entry
1617 * @parent: parent of entry to allocate
1618 * @name: qstr of the name
1619 *
1620 * Allocates a dentry. It returns %NULL if there is insufficient memory
1621 * available. On a success the dentry is returned. The name passed in is
1622 * copied and the copy passed in may be reused after this call.
1623 */
1624struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1625{
1626 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1627 if (!dentry)
1628 return NULL;
1629
1630 spin_lock(&parent->d_lock);
1631 /*
1632 * don't need child lock because it is not subject
1633 * to concurrency here
1634 */
1635 __dget_dlock(parent);
1636 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001637 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001638 spin_unlock(&parent->d_lock);
1639
1640 return dentry;
1641}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001642EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001644/**
1645 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1646 * @sb: the superblock
1647 * @name: qstr of the name
1648 *
1649 * For a filesystem that just pins its dentries in memory and never
1650 * performs lookups at all, return an unhashed IS_ROOT dentry.
1651 */
Nick Piggin4b936882011-01-07 17:50:07 +11001652struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1653{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001654 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001655}
1656EXPORT_SYMBOL(d_alloc_pseudo);
1657
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1659{
1660 struct qstr q;
1661
1662 q.name = name;
1663 q.len = strlen(name);
1664 q.hash = full_name_hash(q.name, q.len);
1665 return d_alloc(parent, &q);
1666}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001667EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
Nick Pigginfb045ad2011-01-07 17:49:55 +11001669void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1670{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001671 WARN_ON_ONCE(dentry->d_op);
1672 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001673 DCACHE_OP_COMPARE |
1674 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001675 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001676 DCACHE_OP_DELETE |
1677 DCACHE_OP_SELECT_INODE));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001678 dentry->d_op = op;
1679 if (!op)
1680 return;
1681 if (op->d_hash)
1682 dentry->d_flags |= DCACHE_OP_HASH;
1683 if (op->d_compare)
1684 dentry->d_flags |= DCACHE_OP_COMPARE;
1685 if (op->d_revalidate)
1686 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001687 if (op->d_weak_revalidate)
1688 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001689 if (op->d_delete)
1690 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001691 if (op->d_prune)
1692 dentry->d_flags |= DCACHE_OP_PRUNE;
David Howells4bacc9c2015-06-18 14:32:31 +01001693 if (op->d_select_inode)
1694 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001695
1696}
1697EXPORT_SYMBOL(d_set_d_op);
1698
David Howellsdf1a0852015-01-29 12:02:28 +00001699
1700/*
1701 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1702 * @dentry - The dentry to mark
1703 *
1704 * Mark a dentry as falling through to the lower layer (as set with
1705 * d_pin_lower()). This flag may be recorded on the medium.
1706 */
1707void d_set_fallthru(struct dentry *dentry)
1708{
1709 spin_lock(&dentry->d_lock);
1710 dentry->d_flags |= DCACHE_FALLTHRU;
1711 spin_unlock(&dentry->d_lock);
1712}
1713EXPORT_SYMBOL(d_set_fallthru);
1714
David Howellsb18825a2013-09-12 19:22:53 +01001715static unsigned d_flags_for_inode(struct inode *inode)
1716{
David Howells44bdb5e2015-01-29 12:02:29 +00001717 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001718
1719 if (!inode)
1720 return DCACHE_MISS_TYPE;
1721
1722 if (S_ISDIR(inode->i_mode)) {
1723 add_flags = DCACHE_DIRECTORY_TYPE;
1724 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1725 if (unlikely(!inode->i_op->lookup))
1726 add_flags = DCACHE_AUTODIR_TYPE;
1727 else
1728 inode->i_opflags |= IOP_LOOKUP;
1729 }
David Howells44bdb5e2015-01-29 12:02:29 +00001730 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001731 }
1732
David Howells44bdb5e2015-01-29 12:02:29 +00001733 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1734 if (unlikely(inode->i_op->follow_link)) {
1735 add_flags = DCACHE_SYMLINK_TYPE;
1736 goto type_determined;
1737 }
1738 inode->i_opflags |= IOP_NOFOLLOW;
1739 }
1740
1741 if (unlikely(!S_ISREG(inode->i_mode)))
1742 add_flags = DCACHE_SPECIAL_TYPE;
1743
1744type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001745 if (unlikely(IS_AUTOMOUNT(inode)))
1746 add_flags |= DCACHE_NEED_AUTOMOUNT;
1747 return add_flags;
1748}
1749
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001750static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1751{
David Howellsb18825a2013-09-12 19:22:53 +01001752 unsigned add_flags = d_flags_for_inode(inode);
1753
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001754 spin_lock(&dentry->d_lock);
David Howellsb18825a2013-09-12 19:22:53 +01001755 if (inode)
Al Viro946e51f2014-10-26 19:19:16 -04001756 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
David Howells4bf46a22015-03-05 14:09:22 +00001757 __d_set_inode_and_type(dentry, inode, add_flags);
Peter Zijlstraa7c6f572015-06-11 14:46:46 +02001758 dentry_rcuwalk_invalidate(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001759 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001760 fsnotify_d_instantiate(dentry, inode);
1761}
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763/**
1764 * d_instantiate - fill in inode information for a dentry
1765 * @entry: dentry to complete
1766 * @inode: inode to attach to this dentry
1767 *
1768 * Fill in inode information in the entry.
1769 *
1770 * This turns negative dentries into productive full members
1771 * of society.
1772 *
1773 * NOTE! This assumes that the inode count has been incremented
1774 * (or otherwise set) by the caller to indicate that it is now
1775 * in use by the dcache.
1776 */
1777
1778void d_instantiate(struct dentry *entry, struct inode * inode)
1779{
Al Viro946e51f2014-10-26 19:19:16 -04001780 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001781 if (inode)
1782 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001783 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001784 if (inode)
1785 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 security_d_instantiate(entry, inode);
1787}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001788EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
1790/**
1791 * d_instantiate_unique - instantiate a non-aliased dentry
1792 * @entry: dentry to instantiate
1793 * @inode: inode to attach to this dentry
1794 *
1795 * Fill in inode information in the entry. On success, it returns NULL.
1796 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001797 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 *
1799 * Note that in order to avoid conflicts with rename() etc, the caller
1800 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001801 *
1802 * This also assumes that the inode count has been incremented
1803 * (or otherwise set) by the caller to indicate that it is now
1804 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 */
David Howells770bfad2006-08-22 20:06:07 -04001806static struct dentry *__d_instantiate_unique(struct dentry *entry,
1807 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808{
1809 struct dentry *alias;
1810 int len = entry->d_name.len;
1811 const char *name = entry->d_name.name;
1812 unsigned int hash = entry->d_name.hash;
1813
David Howells770bfad2006-08-22 20:06:07 -04001814 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001815 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001816 return NULL;
1817 }
1818
Al Viro946e51f2014-10-26 19:19:16 -04001819 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001820 /*
1821 * Don't need alias->d_lock here, because aliases with
1822 * d_parent == entry->d_parent are not subject to name or
1823 * parent changes, because the parent inode i_mutex is held.
1824 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001825 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 continue;
1827 if (alias->d_parent != entry->d_parent)
1828 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001829 if (alias->d_name.len != len)
1830 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001831 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001833 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 return alias;
1835 }
David Howells770bfad2006-08-22 20:06:07 -04001836
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001837 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 return NULL;
1839}
David Howells770bfad2006-08-22 20:06:07 -04001840
1841struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1842{
1843 struct dentry *result;
1844
Al Viro946e51f2014-10-26 19:19:16 -04001845 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001846
Nick Piggin873feea2011-01-07 17:50:06 +11001847 if (inode)
1848 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001849 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001850 if (inode)
1851 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001852
1853 if (!result) {
1854 security_d_instantiate(entry, inode);
1855 return NULL;
1856 }
1857
1858 BUG_ON(!d_unhashed(result));
1859 iput(inode);
1860 return result;
1861}
1862
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863EXPORT_SYMBOL(d_instantiate_unique);
1864
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001865/**
1866 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1867 * @entry: dentry to complete
1868 * @inode: inode to attach to this dentry
1869 *
1870 * Fill in inode information in the entry. If a directory alias is found, then
1871 * return an error (and drop inode). Together with d_materialise_unique() this
1872 * guarantees that a directory inode may never have more than one alias.
1873 */
1874int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1875{
Al Viro946e51f2014-10-26 19:19:16 -04001876 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001877
1878 spin_lock(&inode->i_lock);
1879 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1880 spin_unlock(&inode->i_lock);
1881 iput(inode);
1882 return -EBUSY;
1883 }
1884 __d_instantiate(entry, inode);
1885 spin_unlock(&inode->i_lock);
1886 security_d_instantiate(entry, inode);
1887
1888 return 0;
1889}
1890EXPORT_SYMBOL(d_instantiate_no_diralias);
1891
Al Viroadc0e912012-01-08 16:49:21 -05001892struct dentry *d_make_root(struct inode *root_inode)
1893{
1894 struct dentry *res = NULL;
1895
1896 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001897 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001898
1899 res = __d_alloc(root_inode->i_sb, &name);
1900 if (res)
1901 d_instantiate(res, root_inode);
1902 else
1903 iput(root_inode);
1904 }
1905 return res;
1906}
1907EXPORT_SYMBOL(d_make_root);
1908
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001909static struct dentry * __d_find_any_alias(struct inode *inode)
1910{
1911 struct dentry *alias;
1912
Al Virob3d9b7a2012-06-09 13:51:19 -04001913 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001914 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001915 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001916 __dget(alias);
1917 return alias;
1918}
1919
Sage Weil46f72b32012-01-10 09:04:37 -08001920/**
1921 * d_find_any_alias - find any alias for a given inode
1922 * @inode: inode to find an alias for
1923 *
1924 * If any aliases exist for the given inode, take and return a
1925 * reference for one of them. If no aliases exist, return %NULL.
1926 */
1927struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001928{
1929 struct dentry *de;
1930
1931 spin_lock(&inode->i_lock);
1932 de = __d_find_any_alias(inode);
1933 spin_unlock(&inode->i_lock);
1934 return de;
1935}
Sage Weil46f72b32012-01-10 09:04:37 -08001936EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001937
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001938static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001939{
NeilBrownb911a6b2012-11-08 16:09:37 -08001940 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001941 struct dentry *tmp;
1942 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001943 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001944
1945 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001946 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001947 if (IS_ERR(inode))
1948 return ERR_CAST(inode);
1949
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001950 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001951 if (res)
1952 goto out_iput;
1953
Al Viroa4464db2011-07-07 15:03:58 -04001954 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001955 if (!tmp) {
1956 res = ERR_PTR(-ENOMEM);
1957 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001958 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001959
Nick Piggin873feea2011-01-07 17:50:06 +11001960 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001961 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001962 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001963 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001964 dput(tmp);
1965 goto out_iput;
1966 }
1967
1968 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001969 add_flags = d_flags_for_inode(inode);
1970
1971 if (disconnected)
1972 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001973
Christoph Hellwig9308a612008-08-11 15:49:12 +02001974 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001975 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001976 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001977 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001978 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001979 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001980 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001981 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001982 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001983
Christoph Hellwig9308a612008-08-11 15:49:12 +02001984 return tmp;
1985
1986 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001987 if (res && !IS_ERR(res))
1988 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001989 iput(inode);
1990 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001991}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001992
1993/**
1994 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1995 * @inode: inode to allocate the dentry for
1996 *
1997 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1998 * similar open by handle operations. The returned dentry may be anonymous,
1999 * or may have a full name (if the inode was already in the cache).
2000 *
2001 * When called on a directory inode, we must ensure that the inode only ever
2002 * has one dentry. If a dentry is found, that is returned instead of
2003 * allocating a new one.
2004 *
2005 * On successful return, the reference to the inode has been transferred
2006 * to the dentry. In case of an error the reference on the inode is released.
2007 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2008 * be passed in and the error will be propagated to the return value,
2009 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2010 */
2011struct dentry *d_obtain_alias(struct inode *inode)
2012{
2013 return __d_obtain_alias(inode, 1);
2014}
Benny Halevyadc48722009-02-27 14:02:59 -08002015EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
2017/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002018 * d_obtain_root - find or allocate a dentry for a given inode
2019 * @inode: inode to allocate the dentry for
2020 *
2021 * Obtain an IS_ROOT dentry for the root of a filesystem.
2022 *
2023 * We must ensure that directory inodes only ever have one dentry. If a
2024 * dentry is found, that is returned instead of allocating a new one.
2025 *
2026 * On successful return, the reference to the inode has been transferred
2027 * to the dentry. In case of an error the reference on the inode is
2028 * released. A %NULL or IS_ERR inode may be passed in and will be the
2029 * error will be propagate to the return value, with a %NULL @inode
2030 * replaced by ERR_PTR(-ESTALE).
2031 */
2032struct dentry *d_obtain_root(struct inode *inode)
2033{
2034 return __d_obtain_alias(inode, 0);
2035}
2036EXPORT_SYMBOL(d_obtain_root);
2037
2038/**
Barry Naujok94035402008-05-21 16:50:46 +10002039 * d_add_ci - lookup or allocate new dentry with case-exact name
2040 * @inode: the inode case-insensitive lookup has found
2041 * @dentry: the negative dentry that was passed to the parent's lookup func
2042 * @name: the case-exact name to be associated with the returned dentry
2043 *
2044 * This is to avoid filling the dcache with case-insensitive names to the
2045 * same inode, only the actual correct case is stored in the dcache for
2046 * case-insensitive filesystems.
2047 *
2048 * For a case-insensitive lookup match and if the the case-exact dentry
2049 * already exists in in the dcache, use it and return it.
2050 *
2051 * If no entry exists with the exact case name, allocate new dentry with
2052 * the exact case, and return the spliced entry.
2053 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002054struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002055 struct qstr *name)
2056{
Barry Naujok94035402008-05-21 16:50:46 +10002057 struct dentry *found;
2058 struct dentry *new;
2059
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002060 /*
2061 * First check if a dentry matching the name already exists,
2062 * if not go ahead and create it now.
2063 */
Barry Naujok94035402008-05-21 16:50:46 +10002064 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10002065 if (!found) {
2066 new = d_alloc(dentry->d_parent, name);
2067 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002068 found = ERR_PTR(-ENOMEM);
Al Viro427c77d2014-10-12 18:46:26 -04002069 } else {
2070 found = d_splice_alias(inode, new);
2071 if (found) {
2072 dput(new);
2073 return found;
2074 }
2075 return new;
Barry Naujok94035402008-05-21 16:50:46 +10002076 }
Barry Naujok94035402008-05-21 16:50:46 +10002077 }
Barry Naujok94035402008-05-21 16:50:46 +10002078 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002079 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002080}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002081EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002083/*
2084 * Do the slow-case of the dentry name compare.
2085 *
2086 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002087 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002088 * filesystem can rely on them, and can use the 'name' and
2089 * 'len' information without worrying about walking off the
2090 * end of memory etc.
2091 *
2092 * Thus the read_seqcount_retry() and the "duplicate" info
2093 * in arguments (the low-level filesystem should not look
2094 * at the dentry inode or name contents directly, since
2095 * rename can change them while we're in RCU mode).
2096 */
2097enum slow_d_compare {
2098 D_COMP_OK,
2099 D_COMP_NOMATCH,
2100 D_COMP_SEQRETRY,
2101};
2102
2103static noinline enum slow_d_compare slow_dentry_cmp(
2104 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002105 struct dentry *dentry,
2106 unsigned int seq,
2107 const struct qstr *name)
2108{
2109 int tlen = dentry->d_name.len;
2110 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002111
2112 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2113 cpu_relax();
2114 return D_COMP_SEQRETRY;
2115 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002116 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002117 return D_COMP_NOMATCH;
2118 return D_COMP_OK;
2119}
2120
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002122 * __d_lookup_rcu - search for a dentry (racy, store-free)
2123 * @parent: parent dentry
2124 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002125 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002126 * Returns: dentry, or NULL
2127 *
2128 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2129 * resolution (store-free path walking) design described in
2130 * Documentation/filesystems/path-lookup.txt.
2131 *
2132 * This is not to be used outside core vfs.
2133 *
2134 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2135 * held, and rcu_read_lock held. The returned dentry must not be stored into
2136 * without taking d_lock and checking d_seq sequence count against @seq
2137 * returned here.
2138 *
Linus Torvalds15570082013-09-02 11:38:06 -07002139 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002140 * function.
2141 *
2142 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2143 * the returned dentry, so long as its parent's seqlock is checked after the
2144 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2145 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002146 *
2147 * NOTE! The caller *has* to check the resulting dentry against the sequence
2148 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002149 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002150struct dentry *__d_lookup_rcu(const struct dentry *parent,
2151 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002152 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002153{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002154 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002155 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002156 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002157 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002158 struct dentry *dentry;
2159
2160 /*
2161 * Note: There is significant duplication with __d_lookup_rcu which is
2162 * required to prevent single threaded performance regressions
2163 * especially on architectures where smp_rmb (in seqcounts) are costly.
2164 * Keep the two functions in sync.
2165 */
2166
2167 /*
2168 * The hash list is protected using RCU.
2169 *
2170 * Carefully use d_seq when comparing a candidate dentry, to avoid
2171 * races with d_move().
2172 *
2173 * It is possible that concurrent renames can mess up our list
2174 * walk here and result in missing our dentry, resulting in the
2175 * false-negative result. d_lookup() protects against concurrent
2176 * renames using rename_lock seqlock.
2177 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002178 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002179 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002180 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002181 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002182
Nick Piggin31e6b012011-01-07 17:49:52 +11002183seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002184 /*
2185 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002186 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002187 *
2188 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002189 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002190 *
2191 * NOTE! We do a "raw" seqcount_begin here. That means that
2192 * we don't wait for the sequence count to stabilize if it
2193 * is in the middle of a sequence change. If we do the slow
2194 * dentry compare, we will do seqretries until it is stable,
2195 * and if we end up with a successful lookup, we actually
2196 * want to exit RCU lookup anyway.
2197 */
2198 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002199 if (dentry->d_parent != parent)
2200 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002201 if (d_unhashed(dentry))
2202 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002203
2204 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002205 if (dentry->d_name.hash != hashlen_hash(hashlen))
2206 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002207 *seqp = seq;
2208 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002209 case D_COMP_OK:
2210 return dentry;
2211 case D_COMP_NOMATCH:
2212 continue;
2213 default:
2214 goto seqretry;
2215 }
2216 }
2217
Linus Torvalds26fe5752012-05-10 13:14:12 -07002218 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002219 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002220 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002221 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002222 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002223 }
2224 return NULL;
2225}
2226
2227/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 * d_lookup - search for a dentry
2229 * @parent: parent dentry
2230 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002231 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002233 * d_lookup searches the children of the parent dentry for the name in
2234 * question. If the dentry is found its reference count is incremented and the
2235 * dentry is returned. The caller must use dput to free the entry when it has
2236 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 */
Al Viroda2d8452013-01-24 18:29:34 -05002238struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239{
Nick Piggin31e6b012011-01-07 17:49:52 +11002240 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002241 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
Daeseok Younb8314f92014-08-11 11:46:53 +09002243 do {
2244 seq = read_seqbegin(&rename_lock);
2245 dentry = __d_lookup(parent, name);
2246 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 break;
2248 } while (read_seqretry(&rename_lock, seq));
2249 return dentry;
2250}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002251EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252
Nick Piggin31e6b012011-01-07 17:49:52 +11002253/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002254 * __d_lookup - search for a dentry (racy)
2255 * @parent: parent dentry
2256 * @name: qstr of name we wish to find
2257 * Returns: dentry, or NULL
2258 *
2259 * __d_lookup is like d_lookup, however it may (rarely) return a
2260 * false-negative result due to unrelated rename activity.
2261 *
2262 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2263 * however it must be used carefully, eg. with a following d_lookup in
2264 * the case of failure.
2265 *
2266 * __d_lookup callers must be commented.
2267 */
Al Viroa713ca22013-01-24 18:27:00 -05002268struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269{
2270 unsigned int len = name->len;
2271 unsigned int hash = name->hash;
2272 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002273 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002274 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002275 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002276 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Nick Pigginb04f7842010-08-18 04:37:34 +10002278 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002279 * Note: There is significant duplication with __d_lookup_rcu which is
2280 * required to prevent single threaded performance regressions
2281 * especially on architectures where smp_rmb (in seqcounts) are costly.
2282 * Keep the two functions in sync.
2283 */
2284
2285 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002286 * The hash list is protected using RCU.
2287 *
2288 * Take d_lock when comparing a candidate dentry, to avoid races
2289 * with d_move().
2290 *
2291 * It is possible that concurrent renames can mess up our list
2292 * walk here and result in missing our dentry, resulting in the
2293 * false-negative result. d_lookup() protects against concurrent
2294 * renames using rename_lock seqlock.
2295 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002296 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002297 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 rcu_read_lock();
2299
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002300 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 if (dentry->d_name.hash != hash)
2303 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 if (dentry->d_parent != parent)
2307 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002308 if (d_unhashed(dentry))
2309 goto next;
2310
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 /*
2312 * It is safe to compare names since d_move() cannot
2313 * change the qstr (protected by d_lock).
2314 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002315 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002316 int tlen = dentry->d_name.len;
2317 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002318 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 goto next;
2320 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002321 if (dentry->d_name.len != len)
2322 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002323 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 goto next;
2325 }
2326
Waiman Long98474232013-08-28 18:24:59 -07002327 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002328 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 spin_unlock(&dentry->d_lock);
2330 break;
2331next:
2332 spin_unlock(&dentry->d_lock);
2333 }
2334 rcu_read_unlock();
2335
2336 return found;
2337}
2338
2339/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002340 * d_hash_and_lookup - hash the qstr then search for a dentry
2341 * @dir: Directory to search in
2342 * @name: qstr of name we wish to find
2343 *
Al Viro4f522a22013-02-11 23:20:37 -05002344 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002345 */
2346struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2347{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002348 /*
2349 * Check for a fs-specific hash function. Note that we must
2350 * calculate the standard hash first, as the d_op->d_hash()
2351 * routine may choose to leave the hash value unchanged.
2352 */
2353 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002354 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002355 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002356 if (unlikely(err < 0))
2357 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002358 }
Al Viro4f522a22013-02-11 23:20:37 -05002359 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002360}
Al Viro4f522a22013-02-11 23:20:37 -05002361EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002362
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363/*
2364 * When a file is deleted, we have two options:
2365 * - turn this dentry into a negative dentry
2366 * - unhash this dentry and free it.
2367 *
2368 * Usually, we want to just turn this into
2369 * a negative dentry, but if anybody else is
2370 * currently using the dentry or the inode
2371 * we can't do that and we fall back on removing
2372 * it from the hash queues and waiting for
2373 * it to be deleted later when it has no users
2374 */
2375
2376/**
2377 * d_delete - delete a dentry
2378 * @dentry: The dentry to delete
2379 *
2380 * Turn the dentry into a negative dentry if possible, otherwise
2381 * remove it from the hash queues so it can be deleted later
2382 */
2383
2384void d_delete(struct dentry * dentry)
2385{
Nick Piggin873feea2011-01-07 17:50:06 +11002386 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002387 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 /*
2389 * Are we the only user?
2390 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002391again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002393 inode = dentry->d_inode;
2394 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002395 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002396 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002397 spin_unlock(&dentry->d_lock);
2398 cpu_relax();
2399 goto again;
2400 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002401 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002402 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002403 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 return;
2405 }
2406
2407 if (!d_unhashed(dentry))
2408 __d_drop(dentry);
2409
2410 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002411
2412 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002414EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002416static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002418 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002419 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002420 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002421 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002422 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423}
2424
David Howells770bfad2006-08-22 20:06:07 -04002425static void _d_rehash(struct dentry * entry)
2426{
2427 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2428}
2429
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430/**
2431 * d_rehash - add an entry back to the hash
2432 * @entry: dentry to add to the hash
2433 *
2434 * Adds a dentry to the hash according to its name.
2435 */
2436
2437void d_rehash(struct dentry * entry)
2438{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002440 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002443EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002445/**
2446 * dentry_update_name_case - update case insensitive dentry with a new name
2447 * @dentry: dentry to be updated
2448 * @name: new name
2449 *
2450 * Update a case insensitive dentry with new case of name.
2451 *
2452 * dentry must have been returned by d_lookup with name @name. Old and new
2453 * name lengths must match (ie. no d_compare which allows mismatched name
2454 * lengths).
2455 *
2456 * Parent inode i_mutex must be held over d_lookup and into this call (to
2457 * keep renames and concurrent inserts, and readdir(2) away).
2458 */
2459void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2460{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002461 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002462 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2463
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002464 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002465 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002466 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002467 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002468 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002469}
2470EXPORT_SYMBOL(dentry_update_name_case);
2471
Al Viro8d85b482014-09-29 14:54:27 -04002472static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473{
Al Viro8d85b482014-09-29 14:54:27 -04002474 if (unlikely(dname_external(target))) {
2475 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 /*
2477 * Both external: swap the pointers
2478 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002479 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 } else {
2481 /*
2482 * dentry:internal, target:external. Steal target's
2483 * storage and make target internal.
2484 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002485 memcpy(target->d_iname, dentry->d_name.name,
2486 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 dentry->d_name.name = target->d_name.name;
2488 target->d_name.name = target->d_iname;
2489 }
2490 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002491 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 /*
2493 * dentry:external, target:internal. Give dentry's
2494 * storage to target and make dentry internal
2495 */
2496 memcpy(dentry->d_iname, target->d_name.name,
2497 target->d_name.len + 1);
2498 target->d_name.name = dentry->d_name.name;
2499 dentry->d_name.name = dentry->d_iname;
2500 } else {
2501 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002502 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002504 unsigned int i;
2505 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002506 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2507 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002508 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2509 swap(((long *) &dentry->d_iname)[i],
2510 ((long *) &target->d_iname)[i]);
2511 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 }
2513 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002514 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515}
2516
Al Viro8d85b482014-09-29 14:54:27 -04002517static void copy_name(struct dentry *dentry, struct dentry *target)
2518{
2519 struct external_name *old_name = NULL;
2520 if (unlikely(dname_external(dentry)))
2521 old_name = external_name(dentry);
2522 if (unlikely(dname_external(target))) {
2523 atomic_inc(&external_name(target)->u.count);
2524 dentry->d_name = target->d_name;
2525 } else {
2526 memcpy(dentry->d_iname, target->d_name.name,
2527 target->d_name.len + 1);
2528 dentry->d_name.name = dentry->d_iname;
2529 dentry->d_name.hash_len = target->d_name.hash_len;
2530 }
2531 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2532 kfree_rcu(old_name, u.head);
2533}
2534
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002535static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2536{
2537 /*
2538 * XXXX: do we really need to take target->d_lock?
2539 */
2540 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2541 spin_lock(&target->d_parent->d_lock);
2542 else {
2543 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2544 spin_lock(&dentry->d_parent->d_lock);
2545 spin_lock_nested(&target->d_parent->d_lock,
2546 DENTRY_D_LOCK_NESTED);
2547 } else {
2548 spin_lock(&target->d_parent->d_lock);
2549 spin_lock_nested(&dentry->d_parent->d_lock,
2550 DENTRY_D_LOCK_NESTED);
2551 }
2552 }
2553 if (target < dentry) {
2554 spin_lock_nested(&target->d_lock, 2);
2555 spin_lock_nested(&dentry->d_lock, 3);
2556 } else {
2557 spin_lock_nested(&dentry->d_lock, 2);
2558 spin_lock_nested(&target->d_lock, 3);
2559 }
2560}
2561
Al Viro986c0192014-09-26 23:11:15 -04002562static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002563{
2564 if (target->d_parent != dentry->d_parent)
2565 spin_unlock(&dentry->d_parent->d_lock);
2566 if (target->d_parent != target)
2567 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002568 spin_unlock(&target->d_lock);
2569 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002570}
2571
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002573 * When switching names, the actual string doesn't strictly have to
2574 * be preserved in the target - because we're dropping the target
2575 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002576 * the new name before we switch, unless we are going to rehash
2577 * it. Note that if we *do* unhash the target, we are not allowed
2578 * to rehash it without giving it a new name/hash key - whether
2579 * we swap or overwrite the names here, resulting name won't match
2580 * the reality in filesystem; it's only there for d_path() purposes.
2581 * Note that all of this is happening under rename_lock, so the
2582 * any hash lookup seeing it in the middle of manipulations will
2583 * be discarded anyway. So we do not care what happens to the hash
2584 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002586/*
Al Viro18367502011-07-12 21:42:24 -04002587 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 * @dentry: entry to move
2589 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002590 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 *
2592 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002593 * dcache entries should not be moved in this way. Caller must hold
2594 * rename_lock, the i_mutex of the source and target directories,
2595 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002597static void __d_move(struct dentry *dentry, struct dentry *target,
2598 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 if (!dentry->d_inode)
2601 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2602
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002603 BUG_ON(d_ancestor(dentry, target));
2604 BUG_ON(d_ancestor(target, dentry));
2605
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002606 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Nick Piggin31e6b012011-01-07 17:49:52 +11002608 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002609 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002610
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002611 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2612
2613 /*
2614 * Move the dentry to the target hash queue. Don't bother checking
2615 * for the same hash queue because of how unlikely it is.
2616 */
2617 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002618 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
Miklos Szeredida1ce062014-04-01 17:08:43 +02002620 /*
2621 * Unhash the target (d_delete() is not usable here). If exchanging
2622 * the two dentries, then rehash onto the other's hash queue.
2623 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002625 if (exchange) {
2626 __d_rehash(target,
2627 d_hash(dentry->d_parent, dentry->d_name.hash));
2628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002631 if (exchange)
2632 swap_names(dentry, target);
2633 else
2634 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
Al Viro63cf4272014-09-26 23:06:14 -04002636 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002638 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 dentry->d_parent = target->d_parent;
2640 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002641 list_del_init(&target->d_child);
2642 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002644 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002645 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002646 list_move(&target->d_child, &target->d_parent->d_subdirs);
2647 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002648 if (exchange)
2649 fsnotify_d_move(target);
2650 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 }
2652
Nick Piggin31e6b012011-01-07 17:49:52 +11002653 write_seqcount_end(&target->d_seq);
2654 write_seqcount_end(&dentry->d_seq);
2655
Al Viro986c0192014-09-26 23:11:15 -04002656 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002657}
2658
2659/*
2660 * d_move - move a dentry
2661 * @dentry: entry to move
2662 * @target: new dentry
2663 *
2664 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002665 * dcache entries should not be moved in this way. See the locking
2666 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002667 */
2668void d_move(struct dentry *dentry, struct dentry *target)
2669{
2670 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002671 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002673}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002674EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675
Miklos Szeredida1ce062014-04-01 17:08:43 +02002676/*
2677 * d_exchange - exchange two dentries
2678 * @dentry1: first dentry
2679 * @dentry2: second dentry
2680 */
2681void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2682{
2683 write_seqlock(&rename_lock);
2684
2685 WARN_ON(!dentry1->d_inode);
2686 WARN_ON(!dentry2->d_inode);
2687 WARN_ON(IS_ROOT(dentry1));
2688 WARN_ON(IS_ROOT(dentry2));
2689
2690 __d_move(dentry1, dentry2, true);
2691
2692 write_sequnlock(&rename_lock);
2693}
2694
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002695/**
2696 * d_ancestor - search for an ancestor
2697 * @p1: ancestor dentry
2698 * @p2: child dentry
2699 *
2700 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2701 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002702 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002703struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002704{
2705 struct dentry *p;
2706
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002707 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002708 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002709 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002710 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002711 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002712}
2713
2714/*
2715 * This helper attempts to cope with remotely renamed directories
2716 *
2717 * It assumes that the caller is already holding
Al Viro18367502011-07-12 21:42:24 -04002718 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002719 *
2720 * Note: If ever the locking in lock_rename() changes, then please
2721 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002722 */
Al Virob5ae6b12014-10-12 22:16:02 -04002723static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002724 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002725{
2726 struct mutex *m1 = NULL, *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002727 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002728
2729 /* If alias and dentry share a parent, then no extra locks required */
2730 if (alias->d_parent == dentry->d_parent)
2731 goto out_unalias;
2732
Trond Myklebust9eaef272006-10-21 10:24:20 -07002733 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002734 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2735 goto out_err;
2736 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2737 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2738 goto out_err;
2739 m2 = &alias->d_parent->d_inode->i_mutex;
2740out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002741 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002742 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002743out_err:
Nick Piggin873feea2011-01-07 17:50:06 +11002744 spin_unlock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002745 if (m2)
2746 mutex_unlock(m2);
2747 if (m1)
2748 mutex_unlock(m1);
2749 return ret;
2750}
2751
David Howells770bfad2006-08-22 20:06:07 -04002752/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002753 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2754 * @inode: the inode which may have a disconnected dentry
2755 * @dentry: a negative dentry which we want to point to the inode.
2756 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002757 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2758 * place of the given dentry and return it, else simply d_add the inode
2759 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002760 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002761 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2762 * we should error out: directories can't have multiple aliases.
2763 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002764 * This is needed in the lookup routine of any filesystem that is exportable
2765 * (via knfsd) so that we can build dcache paths to directories effectively.
2766 *
2767 * If a dentry was found and moved, then it is returned. Otherwise NULL
2768 * is returned. This matches the expected return value of ->lookup.
2769 *
2770 * Cluster filesystems may call this function with a negative, hashed dentry.
2771 * In that case, we know that the inode will be a regular file, and also this
2772 * will only occur during atomic_open. So we need to check for the dentry
2773 * being already hashed only in the final case.
2774 */
2775struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2776{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002777 if (IS_ERR(inode))
2778 return ERR_CAST(inode);
2779
David Howells770bfad2006-08-22 20:06:07 -04002780 BUG_ON(!d_unhashed(dentry));
2781
David Howells770bfad2006-08-22 20:06:07 -04002782 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002783 __d_instantiate(dentry, NULL);
Al Virob5ae6b12014-10-12 22:16:02 -04002784 goto out;
David Howells770bfad2006-08-22 20:06:07 -04002785 }
Nick Piggin873feea2011-01-07 17:50:06 +11002786 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002787 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002788 struct dentry *new = __d_find_any_alias(inode);
2789 if (unlikely(new)) {
Al Viro18367502011-07-12 21:42:24 -04002790 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002791 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002792 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002793 spin_unlock(&inode->i_lock);
2794 dput(new);
2795 new = ERR_PTR(-ELOOP);
2796 pr_warn_ratelimited(
2797 "VFS: Lookup of '%s' in %s %s"
2798 " would have caused loop\n",
2799 dentry->d_name.name,
2800 inode->i_sb->s_type->name,
2801 inode->i_sb->s_id);
2802 } else if (!IS_ROOT(new)) {
2803 int err = __d_unalias(inode, dentry, new);
2804 write_sequnlock(&rename_lock);
2805 if (err) {
2806 dput(new);
2807 new = ERR_PTR(err);
2808 }
Al Viro18367502011-07-12 21:42:24 -04002809 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002810 __d_move(new, dentry, false);
2811 write_sequnlock(&rename_lock);
2812 spin_unlock(&inode->i_lock);
2813 security_d_instantiate(new, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002814 }
Al Virob5ae6b12014-10-12 22:16:02 -04002815 iput(inode);
2816 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002817 }
David Howells770bfad2006-08-22 20:06:07 -04002818 }
Al Virob5ae6b12014-10-12 22:16:02 -04002819 /* already taking inode->i_lock, so d_add() by hand */
2820 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002821 spin_unlock(&inode->i_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002822out:
2823 security_d_instantiate(dentry, inode);
2824 d_rehash(dentry);
2825 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002826}
Al Virob5ae6b12014-10-12 22:16:02 -04002827EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002828
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002829static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002830{
2831 *buflen -= namelen;
2832 if (*buflen < 0)
2833 return -ENAMETOOLONG;
2834 *buffer -= namelen;
2835 memcpy(*buffer, str, namelen);
2836 return 0;
2837}
2838
Waiman Long232d2d62013-09-09 12:18:13 -04002839/**
2840 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002841 * @buffer: buffer pointer
2842 * @buflen: allocated length of the buffer
2843 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002844 *
2845 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2846 * make sure that either the old or the new name pointer and length are
2847 * fetched. However, there may be mismatch between length and pointer.
2848 * The length cannot be trusted, we need to copy it byte-by-byte until
2849 * the length is reached or a null byte is found. It also prepends "/" at
2850 * the beginning of the name. The sequence number check at the caller will
2851 * retry it again when a d_move() does happen. So any garbage in the buffer
2852 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002853 *
2854 * Data dependency barrier is needed to make sure that we see that terminating
2855 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002856 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002857static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2858{
Waiman Long232d2d62013-09-09 12:18:13 -04002859 const char *dname = ACCESS_ONCE(name->name);
2860 u32 dlen = ACCESS_ONCE(name->len);
2861 char *p;
2862
Al Viro6d13f692014-09-29 14:46:30 -04002863 smp_read_barrier_depends();
2864
Waiman Long232d2d62013-09-09 12:18:13 -04002865 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002866 if (*buflen < 0)
2867 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002868 p = *buffer -= dlen + 1;
2869 *p++ = '/';
2870 while (dlen--) {
2871 char c = *dname++;
2872 if (!c)
2873 break;
2874 *p++ = c;
2875 }
2876 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002877}
2878
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002880 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002881 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002882 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002883 * @buffer: pointer to the end of the buffer
2884 * @buflen: pointer to buffer length
2885 *
Waiman Long18129972013-09-12 10:55:35 -04002886 * The function will first try to write out the pathname without taking any
2887 * lock other than the RCU read lock to make sure that dentries won't go away.
2888 * It only checks the sequence number of the global rename_lock as any change
2889 * in the dentry's d_seq will be preceded by changes in the rename_lock
2890 * sequence number. If the sequence number had been changed, it will restart
2891 * the whole pathname back-tracing sequence again by taking the rename_lock.
2892 * In this case, there is no need to take the RCU read lock as the recursive
2893 * parent pointer references will keep the dentry chain alive as long as no
2894 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002895 */
Al Viro02125a82011-12-05 08:43:34 -05002896static int prepend_path(const struct path *path,
2897 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002898 char **buffer, int *buflen)
2899{
Al Viroede4ceb2013-11-13 07:45:40 -05002900 struct dentry *dentry;
2901 struct vfsmount *vfsmnt;
2902 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002903 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002904 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002905 char *bptr;
2906 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002907
Al Viro48f5ec22013-09-09 15:22:25 -04002908 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002909restart_mnt:
2910 read_seqbegin_or_lock(&mount_lock, &m_seq);
2911 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002912 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002913restart:
2914 bptr = *buffer;
2915 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002916 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002917 dentry = path->dentry;
2918 vfsmnt = path->mnt;
2919 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002920 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002921 while (dentry != root->dentry || vfsmnt != root->mnt) {
2922 struct dentry * parent;
2923
2924 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002925 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
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 *
3297 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3298 * Returns 0 otherwise.
3299 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3300 */
3301
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003302int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303{
3304 int 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)
3308 return 1;
3309
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))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 result = 1;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003320 else
3321 result = 0;
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,
3409 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
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
3442void __init vfs_caches_init(unsigned long mempages)
3443{
3444 unsigned long reserve;
3445
3446 /* Base hash sizes on available memory, with a reserve equal to
3447 150% of current kernel size */
3448
3449 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3450 mempages -= reserve;
3451
3452 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003453 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454
Denis Cheng74bf17c2007-10-16 23:26:30 -07003455 dcache_init();
3456 inode_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 files_init(mempages);
Denis Cheng74bf17c2007-10-16 23:26:30 -07003458 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 bdev_cache_init();
3460 chrdev_init();
3461}