blob: 8d38cd07b207b9234d0f571bb4a945e05a395c7b [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
Al Viro75a6f822015-07-08 02:42:38 +0100645 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800646 */
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);
Al Viro75a6f822015-07-08 02:42:38 +0100700 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
Linus Torvalds360f5472015-01-09 15:19:03 -0800701
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
Al Viro75a6f822015-07-08 02:42:38 +0100779 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
780 goto kill_it;
781
Linus Torvalds8aab6a22013-09-08 13:26:18 -0700782 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (dentry->d_op->d_delete(dentry))
Nick Piggin61f3dee2011-01-07 17:49:40 +1100784 goto kill_it;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 }
Nick Piggin265ac902010-10-10 05:36:24 -0400786
Linus Torvalds358eec12013-10-31 15:43:02 -0700787 if (!(dentry->d_flags & DCACHE_REFERENCED))
788 dentry->d_flags |= DCACHE_REFERENCED;
Christoph Hellwiga4633352010-10-10 05:36:26 -0400789 dentry_lru_add(dentry);
Nick Piggin265ac902010-10-10 05:36:24 -0400790
Waiman Long98474232013-08-28 18:24:59 -0700791 dentry->d_lockref.count--;
Nick Piggin61f3dee2011-01-07 17:49:40 +1100792 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return;
794
Miklos Szeredid52b9082007-05-08 00:23:46 -0700795kill_it:
Al Viro8cbf74d2014-05-29 09:18:26 -0400796 dentry = dentry_kill(dentry);
Miklos Szeredid52b9082007-05-08 00:23:46 -0700797 if (dentry)
798 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700800EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100803/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100804static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
Waiman Long98474232013-08-28 18:24:59 -0700806 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
808
Nick Piggindc0474b2011-01-07 17:49:43 +1100809static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100810{
Waiman Long98474232013-08-28 18:24:59 -0700811 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100812}
813
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100814struct dentry *dget_parent(struct dentry *dentry)
815{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700816 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100817 struct dentry *ret;
818
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700819 /*
820 * Do optimistic parent lookup without any
821 * locking.
822 */
823 rcu_read_lock();
824 ret = ACCESS_ONCE(dentry->d_parent);
825 gotref = lockref_get_not_zero(&ret->d_lockref);
826 rcu_read_unlock();
827 if (likely(gotref)) {
828 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
829 return ret;
830 dput(ret);
831 }
832
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100833repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100834 /*
835 * Don't need rcu_dereference because we re-check it was correct under
836 * the lock.
837 */
838 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100839 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100840 spin_lock(&ret->d_lock);
841 if (unlikely(ret != dentry->d_parent)) {
842 spin_unlock(&ret->d_lock);
843 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100844 goto repeat;
845 }
Nick Piggina734eb42011-01-07 17:49:44 +1100846 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700847 BUG_ON(!ret->d_lockref.count);
848 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100849 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100850 return ret;
851}
852EXPORT_SYMBOL(dget_parent);
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854/**
855 * d_find_alias - grab a hashed alias of inode
856 * @inode: inode in question
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 *
858 * If inode has a hashed alias, or is a directory and has any alias,
859 * acquire the reference to alias and return it. Otherwise return NULL.
860 * Notice that if inode is a directory there can be only one alias and
861 * it can be unhashed only if it has no children, or if it is the root
Eric W. Biederman3ccb3542014-02-12 16:08:06 -0800862 * of a filesystem, or if the directory was renamed and d_revalidate
863 * was the first vfs operation to notice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 *
NeilBrown21c0d8f2006-10-04 02:16:16 -0700865 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500866 * any other hashed alias over that one.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 */
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500868static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
Nick Pigginda502952011-01-07 17:49:33 +1100870 struct dentry *alias, *discon_alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Nick Pigginda502952011-01-07 17:49:33 +1100872again:
873 discon_alias = NULL;
Al Viro946e51f2014-10-26 19:19:16 -0400874 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +1100875 spin_lock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
NeilBrown21c0d8f2006-10-04 02:16:16 -0700877 if (IS_ROOT(alias) &&
Nick Pigginda502952011-01-07 17:49:33 +1100878 (alias->d_flags & DCACHE_DISCONNECTED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 discon_alias = alias;
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500880 } else {
Nick Piggindc0474b2011-01-07 17:49:43 +1100881 __dget_dlock(alias);
Nick Pigginda502952011-01-07 17:49:33 +1100882 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 return alias;
884 }
885 }
Nick Pigginda502952011-01-07 17:49:33 +1100886 spin_unlock(&alias->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 }
Nick Pigginda502952011-01-07 17:49:33 +1100888 if (discon_alias) {
889 alias = discon_alias;
890 spin_lock(&alias->d_lock);
891 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -0500892 __dget_dlock(alias);
893 spin_unlock(&alias->d_lock);
894 return alias;
Nick Pigginda502952011-01-07 17:49:33 +1100895 }
896 spin_unlock(&alias->d_lock);
897 goto again;
898 }
899 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900}
901
Nick Pigginda502952011-01-07 17:49:33 +1100902struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
David Howells214fda12006-03-25 03:06:36 -0800904 struct dentry *de = NULL;
905
Al Virob3d9b7a2012-06-09 13:51:19 -0400906 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +1100907 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -0500908 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +1100909 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -0800910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 return de;
912}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700913EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914
915/*
916 * Try to kill dentries associated with this inode.
917 * WARNING: you must own a reference to inode.
918 */
919void d_prune_aliases(struct inode *inode)
920{
Domen Puncer0cdca3f2005-09-10 00:27:07 -0700921 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922restart:
Nick Piggin873feea2011-01-07 17:50:06 +1100923 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -0400924 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -0700926 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -0400927 struct dentry *parent = lock_parent(dentry);
928 if (likely(!dentry->d_lockref.count)) {
929 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +0800930 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -0400931 goto restart;
932 }
933 if (parent)
934 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 }
936 spin_unlock(&dentry->d_lock);
937 }
Nick Piggin873feea2011-01-07 17:50:06 +1100938 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700940EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Christoph Hellwig3049cfe2010-10-10 05:36:25 -0400942static void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943{
Al Viro5c47e6d2014-04-29 16:13:18 -0400944 struct dentry *dentry, *parent;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700945
Miklos Szeredi60942f22014-05-02 15:38:39 -0400946 while (!list_empty(list)) {
Al Viroff2fde92014-05-28 13:59:13 -0400947 struct inode *inode;
Miklos Szeredi60942f22014-05-02 15:38:39 -0400948 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +1100949 spin_lock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400950 parent = lock_parent(dentry);
951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 /*
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000953 * The dispose list is isolated and dentries are not accounted
954 * to the LRU here, so we can simply remove it from the list
955 * here regardless of whether it is referenced or not.
956 */
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400957 d_shrink_del(dentry);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000958
959 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 * We found an inuse dentry which was not removed from
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000961 * the LRU because of laziness during lookup. Do not free it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 */
Linus Torvalds360f5472015-01-09 15:19:03 -0800963 if (dentry->d_lockref.count > 0) {
Kentaro Makitada3bbdd2008-07-23 21:27:13 -0700964 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400965 if (parent)
966 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 continue;
968 }
Nick Piggin77812a12011-01-07 17:49:48 +1100969
Al Viro64fd72e2014-05-28 09:48:44 -0400970
971 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
972 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
973 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400974 if (parent)
975 spin_unlock(&parent->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -0400976 if (can_free)
977 dentry_free(dentry);
978 continue;
979 }
980
Al Viroff2fde92014-05-28 13:59:13 -0400981 inode = dentry->d_inode;
982 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400983 d_shrink_add(dentry, list);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000984 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400985 if (parent)
986 spin_unlock(&parent->d_lock);
Al Viro5c47e6d2014-04-29 16:13:18 -0400987 continue;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +1000988 }
Al Viroff2fde92014-05-28 13:59:13 -0400989
Al Viroff2fde92014-05-28 13:59:13 -0400990 __dentry_kill(dentry);
Al Viro046b9612014-05-29 08:54:52 -0400991
Al Viro5c47e6d2014-04-29 16:13:18 -0400992 /*
993 * We need to prune ancestors too. This is necessary to prevent
994 * quadratic behavior of shrink_dcache_parent(), but is also
995 * expected to be beneficial in reducing dentry cache
996 * fragmentation.
997 */
998 dentry = parent;
Al Virob2b80192014-05-29 09:11:45 -0400999 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1000 parent = lock_parent(dentry);
1001 if (dentry->d_lockref.count != 1) {
1002 dentry->d_lockref.count--;
1003 spin_unlock(&dentry->d_lock);
1004 if (parent)
1005 spin_unlock(&parent->d_lock);
1006 break;
1007 }
1008 inode = dentry->d_inode; /* can't be NULL */
1009 if (unlikely(!spin_trylock(&inode->i_lock))) {
1010 spin_unlock(&dentry->d_lock);
1011 if (parent)
1012 spin_unlock(&parent->d_lock);
1013 cpu_relax();
1014 continue;
1015 }
1016 __dentry_kill(dentry);
1017 dentry = parent;
1018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001020}
1021
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001022static enum lru_status dentry_lru_isolate(struct list_head *item,
1023 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001024{
1025 struct list_head *freeable = arg;
1026 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1027
1028
1029 /*
1030 * we are inverting the lru lock/dentry->d_lock here,
1031 * so use a trylock. If we fail to get the lock, just skip
1032 * it
1033 */
1034 if (!spin_trylock(&dentry->d_lock))
1035 return LRU_SKIP;
1036
1037 /*
1038 * Referenced dentries are still in use. If they have active
1039 * counts, just remove them from the LRU. Otherwise give them
1040 * another pass through the LRU.
1041 */
1042 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001043 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001044 spin_unlock(&dentry->d_lock);
1045 return LRU_REMOVED;
1046 }
1047
1048 if (dentry->d_flags & DCACHE_REFERENCED) {
1049 dentry->d_flags &= ~DCACHE_REFERENCED;
1050 spin_unlock(&dentry->d_lock);
1051
1052 /*
1053 * The list move itself will be made by the common LRU code. At
1054 * this point, we've dropped the dentry->d_lock but keep the
1055 * lru lock. This is safe to do, since every list movement is
1056 * protected by the lru lock even if both locks are held.
1057 *
1058 * This is guaranteed by the fact that all LRU management
1059 * functions are intermediated by the LRU API calls like
1060 * list_lru_add and list_lru_del. List movement in this file
1061 * only ever occur through this functions or through callbacks
1062 * like this one, that are called from the LRU API.
1063 *
1064 * The only exceptions to this are functions like
1065 * shrink_dentry_list, and code that first checks for the
1066 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1067 * operating only with stack provided lists after they are
1068 * properly isolated from the main list. It is thus, always a
1069 * local access.
1070 */
1071 return LRU_ROTATE;
1072 }
1073
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001074 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001075 spin_unlock(&dentry->d_lock);
1076
1077 return LRU_REMOVED;
1078}
1079
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001080/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001081 * prune_dcache_sb - shrink the dcache
1082 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001083 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001084 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001085 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1086 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001087 * function.
1088 *
1089 * This function may fail to free any resources if all the dentries are in
1090 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001091 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001092long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001093{
Dave Chinnerf6041562013-08-28 10:18:00 +10001094 LIST_HEAD(dispose);
1095 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001096
Vladimir Davydov503c3582015-02-12 14:58:47 -08001097 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1098 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001099 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001100 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101}
1102
Glauber Costa4e717f52013-08-28 10:18:03 +10001103static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001104 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001105{
Glauber Costa4e717f52013-08-28 10:18:03 +10001106 struct list_head *freeable = arg;
1107 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001108
Glauber Costa4e717f52013-08-28 10:18:03 +10001109 /*
1110 * we are inverting the lru lock/dentry->d_lock here,
1111 * so use a trylock. If we fail to get the lock, just skip
1112 * it
1113 */
1114 if (!spin_trylock(&dentry->d_lock))
1115 return LRU_SKIP;
1116
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001117 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001118 spin_unlock(&dentry->d_lock);
1119
1120 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001121}
1122
Glauber Costa4e717f52013-08-28 10:18:03 +10001123
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001124/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 * shrink_dcache_sb - shrink dcache for a superblock
1126 * @sb: superblock
1127 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001128 * Shrink the dcache for the specified super block. This is used to free
1129 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001131void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
Glauber Costa4e717f52013-08-28 10:18:03 +10001133 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001134
Glauber Costa4e717f52013-08-28 10:18:03 +10001135 do {
1136 LIST_HEAD(dispose);
1137
1138 freed = list_lru_walk(&sb->s_dentry_lru,
1139 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1140
1141 this_cpu_sub(nr_dentry_unused, freed);
1142 shrink_dentry_list(&dispose);
1143 } while (freed > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001145EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001148 * enum d_walk_ret - action to talke during tree walk
1149 * @D_WALK_CONTINUE: contrinue walk
1150 * @D_WALK_QUIT: quit walk
1151 * @D_WALK_NORETRY: quit when retry is needed
1152 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001154enum d_walk_ret {
1155 D_WALK_CONTINUE,
1156 D_WALK_QUIT,
1157 D_WALK_NORETRY,
1158 D_WALK_SKIP,
1159};
1160
1161/**
1162 * d_walk - walk the dentry tree
1163 * @parent: start of walk
1164 * @data: data passed to @enter() and @finish()
1165 * @enter: callback when first entering the dentry
1166 * @finish: callback when successfully finished the walk
1167 *
1168 * The @enter() and @finish() callbacks are called with d_lock held.
1169 */
1170static void d_walk(struct dentry *parent, void *data,
1171 enum d_walk_ret (*enter)(void *, struct dentry *),
1172 void (*finish)(void *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173{
Nick Piggin949854d2011-01-07 17:49:37 +11001174 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001176 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001177 enum d_walk_ret ret;
1178 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001179
Nick Piggin58db63d2011-01-07 17:49:39 +11001180again:
Al Viro48f5ec22013-09-09 15:22:25 -04001181 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001182 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001183 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001184
1185 ret = enter(data, this_parent);
1186 switch (ret) {
1187 case D_WALK_CONTINUE:
1188 break;
1189 case D_WALK_QUIT:
1190 case D_WALK_SKIP:
1191 goto out_unlock;
1192 case D_WALK_NORETRY:
1193 retry = false;
1194 break;
1195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196repeat:
1197 next = this_parent->d_subdirs.next;
1198resume:
1199 while (next != &this_parent->d_subdirs) {
1200 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001201 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001203
1204 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001205
1206 ret = enter(data, dentry);
1207 switch (ret) {
1208 case D_WALK_CONTINUE:
1209 break;
1210 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001211 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001212 goto out_unlock;
1213 case D_WALK_NORETRY:
1214 retry = false;
1215 break;
1216 case D_WALK_SKIP:
1217 spin_unlock(&dentry->d_lock);
1218 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001219 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001222 spin_unlock(&this_parent->d_lock);
1223 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001225 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 goto repeat;
1227 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001228 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 }
1230 /*
1231 * All done at this level ... ascend and resume the search.
1232 */
Al Viroca5358e2014-10-26 19:31:10 -04001233 rcu_read_lock();
1234ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001236 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001237 this_parent = child->d_parent;
1238
Al Viro31dec132013-10-25 17:04:27 -04001239 spin_unlock(&child->d_lock);
1240 spin_lock(&this_parent->d_lock);
1241
Al Viroca5358e2014-10-26 19:31:10 -04001242 /* might go back up the wrong parent if we have had a rename. */
1243 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001244 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001245 /* go into the first sibling still alive */
1246 do {
1247 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001248 if (next == &this_parent->d_subdirs)
1249 goto ascend;
1250 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001251 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001252 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 goto resume;
1254 }
Al Viroca5358e2014-10-26 19:31:10 -04001255 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001256 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001257 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001258 if (finish)
1259 finish(data);
1260
1261out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001262 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001263 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001264 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001265
1266rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001267 spin_unlock(&this_parent->d_lock);
1268 rcu_read_unlock();
1269 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001270 if (!retry)
1271 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001272 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001273 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001275
1276/*
1277 * Search for at least 1 mount point in the dentry's subdirs.
1278 * We descend to the next level whenever the d_subdirs
1279 * list is non-empty and continue searching.
1280 */
1281
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001282static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1283{
1284 int *ret = data;
1285 if (d_mountpoint(dentry)) {
1286 *ret = 1;
1287 return D_WALK_QUIT;
1288 }
1289 return D_WALK_CONTINUE;
1290}
1291
Randy Dunlap69c88dc2013-10-19 14:57:07 -07001292/**
1293 * have_submounts - check for mounts over a dentry
1294 * @parent: dentry to check.
1295 *
1296 * Return true if the parent or its subdirectories contain
1297 * a mount point
1298 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001299int have_submounts(struct dentry *parent)
1300{
1301 int ret = 0;
1302
1303 d_walk(parent, &ret, check_mount, NULL);
1304
1305 return ret;
1306}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001307EXPORT_SYMBOL(have_submounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
1309/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001310 * Called by mount code to set a mountpoint and check if the mountpoint is
1311 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1312 * subtree can become unreachable).
1313 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001314 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001315 * this reason take rename_lock and d_lock on dentry and ancestors.
1316 */
1317int d_set_mounted(struct dentry *dentry)
1318{
1319 struct dentry *p;
1320 int ret = -ENOENT;
1321 write_seqlock(&rename_lock);
1322 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001323 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001324 spin_lock(&p->d_lock);
1325 if (unlikely(d_unhashed(p))) {
1326 spin_unlock(&p->d_lock);
1327 goto out;
1328 }
1329 spin_unlock(&p->d_lock);
1330 }
1331 spin_lock(&dentry->d_lock);
1332 if (!d_unlinked(dentry)) {
1333 dentry->d_flags |= DCACHE_MOUNTED;
1334 ret = 0;
1335 }
1336 spin_unlock(&dentry->d_lock);
1337out:
1338 write_sequnlock(&rename_lock);
1339 return ret;
1340}
1341
1342/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001343 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 * and move any unused dentries to the end of the unused
1345 * list for prune_dcache(). We descend to the next level
1346 * whenever the d_subdirs list is non-empty and continue
1347 * searching.
1348 *
1349 * It returns zero iff there are no unused children,
1350 * otherwise it returns the number of children moved to
1351 * the end of the unused list. This may not be the total
1352 * number of unused children, because select_parent can
1353 * drop the lock and return early due to latency
1354 * constraints.
1355 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001356
1357struct select_data {
1358 struct dentry *start;
1359 struct list_head dispose;
1360 int found;
1361};
1362
1363static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001365 struct select_data *data = _data;
1366 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001368 if (data->start == dentry)
1369 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Al Virofe915222014-05-03 00:02:25 -04001371 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001372 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001373 } else {
1374 if (dentry->d_flags & DCACHE_LRU_LIST)
1375 d_lru_del(dentry);
1376 if (!dentry->d_lockref.count) {
1377 d_shrink_add(dentry, &data->dispose);
1378 data->found++;
1379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 }
1381 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001382 * We can return to the caller if we have found some (this
1383 * ensures forward progress). We'll be coming back to find
1384 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 */
Al Virofe915222014-05-03 00:02:25 -04001386 if (!list_empty(&data->dispose))
1387 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001389 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390}
1391
1392/**
1393 * shrink_dcache_parent - prune dcache
1394 * @parent: parent of entries to prune
1395 *
1396 * Prune the dcache to remove unused children of the parent dentry.
1397 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001398void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001400 for (;;) {
1401 struct select_data data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001403 INIT_LIST_HEAD(&data.dispose);
1404 data.start = parent;
1405 data.found = 0;
1406
1407 d_walk(parent, &data, select_collect, NULL);
1408 if (!data.found)
1409 break;
1410
1411 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001412 cond_resched();
1413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001415EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Al Viro9c8c10e2014-05-02 20:36:10 -04001417static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001418{
Al Viro9c8c10e2014-05-02 20:36:10 -04001419 /* it has busy descendents; complain about those instead */
1420 if (!list_empty(&dentry->d_subdirs))
1421 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001422
Al Viro9c8c10e2014-05-02 20:36:10 -04001423 /* root with refcount 1 is fine */
1424 if (dentry == _data && dentry->d_lockref.count == 1)
1425 return D_WALK_CONTINUE;
1426
1427 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1428 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001429 dentry,
1430 dentry->d_inode ?
1431 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001432 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001433 dentry->d_lockref.count,
1434 dentry->d_sb->s_type->name,
1435 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001436 WARN_ON(1);
1437 return D_WALK_CONTINUE;
1438}
1439
1440static void do_one_tree(struct dentry *dentry)
1441{
1442 shrink_dcache_parent(dentry);
1443 d_walk(dentry, dentry, umount_check, NULL);
1444 d_drop(dentry);
1445 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001446}
1447
1448/*
1449 * destroy the dentries attached to a superblock on unmounting
1450 */
1451void shrink_dcache_for_umount(struct super_block *sb)
1452{
1453 struct dentry *dentry;
1454
Al Viro9c8c10e2014-05-02 20:36:10 -04001455 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001456
1457 dentry = sb->s_root;
1458 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001459 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001460
1461 while (!hlist_bl_empty(&sb->s_anon)) {
Al Viro9c8c10e2014-05-02 20:36:10 -04001462 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1463 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001464 }
1465}
1466
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001467struct detach_data {
1468 struct select_data select;
1469 struct dentry *mountpoint;
1470};
1471static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001472{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001473 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001474
1475 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001476 __dget_dlock(dentry);
1477 data->mountpoint = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001478 return D_WALK_QUIT;
1479 }
1480
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001481 return select_collect(&data->select, dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001482}
1483
1484static void check_and_drop(void *_data)
1485{
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001486 struct detach_data *data = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001487
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001488 if (!data->mountpoint && !data->select.found)
1489 __d_drop(data->select.start);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001490}
1491
1492/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001493 * d_invalidate - detach submounts, prune dcache, and drop
1494 * @dentry: dentry to invalidate (aka detach, prune and drop)
1495 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001496 * no dcache lock.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001497 *
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001498 * The final d_drop is done as an atomic operation relative to
1499 * rename_lock ensuring there are no races with d_set_mounted. This
1500 * ensures there are no unhashed dentries on the path to a mountpoint.
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001501 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001502void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001503{
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001504 /*
1505 * If it's already been dropped, return OK.
1506 */
1507 spin_lock(&dentry->d_lock);
1508 if (d_unhashed(dentry)) {
1509 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001510 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001511 }
1512 spin_unlock(&dentry->d_lock);
1513
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001514 /* Negative dentries can be dropped without further checks */
1515 if (!dentry->d_inode) {
1516 d_drop(dentry);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001517 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001518 }
1519
1520 for (;;) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001521 struct detach_data data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001522
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001523 data.mountpoint = NULL;
1524 INIT_LIST_HEAD(&data.select.dispose);
1525 data.select.start = dentry;
1526 data.select.found = 0;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001527
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001528 d_walk(dentry, &data, detach_and_collect, check_and_drop);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001529
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001530 if (data.select.found)
1531 shrink_dentry_list(&data.select.dispose);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001532
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001533 if (data.mountpoint) {
1534 detach_mounts(data.mountpoint);
1535 dput(data.mountpoint);
1536 }
1537
1538 if (!data.mountpoint && !data.select.found)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001539 break;
1540
1541 cond_resched();
1542 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001543}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001544EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546/**
Al Viroa4464db2011-07-07 15:03:58 -04001547 * __d_alloc - allocate a dcache entry
1548 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 * @name: qstr of the name
1550 *
1551 * Allocates a dentry. It returns %NULL if there is insufficient memory
1552 * available. On a success the dentry is returned. The name passed in is
1553 * copied and the copy passed in may be reused after this call.
1554 */
1555
Al Viroa4464db2011-07-07 15:03:58 -04001556struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557{
1558 struct dentry *dentry;
1559 char *dname;
1560
Mel Gormane12ba742007-10-16 01:25:52 -07001561 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (!dentry)
1563 return NULL;
1564
Linus Torvalds6326c712012-05-21 16:14:04 -07001565 /*
1566 * We guarantee that the inline name is always NUL-terminated.
1567 * This way the memcpy() done by the name switching in rename
1568 * will still always have a NUL at the end, even if we might
1569 * be overwriting an internal NUL character
1570 */
1571 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001573 size_t size = offsetof(struct external_name, name[1]);
1574 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1575 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 kmem_cache_free(dentry_cache, dentry);
1577 return NULL;
1578 }
Al Viro8d85b482014-09-29 14:54:27 -04001579 atomic_set(&p->u.count, 1);
1580 dname = p->name;
Andrey Ryabinindf4c0e32015-02-13 14:39:45 -08001581 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1582 kasan_unpoison_shadow(dname,
1583 round_up(name->len + 1, sizeof(unsigned long)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 } else {
1585 dname = dentry->d_iname;
1586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588 dentry->d_name.len = name->len;
1589 dentry->d_name.hash = name->hash;
1590 memcpy(dname, name->name, name->len);
1591 dname[name->len] = 0;
1592
Linus Torvalds6326c712012-05-21 16:14:04 -07001593 /* Make sure we always see the terminating NUL character */
1594 smp_wmb();
1595 dentry->d_name.name = dname;
1596
Waiman Long98474232013-08-28 18:24:59 -07001597 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001598 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 spin_lock_init(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11001600 seqcount_init(&dentry->d_seq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001602 dentry->d_parent = dentry;
1603 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 dentry->d_op = NULL;
1605 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001606 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 INIT_LIST_HEAD(&dentry->d_lru);
1608 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001609 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1610 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001611 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
Nick Piggin3e880fb2011-01-07 17:49:19 +11001613 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001614
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 return dentry;
1616}
Al Viroa4464db2011-07-07 15:03:58 -04001617
1618/**
1619 * d_alloc - allocate a dcache entry
1620 * @parent: parent of entry to allocate
1621 * @name: qstr of the name
1622 *
1623 * Allocates a dentry. It returns %NULL if there is insufficient memory
1624 * available. On a success the dentry is returned. The name passed in is
1625 * copied and the copy passed in may be reused after this call.
1626 */
1627struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1628{
1629 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1630 if (!dentry)
1631 return NULL;
1632
1633 spin_lock(&parent->d_lock);
1634 /*
1635 * don't need child lock because it is not subject
1636 * to concurrency here
1637 */
1638 __dget_dlock(parent);
1639 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001640 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001641 spin_unlock(&parent->d_lock);
1642
1643 return dentry;
1644}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001645EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001647/**
1648 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1649 * @sb: the superblock
1650 * @name: qstr of the name
1651 *
1652 * For a filesystem that just pins its dentries in memory and never
1653 * performs lookups at all, return an unhashed IS_ROOT dentry.
1654 */
Nick Piggin4b936882011-01-07 17:50:07 +11001655struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1656{
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001657 return __d_alloc(sb, name);
Nick Piggin4b936882011-01-07 17:50:07 +11001658}
1659EXPORT_SYMBOL(d_alloc_pseudo);
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1662{
1663 struct qstr q;
1664
1665 q.name = name;
1666 q.len = strlen(name);
1667 q.hash = full_name_hash(q.name, q.len);
1668 return d_alloc(parent, &q);
1669}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001670EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Nick Pigginfb045ad2011-01-07 17:49:55 +11001672void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1673{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001674 WARN_ON_ONCE(dentry->d_op);
1675 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001676 DCACHE_OP_COMPARE |
1677 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001678 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001679 DCACHE_OP_DELETE |
1680 DCACHE_OP_SELECT_INODE));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001681 dentry->d_op = op;
1682 if (!op)
1683 return;
1684 if (op->d_hash)
1685 dentry->d_flags |= DCACHE_OP_HASH;
1686 if (op->d_compare)
1687 dentry->d_flags |= DCACHE_OP_COMPARE;
1688 if (op->d_revalidate)
1689 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001690 if (op->d_weak_revalidate)
1691 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001692 if (op->d_delete)
1693 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001694 if (op->d_prune)
1695 dentry->d_flags |= DCACHE_OP_PRUNE;
David Howells4bacc9c2015-06-18 14:32:31 +01001696 if (op->d_select_inode)
1697 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001698
1699}
1700EXPORT_SYMBOL(d_set_d_op);
1701
David Howellsdf1a0852015-01-29 12:02:28 +00001702
1703/*
1704 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1705 * @dentry - The dentry to mark
1706 *
1707 * Mark a dentry as falling through to the lower layer (as set with
1708 * d_pin_lower()). This flag may be recorded on the medium.
1709 */
1710void d_set_fallthru(struct dentry *dentry)
1711{
1712 spin_lock(&dentry->d_lock);
1713 dentry->d_flags |= DCACHE_FALLTHRU;
1714 spin_unlock(&dentry->d_lock);
1715}
1716EXPORT_SYMBOL(d_set_fallthru);
1717
David Howellsb18825a2013-09-12 19:22:53 +01001718static unsigned d_flags_for_inode(struct inode *inode)
1719{
David Howells44bdb5e2015-01-29 12:02:29 +00001720 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001721
1722 if (!inode)
1723 return DCACHE_MISS_TYPE;
1724
1725 if (S_ISDIR(inode->i_mode)) {
1726 add_flags = DCACHE_DIRECTORY_TYPE;
1727 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1728 if (unlikely(!inode->i_op->lookup))
1729 add_flags = DCACHE_AUTODIR_TYPE;
1730 else
1731 inode->i_opflags |= IOP_LOOKUP;
1732 }
David Howells44bdb5e2015-01-29 12:02:29 +00001733 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001734 }
1735
David Howells44bdb5e2015-01-29 12:02:29 +00001736 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001737 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001738 add_flags = DCACHE_SYMLINK_TYPE;
1739 goto type_determined;
1740 }
1741 inode->i_opflags |= IOP_NOFOLLOW;
1742 }
1743
1744 if (unlikely(!S_ISREG(inode->i_mode)))
1745 add_flags = DCACHE_SPECIAL_TYPE;
1746
1747type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001748 if (unlikely(IS_AUTOMOUNT(inode)))
1749 add_flags |= DCACHE_NEED_AUTOMOUNT;
1750 return add_flags;
1751}
1752
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001753static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1754{
David Howellsb18825a2013-09-12 19:22:53 +01001755 unsigned add_flags = d_flags_for_inode(inode);
1756
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001757 spin_lock(&dentry->d_lock);
David Howellsb18825a2013-09-12 19:22:53 +01001758 if (inode)
Al Viro946e51f2014-10-26 19:19:16 -04001759 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
David Howells4bf46a22015-03-05 14:09:22 +00001760 __d_set_inode_and_type(dentry, inode, add_flags);
Peter Zijlstraa7c6f572015-06-11 14:46:46 +02001761 dentry_rcuwalk_invalidate(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11001762 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001763 fsnotify_d_instantiate(dentry, inode);
1764}
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766/**
1767 * d_instantiate - fill in inode information for a dentry
1768 * @entry: dentry to complete
1769 * @inode: inode to attach to this dentry
1770 *
1771 * Fill in inode information in the entry.
1772 *
1773 * This turns negative dentries into productive full members
1774 * of society.
1775 *
1776 * NOTE! This assumes that the inode count has been incremented
1777 * (or otherwise set) by the caller to indicate that it is now
1778 * in use by the dcache.
1779 */
1780
1781void d_instantiate(struct dentry *entry, struct inode * inode)
1782{
Al Viro946e51f2014-10-26 19:19:16 -04001783 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Nick Piggin873feea2011-01-07 17:50:06 +11001784 if (inode)
1785 spin_lock(&inode->i_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001786 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001787 if (inode)
1788 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 security_d_instantiate(entry, inode);
1790}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001791EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
1793/**
1794 * d_instantiate_unique - instantiate a non-aliased dentry
1795 * @entry: dentry to instantiate
1796 * @inode: inode to attach to this dentry
1797 *
1798 * Fill in inode information in the entry. On success, it returns NULL.
1799 * If an unhashed alias of "entry" already exists, then we return the
Oleg Drokine866cfa2006-01-09 20:52:51 -08001800 * aliased dentry instead and drop one reference to inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 *
1802 * Note that in order to avoid conflicts with rename() etc, the caller
1803 * had better be holding the parent directory semaphore.
Oleg Drokine866cfa2006-01-09 20:52:51 -08001804 *
1805 * This also assumes that the inode count has been incremented
1806 * (or otherwise set) by the caller to indicate that it is now
1807 * in use by the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 */
David Howells770bfad2006-08-22 20:06:07 -04001809static struct dentry *__d_instantiate_unique(struct dentry *entry,
1810 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811{
1812 struct dentry *alias;
1813 int len = entry->d_name.len;
1814 const char *name = entry->d_name.name;
1815 unsigned int hash = entry->d_name.hash;
1816
David Howells770bfad2006-08-22 20:06:07 -04001817 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001818 __d_instantiate(entry, NULL);
David Howells770bfad2006-08-22 20:06:07 -04001819 return NULL;
1820 }
1821
Al Viro946e51f2014-10-26 19:19:16 -04001822 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Piggin9abca362011-01-07 17:49:36 +11001823 /*
1824 * Don't need alias->d_lock here, because aliases with
1825 * d_parent == entry->d_parent are not subject to name or
1826 * parent changes, because the parent inode i_mutex is held.
1827 */
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001828 if (alias->d_name.hash != hash)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 continue;
1830 if (alias->d_parent != entry->d_parent)
1831 continue;
Linus Torvaldsee983e82012-05-10 12:37:10 -07001832 if (alias->d_name.len != len)
1833 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07001834 if (dentry_cmp(alias, name, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 continue;
Nick Piggindc0474b2011-01-07 17:49:43 +11001836 __dget(alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 return alias;
1838 }
David Howells770bfad2006-08-22 20:06:07 -04001839
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001840 __d_instantiate(entry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 return NULL;
1842}
David Howells770bfad2006-08-22 20:06:07 -04001843
1844struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1845{
1846 struct dentry *result;
1847
Al Viro946e51f2014-10-26 19:19:16 -04001848 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
David Howells770bfad2006-08-22 20:06:07 -04001849
Nick Piggin873feea2011-01-07 17:50:06 +11001850 if (inode)
1851 spin_lock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001852 result = __d_instantiate_unique(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001853 if (inode)
1854 spin_unlock(&inode->i_lock);
David Howells770bfad2006-08-22 20:06:07 -04001855
1856 if (!result) {
1857 security_d_instantiate(entry, inode);
1858 return NULL;
1859 }
1860
1861 BUG_ON(!d_unhashed(result));
1862 iput(inode);
1863 return result;
1864}
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866EXPORT_SYMBOL(d_instantiate_unique);
1867
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001868/**
1869 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1870 * @entry: dentry to complete
1871 * @inode: inode to attach to this dentry
1872 *
1873 * Fill in inode information in the entry. If a directory alias is found, then
1874 * return an error (and drop inode). Together with d_materialise_unique() this
1875 * guarantees that a directory inode may never have more than one alias.
1876 */
1877int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1878{
Al Viro946e51f2014-10-26 19:19:16 -04001879 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Miklos Szeredib70a80e2013-10-01 16:44:54 +02001880
1881 spin_lock(&inode->i_lock);
1882 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1883 spin_unlock(&inode->i_lock);
1884 iput(inode);
1885 return -EBUSY;
1886 }
1887 __d_instantiate(entry, inode);
1888 spin_unlock(&inode->i_lock);
1889 security_d_instantiate(entry, inode);
1890
1891 return 0;
1892}
1893EXPORT_SYMBOL(d_instantiate_no_diralias);
1894
Al Viroadc0e912012-01-08 16:49:21 -05001895struct dentry *d_make_root(struct inode *root_inode)
1896{
1897 struct dentry *res = NULL;
1898
1899 if (root_inode) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07001900 static const struct qstr name = QSTR_INIT("/", 1);
Al Viroadc0e912012-01-08 16:49:21 -05001901
1902 res = __d_alloc(root_inode->i_sb, &name);
1903 if (res)
1904 d_instantiate(res, root_inode);
1905 else
1906 iput(root_inode);
1907 }
1908 return res;
1909}
1910EXPORT_SYMBOL(d_make_root);
1911
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001912static struct dentry * __d_find_any_alias(struct inode *inode)
1913{
1914 struct dentry *alias;
1915
Al Virob3d9b7a2012-06-09 13:51:19 -04001916 if (hlist_empty(&inode->i_dentry))
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001917 return NULL;
Al Viro946e51f2014-10-26 19:19:16 -04001918 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001919 __dget(alias);
1920 return alias;
1921}
1922
Sage Weil46f72b32012-01-10 09:04:37 -08001923/**
1924 * d_find_any_alias - find any alias for a given inode
1925 * @inode: inode to find an alias for
1926 *
1927 * If any aliases exist for the given inode, take and return a
1928 * reference for one of them. If no aliases exist, return %NULL.
1929 */
1930struct dentry *d_find_any_alias(struct inode *inode)
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001931{
1932 struct dentry *de;
1933
1934 spin_lock(&inode->i_lock);
1935 de = __d_find_any_alias(inode);
1936 spin_unlock(&inode->i_lock);
1937 return de;
1938}
Sage Weil46f72b32012-01-10 09:04:37 -08001939EXPORT_SYMBOL(d_find_any_alias);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001940
Fengguang Wu49c7dd22014-07-31 17:59:02 -04001941static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001942{
NeilBrownb911a6b2012-11-08 16:09:37 -08001943 static const struct qstr anonstring = QSTR_INIT("/", 1);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001944 struct dentry *tmp;
1945 struct dentry *res;
David Howellsb18825a2013-09-12 19:22:53 +01001946 unsigned add_flags;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001947
1948 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02001949 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001950 if (IS_ERR(inode))
1951 return ERR_CAST(inode);
1952
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001953 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001954 if (res)
1955 goto out_iput;
1956
Al Viroa4464db2011-07-07 15:03:58 -04001957 tmp = __d_alloc(inode->i_sb, &anonstring);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001958 if (!tmp) {
1959 res = ERR_PTR(-ENOMEM);
1960 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001961 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11001962
Nick Piggin873feea2011-01-07 17:50:06 +11001963 spin_lock(&inode->i_lock);
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05001964 res = __d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001965 if (res) {
Nick Piggin873feea2011-01-07 17:50:06 +11001966 spin_unlock(&inode->i_lock);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001967 dput(tmp);
1968 goto out_iput;
1969 }
1970
1971 /* attach a disconnected dentry */
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001972 add_flags = d_flags_for_inode(inode);
1973
1974 if (disconnected)
1975 add_flags |= DCACHE_DISCONNECTED;
David Howellsb18825a2013-09-12 19:22:53 +01001976
Christoph Hellwig9308a612008-08-11 15:49:12 +02001977 spin_lock(&tmp->d_lock);
David Howells4bf46a22015-03-05 14:09:22 +00001978 __d_set_inode_and_type(tmp, inode, add_flags);
Al Viro946e51f2014-10-26 19:19:16 -04001979 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001980 hlist_bl_lock(&tmp->d_sb->s_anon);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001981 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04001982 hlist_bl_unlock(&tmp->d_sb->s_anon);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001983 spin_unlock(&tmp->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11001984 spin_unlock(&inode->i_lock);
Josef Bacik24ff6662010-11-18 20:52:55 -05001985 security_d_instantiate(tmp, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001986
Christoph Hellwig9308a612008-08-11 15:49:12 +02001987 return tmp;
1988
1989 out_iput:
Josef Bacik24ff6662010-11-18 20:52:55 -05001990 if (res && !IS_ERR(res))
1991 security_d_instantiate(res, inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02001992 iput(inode);
1993 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02001994}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05001995
1996/**
1997 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1998 * @inode: inode to allocate the dentry for
1999 *
2000 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2001 * similar open by handle operations. The returned dentry may be anonymous,
2002 * or may have a full name (if the inode was already in the cache).
2003 *
2004 * When called on a directory inode, we must ensure that the inode only ever
2005 * has one dentry. If a dentry is found, that is returned instead of
2006 * allocating a new one.
2007 *
2008 * On successful return, the reference to the inode has been transferred
2009 * to the dentry. In case of an error the reference on the inode is released.
2010 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2011 * be passed in and the error will be propagated to the return value,
2012 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2013 */
2014struct dentry *d_obtain_alias(struct inode *inode)
2015{
2016 return __d_obtain_alias(inode, 1);
2017}
Benny Halevyadc48722009-02-27 14:02:59 -08002018EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
2020/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002021 * d_obtain_root - find or allocate a dentry for a given inode
2022 * @inode: inode to allocate the dentry for
2023 *
2024 * Obtain an IS_ROOT dentry for the root of a filesystem.
2025 *
2026 * We must ensure that directory inodes only ever have one dentry. If a
2027 * dentry is found, that is returned instead of allocating a new one.
2028 *
2029 * On successful return, the reference to the inode has been transferred
2030 * to the dentry. In case of an error the reference on the inode is
2031 * released. A %NULL or IS_ERR inode may be passed in and will be the
2032 * error will be propagate to the return value, with a %NULL @inode
2033 * replaced by ERR_PTR(-ESTALE).
2034 */
2035struct dentry *d_obtain_root(struct inode *inode)
2036{
2037 return __d_obtain_alias(inode, 0);
2038}
2039EXPORT_SYMBOL(d_obtain_root);
2040
2041/**
Barry Naujok94035402008-05-21 16:50:46 +10002042 * d_add_ci - lookup or allocate new dentry with case-exact name
2043 * @inode: the inode case-insensitive lookup has found
2044 * @dentry: the negative dentry that was passed to the parent's lookup func
2045 * @name: the case-exact name to be associated with the returned dentry
2046 *
2047 * This is to avoid filling the dcache with case-insensitive names to the
2048 * same inode, only the actual correct case is stored in the dcache for
2049 * case-insensitive filesystems.
2050 *
2051 * For a case-insensitive lookup match and if the the case-exact dentry
2052 * already exists in in the dcache, use it and return it.
2053 *
2054 * If no entry exists with the exact case name, allocate new dentry with
2055 * the exact case, and return the spliced entry.
2056 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002057struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002058 struct qstr *name)
2059{
Barry Naujok94035402008-05-21 16:50:46 +10002060 struct dentry *found;
2061 struct dentry *new;
2062
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002063 /*
2064 * First check if a dentry matching the name already exists,
2065 * if not go ahead and create it now.
2066 */
Barry Naujok94035402008-05-21 16:50:46 +10002067 found = d_hash_and_lookup(dentry->d_parent, name);
Barry Naujok94035402008-05-21 16:50:46 +10002068 if (!found) {
2069 new = d_alloc(dentry->d_parent, name);
2070 if (!new) {
Al Viro4f522a22013-02-11 23:20:37 -05002071 found = ERR_PTR(-ENOMEM);
Al Viro427c77d2014-10-12 18:46:26 -04002072 } else {
2073 found = d_splice_alias(inode, new);
2074 if (found) {
2075 dput(new);
2076 return found;
2077 }
2078 return new;
Barry Naujok94035402008-05-21 16:50:46 +10002079 }
Barry Naujok94035402008-05-21 16:50:46 +10002080 }
Barry Naujok94035402008-05-21 16:50:46 +10002081 iput(inode);
Al Viro4f522a22013-02-11 23:20:37 -05002082 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002083}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002084EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002086/*
2087 * Do the slow-case of the dentry name compare.
2088 *
2089 * Unlike the dentry_cmp() function, we need to atomically
Linus Torvaldsda53be12013-05-21 15:22:44 -07002090 * load the name and length information, so that the
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002091 * filesystem can rely on them, and can use the 'name' and
2092 * 'len' information without worrying about walking off the
2093 * end of memory etc.
2094 *
2095 * Thus the read_seqcount_retry() and the "duplicate" info
2096 * in arguments (the low-level filesystem should not look
2097 * at the dentry inode or name contents directly, since
2098 * rename can change them while we're in RCU mode).
2099 */
2100enum slow_d_compare {
2101 D_COMP_OK,
2102 D_COMP_NOMATCH,
2103 D_COMP_SEQRETRY,
2104};
2105
2106static noinline enum slow_d_compare slow_dentry_cmp(
2107 const struct dentry *parent,
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002108 struct dentry *dentry,
2109 unsigned int seq,
2110 const struct qstr *name)
2111{
2112 int tlen = dentry->d_name.len;
2113 const char *tname = dentry->d_name.name;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002114
2115 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2116 cpu_relax();
2117 return D_COMP_SEQRETRY;
2118 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002119 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002120 return D_COMP_NOMATCH;
2121 return D_COMP_OK;
2122}
2123
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002125 * __d_lookup_rcu - search for a dentry (racy, store-free)
2126 * @parent: parent dentry
2127 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002128 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002129 * Returns: dentry, or NULL
2130 *
2131 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2132 * resolution (store-free path walking) design described in
2133 * Documentation/filesystems/path-lookup.txt.
2134 *
2135 * This is not to be used outside core vfs.
2136 *
2137 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2138 * held, and rcu_read_lock held. The returned dentry must not be stored into
2139 * without taking d_lock and checking d_seq sequence count against @seq
2140 * returned here.
2141 *
Linus Torvalds15570082013-09-02 11:38:06 -07002142 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002143 * function.
2144 *
2145 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2146 * the returned dentry, so long as its parent's seqlock is checked after the
2147 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2148 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002149 *
2150 * NOTE! The caller *has* to check the resulting dentry against the sequence
2151 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002152 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002153struct dentry *__d_lookup_rcu(const struct dentry *parent,
2154 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002155 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002156{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002157 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002158 const unsigned char *str = name->name;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002159 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002160 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002161 struct dentry *dentry;
2162
2163 /*
2164 * Note: There is significant duplication with __d_lookup_rcu which is
2165 * required to prevent single threaded performance regressions
2166 * especially on architectures where smp_rmb (in seqcounts) are costly.
2167 * Keep the two functions in sync.
2168 */
2169
2170 /*
2171 * The hash list is protected using RCU.
2172 *
2173 * Carefully use d_seq when comparing a candidate dentry, to avoid
2174 * races with d_move().
2175 *
2176 * It is possible that concurrent renames can mess up our list
2177 * walk here and result in missing our dentry, resulting in the
2178 * false-negative result. d_lookup() protects against concurrent
2179 * renames using rename_lock seqlock.
2180 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002181 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002182 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002183 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002184 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002185
Nick Piggin31e6b012011-01-07 17:49:52 +11002186seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002187 /*
2188 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002189 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002190 *
2191 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002192 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002193 *
2194 * NOTE! We do a "raw" seqcount_begin here. That means that
2195 * we don't wait for the sequence count to stabilize if it
2196 * is in the middle of a sequence change. If we do the slow
2197 * dentry compare, we will do seqretries until it is stable,
2198 * and if we end up with a successful lookup, we actually
2199 * want to exit RCU lookup anyway.
2200 */
2201 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002202 if (dentry->d_parent != parent)
2203 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002204 if (d_unhashed(dentry))
2205 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002206
2207 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Linus Torvalds26fe5752012-05-10 13:14:12 -07002208 if (dentry->d_name.hash != hashlen_hash(hashlen))
2209 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002210 *seqp = seq;
2211 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002212 case D_COMP_OK:
2213 return dentry;
2214 case D_COMP_NOMATCH:
2215 continue;
2216 default:
2217 goto seqretry;
2218 }
2219 }
2220
Linus Torvalds26fe5752012-05-10 13:14:12 -07002221 if (dentry->d_name.hash_len != hashlen)
Linus Torvaldsee983e82012-05-10 12:37:10 -07002222 continue;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002223 *seqp = seq;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002224 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002225 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002226 }
2227 return NULL;
2228}
2229
2230/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 * d_lookup - search for a dentry
2232 * @parent: parent dentry
2233 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002234 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002236 * d_lookup searches the children of the parent dentry for the name in
2237 * question. If the dentry is found its reference count is incremented and the
2238 * dentry is returned. The caller must use dput to free the entry when it has
2239 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 */
Al Viroda2d8452013-01-24 18:29:34 -05002241struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242{
Nick Piggin31e6b012011-01-07 17:49:52 +11002243 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002244 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
Daeseok Younb8314f92014-08-11 11:46:53 +09002246 do {
2247 seq = read_seqbegin(&rename_lock);
2248 dentry = __d_lookup(parent, name);
2249 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250 break;
2251 } while (read_seqretry(&rename_lock, seq));
2252 return dentry;
2253}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002254EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255
Nick Piggin31e6b012011-01-07 17:49:52 +11002256/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002257 * __d_lookup - search for a dentry (racy)
2258 * @parent: parent dentry
2259 * @name: qstr of name we wish to find
2260 * Returns: dentry, or NULL
2261 *
2262 * __d_lookup is like d_lookup, however it may (rarely) return a
2263 * false-negative result due to unrelated rename activity.
2264 *
2265 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2266 * however it must be used carefully, eg. with a following d_lookup in
2267 * the case of failure.
2268 *
2269 * __d_lookup callers must be commented.
2270 */
Al Viroa713ca22013-01-24 18:27:00 -05002271struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272{
2273 unsigned int len = name->len;
2274 unsigned int hash = name->hash;
2275 const unsigned char *str = name->name;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002276 struct hlist_bl_head *b = d_hash(parent, hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002277 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002278 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002279 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
Nick Pigginb04f7842010-08-18 04:37:34 +10002281 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002282 * Note: There is significant duplication with __d_lookup_rcu which is
2283 * required to prevent single threaded performance regressions
2284 * especially on architectures where smp_rmb (in seqcounts) are costly.
2285 * Keep the two functions in sync.
2286 */
2287
2288 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002289 * The hash list is protected using RCU.
2290 *
2291 * Take d_lock when comparing a candidate dentry, to avoid races
2292 * with d_move().
2293 *
2294 * It is possible that concurrent renames can mess up our list
2295 * walk here and result in missing our dentry, resulting in the
2296 * false-negative result. d_lookup() protects against concurrent
2297 * renames using rename_lock seqlock.
2298 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002299 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002300 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301 rcu_read_lock();
2302
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002303 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 if (dentry->d_name.hash != hash)
2306 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307
2308 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 if (dentry->d_parent != parent)
2310 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002311 if (d_unhashed(dentry))
2312 goto next;
2313
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 /*
2315 * It is safe to compare names since d_move() cannot
2316 * change the qstr (protected by d_lock).
2317 */
Nick Pigginfb045ad2011-01-07 17:49:55 +11002318 if (parent->d_flags & DCACHE_OP_COMPARE) {
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002319 int tlen = dentry->d_name.len;
2320 const char *tname = dentry->d_name.name;
Linus Torvaldsda53be12013-05-21 15:22:44 -07002321 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 goto next;
2323 } else {
Linus Torvaldsee983e82012-05-10 12:37:10 -07002324 if (dentry->d_name.len != len)
2325 goto next;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002326 if (dentry_cmp(dentry, str, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 goto next;
2328 }
2329
Waiman Long98474232013-08-28 18:24:59 -07002330 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002331 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 spin_unlock(&dentry->d_lock);
2333 break;
2334next:
2335 spin_unlock(&dentry->d_lock);
2336 }
2337 rcu_read_unlock();
2338
2339 return found;
2340}
2341
2342/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002343 * d_hash_and_lookup - hash the qstr then search for a dentry
2344 * @dir: Directory to search in
2345 * @name: qstr of name we wish to find
2346 *
Al Viro4f522a22013-02-11 23:20:37 -05002347 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002348 */
2349struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2350{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002351 /*
2352 * Check for a fs-specific hash function. Note that we must
2353 * calculate the standard hash first, as the d_op->d_hash()
2354 * routine may choose to leave the hash value unchanged.
2355 */
2356 name->hash = full_name_hash(name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002357 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002358 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002359 if (unlikely(err < 0))
2360 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002361 }
Al Viro4f522a22013-02-11 23:20:37 -05002362 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002363}
Al Viro4f522a22013-02-11 23:20:37 -05002364EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002365
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366/*
2367 * When a file is deleted, we have two options:
2368 * - turn this dentry into a negative dentry
2369 * - unhash this dentry and free it.
2370 *
2371 * Usually, we want to just turn this into
2372 * a negative dentry, but if anybody else is
2373 * currently using the dentry or the inode
2374 * we can't do that and we fall back on removing
2375 * it from the hash queues and waiting for
2376 * it to be deleted later when it has no users
2377 */
2378
2379/**
2380 * d_delete - delete a dentry
2381 * @dentry: The dentry to delete
2382 *
2383 * Turn the dentry into a negative dentry if possible, otherwise
2384 * remove it from the hash queues so it can be deleted later
2385 */
2386
2387void d_delete(struct dentry * dentry)
2388{
Nick Piggin873feea2011-01-07 17:50:06 +11002389 struct inode *inode;
John McCutchan7a91bf72005-08-08 13:52:16 -04002390 int isdir = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 /*
2392 * Are we the only user?
2393 */
Nick Piggin357f8e62011-01-07 17:49:42 +11002394again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 spin_lock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +11002396 inode = dentry->d_inode;
2397 isdir = S_ISDIR(inode->i_mode);
Waiman Long98474232013-08-28 18:24:59 -07002398 if (dentry->d_lockref.count == 1) {
Alan Cox1fe0c022012-09-19 15:49:51 +01002399 if (!spin_trylock(&inode->i_lock)) {
Nick Piggin357f8e62011-01-07 17:49:42 +11002400 spin_unlock(&dentry->d_lock);
2401 cpu_relax();
2402 goto again;
2403 }
Al Viro13e3c5e2010-05-21 16:11:04 -04002404 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002405 dentry_unlink_inode(dentry);
John McCutchan7a91bf72005-08-08 13:52:16 -04002406 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 return;
2408 }
2409
2410 if (!d_unhashed(dentry))
2411 __d_drop(dentry);
2412
2413 spin_unlock(&dentry->d_lock);
John McCutchan7a91bf72005-08-08 13:52:16 -04002414
2415 fsnotify_nameremove(dentry, isdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002417EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002419static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420{
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002421 BUG_ON(!d_unhashed(entry));
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002422 hlist_bl_lock(b);
Linus Torvaldsdea36672011-04-24 07:58:46 -07002423 entry->d_flags |= DCACHE_RCUACCESS;
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002424 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002425 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426}
2427
David Howells770bfad2006-08-22 20:06:07 -04002428static void _d_rehash(struct dentry * entry)
2429{
2430 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2431}
2432
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433/**
2434 * d_rehash - add an entry back to the hash
2435 * @entry: dentry to add to the hash
2436 *
2437 * Adds a dentry to the hash according to its name.
2438 */
2439
2440void d_rehash(struct dentry * entry)
2441{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 spin_lock(&entry->d_lock);
David Howells770bfad2006-08-22 20:06:07 -04002443 _d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002446EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002448/**
2449 * dentry_update_name_case - update case insensitive dentry with a new name
2450 * @dentry: dentry to be updated
2451 * @name: new name
2452 *
2453 * Update a case insensitive dentry with new case of name.
2454 *
2455 * dentry must have been returned by d_lookup with name @name. Old and new
2456 * name lengths must match (ie. no d_compare which allows mismatched name
2457 * lengths).
2458 *
2459 * Parent inode i_mutex must be held over d_lookup and into this call (to
2460 * keep renames and concurrent inserts, and readdir(2) away).
2461 */
2462void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2463{
Linus Torvalds7ebfa572011-04-15 07:34:26 -07002464 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002465 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2466
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002467 spin_lock(&dentry->d_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +11002468 write_seqcount_begin(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002469 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
Nick Piggin31e6b012011-01-07 17:49:52 +11002470 write_seqcount_end(&dentry->d_seq);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002471 spin_unlock(&dentry->d_lock);
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002472}
2473EXPORT_SYMBOL(dentry_update_name_case);
2474
Al Viro8d85b482014-09-29 14:54:27 -04002475static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476{
Al Viro8d85b482014-09-29 14:54:27 -04002477 if (unlikely(dname_external(target))) {
2478 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 /*
2480 * Both external: swap the pointers
2481 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002482 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 } else {
2484 /*
2485 * dentry:internal, target:external. Steal target's
2486 * storage and make target internal.
2487 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002488 memcpy(target->d_iname, dentry->d_name.name,
2489 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 dentry->d_name.name = target->d_name.name;
2491 target->d_name.name = target->d_iname;
2492 }
2493 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002494 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 /*
2496 * dentry:external, target:internal. Give dentry's
2497 * storage to target and make dentry internal
2498 */
2499 memcpy(dentry->d_iname, target->d_name.name,
2500 target->d_name.len + 1);
2501 target->d_name.name = dentry->d_name.name;
2502 dentry->d_name.name = dentry->d_iname;
2503 } else {
2504 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002505 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002507 unsigned int i;
2508 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
Mikulas Patocka08d4f772014-09-05 12:16:01 -04002509 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2510 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002511 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2512 swap(((long *) &dentry->d_iname)[i],
2513 ((long *) &target->d_iname)[i]);
2514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 }
2516 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002517 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518}
2519
Al Viro8d85b482014-09-29 14:54:27 -04002520static void copy_name(struct dentry *dentry, struct dentry *target)
2521{
2522 struct external_name *old_name = NULL;
2523 if (unlikely(dname_external(dentry)))
2524 old_name = external_name(dentry);
2525 if (unlikely(dname_external(target))) {
2526 atomic_inc(&external_name(target)->u.count);
2527 dentry->d_name = target->d_name;
2528 } else {
2529 memcpy(dentry->d_iname, target->d_name.name,
2530 target->d_name.len + 1);
2531 dentry->d_name.name = dentry->d_iname;
2532 dentry->d_name.hash_len = target->d_name.hash_len;
2533 }
2534 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2535 kfree_rcu(old_name, u.head);
2536}
2537
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002538static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2539{
2540 /*
2541 * XXXX: do we really need to take target->d_lock?
2542 */
2543 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2544 spin_lock(&target->d_parent->d_lock);
2545 else {
2546 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2547 spin_lock(&dentry->d_parent->d_lock);
2548 spin_lock_nested(&target->d_parent->d_lock,
2549 DENTRY_D_LOCK_NESTED);
2550 } else {
2551 spin_lock(&target->d_parent->d_lock);
2552 spin_lock_nested(&dentry->d_parent->d_lock,
2553 DENTRY_D_LOCK_NESTED);
2554 }
2555 }
2556 if (target < dentry) {
2557 spin_lock_nested(&target->d_lock, 2);
2558 spin_lock_nested(&dentry->d_lock, 3);
2559 } else {
2560 spin_lock_nested(&dentry->d_lock, 2);
2561 spin_lock_nested(&target->d_lock, 3);
2562 }
2563}
2564
Al Viro986c0192014-09-26 23:11:15 -04002565static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002566{
2567 if (target->d_parent != dentry->d_parent)
2568 spin_unlock(&dentry->d_parent->d_lock);
2569 if (target->d_parent != target)
2570 spin_unlock(&target->d_parent->d_lock);
Al Viro986c0192014-09-26 23:11:15 -04002571 spin_unlock(&target->d_lock);
2572 spin_unlock(&dentry->d_lock);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002573}
2574
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575/*
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002576 * When switching names, the actual string doesn't strictly have to
2577 * be preserved in the target - because we're dropping the target
2578 * anyway. As such, we can just do a simple memcpy() to copy over
Mikhail Efremovd2fa4a82014-09-24 22:14:33 +04002579 * the new name before we switch, unless we are going to rehash
2580 * it. Note that if we *do* unhash the target, we are not allowed
2581 * to rehash it without giving it a new name/hash key - whether
2582 * we swap or overwrite the names here, resulting name won't match
2583 * the reality in filesystem; it's only there for d_path() purposes.
2584 * Note that all of this is happening under rename_lock, so the
2585 * any hash lookup seeing it in the middle of manipulations will
2586 * be discarded anyway. So we do not care what happens to the hash
2587 * key in that case.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002589/*
Al Viro18367502011-07-12 21:42:24 -04002590 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 * @dentry: entry to move
2592 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002593 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 *
2595 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002596 * dcache entries should not be moved in this way. Caller must hold
2597 * rename_lock, the i_mutex of the source and target directories,
2598 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002600static void __d_move(struct dentry *dentry, struct dentry *target,
2601 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 if (!dentry->d_inode)
2604 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2605
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002606 BUG_ON(d_ancestor(dentry, target));
2607 BUG_ON(d_ancestor(target, dentry));
2608
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002609 dentry_lock_for_move(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
Nick Piggin31e6b012011-01-07 17:49:52 +11002611 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002612 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002613
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002614 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2615
2616 /*
2617 * Move the dentry to the target hash queue. Don't bother checking
2618 * for the same hash queue because of how unlikely it is.
2619 */
2620 __d_drop(dentry);
Nick Piggin789680d2011-01-07 17:49:30 +11002621 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Miklos Szeredida1ce062014-04-01 17:08:43 +02002623 /*
2624 * Unhash the target (d_delete() is not usable here). If exchanging
2625 * the two dentries, then rehash onto the other's hash queue.
2626 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 __d_drop(target);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002628 if (exchange) {
2629 __d_rehash(target,
2630 d_hash(dentry->d_parent, dentry->d_name.hash));
2631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 /* Switch the names.. */
Al Viro8d85b482014-09-29 14:54:27 -04002634 if (exchange)
2635 swap_names(dentry, target);
2636 else
2637 copy_name(dentry, target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
Al Viro63cf4272014-09-26 23:06:14 -04002639 /* ... and switch them in the tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 if (IS_ROOT(dentry)) {
Al Viro63cf4272014-09-26 23:06:14 -04002641 /* splicing a tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 dentry->d_parent = target->d_parent;
2643 target->d_parent = target;
Al Viro946e51f2014-10-26 19:19:16 -04002644 list_del_init(&target->d_child);
2645 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 } else {
Al Viro63cf4272014-09-26 23:06:14 -04002647 /* swapping two dentries */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002648 swap(dentry->d_parent, target->d_parent);
Al Viro946e51f2014-10-26 19:19:16 -04002649 list_move(&target->d_child, &target->d_parent->d_subdirs);
2650 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
Al Viro63cf4272014-09-26 23:06:14 -04002651 if (exchange)
2652 fsnotify_d_move(target);
2653 fsnotify_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 }
2655
Nick Piggin31e6b012011-01-07 17:49:52 +11002656 write_seqcount_end(&target->d_seq);
2657 write_seqcount_end(&dentry->d_seq);
2658
Al Viro986c0192014-09-26 23:11:15 -04002659 dentry_unlock_for_move(dentry, target);
Al Viro18367502011-07-12 21:42:24 -04002660}
2661
2662/*
2663 * d_move - move a dentry
2664 * @dentry: entry to move
2665 * @target: new dentry
2666 *
2667 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002668 * dcache entries should not be moved in this way. See the locking
2669 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002670 */
2671void d_move(struct dentry *dentry, struct dentry *target)
2672{
2673 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002674 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002676}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002677EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678
Miklos Szeredida1ce062014-04-01 17:08:43 +02002679/*
2680 * d_exchange - exchange two dentries
2681 * @dentry1: first dentry
2682 * @dentry2: second dentry
2683 */
2684void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2685{
2686 write_seqlock(&rename_lock);
2687
2688 WARN_ON(!dentry1->d_inode);
2689 WARN_ON(!dentry2->d_inode);
2690 WARN_ON(IS_ROOT(dentry1));
2691 WARN_ON(IS_ROOT(dentry2));
2692
2693 __d_move(dentry1, dentry2, true);
2694
2695 write_sequnlock(&rename_lock);
2696}
2697
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002698/**
2699 * d_ancestor - search for an ancestor
2700 * @p1: ancestor dentry
2701 * @p2: child dentry
2702 *
2703 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2704 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07002705 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002706struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002707{
2708 struct dentry *p;
2709
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09002710 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07002711 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002712 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002713 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002714 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002715}
2716
2717/*
2718 * This helper attempts to cope with remotely renamed directories
2719 *
2720 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002721 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07002722 *
2723 * Note: If ever the locking in lock_rename() changes, then please
2724 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07002725 */
Al Virob5ae6b12014-10-12 22:16:02 -04002726static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11002727 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07002728{
2729 struct mutex *m1 = NULL, *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05002730 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002731
2732 /* If alias and dentry share a parent, then no extra locks required */
2733 if (alias->d_parent == dentry->d_parent)
2734 goto out_unalias;
2735
Trond Myklebust9eaef272006-10-21 10:24:20 -07002736 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07002737 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2738 goto out_err;
2739 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2740 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2741 goto out_err;
2742 m2 = &alias->d_parent->d_inode->i_mutex;
2743out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07002744 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04002745 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002746out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07002747 if (m2)
2748 mutex_unlock(m2);
2749 if (m1)
2750 mutex_unlock(m1);
2751 return ret;
2752}
2753
David Howells770bfad2006-08-22 20:06:07 -04002754/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002755 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2756 * @inode: the inode which may have a disconnected dentry
2757 * @dentry: a negative dentry which we want to point to the inode.
2758 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05002759 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2760 * place of the given dentry and return it, else simply d_add the inode
2761 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002762 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05002763 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2764 * we should error out: directories can't have multiple aliases.
2765 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002766 * This is needed in the lookup routine of any filesystem that is exportable
2767 * (via knfsd) so that we can build dcache paths to directories effectively.
2768 *
2769 * If a dentry was found and moved, then it is returned. Otherwise NULL
2770 * is returned. This matches the expected return value of ->lookup.
2771 *
2772 * Cluster filesystems may call this function with a negative, hashed dentry.
2773 * In that case, we know that the inode will be a regular file, and also this
2774 * will only occur during atomic_open. So we need to check for the dentry
2775 * being already hashed only in the final case.
2776 */
2777struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2778{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05002779 if (IS_ERR(inode))
2780 return ERR_CAST(inode);
2781
David Howells770bfad2006-08-22 20:06:07 -04002782 BUG_ON(!d_unhashed(dentry));
2783
David Howells770bfad2006-08-22 20:06:07 -04002784 if (!inode) {
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002785 __d_instantiate(dentry, NULL);
Al Virob5ae6b12014-10-12 22:16:02 -04002786 goto out;
David Howells770bfad2006-08-22 20:06:07 -04002787 }
Nick Piggin873feea2011-01-07 17:50:06 +11002788 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002789 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04002790 struct dentry *new = __d_find_any_alias(inode);
2791 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05002792 /* The reference to new ensures it remains an alias */
2793 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04002794 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002795 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04002796 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002797 dput(new);
2798 new = ERR_PTR(-ELOOP);
2799 pr_warn_ratelimited(
2800 "VFS: Lookup of '%s' in %s %s"
2801 " would have caused loop\n",
2802 dentry->d_name.name,
2803 inode->i_sb->s_type->name,
2804 inode->i_sb->s_id);
2805 } else if (!IS_ROOT(new)) {
2806 int err = __d_unalias(inode, dentry, new);
2807 write_sequnlock(&rename_lock);
2808 if (err) {
2809 dput(new);
2810 new = ERR_PTR(err);
2811 }
Al Viro18367502011-07-12 21:42:24 -04002812 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04002813 __d_move(new, dentry, false);
2814 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002815 security_d_instantiate(new, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002816 }
Al Virob5ae6b12014-10-12 22:16:02 -04002817 iput(inode);
2818 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07002819 }
David Howells770bfad2006-08-22 20:06:07 -04002820 }
Al Virob5ae6b12014-10-12 22:16:02 -04002821 /* already taking inode->i_lock, so d_add() by hand */
2822 __d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002823 spin_unlock(&inode->i_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04002824out:
2825 security_d_instantiate(dentry, inode);
2826 d_rehash(dentry);
2827 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04002828}
Al Virob5ae6b12014-10-12 22:16:02 -04002829EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04002830
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002831static int prepend(char **buffer, int *buflen, const char *str, int namelen)
Ram Pai6092d042008-03-27 13:06:20 +01002832{
2833 *buflen -= namelen;
2834 if (*buflen < 0)
2835 return -ENAMETOOLONG;
2836 *buffer -= namelen;
2837 memcpy(*buffer, str, namelen);
2838 return 0;
2839}
2840
Waiman Long232d2d62013-09-09 12:18:13 -04002841/**
2842 * prepend_name - prepend a pathname in front of current buffer pointer
Waiman Long18129972013-09-12 10:55:35 -04002843 * @buffer: buffer pointer
2844 * @buflen: allocated length of the buffer
2845 * @name: name string and length qstr structure
Waiman Long232d2d62013-09-09 12:18:13 -04002846 *
2847 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2848 * make sure that either the old or the new name pointer and length are
2849 * fetched. However, there may be mismatch between length and pointer.
2850 * The length cannot be trusted, we need to copy it byte-by-byte until
2851 * the length is reached or a null byte is found. It also prepends "/" at
2852 * the beginning of the name. The sequence number check at the caller will
2853 * retry it again when a d_move() does happen. So any garbage in the buffer
2854 * due to mismatched pointer and length will be discarded.
Al Viro6d13f692014-09-29 14:46:30 -04002855 *
2856 * Data dependency barrier is needed to make sure that we see that terminating
2857 * NUL. Alpha strikes again, film at 11...
Waiman Long232d2d62013-09-09 12:18:13 -04002858 */
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002859static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2860{
Waiman Long232d2d62013-09-09 12:18:13 -04002861 const char *dname = ACCESS_ONCE(name->name);
2862 u32 dlen = ACCESS_ONCE(name->len);
2863 char *p;
2864
Al Viro6d13f692014-09-29 14:46:30 -04002865 smp_read_barrier_depends();
2866
Waiman Long232d2d62013-09-09 12:18:13 -04002867 *buflen -= dlen + 1;
Al Viroe8251962014-03-23 00:28:40 -04002868 if (*buflen < 0)
2869 return -ENAMETOOLONG;
Waiman Long232d2d62013-09-09 12:18:13 -04002870 p = *buffer -= dlen + 1;
2871 *p++ = '/';
2872 while (dlen--) {
2873 char c = *dname++;
2874 if (!c)
2875 break;
2876 *p++ = c;
2877 }
2878 return 0;
Miklos Szeredicdd16d02008-06-23 18:11:53 +02002879}
2880
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881/**
Randy Dunlap208898c2010-11-18 15:02:49 -08002882 * prepend_path - Prepend path string to a buffer
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002883 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002884 * @root: root vfsmnt/dentry
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002885 * @buffer: pointer to the end of the buffer
2886 * @buflen: pointer to buffer length
2887 *
Waiman Long18129972013-09-12 10:55:35 -04002888 * The function will first try to write out the pathname without taking any
2889 * lock other than the RCU read lock to make sure that dentries won't go away.
2890 * It only checks the sequence number of the global rename_lock as any change
2891 * in the dentry's d_seq will be preceded by changes in the rename_lock
2892 * sequence number. If the sequence number had been changed, it will restart
2893 * the whole pathname back-tracing sequence again by taking the rename_lock.
2894 * In this case, there is no need to take the RCU read lock as the recursive
2895 * parent pointer references will keep the dentry chain alive as long as no
2896 * rename operation is performed.
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002897 */
Al Viro02125a82011-12-05 08:43:34 -05002898static int prepend_path(const struct path *path,
2899 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002900 char **buffer, int *buflen)
2901{
Al Viroede4ceb2013-11-13 07:45:40 -05002902 struct dentry *dentry;
2903 struct vfsmount *vfsmnt;
2904 struct mount *mnt;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002905 int error = 0;
Al Viro48a066e2013-09-29 22:06:07 -04002906 unsigned seq, m_seq = 0;
Waiman Long232d2d62013-09-09 12:18:13 -04002907 char *bptr;
2908 int blen;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002909
Al Viro48f5ec22013-09-09 15:22:25 -04002910 rcu_read_lock();
Al Viro48a066e2013-09-29 22:06:07 -04002911restart_mnt:
2912 read_seqbegin_or_lock(&mount_lock, &m_seq);
2913 seq = 0;
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002914 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04002915restart:
2916 bptr = *buffer;
2917 blen = *buflen;
Al Viro48a066e2013-09-29 22:06:07 -04002918 error = 0;
Al Viroede4ceb2013-11-13 07:45:40 -05002919 dentry = path->dentry;
2920 vfsmnt = path->mnt;
2921 mnt = real_mount(vfsmnt);
Waiman Long232d2d62013-09-09 12:18:13 -04002922 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002923 while (dentry != root->dentry || vfsmnt != root->mnt) {
2924 struct dentry * parent;
2925
2926 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
Al Viro48a066e2013-09-29 22:06:07 -04002927 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
Eric W. Biedermancde93be2015-08-15 13:36:12 -05002928 /* Escaped? */
2929 if (dentry != vfsmnt->mnt_root) {
2930 bptr = *buffer;
2931 blen = *buflen;
2932 error = 3;
2933 break;
2934 }
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002935 /* Global root? */
Al Viro48a066e2013-09-29 22:06:07 -04002936 if (mnt != parent) {
2937 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2938 mnt = parent;
Waiman Long232d2d62013-09-09 12:18:13 -04002939 vfsmnt = &mnt->mnt;
2940 continue;
2941 }
Waiman Long232d2d62013-09-09 12:18:13 -04002942 if (!error)
2943 error = is_mounted(vfsmnt) ? 1 : 2;
2944 break;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002945 }
2946 parent = dentry->d_parent;
2947 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04002948 error = prepend_name(&bptr, &blen, &dentry->d_name);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002949 if (error)
2950 break;
2951
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002952 dentry = parent;
2953 }
Al Viro48f5ec22013-09-09 15:22:25 -04002954 if (!(seq & 1))
2955 rcu_read_unlock();
2956 if (need_seqretry(&rename_lock, seq)) {
2957 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04002958 goto restart;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002959 }
Al Viro48f5ec22013-09-09 15:22:25 -04002960 done_seqretry(&rename_lock, seq);
Li Zhong4ec6c2a2013-11-13 15:21:51 +08002961
2962 if (!(m_seq & 1))
2963 rcu_read_unlock();
Al Viro48a066e2013-09-29 22:06:07 -04002964 if (need_seqretry(&mount_lock, m_seq)) {
2965 m_seq = 1;
2966 goto restart_mnt;
2967 }
2968 done_seqretry(&mount_lock, m_seq);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002969
Waiman Long232d2d62013-09-09 12:18:13 -04002970 if (error >= 0 && bptr == *buffer) {
2971 if (--blen < 0)
2972 error = -ENAMETOOLONG;
2973 else
2974 *--bptr = '/';
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002975 }
Waiman Long232d2d62013-09-09 12:18:13 -04002976 *buffer = bptr;
2977 *buflen = blen;
Al Viro7ea600b2013-03-26 18:25:57 -04002978 return error;
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002979}
2980
2981/**
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02002982 * __d_path - return the path of a dentry
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002983 * @path: the dentry/vfsmount to report
Al Viro02125a82011-12-05 08:43:34 -05002984 * @root: root vfsmnt/dentry
Randy Dunlapcd956a12010-08-14 13:05:31 -07002985 * @buf: buffer to return value in
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 * @buflen: buffer length
2987 *
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02002988 * Convert a dentry into an ASCII path name.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08002990 * Returns a pointer into the buffer or an error code if the
2991 * path was too long.
Linus Torvalds552ce542007-02-13 12:08:18 -08002992 *
Christoph Hellwigbe148242010-10-10 05:36:21 -04002993 * "buflen" should be positive.
Miklos Szeredi9d1bc6012008-03-27 13:06:21 +01002994 *
Al Viro02125a82011-12-05 08:43:34 -05002995 * If the path is not reachable from the supplied root, return %NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 */
Al Viro02125a82011-12-05 08:43:34 -05002997char *__d_path(const struct path *path,
2998 const struct path *root,
Miklos Szeredif2eb6572010-08-10 11:41:39 +02002999 char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000{
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003001 char *res = buf + buflen;
3002 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003004 prepend(&res, &buflen, "\0", 1);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003005 error = prepend_path(path, root, &res, &buflen);
Christoph Hellwigbe148242010-10-10 05:36:21 -04003006
Al Viro02125a82011-12-05 08:43:34 -05003007 if (error < 0)
3008 return ERR_PTR(error);
3009 if (error > 0)
3010 return NULL;
3011 return res;
3012}
3013
3014char *d_absolute_path(const struct path *path,
3015 char *buf, int buflen)
3016{
3017 struct path root = {};
3018 char *res = buf + buflen;
3019 int error;
3020
3021 prepend(&res, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003022 error = prepend_path(path, &root, &res, &buflen);
Al Viro02125a82011-12-05 08:43:34 -05003023
3024 if (error > 1)
3025 error = -EINVAL;
3026 if (error < 0)
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003027 return ERR_PTR(error);
Miklos Szeredif2eb6572010-08-10 11:41:39 +02003028 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029}
3030
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003031/*
3032 * same as __d_path but appends "(deleted)" for unlinked files.
3033 */
Al Viro02125a82011-12-05 08:43:34 -05003034static int path_with_deleted(const struct path *path,
3035 const struct path *root,
3036 char **buf, int *buflen)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003037{
3038 prepend(buf, buflen, "\0", 1);
3039 if (d_unlinked(path->dentry)) {
3040 int error = prepend(buf, buflen, " (deleted)", 10);
3041 if (error)
3042 return error;
3043 }
3044
3045 return prepend_path(path, root, buf, buflen);
3046}
3047
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003048static int prepend_unreachable(char **buffer, int *buflen)
3049{
3050 return prepend(buffer, buflen, "(unreachable)", 13);
3051}
3052
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003053static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3054{
3055 unsigned seq;
3056
3057 do {
3058 seq = read_seqcount_begin(&fs->seq);
3059 *root = fs->root;
3060 } while (read_seqcount_retry(&fs->seq, seq));
3061}
3062
Jan Bluncka03a8a702008-02-14 19:38:32 -08003063/**
3064 * d_path - return the path of a dentry
Jan Blunckcf28b482008-02-14 19:38:44 -08003065 * @path: path to report
Jan Bluncka03a8a702008-02-14 19:38:32 -08003066 * @buf: buffer to return value in
3067 * @buflen: buffer length
3068 *
3069 * Convert a dentry into an ASCII path name. If the entry has been deleted
3070 * the string " (deleted)" is appended. Note that this is ambiguous.
3071 *
Arjan van de Ven52afeef2008-12-01 14:35:00 -08003072 * Returns a pointer into the buffer or an error code if the path was
3073 * too long. Note: Callers should use the returned pointer, not the passed
3074 * in buffer, to use the name! The implementation often starts at an offset
3075 * into the buffer, and may leave 0 bytes at the start.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003076 *
Miklos Szeredi31f3e0b2008-06-23 18:11:52 +02003077 * "buflen" should be positive.
Jan Bluncka03a8a702008-02-14 19:38:32 -08003078 */
Jan Engelhardt20d4fdc2008-06-09 16:40:36 -07003079char *d_path(const struct path *path, char *buf, int buflen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080{
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003081 char *res = buf + buflen;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003082 struct path root;
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003083 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003085 /*
3086 * We have various synthetic filesystems that never get mounted. On
3087 * these filesystems dentries are never used for lookup purposes, and
3088 * thus don't need to be hashed. They also don't need a name until a
3089 * user wants to identify the object in /proc/pid/fd/. The little hack
3090 * below allows us to generate a name for these objects on demand:
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003091 *
3092 * Some pseudo inodes are mountable. When they are mounted
3093 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3094 * and instead have d_path return the mounted path.
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003095 */
Eric W. Biedermanf48cfdd2013-11-08 16:31:29 -08003096 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3097 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
Jan Blunckcf28b482008-02-14 19:38:44 -08003098 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003099
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003100 rcu_read_lock();
3101 get_fs_root_rcu(current->fs, &root);
Al Viro02125a82011-12-05 08:43:34 -05003102 error = path_with_deleted(path, &root, &res, &buflen);
Linus Torvalds68f0d9d2013-09-12 13:24:55 -07003103 rcu_read_unlock();
3104
Al Viro02125a82011-12-05 08:43:34 -05003105 if (error < 0)
Miklos Szerediffd1f4e2010-08-10 11:41:40 +02003106 res = ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 return res;
3108}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003109EXPORT_SYMBOL(d_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
3111/*
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003112 * Helper function for dentry_operations.d_dname() members
3113 */
3114char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3115 const char *fmt, ...)
3116{
3117 va_list args;
3118 char temp[64];
3119 int sz;
3120
3121 va_start(args, fmt);
3122 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3123 va_end(args);
3124
3125 if (sz > sizeof(temp) || sz > buflen)
3126 return ERR_PTR(-ENAMETOOLONG);
3127
3128 buffer += buflen - sz;
3129 return memcpy(buffer, temp, sz);
3130}
3131
Al Viro118b2302013-08-24 12:08:17 -04003132char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3133{
3134 char *end = buffer + buflen;
3135 /* these dentries are never renamed, so d_lock is not needed */
3136 if (prepend(&end, &buflen, " (deleted)", 11) ||
Waiman Long232d2d62013-09-09 12:18:13 -04003137 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
Al Viro118b2302013-08-24 12:08:17 -04003138 prepend(&end, &buflen, "/", 1))
3139 end = ERR_PTR(-ENAMETOOLONG);
Waiman Long232d2d62013-09-09 12:18:13 -04003140 return end;
Al Viro118b2302013-08-24 12:08:17 -04003141}
David Herrmann31bbe162014-01-03 14:09:47 +01003142EXPORT_SYMBOL(simple_dname);
Al Viro118b2302013-08-24 12:08:17 -04003143
Eric Dumazetc23fbb62007-05-08 00:26:18 -07003144/*
Ram Pai6092d042008-03-27 13:06:20 +01003145 * Write full pathname from the root of the filesystem into the buffer.
3146 */
Al Virof6500802014-01-26 12:37:55 -05003147static char *__dentry_path(struct dentry *d, char *buf, int buflen)
Ram Pai6092d042008-03-27 13:06:20 +01003148{
Al Virof6500802014-01-26 12:37:55 -05003149 struct dentry *dentry;
Waiman Long232d2d62013-09-09 12:18:13 -04003150 char *end, *retval;
3151 int len, seq = 0;
3152 int error = 0;
Ram Pai6092d042008-03-27 13:06:20 +01003153
Al Virof6500802014-01-26 12:37:55 -05003154 if (buflen < 2)
3155 goto Elong;
3156
Al Viro48f5ec22013-09-09 15:22:25 -04003157 rcu_read_lock();
Waiman Long232d2d62013-09-09 12:18:13 -04003158restart:
Al Virof6500802014-01-26 12:37:55 -05003159 dentry = d;
Waiman Long232d2d62013-09-09 12:18:13 -04003160 end = buf + buflen;
3161 len = buflen;
3162 prepend(&end, &len, "\0", 1);
Ram Pai6092d042008-03-27 13:06:20 +01003163 /* Get '/' right */
3164 retval = end-1;
3165 *retval = '/';
Waiman Long232d2d62013-09-09 12:18:13 -04003166 read_seqbegin_or_lock(&rename_lock, &seq);
Miklos Szeredicdd16d02008-06-23 18:11:53 +02003167 while (!IS_ROOT(dentry)) {
3168 struct dentry *parent = dentry->d_parent;
Ram Pai6092d042008-03-27 13:06:20 +01003169
Ram Pai6092d042008-03-27 13:06:20 +01003170 prefetch(parent);
Waiman Long232d2d62013-09-09 12:18:13 -04003171 error = prepend_name(&end, &len, &dentry->d_name);
3172 if (error)
3173 break;
Ram Pai6092d042008-03-27 13:06:20 +01003174
3175 retval = end;
3176 dentry = parent;
3177 }
Al Viro48f5ec22013-09-09 15:22:25 -04003178 if (!(seq & 1))
3179 rcu_read_unlock();
3180 if (need_seqretry(&rename_lock, seq)) {
3181 seq = 1;
Waiman Long232d2d62013-09-09 12:18:13 -04003182 goto restart;
Al Viro48f5ec22013-09-09 15:22:25 -04003183 }
3184 done_seqretry(&rename_lock, seq);
Waiman Long232d2d62013-09-09 12:18:13 -04003185 if (error)
3186 goto Elong;
Al Viroc1031352010-06-06 22:31:14 -04003187 return retval;
3188Elong:
3189 return ERR_PTR(-ENAMETOOLONG);
3190}
Nick Pigginec2447c2011-01-07 17:49:29 +11003191
3192char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3193{
Waiman Long232d2d62013-09-09 12:18:13 -04003194 return __dentry_path(dentry, buf, buflen);
Nick Pigginec2447c2011-01-07 17:49:29 +11003195}
3196EXPORT_SYMBOL(dentry_path_raw);
Al Viroc1031352010-06-06 22:31:14 -04003197
3198char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3199{
3200 char *p = NULL;
3201 char *retval;
3202
Al Viroc1031352010-06-06 22:31:14 -04003203 if (d_unlinked(dentry)) {
3204 p = buf + buflen;
3205 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3206 goto Elong;
3207 buflen++;
3208 }
3209 retval = __dentry_path(dentry, buf, buflen);
Al Viroc1031352010-06-06 22:31:14 -04003210 if (!IS_ERR(retval) && p)
3211 *p = '/'; /* restore '/' overriden with '\0' */
Ram Pai6092d042008-03-27 13:06:20 +01003212 return retval;
3213Elong:
Ram Pai6092d042008-03-27 13:06:20 +01003214 return ERR_PTR(-ENAMETOOLONG);
3215}
3216
Linus Torvalds8b19e342013-09-12 10:35:47 -07003217static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3218 struct path *pwd)
Linus Torvalds57624822013-09-12 10:12:47 -07003219{
Linus Torvalds8b19e342013-09-12 10:35:47 -07003220 unsigned seq;
3221
3222 do {
3223 seq = read_seqcount_begin(&fs->seq);
3224 *root = fs->root;
3225 *pwd = fs->pwd;
3226 } while (read_seqcount_retry(&fs->seq, seq));
Linus Torvalds57624822013-09-12 10:12:47 -07003227}
3228
Ram Pai6092d042008-03-27 13:06:20 +01003229/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230 * NOTE! The user-level library version returns a
3231 * character pointer. The kernel system call just
3232 * returns the length of the buffer filled (which
3233 * includes the ending '\0' character), or a negative
3234 * error value. So libc would do something like
3235 *
3236 * char *getcwd(char * buf, size_t size)
3237 * {
3238 * int retval;
3239 *
3240 * retval = sys_getcwd(buf, size);
3241 * if (retval >= 0)
3242 * return buf;
3243 * errno = -retval;
3244 * return NULL;
3245 * }
3246 */
Heiko Carstens3cdad422009-01-14 14:14:22 +01003247SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248{
Linus Torvalds552ce542007-02-13 12:08:18 -08003249 int error;
Jan Blunck6ac08c32008-02-14 19:34:38 -08003250 struct path pwd, root;
Linus Torvalds3272c542013-09-12 12:40:15 -07003251 char *page = __getname();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
3253 if (!page)
3254 return -ENOMEM;
3255
Linus Torvalds8b19e342013-09-12 10:35:47 -07003256 rcu_read_lock();
3257 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
Linus Torvalds552ce542007-02-13 12:08:18 -08003259 error = -ENOENT;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04003260 if (!d_unlinked(pwd.dentry)) {
Linus Torvalds552ce542007-02-13 12:08:18 -08003261 unsigned long len;
Linus Torvalds3272c542013-09-12 12:40:15 -07003262 char *cwd = page + PATH_MAX;
3263 int buflen = PATH_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003265 prepend(&cwd, &buflen, "\0", 1);
Al Viro02125a82011-12-05 08:43:34 -05003266 error = prepend_path(&pwd, &root, &cwd, &buflen);
Linus Torvaldsff812d72013-09-12 11:57:01 -07003267 rcu_read_unlock();
Linus Torvalds552ce542007-02-13 12:08:18 -08003268
Al Viro02125a82011-12-05 08:43:34 -05003269 if (error < 0)
Linus Torvalds552ce542007-02-13 12:08:18 -08003270 goto out;
3271
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003272 /* Unreachable from current root */
Al Viro02125a82011-12-05 08:43:34 -05003273 if (error > 0) {
Miklos Szeredi8df9d1a2010-08-10 11:41:41 +02003274 error = prepend_unreachable(&cwd, &buflen);
3275 if (error)
3276 goto out;
3277 }
3278
Linus Torvalds552ce542007-02-13 12:08:18 -08003279 error = -ERANGE;
Linus Torvalds3272c542013-09-12 12:40:15 -07003280 len = PATH_MAX + page - cwd;
Linus Torvalds552ce542007-02-13 12:08:18 -08003281 if (len <= size) {
3282 error = len;
3283 if (copy_to_user(buf, cwd, len))
3284 error = -EFAULT;
3285 }
Nick Piggin949854d2011-01-07 17:49:37 +11003286 } else {
Linus Torvaldsff812d72013-09-12 11:57:01 -07003287 rcu_read_unlock();
Nick Piggin949854d2011-01-07 17:49:37 +11003288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
3290out:
Linus Torvalds3272c542013-09-12 12:40:15 -07003291 __putname(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 return error;
3293}
3294
3295/*
3296 * Test whether new_dentry is a subdirectory of old_dentry.
3297 *
3298 * Trivially implemented using the dcache structure
3299 */
3300
3301/**
3302 * is_subdir - is new dentry a subdirectory of old_dentry
3303 * @new_dentry: new dentry
3304 * @old_dentry: old dentry
3305 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003306 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3307 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3309 */
3310
Yaowei Baia6e57872015-11-17 14:40:11 +08003311bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003312{
Yaowei Baia6e57872015-11-17 14:40:11 +08003313 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003314 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003315
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003316 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003317 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003318
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003319 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003322 /*
3323 * Need rcu_readlock to protect against the d_parent trashing
3324 * due to d_move
3325 */
3326 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003327 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003328 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003329 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003330 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003331 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333
3334 return result;
3335}
3336
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003337static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003339 struct dentry *root = data;
3340 if (dentry != root) {
3341 if (d_unhashed(dentry) || !dentry->d_inode)
3342 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003344 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3345 dentry->d_flags |= DCACHE_GENOCIDE;
3346 dentry->d_lockref.count--;
3347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003349 return D_WALK_CONTINUE;
3350}
Nick Piggin58db63d2011-01-07 17:49:39 +11003351
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003352void d_genocide(struct dentry *parent)
3353{
3354 d_walk(parent, parent, d_genocide_kill, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355}
3356
Al Viro60545d02013-06-07 01:20:27 -04003357void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358{
Al Viro60545d02013-06-07 01:20:27 -04003359 inode_dec_link_count(inode);
3360 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003361 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003362 !d_unlinked(dentry));
3363 spin_lock(&dentry->d_parent->d_lock);
3364 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3365 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3366 (unsigned long long)inode->i_ino);
3367 spin_unlock(&dentry->d_lock);
3368 spin_unlock(&dentry->d_parent->d_lock);
3369 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370}
Al Viro60545d02013-06-07 01:20:27 -04003371EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372
3373static __initdata unsigned long dhash_entries;
3374static int __init set_dhash_entries(char *str)
3375{
3376 if (!str)
3377 return 0;
3378 dhash_entries = simple_strtoul(str, &str, 0);
3379 return 1;
3380}
3381__setup("dhash_entries=", set_dhash_entries);
3382
3383static void __init dcache_init_early(void)
3384{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003385 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386
3387 /* If hashes are distributed across NUMA nodes, defer
3388 * hash allocation until vmalloc space is available.
3389 */
3390 if (hashdist)
3391 return;
3392
3393 dentry_hashtable =
3394 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003395 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003396 dhash_entries,
3397 13,
3398 HASH_EARLY,
3399 &d_hash_shift,
3400 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003401 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 0);
3403
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003404 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003405 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406}
3407
Denis Cheng74bf17c2007-10-16 23:26:30 -07003408static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409{
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003410 unsigned int loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411
3412 /*
3413 * A constructor could be added for stable state like the lists,
3414 * but it is probably not worth it because of the cache nature
3415 * of the dcache.
3416 */
Christoph Lameter0a31bd52007-05-06 14:49:57 -07003417 dentry_cache = KMEM_CACHE(dentry,
3418 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
3420 /* Hash may have been set up in dcache_init_early */
3421 if (!hashdist)
3422 return;
3423
3424 dentry_hashtable =
3425 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003426 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427 dhash_entries,
3428 13,
3429 0,
3430 &d_hash_shift,
3431 &d_hash_mask,
Tim Bird31fe62b2012-05-23 13:33:35 +00003432 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433 0);
3434
Dimitri Sivanich074b8512012-02-08 12:39:07 -08003435 for (loop = 0; loop < (1U << d_hash_shift); loop++)
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003436 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003437}
3438
3439/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003440struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003441EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443EXPORT_SYMBOL(d_genocide);
3444
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445void __init vfs_caches_init_early(void)
3446{
3447 dcache_init_early();
3448 inode_init_early();
3449}
3450
Mel Gorman4248b0d2015-08-06 15:46:20 -07003451void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
Paul Mundt20c2df82007-07-20 10:11:58 +09003454 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455
Denis Cheng74bf17c2007-10-16 23:26:30 -07003456 dcache_init();
3457 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003458 files_init();
3459 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003460 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461 bdev_cache_init();
3462 chrdev_init();
3463}