blob: c84269c6e8bf6df3b9e53ed42bb450bd500e2afe [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * fs/dcache.c
4 *
5 * Complete reimplementation
6 * (C) 1997 Thomas Schoebel-Theuer,
7 * with heavy changes by Linus Torvalds
8 */
9
10/*
11 * Notes on the allocation strategy:
12 *
13 * The dcache is a master of the icache - whenever a dcache entry
14 * exists, the inode will always exist. "iput()" is done either when
15 * the dcache entry is deleted or garbage collected.
16 */
17
Al Viro7a5cf792018-03-05 19:15:50 -050018#include <linux/ratelimit.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/string.h>
20#include <linux/mm.h>
21#include <linux/fs.h>
Eric Biggers0bf3d5c12019-03-20 11:39:11 -070022#include <linux/fscrypt.h>
John McCutchan7a91bf72005-08-08 13:52:16 -040023#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
25#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/hash.h>
27#include <linux/cache.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050028#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/security.h>
30#include <linux/seqlock.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -070031#include <linux/memblock.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110032#include <linux/bit_spinlock.h>
33#include <linux/rculist_bl.h>
Dave Chinnerf6041562013-08-28 10:18:00 +100034#include <linux/list_lru.h>
David Howells07f3f052006-09-30 20:52:18 +020035#include "internal.h"
Al Virob2dba1a2011-11-23 19:26:23 -050036#include "mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Nick Piggin789680d2011-01-07 17:49:30 +110038/*
39 * Usage:
Nick Piggin873feea2011-01-07 17:50:06 +110040 * dcache->d_inode->i_lock protects:
Al Viro946e51f2014-10-26 19:19:16 -040041 * - i_dentry, d_u.d_alias, d_inode of aliases
Nick Pigginceb5bdc2011-01-07 17:50:05 +110042 * dcache_hash_bucket lock protects:
43 * - the dcache hash table
NeilBrownf1ee6162017-12-21 09:45:40 +110044 * s_roots bl list spinlock protects:
45 * - the s_roots list (see __d_drop)
Dave Chinner19156842013-08-28 10:17:55 +100046 * dentry->d_sb->s_dentry_lru_lock protects:
Nick Piggin23044502011-01-07 17:49:31 +110047 * - the dcache lru lists and counters
48 * d_lock protects:
49 * - d_flags
50 * - d_name
51 * - d_lru
Nick Pigginb7ab39f2011-01-07 17:49:32 +110052 * - d_count
Nick Pigginda502952011-01-07 17:49:33 +110053 * - d_unhashed()
Nick Piggin2fd6b7f2011-01-07 17:49:34 +110054 * - d_parent and d_subdirs
55 * - childrens' d_child and d_parent
Al Viro946e51f2014-10-26 19:19:16 -040056 * - d_u.d_alias, d_inode
Nick Piggin789680d2011-01-07 17:49:30 +110057 *
58 * Ordering:
Nick Piggin873feea2011-01-07 17:50:06 +110059 * dentry->d_inode->i_lock
Nick Pigginb5c84bf2011-01-07 17:49:38 +110060 * dentry->d_lock
Dave Chinner19156842013-08-28 10:17:55 +100061 * dentry->d_sb->s_dentry_lru_lock
Nick Pigginceb5bdc2011-01-07 17:50:05 +110062 * dcache_hash_bucket lock
NeilBrownf1ee6162017-12-21 09:45:40 +110063 * s_roots lock
Nick Piggin789680d2011-01-07 17:49:30 +110064 *
Nick Pigginda502952011-01-07 17:49:33 +110065 * If there is an ancestor relationship:
66 * dentry->d_parent->...->d_parent->d_lock
67 * ...
68 * dentry->d_parent->d_lock
69 * dentry->d_lock
70 *
71 * If no ancestor relationship:
Al Viro076515f2018-03-10 23:15:52 -050072 * arbitrary, since it's serialized on rename_lock
Nick Piggin789680d2011-01-07 17:49:30 +110073 */
Eric Dumazetfa3536c2006-03-26 01:37:24 -080074int sysctl_vfs_cache_pressure __read_mostly = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
76
Al Viro74c3cbe2007-07-22 08:04:18 -040077__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Nick Piggin949854d2011-01-07 17:49:37 +110079EXPORT_SYMBOL(rename_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Christoph Lametere18b8902006-12-06 20:33:20 -080081static struct kmem_cache *dentry_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
David Howellscdf01222017-07-04 17:25:22 +010083const struct qstr empty_name = QSTR_INIT("", 0);
84EXPORT_SYMBOL(empty_name);
85const struct qstr slash_name = QSTR_INIT("/", 1);
86EXPORT_SYMBOL(slash_name);
Al Viro80e5d1ff52021-04-15 19:46:50 -040087const struct qstr dotdot_name = QSTR_INIT("..", 2);
88EXPORT_SYMBOL(dotdot_name);
David Howellscdf01222017-07-04 17:25:22 +010089
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/*
91 * This is the single most critical data structure when it comes
92 * to the dcache: the hashtable for lookups. Somebody should try
93 * to make this good - I've just made it work.
94 *
95 * This hash-function tries to avoid losing too many bits of hash
96 * information, yet avoid using a prime hash-size or similar.
97 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
Eric Dumazetfa3536c2006-03-26 01:37:24 -080099static unsigned int d_hash_shift __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100100
Linus Torvaldsb07ad992011-04-23 22:32:03 -0700101static struct hlist_bl_head *dentry_hashtable __read_mostly;
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100102
Linus Torvalds8387ff22016-06-10 07:51:30 -0700103static inline struct hlist_bl_head *d_hash(unsigned int hash)
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100104{
Alexey Dobriyan854d3e62017-11-20 18:05:07 +0300105 return dentry_hashtable + (hash >> d_hash_shift);
Nick Pigginceb5bdc2011-01-07 17:50:05 +1100106}
107
Al Viro94bdd652016-04-15 02:42:04 -0400108#define IN_LOOKUP_SHIFT 10
109static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
110
111static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
112 unsigned int hash)
113{
114 hash += (unsigned long) parent / L1_CACHE_BYTES;
115 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
116}
117
Luis Chamberlainc8c0c232022-01-21 22:12:59 -0800118struct dentry_stat_t {
119 long nr_dentry;
120 long nr_unused;
121 long age_limit; /* age in seconds */
122 long want_pages; /* pages requested by system */
123 long nr_negative; /* # of unused negative dentries */
124 long dummy; /* Reserved for future use */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125};
126
Glauber Costa3942c072013-08-28 10:17:53 +1000127static DEFINE_PER_CPU(long, nr_dentry);
Dave Chinner62d36c72013-08-28 10:17:54 +1000128static DEFINE_PER_CPU(long, nr_dentry_unused);
Waiman Longaf0c9af2019-01-30 13:52:38 -0500129static DEFINE_PER_CPU(long, nr_dentry_negative);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400130
131#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
Luis Chamberlainc8c0c232022-01-21 22:12:59 -0800132/* Statistics gathering. */
133static struct dentry_stat_t dentry_stat = {
134 .age_limit = 45,
135};
Dave Chinner62d36c72013-08-28 10:17:54 +1000136
137/*
138 * Here we resort to our own counters instead of using generic per-cpu counters
139 * for consistency with what the vfs inode code does. We are expected to harvest
140 * better code and performance by having our own specialized counters.
141 *
142 * Please note that the loop is done over all possible CPUs, not over all online
143 * CPUs. The reason for this is that we don't want to play games with CPUs going
144 * on and off. If one of them goes off, we will just keep their counters.
145 *
146 * glommer: See cffbc8a for details, and if you ever intend to change this,
147 * please update all vfs counters to match.
148 */
Glauber Costa3942c072013-08-28 10:17:53 +1000149static long get_nr_dentry(void)
Nick Piggin3e880fb2011-01-07 17:49:19 +1100150{
151 int i;
Glauber Costa3942c072013-08-28 10:17:53 +1000152 long sum = 0;
Nick Piggin3e880fb2011-01-07 17:49:19 +1100153 for_each_possible_cpu(i)
154 sum += per_cpu(nr_dentry, i);
155 return sum < 0 ? 0 : sum;
156}
157
Dave Chinner62d36c72013-08-28 10:17:54 +1000158static long get_nr_dentry_unused(void)
159{
160 int i;
161 long sum = 0;
162 for_each_possible_cpu(i)
163 sum += per_cpu(nr_dentry_unused, i);
164 return sum < 0 ? 0 : sum;
165}
166
Waiman Longaf0c9af2019-01-30 13:52:38 -0500167static long get_nr_dentry_negative(void)
168{
169 int i;
170 long sum = 0;
171
172 for_each_possible_cpu(i)
173 sum += per_cpu(nr_dentry_negative, i);
174 return sum < 0 ? 0 : sum;
175}
176
Luis Chamberlainc8c0c232022-01-21 22:12:59 -0800177static int proc_nr_dentry(struct ctl_table *table, int write, void *buffer,
178 size_t *lenp, loff_t *ppos)
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400179{
Nick Piggin3e880fb2011-01-07 17:49:19 +1100180 dentry_stat.nr_dentry = get_nr_dentry();
Dave Chinner62d36c72013-08-28 10:17:54 +1000181 dentry_stat.nr_unused = get_nr_dentry_unused();
Waiman Longaf0c9af2019-01-30 13:52:38 -0500182 dentry_stat.nr_negative = get_nr_dentry_negative();
Glauber Costa3942c072013-08-28 10:17:53 +1000183 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400184}
Luis Chamberlainc8c0c232022-01-21 22:12:59 -0800185
186static struct ctl_table fs_dcache_sysctls[] = {
187 {
188 .procname = "dentry-state",
189 .data = &dentry_stat,
190 .maxlen = 6*sizeof(long),
191 .mode = 0444,
192 .proc_handler = proc_nr_dentry,
193 },
194 { }
195};
196
197static int __init init_fs_dcache_sysctls(void)
198{
199 register_sysctl_init("fs", fs_dcache_sysctls);
200 return 0;
201}
202fs_initcall(init_fs_dcache_sysctls);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -0400203#endif
204
Linus Torvalds5483f182012-03-04 15:51:42 -0800205/*
206 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
207 * The strings are both count bytes long, and count is non-zero.
208 */
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700209#ifdef CONFIG_DCACHE_WORD_ACCESS
210
211#include <asm/word-at-a-time.h>
212/*
213 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
214 * aligned allocation for this particular component. We don't
215 * strictly need the load_unaligned_zeropad() safety, but it
216 * doesn't hurt either.
217 *
218 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
219 * need the careful unaligned handling.
220 */
Linus Torvalds94753db52012-05-10 12:19:19 -0700221static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvalds5483f182012-03-04 15:51:42 -0800222{
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800223 unsigned long a,b,mask;
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800224
225 for (;;) {
Andrey Ryabininbfe7aa62018-02-01 21:00:51 +0300226 a = read_word_at_a_time(cs);
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700227 b = load_unaligned_zeropad(ct);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800228 if (tcount < sizeof(unsigned long))
229 break;
230 if (unlikely(a != b))
231 return 1;
232 cs += sizeof(unsigned long);
233 ct += sizeof(unsigned long);
234 tcount -= sizeof(unsigned long);
235 if (!tcount)
236 return 0;
237 }
Will Deacona5c21dc2013-12-12 17:40:21 +0000238 mask = bytemask_from_count(tcount);
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800239 return unlikely(!!((a ^ b) & mask));
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700240}
241
Linus Torvaldsbfcfaa72012-03-06 11:16:17 -0800242#else
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700243
Linus Torvalds94753db52012-05-10 12:19:19 -0700244static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700245{
Linus Torvalds5483f182012-03-04 15:51:42 -0800246 do {
247 if (*cs != *ct)
248 return 1;
249 cs++;
250 ct++;
251 tcount--;
252 } while (tcount);
253 return 0;
254}
255
Linus Torvaldse419b4c2012-05-03 10:16:43 -0700256#endif
257
Linus Torvalds94753db52012-05-10 12:19:19 -0700258static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
259{
Linus Torvalds94753db52012-05-10 12:19:19 -0700260 /*
261 * Be careful about RCU walk racing with rename:
Will Deacon506458e2017-10-24 11:22:48 +0100262 * use 'READ_ONCE' to fetch the name pointer.
Linus Torvalds94753db52012-05-10 12:19:19 -0700263 *
264 * NOTE! Even if a rename will mean that the length
265 * was not loaded atomically, we don't care. The
266 * RCU walk will check the sequence count eventually,
267 * and catch it. And we won't overrun the buffer,
268 * because we're reading the name pointer atomically,
269 * and a dentry name is guaranteed to be properly
270 * terminated with a NUL byte.
271 *
272 * End result: even if 'len' is wrong, we'll exit
273 * early because the data cannot match (there can
274 * be no NUL in the ct/tcount data)
275 */
Will Deacon506458e2017-10-24 11:22:48 +0100276 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
He Kuangae0a8432016-03-26 09:12:10 +0000277
Linus Torvalds6326c71f2012-05-21 16:14:04 -0700278 return dentry_string_cmp(cs, ct, tcount);
Linus Torvalds94753db52012-05-10 12:19:19 -0700279}
280
Al Viro8d85b482014-09-29 14:54:27 -0400281struct external_name {
282 union {
283 atomic_t count;
284 struct rcu_head head;
285 } u;
286 unsigned char name[];
287};
288
289static inline struct external_name *external_name(struct dentry *dentry)
290{
291 return container_of(dentry->d_name.name, struct external_name, name[0]);
292}
293
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400294static void __d_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
Christoph Hellwig9c82ab92010-10-10 05:36:22 -0400296 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
297
Al Viro8d85b482014-09-29 14:54:27 -0400298 kmem_cache_free(dentry_cache, dentry);
299}
300
301static void __d_free_external(struct rcu_head *head)
302{
303 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Vlastimil Babka2e03b4b2018-10-26 15:05:41 -0700304 kfree(external_name(dentry));
Roman Gushchinf1782c92018-04-10 16:27:44 -0700305 kmem_cache_free(dentry_cache, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306}
307
Al Viro810bb172014-10-12 12:45:37 -0400308static inline int dname_external(const struct dentry *dentry)
309{
310 return dentry->d_name.name != dentry->d_iname;
311}
312
Al Viro49d31c22017-07-07 14:51:19 -0400313void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
314{
315 spin_lock(&dentry->d_lock);
Al Viro230c6402019-04-26 13:07:27 -0400316 name->name = dentry->d_name;
Al Viro49d31c22017-07-07 14:51:19 -0400317 if (unlikely(dname_external(dentry))) {
Al Viro230c6402019-04-26 13:07:27 -0400318 atomic_inc(&external_name(dentry)->u.count);
Al Viro49d31c22017-07-07 14:51:19 -0400319 } else {
Tetsuo Handa6cd00a02018-08-17 15:44:34 -0700320 memcpy(name->inline_name, dentry->d_iname,
321 dentry->d_name.len + 1);
Al Viro230c6402019-04-26 13:07:27 -0400322 name->name.name = name->inline_name;
Al Viro49d31c22017-07-07 14:51:19 -0400323 }
Al Viro230c6402019-04-26 13:07:27 -0400324 spin_unlock(&dentry->d_lock);
Al Viro49d31c22017-07-07 14:51:19 -0400325}
326EXPORT_SYMBOL(take_dentry_name_snapshot);
327
328void release_dentry_name_snapshot(struct name_snapshot *name)
329{
Al Viro230c6402019-04-26 13:07:27 -0400330 if (unlikely(name->name.name != name->inline_name)) {
Al Viro49d31c22017-07-07 14:51:19 -0400331 struct external_name *p;
Al Viro230c6402019-04-26 13:07:27 -0400332 p = container_of(name->name.name, struct external_name, name[0]);
Al Viro49d31c22017-07-07 14:51:19 -0400333 if (unlikely(atomic_dec_and_test(&p->u.count)))
Vlastimil Babka2e03b4b2018-10-26 15:05:41 -0700334 kfree_rcu(p, u.head);
Al Viro49d31c22017-07-07 14:51:19 -0400335 }
336}
337EXPORT_SYMBOL(release_dentry_name_snapshot);
338
David Howells4bf46a22015-03-05 14:09:22 +0000339static inline void __d_set_inode_and_type(struct dentry *dentry,
340 struct inode *inode,
341 unsigned type_flags)
342{
343 unsigned flags;
344
345 dentry->d_inode = inode;
David Howells4bf46a22015-03-05 14:09:22 +0000346 flags = READ_ONCE(dentry->d_flags);
347 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
348 flags |= type_flags;
Al Viro2fa6b1e2019-11-12 16:13:06 -0500349 smp_store_release(&dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000350}
351
David Howells4bf46a22015-03-05 14:09:22 +0000352static inline void __d_clear_type_and_inode(struct dentry *dentry)
353{
354 unsigned flags = READ_ONCE(dentry->d_flags);
355
356 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
357 WRITE_ONCE(dentry->d_flags, flags);
David Howells4bf46a22015-03-05 14:09:22 +0000358 dentry->d_inode = NULL;
Waiman Longaf0c9af2019-01-30 13:52:38 -0500359 if (dentry->d_flags & DCACHE_LRU_LIST)
360 this_cpu_inc(nr_dentry_negative);
David Howells4bf46a22015-03-05 14:09:22 +0000361}
362
Al Virob4f03542014-04-29 23:40:14 -0400363static void dentry_free(struct dentry *dentry)
364{
Al Viro946e51f2014-10-26 19:19:16 -0400365 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Al Viro8d85b482014-09-29 14:54:27 -0400366 if (unlikely(dname_external(dentry))) {
367 struct external_name *p = external_name(dentry);
368 if (likely(atomic_dec_and_test(&p->u.count))) {
369 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
370 return;
371 }
372 }
Al Virob4f03542014-04-29 23:40:14 -0400373 /* if dentry was never visible to RCU, immediate free is OK */
Al Viro5467a682019-03-15 22:23:19 -0400374 if (dentry->d_flags & DCACHE_NORCU)
Al Virob4f03542014-04-29 23:40:14 -0400375 __d_free(&dentry->d_u.d_rcu);
376 else
377 call_rcu(&dentry->d_u.d_rcu, __d_free);
378}
379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380/*
381 * Release the dentry's inode, using the filesystem
Al Viro550dce02016-05-29 20:13:30 -0400382 * d_iput() operation if defined.
Nick Piggin31e6b012011-01-07 17:49:52 +1100383 */
384static void dentry_unlink_inode(struct dentry * dentry)
385 __releases(dentry->d_lock)
Nick Piggin873feea2011-01-07 17:50:06 +1100386 __releases(dentry->d_inode->i_lock)
Nick Piggin31e6b012011-01-07 17:49:52 +1100387{
388 struct inode *inode = dentry->d_inode;
Al Viroa528aca2016-02-29 12:12:46 -0500389
Al Viro4c0d7cd2018-08-09 10:15:54 -0400390 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +0000391 __d_clear_type_and_inode(dentry);
Al Viro946e51f2014-10-26 19:19:16 -0400392 hlist_del_init(&dentry->d_u.d_alias);
Al Viro4c0d7cd2018-08-09 10:15:54 -0400393 raw_write_seqcount_end(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +1100394 spin_unlock(&dentry->d_lock);
Nick Piggin873feea2011-01-07 17:50:06 +1100395 spin_unlock(&inode->i_lock);
Nick Piggin31e6b012011-01-07 17:49:52 +1100396 if (!inode->i_nlink)
397 fsnotify_inoderemove(inode);
398 if (dentry->d_op && dentry->d_op->d_iput)
399 dentry->d_op->d_iput(dentry, inode);
400 else
401 iput(inode);
402}
403
404/*
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400405 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
406 * is in use - which includes both the "real" per-superblock
407 * LRU list _and_ the DCACHE_SHRINK_LIST use.
408 *
409 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
410 * on the shrink list (ie not on the superblock LRU list).
411 *
412 * The per-cpu "nr_dentry_unused" counters are updated with
413 * the DCACHE_LRU_LIST bit.
414 *
Waiman Longaf0c9af2019-01-30 13:52:38 -0500415 * The per-cpu "nr_dentry_negative" counters are only updated
416 * when deleted from or added to the per-superblock LRU list, not
417 * from/to the shrink list. That is to avoid an unneeded dec/inc
418 * pair when moving from LRU to shrink list in select_collect().
419 *
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400420 * These helper functions make sure we always follow the
421 * rules. d_lock must be held by the caller.
422 */
423#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
424static void d_lru_add(struct dentry *dentry)
425{
426 D_FLAG_VERIFY(dentry, 0);
427 dentry->d_flags |= DCACHE_LRU_LIST;
428 this_cpu_inc(nr_dentry_unused);
Waiman Longaf0c9af2019-01-30 13:52:38 -0500429 if (d_is_negative(dentry))
430 this_cpu_inc(nr_dentry_negative);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400431 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
432}
433
434static void d_lru_del(struct dentry *dentry)
435{
436 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
437 dentry->d_flags &= ~DCACHE_LRU_LIST;
438 this_cpu_dec(nr_dentry_unused);
Waiman Longaf0c9af2019-01-30 13:52:38 -0500439 if (d_is_negative(dentry))
440 this_cpu_dec(nr_dentry_negative);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400441 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
442}
443
444static void d_shrink_del(struct dentry *dentry)
445{
446 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
447 list_del_init(&dentry->d_lru);
448 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
449 this_cpu_dec(nr_dentry_unused);
450}
451
452static void d_shrink_add(struct dentry *dentry, struct list_head *list)
453{
454 D_FLAG_VERIFY(dentry, 0);
455 list_add(&dentry->d_lru, list);
456 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
457 this_cpu_inc(nr_dentry_unused);
458}
459
460/*
461 * These can only be called under the global LRU lock, ie during the
462 * callback for freeing the LRU list. "isolate" removes it from the
463 * LRU lists entirely, while shrink_move moves it to the indicated
464 * private list.
465 */
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800466static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400467{
468 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
469 dentry->d_flags &= ~DCACHE_LRU_LIST;
470 this_cpu_dec(nr_dentry_unused);
Waiman Longaf0c9af2019-01-30 13:52:38 -0500471 if (d_is_negative(dentry))
472 this_cpu_dec(nr_dentry_negative);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800473 list_lru_isolate(lru, &dentry->d_lru);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400474}
475
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800476static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
477 struct list_head *list)
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400478{
479 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
480 dentry->d_flags |= DCACHE_SHRINK_LIST;
Waiman Longaf0c9af2019-01-30 13:52:38 -0500481 if (d_is_negative(dentry))
482 this_cpu_dec(nr_dentry_negative);
Vladimir Davydov3f97b162015-02-12 14:59:35 -0800483 list_lru_isolate_move(lru, &dentry->d_lru, list);
Linus Torvalds89dc77b2013-09-13 22:55:10 -0400484}
485
NeilBrown61647822017-11-10 15:45:41 +1100486static void ___d_drop(struct dentry *dentry)
Nick Piggin789680d2011-01-07 17:49:30 +1100487{
Al Viro0632a9a2018-03-07 00:49:10 -0500488 struct hlist_bl_head *b;
489 /*
490 * Hashed dentries are normally on the dentry hashtable,
491 * with the exception of those newly allocated by
492 * d_obtain_root, which are always IS_ROOT:
493 */
494 if (unlikely(IS_ROOT(dentry)))
495 b = &dentry->d_sb->s_roots;
496 else
497 b = d_hash(dentry->d_name.hash);
Al Virob61625d2013-10-04 11:09:01 -0400498
Al Viro0632a9a2018-03-07 00:49:10 -0500499 hlist_bl_lock(b);
500 __hlist_bl_del(&dentry->d_hash);
501 hlist_bl_unlock(b);
Nick Piggin789680d2011-01-07 17:49:30 +1100502}
NeilBrown61647822017-11-10 15:45:41 +1100503
504void __d_drop(struct dentry *dentry)
505{
Al Viro0632a9a2018-03-07 00:49:10 -0500506 if (!d_unhashed(dentry)) {
507 ___d_drop(dentry);
508 dentry->d_hash.pprev = NULL;
509 write_seqcount_invalidate(&dentry->d_seq);
510 }
NeilBrown61647822017-11-10 15:45:41 +1100511}
Nick Piggin789680d2011-01-07 17:49:30 +1100512EXPORT_SYMBOL(__d_drop);
513
Mauro Carvalho Chehab961f3c82021-01-14 09:04:39 +0100514/**
515 * d_drop - drop a dentry
516 * @dentry: dentry to drop
517 *
518 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
519 * be found through a VFS lookup any more. Note that this is different from
520 * deleting the dentry - d_delete will try to mark the dentry negative if
521 * possible, giving a successful _negative_ lookup, while d_drop will
522 * just make the cache lookup fail.
523 *
524 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
525 * reason (NFS timeouts or autofs deletes).
526 *
527 * __d_drop requires dentry->d_lock
528 *
529 * ___d_drop doesn't mark dentry as "unhashed"
530 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
531 */
Nick Piggin789680d2011-01-07 17:49:30 +1100532void d_drop(struct dentry *dentry)
533{
Nick Piggin789680d2011-01-07 17:49:30 +1100534 spin_lock(&dentry->d_lock);
535 __d_drop(dentry);
536 spin_unlock(&dentry->d_lock);
Nick Piggin789680d2011-01-07 17:49:30 +1100537}
538EXPORT_SYMBOL(d_drop);
539
Al Viroba65dc52016-06-10 11:32:47 -0400540static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
541{
542 struct dentry *next;
543 /*
544 * Inform d_walk() and shrink_dentry_list() that we are no longer
545 * attached to the dentry tree
546 */
547 dentry->d_flags |= DCACHE_DENTRY_KILLED;
548 if (unlikely(list_empty(&dentry->d_child)))
549 return;
550 __list_del_entry(&dentry->d_child);
551 /*
552 * Cursors can move around the list of children. While we'd been
553 * a normal list member, it didn't matter - ->d_child.next would've
554 * been updated. However, from now on it won't be and for the
555 * things like d_walk() it might end up with a nasty surprise.
556 * Normally d_walk() doesn't care about cursors moving around -
557 * ->d_lock on parent prevents that and since a cursor has no children
558 * of its own, we get through it without ever unlocking the parent.
559 * There is one exception, though - if we ascend from a child that
560 * gets killed as soon as we unlock it, the next sibling is found
561 * using the value left in its ->d_child.next. And if _that_
562 * pointed to a cursor, and cursor got moved (e.g. by lseek())
563 * before d_walk() regains parent->d_lock, we'll end up skipping
564 * everything the cursor had been moved past.
565 *
566 * Solution: make sure that the pointer left behind in ->d_child.next
567 * points to something that won't be moving around. I.e. skip the
568 * cursors.
569 */
570 while (dentry->d_child.next != &parent->d_subdirs) {
571 next = list_entry(dentry->d_child.next, struct dentry, d_child);
572 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
573 break;
574 dentry->d_child.next = next->d_child.next;
575 }
576}
577
Al Viroe55fd012014-05-28 13:51:12 -0400578static void __dentry_kill(struct dentry *dentry)
Nick Piggin77812a12011-01-07 17:49:48 +1100579{
Al Viro41edf2782014-05-01 10:30:00 -0400580 struct dentry *parent = NULL;
581 bool can_free = true;
Al Viro41edf2782014-05-01 10:30:00 -0400582 if (!IS_ROOT(dentry))
Nick Piggin77812a12011-01-07 17:49:48 +1100583 parent = dentry->d_parent;
Nick Piggin31e6b012011-01-07 17:49:52 +1100584
Linus Torvalds0d984392013-09-08 13:46:52 -0700585 /*
586 * The dentry is now unrecoverably dead to the world.
587 */
588 lockref_mark_dead(&dentry->d_lockref);
589
Sage Weilf0023bc2011-10-28 10:02:42 -0700590 /*
Sage Weilf0023bc2011-10-28 10:02:42 -0700591 * inform the fs via d_prune that this dentry is about to be
592 * unhashed and destroyed.
593 */
Al Viro29266202014-05-30 11:39:02 -0400594 if (dentry->d_flags & DCACHE_OP_PRUNE)
Yan, Zheng61572bb2013-04-15 14:13:21 +0800595 dentry->d_op->d_prune(dentry);
596
Al Viro01b60352014-04-29 23:42:52 -0400597 if (dentry->d_flags & DCACHE_LRU_LIST) {
598 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
599 d_lru_del(dentry);
Al Viro01b60352014-04-29 23:42:52 -0400600 }
Nick Piggin77812a12011-01-07 17:49:48 +1100601 /* if it was on the hash then remove it */
602 __d_drop(dentry);
Al Viroba65dc52016-06-10 11:32:47 -0400603 dentry_unlist(dentry, parent);
Al Viro03b3b882014-04-29 15:45:28 -0400604 if (parent)
605 spin_unlock(&parent->d_lock);
Al Viro550dce02016-05-29 20:13:30 -0400606 if (dentry->d_inode)
607 dentry_unlink_inode(dentry);
608 else
609 spin_unlock(&dentry->d_lock);
Al Viro03b3b882014-04-29 15:45:28 -0400610 this_cpu_dec(nr_dentry);
611 if (dentry->d_op && dentry->d_op->d_release)
612 dentry->d_op->d_release(dentry);
613
Al Viro41edf2782014-05-01 10:30:00 -0400614 spin_lock(&dentry->d_lock);
615 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
616 dentry->d_flags |= DCACHE_MAY_FREE;
617 can_free = false;
618 }
619 spin_unlock(&dentry->d_lock);
Al Viro41edf2782014-05-01 10:30:00 -0400620 if (likely(can_free))
621 dentry_free(dentry);
Al Viro9c5f1d32018-04-15 18:28:48 -0400622 cond_resched();
Al Viroe55fd012014-05-28 13:51:12 -0400623}
624
Al Viro8b987a42018-02-23 22:11:34 -0500625static struct dentry *__lock_parent(struct dentry *dentry)
Al Viro046b9612014-05-29 08:54:52 -0400626{
Al Viro8b987a42018-02-23 22:11:34 -0500627 struct dentry *parent;
Al Viro046b9612014-05-29 08:54:52 -0400628 rcu_read_lock();
Al Viroc2338f22014-06-12 00:29:13 -0400629 spin_unlock(&dentry->d_lock);
Al Viro046b9612014-05-29 08:54:52 -0400630again:
Mark Rutland66702eb2017-10-23 14:07:14 -0700631 parent = READ_ONCE(dentry->d_parent);
Al Viro046b9612014-05-29 08:54:52 -0400632 spin_lock(&parent->d_lock);
633 /*
634 * We can't blindly lock dentry until we are sure
635 * that we won't violate the locking order.
636 * Any changes of dentry->d_parent must have
637 * been done with parent->d_lock held, so
638 * spin_lock() above is enough of a barrier
639 * for checking if it's still our child.
640 */
641 if (unlikely(parent != dentry->d_parent)) {
642 spin_unlock(&parent->d_lock);
643 goto again;
644 }
Al Viro3b821402018-02-23 20:47:17 -0500645 rcu_read_unlock();
Al Viro65d8eb52018-02-23 22:07:35 -0500646 if (parent != dentry)
647 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
648 else
649 parent = NULL;
Al Viro046b9612014-05-29 08:54:52 -0400650 return parent;
651}
652
Al Viro8b987a42018-02-23 22:11:34 -0500653static inline struct dentry *lock_parent(struct dentry *dentry)
654{
655 struct dentry *parent = dentry->d_parent;
656 if (IS_ROOT(dentry))
657 return NULL;
658 if (likely(spin_trylock(&parent->d_lock)))
659 return parent;
660 return __lock_parent(dentry);
661}
662
Al Viroa3385792018-02-23 21:07:24 -0500663static inline bool retain_dentry(struct dentry *dentry)
664{
665 WARN_ON(d_in_lookup(dentry));
666
667 /* Unreachable? Get rid of it */
668 if (unlikely(d_unhashed(dentry)))
669 return false;
670
671 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
672 return false;
673
674 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
675 if (dentry->d_op->d_delete(dentry))
676 return false;
677 }
Ira Weiny2c567af2020-04-30 07:41:37 -0700678
679 if (unlikely(dentry->d_flags & DCACHE_DONTCACHE))
680 return false;
681
Al Viro62d99562018-03-06 21:37:31 -0500682 /* retain; LRU fodder */
683 dentry->d_lockref.count--;
684 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
685 d_lru_add(dentry);
686 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
687 dentry->d_flags |= DCACHE_REFERENCED;
Al Viroa3385792018-02-23 21:07:24 -0500688 return true;
689}
690
Ira Weiny2c567af2020-04-30 07:41:37 -0700691void d_mark_dontcache(struct inode *inode)
692{
693 struct dentry *de;
694
695 spin_lock(&inode->i_lock);
696 hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
697 spin_lock(&de->d_lock);
698 de->d_flags |= DCACHE_DONTCACHE;
699 spin_unlock(&de->d_lock);
700 }
701 inode->i_state |= I_DONTCACHE;
702 spin_unlock(&inode->i_lock);
703}
704EXPORT_SYMBOL(d_mark_dontcache);
705
Linus Torvalds360f5472015-01-09 15:19:03 -0800706/*
John Ognessc1d0c1a2018-02-23 00:50:21 +0100707 * Finish off a dentry we've decided to kill.
708 * dentry->d_lock must be held, returns with it unlocked.
709 * Returns dentry requiring refcount drop, or NULL if we're done.
710 */
711static struct dentry *dentry_kill(struct dentry *dentry)
712 __releases(dentry->d_lock)
713{
714 struct inode *inode = dentry->d_inode;
715 struct dentry *parent = NULL;
716
717 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
Al Virof657a662018-02-23 21:25:42 -0500718 goto slow_positive;
John Ognessc1d0c1a2018-02-23 00:50:21 +0100719
720 if (!IS_ROOT(dentry)) {
721 parent = dentry->d_parent;
722 if (unlikely(!spin_trylock(&parent->d_lock))) {
Al Virof657a662018-02-23 21:25:42 -0500723 parent = __lock_parent(dentry);
724 if (likely(inode || !dentry->d_inode))
725 goto got_locks;
726 /* negative that became positive */
727 if (parent)
728 spin_unlock(&parent->d_lock);
729 inode = dentry->d_inode;
730 goto slow_positive;
John Ognessc1d0c1a2018-02-23 00:50:21 +0100731 }
732 }
John Ognessc1d0c1a2018-02-23 00:50:21 +0100733 __dentry_kill(dentry);
734 return parent;
735
Al Virof657a662018-02-23 21:25:42 -0500736slow_positive:
John Ognessc1d0c1a2018-02-23 00:50:21 +0100737 spin_unlock(&dentry->d_lock);
Al Virof657a662018-02-23 21:25:42 -0500738 spin_lock(&inode->i_lock);
739 spin_lock(&dentry->d_lock);
740 parent = lock_parent(dentry);
741got_locks:
742 if (unlikely(dentry->d_lockref.count != 1)) {
743 dentry->d_lockref.count--;
744 } else if (likely(!retain_dentry(dentry))) {
745 __dentry_kill(dentry);
746 return parent;
747 }
748 /* we are keeping it, after all */
749 if (inode)
750 spin_unlock(&inode->i_lock);
751 if (parent)
752 spin_unlock(&parent->d_lock);
753 spin_unlock(&dentry->d_lock);
754 return NULL;
John Ognessc1d0c1a2018-02-23 00:50:21 +0100755}
756
757/*
Linus Torvalds360f5472015-01-09 15:19:03 -0800758 * Try to do a lockless dput(), and return whether that was successful.
759 *
760 * If unsuccessful, we return false, having already taken the dentry lock.
761 *
762 * The caller needs to hold the RCU read lock, so that the dentry is
763 * guaranteed to stay around even if the refcount goes down to zero!
764 */
765static inline bool fast_dput(struct dentry *dentry)
766{
767 int ret;
768 unsigned int d_flags;
769
770 /*
771 * If we have a d_op->d_delete() operation, we sould not
Al Viro75a6f822015-07-08 02:42:38 +0100772 * let the dentry count go to zero, so use "put_or_lock".
Linus Torvalds360f5472015-01-09 15:19:03 -0800773 */
774 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
775 return lockref_put_or_lock(&dentry->d_lockref);
776
777 /*
778 * .. otherwise, we can try to just decrement the
779 * lockref optimistically.
780 */
781 ret = lockref_put_return(&dentry->d_lockref);
782
783 /*
784 * If the lockref_put_return() failed due to the lock being held
785 * by somebody else, the fast path has failed. We will need to
786 * get the lock, and then check the count again.
787 */
788 if (unlikely(ret < 0)) {
789 spin_lock(&dentry->d_lock);
790 if (dentry->d_lockref.count > 1) {
791 dentry->d_lockref.count--;
792 spin_unlock(&dentry->d_lock);
Gustavo A. R. Silva79644102018-08-01 19:39:05 -0500793 return true;
Linus Torvalds360f5472015-01-09 15:19:03 -0800794 }
Gustavo A. R. Silva79644102018-08-01 19:39:05 -0500795 return false;
Linus Torvalds360f5472015-01-09 15:19:03 -0800796 }
797
798 /*
799 * If we weren't the last ref, we're done.
800 */
801 if (ret)
Gustavo A. R. Silva79644102018-08-01 19:39:05 -0500802 return true;
Linus Torvalds360f5472015-01-09 15:19:03 -0800803
804 /*
805 * Careful, careful. The reference count went down
806 * to zero, but we don't hold the dentry lock, so
807 * somebody else could get it again, and do another
808 * dput(), and we need to not race with that.
809 *
810 * However, there is a very special and common case
811 * where we don't care, because there is nothing to
812 * do: the dentry is still hashed, it does not have
813 * a 'delete' op, and it's referenced and already on
814 * the LRU list.
815 *
816 * NOTE! Since we aren't locked, these values are
817 * not "stable". However, it is sufficient that at
818 * some point after we dropped the reference the
819 * dentry was hashed and the flags had the proper
820 * value. Other dentry users may have re-gotten
821 * a reference to the dentry and change that, but
822 * our work is done - we can leave the dentry
823 * around with a zero refcount.
Hao Li77573fa2020-12-08 10:10:50 +0800824 *
825 * Nevertheless, there are two cases that we should kill
826 * the dentry anyway.
827 * 1. free disconnected dentries as soon as their refcount
828 * reached zero.
829 * 2. free dentries if they should not be cached.
Linus Torvalds360f5472015-01-09 15:19:03 -0800830 */
831 smp_rmb();
Mark Rutland66702eb2017-10-23 14:07:14 -0700832 d_flags = READ_ONCE(dentry->d_flags);
Hao Li77573fa2020-12-08 10:10:50 +0800833 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
834 DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
Linus Torvalds360f5472015-01-09 15:19:03 -0800835
836 /* Nothing to do? Dropping the reference was all we needed? */
837 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
Gustavo A. R. Silva79644102018-08-01 19:39:05 -0500838 return true;
Linus Torvalds360f5472015-01-09 15:19:03 -0800839
840 /*
841 * Not the fast normal case? Get the lock. We've already decremented
842 * the refcount, but we'll need to re-check the situation after
843 * getting the lock.
844 */
845 spin_lock(&dentry->d_lock);
846
847 /*
848 * Did somebody else grab a reference to it in the meantime, and
849 * we're no longer the last user after all? Alternatively, somebody
850 * else could have killed it and marked it dead. Either way, we
851 * don't need to do anything else.
852 */
853 if (dentry->d_lockref.count) {
854 spin_unlock(&dentry->d_lock);
Gustavo A. R. Silva79644102018-08-01 19:39:05 -0500855 return true;
Linus Torvalds360f5472015-01-09 15:19:03 -0800856 }
857
858 /*
859 * Re-get the reference we optimistically dropped. We hold the
860 * lock, and we just tested that it was zero, so we can just
861 * set it to 1.
862 */
863 dentry->d_lockref.count = 1;
Gustavo A. R. Silva79644102018-08-01 19:39:05 -0500864 return false;
Linus Torvalds360f5472015-01-09 15:19:03 -0800865}
866
867
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868/*
869 * This is dput
870 *
871 * This is complicated by the fact that we do not want to put
872 * dentries that are no longer on any hash chain on the unused
873 * list: we'd much rather just get rid of them immediately.
874 *
875 * However, that implies that we have to traverse the dentry
876 * tree upwards to the parents which might _also_ now be
877 * scheduled for deletion (it may have been only waiting for
878 * its last child to go away).
879 *
880 * This tail recursion is done by hand as we don't want to depend
881 * on the compiler to always get this right (gcc generally doesn't).
882 * Real recursion would eat up our stack space.
883 */
884
885/*
886 * dput - release a dentry
887 * @dentry: dentry to release
888 *
889 * Release a dentry. This will drop the usage count and if appropriate
890 * call the dentry unlink method as well as removing it from the queues and
891 * releasing its resources. If the parent dentries were scheduled for release
892 * they too may now get deleted.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894void dput(struct dentry *dentry)
895{
Al Viro1088a642018-04-15 18:31:03 -0400896 while (dentry) {
897 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
Al Viro1088a642018-04-15 18:31:03 -0400899 rcu_read_lock();
900 if (likely(fast_dput(dentry))) {
901 rcu_read_unlock();
902 return;
903 }
Wei Fang47be6182016-07-06 11:32:20 +0800904
Al Viro1088a642018-04-15 18:31:03 -0400905 /* Slow case: now with the dentry lock held */
Linus Torvalds360f5472015-01-09 15:19:03 -0800906 rcu_read_unlock();
Linus Torvalds360f5472015-01-09 15:19:03 -0800907
Al Viro1088a642018-04-15 18:31:03 -0400908 if (likely(retain_dentry(dentry))) {
909 spin_unlock(&dentry->d_lock);
910 return;
911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Al Viro1088a642018-04-15 18:31:03 -0400913 dentry = dentry_kill(dentry);
Wei Fang47be6182016-07-06 11:32:20 +0800914 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -0700916EXPORT_SYMBOL(dput);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Al Viro9bdebc22019-06-29 18:31:24 -0400918static void __dput_to_list(struct dentry *dentry, struct list_head *list)
919__must_hold(&dentry->d_lock)
920{
921 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
922 /* let the owner of the list it's on deal with it */
923 --dentry->d_lockref.count;
924 } else {
925 if (dentry->d_flags & DCACHE_LRU_LIST)
926 d_lru_del(dentry);
927 if (!--dentry->d_lockref.count)
928 d_shrink_add(dentry, list);
929 }
930}
931
932void dput_to_list(struct dentry *dentry, struct list_head *list)
933{
934 rcu_read_lock();
935 if (likely(fast_dput(dentry))) {
936 rcu_read_unlock();
937 return;
938 }
939 rcu_read_unlock();
940 if (!retain_dentry(dentry))
941 __dput_to_list(dentry, list);
942 spin_unlock(&dentry->d_lock);
943}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Nick Pigginb5c84bf2011-01-07 17:49:38 +1100945/* This must be called with d_lock held */
Nick Piggindc0474b2011-01-07 17:49:43 +1100946static inline void __dget_dlock(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Waiman Long98474232013-08-28 18:24:59 -0700948 dentry->d_lockref.count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
Nick Piggindc0474b2011-01-07 17:49:43 +1100951static inline void __dget(struct dentry *dentry)
Nick Piggin23044502011-01-07 17:49:31 +1100952{
Waiman Long98474232013-08-28 18:24:59 -0700953 lockref_get(&dentry->d_lockref);
Nick Piggin23044502011-01-07 17:49:31 +1100954}
955
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100956struct dentry *dget_parent(struct dentry *dentry)
957{
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700958 int gotref;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100959 struct dentry *ret;
Al Viroe8400932019-10-31 01:43:31 -0400960 unsigned seq;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100961
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700962 /*
963 * Do optimistic parent lookup without any
964 * locking.
965 */
966 rcu_read_lock();
Al Viroe8400932019-10-31 01:43:31 -0400967 seq = raw_seqcount_begin(&dentry->d_seq);
Mark Rutland66702eb2017-10-23 14:07:14 -0700968 ret = READ_ONCE(dentry->d_parent);
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700969 gotref = lockref_get_not_zero(&ret->d_lockref);
970 rcu_read_unlock();
971 if (likely(gotref)) {
Al Viroe8400932019-10-31 01:43:31 -0400972 if (!read_seqcount_retry(&dentry->d_seq, seq))
Waiman Longdf3d0bb2013-09-02 11:29:22 -0700973 return ret;
974 dput(ret);
975 }
976
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100977repeat:
Nick Piggina734eb42011-01-07 17:49:44 +1100978 /*
979 * Don't need rcu_dereference because we re-check it was correct under
980 * the lock.
981 */
982 rcu_read_lock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100983 ret = dentry->d_parent;
Nick Piggina734eb42011-01-07 17:49:44 +1100984 spin_lock(&ret->d_lock);
985 if (unlikely(ret != dentry->d_parent)) {
986 spin_unlock(&ret->d_lock);
987 rcu_read_unlock();
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100988 goto repeat;
989 }
Nick Piggina734eb42011-01-07 17:49:44 +1100990 rcu_read_unlock();
Waiman Long98474232013-08-28 18:24:59 -0700991 BUG_ON(!ret->d_lockref.count);
992 ret->d_lockref.count++;
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100993 spin_unlock(&ret->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +1100994 return ret;
995}
996EXPORT_SYMBOL(dget_parent);
997
Al Viro61fec492018-04-25 10:52:25 -0400998static struct dentry * __d_find_any_alias(struct inode *inode)
999{
1000 struct dentry *alias;
1001
1002 if (hlist_empty(&inode->i_dentry))
1003 return NULL;
1004 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1005 __dget(alias);
1006 return alias;
1007}
1008
1009/**
1010 * d_find_any_alias - find any alias for a given inode
1011 * @inode: inode to find an alias for
1012 *
1013 * If any aliases exist for the given inode, take and return a
1014 * reference for one of them. If no aliases exist, return %NULL.
1015 */
1016struct dentry *d_find_any_alias(struct inode *inode)
1017{
1018 struct dentry *de;
1019
1020 spin_lock(&inode->i_lock);
1021 de = __d_find_any_alias(inode);
1022 spin_unlock(&inode->i_lock);
1023 return de;
1024}
1025EXPORT_SYMBOL(d_find_any_alias);
1026
J. Bruce Fields52ed46f2014-01-16 11:15:51 -05001027static struct dentry *__d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
Al Viro61fec492018-04-25 10:52:25 -04001029 struct dentry *alias;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Al Viro61fec492018-04-25 10:52:25 -04001031 if (S_ISDIR(inode->i_mode))
1032 return __d_find_any_alias(inode);
1033
Al Viro946e51f2014-10-26 19:19:16 -04001034 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
Nick Pigginda502952011-01-07 17:49:33 +11001035 spin_lock(&alias->d_lock);
Al Viro61fec492018-04-25 10:52:25 -04001036 if (!d_unhashed(alias)) {
J. Bruce Fields8d80d7d2014-01-16 17:17:31 -05001037 __dget_dlock(alias);
1038 spin_unlock(&alias->d_lock);
1039 return alias;
Nick Pigginda502952011-01-07 17:49:33 +11001040 }
1041 spin_unlock(&alias->d_lock);
Nick Pigginda502952011-01-07 17:49:33 +11001042 }
1043 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044}
1045
Mauro Carvalho Chehab961f3c82021-01-14 09:04:39 +01001046/**
1047 * d_find_alias - grab a hashed alias of inode
1048 * @inode: inode in question
1049 *
1050 * If inode has a hashed alias, or is a directory and has any alias,
1051 * acquire the reference to alias and return it. Otherwise return NULL.
1052 * Notice that if inode is a directory there can be only one alias and
1053 * it can be unhashed only if it has no children, or if it is the root
1054 * of a filesystem, or if the directory was renamed and d_revalidate
1055 * was the first vfs operation to notice.
1056 *
1057 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1058 * any other hashed alias over that one.
1059 */
Nick Pigginda502952011-01-07 17:49:33 +11001060struct dentry *d_find_alias(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
David Howells214fda12006-03-25 03:06:36 -08001062 struct dentry *de = NULL;
1063
Al Virob3d9b7a2012-06-09 13:51:19 -04001064 if (!hlist_empty(&inode->i_dentry)) {
Nick Piggin873feea2011-01-07 17:50:06 +11001065 spin_lock(&inode->i_lock);
J. Bruce Fields52ed46f2014-01-16 11:15:51 -05001066 de = __d_find_alias(inode);
Nick Piggin873feea2011-01-07 17:50:06 +11001067 spin_unlock(&inode->i_lock);
David Howells214fda12006-03-25 03:06:36 -08001068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 return de;
1070}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001071EXPORT_SYMBOL(d_find_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
1073/*
Al Virobca585d2021-01-05 14:13:52 -05001074 * Caller MUST be holding rcu_read_lock() and be guaranteed
1075 * that inode won't get freed until rcu_read_unlock().
1076 */
1077struct dentry *d_find_alias_rcu(struct inode *inode)
1078{
1079 struct hlist_head *l = &inode->i_dentry;
1080 struct dentry *de = NULL;
1081
1082 spin_lock(&inode->i_lock);
1083 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1084 // used without having I_FREEING set, which means no aliases left
1085 if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1086 if (S_ISDIR(inode->i_mode)) {
1087 de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1088 } else {
1089 hlist_for_each_entry(de, l, d_u.d_alias)
1090 if (!d_unhashed(de))
1091 break;
1092 }
1093 }
1094 spin_unlock(&inode->i_lock);
1095 return de;
1096}
1097
1098/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 * Try to kill dentries associated with this inode.
1100 * WARNING: you must own a reference to inode.
1101 */
1102void d_prune_aliases(struct inode *inode)
1103{
Domen Puncer0cdca3f2005-09-10 00:27:07 -07001104 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105restart:
Nick Piggin873feea2011-01-07 17:50:06 +11001106 spin_lock(&inode->i_lock);
Al Viro946e51f2014-10-26 19:19:16 -04001107 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 spin_lock(&dentry->d_lock);
Waiman Long98474232013-08-28 18:24:59 -07001109 if (!dentry->d_lockref.count) {
Al Viro29355c32014-05-30 11:25:30 -04001110 struct dentry *parent = lock_parent(dentry);
1111 if (likely(!dentry->d_lockref.count)) {
1112 __dentry_kill(dentry);
Yan, Zheng4a7795d2014-11-19 15:50:34 +08001113 dput(parent);
Al Viro29355c32014-05-30 11:25:30 -04001114 goto restart;
1115 }
1116 if (parent)
1117 spin_unlock(&parent->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
1119 spin_unlock(&dentry->d_lock);
1120 }
Nick Piggin873feea2011-01-07 17:50:06 +11001121 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001123EXPORT_SYMBOL(d_prune_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
Al Viro3b3f09f2018-02-23 21:54:18 -05001125/*
1126 * Lock a dentry from shrink list.
John Ogness8f04da22018-02-23 00:50:24 +01001127 * Called under rcu_read_lock() and dentry->d_lock; the former
1128 * guarantees that nothing we access will be freed under us.
Al Viro3b3f09f2018-02-23 21:54:18 -05001129 * Note that dentry is *not* protected from concurrent dentry_kill(),
John Ogness8f04da22018-02-23 00:50:24 +01001130 * d_delete(), etc.
1131 *
Al Viro3b3f09f2018-02-23 21:54:18 -05001132 * Return false if dentry has been disrupted or grabbed, leaving
1133 * the caller to kick it off-list. Otherwise, return true and have
1134 * that dentry's inode and parent both locked.
1135 */
1136static bool shrink_lock_dentry(struct dentry *dentry)
1137{
1138 struct inode *inode;
1139 struct dentry *parent;
1140
1141 if (dentry->d_lockref.count)
1142 return false;
1143
1144 inode = dentry->d_inode;
1145 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
Al Viro3b3f09f2018-02-23 21:54:18 -05001146 spin_unlock(&dentry->d_lock);
1147 spin_lock(&inode->i_lock);
1148 spin_lock(&dentry->d_lock);
1149 if (unlikely(dentry->d_lockref.count))
1150 goto out;
1151 /* changed inode means that somebody had grabbed it */
1152 if (unlikely(inode != dentry->d_inode))
1153 goto out;
Al Viro3b3f09f2018-02-23 21:54:18 -05001154 }
1155
1156 parent = dentry->d_parent;
1157 if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1158 return true;
1159
Al Viro3b3f09f2018-02-23 21:54:18 -05001160 spin_unlock(&dentry->d_lock);
Al Viro3b3f09f2018-02-23 21:54:18 -05001161 spin_lock(&parent->d_lock);
1162 if (unlikely(parent != dentry->d_parent)) {
1163 spin_unlock(&parent->d_lock);
1164 spin_lock(&dentry->d_lock);
1165 goto out;
1166 }
1167 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
John Ogness8f04da22018-02-23 00:50:24 +01001168 if (likely(!dentry->d_lockref.count))
Al Viro3b3f09f2018-02-23 21:54:18 -05001169 return true;
Al Viro3b3f09f2018-02-23 21:54:18 -05001170 spin_unlock(&parent->d_lock);
1171out:
1172 if (inode)
1173 spin_unlock(&inode->i_lock);
Al Viro3b3f09f2018-02-23 21:54:18 -05001174 return false;
1175}
1176
Al Viro9bdebc22019-06-29 18:31:24 -04001177void shrink_dentry_list(struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
Miklos Szeredi60942f22014-05-02 15:38:39 -04001179 while (!list_empty(list)) {
Al Viro3b3f09f2018-02-23 21:54:18 -05001180 struct dentry *dentry, *parent;
Al Viro3b3f09f2018-02-23 21:54:18 -05001181
Miklos Szeredi60942f22014-05-02 15:38:39 -04001182 dentry = list_entry(list->prev, struct dentry, d_lru);
Nick Pigginec336792011-01-07 17:49:47 +11001183 spin_lock(&dentry->d_lock);
John Ogness8f04da22018-02-23 00:50:24 +01001184 rcu_read_lock();
Al Viro3b3f09f2018-02-23 21:54:18 -05001185 if (!shrink_lock_dentry(dentry)) {
1186 bool can_free = false;
John Ogness8f04da22018-02-23 00:50:24 +01001187 rcu_read_unlock();
Al Viro3b3f09f2018-02-23 21:54:18 -05001188 d_shrink_del(dentry);
1189 if (dentry->d_lockref.count < 0)
1190 can_free = dentry->d_flags & DCACHE_MAY_FREE;
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001191 spin_unlock(&dentry->d_lock);
Al Viro64fd72e2014-05-28 09:48:44 -04001192 if (can_free)
1193 dentry_free(dentry);
1194 continue;
1195 }
John Ogness8f04da22018-02-23 00:50:24 +01001196 rcu_read_unlock();
Al Viro3b3f09f2018-02-23 21:54:18 -05001197 d_shrink_del(dentry);
1198 parent = dentry->d_parent;
Al Viro9bdebc22019-06-29 18:31:24 -04001199 if (parent != dentry)
1200 __dput_to_list(parent, list);
Al Viroff2fde92014-05-28 13:59:13 -04001201 __dentry_kill(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 }
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001203}
1204
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001205static enum lru_status dentry_lru_isolate(struct list_head *item,
1206 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerf6041562013-08-28 10:18:00 +10001207{
1208 struct list_head *freeable = arg;
1209 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1210
1211
1212 /*
1213 * we are inverting the lru lock/dentry->d_lock here,
1214 * so use a trylock. If we fail to get the lock, just skip
1215 * it
1216 */
1217 if (!spin_trylock(&dentry->d_lock))
1218 return LRU_SKIP;
1219
1220 /*
1221 * Referenced dentries are still in use. If they have active
1222 * counts, just remove them from the LRU. Otherwise give them
1223 * another pass through the LRU.
1224 */
1225 if (dentry->d_lockref.count) {
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001226 d_lru_isolate(lru, dentry);
Dave Chinnerf6041562013-08-28 10:18:00 +10001227 spin_unlock(&dentry->d_lock);
1228 return LRU_REMOVED;
1229 }
1230
1231 if (dentry->d_flags & DCACHE_REFERENCED) {
1232 dentry->d_flags &= ~DCACHE_REFERENCED;
1233 spin_unlock(&dentry->d_lock);
1234
1235 /*
1236 * The list move itself will be made by the common LRU code. At
1237 * this point, we've dropped the dentry->d_lock but keep the
1238 * lru lock. This is safe to do, since every list movement is
1239 * protected by the lru lock even if both locks are held.
1240 *
1241 * This is guaranteed by the fact that all LRU management
1242 * functions are intermediated by the LRU API calls like
1243 * list_lru_add and list_lru_del. List movement in this file
1244 * only ever occur through this functions or through callbacks
1245 * like this one, that are called from the LRU API.
1246 *
1247 * The only exceptions to this are functions like
1248 * shrink_dentry_list, and code that first checks for the
1249 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1250 * operating only with stack provided lists after they are
1251 * properly isolated from the main list. It is thus, always a
1252 * local access.
1253 */
1254 return LRU_ROTATE;
1255 }
1256
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001257 d_lru_shrink_move(lru, dentry, freeable);
Dave Chinnerf6041562013-08-28 10:18:00 +10001258 spin_unlock(&dentry->d_lock);
1259
1260 return LRU_REMOVED;
1261}
1262
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001263/**
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001264 * prune_dcache_sb - shrink the dcache
1265 * @sb: superblock
Vladimir Davydov503c3582015-02-12 14:58:47 -08001266 * @sc: shrink control, passed to list_lru_shrink_walk()
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001267 *
Vladimir Davydov503c3582015-02-12 14:58:47 -08001268 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1269 * is done when we need more memory and called from the superblock shrinker
Dave Chinnerb48f03b2011-08-23 18:56:24 +10001270 * function.
1271 *
1272 * This function may fail to free any resources if all the dentries are in
1273 * use.
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001274 */
Vladimir Davydov503c3582015-02-12 14:58:47 -08001275long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001276{
Dave Chinnerf6041562013-08-28 10:18:00 +10001277 LIST_HEAD(dispose);
1278 long freed;
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001279
Vladimir Davydov503c3582015-02-12 14:58:47 -08001280 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1281 dentry_lru_isolate, &dispose);
Dave Chinnerf6041562013-08-28 10:18:00 +10001282 shrink_dentry_list(&dispose);
Dave Chinner0a234c62013-08-28 10:17:57 +10001283 return freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284}
1285
Glauber Costa4e717f52013-08-28 10:18:03 +10001286static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001287 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001288{
Glauber Costa4e717f52013-08-28 10:18:03 +10001289 struct list_head *freeable = arg;
1290 struct dentry *dentry = container_of(item, struct dentry, d_lru);
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001291
Glauber Costa4e717f52013-08-28 10:18:03 +10001292 /*
1293 * we are inverting the lru lock/dentry->d_lock here,
1294 * so use a trylock. If we fail to get the lock, just skip
1295 * it
1296 */
1297 if (!spin_trylock(&dentry->d_lock))
1298 return LRU_SKIP;
1299
Vladimir Davydov3f97b162015-02-12 14:59:35 -08001300 d_lru_shrink_move(lru, dentry, freeable);
Glauber Costa4e717f52013-08-28 10:18:03 +10001301 spin_unlock(&dentry->d_lock);
1302
1303 return LRU_REMOVED;
Dave Chinnerdd1f6b22013-08-28 10:17:55 +10001304}
1305
Glauber Costa4e717f52013-08-28 10:18:03 +10001306
Kentaro Makitada3bbdd2008-07-23 21:27:13 -07001307/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 * shrink_dcache_sb - shrink dcache for a superblock
1309 * @sb: superblock
1310 *
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001311 * Shrink the dcache for the specified super block. This is used to free
1312 * the dcache before unmounting a file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 */
Christoph Hellwig3049cfe2010-10-10 05:36:25 -04001314void shrink_dcache_sb(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Glauber Costa4e717f52013-08-28 10:18:03 +10001316 do {
1317 LIST_HEAD(dispose);
1318
Waiman Long1dbd4492019-01-30 13:52:36 -05001319 list_lru_walk(&sb->s_dentry_lru,
Sahitya Tummalab17c0702017-07-10 15:50:00 -07001320 dentry_lru_isolate_shrink, &dispose, 1024);
Glauber Costa4e717f52013-08-28 10:18:03 +10001321 shrink_dentry_list(&dispose);
Sahitya Tummalab17c0702017-07-10 15:50:00 -07001322 } while (list_lru_count(&sb->s_dentry_lru) > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001324EXPORT_SYMBOL(shrink_dcache_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326/**
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001327 * enum d_walk_ret - action to talke during tree walk
1328 * @D_WALK_CONTINUE: contrinue walk
1329 * @D_WALK_QUIT: quit walk
1330 * @D_WALK_NORETRY: quit when retry is needed
1331 * @D_WALK_SKIP: skip this dentry and its children
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001333enum d_walk_ret {
1334 D_WALK_CONTINUE,
1335 D_WALK_QUIT,
1336 D_WALK_NORETRY,
1337 D_WALK_SKIP,
1338};
1339
1340/**
1341 * d_walk - walk the dentry tree
1342 * @parent: start of walk
1343 * @data: data passed to @enter() and @finish()
1344 * @enter: callback when first entering the dentry
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001345 *
Al Viro3a8e3612018-04-15 18:27:23 -04001346 * The @enter() callbacks are called with d_lock held.
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001347 */
1348static void d_walk(struct dentry *parent, void *data,
Al Viro3a8e3612018-04-15 18:27:23 -04001349 enum d_walk_ret (*enter)(void *, struct dentry *))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350{
Nick Piggin949854d2011-01-07 17:49:37 +11001351 struct dentry *this_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct list_head *next;
Al Viro48f5ec22013-09-09 15:22:25 -04001353 unsigned seq = 0;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001354 enum d_walk_ret ret;
1355 bool retry = true;
Nick Piggin949854d2011-01-07 17:49:37 +11001356
Nick Piggin58db63d2011-01-07 17:49:39 +11001357again:
Al Viro48f5ec22013-09-09 15:22:25 -04001358 read_seqbegin_or_lock(&rename_lock, &seq);
Nick Piggin58db63d2011-01-07 17:49:39 +11001359 this_parent = parent;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001360 spin_lock(&this_parent->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001361
1362 ret = enter(data, this_parent);
1363 switch (ret) {
1364 case D_WALK_CONTINUE:
1365 break;
1366 case D_WALK_QUIT:
1367 case D_WALK_SKIP:
1368 goto out_unlock;
1369 case D_WALK_NORETRY:
1370 retry = false;
1371 break;
1372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373repeat:
1374 next = this_parent->d_subdirs.next;
1375resume:
1376 while (next != &this_parent->d_subdirs) {
1377 struct list_head *tmp = next;
Al Viro946e51f2014-10-26 19:19:16 -04001378 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 next = tmp->next;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001380
Al Viroba65dc52016-06-10 11:32:47 -04001381 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1382 continue;
1383
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001384 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001385
1386 ret = enter(data, dentry);
1387 switch (ret) {
1388 case D_WALK_CONTINUE:
1389 break;
1390 case D_WALK_QUIT:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001391 spin_unlock(&dentry->d_lock);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001392 goto out_unlock;
1393 case D_WALK_NORETRY:
1394 retry = false;
1395 break;
1396 case D_WALK_SKIP:
1397 spin_unlock(&dentry->d_lock);
1398 continue;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001399 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (!list_empty(&dentry->d_subdirs)) {
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001402 spin_unlock(&this_parent->d_lock);
Qian Cai5facae42019-09-19 12:09:40 -04001403 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 this_parent = dentry;
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001405 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 goto repeat;
1407 }
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001408 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 }
1410 /*
1411 * All done at this level ... ascend and resume the search.
1412 */
Al Viroca5358e2014-10-26 19:31:10 -04001413 rcu_read_lock();
1414ascend:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 if (this_parent != parent) {
Linus Torvaldsc826cb72011-03-15 15:29:21 -07001416 struct dentry *child = this_parent;
Al Viro31dec132013-10-25 17:04:27 -04001417 this_parent = child->d_parent;
1418
Al Viro31dec132013-10-25 17:04:27 -04001419 spin_unlock(&child->d_lock);
1420 spin_lock(&this_parent->d_lock);
1421
Al Viroca5358e2014-10-26 19:31:10 -04001422 /* might go back up the wrong parent if we have had a rename. */
1423 if (need_seqretry(&rename_lock, seq))
Nick Piggin949854d2011-01-07 17:49:37 +11001424 goto rename_retry;
Al Viro21591842015-05-28 23:09:19 -04001425 /* go into the first sibling still alive */
1426 do {
1427 next = child->d_child.next;
Al Viroca5358e2014-10-26 19:31:10 -04001428 if (next == &this_parent->d_subdirs)
1429 goto ascend;
1430 child = list_entry(next, struct dentry, d_child);
Al Viro21591842015-05-28 23:09:19 -04001431 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
Al Viro31dec132013-10-25 17:04:27 -04001432 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 goto resume;
1434 }
Al Viroca5358e2014-10-26 19:31:10 -04001435 if (need_seqretry(&rename_lock, seq))
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001436 goto rename_retry;
Al Viroca5358e2014-10-26 19:31:10 -04001437 rcu_read_unlock();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001438
1439out_unlock:
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11001440 spin_unlock(&this_parent->d_lock);
Al Viro48f5ec22013-09-09 15:22:25 -04001441 done_seqretry(&rename_lock, seq);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001442 return;
Nick Piggin58db63d2011-01-07 17:49:39 +11001443
1444rename_retry:
Al Viroca5358e2014-10-26 19:31:10 -04001445 spin_unlock(&this_parent->d_lock);
1446 rcu_read_unlock();
1447 BUG_ON(seq & 1);
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001448 if (!retry)
1449 return;
Al Viro48f5ec22013-09-09 15:22:25 -04001450 seq = 1;
Nick Piggin58db63d2011-01-07 17:49:39 +11001451 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001453
Ian Kent01619492016-11-24 08:03:41 +11001454struct check_mount {
1455 struct vfsmount *mnt;
1456 unsigned int mounted;
1457};
1458
1459static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1460{
1461 struct check_mount *info = data;
1462 struct path path = { .mnt = info->mnt, .dentry = dentry };
1463
1464 if (likely(!d_mountpoint(dentry)))
1465 return D_WALK_CONTINUE;
1466 if (__path_is_mountpoint(&path)) {
1467 info->mounted = 1;
1468 return D_WALK_QUIT;
1469 }
1470 return D_WALK_CONTINUE;
1471}
1472
1473/**
1474 * path_has_submounts - check for mounts over a dentry in the
1475 * current namespace.
1476 * @parent: path to check.
1477 *
1478 * Return true if the parent or its subdirectories contain
1479 * a mount point in the current namespace.
1480 */
1481int path_has_submounts(const struct path *parent)
1482{
1483 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1484
1485 read_seqlock_excl(&mount_lock);
Al Viro3a8e3612018-04-15 18:27:23 -04001486 d_walk(parent->dentry, &data, path_check_mount);
Ian Kent01619492016-11-24 08:03:41 +11001487 read_sequnlock_excl(&mount_lock);
1488
1489 return data.mounted;
1490}
1491EXPORT_SYMBOL(path_has_submounts);
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493/*
Miklos Szeredieed81002013-09-05 14:39:11 +02001494 * Called by mount code to set a mountpoint and check if the mountpoint is
1495 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1496 * subtree can become unreachable).
1497 *
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001498 * Only one of d_invalidate() and d_set_mounted() must succeed. For
Miklos Szeredieed81002013-09-05 14:39:11 +02001499 * this reason take rename_lock and d_lock on dentry and ancestors.
1500 */
1501int d_set_mounted(struct dentry *dentry)
1502{
1503 struct dentry *p;
1504 int ret = -ENOENT;
1505 write_seqlock(&rename_lock);
1506 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001507 /* Need exclusion wrt. d_invalidate() */
Miklos Szeredieed81002013-09-05 14:39:11 +02001508 spin_lock(&p->d_lock);
1509 if (unlikely(d_unhashed(p))) {
1510 spin_unlock(&p->d_lock);
1511 goto out;
1512 }
1513 spin_unlock(&p->d_lock);
1514 }
1515 spin_lock(&dentry->d_lock);
1516 if (!d_unlinked(dentry)) {
Eric W. Biederman3895dbf2017-01-03 14:18:43 +13001517 ret = -EBUSY;
1518 if (!d_mountpoint(dentry)) {
1519 dentry->d_flags |= DCACHE_MOUNTED;
1520 ret = 0;
1521 }
Miklos Szeredieed81002013-09-05 14:39:11 +02001522 }
1523 spin_unlock(&dentry->d_lock);
1524out:
1525 write_sequnlock(&rename_lock);
1526 return ret;
1527}
1528
1529/*
J. Bruce Fieldsfd517902012-09-18 16:35:51 -04001530 * Search the dentry child list of the specified parent,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 * and move any unused dentries to the end of the unused
1532 * list for prune_dcache(). We descend to the next level
1533 * whenever the d_subdirs list is non-empty and continue
1534 * searching.
1535 *
1536 * It returns zero iff there are no unused children,
1537 * otherwise it returns the number of children moved to
1538 * the end of the unused list. This may not be the total
1539 * number of unused children, because select_parent can
1540 * drop the lock and return early due to latency
1541 * constraints.
1542 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001543
1544struct select_data {
1545 struct dentry *start;
Al Viro9bdebc22019-06-29 18:31:24 -04001546 union {
1547 long found;
1548 struct dentry *victim;
1549 };
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001550 struct list_head dispose;
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001551};
1552
1553static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001555 struct select_data *data = _data;
1556 enum d_walk_ret ret = D_WALK_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001558 if (data->start == dentry)
1559 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
Al Virofe915222014-05-03 00:02:25 -04001561 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001562 data->found++;
Al Virofe915222014-05-03 00:02:25 -04001563 } else {
1564 if (dentry->d_flags & DCACHE_LRU_LIST)
1565 d_lru_del(dentry);
1566 if (!dentry->d_lockref.count) {
1567 d_shrink_add(dentry, &data->dispose);
1568 data->found++;
1569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 }
1571 /*
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001572 * We can return to the caller if we have found some (this
1573 * ensures forward progress). We'll be coming back to find
1574 * the rest.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 */
Al Virofe915222014-05-03 00:02:25 -04001576 if (!list_empty(&data->dispose))
1577 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578out:
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001579 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580}
1581
Al Viro9bdebc22019-06-29 18:31:24 -04001582static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1583{
1584 struct select_data *data = _data;
1585 enum d_walk_ret ret = D_WALK_CONTINUE;
1586
1587 if (data->start == dentry)
1588 goto out;
1589
1590 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1591 if (!dentry->d_lockref.count) {
1592 rcu_read_lock();
1593 data->victim = dentry;
1594 return D_WALK_QUIT;
1595 }
1596 } else {
1597 if (dentry->d_flags & DCACHE_LRU_LIST)
1598 d_lru_del(dentry);
1599 if (!dentry->d_lockref.count)
1600 d_shrink_add(dentry, &data->dispose);
1601 }
1602 /*
1603 * We can return to the caller if we have found some (this
1604 * ensures forward progress). We'll be coming back to find
1605 * the rest.
1606 */
1607 if (!list_empty(&data->dispose))
1608 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1609out:
1610 return ret;
1611}
1612
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613/**
1614 * shrink_dcache_parent - prune dcache
1615 * @parent: parent of entries to prune
1616 *
1617 * Prune the dcache to remove unused children of the parent dentry.
1618 */
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001619void shrink_dcache_parent(struct dentry *parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001621 for (;;) {
Al Viro9bdebc22019-06-29 18:31:24 -04001622 struct select_data data = {.start = parent};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001624 INIT_LIST_HEAD(&data.dispose);
Al Viro3a8e3612018-04-15 18:27:23 -04001625 d_walk(parent, &data, select_collect);
Al Viro4fb48872018-04-19 23:58:48 -04001626
1627 if (!list_empty(&data.dispose)) {
1628 shrink_dentry_list(&data.dispose);
1629 continue;
1630 }
1631
1632 cond_resched();
Miklos Szeredidb14fc32013-09-05 11:44:35 +02001633 if (!data.found)
1634 break;
Al Viro9bdebc22019-06-29 18:31:24 -04001635 data.victim = NULL;
1636 d_walk(parent, &data, select_collect2);
1637 if (data.victim) {
1638 struct dentry *parent;
1639 spin_lock(&data.victim->d_lock);
1640 if (!shrink_lock_dentry(data.victim)) {
1641 spin_unlock(&data.victim->d_lock);
1642 rcu_read_unlock();
1643 } else {
1644 rcu_read_unlock();
1645 parent = data.victim->d_parent;
1646 if (parent != data.victim)
1647 __dput_to_list(parent, &data.dispose);
1648 __dentry_kill(data.victim);
1649 }
1650 }
1651 if (!list_empty(&data.dispose))
1652 shrink_dentry_list(&data.dispose);
Greg Thelen421348f2013-04-30 15:26:48 -07001653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001655EXPORT_SYMBOL(shrink_dcache_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
Al Viro9c8c10e2014-05-02 20:36:10 -04001657static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
Al Viro42c32602013-11-08 12:31:16 -05001658{
Al Viro9c8c10e2014-05-02 20:36:10 -04001659 /* it has busy descendents; complain about those instead */
1660 if (!list_empty(&dentry->d_subdirs))
1661 return D_WALK_CONTINUE;
Al Viro42c32602013-11-08 12:31:16 -05001662
Al Viro9c8c10e2014-05-02 20:36:10 -04001663 /* root with refcount 1 is fine */
1664 if (dentry == _data && dentry->d_lockref.count == 1)
1665 return D_WALK_CONTINUE;
1666
1667 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1668 " still in use (%d) [unmount of %s %s]\n",
Al Viro42c32602013-11-08 12:31:16 -05001669 dentry,
1670 dentry->d_inode ?
1671 dentry->d_inode->i_ino : 0UL,
Al Viro9c8c10e2014-05-02 20:36:10 -04001672 dentry,
Al Viro42c32602013-11-08 12:31:16 -05001673 dentry->d_lockref.count,
1674 dentry->d_sb->s_type->name,
1675 dentry->d_sb->s_id);
Al Viro9c8c10e2014-05-02 20:36:10 -04001676 WARN_ON(1);
1677 return D_WALK_CONTINUE;
1678}
1679
1680static void do_one_tree(struct dentry *dentry)
1681{
1682 shrink_dcache_parent(dentry);
Al Viro3a8e3612018-04-15 18:27:23 -04001683 d_walk(dentry, dentry, umount_check);
Al Viro9c8c10e2014-05-02 20:36:10 -04001684 d_drop(dentry);
1685 dput(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001686}
1687
1688/*
1689 * destroy the dentries attached to a superblock on unmounting
1690 */
1691void shrink_dcache_for_umount(struct super_block *sb)
1692{
1693 struct dentry *dentry;
1694
Al Viro9c8c10e2014-05-02 20:36:10 -04001695 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
Al Viro42c32602013-11-08 12:31:16 -05001696
1697 dentry = sb->s_root;
1698 sb->s_root = NULL;
Al Viro9c8c10e2014-05-02 20:36:10 -04001699 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001700
NeilBrownf1ee6162017-12-21 09:45:40 +11001701 while (!hlist_bl_empty(&sb->s_roots)) {
1702 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
Al Viro9c8c10e2014-05-02 20:36:10 -04001703 do_one_tree(dentry);
Al Viro42c32602013-11-08 12:31:16 -05001704 }
1705}
1706
Al Viroff17fa52018-04-15 18:21:47 -04001707static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001708{
Al Viroff17fa52018-04-15 18:21:47 -04001709 struct dentry **victim = _data;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001710 if (d_mountpoint(dentry)) {
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001711 __dget_dlock(dentry);
Al Viroff17fa52018-04-15 18:21:47 -04001712 *victim = dentry;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001713 return D_WALK_QUIT;
1714 }
Al Viroff17fa52018-04-15 18:21:47 -04001715 return D_WALK_CONTINUE;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001716}
1717
1718/**
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001719 * d_invalidate - detach submounts, prune dcache, and drop
1720 * @dentry: dentry to invalidate (aka detach, prune and drop)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001721 */
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001722void d_invalidate(struct dentry *dentry)
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001723{
Al Viroff17fa52018-04-15 18:21:47 -04001724 bool had_submounts = false;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001725 spin_lock(&dentry->d_lock);
1726 if (d_unhashed(dentry)) {
1727 spin_unlock(&dentry->d_lock);
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001728 return;
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001729 }
Al Viroff17fa52018-04-15 18:21:47 -04001730 __d_drop(dentry);
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001731 spin_unlock(&dentry->d_lock);
1732
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001733 /* Negative dentries can be dropped without further checks */
Al Viroff17fa52018-04-15 18:21:47 -04001734 if (!dentry->d_inode)
Eric W. Biederman5542aa22014-02-13 09:46:25 -08001735 return;
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001736
Al Viroff17fa52018-04-15 18:21:47 -04001737 shrink_dcache_parent(dentry);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001738 for (;;) {
Al Viroff17fa52018-04-15 18:21:47 -04001739 struct dentry *victim = NULL;
Al Viro3a8e3612018-04-15 18:27:23 -04001740 d_walk(dentry, &victim, find_submount);
Al Viroff17fa52018-04-15 18:21:47 -04001741 if (!victim) {
1742 if (had_submounts)
1743 shrink_dcache_parent(dentry);
Al Viro81be24d2017-06-03 07:20:09 +01001744 return;
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07001745 }
Al Viroff17fa52018-04-15 18:21:47 -04001746 had_submounts = true;
1747 detach_mounts(victim);
1748 dput(victim);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001749 }
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001750}
Eric W. Biederman1ffe46d2014-02-13 09:39:37 -08001751EXPORT_SYMBOL(d_invalidate);
Miklos Szeredi848ac1142013-09-05 11:44:36 +02001752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753/**
Al Viroa4464db2011-07-07 15:03:58 -04001754 * __d_alloc - allocate a dcache entry
1755 * @sb: filesystem it will belong to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 * @name: qstr of the name
1757 *
1758 * Allocates a dentry. It returns %NULL if there is insufficient memory
1759 * available. On a success the dentry is returned. The name passed in is
1760 * copied and the copy passed in may be reused after this call.
1761 */
1762
Al Viro5c8b0df2019-10-25 14:08:24 -04001763static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
1765 struct dentry *dentry;
1766 char *dname;
Miklos Szeredi285b1022016-06-28 11:47:32 +02001767 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Mel Gormane12ba742007-10-16 01:25:52 -07001769 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 if (!dentry)
1771 return NULL;
1772
Linus Torvalds6326c71f2012-05-21 16:14:04 -07001773 /*
1774 * We guarantee that the inline name is always NUL-terminated.
1775 * This way the memcpy() done by the name switching in rename
1776 * will still always have a NUL at the end, even if we might
1777 * be overwriting an internal NUL character
1778 */
1779 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
Al Viro798434b2016-03-24 20:38:43 -04001780 if (unlikely(!name)) {
David Howellscdf01222017-07-04 17:25:22 +01001781 name = &slash_name;
Al Viro798434b2016-03-24 20:38:43 -04001782 dname = dentry->d_iname;
1783 } else if (name->len > DNAME_INLINE_LEN-1) {
Al Viro8d85b482014-09-29 14:54:27 -04001784 size_t size = offsetof(struct external_name, name[1]);
Vlastimil Babka2e03b4b2018-10-26 15:05:41 -07001785 struct external_name *p = kmalloc(size + name->len,
1786 GFP_KERNEL_ACCOUNT |
1787 __GFP_RECLAIMABLE);
1788 if (!p) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 kmem_cache_free(dentry_cache, dentry);
1790 return NULL;
1791 }
Vlastimil Babka2e03b4b2018-10-26 15:05:41 -07001792 atomic_set(&p->u.count, 1);
1793 dname = p->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 } else {
1795 dname = dentry->d_iname;
1796 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
1798 dentry->d_name.len = name->len;
1799 dentry->d_name.hash = name->hash;
1800 memcpy(dname, name->name, name->len);
1801 dname[name->len] = 0;
1802
Linus Torvalds6326c71f2012-05-21 16:14:04 -07001803 /* Make sure we always see the terminating NUL character */
Paul E. McKenney7088efa2017-10-09 10:04:27 -07001804 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
Linus Torvalds6326c71f2012-05-21 16:14:04 -07001805
Waiman Long98474232013-08-28 18:24:59 -07001806 dentry->d_lockref.count = 1;
Linus Torvaldsdea36672011-04-24 07:58:46 -07001807 dentry->d_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 spin_lock_init(&dentry->d_lock);
Ahmed S. Darwish26475372020-07-20 17:55:24 +02001809 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 dentry->d_inode = NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001811 dentry->d_parent = dentry;
1812 dentry->d_sb = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 dentry->d_op = NULL;
1814 dentry->d_fsdata = NULL;
Nick Pigginceb5bdc2011-01-07 17:50:05 +11001815 INIT_HLIST_BL_NODE(&dentry->d_hash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 INIT_LIST_HEAD(&dentry->d_lru);
1817 INIT_LIST_HEAD(&dentry->d_subdirs);
Al Viro946e51f2014-10-26 19:19:16 -04001818 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1819 INIT_LIST_HEAD(&dentry->d_child);
Al Viroa4464db2011-07-07 15:03:58 -04001820 d_set_d_op(dentry, dentry->d_sb->s_d_op);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
Miklos Szeredi285b1022016-06-28 11:47:32 +02001822 if (dentry->d_op && dentry->d_op->d_init) {
1823 err = dentry->d_op->d_init(dentry);
1824 if (err) {
1825 if (dname_external(dentry))
1826 kfree(external_name(dentry));
1827 kmem_cache_free(dentry_cache, dentry);
1828 return NULL;
1829 }
1830 }
1831
Nick Piggin3e880fb2011-01-07 17:49:19 +11001832 this_cpu_inc(nr_dentry);
Christoph Hellwig312d3ca2010-10-10 05:36:23 -04001833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 return dentry;
1835}
Al Viroa4464db2011-07-07 15:03:58 -04001836
1837/**
1838 * d_alloc - allocate a dcache entry
1839 * @parent: parent of entry to allocate
1840 * @name: qstr of the name
1841 *
1842 * Allocates a dentry. It returns %NULL if there is insufficient memory
1843 * available. On a success the dentry is returned. The name passed in is
1844 * copied and the copy passed in may be reused after this call.
1845 */
1846struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1847{
1848 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1849 if (!dentry)
1850 return NULL;
Al Viroa4464db2011-07-07 15:03:58 -04001851 spin_lock(&parent->d_lock);
1852 /*
1853 * don't need child lock because it is not subject
1854 * to concurrency here
1855 */
1856 __dget_dlock(parent);
1857 dentry->d_parent = parent;
Al Viro946e51f2014-10-26 19:19:16 -04001858 list_add(&dentry->d_child, &parent->d_subdirs);
Al Viroa4464db2011-07-07 15:03:58 -04001859 spin_unlock(&parent->d_lock);
1860
1861 return dentry;
1862}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07001863EXPORT_SYMBOL(d_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Miklos Szeredif9c34672018-01-19 11:39:52 +01001865struct dentry *d_alloc_anon(struct super_block *sb)
1866{
1867 return __d_alloc(sb, NULL);
1868}
1869EXPORT_SYMBOL(d_alloc_anon);
1870
Al Viroba65dc52016-06-10 11:32:47 -04001871struct dentry *d_alloc_cursor(struct dentry * parent)
1872{
Miklos Szeredif9c34672018-01-19 11:39:52 +01001873 struct dentry *dentry = d_alloc_anon(parent->d_sb);
Al Viroba65dc52016-06-10 11:32:47 -04001874 if (dentry) {
Al Viro5467a682019-03-15 22:23:19 -04001875 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
Al Viroba65dc52016-06-10 11:32:47 -04001876 dentry->d_parent = dget(parent);
1877 }
1878 return dentry;
1879}
1880
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001881/**
1882 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1883 * @sb: the superblock
1884 * @name: qstr of the name
1885 *
1886 * For a filesystem that just pins its dentries in memory and never
1887 * performs lookups at all, return an unhashed IS_ROOT dentry.
Al Viro5467a682019-03-15 22:23:19 -04001888 * This is used for pipes, sockets et.al. - the stuff that should
1889 * never be anyone's children or parents. Unlike all other
1890 * dentries, these will not have RCU delay between dropping the
1891 * last reference and freeing them.
Al Viroab1152d2019-03-15 22:58:11 -04001892 *
1893 * The only user is alloc_file_pseudo() and that's what should
1894 * be considered a public interface. Don't use directly.
J. Bruce Fieldse1a24bb2012-06-29 16:20:47 -04001895 */
Nick Piggin4b936882011-01-07 17:50:07 +11001896struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1897{
Al Viro5467a682019-03-15 22:23:19 -04001898 struct dentry *dentry = __d_alloc(sb, name);
1899 if (likely(dentry))
1900 dentry->d_flags |= DCACHE_NORCU;
1901 return dentry;
Nick Piggin4b936882011-01-07 17:50:07 +11001902}
Nick Piggin4b936882011-01-07 17:50:07 +11001903
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1905{
1906 struct qstr q;
1907
1908 q.name = name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07001909 q.hash_len = hashlen_string(parent, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 return d_alloc(parent, &q);
1911}
H Hartley Sweetenef26ca92009-09-29 20:09:42 -04001912EXPORT_SYMBOL(d_alloc_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
Nick Pigginfb045ad2011-01-07 17:49:55 +11001914void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1915{
Linus Torvalds6f7f7ca2011-01-14 13:26:18 -08001916 WARN_ON_ONCE(dentry->d_op);
1917 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
Nick Pigginfb045ad2011-01-07 17:49:55 +11001918 DCACHE_OP_COMPARE |
1919 DCACHE_OP_REVALIDATE |
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001920 DCACHE_OP_WEAK_REVALIDATE |
David Howells4bacc9c2015-06-18 14:32:31 +01001921 DCACHE_OP_DELETE |
Miklos Szeredid101a122016-03-26 16:14:37 -04001922 DCACHE_OP_REAL));
Nick Pigginfb045ad2011-01-07 17:49:55 +11001923 dentry->d_op = op;
1924 if (!op)
1925 return;
1926 if (op->d_hash)
1927 dentry->d_flags |= DCACHE_OP_HASH;
1928 if (op->d_compare)
1929 dentry->d_flags |= DCACHE_OP_COMPARE;
1930 if (op->d_revalidate)
1931 dentry->d_flags |= DCACHE_OP_REVALIDATE;
Jeff Laytonecf3d1f2013-02-20 11:19:05 -05001932 if (op->d_weak_revalidate)
1933 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001934 if (op->d_delete)
1935 dentry->d_flags |= DCACHE_OP_DELETE;
Sage Weilf0023bc2011-10-28 10:02:42 -07001936 if (op->d_prune)
1937 dentry->d_flags |= DCACHE_OP_PRUNE;
Miklos Szeredid101a122016-03-26 16:14:37 -04001938 if (op->d_real)
1939 dentry->d_flags |= DCACHE_OP_REAL;
Nick Pigginfb045ad2011-01-07 17:49:55 +11001940
1941}
1942EXPORT_SYMBOL(d_set_d_op);
1943
David Howellsdf1a0852015-01-29 12:02:28 +00001944
1945/*
1946 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1947 * @dentry - The dentry to mark
1948 *
1949 * Mark a dentry as falling through to the lower layer (as set with
1950 * d_pin_lower()). This flag may be recorded on the medium.
1951 */
1952void d_set_fallthru(struct dentry *dentry)
1953{
1954 spin_lock(&dentry->d_lock);
1955 dentry->d_flags |= DCACHE_FALLTHRU;
1956 spin_unlock(&dentry->d_lock);
1957}
1958EXPORT_SYMBOL(d_set_fallthru);
1959
David Howellsb18825a2013-09-12 19:22:53 +01001960static unsigned d_flags_for_inode(struct inode *inode)
1961{
David Howells44bdb5e2015-01-29 12:02:29 +00001962 unsigned add_flags = DCACHE_REGULAR_TYPE;
David Howellsb18825a2013-09-12 19:22:53 +01001963
1964 if (!inode)
1965 return DCACHE_MISS_TYPE;
1966
1967 if (S_ISDIR(inode->i_mode)) {
1968 add_flags = DCACHE_DIRECTORY_TYPE;
1969 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1970 if (unlikely(!inode->i_op->lookup))
1971 add_flags = DCACHE_AUTODIR_TYPE;
1972 else
1973 inode->i_opflags |= IOP_LOOKUP;
1974 }
David Howells44bdb5e2015-01-29 12:02:29 +00001975 goto type_determined;
David Howellsb18825a2013-09-12 19:22:53 +01001976 }
1977
David Howells44bdb5e2015-01-29 12:02:29 +00001978 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
Al Viro6b255392015-11-17 10:20:54 -05001979 if (unlikely(inode->i_op->get_link)) {
David Howells44bdb5e2015-01-29 12:02:29 +00001980 add_flags = DCACHE_SYMLINK_TYPE;
1981 goto type_determined;
1982 }
1983 inode->i_opflags |= IOP_NOFOLLOW;
1984 }
1985
1986 if (unlikely(!S_ISREG(inode->i_mode)))
1987 add_flags = DCACHE_SPECIAL_TYPE;
1988
1989type_determined:
David Howellsb18825a2013-09-12 19:22:53 +01001990 if (unlikely(IS_AUTOMOUNT(inode)))
1991 add_flags |= DCACHE_NEED_AUTOMOUNT;
1992 return add_flags;
1993}
1994
OGAWA Hirofumi360da902008-10-16 07:50:28 +09001995static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1996{
David Howellsb18825a2013-09-12 19:22:53 +01001997 unsigned add_flags = d_flags_for_inode(inode);
Al Viro85c7f812016-04-14 19:52:13 -04001998 WARN_ON(d_in_lookup(dentry));
David Howellsb18825a2013-09-12 19:22:53 +01001999
Nick Pigginb23fb0a2011-01-07 17:49:35 +11002000 spin_lock(&dentry->d_lock);
Waiman Longaf0c9af2019-01-30 13:52:38 -05002001 /*
2002 * Decrement negative dentry count if it was in the LRU list.
2003 */
2004 if (dentry->d_flags & DCACHE_LRU_LIST)
2005 this_cpu_dec(nr_dentry_negative);
Al Virode689f52016-03-09 18:05:42 -05002006 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Al Viroa528aca2016-02-29 12:12:46 -05002007 raw_write_seqcount_begin(&dentry->d_seq);
David Howells4bf46a22015-03-05 14:09:22 +00002008 __d_set_inode_and_type(dentry, inode, add_flags);
Al Viroa528aca2016-02-29 12:12:46 -05002009 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002010 fsnotify_update_flags(dentry);
Nick Pigginb23fb0a2011-01-07 17:49:35 +11002011 spin_unlock(&dentry->d_lock);
OGAWA Hirofumi360da902008-10-16 07:50:28 +09002012}
2013
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014/**
2015 * d_instantiate - fill in inode information for a dentry
2016 * @entry: dentry to complete
2017 * @inode: inode to attach to this dentry
2018 *
2019 * Fill in inode information in the entry.
2020 *
2021 * This turns negative dentries into productive full members
2022 * of society.
2023 *
2024 * NOTE! This assumes that the inode count has been incremented
2025 * (or otherwise set) by the caller to indicate that it is now
2026 * in use by the dcache.
2027 */
2028
2029void d_instantiate(struct dentry *entry, struct inode * inode)
2030{
Al Viro946e51f2014-10-26 19:19:16 -04002031 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
Al Virode689f52016-03-09 18:05:42 -05002032 if (inode) {
Al Virob9680912016-04-11 00:53:26 -04002033 security_d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002034 spin_lock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05002035 __d_instantiate(entry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11002036 spin_unlock(&inode->i_lock);
Al Virode689f52016-03-09 18:05:42 -05002037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002039EXPORT_SYMBOL(d_instantiate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Al Viro1e2e5472018-05-04 08:23:01 -04002041/*
2042 * This should be equivalent to d_instantiate() + unlock_new_inode(),
2043 * with lockdep-related part of unlock_new_inode() done before
2044 * anything else. Use that instead of open-coding d_instantiate()/
2045 * unlock_new_inode() combinations.
2046 */
2047void d_instantiate_new(struct dentry *entry, struct inode *inode)
2048{
2049 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
2050 BUG_ON(!inode);
2051 lockdep_annotate_inode_mutex_key(inode);
2052 security_d_instantiate(entry, inode);
2053 spin_lock(&inode->i_lock);
2054 __d_instantiate(entry, inode);
2055 WARN_ON(!(inode->i_state & I_NEW));
Al Viroc2b6d622018-06-28 15:53:17 -04002056 inode->i_state &= ~I_NEW & ~I_CREATING;
Al Viro1e2e5472018-05-04 08:23:01 -04002057 smp_mb();
2058 wake_up_bit(&inode->i_state, __I_NEW);
2059 spin_unlock(&inode->i_lock);
2060}
2061EXPORT_SYMBOL(d_instantiate_new);
2062
Al Viroadc0e91a2012-01-08 16:49:21 -05002063struct dentry *d_make_root(struct inode *root_inode)
2064{
2065 struct dentry *res = NULL;
2066
2067 if (root_inode) {
Miklos Szeredif9c34672018-01-19 11:39:52 +01002068 res = d_alloc_anon(root_inode->i_sb);
Al Viro5467a682019-03-15 22:23:19 -04002069 if (res)
Al Viroadc0e91a2012-01-08 16:49:21 -05002070 d_instantiate(res, root_inode);
Al Viro5467a682019-03-15 22:23:19 -04002071 else
Al Viroadc0e91a2012-01-08 16:49:21 -05002072 iput(root_inode);
2073 }
2074 return res;
2075}
2076EXPORT_SYMBOL(d_make_root);
2077
Miklos Szeredif9c34672018-01-19 11:39:52 +01002078static struct dentry *__d_instantiate_anon(struct dentry *dentry,
2079 struct inode *inode,
2080 bool disconnected)
2081{
2082 struct dentry *res;
2083 unsigned add_flags;
2084
2085 security_d_instantiate(dentry, inode);
2086 spin_lock(&inode->i_lock);
2087 res = __d_find_any_alias(inode);
2088 if (res) {
2089 spin_unlock(&inode->i_lock);
2090 dput(dentry);
2091 goto out_iput;
2092 }
2093
2094 /* attach a disconnected dentry */
2095 add_flags = d_flags_for_inode(inode);
2096
2097 if (disconnected)
2098 add_flags |= DCACHE_DISCONNECTED;
2099
2100 spin_lock(&dentry->d_lock);
2101 __d_set_inode_and_type(dentry, inode, add_flags);
2102 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
Linus Torvalds139351f2018-02-05 13:05:20 -08002103 if (!disconnected) {
2104 hlist_bl_lock(&dentry->d_sb->s_roots);
2105 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
2106 hlist_bl_unlock(&dentry->d_sb->s_roots);
2107 }
Miklos Szeredif9c34672018-01-19 11:39:52 +01002108 spin_unlock(&dentry->d_lock);
2109 spin_unlock(&inode->i_lock);
2110
2111 return dentry;
2112
2113 out_iput:
2114 iput(inode);
2115 return res;
2116}
2117
2118struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2119{
2120 return __d_instantiate_anon(dentry, inode, true);
2121}
2122EXPORT_SYMBOL(d_instantiate_anon);
2123
2124static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002125{
Christoph Hellwig9308a612008-08-11 15:49:12 +02002126 struct dentry *tmp;
2127 struct dentry *res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002128
2129 if (!inode)
Christoph Hellwig44003722008-08-11 15:49:04 +02002130 return ERR_PTR(-ESTALE);
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002131 if (IS_ERR(inode))
2132 return ERR_CAST(inode);
2133
J. Bruce Fieldsd891eed2011-01-18 15:45:09 -05002134 res = d_find_any_alias(inode);
Christoph Hellwig9308a612008-08-11 15:49:12 +02002135 if (res)
2136 goto out_iput;
2137
Miklos Szeredif9c34672018-01-19 11:39:52 +01002138 tmp = d_alloc_anon(inode->i_sb);
Christoph Hellwig9308a612008-08-11 15:49:12 +02002139 if (!tmp) {
2140 res = ERR_PTR(-ENOMEM);
2141 goto out_iput;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002142 }
Nick Pigginb5c84bf2011-01-07 17:49:38 +11002143
Miklos Szeredif9c34672018-01-19 11:39:52 +01002144 return __d_instantiate_anon(tmp, inode, disconnected);
Christoph Hellwig9308a612008-08-11 15:49:12 +02002145
Miklos Szeredif9c34672018-01-19 11:39:52 +01002146out_iput:
Christoph Hellwig9308a612008-08-11 15:49:12 +02002147 iput(inode);
2148 return res;
Christoph Hellwig4ea3ada2008-08-11 15:48:57 +02002149}
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002150
2151/**
2152 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2153 * @inode: inode to allocate the dentry for
2154 *
2155 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2156 * similar open by handle operations. The returned dentry may be anonymous,
2157 * or may have a full name (if the inode was already in the cache).
2158 *
2159 * When called on a directory inode, we must ensure that the inode only ever
2160 * has one dentry. If a dentry is found, that is returned instead of
2161 * allocating a new one.
2162 *
2163 * On successful return, the reference to the inode has been transferred
2164 * to the dentry. In case of an error the reference on the inode is released.
2165 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2166 * be passed in and the error will be propagated to the return value,
2167 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2168 */
2169struct dentry *d_obtain_alias(struct inode *inode)
2170{
Miklos Szeredif9c34672018-01-19 11:39:52 +01002171 return __d_obtain_alias(inode, true);
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002172}
Benny Halevyadc48722009-02-27 14:02:59 -08002173EXPORT_SYMBOL(d_obtain_alias);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
2175/**
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002176 * d_obtain_root - find or allocate a dentry for a given inode
2177 * @inode: inode to allocate the dentry for
2178 *
2179 * Obtain an IS_ROOT dentry for the root of a filesystem.
2180 *
2181 * We must ensure that directory inodes only ever have one dentry. If a
2182 * dentry is found, that is returned instead of allocating a new one.
2183 *
2184 * On successful return, the reference to the inode has been transferred
2185 * to the dentry. In case of an error the reference on the inode is
2186 * released. A %NULL or IS_ERR inode may be passed in and will be the
2187 * error will be propagate to the return value, with a %NULL @inode
2188 * replaced by ERR_PTR(-ESTALE).
2189 */
2190struct dentry *d_obtain_root(struct inode *inode)
2191{
Miklos Szeredif9c34672018-01-19 11:39:52 +01002192 return __d_obtain_alias(inode, false);
J. Bruce Fields1a0a3972014-02-14 17:35:37 -05002193}
2194EXPORT_SYMBOL(d_obtain_root);
2195
2196/**
Barry Naujok94035402008-05-21 16:50:46 +10002197 * d_add_ci - lookup or allocate new dentry with case-exact name
2198 * @inode: the inode case-insensitive lookup has found
2199 * @dentry: the negative dentry that was passed to the parent's lookup func
2200 * @name: the case-exact name to be associated with the returned dentry
2201 *
2202 * This is to avoid filling the dcache with case-insensitive names to the
2203 * same inode, only the actual correct case is stored in the dcache for
2204 * case-insensitive filesystems.
2205 *
Randy Dunlap3d742d42021-02-24 12:00:48 -08002206 * For a case-insensitive lookup match and if the case-exact dentry
2207 * already exists in the dcache, use it and return it.
Barry Naujok94035402008-05-21 16:50:46 +10002208 *
2209 * If no entry exists with the exact case name, allocate new dentry with
2210 * the exact case, and return the spliced entry.
2211 */
Christoph Hellwige45b5902008-08-07 23:49:07 +02002212struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
Barry Naujok94035402008-05-21 16:50:46 +10002213 struct qstr *name)
2214{
Al Virod9171b92016-04-15 03:33:13 -04002215 struct dentry *found, *res;
Barry Naujok94035402008-05-21 16:50:46 +10002216
Christoph Hellwigb6520c82009-01-05 19:10:37 +01002217 /*
2218 * First check if a dentry matching the name already exists,
2219 * if not go ahead and create it now.
2220 */
Barry Naujok94035402008-05-21 16:50:46 +10002221 found = d_hash_and_lookup(dentry->d_parent, name);
Al Virod9171b92016-04-15 03:33:13 -04002222 if (found) {
2223 iput(inode);
2224 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002225 }
Al Virod9171b92016-04-15 03:33:13 -04002226 if (d_in_lookup(dentry)) {
2227 found = d_alloc_parallel(dentry->d_parent, name,
2228 dentry->d_wait);
2229 if (IS_ERR(found) || !d_in_lookup(found)) {
2230 iput(inode);
2231 return found;
2232 }
2233 } else {
2234 found = d_alloc(dentry->d_parent, name);
2235 if (!found) {
2236 iput(inode);
2237 return ERR_PTR(-ENOMEM);
2238 }
2239 }
2240 res = d_splice_alias(inode, found);
2241 if (res) {
2242 dput(found);
2243 return res;
2244 }
Al Viro4f522a22013-02-11 23:20:37 -05002245 return found;
Barry Naujok94035402008-05-21 16:50:46 +10002246}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002247EXPORT_SYMBOL(d_add_ci);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002249
Al Virod4c91a82016-06-25 23:33:49 -04002250static inline bool d_same_name(const struct dentry *dentry,
2251 const struct dentry *parent,
2252 const struct qstr *name)
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002253{
Al Virod4c91a82016-06-25 23:33:49 -04002254 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2255 if (dentry->d_name.len != name->len)
2256 return false;
2257 return dentry_cmp(dentry, name->name, name->len) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002258 }
Al Viro6fa67e72016-07-31 16:37:25 -04002259 return parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002260 dentry->d_name.len, dentry->d_name.name,
2261 name) == 0;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002262}
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264/**
Nick Piggin31e6b012011-01-07 17:49:52 +11002265 * __d_lookup_rcu - search for a dentry (racy, store-free)
2266 * @parent: parent dentry
2267 * @name: qstr of name we wish to find
Randy Dunlap1f1e6e52012-03-18 21:23:05 -07002268 * @seqp: returns d_seq value at the point where the dentry was found
Nick Piggin31e6b012011-01-07 17:49:52 +11002269 * Returns: dentry, or NULL
2270 *
2271 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2272 * resolution (store-free path walking) design described in
2273 * Documentation/filesystems/path-lookup.txt.
2274 *
2275 * This is not to be used outside core vfs.
2276 *
2277 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2278 * held, and rcu_read_lock held. The returned dentry must not be stored into
2279 * without taking d_lock and checking d_seq sequence count against @seq
2280 * returned here.
2281 *
Linus Torvalds15570082013-09-02 11:38:06 -07002282 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
Nick Piggin31e6b012011-01-07 17:49:52 +11002283 * function.
2284 *
2285 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2286 * the returned dentry, so long as its parent's seqlock is checked after the
2287 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2288 * is formed, giving integrity down the path walk.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002289 *
2290 * NOTE! The caller *has* to check the resulting dentry against the sequence
2291 * number we've returned before using any of the resulting dentry state!
Nick Piggin31e6b012011-01-07 17:49:52 +11002292 */
Linus Torvalds8966be92012-03-02 14:23:30 -08002293struct dentry *__d_lookup_rcu(const struct dentry *parent,
2294 const struct qstr *name,
Linus Torvaldsda53be12013-05-21 15:22:44 -07002295 unsigned *seqp)
Nick Piggin31e6b012011-01-07 17:49:52 +11002296{
Linus Torvalds26fe5752012-05-10 13:14:12 -07002297 u64 hashlen = name->hash_len;
Nick Piggin31e6b012011-01-07 17:49:52 +11002298 const unsigned char *str = name->name;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002299 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002300 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002301 struct dentry *dentry;
2302
2303 /*
2304 * Note: There is significant duplication with __d_lookup_rcu which is
2305 * required to prevent single threaded performance regressions
2306 * especially on architectures where smp_rmb (in seqcounts) are costly.
2307 * Keep the two functions in sync.
2308 */
2309
2310 /*
2311 * The hash list is protected using RCU.
2312 *
2313 * Carefully use d_seq when comparing a candidate dentry, to avoid
2314 * races with d_move().
2315 *
2316 * It is possible that concurrent renames can mess up our list
2317 * walk here and result in missing our dentry, resulting in the
2318 * false-negative result. d_lookup() protects against concurrent
2319 * renames using rename_lock seqlock.
2320 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002321 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Piggin31e6b012011-01-07 17:49:52 +11002322 */
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002323 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds8966be92012-03-02 14:23:30 -08002324 unsigned seq;
Nick Piggin31e6b012011-01-07 17:49:52 +11002325
Nick Piggin31e6b012011-01-07 17:49:52 +11002326seqretry:
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002327 /*
2328 * The dentry sequence count protects us from concurrent
Linus Torvaldsda53be12013-05-21 15:22:44 -07002329 * renames, and thus protects parent and name fields.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002330 *
2331 * The caller must perform a seqcount check in order
Linus Torvaldsda53be12013-05-21 15:22:44 -07002332 * to do anything useful with the returned dentry.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002333 *
2334 * NOTE! We do a "raw" seqcount_begin here. That means that
2335 * we don't wait for the sequence count to stabilize if it
2336 * is in the middle of a sequence change. If we do the slow
2337 * dentry compare, we will do seqretries until it is stable,
2338 * and if we end up with a successful lookup, we actually
2339 * want to exit RCU lookup anyway.
Al Virod4c91a82016-06-25 23:33:49 -04002340 *
2341 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2342 * we are still guaranteed NUL-termination of ->d_name.name.
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002343 */
2344 seq = raw_seqcount_begin(&dentry->d_seq);
Nick Piggin31e6b012011-01-07 17:49:52 +11002345 if (dentry->d_parent != parent)
2346 continue;
Linus Torvalds2e321802012-05-21 18:48:10 -07002347 if (d_unhashed(dentry))
2348 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002349
2350 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
Al Virod4c91a82016-06-25 23:33:49 -04002351 int tlen;
2352 const char *tname;
Linus Torvalds26fe5752012-05-10 13:14:12 -07002353 if (dentry->d_name.hash != hashlen_hash(hashlen))
2354 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002355 tlen = dentry->d_name.len;
2356 tname = dentry->d_name.name;
2357 /* we want a consistent (name,len) pair */
2358 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2359 cpu_relax();
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002360 goto seqretry;
2361 }
Al Viro6fa67e72016-07-31 16:37:25 -04002362 if (parent->d_op->d_compare(dentry,
Al Virod4c91a82016-06-25 23:33:49 -04002363 tlen, tname, name) != 0)
2364 continue;
2365 } else {
2366 if (dentry->d_name.hash_len != hashlen)
2367 continue;
2368 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2369 continue;
Linus Torvalds12f8ad42012-05-04 14:59:14 -07002370 }
Linus Torvaldsda53be12013-05-21 15:22:44 -07002371 *seqp = seq;
Al Virod4c91a82016-06-25 23:33:49 -04002372 return dentry;
Nick Piggin31e6b012011-01-07 17:49:52 +11002373 }
2374 return NULL;
2375}
2376
2377/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 * d_lookup - search for a dentry
2379 * @parent: parent dentry
2380 * @name: qstr of name we wish to find
Nick Pigginb04f7842010-08-18 04:37:34 +10002381 * Returns: dentry, or NULL
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 *
Nick Pigginb04f7842010-08-18 04:37:34 +10002383 * d_lookup searches the children of the parent dentry for the name in
2384 * question. If the dentry is found its reference count is incremented and the
2385 * dentry is returned. The caller must use dput to free the entry when it has
2386 * finished using it. %NULL is returned if the dentry does not exist.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 */
Al Viroda2d8452013-01-24 18:29:34 -05002388struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389{
Nick Piggin31e6b012011-01-07 17:49:52 +11002390 struct dentry *dentry;
Nick Piggin949854d2011-01-07 17:49:37 +11002391 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392
Daeseok Younb8314f92014-08-11 11:46:53 +09002393 do {
2394 seq = read_seqbegin(&rename_lock);
2395 dentry = __d_lookup(parent, name);
2396 if (dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 break;
2398 } while (read_seqretry(&rename_lock, seq));
2399 return dentry;
2400}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002401EXPORT_SYMBOL(d_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402
Nick Piggin31e6b012011-01-07 17:49:52 +11002403/**
Nick Pigginb04f7842010-08-18 04:37:34 +10002404 * __d_lookup - search for a dentry (racy)
2405 * @parent: parent dentry
2406 * @name: qstr of name we wish to find
2407 * Returns: dentry, or NULL
2408 *
2409 * __d_lookup is like d_lookup, however it may (rarely) return a
2410 * false-negative result due to unrelated rename activity.
2411 *
2412 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2413 * however it must be used carefully, eg. with a following d_lookup in
2414 * the case of failure.
2415 *
2416 * __d_lookup callers must be commented.
2417 */
Al Viroa713ca2a2013-01-24 18:27:00 -05002418struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 unsigned int hash = name->hash;
Linus Torvalds8387ff22016-06-10 07:51:30 -07002421 struct hlist_bl_head *b = d_hash(hash);
Nick Pigginceb5bdc2011-01-07 17:50:05 +11002422 struct hlist_bl_node *node;
Nick Piggin31e6b012011-01-07 17:49:52 +11002423 struct dentry *found = NULL;
Paul E. McKenney665a7582005-11-07 00:59:17 -08002424 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Nick Pigginb04f7842010-08-18 04:37:34 +10002426 /*
Nick Piggin31e6b012011-01-07 17:49:52 +11002427 * Note: There is significant duplication with __d_lookup_rcu which is
2428 * required to prevent single threaded performance regressions
2429 * especially on architectures where smp_rmb (in seqcounts) are costly.
2430 * Keep the two functions in sync.
2431 */
2432
2433 /*
Nick Pigginb04f7842010-08-18 04:37:34 +10002434 * The hash list is protected using RCU.
2435 *
2436 * Take d_lock when comparing a candidate dentry, to avoid races
2437 * with d_move().
2438 *
2439 * It is possible that concurrent renames can mess up our list
2440 * walk here and result in missing our dentry, resulting in the
2441 * false-negative result. d_lookup() protects against concurrent
2442 * renames using rename_lock seqlock.
2443 *
Namhyung Kimb0a4bb82011-01-22 15:31:32 +09002444 * See Documentation/filesystems/path-lookup.txt for more details.
Nick Pigginb04f7842010-08-18 04:37:34 +10002445 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 rcu_read_lock();
2447
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002448 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 if (dentry->d_name.hash != hash)
2451 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
2453 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 if (dentry->d_parent != parent)
2455 goto next;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002456 if (d_unhashed(dentry))
2457 goto next;
2458
Al Virod4c91a82016-06-25 23:33:49 -04002459 if (!d_same_name(dentry, parent, name))
2460 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461
Waiman Long98474232013-08-28 18:24:59 -07002462 dentry->d_lockref.count++;
Linus Torvaldsd0185c02008-09-29 07:42:57 -07002463 found = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 spin_unlock(&dentry->d_lock);
2465 break;
2466next:
2467 spin_unlock(&dentry->d_lock);
2468 }
2469 rcu_read_unlock();
2470
2471 return found;
2472}
2473
2474/**
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002475 * d_hash_and_lookup - hash the qstr then search for a dentry
2476 * @dir: Directory to search in
2477 * @name: qstr of name we wish to find
2478 *
Al Viro4f522a22013-02-11 23:20:37 -05002479 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002480 */
2481struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2482{
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002483 /*
2484 * Check for a fs-specific hash function. Note that we must
2485 * calculate the standard hash first, as the d_op->d_hash()
2486 * routine may choose to leave the hash value unchanged.
2487 */
Linus Torvalds8387ff22016-06-10 07:51:30 -07002488 name->hash = full_name_hash(dir, name->name, name->len);
Nick Pigginfb045ad2011-01-07 17:49:55 +11002489 if (dir->d_flags & DCACHE_OP_HASH) {
Linus Torvaldsda53be12013-05-21 15:22:44 -07002490 int err = dir->d_op->d_hash(dir, name);
Al Viro4f522a22013-02-11 23:20:37 -05002491 if (unlikely(err < 0))
2492 return ERR_PTR(err);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002493 }
Al Viro4f522a22013-02-11 23:20:37 -05002494 return d_lookup(dir, name);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002495}
Al Viro4f522a22013-02-11 23:20:37 -05002496EXPORT_SYMBOL(d_hash_and_lookup);
Eric W. Biederman3e7e2412006-03-31 02:31:43 -08002497
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498/*
2499 * When a file is deleted, we have two options:
2500 * - turn this dentry into a negative dentry
2501 * - unhash this dentry and free it.
2502 *
2503 * Usually, we want to just turn this into
2504 * a negative dentry, but if anybody else is
2505 * currently using the dentry or the inode
2506 * we can't do that and we fall back on removing
2507 * it from the hash queues and waiting for
2508 * it to be deleted later when it has no users
2509 */
2510
2511/**
2512 * d_delete - delete a dentry
2513 * @dentry: The dentry to delete
2514 *
2515 * Turn the dentry into a negative dentry if possible, otherwise
2516 * remove it from the hash queues so it can be deleted later
2517 */
2518
2519void d_delete(struct dentry * dentry)
2520{
Al Viroc19457f2018-02-23 21:02:31 -05002521 struct inode *inode = dentry->d_inode;
Al Viroc19457f2018-02-23 21:02:31 -05002522
2523 spin_lock(&inode->i_lock);
2524 spin_lock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 /*
2526 * Are we the only user?
2527 */
Waiman Long98474232013-08-28 18:24:59 -07002528 if (dentry->d_lockref.count == 1) {
Al Viro13e3c5e2010-05-21 16:11:04 -04002529 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
Nick Piggin31e6b012011-01-07 17:49:52 +11002530 dentry_unlink_inode(dentry);
Al Viroc19457f2018-02-23 21:02:31 -05002531 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 __d_drop(dentry);
Al Viroc19457f2018-02-23 21:02:31 -05002533 spin_unlock(&dentry->d_lock);
2534 spin_unlock(&inode->i_lock);
2535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002537EXPORT_SYMBOL(d_delete);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538
Al Viro15d3c582016-07-29 17:45:21 -04002539static void __d_rehash(struct dentry *entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540{
Al Viro15d3c582016-07-29 17:45:21 -04002541 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
NeilBrown61647822017-11-10 15:45:41 +11002542
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002543 hlist_bl_lock(b);
Linus Torvaldsb07ad992011-04-23 22:32:03 -07002544 hlist_bl_add_head_rcu(&entry->d_hash, b);
Christoph Hellwig1879fd62011-04-25 14:01:36 -04002545 hlist_bl_unlock(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546}
2547
2548/**
2549 * d_rehash - add an entry back to the hash
2550 * @entry: dentry to add to the hash
2551 *
2552 * Adds a dentry to the hash according to its name.
2553 */
2554
2555void d_rehash(struct dentry * entry)
2556{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 spin_lock(&entry->d_lock);
Al Viro15d3c582016-07-29 17:45:21 -04002558 __d_rehash(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 spin_unlock(&entry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002561EXPORT_SYMBOL(d_rehash);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562
Al Viro84e710d2016-04-15 00:58:55 -04002563static inline unsigned start_dir_add(struct inode *dir)
2564{
2565
2566 for (;;) {
2567 unsigned n = dir->i_dir_seq;
2568 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2569 return n;
2570 cpu_relax();
2571 }
2572}
2573
2574static inline void end_dir_add(struct inode *dir, unsigned n)
2575{
2576 smp_store_release(&dir->i_dir_seq, n + 2);
2577}
2578
Al Virod9171b92016-04-15 03:33:13 -04002579static void d_wait_lookup(struct dentry *dentry)
2580{
2581 if (d_in_lookup(dentry)) {
2582 DECLARE_WAITQUEUE(wait, current);
2583 add_wait_queue(dentry->d_wait, &wait);
2584 do {
2585 set_current_state(TASK_UNINTERRUPTIBLE);
2586 spin_unlock(&dentry->d_lock);
2587 schedule();
2588 spin_lock(&dentry->d_lock);
2589 } while (d_in_lookup(dentry));
2590 }
2591}
2592
Al Viro94bdd652016-04-15 02:42:04 -04002593struct dentry *d_alloc_parallel(struct dentry *parent,
Al Virod9171b92016-04-15 03:33:13 -04002594 const struct qstr *name,
2595 wait_queue_head_t *wq)
Al Viro94bdd652016-04-15 02:42:04 -04002596{
Al Viro94bdd652016-04-15 02:42:04 -04002597 unsigned int hash = name->hash;
Al Viro94bdd652016-04-15 02:42:04 -04002598 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2599 struct hlist_bl_node *node;
2600 struct dentry *new = d_alloc(parent, name);
2601 struct dentry *dentry;
2602 unsigned seq, r_seq, d_seq;
2603
2604 if (unlikely(!new))
2605 return ERR_PTR(-ENOMEM);
2606
2607retry:
2608 rcu_read_lock();
Will Deacon015555f2018-02-19 14:55:54 +00002609 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
Al Viro94bdd652016-04-15 02:42:04 -04002610 r_seq = read_seqbegin(&rename_lock);
2611 dentry = __d_lookup_rcu(parent, name, &d_seq);
2612 if (unlikely(dentry)) {
2613 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2614 rcu_read_unlock();
2615 goto retry;
2616 }
2617 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2618 rcu_read_unlock();
2619 dput(dentry);
2620 goto retry;
2621 }
2622 rcu_read_unlock();
2623 dput(new);
2624 return dentry;
2625 }
2626 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2627 rcu_read_unlock();
2628 goto retry;
2629 }
Will Deacon015555f2018-02-19 14:55:54 +00002630
2631 if (unlikely(seq & 1)) {
2632 rcu_read_unlock();
2633 goto retry;
2634 }
2635
Al Viro94bdd652016-04-15 02:42:04 -04002636 hlist_bl_lock(b);
Will Deacon8cc07c82018-02-19 14:55:55 +00002637 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
Al Viro94bdd652016-04-15 02:42:04 -04002638 hlist_bl_unlock(b);
2639 rcu_read_unlock();
2640 goto retry;
2641 }
Al Viro94bdd652016-04-15 02:42:04 -04002642 /*
2643 * No changes for the parent since the beginning of d_lookup().
2644 * Since all removals from the chain happen with hlist_bl_lock(),
2645 * any potential in-lookup matches are going to stay here until
2646 * we unlock the chain. All fields are stable in everything
2647 * we encounter.
2648 */
2649 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2650 if (dentry->d_name.hash != hash)
2651 continue;
2652 if (dentry->d_parent != parent)
2653 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002654 if (!d_same_name(dentry, parent, name))
2655 continue;
Al Viro94bdd652016-04-15 02:42:04 -04002656 hlist_bl_unlock(b);
Al Viroe7d6ef92016-06-20 01:35:59 -04002657 /* now we can try to grab a reference */
2658 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2659 rcu_read_unlock();
2660 goto retry;
2661 }
2662
2663 rcu_read_unlock();
2664 /*
2665 * somebody is likely to be still doing lookup for it;
2666 * wait for them to finish
2667 */
Al Virod9171b92016-04-15 03:33:13 -04002668 spin_lock(&dentry->d_lock);
2669 d_wait_lookup(dentry);
2670 /*
2671 * it's not in-lookup anymore; in principle we should repeat
2672 * everything from dcache lookup, but it's likely to be what
2673 * d_lookup() would've found anyway. If it is, just return it;
2674 * otherwise we really have to repeat the whole thing.
2675 */
2676 if (unlikely(dentry->d_name.hash != hash))
2677 goto mismatch;
2678 if (unlikely(dentry->d_parent != parent))
2679 goto mismatch;
2680 if (unlikely(d_unhashed(dentry)))
2681 goto mismatch;
Al Virod4c91a82016-06-25 23:33:49 -04002682 if (unlikely(!d_same_name(dentry, parent, name)))
2683 goto mismatch;
Al Virod9171b92016-04-15 03:33:13 -04002684 /* OK, it *is* a hashed match; return it */
2685 spin_unlock(&dentry->d_lock);
Al Viro94bdd652016-04-15 02:42:04 -04002686 dput(new);
2687 return dentry;
2688 }
Al Viroe7d6ef92016-06-20 01:35:59 -04002689 rcu_read_unlock();
Al Viro94bdd652016-04-15 02:42:04 -04002690 /* we can't take ->d_lock here; it's OK, though. */
2691 new->d_flags |= DCACHE_PAR_LOOKUP;
Al Virod9171b92016-04-15 03:33:13 -04002692 new->d_wait = wq;
Al Viro94bdd652016-04-15 02:42:04 -04002693 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2694 hlist_bl_unlock(b);
2695 return new;
Al Virod9171b92016-04-15 03:33:13 -04002696mismatch:
2697 spin_unlock(&dentry->d_lock);
2698 dput(dentry);
2699 goto retry;
Al Viro94bdd652016-04-15 02:42:04 -04002700}
2701EXPORT_SYMBOL(d_alloc_parallel);
2702
Al Viro85c7f812016-04-14 19:52:13 -04002703void __d_lookup_done(struct dentry *dentry)
2704{
Al Viro94bdd652016-04-15 02:42:04 -04002705 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2706 dentry->d_name.hash);
2707 hlist_bl_lock(b);
Al Viro85c7f812016-04-14 19:52:13 -04002708 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
Al Viro94bdd652016-04-15 02:42:04 -04002709 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
Al Virod9171b92016-04-15 03:33:13 -04002710 wake_up_all(dentry->d_wait);
2711 dentry->d_wait = NULL;
Al Viro94bdd652016-04-15 02:42:04 -04002712 hlist_bl_unlock(b);
2713 INIT_HLIST_NODE(&dentry->d_u.d_alias);
Al Virod9171b92016-04-15 03:33:13 -04002714 INIT_LIST_HEAD(&dentry->d_lru);
Al Viro85c7f812016-04-14 19:52:13 -04002715}
2716EXPORT_SYMBOL(__d_lookup_done);
Al Viroed782b52016-03-09 19:52:39 -05002717
2718/* inode->i_lock held if inode is non-NULL */
2719
2720static inline void __d_add(struct dentry *dentry, struct inode *inode)
2721{
Al Viro84e710d2016-04-15 00:58:55 -04002722 struct inode *dir = NULL;
2723 unsigned n;
Al Viro0568d702016-04-14 19:40:56 -04002724 spin_lock(&dentry->d_lock);
Al Viro84e710d2016-04-15 00:58:55 -04002725 if (unlikely(d_in_lookup(dentry))) {
2726 dir = dentry->d_parent->d_inode;
2727 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002728 __d_lookup_done(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002729 }
Al Viroed782b52016-03-09 19:52:39 -05002730 if (inode) {
Al Viro0568d702016-04-14 19:40:56 -04002731 unsigned add_flags = d_flags_for_inode(inode);
2732 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2733 raw_write_seqcount_begin(&dentry->d_seq);
2734 __d_set_inode_and_type(dentry, inode, add_flags);
2735 raw_write_seqcount_end(&dentry->d_seq);
Al Viroaffda482016-05-29 18:35:12 -04002736 fsnotify_update_flags(dentry);
Al Viroed782b52016-03-09 19:52:39 -05002737 }
Al Viro15d3c582016-07-29 17:45:21 -04002738 __d_rehash(dentry);
Al Viro84e710d2016-04-15 00:58:55 -04002739 if (dir)
2740 end_dir_add(dir, n);
Al Viro0568d702016-04-14 19:40:56 -04002741 spin_unlock(&dentry->d_lock);
2742 if (inode)
2743 spin_unlock(&inode->i_lock);
Al Viroed782b52016-03-09 19:52:39 -05002744}
2745
Nick Pigginfb2d5b82011-01-07 17:49:26 +11002746/**
Al Viro34d0d192016-03-08 21:01:03 -05002747 * d_add - add dentry to hash queues
2748 * @entry: dentry to add
2749 * @inode: The inode to attach to this dentry
2750 *
2751 * This adds the entry to the hash queues and initializes @inode.
2752 * The entry was actually filled in earlier during d_alloc().
2753 */
2754
2755void d_add(struct dentry *entry, struct inode *inode)
2756{
Al Virob9680912016-04-11 00:53:26 -04002757 if (inode) {
2758 security_d_instantiate(entry, inode);
Al Viroed782b52016-03-09 19:52:39 -05002759 spin_lock(&inode->i_lock);
Al Virob9680912016-04-11 00:53:26 -04002760 }
Al Viroed782b52016-03-09 19:52:39 -05002761 __d_add(entry, inode);
Al Viro34d0d192016-03-08 21:01:03 -05002762}
2763EXPORT_SYMBOL(d_add);
2764
2765/**
Al Viro668d0cd2016-03-08 12:44:17 -05002766 * d_exact_alias - find and hash an exact unhashed alias
2767 * @entry: dentry to add
2768 * @inode: The inode to go with this dentry
2769 *
2770 * If an unhashed dentry with the same name/parent and desired
2771 * inode already exists, hash and return it. Otherwise, return
2772 * NULL.
2773 *
2774 * Parent directory should be locked.
2775 */
2776struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2777{
2778 struct dentry *alias;
Al Viro668d0cd2016-03-08 12:44:17 -05002779 unsigned int hash = entry->d_name.hash;
2780
2781 spin_lock(&inode->i_lock);
2782 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2783 /*
2784 * Don't need alias->d_lock here, because aliases with
2785 * d_parent == entry->d_parent are not subject to name or
2786 * parent changes, because the parent inode i_mutex is held.
2787 */
2788 if (alias->d_name.hash != hash)
2789 continue;
2790 if (alias->d_parent != entry->d_parent)
2791 continue;
Al Virod4c91a82016-06-25 23:33:49 -04002792 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
Al Viro668d0cd2016-03-08 12:44:17 -05002793 continue;
2794 spin_lock(&alias->d_lock);
2795 if (!d_unhashed(alias)) {
2796 spin_unlock(&alias->d_lock);
2797 alias = NULL;
2798 } else {
2799 __dget_dlock(alias);
Al Viro15d3c582016-07-29 17:45:21 -04002800 __d_rehash(alias);
Al Viro668d0cd2016-03-08 12:44:17 -05002801 spin_unlock(&alias->d_lock);
2802 }
2803 spin_unlock(&inode->i_lock);
2804 return alias;
2805 }
2806 spin_unlock(&inode->i_lock);
2807 return NULL;
2808}
2809EXPORT_SYMBOL(d_exact_alias);
2810
Al Viro8d85b482014-09-29 14:54:27 -04002811static void swap_names(struct dentry *dentry, struct dentry *target)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812{
Al Viro8d85b482014-09-29 14:54:27 -04002813 if (unlikely(dname_external(target))) {
2814 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 /*
2816 * Both external: swap the pointers
2817 */
Wu Fengguang9a8d5bb2009-01-07 18:09:14 -08002818 swap(target->d_name.name, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 } else {
2820 /*
2821 * dentry:internal, target:external. Steal target's
2822 * storage and make target internal.
2823 */
J. Bruce Fields321bcf92007-10-21 16:41:38 -07002824 memcpy(target->d_iname, dentry->d_name.name,
2825 dentry->d_name.len + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826 dentry->d_name.name = target->d_name.name;
2827 target->d_name.name = target->d_iname;
2828 }
2829 } else {
Al Viro8d85b482014-09-29 14:54:27 -04002830 if (unlikely(dname_external(dentry))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 /*
2832 * dentry:external, target:internal. Give dentry's
2833 * storage to target and make dentry internal
2834 */
2835 memcpy(dentry->d_iname, target->d_name.name,
2836 target->d_name.len + 1);
2837 target->d_name.name = dentry->d_name.name;
2838 dentry->d_name.name = dentry->d_iname;
2839 } else {
2840 /*
Miklos Szeredida1ce062014-04-01 17:08:43 +02002841 * Both are internal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002843 unsigned int i;
2844 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2845 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2846 swap(((long *) &dentry->d_iname)[i],
2847 ((long *) &target->d_iname)[i]);
2848 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002849 }
2850 }
Linus Torvaldsa28ddb82014-09-24 12:27:39 -07002851 swap(dentry->d_name.hash_len, target->d_name.hash_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852}
2853
Al Viro8d85b482014-09-29 14:54:27 -04002854static void copy_name(struct dentry *dentry, struct dentry *target)
2855{
2856 struct external_name *old_name = NULL;
2857 if (unlikely(dname_external(dentry)))
2858 old_name = external_name(dentry);
2859 if (unlikely(dname_external(target))) {
2860 atomic_inc(&external_name(target)->u.count);
2861 dentry->d_name = target->d_name;
2862 } else {
2863 memcpy(dentry->d_iname, target->d_name.name,
2864 target->d_name.len + 1);
2865 dentry->d_name.name = dentry->d_iname;
2866 dentry->d_name.hash_len = target->d_name.hash_len;
2867 }
2868 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
Vlastimil Babka2e03b4b2018-10-26 15:05:41 -07002869 kfree_rcu(old_name, u.head);
Al Viro8d85b482014-09-29 14:54:27 -04002870}
2871
Trond Myklebust9eaef272006-10-21 10:24:20 -07002872/*
Al Viro18367502011-07-12 21:42:24 -04002873 * __d_move - move a dentry
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 * @dentry: entry to move
2875 * @target: new dentry
Miklos Szeredida1ce062014-04-01 17:08:43 +02002876 * @exchange: exchange the two dentries
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 *
2878 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002879 * dcache entries should not be moved in this way. Caller must hold
2880 * rename_lock, the i_mutex of the source and target directories,
2881 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882 */
Miklos Szeredida1ce062014-04-01 17:08:43 +02002883static void __d_move(struct dentry *dentry, struct dentry *target,
2884 bool exchange)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885{
Al Viro42177002018-03-11 15:15:46 -04002886 struct dentry *old_parent, *p;
Al Viro84e710d2016-04-15 00:58:55 -04002887 struct inode *dir = NULL;
2888 unsigned n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
Al Viro42177002018-03-11 15:15:46 -04002890 WARN_ON(!dentry->d_inode);
2891 if (WARN_ON(dentry == target))
2892 return;
2893
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002894 BUG_ON(d_ancestor(target, dentry));
Al Viro42177002018-03-11 15:15:46 -04002895 old_parent = dentry->d_parent;
2896 p = d_ancestor(old_parent, target);
2897 if (IS_ROOT(dentry)) {
2898 BUG_ON(p);
2899 spin_lock(&target->d_parent->d_lock);
2900 } else if (!p) {
2901 /* target is not a descendent of dentry->d_parent */
2902 spin_lock(&target->d_parent->d_lock);
2903 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2904 } else {
2905 BUG_ON(p == dentry);
2906 spin_lock(&old_parent->d_lock);
2907 if (p != target)
2908 spin_lock_nested(&target->d_parent->d_lock,
2909 DENTRY_D_LOCK_NESTED);
2910 }
2911 spin_lock_nested(&dentry->d_lock, 2);
2912 spin_lock_nested(&target->d_lock, 3);
Nick Piggin2fd6b7f2011-01-07 17:49:34 +11002913
Al Viro84e710d2016-04-15 00:58:55 -04002914 if (unlikely(d_in_lookup(target))) {
2915 dir = target->d_parent->d_inode;
2916 n = start_dir_add(dir);
Al Viro85c7f812016-04-14 19:52:13 -04002917 __d_lookup_done(target);
Al Viro84e710d2016-04-15 00:58:55 -04002918 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
Nick Piggin31e6b012011-01-07 17:49:52 +11002920 write_seqcount_begin(&dentry->d_seq);
John Stultz1ca7d672013-10-07 15:51:59 -07002921 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
Nick Piggin31e6b012011-01-07 17:49:52 +11002922
Al Viro15d3c582016-07-29 17:45:21 -04002923 /* unhash both */
Al Viro0632a9a2018-03-07 00:49:10 -05002924 if (!d_unhashed(dentry))
2925 ___d_drop(dentry);
2926 if (!d_unhashed(target))
2927 ___d_drop(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
Al Viro63cf4272014-09-26 23:06:14 -04002929 /* ... and switch them in the tree */
Al Viro076515f2018-03-10 23:15:52 -05002930 dentry->d_parent = target->d_parent;
2931 if (!exchange) {
2932 copy_name(dentry, target);
2933 target->d_hash.pprev = NULL;
2934 dentry->d_parent->d_lockref.count++;
Al Viro5467a682019-03-15 22:23:19 -04002935 if (dentry != old_parent) /* wasn't IS_ROOT */
Al Viro076515f2018-03-10 23:15:52 -05002936 WARN_ON(!--old_parent->d_lockref.count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 } else {
Al Viro076515f2018-03-10 23:15:52 -05002938 target->d_parent = old_parent;
2939 swap_names(dentry, target);
Al Viro946e51f2014-10-26 19:19:16 -04002940 list_move(&target->d_child, &target->d_parent->d_subdirs);
Al Viro076515f2018-03-10 23:15:52 -05002941 __d_rehash(target);
2942 fsnotify_update_flags(target);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002943 }
Al Viro076515f2018-03-10 23:15:52 -05002944 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2945 __d_rehash(dentry);
2946 fsnotify_update_flags(dentry);
Eric Biggers0bf3d5c12019-03-20 11:39:11 -07002947 fscrypt_handle_d_move(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948
Nick Piggin31e6b012011-01-07 17:49:52 +11002949 write_seqcount_end(&target->d_seq);
2950 write_seqcount_end(&dentry->d_seq);
2951
Al Viro84e710d2016-04-15 00:58:55 -04002952 if (dir)
2953 end_dir_add(dir, n);
Al Viro076515f2018-03-10 23:15:52 -05002954
2955 if (dentry->d_parent != old_parent)
2956 spin_unlock(&dentry->d_parent->d_lock);
2957 if (dentry != old_parent)
2958 spin_unlock(&old_parent->d_lock);
2959 spin_unlock(&target->d_lock);
2960 spin_unlock(&dentry->d_lock);
Al Viro18367502011-07-12 21:42:24 -04002961}
2962
2963/*
2964 * d_move - move a dentry
2965 * @dentry: entry to move
2966 * @target: new dentry
2967 *
2968 * Update the dcache to reflect the move of a file name. Negative
Jeff Laytonc46c8872011-07-26 13:33:16 -04002969 * dcache entries should not be moved in this way. See the locking
2970 * requirements for __d_move.
Al Viro18367502011-07-12 21:42:24 -04002971 */
2972void d_move(struct dentry *dentry, struct dentry *target)
2973{
2974 write_seqlock(&rename_lock);
Miklos Szeredida1ce062014-04-01 17:08:43 +02002975 __d_move(dentry, target, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07002977}
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07002978EXPORT_SYMBOL(d_move);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979
Miklos Szeredida1ce062014-04-01 17:08:43 +02002980/*
2981 * d_exchange - exchange two dentries
2982 * @dentry1: first dentry
2983 * @dentry2: second dentry
2984 */
2985void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2986{
2987 write_seqlock(&rename_lock);
2988
2989 WARN_ON(!dentry1->d_inode);
2990 WARN_ON(!dentry2->d_inode);
2991 WARN_ON(IS_ROOT(dentry1));
2992 WARN_ON(IS_ROOT(dentry2));
2993
2994 __d_move(dentry1, dentry2, true);
2995
2996 write_sequnlock(&rename_lock);
2997}
2998
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09002999/**
3000 * d_ancestor - search for an ancestor
3001 * @p1: ancestor dentry
3002 * @p2: child dentry
3003 *
3004 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
3005 * an ancestor of p2, else NULL.
Trond Myklebust9eaef272006-10-21 10:24:20 -07003006 */
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003007struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
Trond Myklebust9eaef272006-10-21 10:24:20 -07003008{
3009 struct dentry *p;
3010
OGAWA Hirofumi871c0062008-10-16 07:50:27 +09003011 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
Trond Myklebust9eaef272006-10-21 10:24:20 -07003012 if (p->d_parent == p1)
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003013 return p;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003014 }
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003015 return NULL;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003016}
3017
3018/*
3019 * This helper attempts to cope with remotely renamed directories
3020 *
3021 * It assumes that the caller is already holding
Eric W. Biedermana03e2832015-08-15 13:36:41 -05003022 * dentry->d_parent->d_inode->i_mutex, and rename_lock
Trond Myklebust9eaef272006-10-21 10:24:20 -07003023 *
3024 * Note: If ever the locking in lock_rename() changes, then please
3025 * remember to update this too...
Trond Myklebust9eaef272006-10-21 10:24:20 -07003026 */
Al Virob5ae6b12014-10-12 22:16:02 -04003027static int __d_unalias(struct inode *inode,
Nick Piggin873feea2011-01-07 17:50:06 +11003028 struct dentry *dentry, struct dentry *alias)
Trond Myklebust9eaef272006-10-21 10:24:20 -07003029{
Al Viro9902af72016-04-15 15:08:36 -04003030 struct mutex *m1 = NULL;
3031 struct rw_semaphore *m2 = NULL;
J. Bruce Fields3d330dc2015-02-10 10:55:53 -05003032 int ret = -ESTALE;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003033
3034 /* If alias and dentry share a parent, then no extra locks required */
3035 if (alias->d_parent == dentry->d_parent)
3036 goto out_unalias;
3037
Trond Myklebust9eaef272006-10-21 10:24:20 -07003038 /* See lock_rename() */
Trond Myklebust9eaef272006-10-21 10:24:20 -07003039 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
3040 goto out_err;
3041 m1 = &dentry->d_sb->s_vfs_rename_mutex;
Al Viro9902af72016-04-15 15:08:36 -04003042 if (!inode_trylock_shared(alias->d_parent->d_inode))
Trond Myklebust9eaef272006-10-21 10:24:20 -07003043 goto out_err;
Al Viro9902af72016-04-15 15:08:36 -04003044 m2 = &alias->d_parent->d_inode->i_rwsem;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003045out_unalias:
Eric W. Biederman8ed936b2013-10-01 18:33:48 -07003046 __d_move(alias, dentry, false);
Al Virob5ae6b12014-10-12 22:16:02 -04003047 ret = 0;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003048out_err:
Trond Myklebust9eaef272006-10-21 10:24:20 -07003049 if (m2)
Al Viro9902af72016-04-15 15:08:36 -04003050 up_read(m2);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003051 if (m1)
3052 mutex_unlock(m1);
3053 return ret;
3054}
3055
David Howells770bfad2006-08-22 20:06:07 -04003056/**
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003057 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3058 * @inode: the inode which may have a disconnected dentry
3059 * @dentry: a negative dentry which we want to point to the inode.
3060 *
J. Bruce Fieldsda093a92014-02-17 18:03:57 -05003061 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3062 * place of the given dentry and return it, else simply d_add the inode
3063 * to the dentry and return NULL.
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003064 *
J. Bruce Fields908790f2014-02-17 17:58:42 -05003065 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3066 * we should error out: directories can't have multiple aliases.
3067 *
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003068 * This is needed in the lookup routine of any filesystem that is exportable
3069 * (via knfsd) so that we can build dcache paths to directories effectively.
3070 *
3071 * If a dentry was found and moved, then it is returned. Otherwise NULL
3072 * is returned. This matches the expected return value of ->lookup.
3073 *
3074 * Cluster filesystems may call this function with a negative, hashed dentry.
3075 * In that case, we know that the inode will be a regular file, and also this
3076 * will only occur during atomic_open. So we need to check for the dentry
3077 * being already hashed only in the final case.
3078 */
3079struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3080{
J. Bruce Fields3f70bd52014-02-18 14:11:26 -05003081 if (IS_ERR(inode))
3082 return ERR_CAST(inode);
3083
David Howells770bfad2006-08-22 20:06:07 -04003084 BUG_ON(!d_unhashed(dentry));
3085
Al Virode689f52016-03-09 18:05:42 -05003086 if (!inode)
Al Virob5ae6b12014-10-12 22:16:02 -04003087 goto out;
Al Virode689f52016-03-09 18:05:42 -05003088
Al Virob9680912016-04-11 00:53:26 -04003089 security_d_instantiate(dentry, inode);
Nick Piggin873feea2011-01-07 17:50:06 +11003090 spin_lock(&inode->i_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003091 if (S_ISDIR(inode->i_mode)) {
Al Virob5ae6b12014-10-12 22:16:02 -04003092 struct dentry *new = __d_find_any_alias(inode);
3093 if (unlikely(new)) {
Eric W. Biedermana03e2832015-08-15 13:36:41 -05003094 /* The reference to new ensures it remains an alias */
3095 spin_unlock(&inode->i_lock);
Al Viro18367502011-07-12 21:42:24 -04003096 write_seqlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003097 if (unlikely(d_ancestor(new, dentry))) {
Al Viro18367502011-07-12 21:42:24 -04003098 write_sequnlock(&rename_lock);
Al Virob5ae6b12014-10-12 22:16:02 -04003099 dput(new);
3100 new = ERR_PTR(-ELOOP);
3101 pr_warn_ratelimited(
3102 "VFS: Lookup of '%s' in %s %s"
3103 " would have caused loop\n",
3104 dentry->d_name.name,
3105 inode->i_sb->s_type->name,
3106 inode->i_sb->s_id);
3107 } else if (!IS_ROOT(new)) {
Al Viro076515f2018-03-10 23:15:52 -05003108 struct dentry *old_parent = dget(new->d_parent);
Al Virob5ae6b12014-10-12 22:16:02 -04003109 int err = __d_unalias(inode, dentry, new);
3110 write_sequnlock(&rename_lock);
3111 if (err) {
3112 dput(new);
3113 new = ERR_PTR(err);
3114 }
Al Viro076515f2018-03-10 23:15:52 -05003115 dput(old_parent);
Al Viro18367502011-07-12 21:42:24 -04003116 } else {
Al Virob5ae6b12014-10-12 22:16:02 -04003117 __d_move(new, dentry, false);
3118 write_sequnlock(&rename_lock);
Trond Myklebust9eaef272006-10-21 10:24:20 -07003119 }
Al Virob5ae6b12014-10-12 22:16:02 -04003120 iput(inode);
3121 return new;
Trond Myklebust9eaef272006-10-21 10:24:20 -07003122 }
David Howells770bfad2006-08-22 20:06:07 -04003123 }
Al Virob5ae6b12014-10-12 22:16:02 -04003124out:
Al Viroed782b52016-03-09 19:52:39 -05003125 __d_add(dentry, inode);
Al Virob5ae6b12014-10-12 22:16:02 -04003126 return NULL;
David Howells770bfad2006-08-22 20:06:07 -04003127}
Al Virob5ae6b12014-10-12 22:16:02 -04003128EXPORT_SYMBOL(d_splice_alias);
David Howells770bfad2006-08-22 20:06:07 -04003129
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130/*
3131 * Test whether new_dentry is a subdirectory of old_dentry.
3132 *
3133 * Trivially implemented using the dcache structure
3134 */
3135
3136/**
3137 * is_subdir - is new dentry a subdirectory of old_dentry
3138 * @new_dentry: new dentry
3139 * @old_dentry: old dentry
3140 *
Yaowei Baia6e57872015-11-17 14:40:11 +08003141 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3142 * Returns false otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3144 */
3145
Yaowei Baia6e57872015-11-17 14:40:11 +08003146bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147{
Yaowei Baia6e57872015-11-17 14:40:11 +08003148 bool result;
Nick Piggin949854d2011-01-07 17:49:37 +11003149 unsigned seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003151 if (new_dentry == old_dentry)
Yaowei Baia6e57872015-11-17 14:40:11 +08003152 return true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003153
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003154 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 /* for restarting inner loop in case of seq retry */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003156 seq = read_seqbegin(&rename_lock);
Nick Piggin949854d2011-01-07 17:49:37 +11003157 /*
3158 * Need rcu_readlock to protect against the d_parent trashing
3159 * due to d_move
3160 */
3161 rcu_read_lock();
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003162 if (d_ancestor(old_dentry, new_dentry))
Yaowei Baia6e57872015-11-17 14:40:11 +08003163 result = true;
OGAWA Hirofumie2761a12008-10-16 07:50:28 +09003164 else
Yaowei Baia6e57872015-11-17 14:40:11 +08003165 result = false;
Nick Piggin949854d2011-01-07 17:49:37 +11003166 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 } while (read_seqretry(&rename_lock, seq));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003168
3169 return result;
3170}
Amir Goldsteine8f9e5b2018-01-11 11:33:24 +02003171EXPORT_SYMBOL(is_subdir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003173static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003174{
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003175 struct dentry *root = data;
3176 if (dentry != root) {
3177 if (d_unhashed(dentry) || !dentry->d_inode)
3178 return D_WALK_SKIP;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179
Miklos Szeredi01ddc4e2013-09-05 11:44:34 +02003180 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3181 dentry->d_flags |= DCACHE_GENOCIDE;
3182 dentry->d_lockref.count--;
3183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003184 }
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003185 return D_WALK_CONTINUE;
3186}
Nick Piggin58db63d2011-01-07 17:49:39 +11003187
Miklos Szeredidb14fc32013-09-05 11:44:35 +02003188void d_genocide(struct dentry *parent)
3189{
Al Viro3a8e3612018-04-15 18:27:23 -04003190 d_walk(parent, parent, d_genocide_kill);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003191}
3192
Al Virocbd4a5b2018-03-29 15:08:21 -04003193EXPORT_SYMBOL(d_genocide);
3194
Al Viro60545d02013-06-07 01:20:27 -04003195void d_tmpfile(struct dentry *dentry, struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196{
Al Viro60545d02013-06-07 01:20:27 -04003197 inode_dec_link_count(inode);
3198 BUG_ON(dentry->d_name.name != dentry->d_iname ||
Al Viro946e51f2014-10-26 19:19:16 -04003199 !hlist_unhashed(&dentry->d_u.d_alias) ||
Al Viro60545d02013-06-07 01:20:27 -04003200 !d_unlinked(dentry));
3201 spin_lock(&dentry->d_parent->d_lock);
3202 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3203 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3204 (unsigned long long)inode->i_ino);
3205 spin_unlock(&dentry->d_lock);
3206 spin_unlock(&dentry->d_parent->d_lock);
3207 d_instantiate(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208}
Al Viro60545d02013-06-07 01:20:27 -04003209EXPORT_SYMBOL(d_tmpfile);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210
3211static __initdata unsigned long dhash_entries;
3212static int __init set_dhash_entries(char *str)
3213{
3214 if (!str)
3215 return 0;
3216 dhash_entries = simple_strtoul(str, &str, 0);
3217 return 1;
3218}
3219__setup("dhash_entries=", set_dhash_entries);
3220
3221static void __init dcache_init_early(void)
3222{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003223 /* If hashes are distributed across NUMA nodes, defer
3224 * hash allocation until vmalloc space is available.
3225 */
3226 if (hashdist)
3227 return;
3228
3229 dentry_hashtable =
3230 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003231 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003232 dhash_entries,
3233 13,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003234 HASH_EARLY | HASH_ZERO,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003235 &d_hash_shift,
Alexey Dobriyanb35d7862017-11-20 18:05:52 +03003236 NULL,
Tim Bird31fe62b2012-05-23 13:33:35 +00003237 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 0);
Alexey Dobriyan854d3e62017-11-20 18:05:07 +03003239 d_hash_shift = 32 - d_hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003240}
3241
Denis Cheng74bf17c2007-10-16 23:26:30 -07003242static void __init dcache_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243{
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003244 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245 * A constructor could be added for stable state like the lists,
3246 * but it is probably not worth it because of the cache nature
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003247 * of the dcache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003248 */
David Windsor80344262017-06-10 22:50:44 -04003249 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3250 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3251 d_iname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
3253 /* Hash may have been set up in dcache_init_early */
3254 if (!hashdist)
3255 return;
3256
3257 dentry_hashtable =
3258 alloc_large_system_hash("Dentry cache",
Linus Torvaldsb07ad992011-04-23 22:32:03 -07003259 sizeof(struct hlist_bl_head),
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 dhash_entries,
3261 13,
Pavel Tatashin3d375d72017-07-06 15:39:11 -07003262 HASH_ZERO,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 &d_hash_shift,
Alexey Dobriyanb35d7862017-11-20 18:05:52 +03003264 NULL,
Tim Bird31fe62b2012-05-23 13:33:35 +00003265 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 0);
Alexey Dobriyan854d3e62017-11-20 18:05:07 +03003267 d_hash_shift = 32 - d_hash_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003268}
3269
3270/* SLAB cache for __getname() consumers */
Christoph Lametere18b8902006-12-06 20:33:20 -08003271struct kmem_cache *names_cachep __read_mostly;
H Hartley Sweetenec4f8602010-01-05 13:45:18 -07003272EXPORT_SYMBOL(names_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273
Linus Torvalds1da177e2005-04-16 15:20:36 -07003274void __init vfs_caches_init_early(void)
3275{
Sebastian Andrzej Siewior69163632017-06-27 18:19:11 +02003276 int i;
3277
3278 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3279 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3280
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 dcache_init_early();
3282 inode_init_early();
3283}
3284
Mel Gorman4248b0d2015-08-06 15:46:20 -07003285void __init vfs_caches_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286{
David Windsor6a9b8822017-06-10 22:50:30 -04003287 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3288 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289
Denis Cheng74bf17c2007-10-16 23:26:30 -07003290 dcache_init();
3291 inode_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -07003292 files_init();
3293 files_maxfiles_init();
Denis Cheng74bf17c2007-10-16 23:26:30 -07003294 mnt_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 bdev_cache_init();
3296 chrdev_init();
3297}