1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * fs/dcache.c 3 * fs/dcache.c
4 * 4 *
5 * Complete reimplementation 5 * Complete reimplementation
6 * (C) 1997 Thomas Schoebel-Theuer, 6 * (C) 1997 Thomas Schoebel-Theuer,
7 * with heavy changes by Linus Torvalds 7 * with heavy changes by Linus Torvalds
8 */ 8 */
9 9
10 /* 10 /*
11 * Notes on the allocation strategy: 11 * Notes on the allocation strategy:
12 * 12 *
13 * The dcache is a master of the icache - when 13 * The dcache is a master of the icache - whenever a dcache entry
14 * exists, the inode will always exist. "iput( 14 * exists, the inode will always exist. "iput()" is done either when
15 * the dcache entry is deleted or garbage coll 15 * the dcache entry is deleted or garbage collected.
16 */ 16 */
17 17
18 #include <linux/ratelimit.h> 18 #include <linux/ratelimit.h>
19 #include <linux/string.h> 19 #include <linux/string.h>
20 #include <linux/mm.h> 20 #include <linux/mm.h>
21 #include <linux/fs.h> 21 #include <linux/fs.h>
22 #include <linux/fscrypt.h> 22 #include <linux/fscrypt.h>
23 #include <linux/fsnotify.h> 23 #include <linux/fsnotify.h>
24 #include <linux/slab.h> 24 #include <linux/slab.h>
25 #include <linux/init.h> 25 #include <linux/init.h>
26 #include <linux/hash.h> 26 #include <linux/hash.h>
27 #include <linux/cache.h> 27 #include <linux/cache.h>
28 #include <linux/export.h> 28 #include <linux/export.h>
29 #include <linux/security.h> 29 #include <linux/security.h>
30 #include <linux/seqlock.h> 30 #include <linux/seqlock.h>
31 #include <linux/memblock.h> 31 #include <linux/memblock.h>
32 #include <linux/bit_spinlock.h> 32 #include <linux/bit_spinlock.h>
33 #include <linux/rculist_bl.h> 33 #include <linux/rculist_bl.h>
34 #include <linux/list_lru.h> 34 #include <linux/list_lru.h>
35 #include "internal.h" 35 #include "internal.h"
36 #include "mount.h" 36 #include "mount.h"
37 37
38 #include <asm/runtime-const.h> 38 #include <asm/runtime-const.h>
39 39
40 /* 40 /*
41 * Usage: 41 * Usage:
42 * dcache->d_inode->i_lock protects: 42 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_u.d_alias, d_inode of alias 43 * - i_dentry, d_u.d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects: 44 * dcache_hash_bucket lock protects:
45 * - the dcache hash table 45 * - the dcache hash table
46 * s_roots bl list spinlock protects: 46 * s_roots bl list spinlock protects:
47 * - the s_roots list (see __d_drop) 47 * - the s_roots list (see __d_drop)
48 * dentry->d_sb->s_dentry_lru_lock protects: 48 * dentry->d_sb->s_dentry_lru_lock protects:
49 * - the dcache lru lists and counters 49 * - the dcache lru lists and counters
50 * d_lock protects: 50 * d_lock protects:
51 * - d_flags 51 * - d_flags
52 * - d_name 52 * - d_name
53 * - d_lru 53 * - d_lru
54 * - d_count 54 * - d_count
55 * - d_unhashed() 55 * - d_unhashed()
56 * - d_parent and d_chilren 56 * - d_parent and d_chilren
57 * - childrens' d_sib and d_parent 57 * - childrens' d_sib and d_parent
58 * - d_u.d_alias, d_inode 58 * - d_u.d_alias, d_inode
59 * 59 *
60 * Ordering: 60 * Ordering:
61 * dentry->d_inode->i_lock 61 * dentry->d_inode->i_lock
62 * dentry->d_lock 62 * dentry->d_lock
63 * dentry->d_sb->s_dentry_lru_lock 63 * dentry->d_sb->s_dentry_lru_lock
64 * dcache_hash_bucket lock 64 * dcache_hash_bucket lock
65 * s_roots lock 65 * s_roots lock
66 * 66 *
67 * If there is an ancestor relationship: 67 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock 68 * dentry->d_parent->...->d_parent->d_lock
69 * ... 69 * ...
70 * dentry->d_parent->d_lock 70 * dentry->d_parent->d_lock
71 * dentry->d_lock 71 * dentry->d_lock
72 * 72 *
73 * If no ancestor relationship: 73 * If no ancestor relationship:
74 * arbitrary, since it's serialized on rename_ 74 * arbitrary, since it's serialized on rename_lock
75 */ 75 */
76 int sysctl_vfs_cache_pressure __read_mostly = 76 int sysctl_vfs_cache_pressure __read_mostly = 100;
77 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); 77 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
78 78
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rena 79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
80 80
81 EXPORT_SYMBOL(rename_lock); 81 EXPORT_SYMBOL(rename_lock);
82 82
83 static struct kmem_cache *dentry_cache __ro_af 83 static struct kmem_cache *dentry_cache __ro_after_init;
84 84
85 const struct qstr empty_name = QSTR_INIT("", 0 85 const struct qstr empty_name = QSTR_INIT("", 0);
86 EXPORT_SYMBOL(empty_name); 86 EXPORT_SYMBOL(empty_name);
87 const struct qstr slash_name = QSTR_INIT("/", 87 const struct qstr slash_name = QSTR_INIT("/", 1);
88 EXPORT_SYMBOL(slash_name); 88 EXPORT_SYMBOL(slash_name);
89 const struct qstr dotdot_name = QSTR_INIT(".." 89 const struct qstr dotdot_name = QSTR_INIT("..", 2);
90 EXPORT_SYMBOL(dotdot_name); 90 EXPORT_SYMBOL(dotdot_name);
91 91
92 /* 92 /*
93 * This is the single most critical data struc 93 * This is the single most critical data structure when it comes
94 * to the dcache: the hashtable for lookups. S 94 * to the dcache: the hashtable for lookups. Somebody should try
95 * to make this good - I've just made it work. 95 * to make this good - I've just made it work.
96 * 96 *
97 * This hash-function tries to avoid losing to 97 * This hash-function tries to avoid losing too many bits of hash
98 * information, yet avoid using a prime hash-s 98 * information, yet avoid using a prime hash-size or similar.
99 * 99 *
100 * Marking the variables "used" ensures that t 100 * Marking the variables "used" ensures that the compiler doesn't
101 * optimize them away completely on architectu 101 * optimize them away completely on architectures with runtime
102 * constant infrastructure, this allows debugg 102 * constant infrastructure, this allows debuggers to see their
103 * values. But updating these values has no ef 103 * values. But updating these values has no effect on those arches.
104 */ 104 */
105 105
106 static unsigned int d_hash_shift __ro_after_in 106 static unsigned int d_hash_shift __ro_after_init __used;
107 107
108 static struct hlist_bl_head *dentry_hashtable 108 static struct hlist_bl_head *dentry_hashtable __ro_after_init __used;
109 109
110 static inline struct hlist_bl_head *d_hash(uns 110 static inline struct hlist_bl_head *d_hash(unsigned long hashlen)
111 { 111 {
112 return runtime_const_ptr(dentry_hashta 112 return runtime_const_ptr(dentry_hashtable) +
113 runtime_const_shift_right_32(h 113 runtime_const_shift_right_32(hashlen, d_hash_shift);
114 } 114 }
115 115
116 #define IN_LOOKUP_SHIFT 10 116 #define IN_LOOKUP_SHIFT 10
117 static struct hlist_bl_head in_lookup_hashtabl 117 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
118 118
119 static inline struct hlist_bl_head *in_lookup_ 119 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
120 unsign 120 unsigned int hash)
121 { 121 {
122 hash += (unsigned long) parent / L1_CA 122 hash += (unsigned long) parent / L1_CACHE_BYTES;
123 return in_lookup_hashtable + hash_32(h 123 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
124 } 124 }
125 125
126 struct dentry_stat_t { 126 struct dentry_stat_t {
127 long nr_dentry; 127 long nr_dentry;
128 long nr_unused; 128 long nr_unused;
129 long age_limit; /* age in seco 129 long age_limit; /* age in seconds */
130 long want_pages; /* pages reque 130 long want_pages; /* pages requested by system */
131 long nr_negative; /* # of unused 131 long nr_negative; /* # of unused negative dentries */
132 long dummy; /* Reserved fo 132 long dummy; /* Reserved for future use */
133 }; 133 };
134 134
135 static DEFINE_PER_CPU(long, nr_dentry); 135 static DEFINE_PER_CPU(long, nr_dentry);
136 static DEFINE_PER_CPU(long, nr_dentry_unused); 136 static DEFINE_PER_CPU(long, nr_dentry_unused);
137 static DEFINE_PER_CPU(long, nr_dentry_negative 137 static DEFINE_PER_CPU(long, nr_dentry_negative);
138 138
139 #if defined(CONFIG_SYSCTL) && defined(CONFIG_P 139 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
140 /* Statistics gathering. */ 140 /* Statistics gathering. */
141 static struct dentry_stat_t dentry_stat = { 141 static struct dentry_stat_t dentry_stat = {
142 .age_limit = 45, 142 .age_limit = 45,
143 }; 143 };
144 144
145 /* 145 /*
146 * Here we resort to our own counters instead 146 * Here we resort to our own counters instead of using generic per-cpu counters
147 * for consistency with what the vfs inode cod 147 * for consistency with what the vfs inode code does. We are expected to harvest
148 * better code and performance by having our o 148 * better code and performance by having our own specialized counters.
149 * 149 *
150 * Please note that the loop is done over all 150 * Please note that the loop is done over all possible CPUs, not over all online
151 * CPUs. The reason for this is that we don't 151 * CPUs. The reason for this is that we don't want to play games with CPUs going
152 * on and off. If one of them goes off, we wil 152 * on and off. If one of them goes off, we will just keep their counters.
153 * 153 *
154 * glommer: See cffbc8a for details, and if yo 154 * glommer: See cffbc8a for details, and if you ever intend to change this,
155 * please update all vfs counters to match. 155 * please update all vfs counters to match.
156 */ 156 */
157 static long get_nr_dentry(void) 157 static long get_nr_dentry(void)
158 { 158 {
159 int i; 159 int i;
160 long sum = 0; 160 long sum = 0;
161 for_each_possible_cpu(i) 161 for_each_possible_cpu(i)
162 sum += per_cpu(nr_dentry, i); 162 sum += per_cpu(nr_dentry, i);
163 return sum < 0 ? 0 : sum; 163 return sum < 0 ? 0 : sum;
164 } 164 }
165 165
166 static long get_nr_dentry_unused(void) 166 static long get_nr_dentry_unused(void)
167 { 167 {
168 int i; 168 int i;
169 long sum = 0; 169 long sum = 0;
170 for_each_possible_cpu(i) 170 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry_unuse 171 sum += per_cpu(nr_dentry_unused, i);
172 return sum < 0 ? 0 : sum; 172 return sum < 0 ? 0 : sum;
173 } 173 }
174 174
175 static long get_nr_dentry_negative(void) 175 static long get_nr_dentry_negative(void)
176 { 176 {
177 int i; 177 int i;
178 long sum = 0; 178 long sum = 0;
179 179
180 for_each_possible_cpu(i) 180 for_each_possible_cpu(i)
181 sum += per_cpu(nr_dentry_negat 181 sum += per_cpu(nr_dentry_negative, i);
182 return sum < 0 ? 0 : sum; 182 return sum < 0 ? 0 : sum;
183 } 183 }
184 184
185 static int proc_nr_dentry(const struct ctl_tab 185 static int proc_nr_dentry(const struct ctl_table *table, int write, void *buffer,
186 size_t *lenp, loff_t 186 size_t *lenp, loff_t *ppos)
187 { 187 {
188 dentry_stat.nr_dentry = get_nr_dentry( 188 dentry_stat.nr_dentry = get_nr_dentry();
189 dentry_stat.nr_unused = get_nr_dentry_ 189 dentry_stat.nr_unused = get_nr_dentry_unused();
190 dentry_stat.nr_negative = get_nr_dentr 190 dentry_stat.nr_negative = get_nr_dentry_negative();
191 return proc_doulongvec_minmax(table, w 191 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
192 } 192 }
193 193
194 static struct ctl_table fs_dcache_sysctls[] = 194 static struct ctl_table fs_dcache_sysctls[] = {
195 { 195 {
196 .procname = "dentry-stat 196 .procname = "dentry-state",
197 .data = &dentry_stat 197 .data = &dentry_stat,
198 .maxlen = 6*sizeof(lon 198 .maxlen = 6*sizeof(long),
199 .mode = 0444, 199 .mode = 0444,
200 .proc_handler = proc_nr_dent 200 .proc_handler = proc_nr_dentry,
201 }, 201 },
202 }; 202 };
203 203
204 static int __init init_fs_dcache_sysctls(void) 204 static int __init init_fs_dcache_sysctls(void)
205 { 205 {
206 register_sysctl_init("fs", fs_dcache_s 206 register_sysctl_init("fs", fs_dcache_sysctls);
207 return 0; 207 return 0;
208 } 208 }
209 fs_initcall(init_fs_dcache_sysctls); 209 fs_initcall(init_fs_dcache_sysctls);
210 #endif 210 #endif
211 211
212 /* 212 /*
213 * Compare 2 name strings, return 0 if they ma 213 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
214 * The strings are both count bytes long, and 214 * The strings are both count bytes long, and count is non-zero.
215 */ 215 */
216 #ifdef CONFIG_DCACHE_WORD_ACCESS 216 #ifdef CONFIG_DCACHE_WORD_ACCESS
217 217
218 #include <asm/word-at-a-time.h> 218 #include <asm/word-at-a-time.h>
219 /* 219 /*
220 * NOTE! 'cs' and 'scount' come from a dentry, 220 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
221 * aligned allocation for this particular comp 221 * aligned allocation for this particular component. We don't
222 * strictly need the load_unaligned_zeropad() 222 * strictly need the load_unaligned_zeropad() safety, but it
223 * doesn't hurt either. 223 * doesn't hurt either.
224 * 224 *
225 * In contrast, 'ct' and 'tcount' can be from 225 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
226 * need the careful unaligned handling. 226 * need the careful unaligned handling.
227 */ 227 */
228 static inline int dentry_string_cmp(const unsi 228 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
229 { 229 {
230 unsigned long a,b,mask; 230 unsigned long a,b,mask;
231 231
232 for (;;) { 232 for (;;) {
233 a = read_word_at_a_time(cs); 233 a = read_word_at_a_time(cs);
234 b = load_unaligned_zeropad(ct) 234 b = load_unaligned_zeropad(ct);
235 if (tcount < sizeof(unsigned l 235 if (tcount < sizeof(unsigned long))
236 break; 236 break;
237 if (unlikely(a != b)) 237 if (unlikely(a != b))
238 return 1; 238 return 1;
239 cs += sizeof(unsigned long); 239 cs += sizeof(unsigned long);
240 ct += sizeof(unsigned long); 240 ct += sizeof(unsigned long);
241 tcount -= sizeof(unsigned long 241 tcount -= sizeof(unsigned long);
242 if (!tcount) 242 if (!tcount)
243 return 0; 243 return 0;
244 } 244 }
245 mask = bytemask_from_count(tcount); 245 mask = bytemask_from_count(tcount);
246 return unlikely(!!((a ^ b) & mask)); 246 return unlikely(!!((a ^ b) & mask));
247 } 247 }
248 248
249 #else 249 #else
250 250
251 static inline int dentry_string_cmp(const unsi 251 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
252 { 252 {
253 do { 253 do {
254 if (*cs != *ct) 254 if (*cs != *ct)
255 return 1; 255 return 1;
256 cs++; 256 cs++;
257 ct++; 257 ct++;
258 tcount--; 258 tcount--;
259 } while (tcount); 259 } while (tcount);
260 return 0; 260 return 0;
261 } 261 }
262 262
263 #endif 263 #endif
264 264
265 static inline int dentry_cmp(const struct dent 265 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
266 { 266 {
267 /* 267 /*
268 * Be careful about RCU walk racing wi 268 * Be careful about RCU walk racing with rename:
269 * use 'READ_ONCE' to fetch the name p 269 * use 'READ_ONCE' to fetch the name pointer.
270 * 270 *
271 * NOTE! Even if a rename will mean th 271 * NOTE! Even if a rename will mean that the length
272 * was not loaded atomically, we don't 272 * was not loaded atomically, we don't care. The
273 * RCU walk will check the sequence co 273 * RCU walk will check the sequence count eventually,
274 * and catch it. And we won't overrun 274 * and catch it. And we won't overrun the buffer,
275 * because we're reading the name poin 275 * because we're reading the name pointer atomically,
276 * and a dentry name is guaranteed to 276 * and a dentry name is guaranteed to be properly
277 * terminated with a NUL byte. 277 * terminated with a NUL byte.
278 * 278 *
279 * End result: even if 'len' is wrong, 279 * End result: even if 'len' is wrong, we'll exit
280 * early because the data cannot match 280 * early because the data cannot match (there can
281 * be no NUL in the ct/tcount data) 281 * be no NUL in the ct/tcount data)
282 */ 282 */
283 const unsigned char *cs = READ_ONCE(de 283 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
284 284
285 return dentry_string_cmp(cs, ct, tcoun 285 return dentry_string_cmp(cs, ct, tcount);
286 } 286 }
287 287
288 struct external_name { 288 struct external_name {
289 union { 289 union {
290 atomic_t count; 290 atomic_t count;
291 struct rcu_head head; 291 struct rcu_head head;
292 } u; 292 } u;
293 unsigned char name[]; 293 unsigned char name[];
294 }; 294 };
295 295
296 static inline struct external_name *external_n 296 static inline struct external_name *external_name(struct dentry *dentry)
297 { 297 {
298 return container_of(dentry->d_name.nam 298 return container_of(dentry->d_name.name, struct external_name, name[0]);
299 } 299 }
300 300
301 static void __d_free(struct rcu_head *head) 301 static void __d_free(struct rcu_head *head)
302 { 302 {
303 struct dentry *dentry = container_of(h 303 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
304 304
305 kmem_cache_free(dentry_cache, dentry); 305 kmem_cache_free(dentry_cache, dentry);
306 } 306 }
307 307
308 static void __d_free_external(struct rcu_head 308 static void __d_free_external(struct rcu_head *head)
309 { 309 {
310 struct dentry *dentry = container_of(h 310 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
311 kfree(external_name(dentry)); 311 kfree(external_name(dentry));
312 kmem_cache_free(dentry_cache, dentry); 312 kmem_cache_free(dentry_cache, dentry);
313 } 313 }
314 314
315 static inline int dname_external(const struct 315 static inline int dname_external(const struct dentry *dentry)
316 { 316 {
317 return dentry->d_name.name != dentry-> 317 return dentry->d_name.name != dentry->d_iname;
318 } 318 }
319 319
320 void take_dentry_name_snapshot(struct name_sna 320 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
321 { 321 {
322 spin_lock(&dentry->d_lock); 322 spin_lock(&dentry->d_lock);
323 name->name = dentry->d_name; 323 name->name = dentry->d_name;
324 if (unlikely(dname_external(dentry))) 324 if (unlikely(dname_external(dentry))) {
325 atomic_inc(&external_name(dent 325 atomic_inc(&external_name(dentry)->u.count);
326 } else { 326 } else {
327 memcpy(name->inline_name, dent 327 memcpy(name->inline_name, dentry->d_iname,
328 dentry->d_name.len + 1) 328 dentry->d_name.len + 1);
329 name->name.name = name->inline 329 name->name.name = name->inline_name;
330 } 330 }
331 spin_unlock(&dentry->d_lock); 331 spin_unlock(&dentry->d_lock);
332 } 332 }
333 EXPORT_SYMBOL(take_dentry_name_snapshot); 333 EXPORT_SYMBOL(take_dentry_name_snapshot);
334 334
335 void release_dentry_name_snapshot(struct name_ 335 void release_dentry_name_snapshot(struct name_snapshot *name)
336 { 336 {
337 if (unlikely(name->name.name != name-> 337 if (unlikely(name->name.name != name->inline_name)) {
338 struct external_name *p; 338 struct external_name *p;
339 p = container_of(name->name.na 339 p = container_of(name->name.name, struct external_name, name[0]);
340 if (unlikely(atomic_dec_and_te 340 if (unlikely(atomic_dec_and_test(&p->u.count)))
341 kfree_rcu(p, u.head); 341 kfree_rcu(p, u.head);
342 } 342 }
343 } 343 }
344 EXPORT_SYMBOL(release_dentry_name_snapshot); 344 EXPORT_SYMBOL(release_dentry_name_snapshot);
345 345
346 static inline void __d_set_inode_and_type(stru 346 static inline void __d_set_inode_and_type(struct dentry *dentry,
347 stru 347 struct inode *inode,
348 unsi 348 unsigned type_flags)
349 { 349 {
350 unsigned flags; 350 unsigned flags;
351 351
352 dentry->d_inode = inode; 352 dentry->d_inode = inode;
353 flags = READ_ONCE(dentry->d_flags); 353 flags = READ_ONCE(dentry->d_flags);
354 flags &= ~DCACHE_ENTRY_TYPE; 354 flags &= ~DCACHE_ENTRY_TYPE;
355 flags |= type_flags; 355 flags |= type_flags;
356 smp_store_release(&dentry->d_flags, fl 356 smp_store_release(&dentry->d_flags, flags);
357 } 357 }
358 358
359 static inline void __d_clear_type_and_inode(st 359 static inline void __d_clear_type_and_inode(struct dentry *dentry)
360 { 360 {
361 unsigned flags = READ_ONCE(dentry->d_f 361 unsigned flags = READ_ONCE(dentry->d_flags);
362 362
363 flags &= ~DCACHE_ENTRY_TYPE; 363 flags &= ~DCACHE_ENTRY_TYPE;
364 WRITE_ONCE(dentry->d_flags, flags); 364 WRITE_ONCE(dentry->d_flags, flags);
365 dentry->d_inode = NULL; 365 dentry->d_inode = NULL;
366 /* 366 /*
367 * The negative counter only tracks de 367 * The negative counter only tracks dentries on the LRU. Don't inc if
368 * d_lru is on another list. 368 * d_lru is on another list.
369 */ 369 */
370 if ((flags & (DCACHE_LRU_LIST|DCACHE_S 370 if ((flags & (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
371 this_cpu_inc(nr_dentry_negativ 371 this_cpu_inc(nr_dentry_negative);
372 } 372 }
373 373
374 static void dentry_free(struct dentry *dentry) 374 static void dentry_free(struct dentry *dentry)
375 { 375 {
376 WARN_ON(!hlist_unhashed(&dentry->d_u.d 376 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
377 if (unlikely(dname_external(dentry))) 377 if (unlikely(dname_external(dentry))) {
378 struct external_name *p = exte 378 struct external_name *p = external_name(dentry);
379 if (likely(atomic_dec_and_test 379 if (likely(atomic_dec_and_test(&p->u.count))) {
380 call_rcu(&dentry->d_u. 380 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
381 return; 381 return;
382 } 382 }
383 } 383 }
384 /* if dentry was never visible to RCU, 384 /* if dentry was never visible to RCU, immediate free is OK */
385 if (dentry->d_flags & DCACHE_NORCU) 385 if (dentry->d_flags & DCACHE_NORCU)
386 __d_free(&dentry->d_u.d_rcu); 386 __d_free(&dentry->d_u.d_rcu);
387 else 387 else
388 call_rcu(&dentry->d_u.d_rcu, _ 388 call_rcu(&dentry->d_u.d_rcu, __d_free);
389 } 389 }
390 390
391 /* 391 /*
392 * Release the dentry's inode, using the files 392 * Release the dentry's inode, using the filesystem
393 * d_iput() operation if defined. 393 * d_iput() operation if defined.
394 */ 394 */
395 static void dentry_unlink_inode(struct dentry 395 static void dentry_unlink_inode(struct dentry * dentry)
396 __releases(dentry->d_lock) 396 __releases(dentry->d_lock)
397 __releases(dentry->d_inode->i_lock) 397 __releases(dentry->d_inode->i_lock)
398 { 398 {
399 struct inode *inode = dentry->d_inode; 399 struct inode *inode = dentry->d_inode;
400 400
401 raw_write_seqcount_begin(&dentry->d_se 401 raw_write_seqcount_begin(&dentry->d_seq);
402 __d_clear_type_and_inode(dentry); 402 __d_clear_type_and_inode(dentry);
403 hlist_del_init(&dentry->d_u.d_alias); 403 hlist_del_init(&dentry->d_u.d_alias);
404 raw_write_seqcount_end(&dentry->d_seq) 404 raw_write_seqcount_end(&dentry->d_seq);
405 spin_unlock(&dentry->d_lock); 405 spin_unlock(&dentry->d_lock);
406 spin_unlock(&inode->i_lock); 406 spin_unlock(&inode->i_lock);
407 if (!inode->i_nlink) 407 if (!inode->i_nlink)
408 fsnotify_inoderemove(inode); 408 fsnotify_inoderemove(inode);
409 if (dentry->d_op && dentry->d_op->d_ip 409 if (dentry->d_op && dentry->d_op->d_iput)
410 dentry->d_op->d_iput(dentry, i 410 dentry->d_op->d_iput(dentry, inode);
411 else 411 else
412 iput(inode); 412 iput(inode);
413 } 413 }
414 414
415 /* 415 /*
416 * The DCACHE_LRU_LIST bit is set whenever the 416 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
417 * is in use - which includes both the "real" 417 * is in use - which includes both the "real" per-superblock
418 * LRU list _and_ the DCACHE_SHRINK_LIST use. 418 * LRU list _and_ the DCACHE_SHRINK_LIST use.
419 * 419 *
420 * The DCACHE_SHRINK_LIST bit is set whenever 420 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
421 * on the shrink list (ie not on the superbloc 421 * on the shrink list (ie not on the superblock LRU list).
422 * 422 *
423 * The per-cpu "nr_dentry_unused" counters are 423 * The per-cpu "nr_dentry_unused" counters are updated with
424 * the DCACHE_LRU_LIST bit. 424 * the DCACHE_LRU_LIST bit.
425 * 425 *
426 * The per-cpu "nr_dentry_negative" counters a 426 * The per-cpu "nr_dentry_negative" counters are only updated
427 * when deleted from or added to the per-super 427 * when deleted from or added to the per-superblock LRU list, not
428 * from/to the shrink list. That is to avoid a 428 * from/to the shrink list. That is to avoid an unneeded dec/inc
429 * pair when moving from LRU to shrink list in 429 * pair when moving from LRU to shrink list in select_collect().
430 * 430 *
431 * These helper functions make sure we always 431 * These helper functions make sure we always follow the
432 * rules. d_lock must be held by the caller. 432 * rules. d_lock must be held by the caller.
433 */ 433 */
434 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(( 434 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
435 static void d_lru_add(struct dentry *dentry) 435 static void d_lru_add(struct dentry *dentry)
436 { 436 {
437 D_FLAG_VERIFY(dentry, 0); 437 D_FLAG_VERIFY(dentry, 0);
438 dentry->d_flags |= DCACHE_LRU_LIST; 438 dentry->d_flags |= DCACHE_LRU_LIST;
439 this_cpu_inc(nr_dentry_unused); 439 this_cpu_inc(nr_dentry_unused);
440 if (d_is_negative(dentry)) 440 if (d_is_negative(dentry))
441 this_cpu_inc(nr_dentry_negativ 441 this_cpu_inc(nr_dentry_negative);
442 WARN_ON_ONCE(!list_lru_add_obj( 442 WARN_ON_ONCE(!list_lru_add_obj(
443 &dentry->d_sb->s_dentr 443 &dentry->d_sb->s_dentry_lru, &dentry->d_lru));
444 } 444 }
445 445
446 static void d_lru_del(struct dentry *dentry) 446 static void d_lru_del(struct dentry *dentry)
447 { 447 {
448 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 448 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
449 dentry->d_flags &= ~DCACHE_LRU_LIST; 449 dentry->d_flags &= ~DCACHE_LRU_LIST;
450 this_cpu_dec(nr_dentry_unused); 450 this_cpu_dec(nr_dentry_unused);
451 if (d_is_negative(dentry)) 451 if (d_is_negative(dentry))
452 this_cpu_dec(nr_dentry_negativ 452 this_cpu_dec(nr_dentry_negative);
453 WARN_ON_ONCE(!list_lru_del_obj( 453 WARN_ON_ONCE(!list_lru_del_obj(
454 &dentry->d_sb->s_dentr 454 &dentry->d_sb->s_dentry_lru, &dentry->d_lru));
455 } 455 }
456 456
457 static void d_shrink_del(struct dentry *dentry 457 static void d_shrink_del(struct dentry *dentry)
458 { 458 {
459 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LI 459 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
460 list_del_init(&dentry->d_lru); 460 list_del_init(&dentry->d_lru);
461 dentry->d_flags &= ~(DCACHE_SHRINK_LIS 461 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
462 this_cpu_dec(nr_dentry_unused); 462 this_cpu_dec(nr_dentry_unused);
463 } 463 }
464 464
465 static void d_shrink_add(struct dentry *dentry 465 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
466 { 466 {
467 D_FLAG_VERIFY(dentry, 0); 467 D_FLAG_VERIFY(dentry, 0);
468 list_add(&dentry->d_lru, list); 468 list_add(&dentry->d_lru, list);
469 dentry->d_flags |= DCACHE_SHRINK_LIST 469 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
470 this_cpu_inc(nr_dentry_unused); 470 this_cpu_inc(nr_dentry_unused);
471 } 471 }
472 472
473 /* 473 /*
474 * These can only be called under the global L 474 * These can only be called under the global LRU lock, ie during the
475 * callback for freeing the LRU list. "isolate 475 * callback for freeing the LRU list. "isolate" removes it from the
476 * LRU lists entirely, while shrink_move moves 476 * LRU lists entirely, while shrink_move moves it to the indicated
477 * private list. 477 * private list.
478 */ 478 */
479 static void d_lru_isolate(struct list_lru_one 479 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
480 { 480 {
481 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 481 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
482 dentry->d_flags &= ~DCACHE_LRU_LIST; 482 dentry->d_flags &= ~DCACHE_LRU_LIST;
483 this_cpu_dec(nr_dentry_unused); 483 this_cpu_dec(nr_dentry_unused);
484 if (d_is_negative(dentry)) 484 if (d_is_negative(dentry))
485 this_cpu_dec(nr_dentry_negativ 485 this_cpu_dec(nr_dentry_negative);
486 list_lru_isolate(lru, &dentry->d_lru); 486 list_lru_isolate(lru, &dentry->d_lru);
487 } 487 }
488 488
489 static void d_lru_shrink_move(struct list_lru_ 489 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
490 struct list_head 490 struct list_head *list)
491 { 491 {
492 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 492 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
493 dentry->d_flags |= DCACHE_SHRINK_LIST; 493 dentry->d_flags |= DCACHE_SHRINK_LIST;
494 if (d_is_negative(dentry)) 494 if (d_is_negative(dentry))
495 this_cpu_dec(nr_dentry_negativ 495 this_cpu_dec(nr_dentry_negative);
496 list_lru_isolate_move(lru, &dentry->d_ 496 list_lru_isolate_move(lru, &dentry->d_lru, list);
497 } 497 }
498 498
499 static void ___d_drop(struct dentry *dentry) 499 static void ___d_drop(struct dentry *dentry)
500 { 500 {
501 struct hlist_bl_head *b; 501 struct hlist_bl_head *b;
502 /* 502 /*
503 * Hashed dentries are normally on the 503 * Hashed dentries are normally on the dentry hashtable,
504 * with the exception of those newly a 504 * with the exception of those newly allocated by
505 * d_obtain_root, which are always IS_ 505 * d_obtain_root, which are always IS_ROOT:
506 */ 506 */
507 if (unlikely(IS_ROOT(dentry))) 507 if (unlikely(IS_ROOT(dentry)))
508 b = &dentry->d_sb->s_roots; 508 b = &dentry->d_sb->s_roots;
509 else 509 else
510 b = d_hash(dentry->d_name.hash 510 b = d_hash(dentry->d_name.hash);
511 511
512 hlist_bl_lock(b); 512 hlist_bl_lock(b);
513 __hlist_bl_del(&dentry->d_hash); 513 __hlist_bl_del(&dentry->d_hash);
514 hlist_bl_unlock(b); 514 hlist_bl_unlock(b);
515 } 515 }
516 516
517 void __d_drop(struct dentry *dentry) 517 void __d_drop(struct dentry *dentry)
518 { 518 {
519 if (!d_unhashed(dentry)) { 519 if (!d_unhashed(dentry)) {
520 ___d_drop(dentry); 520 ___d_drop(dentry);
521 dentry->d_hash.pprev = NULL; 521 dentry->d_hash.pprev = NULL;
522 write_seqcount_invalidate(&den 522 write_seqcount_invalidate(&dentry->d_seq);
523 } 523 }
524 } 524 }
525 EXPORT_SYMBOL(__d_drop); 525 EXPORT_SYMBOL(__d_drop);
526 526
527 /** 527 /**
528 * d_drop - drop a dentry 528 * d_drop - drop a dentry
529 * @dentry: dentry to drop 529 * @dentry: dentry to drop
530 * 530 *
531 * d_drop() unhashes the entry from the parent 531 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
532 * be found through a VFS lookup any more. Not 532 * be found through a VFS lookup any more. Note that this is different from
533 * deleting the dentry - d_delete will try to 533 * deleting the dentry - d_delete will try to mark the dentry negative if
534 * possible, giving a successful _negative_ lo 534 * possible, giving a successful _negative_ lookup, while d_drop will
535 * just make the cache lookup fail. 535 * just make the cache lookup fail.
536 * 536 *
537 * d_drop() is used mainly for stuff that want 537 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
538 * reason (NFS timeouts or autofs deletes). 538 * reason (NFS timeouts or autofs deletes).
539 * 539 *
540 * __d_drop requires dentry->d_lock 540 * __d_drop requires dentry->d_lock
541 * 541 *
542 * ___d_drop doesn't mark dentry as "unhashed" 542 * ___d_drop doesn't mark dentry as "unhashed"
543 * (dentry->d_hash.pprev will be LIST_POISON2, 543 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
544 */ 544 */
545 void d_drop(struct dentry *dentry) 545 void d_drop(struct dentry *dentry)
546 { 546 {
547 spin_lock(&dentry->d_lock); 547 spin_lock(&dentry->d_lock);
548 __d_drop(dentry); 548 __d_drop(dentry);
549 spin_unlock(&dentry->d_lock); 549 spin_unlock(&dentry->d_lock);
550 } 550 }
551 EXPORT_SYMBOL(d_drop); 551 EXPORT_SYMBOL(d_drop);
552 552
553 static inline void dentry_unlist(struct dentry 553 static inline void dentry_unlist(struct dentry *dentry)
554 { 554 {
555 struct dentry *next; 555 struct dentry *next;
556 /* 556 /*
557 * Inform d_walk() and shrink_dentry_l 557 * Inform d_walk() and shrink_dentry_list() that we are no longer
558 * attached to the dentry tree 558 * attached to the dentry tree
559 */ 559 */
560 dentry->d_flags |= DCACHE_DENTRY_KILLE 560 dentry->d_flags |= DCACHE_DENTRY_KILLED;
561 if (unlikely(hlist_unhashed(&dentry->d 561 if (unlikely(hlist_unhashed(&dentry->d_sib)))
562 return; 562 return;
563 __hlist_del(&dentry->d_sib); 563 __hlist_del(&dentry->d_sib);
564 /* 564 /*
565 * Cursors can move around the list of 565 * Cursors can move around the list of children. While we'd been
566 * a normal list member, it didn't mat 566 * a normal list member, it didn't matter - ->d_sib.next would've
567 * been updated. However, from now on 567 * been updated. However, from now on it won't be and for the
568 * things like d_walk() it might end u 568 * things like d_walk() it might end up with a nasty surprise.
569 * Normally d_walk() doesn't care abou 569 * Normally d_walk() doesn't care about cursors moving around -
570 * ->d_lock on parent prevents that an 570 * ->d_lock on parent prevents that and since a cursor has no children
571 * of its own, we get through it witho 571 * of its own, we get through it without ever unlocking the parent.
572 * There is one exception, though - if 572 * There is one exception, though - if we ascend from a child that
573 * gets killed as soon as we unlock it 573 * gets killed as soon as we unlock it, the next sibling is found
574 * using the value left in its ->d_sib 574 * using the value left in its ->d_sib.next. And if _that_
575 * pointed to a cursor, and cursor got 575 * pointed to a cursor, and cursor got moved (e.g. by lseek())
576 * before d_walk() regains parent->d_l 576 * before d_walk() regains parent->d_lock, we'll end up skipping
577 * everything the cursor had been move 577 * everything the cursor had been moved past.
578 * 578 *
579 * Solution: make sure that the pointe 579 * Solution: make sure that the pointer left behind in ->d_sib.next
580 * points to something that won't be m 580 * points to something that won't be moving around. I.e. skip the
581 * cursors. 581 * cursors.
582 */ 582 */
583 while (dentry->d_sib.next) { 583 while (dentry->d_sib.next) {
584 next = hlist_entry(dentry->d_s 584 next = hlist_entry(dentry->d_sib.next, struct dentry, d_sib);
585 if (likely(!(next->d_flags & D 585 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
586 break; 586 break;
587 dentry->d_sib.next = next->d_s 587 dentry->d_sib.next = next->d_sib.next;
588 } 588 }
589 } 589 }
590 590
591 static struct dentry *__dentry_kill(struct den 591 static struct dentry *__dentry_kill(struct dentry *dentry)
592 { 592 {
593 struct dentry *parent = NULL; 593 struct dentry *parent = NULL;
594 bool can_free = true; 594 bool can_free = true;
595 595
596 /* 596 /*
597 * The dentry is now unrecoverably dea 597 * The dentry is now unrecoverably dead to the world.
598 */ 598 */
599 lockref_mark_dead(&dentry->d_lockref); 599 lockref_mark_dead(&dentry->d_lockref);
600 600
601 /* 601 /*
602 * inform the fs via d_prune that this 602 * inform the fs via d_prune that this dentry is about to be
603 * unhashed and destroyed. 603 * unhashed and destroyed.
604 */ 604 */
605 if (dentry->d_flags & DCACHE_OP_PRUNE) 605 if (dentry->d_flags & DCACHE_OP_PRUNE)
606 dentry->d_op->d_prune(dentry); 606 dentry->d_op->d_prune(dentry);
607 607
608 if (dentry->d_flags & DCACHE_LRU_LIST) 608 if (dentry->d_flags & DCACHE_LRU_LIST) {
609 if (!(dentry->d_flags & DCACHE 609 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
610 d_lru_del(dentry); 610 d_lru_del(dentry);
611 } 611 }
612 /* if it was on the hash then remove i 612 /* if it was on the hash then remove it */
613 __d_drop(dentry); 613 __d_drop(dentry);
614 if (dentry->d_inode) 614 if (dentry->d_inode)
615 dentry_unlink_inode(dentry); 615 dentry_unlink_inode(dentry);
616 else 616 else
617 spin_unlock(&dentry->d_lock); 617 spin_unlock(&dentry->d_lock);
618 this_cpu_dec(nr_dentry); 618 this_cpu_dec(nr_dentry);
619 if (dentry->d_op && dentry->d_op->d_re 619 if (dentry->d_op && dentry->d_op->d_release)
620 dentry->d_op->d_release(dentry 620 dentry->d_op->d_release(dentry);
621 621
622 cond_resched(); 622 cond_resched();
623 /* now that it's negative, ->d_parent 623 /* now that it's negative, ->d_parent is stable */
624 if (!IS_ROOT(dentry)) { 624 if (!IS_ROOT(dentry)) {
625 parent = dentry->d_parent; 625 parent = dentry->d_parent;
626 spin_lock(&parent->d_lock); 626 spin_lock(&parent->d_lock);
627 } 627 }
628 spin_lock_nested(&dentry->d_lock, DENT 628 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
629 dentry_unlist(dentry); 629 dentry_unlist(dentry);
630 if (dentry->d_flags & DCACHE_SHRINK_LI 630 if (dentry->d_flags & DCACHE_SHRINK_LIST)
631 can_free = false; 631 can_free = false;
632 spin_unlock(&dentry->d_lock); 632 spin_unlock(&dentry->d_lock);
633 if (likely(can_free)) 633 if (likely(can_free))
634 dentry_free(dentry); 634 dentry_free(dentry);
635 if (parent && --parent->d_lockref.coun 635 if (parent && --parent->d_lockref.count) {
636 spin_unlock(&parent->d_lock); 636 spin_unlock(&parent->d_lock);
637 return NULL; 637 return NULL;
638 } 638 }
639 return parent; 639 return parent;
640 } 640 }
641 641
642 /* 642 /*
643 * Lock a dentry for feeding it to __dentry_ki 643 * Lock a dentry for feeding it to __dentry_kill().
644 * Called under rcu_read_lock() and dentry->d_ 644 * Called under rcu_read_lock() and dentry->d_lock; the former
645 * guarantees that nothing we access will be f 645 * guarantees that nothing we access will be freed under us.
646 * Note that dentry is *not* protected from co 646 * Note that dentry is *not* protected from concurrent dentry_kill(),
647 * d_delete(), etc. 647 * d_delete(), etc.
648 * 648 *
649 * Return false if dentry is busy. Otherwise, 649 * Return false if dentry is busy. Otherwise, return true and have
650 * that dentry's inode locked. 650 * that dentry's inode locked.
651 */ 651 */
652 652
653 static bool lock_for_kill(struct dentry *dentr 653 static bool lock_for_kill(struct dentry *dentry)
654 { 654 {
655 struct inode *inode = dentry->d_inode; 655 struct inode *inode = dentry->d_inode;
656 656
657 if (unlikely(dentry->d_lockref.count)) 657 if (unlikely(dentry->d_lockref.count))
658 return false; 658 return false;
659 659
660 if (!inode || likely(spin_trylock(&ino 660 if (!inode || likely(spin_trylock(&inode->i_lock)))
661 return true; 661 return true;
662 662
663 do { 663 do {
664 spin_unlock(&dentry->d_lock); 664 spin_unlock(&dentry->d_lock);
665 spin_lock(&inode->i_lock); 665 spin_lock(&inode->i_lock);
666 spin_lock(&dentry->d_lock); 666 spin_lock(&dentry->d_lock);
667 if (likely(inode == dentry->d_ 667 if (likely(inode == dentry->d_inode))
668 break; 668 break;
669 spin_unlock(&inode->i_lock); 669 spin_unlock(&inode->i_lock);
670 inode = dentry->d_inode; 670 inode = dentry->d_inode;
671 } while (inode); 671 } while (inode);
672 if (likely(!dentry->d_lockref.count)) 672 if (likely(!dentry->d_lockref.count))
673 return true; 673 return true;
674 if (inode) 674 if (inode)
675 spin_unlock(&inode->i_lock); 675 spin_unlock(&inode->i_lock);
676 return false; 676 return false;
677 } 677 }
678 678
679 /* 679 /*
680 * Decide if dentry is worth retaining. Usual 680 * Decide if dentry is worth retaining. Usually this is called with dentry
681 * locked; if not locked, we are more limited 681 * locked; if not locked, we are more limited and might not be able to tell
682 * without a lock. False in this case means " 682 * without a lock. False in this case means "punt to locked path and recheck".
683 * 683 *
684 * In case we aren't locked, these predicates 684 * In case we aren't locked, these predicates are not "stable". However, it is
685 * sufficient that at some point after we drop 685 * sufficient that at some point after we dropped the reference the dentry was
686 * hashed and the flags had the proper value. 686 * hashed and the flags had the proper value. Other dentry users may have
687 * re-gotten a reference to the dentry and cha 687 * re-gotten a reference to the dentry and change that, but our work is done -
688 * we can leave the dentry around with a zero 688 * we can leave the dentry around with a zero refcount.
689 */ 689 */
690 static inline bool retain_dentry(struct dentry 690 static inline bool retain_dentry(struct dentry *dentry, bool locked)
691 { 691 {
692 unsigned int d_flags; 692 unsigned int d_flags;
693 693
694 smp_rmb(); 694 smp_rmb();
695 d_flags = READ_ONCE(dentry->d_flags); 695 d_flags = READ_ONCE(dentry->d_flags);
696 696
697 // Unreachable? Nobody would be able t 697 // Unreachable? Nobody would be able to look it up, no point retaining
698 if (unlikely(d_unhashed(dentry))) 698 if (unlikely(d_unhashed(dentry)))
699 return false; 699 return false;
700 700
701 // Same if it's disconnected 701 // Same if it's disconnected
702 if (unlikely(d_flags & DCACHE_DISCONNE 702 if (unlikely(d_flags & DCACHE_DISCONNECTED))
703 return false; 703 return false;
704 704
705 // ->d_delete() might tell us not to b 705 // ->d_delete() might tell us not to bother, but that requires
706 // ->d_lock; can't decide without it 706 // ->d_lock; can't decide without it
707 if (unlikely(d_flags & DCACHE_OP_DELET 707 if (unlikely(d_flags & DCACHE_OP_DELETE)) {
708 if (!locked || dentry->d_op->d 708 if (!locked || dentry->d_op->d_delete(dentry))
709 return false; 709 return false;
710 } 710 }
711 711
712 // Explicitly told not to bother 712 // Explicitly told not to bother
713 if (unlikely(d_flags & DCACHE_DONTCACH 713 if (unlikely(d_flags & DCACHE_DONTCACHE))
714 return false; 714 return false;
715 715
716 // At this point it looks like we ough 716 // At this point it looks like we ought to keep it. We also might
717 // need to do something - put it on LR 717 // need to do something - put it on LRU if it wasn't there already
718 // and mark it referenced if it was on 718 // and mark it referenced if it was on LRU, but not marked yet.
719 // Unfortunately, both actions require 719 // Unfortunately, both actions require ->d_lock, so in lockless
720 // case we'd have to punt rather than 720 // case we'd have to punt rather than doing those.
721 if (unlikely(!(d_flags & DCACHE_LRU_LI 721 if (unlikely(!(d_flags & DCACHE_LRU_LIST))) {
722 if (!locked) 722 if (!locked)
723 return false; 723 return false;
724 d_lru_add(dentry); 724 d_lru_add(dentry);
725 } else if (unlikely(!(d_flags & DCACHE 725 } else if (unlikely(!(d_flags & DCACHE_REFERENCED))) {
726 if (!locked) 726 if (!locked)
727 return false; 727 return false;
728 dentry->d_flags |= DCACHE_REFE 728 dentry->d_flags |= DCACHE_REFERENCED;
729 } 729 }
730 return true; 730 return true;
731 } 731 }
732 732
733 void d_mark_dontcache(struct inode *inode) 733 void d_mark_dontcache(struct inode *inode)
734 { 734 {
735 struct dentry *de; 735 struct dentry *de;
736 736
737 spin_lock(&inode->i_lock); 737 spin_lock(&inode->i_lock);
738 hlist_for_each_entry(de, &inode->i_den 738 hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
739 spin_lock(&de->d_lock); 739 spin_lock(&de->d_lock);
740 de->d_flags |= DCACHE_DONTCACH 740 de->d_flags |= DCACHE_DONTCACHE;
741 spin_unlock(&de->d_lock); 741 spin_unlock(&de->d_lock);
742 } 742 }
743 inode->i_state |= I_DONTCACHE; 743 inode->i_state |= I_DONTCACHE;
744 spin_unlock(&inode->i_lock); 744 spin_unlock(&inode->i_lock);
745 } 745 }
746 EXPORT_SYMBOL(d_mark_dontcache); 746 EXPORT_SYMBOL(d_mark_dontcache);
747 747
748 /* 748 /*
749 * Try to do a lockless dput(), and return whe 749 * Try to do a lockless dput(), and return whether that was successful.
750 * 750 *
751 * If unsuccessful, we return false, having al 751 * If unsuccessful, we return false, having already taken the dentry lock.
752 * In that case refcount is guaranteed to be z 752 * In that case refcount is guaranteed to be zero and we have already
753 * decided that it's not worth keeping around. 753 * decided that it's not worth keeping around.
754 * 754 *
755 * The caller needs to hold the RCU read lock, 755 * The caller needs to hold the RCU read lock, so that the dentry is
756 * guaranteed to stay around even if the refco 756 * guaranteed to stay around even if the refcount goes down to zero!
757 */ 757 */
758 static inline bool fast_dput(struct dentry *de 758 static inline bool fast_dput(struct dentry *dentry)
759 { 759 {
760 int ret; 760 int ret;
761 761
762 /* 762 /*
763 * try to decrement the lockref optimi 763 * try to decrement the lockref optimistically.
764 */ 764 */
765 ret = lockref_put_return(&dentry->d_lo 765 ret = lockref_put_return(&dentry->d_lockref);
766 766
767 /* 767 /*
768 * If the lockref_put_return() failed 768 * If the lockref_put_return() failed due to the lock being held
769 * by somebody else, the fast path has 769 * by somebody else, the fast path has failed. We will need to
770 * get the lock, and then check the co 770 * get the lock, and then check the count again.
771 */ 771 */
772 if (unlikely(ret < 0)) { 772 if (unlikely(ret < 0)) {
773 spin_lock(&dentry->d_lock); 773 spin_lock(&dentry->d_lock);
774 if (WARN_ON_ONCE(dentry->d_loc 774 if (WARN_ON_ONCE(dentry->d_lockref.count <= 0)) {
775 spin_unlock(&dentry->d 775 spin_unlock(&dentry->d_lock);
776 return true; 776 return true;
777 } 777 }
778 dentry->d_lockref.count--; 778 dentry->d_lockref.count--;
779 goto locked; 779 goto locked;
780 } 780 }
781 781
782 /* 782 /*
783 * If we weren't the last ref, we're d 783 * If we weren't the last ref, we're done.
784 */ 784 */
785 if (ret) 785 if (ret)
786 return true; 786 return true;
787 787
788 /* 788 /*
789 * Can we decide that decrement of ref 789 * Can we decide that decrement of refcount is all we needed without
790 * taking the lock? There's a very co 790 * taking the lock? There's a very common case when it's all we need -
791 * dentry looks like it ought to be re 791 * dentry looks like it ought to be retained and there's nothing else
792 * to do. 792 * to do.
793 */ 793 */
794 if (retain_dentry(dentry, false)) 794 if (retain_dentry(dentry, false))
795 return true; 795 return true;
796 796
797 /* 797 /*
798 * Either not worth retaining or we ca 798 * Either not worth retaining or we can't tell without the lock.
799 * Get the lock, then. We've already 799 * Get the lock, then. We've already decremented the refcount to 0,
800 * but we'll need to re-check the situ 800 * but we'll need to re-check the situation after getting the lock.
801 */ 801 */
802 spin_lock(&dentry->d_lock); 802 spin_lock(&dentry->d_lock);
803 803
804 /* 804 /*
805 * Did somebody else grab a reference 805 * Did somebody else grab a reference to it in the meantime, and
806 * we're no longer the last user after 806 * we're no longer the last user after all? Alternatively, somebody
807 * else could have killed it and marke 807 * else could have killed it and marked it dead. Either way, we
808 * don't need to do anything else. 808 * don't need to do anything else.
809 */ 809 */
810 locked: 810 locked:
811 if (dentry->d_lockref.count || retain_ 811 if (dentry->d_lockref.count || retain_dentry(dentry, true)) {
812 spin_unlock(&dentry->d_lock); 812 spin_unlock(&dentry->d_lock);
813 return true; 813 return true;
814 } 814 }
815 return false; 815 return false;
816 } 816 }
817 817
818 818
819 /* 819 /*
820 * This is dput 820 * This is dput
821 * 821 *
822 * This is complicated by the fact that we do 822 * This is complicated by the fact that we do not want to put
823 * dentries that are no longer on any hash cha 823 * dentries that are no longer on any hash chain on the unused
824 * list: we'd much rather just get rid of them 824 * list: we'd much rather just get rid of them immediately.
825 * 825 *
826 * However, that implies that we have to trave 826 * However, that implies that we have to traverse the dentry
827 * tree upwards to the parents which might _al 827 * tree upwards to the parents which might _also_ now be
828 * scheduled for deletion (it may have been on 828 * scheduled for deletion (it may have been only waiting for
829 * its last child to go away). 829 * its last child to go away).
830 * 830 *
831 * This tail recursion is done by hand as we d 831 * This tail recursion is done by hand as we don't want to depend
832 * on the compiler to always get this right (g 832 * on the compiler to always get this right (gcc generally doesn't).
833 * Real recursion would eat up our stack space 833 * Real recursion would eat up our stack space.
834 */ 834 */
835 835
836 /* 836 /*
837 * dput - release a dentry 837 * dput - release a dentry
838 * @dentry: dentry to release 838 * @dentry: dentry to release
839 * 839 *
840 * Release a dentry. This will drop the usage 840 * Release a dentry. This will drop the usage count and if appropriate
841 * call the dentry unlink method as well as re 841 * call the dentry unlink method as well as removing it from the queues and
842 * releasing its resources. If the parent dent 842 * releasing its resources. If the parent dentries were scheduled for release
843 * they too may now get deleted. 843 * they too may now get deleted.
844 */ 844 */
845 void dput(struct dentry *dentry) 845 void dput(struct dentry *dentry)
846 { 846 {
847 if (!dentry) 847 if (!dentry)
848 return; 848 return;
849 might_sleep(); 849 might_sleep();
850 rcu_read_lock(); 850 rcu_read_lock();
851 if (likely(fast_dput(dentry))) { 851 if (likely(fast_dput(dentry))) {
852 rcu_read_unlock(); 852 rcu_read_unlock();
853 return; 853 return;
854 } 854 }
855 while (lock_for_kill(dentry)) { 855 while (lock_for_kill(dentry)) {
856 rcu_read_unlock(); 856 rcu_read_unlock();
857 dentry = __dentry_kill(dentry) 857 dentry = __dentry_kill(dentry);
858 if (!dentry) 858 if (!dentry)
859 return; 859 return;
860 if (retain_dentry(dentry, true 860 if (retain_dentry(dentry, true)) {
861 spin_unlock(&dentry->d 861 spin_unlock(&dentry->d_lock);
862 return; 862 return;
863 } 863 }
864 rcu_read_lock(); 864 rcu_read_lock();
865 } 865 }
866 rcu_read_unlock(); 866 rcu_read_unlock();
867 spin_unlock(&dentry->d_lock); 867 spin_unlock(&dentry->d_lock);
868 } 868 }
869 EXPORT_SYMBOL(dput); 869 EXPORT_SYMBOL(dput);
870 870
871 static void to_shrink_list(struct dentry *dent 871 static void to_shrink_list(struct dentry *dentry, struct list_head *list)
872 __must_hold(&dentry->d_lock) 872 __must_hold(&dentry->d_lock)
873 { 873 {
874 if (!(dentry->d_flags & DCACHE_SHRINK_ 874 if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
875 if (dentry->d_flags & DCACHE_L 875 if (dentry->d_flags & DCACHE_LRU_LIST)
876 d_lru_del(dentry); 876 d_lru_del(dentry);
877 d_shrink_add(dentry, list); 877 d_shrink_add(dentry, list);
878 } 878 }
879 } 879 }
880 880
881 void dput_to_list(struct dentry *dentry, struc 881 void dput_to_list(struct dentry *dentry, struct list_head *list)
882 { 882 {
883 rcu_read_lock(); 883 rcu_read_lock();
884 if (likely(fast_dput(dentry))) { 884 if (likely(fast_dput(dentry))) {
885 rcu_read_unlock(); 885 rcu_read_unlock();
886 return; 886 return;
887 } 887 }
888 rcu_read_unlock(); 888 rcu_read_unlock();
889 to_shrink_list(dentry, list); 889 to_shrink_list(dentry, list);
890 spin_unlock(&dentry->d_lock); 890 spin_unlock(&dentry->d_lock);
891 } 891 }
892 892
893 struct dentry *dget_parent(struct dentry *dent 893 struct dentry *dget_parent(struct dentry *dentry)
894 { 894 {
895 int gotref; 895 int gotref;
896 struct dentry *ret; 896 struct dentry *ret;
897 unsigned seq; 897 unsigned seq;
898 898
899 /* 899 /*
900 * Do optimistic parent lookup without 900 * Do optimistic parent lookup without any
901 * locking. 901 * locking.
902 */ 902 */
903 rcu_read_lock(); 903 rcu_read_lock();
904 seq = raw_seqcount_begin(&dentry->d_se 904 seq = raw_seqcount_begin(&dentry->d_seq);
905 ret = READ_ONCE(dentry->d_parent); 905 ret = READ_ONCE(dentry->d_parent);
906 gotref = lockref_get_not_zero(&ret->d_ 906 gotref = lockref_get_not_zero(&ret->d_lockref);
907 rcu_read_unlock(); 907 rcu_read_unlock();
908 if (likely(gotref)) { 908 if (likely(gotref)) {
909 if (!read_seqcount_retry(&dent 909 if (!read_seqcount_retry(&dentry->d_seq, seq))
910 return ret; 910 return ret;
911 dput(ret); 911 dput(ret);
912 } 912 }
913 913
914 repeat: 914 repeat:
915 /* 915 /*
916 * Don't need rcu_dereference because 916 * Don't need rcu_dereference because we re-check it was correct under
917 * the lock. 917 * the lock.
918 */ 918 */
919 rcu_read_lock(); 919 rcu_read_lock();
920 ret = dentry->d_parent; 920 ret = dentry->d_parent;
921 spin_lock(&ret->d_lock); 921 spin_lock(&ret->d_lock);
922 if (unlikely(ret != dentry->d_parent)) 922 if (unlikely(ret != dentry->d_parent)) {
923 spin_unlock(&ret->d_lock); 923 spin_unlock(&ret->d_lock);
924 rcu_read_unlock(); 924 rcu_read_unlock();
925 goto repeat; 925 goto repeat;
926 } 926 }
927 rcu_read_unlock(); 927 rcu_read_unlock();
928 BUG_ON(!ret->d_lockref.count); 928 BUG_ON(!ret->d_lockref.count);
929 ret->d_lockref.count++; 929 ret->d_lockref.count++;
930 spin_unlock(&ret->d_lock); 930 spin_unlock(&ret->d_lock);
931 return ret; 931 return ret;
932 } 932 }
933 EXPORT_SYMBOL(dget_parent); 933 EXPORT_SYMBOL(dget_parent);
934 934
935 static struct dentry * __d_find_any_alias(stru 935 static struct dentry * __d_find_any_alias(struct inode *inode)
936 { 936 {
937 struct dentry *alias; 937 struct dentry *alias;
938 938
939 if (hlist_empty(&inode->i_dentry)) 939 if (hlist_empty(&inode->i_dentry))
940 return NULL; 940 return NULL;
941 alias = hlist_entry(inode->i_dentry.fi 941 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
942 lockref_get(&alias->d_lockref); 942 lockref_get(&alias->d_lockref);
943 return alias; 943 return alias;
944 } 944 }
945 945
946 /** 946 /**
947 * d_find_any_alias - find any alias for a giv 947 * d_find_any_alias - find any alias for a given inode
948 * @inode: inode to find an alias for 948 * @inode: inode to find an alias for
949 * 949 *
950 * If any aliases exist for the given inode, t 950 * If any aliases exist for the given inode, take and return a
951 * reference for one of them. If no aliases e 951 * reference for one of them. If no aliases exist, return %NULL.
952 */ 952 */
953 struct dentry *d_find_any_alias(struct inode * 953 struct dentry *d_find_any_alias(struct inode *inode)
954 { 954 {
955 struct dentry *de; 955 struct dentry *de;
956 956
957 spin_lock(&inode->i_lock); 957 spin_lock(&inode->i_lock);
958 de = __d_find_any_alias(inode); 958 de = __d_find_any_alias(inode);
959 spin_unlock(&inode->i_lock); 959 spin_unlock(&inode->i_lock);
960 return de; 960 return de;
961 } 961 }
962 EXPORT_SYMBOL(d_find_any_alias); 962 EXPORT_SYMBOL(d_find_any_alias);
963 963
964 static struct dentry *__d_find_alias(struct in 964 static struct dentry *__d_find_alias(struct inode *inode)
965 { 965 {
966 struct dentry *alias; 966 struct dentry *alias;
967 967
968 if (S_ISDIR(inode->i_mode)) 968 if (S_ISDIR(inode->i_mode))
969 return __d_find_any_alias(inod 969 return __d_find_any_alias(inode);
970 970
971 hlist_for_each_entry(alias, &inode->i_ 971 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
972 spin_lock(&alias->d_lock); 972 spin_lock(&alias->d_lock);
973 if (!d_unhashed(alias)) { 973 if (!d_unhashed(alias)) {
974 dget_dlock(alias); 974 dget_dlock(alias);
975 spin_unlock(&alias->d_ 975 spin_unlock(&alias->d_lock);
976 return alias; 976 return alias;
977 } 977 }
978 spin_unlock(&alias->d_lock); 978 spin_unlock(&alias->d_lock);
979 } 979 }
980 return NULL; 980 return NULL;
981 } 981 }
982 982
983 /** 983 /**
984 * d_find_alias - grab a hashed alias of inode 984 * d_find_alias - grab a hashed alias of inode
985 * @inode: inode in question 985 * @inode: inode in question
986 * 986 *
987 * If inode has a hashed alias, or is a direct 987 * If inode has a hashed alias, or is a directory and has any alias,
988 * acquire the reference to alias and return i 988 * acquire the reference to alias and return it. Otherwise return NULL.
989 * Notice that if inode is a directory there c 989 * Notice that if inode is a directory there can be only one alias and
990 * it can be unhashed only if it has no childr 990 * it can be unhashed only if it has no children, or if it is the root
991 * of a filesystem, or if the directory was re 991 * of a filesystem, or if the directory was renamed and d_revalidate
992 * was the first vfs operation to notice. 992 * was the first vfs operation to notice.
993 * 993 *
994 * If the inode has an IS_ROOT, DCACHE_DISCONN 994 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
995 * any other hashed alias over that one. 995 * any other hashed alias over that one.
996 */ 996 */
997 struct dentry *d_find_alias(struct inode *inod 997 struct dentry *d_find_alias(struct inode *inode)
998 { 998 {
999 struct dentry *de = NULL; 999 struct dentry *de = NULL;
1000 1000
1001 if (!hlist_empty(&inode->i_dentry)) { 1001 if (!hlist_empty(&inode->i_dentry)) {
1002 spin_lock(&inode->i_lock); 1002 spin_lock(&inode->i_lock);
1003 de = __d_find_alias(inode); 1003 de = __d_find_alias(inode);
1004 spin_unlock(&inode->i_lock); 1004 spin_unlock(&inode->i_lock);
1005 } 1005 }
1006 return de; 1006 return de;
1007 } 1007 }
1008 EXPORT_SYMBOL(d_find_alias); 1008 EXPORT_SYMBOL(d_find_alias);
1009 1009
1010 /* 1010 /*
1011 * Caller MUST be holding rcu_read_lock() an 1011 * Caller MUST be holding rcu_read_lock() and be guaranteed
1012 * that inode won't get freed until rcu_read 1012 * that inode won't get freed until rcu_read_unlock().
1013 */ 1013 */
1014 struct dentry *d_find_alias_rcu(struct inode 1014 struct dentry *d_find_alias_rcu(struct inode *inode)
1015 { 1015 {
1016 struct hlist_head *l = &inode->i_dent 1016 struct hlist_head *l = &inode->i_dentry;
1017 struct dentry *de = NULL; 1017 struct dentry *de = NULL;
1018 1018
1019 spin_lock(&inode->i_lock); 1019 spin_lock(&inode->i_lock);
1020 // ->i_dentry and ->i_rcu are colocat 1020 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1021 // used without having I_FREEING set, 1021 // used without having I_FREEING set, which means no aliases left
1022 if (likely(!(inode->i_state & I_FREEI 1022 if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1023 if (S_ISDIR(inode->i_mode)) { 1023 if (S_ISDIR(inode->i_mode)) {
1024 de = hlist_entry(l->f 1024 de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1025 } else { 1025 } else {
1026 hlist_for_each_entry( 1026 hlist_for_each_entry(de, l, d_u.d_alias)
1027 if (!d_unhash 1027 if (!d_unhashed(de))
1028 break 1028 break;
1029 } 1029 }
1030 } 1030 }
1031 spin_unlock(&inode->i_lock); 1031 spin_unlock(&inode->i_lock);
1032 return de; 1032 return de;
1033 } 1033 }
1034 1034
1035 /* 1035 /*
1036 * Try to kill dentries associated with 1036 * Try to kill dentries associated with this inode.
1037 * WARNING: you must own a reference to inode 1037 * WARNING: you must own a reference to inode.
1038 */ 1038 */
1039 void d_prune_aliases(struct inode *inode) 1039 void d_prune_aliases(struct inode *inode)
1040 { 1040 {
1041 LIST_HEAD(dispose); 1041 LIST_HEAD(dispose);
1042 struct dentry *dentry; 1042 struct dentry *dentry;
1043 1043
1044 spin_lock(&inode->i_lock); 1044 spin_lock(&inode->i_lock);
1045 hlist_for_each_entry(dentry, &inode-> 1045 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1046 spin_lock(&dentry->d_lock); 1046 spin_lock(&dentry->d_lock);
1047 if (!dentry->d_lockref.count) 1047 if (!dentry->d_lockref.count)
1048 to_shrink_list(dentry 1048 to_shrink_list(dentry, &dispose);
1049 spin_unlock(&dentry->d_lock); 1049 spin_unlock(&dentry->d_lock);
1050 } 1050 }
1051 spin_unlock(&inode->i_lock); 1051 spin_unlock(&inode->i_lock);
1052 shrink_dentry_list(&dispose); 1052 shrink_dentry_list(&dispose);
1053 } 1053 }
1054 EXPORT_SYMBOL(d_prune_aliases); 1054 EXPORT_SYMBOL(d_prune_aliases);
1055 1055
1056 static inline void shrink_kill(struct dentry 1056 static inline void shrink_kill(struct dentry *victim)
1057 { 1057 {
1058 do { 1058 do {
1059 rcu_read_unlock(); 1059 rcu_read_unlock();
1060 victim = __dentry_kill(victim 1060 victim = __dentry_kill(victim);
1061 rcu_read_lock(); 1061 rcu_read_lock();
1062 } while (victim && lock_for_kill(vict 1062 } while (victim && lock_for_kill(victim));
1063 rcu_read_unlock(); 1063 rcu_read_unlock();
1064 if (victim) 1064 if (victim)
1065 spin_unlock(&victim->d_lock); 1065 spin_unlock(&victim->d_lock);
1066 } 1066 }
1067 1067
1068 void shrink_dentry_list(struct list_head *lis 1068 void shrink_dentry_list(struct list_head *list)
1069 { 1069 {
1070 while (!list_empty(list)) { 1070 while (!list_empty(list)) {
1071 struct dentry *dentry; 1071 struct dentry *dentry;
1072 1072
1073 dentry = list_entry(list->pre 1073 dentry = list_entry(list->prev, struct dentry, d_lru);
1074 spin_lock(&dentry->d_lock); 1074 spin_lock(&dentry->d_lock);
1075 rcu_read_lock(); 1075 rcu_read_lock();
1076 if (!lock_for_kill(dentry)) { 1076 if (!lock_for_kill(dentry)) {
1077 bool can_free; 1077 bool can_free;
1078 rcu_read_unlock(); 1078 rcu_read_unlock();
1079 d_shrink_del(dentry); 1079 d_shrink_del(dentry);
1080 can_free = dentry->d_ 1080 can_free = dentry->d_flags & DCACHE_DENTRY_KILLED;
1081 spin_unlock(&dentry-> 1081 spin_unlock(&dentry->d_lock);
1082 if (can_free) 1082 if (can_free)
1083 dentry_free(d 1083 dentry_free(dentry);
1084 continue; 1084 continue;
1085 } 1085 }
1086 d_shrink_del(dentry); 1086 d_shrink_del(dentry);
1087 shrink_kill(dentry); 1087 shrink_kill(dentry);
1088 } 1088 }
1089 } 1089 }
1090 1090
1091 static enum lru_status dentry_lru_isolate(str 1091 static enum lru_status dentry_lru_isolate(struct list_head *item,
1092 struct list_lru_one *lru, spi 1092 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1093 { 1093 {
1094 struct list_head *freeable = arg; 1094 struct list_head *freeable = arg;
1095 struct dentry *dentry = container_o 1095 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1096 1096
1097 1097
1098 /* 1098 /*
1099 * we are inverting the lru lock/dent 1099 * we are inverting the lru lock/dentry->d_lock here,
1100 * so use a trylock. If we fail to ge 1100 * so use a trylock. If we fail to get the lock, just skip
1101 * it 1101 * it
1102 */ 1102 */
1103 if (!spin_trylock(&dentry->d_lock)) 1103 if (!spin_trylock(&dentry->d_lock))
1104 return LRU_SKIP; 1104 return LRU_SKIP;
1105 1105
1106 /* 1106 /*
1107 * Referenced dentries are still in u 1107 * Referenced dentries are still in use. If they have active
1108 * counts, just remove them from the 1108 * counts, just remove them from the LRU. Otherwise give them
1109 * another pass through the LRU. 1109 * another pass through the LRU.
1110 */ 1110 */
1111 if (dentry->d_lockref.count) { 1111 if (dentry->d_lockref.count) {
1112 d_lru_isolate(lru, dentry); 1112 d_lru_isolate(lru, dentry);
1113 spin_unlock(&dentry->d_lock); 1113 spin_unlock(&dentry->d_lock);
1114 return LRU_REMOVED; 1114 return LRU_REMOVED;
1115 } 1115 }
1116 1116
1117 if (dentry->d_flags & DCACHE_REFERENC 1117 if (dentry->d_flags & DCACHE_REFERENCED) {
1118 dentry->d_flags &= ~DCACHE_RE 1118 dentry->d_flags &= ~DCACHE_REFERENCED;
1119 spin_unlock(&dentry->d_lock); 1119 spin_unlock(&dentry->d_lock);
1120 1120
1121 /* 1121 /*
1122 * The list move itself will 1122 * The list move itself will be made by the common LRU code. At
1123 * this point, we've dropped 1123 * this point, we've dropped the dentry->d_lock but keep the
1124 * lru lock. This is safe to 1124 * lru lock. This is safe to do, since every list movement is
1125 * protected by the lru lock 1125 * protected by the lru lock even if both locks are held.
1126 * 1126 *
1127 * This is guaranteed by the 1127 * This is guaranteed by the fact that all LRU management
1128 * functions are intermediate 1128 * functions are intermediated by the LRU API calls like
1129 * list_lru_add_obj and list_ 1129 * list_lru_add_obj and list_lru_del_obj. List movement in this file
1130 * only ever occur through th 1130 * only ever occur through this functions or through callbacks
1131 * like this one, that are ca 1131 * like this one, that are called from the LRU API.
1132 * 1132 *
1133 * The only exceptions to thi 1133 * The only exceptions to this are functions like
1134 * shrink_dentry_list, and co 1134 * shrink_dentry_list, and code that first checks for the
1135 * DCACHE_SHRINK_LIST flag. 1135 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1136 * operating only with stack 1136 * operating only with stack provided lists after they are
1137 * properly isolated from the 1137 * properly isolated from the main list. It is thus, always a
1138 * local access. 1138 * local access.
1139 */ 1139 */
1140 return LRU_ROTATE; 1140 return LRU_ROTATE;
1141 } 1141 }
1142 1142
1143 d_lru_shrink_move(lru, dentry, freeab 1143 d_lru_shrink_move(lru, dentry, freeable);
1144 spin_unlock(&dentry->d_lock); 1144 spin_unlock(&dentry->d_lock);
1145 1145
1146 return LRU_REMOVED; 1146 return LRU_REMOVED;
1147 } 1147 }
1148 1148
1149 /** 1149 /**
1150 * prune_dcache_sb - shrink the dcache 1150 * prune_dcache_sb - shrink the dcache
1151 * @sb: superblock 1151 * @sb: superblock
1152 * @sc: shrink control, passed to list_lru_sh 1152 * @sc: shrink control, passed to list_lru_shrink_walk()
1153 * 1153 *
1154 * Attempt to shrink the superblock dcache LR 1154 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1155 * is done when we need more memory and calle 1155 * is done when we need more memory and called from the superblock shrinker
1156 * function. 1156 * function.
1157 * 1157 *
1158 * This function may fail to free any resourc 1158 * This function may fail to free any resources if all the dentries are in
1159 * use. 1159 * use.
1160 */ 1160 */
1161 long prune_dcache_sb(struct super_block *sb, 1161 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1162 { 1162 {
1163 LIST_HEAD(dispose); 1163 LIST_HEAD(dispose);
1164 long freed; 1164 long freed;
1165 1165
1166 freed = list_lru_shrink_walk(&sb->s_d 1166 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1167 dentry_l 1167 dentry_lru_isolate, &dispose);
1168 shrink_dentry_list(&dispose); 1168 shrink_dentry_list(&dispose);
1169 return freed; 1169 return freed;
1170 } 1170 }
1171 1171
1172 static enum lru_status dentry_lru_isolate_shr 1172 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1173 struct list_lru_one *lru, spi 1173 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1174 { 1174 {
1175 struct list_head *freeable = arg; 1175 struct list_head *freeable = arg;
1176 struct dentry *dentry = container_o 1176 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1177 1177
1178 /* 1178 /*
1179 * we are inverting the lru lock/dent 1179 * we are inverting the lru lock/dentry->d_lock here,
1180 * so use a trylock. If we fail to ge 1180 * so use a trylock. If we fail to get the lock, just skip
1181 * it 1181 * it
1182 */ 1182 */
1183 if (!spin_trylock(&dentry->d_lock)) 1183 if (!spin_trylock(&dentry->d_lock))
1184 return LRU_SKIP; 1184 return LRU_SKIP;
1185 1185
1186 d_lru_shrink_move(lru, dentry, freeab 1186 d_lru_shrink_move(lru, dentry, freeable);
1187 spin_unlock(&dentry->d_lock); 1187 spin_unlock(&dentry->d_lock);
1188 1188
1189 return LRU_REMOVED; 1189 return LRU_REMOVED;
1190 } 1190 }
1191 1191
1192 1192
1193 /** 1193 /**
1194 * shrink_dcache_sb - shrink dcache for a sup 1194 * shrink_dcache_sb - shrink dcache for a superblock
1195 * @sb: superblock 1195 * @sb: superblock
1196 * 1196 *
1197 * Shrink the dcache for the specified super 1197 * Shrink the dcache for the specified super block. This is used to free
1198 * the dcache before unmounting a file system 1198 * the dcache before unmounting a file system.
1199 */ 1199 */
1200 void shrink_dcache_sb(struct super_block *sb) 1200 void shrink_dcache_sb(struct super_block *sb)
1201 { 1201 {
1202 do { 1202 do {
1203 LIST_HEAD(dispose); 1203 LIST_HEAD(dispose);
1204 1204
1205 list_lru_walk(&sb->s_dentry_l 1205 list_lru_walk(&sb->s_dentry_lru,
1206 dentry_lru_isolate_sh 1206 dentry_lru_isolate_shrink, &dispose, 1024);
1207 shrink_dentry_list(&dispose); 1207 shrink_dentry_list(&dispose);
1208 } while (list_lru_count(&sb->s_dentry 1208 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1209 } 1209 }
1210 EXPORT_SYMBOL(shrink_dcache_sb); 1210 EXPORT_SYMBOL(shrink_dcache_sb);
1211 1211
1212 /** 1212 /**
1213 * enum d_walk_ret - action to talke during t 1213 * enum d_walk_ret - action to talke during tree walk
1214 * @D_WALK_CONTINUE: contrinue walk 1214 * @D_WALK_CONTINUE: contrinue walk
1215 * @D_WALK_QUIT: quit walk 1215 * @D_WALK_QUIT: quit walk
1216 * @D_WALK_NORETRY: quit when retry is ne 1216 * @D_WALK_NORETRY: quit when retry is needed
1217 * @D_WALK_SKIP: skip this dentry and 1217 * @D_WALK_SKIP: skip this dentry and its children
1218 */ 1218 */
1219 enum d_walk_ret { 1219 enum d_walk_ret {
1220 D_WALK_CONTINUE, 1220 D_WALK_CONTINUE,
1221 D_WALK_QUIT, 1221 D_WALK_QUIT,
1222 D_WALK_NORETRY, 1222 D_WALK_NORETRY,
1223 D_WALK_SKIP, 1223 D_WALK_SKIP,
1224 }; 1224 };
1225 1225
1226 /** 1226 /**
1227 * d_walk - walk the dentry tree 1227 * d_walk - walk the dentry tree
1228 * @parent: start of walk 1228 * @parent: start of walk
1229 * @data: data passed to @enter() and @ 1229 * @data: data passed to @enter() and @finish()
1230 * @enter: callback when first entering 1230 * @enter: callback when first entering the dentry
1231 * 1231 *
1232 * The @enter() callbacks are called with d_l 1232 * The @enter() callbacks are called with d_lock held.
1233 */ 1233 */
1234 static void d_walk(struct dentry *parent, voi 1234 static void d_walk(struct dentry *parent, void *data,
1235 enum d_walk_ret (*enter)(v 1235 enum d_walk_ret (*enter)(void *, struct dentry *))
1236 { 1236 {
1237 struct dentry *this_parent, *dentry; 1237 struct dentry *this_parent, *dentry;
1238 unsigned seq = 0; 1238 unsigned seq = 0;
1239 enum d_walk_ret ret; 1239 enum d_walk_ret ret;
1240 bool retry = true; 1240 bool retry = true;
1241 1241
1242 again: 1242 again:
1243 read_seqbegin_or_lock(&rename_lock, & 1243 read_seqbegin_or_lock(&rename_lock, &seq);
1244 this_parent = parent; 1244 this_parent = parent;
1245 spin_lock(&this_parent->d_lock); 1245 spin_lock(&this_parent->d_lock);
1246 1246
1247 ret = enter(data, this_parent); 1247 ret = enter(data, this_parent);
1248 switch (ret) { 1248 switch (ret) {
1249 case D_WALK_CONTINUE: 1249 case D_WALK_CONTINUE:
1250 break; 1250 break;
1251 case D_WALK_QUIT: 1251 case D_WALK_QUIT:
1252 case D_WALK_SKIP: 1252 case D_WALK_SKIP:
1253 goto out_unlock; 1253 goto out_unlock;
1254 case D_WALK_NORETRY: 1254 case D_WALK_NORETRY:
1255 retry = false; 1255 retry = false;
1256 break; 1256 break;
1257 } 1257 }
1258 repeat: 1258 repeat:
1259 dentry = d_first_child(this_parent); 1259 dentry = d_first_child(this_parent);
1260 resume: 1260 resume:
1261 hlist_for_each_entry_from(dentry, d_s 1261 hlist_for_each_entry_from(dentry, d_sib) {
1262 if (unlikely(dentry->d_flags 1262 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1263 continue; 1263 continue;
1264 1264
1265 spin_lock_nested(&dentry->d_l 1265 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1266 1266
1267 ret = enter(data, dentry); 1267 ret = enter(data, dentry);
1268 switch (ret) { 1268 switch (ret) {
1269 case D_WALK_CONTINUE: 1269 case D_WALK_CONTINUE:
1270 break; 1270 break;
1271 case D_WALK_QUIT: 1271 case D_WALK_QUIT:
1272 spin_unlock(&dentry-> 1272 spin_unlock(&dentry->d_lock);
1273 goto out_unlock; 1273 goto out_unlock;
1274 case D_WALK_NORETRY: 1274 case D_WALK_NORETRY:
1275 retry = false; 1275 retry = false;
1276 break; 1276 break;
1277 case D_WALK_SKIP: 1277 case D_WALK_SKIP:
1278 spin_unlock(&dentry-> 1278 spin_unlock(&dentry->d_lock);
1279 continue; 1279 continue;
1280 } 1280 }
1281 1281
1282 if (!hlist_empty(&dentry->d_c 1282 if (!hlist_empty(&dentry->d_children)) {
1283 spin_unlock(&this_par 1283 spin_unlock(&this_parent->d_lock);
1284 spin_release(&dentry- 1284 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1285 this_parent = dentry; 1285 this_parent = dentry;
1286 spin_acquire(&this_pa 1286 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1287 goto repeat; 1287 goto repeat;
1288 } 1288 }
1289 spin_unlock(&dentry->d_lock); 1289 spin_unlock(&dentry->d_lock);
1290 } 1290 }
1291 /* 1291 /*
1292 * All done at this level ... ascend 1292 * All done at this level ... ascend and resume the search.
1293 */ 1293 */
1294 rcu_read_lock(); 1294 rcu_read_lock();
1295 ascend: 1295 ascend:
1296 if (this_parent != parent) { 1296 if (this_parent != parent) {
1297 dentry = this_parent; 1297 dentry = this_parent;
1298 this_parent = dentry->d_paren 1298 this_parent = dentry->d_parent;
1299 1299
1300 spin_unlock(&dentry->d_lock); 1300 spin_unlock(&dentry->d_lock);
1301 spin_lock(&this_parent->d_loc 1301 spin_lock(&this_parent->d_lock);
1302 1302
1303 /* might go back up the wrong 1303 /* might go back up the wrong parent if we have had a rename. */
1304 if (need_seqretry(&rename_loc 1304 if (need_seqretry(&rename_lock, seq))
1305 goto rename_retry; 1305 goto rename_retry;
1306 /* go into the first sibling 1306 /* go into the first sibling still alive */
1307 hlist_for_each_entry_continue 1307 hlist_for_each_entry_continue(dentry, d_sib) {
1308 if (likely(!(dentry-> 1308 if (likely(!(dentry->d_flags & DCACHE_DENTRY_KILLED))) {
1309 rcu_read_unlo 1309 rcu_read_unlock();
1310 goto resume; 1310 goto resume;
1311 } 1311 }
1312 } 1312 }
1313 goto ascend; 1313 goto ascend;
1314 } 1314 }
1315 if (need_seqretry(&rename_lock, seq)) 1315 if (need_seqretry(&rename_lock, seq))
1316 goto rename_retry; 1316 goto rename_retry;
1317 rcu_read_unlock(); 1317 rcu_read_unlock();
1318 1318
1319 out_unlock: 1319 out_unlock:
1320 spin_unlock(&this_parent->d_lock); 1320 spin_unlock(&this_parent->d_lock);
1321 done_seqretry(&rename_lock, seq); 1321 done_seqretry(&rename_lock, seq);
1322 return; 1322 return;
1323 1323
1324 rename_retry: 1324 rename_retry:
1325 spin_unlock(&this_parent->d_lock); 1325 spin_unlock(&this_parent->d_lock);
1326 rcu_read_unlock(); 1326 rcu_read_unlock();
1327 BUG_ON(seq & 1); 1327 BUG_ON(seq & 1);
1328 if (!retry) 1328 if (!retry)
1329 return; 1329 return;
1330 seq = 1; 1330 seq = 1;
1331 goto again; 1331 goto again;
1332 } 1332 }
1333 1333
1334 struct check_mount { 1334 struct check_mount {
1335 struct vfsmount *mnt; 1335 struct vfsmount *mnt;
1336 unsigned int mounted; 1336 unsigned int mounted;
1337 }; 1337 };
1338 1338
1339 static enum d_walk_ret path_check_mount(void 1339 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1340 { 1340 {
1341 struct check_mount *info = data; 1341 struct check_mount *info = data;
1342 struct path path = { .mnt = info->mnt 1342 struct path path = { .mnt = info->mnt, .dentry = dentry };
1343 1343
1344 if (likely(!d_mountpoint(dentry))) 1344 if (likely(!d_mountpoint(dentry)))
1345 return D_WALK_CONTINUE; 1345 return D_WALK_CONTINUE;
1346 if (__path_is_mountpoint(&path)) { 1346 if (__path_is_mountpoint(&path)) {
1347 info->mounted = 1; 1347 info->mounted = 1;
1348 return D_WALK_QUIT; 1348 return D_WALK_QUIT;
1349 } 1349 }
1350 return D_WALK_CONTINUE; 1350 return D_WALK_CONTINUE;
1351 } 1351 }
1352 1352
1353 /** 1353 /**
1354 * path_has_submounts - check for mounts over 1354 * path_has_submounts - check for mounts over a dentry in the
1355 * current namespace. 1355 * current namespace.
1356 * @parent: path to check. 1356 * @parent: path to check.
1357 * 1357 *
1358 * Return true if the parent or its subdirect 1358 * Return true if the parent or its subdirectories contain
1359 * a mount point in the current namespace. 1359 * a mount point in the current namespace.
1360 */ 1360 */
1361 int path_has_submounts(const struct path *par 1361 int path_has_submounts(const struct path *parent)
1362 { 1362 {
1363 struct check_mount data = { .mnt = pa 1363 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1364 1364
1365 read_seqlock_excl(&mount_lock); 1365 read_seqlock_excl(&mount_lock);
1366 d_walk(parent->dentry, &data, path_ch 1366 d_walk(parent->dentry, &data, path_check_mount);
1367 read_sequnlock_excl(&mount_lock); 1367 read_sequnlock_excl(&mount_lock);
1368 1368
1369 return data.mounted; 1369 return data.mounted;
1370 } 1370 }
1371 EXPORT_SYMBOL(path_has_submounts); 1371 EXPORT_SYMBOL(path_has_submounts);
1372 1372
1373 /* 1373 /*
1374 * Called by mount code to set a mountpoint a 1374 * Called by mount code to set a mountpoint and check if the mountpoint is
1375 * reachable (e.g. NFS can unhash a directory 1375 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1376 * subtree can become unreachable). 1376 * subtree can become unreachable).
1377 * 1377 *
1378 * Only one of d_invalidate() and d_set_mount 1378 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1379 * this reason take rename_lock and d_lock on 1379 * this reason take rename_lock and d_lock on dentry and ancestors.
1380 */ 1380 */
1381 int d_set_mounted(struct dentry *dentry) 1381 int d_set_mounted(struct dentry *dentry)
1382 { 1382 {
1383 struct dentry *p; 1383 struct dentry *p;
1384 int ret = -ENOENT; 1384 int ret = -ENOENT;
1385 write_seqlock(&rename_lock); 1385 write_seqlock(&rename_lock);
1386 for (p = dentry->d_parent; !IS_ROOT(p 1386 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1387 /* Need exclusion wrt. d_inva 1387 /* Need exclusion wrt. d_invalidate() */
1388 spin_lock(&p->d_lock); 1388 spin_lock(&p->d_lock);
1389 if (unlikely(d_unhashed(p))) 1389 if (unlikely(d_unhashed(p))) {
1390 spin_unlock(&p->d_loc 1390 spin_unlock(&p->d_lock);
1391 goto out; 1391 goto out;
1392 } 1392 }
1393 spin_unlock(&p->d_lock); 1393 spin_unlock(&p->d_lock);
1394 } 1394 }
1395 spin_lock(&dentry->d_lock); 1395 spin_lock(&dentry->d_lock);
1396 if (!d_unlinked(dentry)) { 1396 if (!d_unlinked(dentry)) {
1397 ret = -EBUSY; 1397 ret = -EBUSY;
1398 if (!d_mountpoint(dentry)) { 1398 if (!d_mountpoint(dentry)) {
1399 dentry->d_flags |= DC 1399 dentry->d_flags |= DCACHE_MOUNTED;
1400 ret = 0; 1400 ret = 0;
1401 } 1401 }
1402 } 1402 }
1403 spin_unlock(&dentry->d_lock); 1403 spin_unlock(&dentry->d_lock);
1404 out: 1404 out:
1405 write_sequnlock(&rename_lock); 1405 write_sequnlock(&rename_lock);
1406 return ret; 1406 return ret;
1407 } 1407 }
1408 1408
1409 /* 1409 /*
1410 * Search the dentry child list of the specif 1410 * Search the dentry child list of the specified parent,
1411 * and move any unused dentries to the end of 1411 * and move any unused dentries to the end of the unused
1412 * list for prune_dcache(). We descend to the 1412 * list for prune_dcache(). We descend to the next level
1413 * whenever the d_children list is non-empty 1413 * whenever the d_children list is non-empty and continue
1414 * searching. 1414 * searching.
1415 * 1415 *
1416 * It returns zero iff there are no unused ch 1416 * It returns zero iff there are no unused children,
1417 * otherwise it returns the number of childr 1417 * otherwise it returns the number of children moved to
1418 * the end of the unused list. This may not b 1418 * the end of the unused list. This may not be the total
1419 * number of unused children, because select_ 1419 * number of unused children, because select_parent can
1420 * drop the lock and return early due to late 1420 * drop the lock and return early due to latency
1421 * constraints. 1421 * constraints.
1422 */ 1422 */
1423 1423
1424 struct select_data { 1424 struct select_data {
1425 struct dentry *start; 1425 struct dentry *start;
1426 union { 1426 union {
1427 long found; 1427 long found;
1428 struct dentry *victim; 1428 struct dentry *victim;
1429 }; 1429 };
1430 struct list_head dispose; 1430 struct list_head dispose;
1431 }; 1431 };
1432 1432
1433 static enum d_walk_ret select_collect(void *_ 1433 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1434 { 1434 {
1435 struct select_data *data = _data; 1435 struct select_data *data = _data;
1436 enum d_walk_ret ret = D_WALK_CONTINUE 1436 enum d_walk_ret ret = D_WALK_CONTINUE;
1437 1437
1438 if (data->start == dentry) 1438 if (data->start == dentry)
1439 goto out; 1439 goto out;
1440 1440
1441 if (dentry->d_flags & DCACHE_SHRINK_L 1441 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1442 data->found++; 1442 data->found++;
1443 } else if (!dentry->d_lockref.count) 1443 } else if (!dentry->d_lockref.count) {
1444 to_shrink_list(dentry, &data- 1444 to_shrink_list(dentry, &data->dispose);
1445 data->found++; 1445 data->found++;
1446 } else if (dentry->d_lockref.count < 1446 } else if (dentry->d_lockref.count < 0) {
1447 data->found++; 1447 data->found++;
1448 } 1448 }
1449 /* 1449 /*
1450 * We can return to the caller if we 1450 * We can return to the caller if we have found some (this
1451 * ensures forward progress). We'll b 1451 * ensures forward progress). We'll be coming back to find
1452 * the rest. 1452 * the rest.
1453 */ 1453 */
1454 if (!list_empty(&data->dispose)) 1454 if (!list_empty(&data->dispose))
1455 ret = need_resched() ? D_WALK 1455 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1456 out: 1456 out:
1457 return ret; 1457 return ret;
1458 } 1458 }
1459 1459
1460 static enum d_walk_ret select_collect2(void * 1460 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1461 { 1461 {
1462 struct select_data *data = _data; 1462 struct select_data *data = _data;
1463 enum d_walk_ret ret = D_WALK_CONTINUE 1463 enum d_walk_ret ret = D_WALK_CONTINUE;
1464 1464
1465 if (data->start == dentry) 1465 if (data->start == dentry)
1466 goto out; 1466 goto out;
1467 1467
1468 if (!dentry->d_lockref.count) { 1468 if (!dentry->d_lockref.count) {
1469 if (dentry->d_flags & DCACHE_ 1469 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1470 rcu_read_lock(); 1470 rcu_read_lock();
1471 data->victim = dentry 1471 data->victim = dentry;
1472 return D_WALK_QUIT; 1472 return D_WALK_QUIT;
1473 } 1473 }
1474 to_shrink_list(dentry, &data- 1474 to_shrink_list(dentry, &data->dispose);
1475 } 1475 }
1476 /* 1476 /*
1477 * We can return to the caller if we 1477 * We can return to the caller if we have found some (this
1478 * ensures forward progress). We'll b 1478 * ensures forward progress). We'll be coming back to find
1479 * the rest. 1479 * the rest.
1480 */ 1480 */
1481 if (!list_empty(&data->dispose)) 1481 if (!list_empty(&data->dispose))
1482 ret = need_resched() ? D_WALK 1482 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1483 out: 1483 out:
1484 return ret; 1484 return ret;
1485 } 1485 }
1486 1486
1487 /** 1487 /**
1488 * shrink_dcache_parent - prune dcache 1488 * shrink_dcache_parent - prune dcache
1489 * @parent: parent of entries to prune 1489 * @parent: parent of entries to prune
1490 * 1490 *
1491 * Prune the dcache to remove unused children 1491 * Prune the dcache to remove unused children of the parent dentry.
1492 */ 1492 */
1493 void shrink_dcache_parent(struct dentry *pare 1493 void shrink_dcache_parent(struct dentry *parent)
1494 { 1494 {
1495 for (;;) { 1495 for (;;) {
1496 struct select_data data = {.s 1496 struct select_data data = {.start = parent};
1497 1497
1498 INIT_LIST_HEAD(&data.dispose) 1498 INIT_LIST_HEAD(&data.dispose);
1499 d_walk(parent, &data, select_ 1499 d_walk(parent, &data, select_collect);
1500 1500
1501 if (!list_empty(&data.dispose 1501 if (!list_empty(&data.dispose)) {
1502 shrink_dentry_list(&d 1502 shrink_dentry_list(&data.dispose);
1503 continue; 1503 continue;
1504 } 1504 }
1505 1505
1506 cond_resched(); 1506 cond_resched();
1507 if (!data.found) 1507 if (!data.found)
1508 break; 1508 break;
1509 data.victim = NULL; 1509 data.victim = NULL;
1510 d_walk(parent, &data, select_ 1510 d_walk(parent, &data, select_collect2);
1511 if (data.victim) { 1511 if (data.victim) {
1512 spin_lock(&data.victi 1512 spin_lock(&data.victim->d_lock);
1513 if (!lock_for_kill(da 1513 if (!lock_for_kill(data.victim)) {
1514 spin_unlock(& 1514 spin_unlock(&data.victim->d_lock);
1515 rcu_read_unlo 1515 rcu_read_unlock();
1516 } else { 1516 } else {
1517 shrink_kill(d 1517 shrink_kill(data.victim);
1518 } 1518 }
1519 } 1519 }
1520 if (!list_empty(&data.dispose 1520 if (!list_empty(&data.dispose))
1521 shrink_dentry_list(&d 1521 shrink_dentry_list(&data.dispose);
1522 } 1522 }
1523 } 1523 }
1524 EXPORT_SYMBOL(shrink_dcache_parent); 1524 EXPORT_SYMBOL(shrink_dcache_parent);
1525 1525
1526 static enum d_walk_ret umount_check(void *_da 1526 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1527 { 1527 {
1528 /* it has busy descendents; complain 1528 /* it has busy descendents; complain about those instead */
1529 if (!hlist_empty(&dentry->d_children) 1529 if (!hlist_empty(&dentry->d_children))
1530 return D_WALK_CONTINUE; 1530 return D_WALK_CONTINUE;
1531 1531
1532 /* root with refcount 1 is fine */ 1532 /* root with refcount 1 is fine */
1533 if (dentry == _data && dentry->d_lock 1533 if (dentry == _data && dentry->d_lockref.count == 1)
1534 return D_WALK_CONTINUE; 1534 return D_WALK_CONTINUE;
1535 1535
1536 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} 1536 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} "
1537 " still in use (%d) [ 1537 " still in use (%d) [unmount of %s %s]\n",
1538 dentry, 1538 dentry,
1539 dentry->d_inode ? 1539 dentry->d_inode ?
1540 dentry->d_inode->i_ino 1540 dentry->d_inode->i_ino : 0UL,
1541 dentry, 1541 dentry,
1542 dentry->d_lockref.coun 1542 dentry->d_lockref.count,
1543 dentry->d_sb->s_type-> 1543 dentry->d_sb->s_type->name,
1544 dentry->d_sb->s_id); 1544 dentry->d_sb->s_id);
1545 return D_WALK_CONTINUE; 1545 return D_WALK_CONTINUE;
1546 } 1546 }
1547 1547
1548 static void do_one_tree(struct dentry *dentry 1548 static void do_one_tree(struct dentry *dentry)
1549 { 1549 {
1550 shrink_dcache_parent(dentry); 1550 shrink_dcache_parent(dentry);
1551 d_walk(dentry, dentry, umount_check); 1551 d_walk(dentry, dentry, umount_check);
1552 d_drop(dentry); 1552 d_drop(dentry);
1553 dput(dentry); 1553 dput(dentry);
1554 } 1554 }
1555 1555
1556 /* 1556 /*
1557 * destroy the dentries attached to a superbl 1557 * destroy the dentries attached to a superblock on unmounting
1558 */ 1558 */
1559 void shrink_dcache_for_umount(struct super_bl 1559 void shrink_dcache_for_umount(struct super_block *sb)
1560 { 1560 {
1561 struct dentry *dentry; 1561 struct dentry *dentry;
1562 1562
1563 rwsem_assert_held_write(&sb->s_umount 1563 rwsem_assert_held_write(&sb->s_umount);
1564 1564
1565 dentry = sb->s_root; 1565 dentry = sb->s_root;
1566 sb->s_root = NULL; 1566 sb->s_root = NULL;
1567 do_one_tree(dentry); 1567 do_one_tree(dentry);
1568 1568
1569 while (!hlist_bl_empty(&sb->s_roots)) 1569 while (!hlist_bl_empty(&sb->s_roots)) {
1570 dentry = dget(hlist_bl_entry( 1570 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1571 do_one_tree(dentry); 1571 do_one_tree(dentry);
1572 } 1572 }
1573 } 1573 }
1574 1574
1575 static enum d_walk_ret find_submount(void *_d 1575 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1576 { 1576 {
1577 struct dentry **victim = _data; 1577 struct dentry **victim = _data;
1578 if (d_mountpoint(dentry)) { 1578 if (d_mountpoint(dentry)) {
1579 *victim = dget_dlock(dentry); 1579 *victim = dget_dlock(dentry);
1580 return D_WALK_QUIT; 1580 return D_WALK_QUIT;
1581 } 1581 }
1582 return D_WALK_CONTINUE; 1582 return D_WALK_CONTINUE;
1583 } 1583 }
1584 1584
1585 /** 1585 /**
1586 * d_invalidate - detach submounts, prune dca 1586 * d_invalidate - detach submounts, prune dcache, and drop
1587 * @dentry: dentry to invalidate (aka detach, 1587 * @dentry: dentry to invalidate (aka detach, prune and drop)
1588 */ 1588 */
1589 void d_invalidate(struct dentry *dentry) 1589 void d_invalidate(struct dentry *dentry)
1590 { 1590 {
1591 bool had_submounts = false; 1591 bool had_submounts = false;
1592 spin_lock(&dentry->d_lock); 1592 spin_lock(&dentry->d_lock);
1593 if (d_unhashed(dentry)) { 1593 if (d_unhashed(dentry)) {
1594 spin_unlock(&dentry->d_lock); 1594 spin_unlock(&dentry->d_lock);
1595 return; 1595 return;
1596 } 1596 }
1597 __d_drop(dentry); 1597 __d_drop(dentry);
1598 spin_unlock(&dentry->d_lock); 1598 spin_unlock(&dentry->d_lock);
1599 1599
1600 /* Negative dentries can be dropped w 1600 /* Negative dentries can be dropped without further checks */
1601 if (!dentry->d_inode) 1601 if (!dentry->d_inode)
1602 return; 1602 return;
1603 1603
1604 shrink_dcache_parent(dentry); 1604 shrink_dcache_parent(dentry);
1605 for (;;) { 1605 for (;;) {
1606 struct dentry *victim = NULL; 1606 struct dentry *victim = NULL;
1607 d_walk(dentry, &victim, find_ 1607 d_walk(dentry, &victim, find_submount);
1608 if (!victim) { 1608 if (!victim) {
1609 if (had_submounts) 1609 if (had_submounts)
1610 shrink_dcache 1610 shrink_dcache_parent(dentry);
1611 return; 1611 return;
1612 } 1612 }
1613 had_submounts = true; 1613 had_submounts = true;
1614 detach_mounts(victim); 1614 detach_mounts(victim);
1615 dput(victim); 1615 dput(victim);
1616 } 1616 }
1617 } 1617 }
1618 EXPORT_SYMBOL(d_invalidate); 1618 EXPORT_SYMBOL(d_invalidate);
1619 1619
1620 /** 1620 /**
1621 * __d_alloc - allocate a dcache ent 1621 * __d_alloc - allocate a dcache entry
1622 * @sb: filesystem it will belong to 1622 * @sb: filesystem it will belong to
1623 * @name: qstr of the name 1623 * @name: qstr of the name
1624 * 1624 *
1625 * Allocates a dentry. It returns %NULL if th 1625 * Allocates a dentry. It returns %NULL if there is insufficient memory
1626 * available. On a success the dentry is retu 1626 * available. On a success the dentry is returned. The name passed in is
1627 * copied and the copy passed in may be reuse 1627 * copied and the copy passed in may be reused after this call.
1628 */ 1628 */
1629 1629
1630 static struct dentry *__d_alloc(struct super_ 1630 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1631 { 1631 {
1632 struct dentry *dentry; 1632 struct dentry *dentry;
1633 char *dname; 1633 char *dname;
1634 int err; 1634 int err;
1635 1635
1636 dentry = kmem_cache_alloc_lru(dentry_ 1636 dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru,
1637 GFP_KER 1637 GFP_KERNEL);
1638 if (!dentry) 1638 if (!dentry)
1639 return NULL; 1639 return NULL;
1640 1640
1641 /* 1641 /*
1642 * We guarantee that the inline name 1642 * We guarantee that the inline name is always NUL-terminated.
1643 * This way the memcpy() done by the 1643 * This way the memcpy() done by the name switching in rename
1644 * will still always have a NUL at th 1644 * will still always have a NUL at the end, even if we might
1645 * be overwriting an internal NUL cha 1645 * be overwriting an internal NUL character
1646 */ 1646 */
1647 dentry->d_iname[DNAME_INLINE_LEN-1] = 1647 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1648 if (unlikely(!name)) { 1648 if (unlikely(!name)) {
1649 name = &slash_name; 1649 name = &slash_name;
1650 dname = dentry->d_iname; 1650 dname = dentry->d_iname;
1651 } else if (name->len > DNAME_INLINE_L 1651 } else if (name->len > DNAME_INLINE_LEN-1) {
1652 size_t size = offsetof(struct 1652 size_t size = offsetof(struct external_name, name[1]);
1653 struct external_name *p = kma 1653 struct external_name *p = kmalloc(size + name->len,
1654 1654 GFP_KERNEL_ACCOUNT |
1655 1655 __GFP_RECLAIMABLE);
1656 if (!p) { 1656 if (!p) {
1657 kmem_cache_free(dentr 1657 kmem_cache_free(dentry_cache, dentry);
1658 return NULL; 1658 return NULL;
1659 } 1659 }
1660 atomic_set(&p->u.count, 1); 1660 atomic_set(&p->u.count, 1);
1661 dname = p->name; 1661 dname = p->name;
1662 } else { 1662 } else {
1663 dname = dentry->d_iname; 1663 dname = dentry->d_iname;
1664 } 1664 }
1665 1665
1666 dentry->d_name.len = name->len; 1666 dentry->d_name.len = name->len;
1667 dentry->d_name.hash = name->hash; 1667 dentry->d_name.hash = name->hash;
1668 memcpy(dname, name->name, name->len); 1668 memcpy(dname, name->name, name->len);
1669 dname[name->len] = 0; 1669 dname[name->len] = 0;
1670 1670
1671 /* Make sure we always see the termin 1671 /* Make sure we always see the terminating NUL character */
1672 smp_store_release(&dentry->d_name.nam 1672 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1673 1673
1674 dentry->d_lockref.count = 1; 1674 dentry->d_lockref.count = 1;
1675 dentry->d_flags = 0; 1675 dentry->d_flags = 0;
1676 spin_lock_init(&dentry->d_lock); 1676 spin_lock_init(&dentry->d_lock);
1677 seqcount_spinlock_init(&dentry->d_seq 1677 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1678 dentry->d_inode = NULL; 1678 dentry->d_inode = NULL;
1679 dentry->d_parent = dentry; 1679 dentry->d_parent = dentry;
1680 dentry->d_sb = sb; 1680 dentry->d_sb = sb;
1681 dentry->d_op = NULL; 1681 dentry->d_op = NULL;
1682 dentry->d_fsdata = NULL; 1682 dentry->d_fsdata = NULL;
1683 INIT_HLIST_BL_NODE(&dentry->d_hash); 1683 INIT_HLIST_BL_NODE(&dentry->d_hash);
1684 INIT_LIST_HEAD(&dentry->d_lru); 1684 INIT_LIST_HEAD(&dentry->d_lru);
1685 INIT_HLIST_HEAD(&dentry->d_children); 1685 INIT_HLIST_HEAD(&dentry->d_children);
1686 INIT_HLIST_NODE(&dentry->d_u.d_alias) 1686 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1687 INIT_HLIST_NODE(&dentry->d_sib); 1687 INIT_HLIST_NODE(&dentry->d_sib);
1688 d_set_d_op(dentry, dentry->d_sb->s_d_ 1688 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1689 1689
1690 if (dentry->d_op && dentry->d_op->d_i 1690 if (dentry->d_op && dentry->d_op->d_init) {
1691 err = dentry->d_op->d_init(de 1691 err = dentry->d_op->d_init(dentry);
1692 if (err) { 1692 if (err) {
1693 if (dname_external(de 1693 if (dname_external(dentry))
1694 kfree(externa 1694 kfree(external_name(dentry));
1695 kmem_cache_free(dentr 1695 kmem_cache_free(dentry_cache, dentry);
1696 return NULL; 1696 return NULL;
1697 } 1697 }
1698 } 1698 }
1699 1699
1700 this_cpu_inc(nr_dentry); 1700 this_cpu_inc(nr_dentry);
1701 1701
1702 return dentry; 1702 return dentry;
1703 } 1703 }
1704 1704
1705 /** 1705 /**
1706 * d_alloc - allocate a dcache ent 1706 * d_alloc - allocate a dcache entry
1707 * @parent: parent of entry to allocate 1707 * @parent: parent of entry to allocate
1708 * @name: qstr of the name 1708 * @name: qstr of the name
1709 * 1709 *
1710 * Allocates a dentry. It returns %NULL if th 1710 * Allocates a dentry. It returns %NULL if there is insufficient memory
1711 * available. On a success the dentry is retu 1711 * available. On a success the dentry is returned. The name passed in is
1712 * copied and the copy passed in may be reuse 1712 * copied and the copy passed in may be reused after this call.
1713 */ 1713 */
1714 struct dentry *d_alloc(struct dentry * parent 1714 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1715 { 1715 {
1716 struct dentry *dentry = __d_alloc(par 1716 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1717 if (!dentry) 1717 if (!dentry)
1718 return NULL; 1718 return NULL;
1719 spin_lock(&parent->d_lock); 1719 spin_lock(&parent->d_lock);
1720 /* 1720 /*
1721 * don't need child lock because it i 1721 * don't need child lock because it is not subject
1722 * to concurrency here 1722 * to concurrency here
1723 */ 1723 */
1724 dentry->d_parent = dget_dlock(parent) 1724 dentry->d_parent = dget_dlock(parent);
1725 hlist_add_head(&dentry->d_sib, &paren 1725 hlist_add_head(&dentry->d_sib, &parent->d_children);
1726 spin_unlock(&parent->d_lock); 1726 spin_unlock(&parent->d_lock);
1727 1727
1728 return dentry; 1728 return dentry;
1729 } 1729 }
1730 EXPORT_SYMBOL(d_alloc); 1730 EXPORT_SYMBOL(d_alloc);
1731 1731
1732 struct dentry *d_alloc_anon(struct super_bloc 1732 struct dentry *d_alloc_anon(struct super_block *sb)
1733 { 1733 {
1734 return __d_alloc(sb, NULL); 1734 return __d_alloc(sb, NULL);
1735 } 1735 }
1736 EXPORT_SYMBOL(d_alloc_anon); 1736 EXPORT_SYMBOL(d_alloc_anon);
1737 1737
1738 struct dentry *d_alloc_cursor(struct dentry * 1738 struct dentry *d_alloc_cursor(struct dentry * parent)
1739 { 1739 {
1740 struct dentry *dentry = d_alloc_anon( 1740 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1741 if (dentry) { 1741 if (dentry) {
1742 dentry->d_flags |= DCACHE_DEN 1742 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1743 dentry->d_parent = dget(paren 1743 dentry->d_parent = dget(parent);
1744 } 1744 }
1745 return dentry; 1745 return dentry;
1746 } 1746 }
1747 1747
1748 /** 1748 /**
1749 * d_alloc_pseudo - allocate a dentry (for lo 1749 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1750 * @sb: the superblock 1750 * @sb: the superblock
1751 * @name: qstr of the name 1751 * @name: qstr of the name
1752 * 1752 *
1753 * For a filesystem that just pins its dentri 1753 * For a filesystem that just pins its dentries in memory and never
1754 * performs lookups at all, return an unhashe 1754 * performs lookups at all, return an unhashed IS_ROOT dentry.
1755 * This is used for pipes, sockets et.al. - t 1755 * This is used for pipes, sockets et.al. - the stuff that should
1756 * never be anyone's children or parents. Un 1756 * never be anyone's children or parents. Unlike all other
1757 * dentries, these will not have RCU delay be 1757 * dentries, these will not have RCU delay between dropping the
1758 * last reference and freeing them. 1758 * last reference and freeing them.
1759 * 1759 *
1760 * The only user is alloc_file_pseudo() and t 1760 * The only user is alloc_file_pseudo() and that's what should
1761 * be considered a public interface. Don't u 1761 * be considered a public interface. Don't use directly.
1762 */ 1762 */
1763 struct dentry *d_alloc_pseudo(struct super_bl 1763 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1764 { 1764 {
1765 static const struct dentry_operations 1765 static const struct dentry_operations anon_ops = {
1766 .d_dname = simple_dname 1766 .d_dname = simple_dname
1767 }; 1767 };
1768 struct dentry *dentry = __d_alloc(sb, 1768 struct dentry *dentry = __d_alloc(sb, name);
1769 if (likely(dentry)) { 1769 if (likely(dentry)) {
1770 dentry->d_flags |= DCACHE_NOR 1770 dentry->d_flags |= DCACHE_NORCU;
1771 if (!sb->s_d_op) 1771 if (!sb->s_d_op)
1772 d_set_d_op(dentry, &a 1772 d_set_d_op(dentry, &anon_ops);
1773 } 1773 }
1774 return dentry; 1774 return dentry;
1775 } 1775 }
1776 1776
1777 struct dentry *d_alloc_name(struct dentry *pa 1777 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1778 { 1778 {
1779 struct qstr q; 1779 struct qstr q;
1780 1780
1781 q.name = name; 1781 q.name = name;
1782 q.hash_len = hashlen_string(parent, n 1782 q.hash_len = hashlen_string(parent, name);
1783 return d_alloc(parent, &q); 1783 return d_alloc(parent, &q);
1784 } 1784 }
1785 EXPORT_SYMBOL(d_alloc_name); 1785 EXPORT_SYMBOL(d_alloc_name);
1786 1786
1787 void d_set_d_op(struct dentry *dentry, const 1787 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1788 { 1788 {
1789 WARN_ON_ONCE(dentry->d_op); 1789 WARN_ON_ONCE(dentry->d_op);
1790 WARN_ON_ONCE(dentry->d_flags & (DCACH 1790 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1791 DCACHE_OP_COM 1791 DCACHE_OP_COMPARE |
1792 DCACHE_OP_REV 1792 DCACHE_OP_REVALIDATE |
1793 DCACHE_OP_WEA 1793 DCACHE_OP_WEAK_REVALIDATE |
1794 DCACHE_OP_DEL 1794 DCACHE_OP_DELETE |
1795 DCACHE_OP_REA 1795 DCACHE_OP_REAL));
1796 dentry->d_op = op; 1796 dentry->d_op = op;
1797 if (!op) 1797 if (!op)
1798 return; 1798 return;
1799 if (op->d_hash) 1799 if (op->d_hash)
1800 dentry->d_flags |= DCACHE_OP_ 1800 dentry->d_flags |= DCACHE_OP_HASH;
1801 if (op->d_compare) 1801 if (op->d_compare)
1802 dentry->d_flags |= DCACHE_OP_ 1802 dentry->d_flags |= DCACHE_OP_COMPARE;
1803 if (op->d_revalidate) 1803 if (op->d_revalidate)
1804 dentry->d_flags |= DCACHE_OP_ 1804 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1805 if (op->d_weak_revalidate) 1805 if (op->d_weak_revalidate)
1806 dentry->d_flags |= DCACHE_OP_ 1806 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1807 if (op->d_delete) 1807 if (op->d_delete)
1808 dentry->d_flags |= DCACHE_OP_ 1808 dentry->d_flags |= DCACHE_OP_DELETE;
1809 if (op->d_prune) 1809 if (op->d_prune)
1810 dentry->d_flags |= DCACHE_OP_ 1810 dentry->d_flags |= DCACHE_OP_PRUNE;
1811 if (op->d_real) 1811 if (op->d_real)
1812 dentry->d_flags |= DCACHE_OP_ 1812 dentry->d_flags |= DCACHE_OP_REAL;
1813 1813
1814 } 1814 }
1815 EXPORT_SYMBOL(d_set_d_op); 1815 EXPORT_SYMBOL(d_set_d_op);
1816 1816
1817 static unsigned d_flags_for_inode(struct inod 1817 static unsigned d_flags_for_inode(struct inode *inode)
1818 { 1818 {
1819 unsigned add_flags = DCACHE_REGULAR_T 1819 unsigned add_flags = DCACHE_REGULAR_TYPE;
1820 1820
1821 if (!inode) 1821 if (!inode)
1822 return DCACHE_MISS_TYPE; 1822 return DCACHE_MISS_TYPE;
1823 1823
1824 if (S_ISDIR(inode->i_mode)) { 1824 if (S_ISDIR(inode->i_mode)) {
1825 add_flags = DCACHE_DIRECTORY_ 1825 add_flags = DCACHE_DIRECTORY_TYPE;
1826 if (unlikely(!(inode->i_opfla 1826 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1827 if (unlikely(!inode-> 1827 if (unlikely(!inode->i_op->lookup))
1828 add_flags = D 1828 add_flags = DCACHE_AUTODIR_TYPE;
1829 else 1829 else
1830 inode->i_opfl 1830 inode->i_opflags |= IOP_LOOKUP;
1831 } 1831 }
1832 goto type_determined; 1832 goto type_determined;
1833 } 1833 }
1834 1834
1835 if (unlikely(!(inode->i_opflags & IOP 1835 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1836 if (unlikely(inode->i_op->get 1836 if (unlikely(inode->i_op->get_link)) {
1837 add_flags = DCACHE_SY 1837 add_flags = DCACHE_SYMLINK_TYPE;
1838 goto type_determined; 1838 goto type_determined;
1839 } 1839 }
1840 inode->i_opflags |= IOP_NOFOL 1840 inode->i_opflags |= IOP_NOFOLLOW;
1841 } 1841 }
1842 1842
1843 if (unlikely(!S_ISREG(inode->i_mode)) 1843 if (unlikely(!S_ISREG(inode->i_mode)))
1844 add_flags = DCACHE_SPECIAL_TY 1844 add_flags = DCACHE_SPECIAL_TYPE;
1845 1845
1846 type_determined: 1846 type_determined:
1847 if (unlikely(IS_AUTOMOUNT(inode))) 1847 if (unlikely(IS_AUTOMOUNT(inode)))
1848 add_flags |= DCACHE_NEED_AUTO 1848 add_flags |= DCACHE_NEED_AUTOMOUNT;
1849 return add_flags; 1849 return add_flags;
1850 } 1850 }
1851 1851
1852 static void __d_instantiate(struct dentry *de 1852 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1853 { 1853 {
1854 unsigned add_flags = d_flags_for_inod 1854 unsigned add_flags = d_flags_for_inode(inode);
1855 WARN_ON(d_in_lookup(dentry)); 1855 WARN_ON(d_in_lookup(dentry));
1856 1856
1857 spin_lock(&dentry->d_lock); 1857 spin_lock(&dentry->d_lock);
1858 /* 1858 /*
1859 * The negative counter only tracks d 1859 * The negative counter only tracks dentries on the LRU. Don't dec if
1860 * d_lru is on another list. 1860 * d_lru is on another list.
1861 */ 1861 */
1862 if ((dentry->d_flags & 1862 if ((dentry->d_flags &
1863 (DCACHE_LRU_LIST|DCACHE_SHRINK_L 1863 (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
1864 this_cpu_dec(nr_dentry_negati 1864 this_cpu_dec(nr_dentry_negative);
1865 hlist_add_head(&dentry->d_u.d_alias, 1865 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1866 raw_write_seqcount_begin(&dentry->d_s 1866 raw_write_seqcount_begin(&dentry->d_seq);
1867 __d_set_inode_and_type(dentry, inode, 1867 __d_set_inode_and_type(dentry, inode, add_flags);
1868 raw_write_seqcount_end(&dentry->d_seq 1868 raw_write_seqcount_end(&dentry->d_seq);
1869 fsnotify_update_flags(dentry); 1869 fsnotify_update_flags(dentry);
1870 spin_unlock(&dentry->d_lock); 1870 spin_unlock(&dentry->d_lock);
1871 } 1871 }
1872 1872
1873 /** 1873 /**
1874 * d_instantiate - fill in inode information 1874 * d_instantiate - fill in inode information for a dentry
1875 * @entry: dentry to complete 1875 * @entry: dentry to complete
1876 * @inode: inode to attach to this dentry 1876 * @inode: inode to attach to this dentry
1877 * 1877 *
1878 * Fill in inode information in the entry. 1878 * Fill in inode information in the entry.
1879 * 1879 *
1880 * This turns negative dentries into producti 1880 * This turns negative dentries into productive full members
1881 * of society. 1881 * of society.
1882 * 1882 *
1883 * NOTE! This assumes that the inode count ha 1883 * NOTE! This assumes that the inode count has been incremented
1884 * (or otherwise set) by the caller to indica 1884 * (or otherwise set) by the caller to indicate that it is now
1885 * in use by the dcache. 1885 * in use by the dcache.
1886 */ 1886 */
1887 1887
1888 void d_instantiate(struct dentry *entry, stru 1888 void d_instantiate(struct dentry *entry, struct inode * inode)
1889 { 1889 {
1890 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 1890 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1891 if (inode) { 1891 if (inode) {
1892 security_d_instantiate(entry, 1892 security_d_instantiate(entry, inode);
1893 spin_lock(&inode->i_lock); 1893 spin_lock(&inode->i_lock);
1894 __d_instantiate(entry, inode) 1894 __d_instantiate(entry, inode);
1895 spin_unlock(&inode->i_lock); 1895 spin_unlock(&inode->i_lock);
1896 } 1896 }
1897 } 1897 }
1898 EXPORT_SYMBOL(d_instantiate); 1898 EXPORT_SYMBOL(d_instantiate);
1899 1899
1900 /* 1900 /*
1901 * This should be equivalent to d_instantiate 1901 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1902 * with lockdep-related part of unlock_new_in 1902 * with lockdep-related part of unlock_new_inode() done before
1903 * anything else. Use that instead of open-c 1903 * anything else. Use that instead of open-coding d_instantiate()/
1904 * unlock_new_inode() combinations. 1904 * unlock_new_inode() combinations.
1905 */ 1905 */
1906 void d_instantiate_new(struct dentry *entry, 1906 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1907 { 1907 {
1908 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 1908 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1909 BUG_ON(!inode); 1909 BUG_ON(!inode);
1910 lockdep_annotate_inode_mutex_key(inod 1910 lockdep_annotate_inode_mutex_key(inode);
1911 security_d_instantiate(entry, inode); 1911 security_d_instantiate(entry, inode);
1912 spin_lock(&inode->i_lock); 1912 spin_lock(&inode->i_lock);
1913 __d_instantiate(entry, inode); 1913 __d_instantiate(entry, inode);
1914 WARN_ON(!(inode->i_state & I_NEW)); 1914 WARN_ON(!(inode->i_state & I_NEW));
1915 inode->i_state &= ~I_NEW & ~I_CREATIN 1915 inode->i_state &= ~I_NEW & ~I_CREATING;
1916 /* 1916 /*
1917 * Pairs with the barrier in prepare_ 1917 * Pairs with the barrier in prepare_to_wait_event() to make sure
1918 * ___wait_var_event() either sees th 1918 * ___wait_var_event() either sees the bit cleared or
1919 * waitqueue_active() check in wake_u 1919 * waitqueue_active() check in wake_up_var() sees the waiter.
1920 */ 1920 */
1921 smp_mb(); 1921 smp_mb();
1922 inode_wake_up_bit(inode, __I_NEW); 1922 inode_wake_up_bit(inode, __I_NEW);
1923 spin_unlock(&inode->i_lock); 1923 spin_unlock(&inode->i_lock);
1924 } 1924 }
1925 EXPORT_SYMBOL(d_instantiate_new); 1925 EXPORT_SYMBOL(d_instantiate_new);
1926 1926
1927 struct dentry *d_make_root(struct inode *root 1927 struct dentry *d_make_root(struct inode *root_inode)
1928 { 1928 {
1929 struct dentry *res = NULL; 1929 struct dentry *res = NULL;
1930 1930
1931 if (root_inode) { 1931 if (root_inode) {
1932 res = d_alloc_anon(root_inode 1932 res = d_alloc_anon(root_inode->i_sb);
1933 if (res) 1933 if (res)
1934 d_instantiate(res, ro 1934 d_instantiate(res, root_inode);
1935 else 1935 else
1936 iput(root_inode); 1936 iput(root_inode);
1937 } 1937 }
1938 return res; 1938 return res;
1939 } 1939 }
1940 EXPORT_SYMBOL(d_make_root); 1940 EXPORT_SYMBOL(d_make_root);
1941 1941
1942 static struct dentry *__d_obtain_alias(struct 1942 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
1943 { 1943 {
1944 struct super_block *sb; 1944 struct super_block *sb;
1945 struct dentry *new, *res; 1945 struct dentry *new, *res;
1946 1946
1947 if (!inode) 1947 if (!inode)
1948 return ERR_PTR(-ESTALE); 1948 return ERR_PTR(-ESTALE);
1949 if (IS_ERR(inode)) 1949 if (IS_ERR(inode))
1950 return ERR_CAST(inode); 1950 return ERR_CAST(inode);
1951 1951
1952 sb = inode->i_sb; 1952 sb = inode->i_sb;
1953 1953
1954 res = d_find_any_alias(inode); /* exi 1954 res = d_find_any_alias(inode); /* existing alias? */
1955 if (res) 1955 if (res)
1956 goto out; 1956 goto out;
1957 1957
1958 new = d_alloc_anon(sb); 1958 new = d_alloc_anon(sb);
1959 if (!new) { 1959 if (!new) {
1960 res = ERR_PTR(-ENOMEM); 1960 res = ERR_PTR(-ENOMEM);
1961 goto out; 1961 goto out;
1962 } 1962 }
1963 1963
1964 security_d_instantiate(new, inode); 1964 security_d_instantiate(new, inode);
1965 spin_lock(&inode->i_lock); 1965 spin_lock(&inode->i_lock);
1966 res = __d_find_any_alias(inode); /* r 1966 res = __d_find_any_alias(inode); /* recheck under lock */
1967 if (likely(!res)) { /* still no alias 1967 if (likely(!res)) { /* still no alias, attach a disconnected dentry */
1968 unsigned add_flags = d_flags_ 1968 unsigned add_flags = d_flags_for_inode(inode);
1969 1969
1970 if (disconnected) 1970 if (disconnected)
1971 add_flags |= DCACHE_D 1971 add_flags |= DCACHE_DISCONNECTED;
1972 1972
1973 spin_lock(&new->d_lock); 1973 spin_lock(&new->d_lock);
1974 __d_set_inode_and_type(new, i 1974 __d_set_inode_and_type(new, inode, add_flags);
1975 hlist_add_head(&new->d_u.d_al 1975 hlist_add_head(&new->d_u.d_alias, &inode->i_dentry);
1976 if (!disconnected) { 1976 if (!disconnected) {
1977 hlist_bl_lock(&sb->s_ 1977 hlist_bl_lock(&sb->s_roots);
1978 hlist_bl_add_head(&ne 1978 hlist_bl_add_head(&new->d_hash, &sb->s_roots);
1979 hlist_bl_unlock(&sb-> 1979 hlist_bl_unlock(&sb->s_roots);
1980 } 1980 }
1981 spin_unlock(&new->d_lock); 1981 spin_unlock(&new->d_lock);
1982 spin_unlock(&inode->i_lock); 1982 spin_unlock(&inode->i_lock);
1983 inode = NULL; /* consumed by 1983 inode = NULL; /* consumed by new->d_inode */
1984 res = new; 1984 res = new;
1985 } else { 1985 } else {
1986 spin_unlock(&inode->i_lock); 1986 spin_unlock(&inode->i_lock);
1987 dput(new); 1987 dput(new);
1988 } 1988 }
1989 1989
1990 out: 1990 out:
1991 iput(inode); 1991 iput(inode);
1992 return res; 1992 return res;
1993 } 1993 }
1994 1994
1995 /** 1995 /**
1996 * d_obtain_alias - find or allocate a DISCON 1996 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1997 * @inode: inode to allocate the dentry for 1997 * @inode: inode to allocate the dentry for
1998 * 1998 *
1999 * Obtain a dentry for an inode resulting fro 1999 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2000 * similar open by handle operations. The re 2000 * similar open by handle operations. The returned dentry may be anonymous,
2001 * or may have a full name (if the inode was 2001 * or may have a full name (if the inode was already in the cache).
2002 * 2002 *
2003 * When called on a directory inode, we must 2003 * When called on a directory inode, we must ensure that the inode only ever
2004 * has one dentry. If a dentry is found, tha 2004 * has one dentry. If a dentry is found, that is returned instead of
2005 * allocating a new one. 2005 * allocating a new one.
2006 * 2006 *
2007 * On successful return, the reference to the 2007 * On successful return, the reference to the inode has been transferred
2008 * to the dentry. In case of an error the re 2008 * to the dentry. In case of an error the reference on the inode is released.
2009 * To make it easier to use in export operati 2009 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2010 * be passed in and the error will be propaga 2010 * be passed in and the error will be propagated to the return value,
2011 * with a %NULL @inode replaced by ERR_PTR(-E 2011 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2012 */ 2012 */
2013 struct dentry *d_obtain_alias(struct inode *i 2013 struct dentry *d_obtain_alias(struct inode *inode)
2014 { 2014 {
2015 return __d_obtain_alias(inode, true); 2015 return __d_obtain_alias(inode, true);
2016 } 2016 }
2017 EXPORT_SYMBOL(d_obtain_alias); 2017 EXPORT_SYMBOL(d_obtain_alias);
2018 2018
2019 /** 2019 /**
2020 * d_obtain_root - find or allocate a dentry 2020 * d_obtain_root - find or allocate a dentry for a given inode
2021 * @inode: inode to allocate the dentry for 2021 * @inode: inode to allocate the dentry for
2022 * 2022 *
2023 * Obtain an IS_ROOT dentry for the root of a 2023 * Obtain an IS_ROOT dentry for the root of a filesystem.
2024 * 2024 *
2025 * We must ensure that directory inodes only 2025 * We must ensure that directory inodes only ever have one dentry. If a
2026 * dentry is found, that is returned instead 2026 * dentry is found, that is returned instead of allocating a new one.
2027 * 2027 *
2028 * On successful return, the reference to the 2028 * On successful return, the reference to the inode has been transferred
2029 * to the dentry. In case of an error the re 2029 * to the dentry. In case of an error the reference on the inode is
2030 * released. A %NULL or IS_ERR inode may be 2030 * released. A %NULL or IS_ERR inode may be passed in and will be the
2031 * error will be propagate to the return valu 2031 * error will be propagate to the return value, with a %NULL @inode
2032 * replaced by ERR_PTR(-ESTALE). 2032 * replaced by ERR_PTR(-ESTALE).
2033 */ 2033 */
2034 struct dentry *d_obtain_root(struct inode *in 2034 struct dentry *d_obtain_root(struct inode *inode)
2035 { 2035 {
2036 return __d_obtain_alias(inode, false) 2036 return __d_obtain_alias(inode, false);
2037 } 2037 }
2038 EXPORT_SYMBOL(d_obtain_root); 2038 EXPORT_SYMBOL(d_obtain_root);
2039 2039
2040 /** 2040 /**
2041 * d_add_ci - lookup or allocate new dentry w 2041 * d_add_ci - lookup or allocate new dentry with case-exact name
2042 * @inode: the inode case-insensitive lookup 2042 * @inode: the inode case-insensitive lookup has found
2043 * @dentry: the negative dentry that was pass 2043 * @dentry: the negative dentry that was passed to the parent's lookup func
2044 * @name: the case-exact name to be associa 2044 * @name: the case-exact name to be associated with the returned dentry
2045 * 2045 *
2046 * This is to avoid filling the dcache with c 2046 * This is to avoid filling the dcache with case-insensitive names to the
2047 * same inode, only the actual correct case i 2047 * same inode, only the actual correct case is stored in the dcache for
2048 * case-insensitive filesystems. 2048 * case-insensitive filesystems.
2049 * 2049 *
2050 * For a case-insensitive lookup match and if 2050 * For a case-insensitive lookup match and if the case-exact dentry
2051 * already exists in the dcache, use it and r 2051 * already exists in the dcache, use it and return it.
2052 * 2052 *
2053 * If no entry exists with the exact case nam 2053 * If no entry exists with the exact case name, allocate new dentry with
2054 * the exact case, and return the spliced ent 2054 * the exact case, and return the spliced entry.
2055 */ 2055 */
2056 struct dentry *d_add_ci(struct dentry *dentry 2056 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2057 struct qstr *name) 2057 struct qstr *name)
2058 { 2058 {
2059 struct dentry *found, *res; 2059 struct dentry *found, *res;
2060 2060
2061 /* 2061 /*
2062 * First check if a dentry matching t 2062 * First check if a dentry matching the name already exists,
2063 * if not go ahead and create it now. 2063 * if not go ahead and create it now.
2064 */ 2064 */
2065 found = d_hash_and_lookup(dentry->d_p 2065 found = d_hash_and_lookup(dentry->d_parent, name);
2066 if (found) { 2066 if (found) {
2067 iput(inode); 2067 iput(inode);
2068 return found; 2068 return found;
2069 } 2069 }
2070 if (d_in_lookup(dentry)) { 2070 if (d_in_lookup(dentry)) {
2071 found = d_alloc_parallel(dent 2071 found = d_alloc_parallel(dentry->d_parent, name,
2072 dentr 2072 dentry->d_wait);
2073 if (IS_ERR(found) || !d_in_lo 2073 if (IS_ERR(found) || !d_in_lookup(found)) {
2074 iput(inode); 2074 iput(inode);
2075 return found; 2075 return found;
2076 } 2076 }
2077 } else { 2077 } else {
2078 found = d_alloc(dentry->d_par 2078 found = d_alloc(dentry->d_parent, name);
2079 if (!found) { 2079 if (!found) {
2080 iput(inode); 2080 iput(inode);
2081 return ERR_PTR(-ENOME 2081 return ERR_PTR(-ENOMEM);
2082 } 2082 }
2083 } 2083 }
2084 res = d_splice_alias(inode, found); 2084 res = d_splice_alias(inode, found);
2085 if (res) { 2085 if (res) {
2086 d_lookup_done(found); 2086 d_lookup_done(found);
2087 dput(found); 2087 dput(found);
2088 return res; 2088 return res;
2089 } 2089 }
2090 return found; 2090 return found;
2091 } 2091 }
2092 EXPORT_SYMBOL(d_add_ci); 2092 EXPORT_SYMBOL(d_add_ci);
2093 2093
2094 /** 2094 /**
2095 * d_same_name - compare dentry name with cas 2095 * d_same_name - compare dentry name with case-exact name
2096 * @parent: parent dentry 2096 * @parent: parent dentry
2097 * @dentry: the negative dentry that was pass 2097 * @dentry: the negative dentry that was passed to the parent's lookup func
2098 * @name: the case-exact name to be associa 2098 * @name: the case-exact name to be associated with the returned dentry
2099 * 2099 *
2100 * Return: true if names are same, or false 2100 * Return: true if names are same, or false
2101 */ 2101 */
2102 bool d_same_name(const struct dentry *dentry, 2102 bool d_same_name(const struct dentry *dentry, const struct dentry *parent,
2103 const struct qstr *name) 2103 const struct qstr *name)
2104 { 2104 {
2105 if (likely(!(parent->d_flags & DCACHE 2105 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2106 if (dentry->d_name.len != nam 2106 if (dentry->d_name.len != name->len)
2107 return false; 2107 return false;
2108 return dentry_cmp(dentry, nam 2108 return dentry_cmp(dentry, name->name, name->len) == 0;
2109 } 2109 }
2110 return parent->d_op->d_compare(dentry 2110 return parent->d_op->d_compare(dentry,
2111 dentry 2111 dentry->d_name.len, dentry->d_name.name,
2112 name) 2112 name) == 0;
2113 } 2113 }
2114 EXPORT_SYMBOL_GPL(d_same_name); 2114 EXPORT_SYMBOL_GPL(d_same_name);
2115 2115
2116 /* 2116 /*
2117 * This is __d_lookup_rcu() when the parent d 2117 * This is __d_lookup_rcu() when the parent dentry has
2118 * DCACHE_OP_COMPARE, which makes things much 2118 * DCACHE_OP_COMPARE, which makes things much nastier.
2119 */ 2119 */
2120 static noinline struct dentry *__d_lookup_rcu 2120 static noinline struct dentry *__d_lookup_rcu_op_compare(
2121 const struct dentry *parent, 2121 const struct dentry *parent,
2122 const struct qstr *name, 2122 const struct qstr *name,
2123 unsigned *seqp) 2123 unsigned *seqp)
2124 { 2124 {
2125 u64 hashlen = name->hash_len; 2125 u64 hashlen = name->hash_len;
2126 struct hlist_bl_head *b = d_hash(hash 2126 struct hlist_bl_head *b = d_hash(hashlen);
2127 struct hlist_bl_node *node; 2127 struct hlist_bl_node *node;
2128 struct dentry *dentry; 2128 struct dentry *dentry;
2129 2129
2130 hlist_bl_for_each_entry_rcu(dentry, n 2130 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2131 int tlen; 2131 int tlen;
2132 const char *tname; 2132 const char *tname;
2133 unsigned seq; 2133 unsigned seq;
2134 2134
2135 seqretry: 2135 seqretry:
2136 seq = raw_seqcount_begin(&den 2136 seq = raw_seqcount_begin(&dentry->d_seq);
2137 if (dentry->d_parent != paren 2137 if (dentry->d_parent != parent)
2138 continue; 2138 continue;
2139 if (d_unhashed(dentry)) 2139 if (d_unhashed(dentry))
2140 continue; 2140 continue;
2141 if (dentry->d_name.hash != ha 2141 if (dentry->d_name.hash != hashlen_hash(hashlen))
2142 continue; 2142 continue;
2143 tlen = dentry->d_name.len; 2143 tlen = dentry->d_name.len;
2144 tname = dentry->d_name.name; 2144 tname = dentry->d_name.name;
2145 /* we want a consistent (name 2145 /* we want a consistent (name,len) pair */
2146 if (read_seqcount_retry(&dent 2146 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2147 cpu_relax(); 2147 cpu_relax();
2148 goto seqretry; 2148 goto seqretry;
2149 } 2149 }
2150 if (parent->d_op->d_compare(d 2150 if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0)
2151 continue; 2151 continue;
2152 *seqp = seq; 2152 *seqp = seq;
2153 return dentry; 2153 return dentry;
2154 } 2154 }
2155 return NULL; 2155 return NULL;
2156 } 2156 }
2157 2157
2158 /** 2158 /**
2159 * __d_lookup_rcu - search for a dentry (racy 2159 * __d_lookup_rcu - search for a dentry (racy, store-free)
2160 * @parent: parent dentry 2160 * @parent: parent dentry
2161 * @name: qstr of name we wish to find 2161 * @name: qstr of name we wish to find
2162 * @seqp: returns d_seq value at the point wh 2162 * @seqp: returns d_seq value at the point where the dentry was found
2163 * Returns: dentry, or NULL 2163 * Returns: dentry, or NULL
2164 * 2164 *
2165 * __d_lookup_rcu is the dcache lookup functi 2165 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2166 * resolution (store-free path walking) desig 2166 * resolution (store-free path walking) design described in
2167 * Documentation/filesystems/path-lookup.txt. 2167 * Documentation/filesystems/path-lookup.txt.
2168 * 2168 *
2169 * This is not to be used outside core vfs. 2169 * This is not to be used outside core vfs.
2170 * 2170 *
2171 * __d_lookup_rcu must only be used in rcu-wa 2171 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2172 * held, and rcu_read_lock held. The returned 2172 * held, and rcu_read_lock held. The returned dentry must not be stored into
2173 * without taking d_lock and checking d_seq s 2173 * without taking d_lock and checking d_seq sequence count against @seq
2174 * returned here. 2174 * returned here.
2175 * 2175 *
2176 * Alternatively, __d_lookup_rcu may be calle 2176 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2177 * the returned dentry, so long as its parent 2177 * the returned dentry, so long as its parent's seqlock is checked after the
2178 * child is looked up. Thus, an interlocking 2178 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2179 * is formed, giving integrity down the path 2179 * is formed, giving integrity down the path walk.
2180 * 2180 *
2181 * NOTE! The caller *has* to check the result 2181 * NOTE! The caller *has* to check the resulting dentry against the sequence
2182 * number we've returned before using any of 2182 * number we've returned before using any of the resulting dentry state!
2183 */ 2183 */
2184 struct dentry *__d_lookup_rcu(const struct de 2184 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2185 const struct 2185 const struct qstr *name,
2186 unsigned *seq 2186 unsigned *seqp)
2187 { 2187 {
2188 u64 hashlen = name->hash_len; 2188 u64 hashlen = name->hash_len;
2189 const unsigned char *str = name->name 2189 const unsigned char *str = name->name;
2190 struct hlist_bl_head *b = d_hash(hash 2190 struct hlist_bl_head *b = d_hash(hashlen);
2191 struct hlist_bl_node *node; 2191 struct hlist_bl_node *node;
2192 struct dentry *dentry; 2192 struct dentry *dentry;
2193 2193
2194 /* 2194 /*
2195 * Note: There is significant duplica 2195 * Note: There is significant duplication with __d_lookup_rcu which is
2196 * required to prevent single threade 2196 * required to prevent single threaded performance regressions
2197 * especially on architectures where 2197 * especially on architectures where smp_rmb (in seqcounts) are costly.
2198 * Keep the two functions in sync. 2198 * Keep the two functions in sync.
2199 */ 2199 */
2200 2200
2201 if (unlikely(parent->d_flags & DCACHE 2201 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE))
2202 return __d_lookup_rcu_op_comp 2202 return __d_lookup_rcu_op_compare(parent, name, seqp);
2203 2203
2204 /* 2204 /*
2205 * The hash list is protected using R 2205 * The hash list is protected using RCU.
2206 * 2206 *
2207 * Carefully use d_seq when comparing 2207 * Carefully use d_seq when comparing a candidate dentry, to avoid
2208 * races with d_move(). 2208 * races with d_move().
2209 * 2209 *
2210 * It is possible that concurrent ren 2210 * It is possible that concurrent renames can mess up our list
2211 * walk here and result in missing ou 2211 * walk here and result in missing our dentry, resulting in the
2212 * false-negative result. d_lookup() 2212 * false-negative result. d_lookup() protects against concurrent
2213 * renames using rename_lock seqlock. 2213 * renames using rename_lock seqlock.
2214 * 2214 *
2215 * See Documentation/filesystems/path 2215 * See Documentation/filesystems/path-lookup.txt for more details.
2216 */ 2216 */
2217 hlist_bl_for_each_entry_rcu(dentry, n 2217 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2218 unsigned seq; 2218 unsigned seq;
2219 2219
2220 /* 2220 /*
2221 * The dentry sequence count 2221 * The dentry sequence count protects us from concurrent
2222 * renames, and thus protects 2222 * renames, and thus protects parent and name fields.
2223 * 2223 *
2224 * The caller must perform a 2224 * The caller must perform a seqcount check in order
2225 * to do anything useful with 2225 * to do anything useful with the returned dentry.
2226 * 2226 *
2227 * NOTE! We do a "raw" seqcou 2227 * NOTE! We do a "raw" seqcount_begin here. That means that
2228 * we don't wait for the sequ 2228 * we don't wait for the sequence count to stabilize if it
2229 * is in the middle of a sequ 2229 * is in the middle of a sequence change. If we do the slow
2230 * dentry compare, we will do 2230 * dentry compare, we will do seqretries until it is stable,
2231 * and if we end up with a su 2231 * and if we end up with a successful lookup, we actually
2232 * want to exit RCU lookup an 2232 * want to exit RCU lookup anyway.
2233 * 2233 *
2234 * Note that raw_seqcount_beg 2234 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2235 * we are still guaranteed NU 2235 * we are still guaranteed NUL-termination of ->d_name.name.
2236 */ 2236 */
2237 seq = raw_seqcount_begin(&den 2237 seq = raw_seqcount_begin(&dentry->d_seq);
2238 if (dentry->d_parent != paren 2238 if (dentry->d_parent != parent)
2239 continue; 2239 continue;
2240 if (d_unhashed(dentry)) 2240 if (d_unhashed(dentry))
2241 continue; 2241 continue;
2242 if (dentry->d_name.hash_len ! 2242 if (dentry->d_name.hash_len != hashlen)
2243 continue; 2243 continue;
2244 if (dentry_cmp(dentry, str, h 2244 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2245 continue; 2245 continue;
2246 *seqp = seq; 2246 *seqp = seq;
2247 return dentry; 2247 return dentry;
2248 } 2248 }
2249 return NULL; 2249 return NULL;
2250 } 2250 }
2251 2251
2252 /** 2252 /**
2253 * d_lookup - search for a dentry 2253 * d_lookup - search for a dentry
2254 * @parent: parent dentry 2254 * @parent: parent dentry
2255 * @name: qstr of name we wish to find 2255 * @name: qstr of name we wish to find
2256 * Returns: dentry, or NULL 2256 * Returns: dentry, or NULL
2257 * 2257 *
2258 * d_lookup searches the children of the pare 2258 * d_lookup searches the children of the parent dentry for the name in
2259 * question. If the dentry is found its refer 2259 * question. If the dentry is found its reference count is incremented and the
2260 * dentry is returned. The caller must use dp 2260 * dentry is returned. The caller must use dput to free the entry when it has
2261 * finished using it. %NULL is returned if th 2261 * finished using it. %NULL is returned if the dentry does not exist.
2262 */ 2262 */
2263 struct dentry *d_lookup(const struct dentry * 2263 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2264 { 2264 {
2265 struct dentry *dentry; 2265 struct dentry *dentry;
2266 unsigned seq; 2266 unsigned seq;
2267 2267
2268 do { 2268 do {
2269 seq = read_seqbegin(&rename_l 2269 seq = read_seqbegin(&rename_lock);
2270 dentry = __d_lookup(parent, n 2270 dentry = __d_lookup(parent, name);
2271 if (dentry) 2271 if (dentry)
2272 break; 2272 break;
2273 } while (read_seqretry(&rename_lock, 2273 } while (read_seqretry(&rename_lock, seq));
2274 return dentry; 2274 return dentry;
2275 } 2275 }
2276 EXPORT_SYMBOL(d_lookup); 2276 EXPORT_SYMBOL(d_lookup);
2277 2277
2278 /** 2278 /**
2279 * __d_lookup - search for a dentry (racy) 2279 * __d_lookup - search for a dentry (racy)
2280 * @parent: parent dentry 2280 * @parent: parent dentry
2281 * @name: qstr of name we wish to find 2281 * @name: qstr of name we wish to find
2282 * Returns: dentry, or NULL 2282 * Returns: dentry, or NULL
2283 * 2283 *
2284 * __d_lookup is like d_lookup, however it ma 2284 * __d_lookup is like d_lookup, however it may (rarely) return a
2285 * false-negative result due to unrelated ren 2285 * false-negative result due to unrelated rename activity.
2286 * 2286 *
2287 * __d_lookup is slightly faster by avoiding 2287 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2288 * however it must be used carefully, eg. wit 2288 * however it must be used carefully, eg. with a following d_lookup in
2289 * the case of failure. 2289 * the case of failure.
2290 * 2290 *
2291 * __d_lookup callers must be commented. 2291 * __d_lookup callers must be commented.
2292 */ 2292 */
2293 struct dentry *__d_lookup(const struct dentry 2293 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2294 { 2294 {
2295 unsigned int hash = name->hash; 2295 unsigned int hash = name->hash;
2296 struct hlist_bl_head *b = d_hash(hash 2296 struct hlist_bl_head *b = d_hash(hash);
2297 struct hlist_bl_node *node; 2297 struct hlist_bl_node *node;
2298 struct dentry *found = NULL; 2298 struct dentry *found = NULL;
2299 struct dentry *dentry; 2299 struct dentry *dentry;
2300 2300
2301 /* 2301 /*
2302 * Note: There is significant duplica 2302 * Note: There is significant duplication with __d_lookup_rcu which is
2303 * required to prevent single threade 2303 * required to prevent single threaded performance regressions
2304 * especially on architectures where 2304 * especially on architectures where smp_rmb (in seqcounts) are costly.
2305 * Keep the two functions in sync. 2305 * Keep the two functions in sync.
2306 */ 2306 */
2307 2307
2308 /* 2308 /*
2309 * The hash list is protected using R 2309 * The hash list is protected using RCU.
2310 * 2310 *
2311 * Take d_lock when comparing a candi 2311 * Take d_lock when comparing a candidate dentry, to avoid races
2312 * with d_move(). 2312 * with d_move().
2313 * 2313 *
2314 * It is possible that concurrent ren 2314 * It is possible that concurrent renames can mess up our list
2315 * walk here and result in missing ou 2315 * walk here and result in missing our dentry, resulting in the
2316 * false-negative result. d_lookup() 2316 * false-negative result. d_lookup() protects against concurrent
2317 * renames using rename_lock seqlock. 2317 * renames using rename_lock seqlock.
2318 * 2318 *
2319 * See Documentation/filesystems/path 2319 * See Documentation/filesystems/path-lookup.txt for more details.
2320 */ 2320 */
2321 rcu_read_lock(); 2321 rcu_read_lock();
2322 2322
2323 hlist_bl_for_each_entry_rcu(dentry, n 2323 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2324 2324
2325 if (dentry->d_name.hash != ha 2325 if (dentry->d_name.hash != hash)
2326 continue; 2326 continue;
2327 2327
2328 spin_lock(&dentry->d_lock); 2328 spin_lock(&dentry->d_lock);
2329 if (dentry->d_parent != paren 2329 if (dentry->d_parent != parent)
2330 goto next; 2330 goto next;
2331 if (d_unhashed(dentry)) 2331 if (d_unhashed(dentry))
2332 goto next; 2332 goto next;
2333 2333
2334 if (!d_same_name(dentry, pare 2334 if (!d_same_name(dentry, parent, name))
2335 goto next; 2335 goto next;
2336 2336
2337 dentry->d_lockref.count++; 2337 dentry->d_lockref.count++;
2338 found = dentry; 2338 found = dentry;
2339 spin_unlock(&dentry->d_lock); 2339 spin_unlock(&dentry->d_lock);
2340 break; 2340 break;
2341 next: 2341 next:
2342 spin_unlock(&dentry->d_lock); 2342 spin_unlock(&dentry->d_lock);
2343 } 2343 }
2344 rcu_read_unlock(); 2344 rcu_read_unlock();
2345 2345
2346 return found; 2346 return found;
2347 } 2347 }
2348 2348
2349 /** 2349 /**
2350 * d_hash_and_lookup - hash the qstr then sea 2350 * d_hash_and_lookup - hash the qstr then search for a dentry
2351 * @dir: Directory to search in 2351 * @dir: Directory to search in
2352 * @name: qstr of name we wish to find 2352 * @name: qstr of name we wish to find
2353 * 2353 *
2354 * On lookup failure NULL is returned; on bad 2354 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2355 */ 2355 */
2356 struct dentry *d_hash_and_lookup(struct dentr 2356 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2357 { 2357 {
2358 /* 2358 /*
2359 * Check for a fs-specific hash funct 2359 * Check for a fs-specific hash function. Note that we must
2360 * calculate the standard hash first, 2360 * calculate the standard hash first, as the d_op->d_hash()
2361 * routine may choose to leave the ha 2361 * routine may choose to leave the hash value unchanged.
2362 */ 2362 */
2363 name->hash = full_name_hash(dir, name 2363 name->hash = full_name_hash(dir, name->name, name->len);
2364 if (dir->d_flags & DCACHE_OP_HASH) { 2364 if (dir->d_flags & DCACHE_OP_HASH) {
2365 int err = dir->d_op->d_hash(d 2365 int err = dir->d_op->d_hash(dir, name);
2366 if (unlikely(err < 0)) 2366 if (unlikely(err < 0))
2367 return ERR_PTR(err); 2367 return ERR_PTR(err);
2368 } 2368 }
2369 return d_lookup(dir, name); 2369 return d_lookup(dir, name);
2370 } 2370 }
2371 EXPORT_SYMBOL(d_hash_and_lookup); 2371 EXPORT_SYMBOL(d_hash_and_lookup);
2372 2372
2373 /* 2373 /*
2374 * When a file is deleted, we have two option 2374 * When a file is deleted, we have two options:
2375 * - turn this dentry into a negative dentry 2375 * - turn this dentry into a negative dentry
2376 * - unhash this dentry and free it. 2376 * - unhash this dentry and free it.
2377 * 2377 *
2378 * Usually, we want to just turn this into 2378 * Usually, we want to just turn this into
2379 * a negative dentry, but if anybody else is 2379 * a negative dentry, but if anybody else is
2380 * currently using the dentry or the inode 2380 * currently using the dentry or the inode
2381 * we can't do that and we fall back on remov 2381 * we can't do that and we fall back on removing
2382 * it from the hash queues and waiting for 2382 * it from the hash queues and waiting for
2383 * it to be deleted later when it has no user 2383 * it to be deleted later when it has no users
2384 */ 2384 */
2385 2385
2386 /** 2386 /**
2387 * d_delete - delete a dentry 2387 * d_delete - delete a dentry
2388 * @dentry: The dentry to delete 2388 * @dentry: The dentry to delete
2389 * 2389 *
2390 * Turn the dentry into a negative dentry if 2390 * Turn the dentry into a negative dentry if possible, otherwise
2391 * remove it from the hash queues so it can b 2391 * remove it from the hash queues so it can be deleted later
2392 */ 2392 */
2393 2393
2394 void d_delete(struct dentry * dentry) 2394 void d_delete(struct dentry * dentry)
2395 { 2395 {
2396 struct inode *inode = dentry->d_inode 2396 struct inode *inode = dentry->d_inode;
2397 2397
2398 spin_lock(&inode->i_lock); 2398 spin_lock(&inode->i_lock);
2399 spin_lock(&dentry->d_lock); 2399 spin_lock(&dentry->d_lock);
2400 /* 2400 /*
2401 * Are we the only user? 2401 * Are we the only user?
2402 */ 2402 */
2403 if (dentry->d_lockref.count == 1) { 2403 if (dentry->d_lockref.count == 1) {
2404 dentry->d_flags &= ~DCACHE_CA 2404 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2405 dentry_unlink_inode(dentry); 2405 dentry_unlink_inode(dentry);
2406 } else { 2406 } else {
2407 __d_drop(dentry); 2407 __d_drop(dentry);
2408 spin_unlock(&dentry->d_lock); 2408 spin_unlock(&dentry->d_lock);
2409 spin_unlock(&inode->i_lock); 2409 spin_unlock(&inode->i_lock);
2410 } 2410 }
2411 } 2411 }
2412 EXPORT_SYMBOL(d_delete); 2412 EXPORT_SYMBOL(d_delete);
2413 2413
2414 static void __d_rehash(struct dentry *entry) 2414 static void __d_rehash(struct dentry *entry)
2415 { 2415 {
2416 struct hlist_bl_head *b = d_hash(entr 2416 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2417 2417
2418 hlist_bl_lock(b); 2418 hlist_bl_lock(b);
2419 hlist_bl_add_head_rcu(&entry->d_hash, 2419 hlist_bl_add_head_rcu(&entry->d_hash, b);
2420 hlist_bl_unlock(b); 2420 hlist_bl_unlock(b);
2421 } 2421 }
2422 2422
2423 /** 2423 /**
2424 * d_rehash - add an entry back to the ha 2424 * d_rehash - add an entry back to the hash
2425 * @entry: dentry to add to the hash 2425 * @entry: dentry to add to the hash
2426 * 2426 *
2427 * Adds a dentry to the hash according to its 2427 * Adds a dentry to the hash according to its name.
2428 */ 2428 */
2429 2429
2430 void d_rehash(struct dentry * entry) 2430 void d_rehash(struct dentry * entry)
2431 { 2431 {
2432 spin_lock(&entry->d_lock); 2432 spin_lock(&entry->d_lock);
2433 __d_rehash(entry); 2433 __d_rehash(entry);
2434 spin_unlock(&entry->d_lock); 2434 spin_unlock(&entry->d_lock);
2435 } 2435 }
2436 EXPORT_SYMBOL(d_rehash); 2436 EXPORT_SYMBOL(d_rehash);
2437 2437
2438 static inline unsigned start_dir_add(struct i 2438 static inline unsigned start_dir_add(struct inode *dir)
2439 { 2439 {
2440 preempt_disable_nested(); 2440 preempt_disable_nested();
2441 for (;;) { 2441 for (;;) {
2442 unsigned n = dir->i_dir_seq; 2442 unsigned n = dir->i_dir_seq;
2443 if (!(n & 1) && cmpxchg(&dir- 2443 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2444 return n; 2444 return n;
2445 cpu_relax(); 2445 cpu_relax();
2446 } 2446 }
2447 } 2447 }
2448 2448
2449 static inline void end_dir_add(struct inode * 2449 static inline void end_dir_add(struct inode *dir, unsigned int n,
2450 wait_queue_hea 2450 wait_queue_head_t *d_wait)
2451 { 2451 {
2452 smp_store_release(&dir->i_dir_seq, n 2452 smp_store_release(&dir->i_dir_seq, n + 2);
2453 preempt_enable_nested(); 2453 preempt_enable_nested();
2454 wake_up_all(d_wait); 2454 wake_up_all(d_wait);
2455 } 2455 }
2456 2456
2457 static void d_wait_lookup(struct dentry *dent 2457 static void d_wait_lookup(struct dentry *dentry)
2458 { 2458 {
2459 if (d_in_lookup(dentry)) { 2459 if (d_in_lookup(dentry)) {
2460 DECLARE_WAITQUEUE(wait, curre 2460 DECLARE_WAITQUEUE(wait, current);
2461 add_wait_queue(dentry->d_wait 2461 add_wait_queue(dentry->d_wait, &wait);
2462 do { 2462 do {
2463 set_current_state(TAS 2463 set_current_state(TASK_UNINTERRUPTIBLE);
2464 spin_unlock(&dentry-> 2464 spin_unlock(&dentry->d_lock);
2465 schedule(); 2465 schedule();
2466 spin_lock(&dentry->d_ 2466 spin_lock(&dentry->d_lock);
2467 } while (d_in_lookup(dentry)) 2467 } while (d_in_lookup(dentry));
2468 } 2468 }
2469 } 2469 }
2470 2470
2471 struct dentry *d_alloc_parallel(struct dentry 2471 struct dentry *d_alloc_parallel(struct dentry *parent,
2472 const struct 2472 const struct qstr *name,
2473 wait_queue_he 2473 wait_queue_head_t *wq)
2474 { 2474 {
2475 unsigned int hash = name->hash; 2475 unsigned int hash = name->hash;
2476 struct hlist_bl_head *b = in_lookup_h 2476 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2477 struct hlist_bl_node *node; 2477 struct hlist_bl_node *node;
2478 struct dentry *new = d_alloc(parent, 2478 struct dentry *new = d_alloc(parent, name);
2479 struct dentry *dentry; 2479 struct dentry *dentry;
2480 unsigned seq, r_seq, d_seq; 2480 unsigned seq, r_seq, d_seq;
2481 2481
2482 if (unlikely(!new)) 2482 if (unlikely(!new))
2483 return ERR_PTR(-ENOMEM); 2483 return ERR_PTR(-ENOMEM);
2484 2484
2485 retry: 2485 retry:
2486 rcu_read_lock(); 2486 rcu_read_lock();
2487 seq = smp_load_acquire(&parent->d_ino 2487 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2488 r_seq = read_seqbegin(&rename_lock); 2488 r_seq = read_seqbegin(&rename_lock);
2489 dentry = __d_lookup_rcu(parent, name, 2489 dentry = __d_lookup_rcu(parent, name, &d_seq);
2490 if (unlikely(dentry)) { 2490 if (unlikely(dentry)) {
2491 if (!lockref_get_not_dead(&de 2491 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2492 rcu_read_unlock(); 2492 rcu_read_unlock();
2493 goto retry; 2493 goto retry;
2494 } 2494 }
2495 if (read_seqcount_retry(&dent 2495 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2496 rcu_read_unlock(); 2496 rcu_read_unlock();
2497 dput(dentry); 2497 dput(dentry);
2498 goto retry; 2498 goto retry;
2499 } 2499 }
2500 rcu_read_unlock(); 2500 rcu_read_unlock();
2501 dput(new); 2501 dput(new);
2502 return dentry; 2502 return dentry;
2503 } 2503 }
2504 if (unlikely(read_seqretry(&rename_lo 2504 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2505 rcu_read_unlock(); 2505 rcu_read_unlock();
2506 goto retry; 2506 goto retry;
2507 } 2507 }
2508 2508
2509 if (unlikely(seq & 1)) { 2509 if (unlikely(seq & 1)) {
2510 rcu_read_unlock(); 2510 rcu_read_unlock();
2511 goto retry; 2511 goto retry;
2512 } 2512 }
2513 2513
2514 hlist_bl_lock(b); 2514 hlist_bl_lock(b);
2515 if (unlikely(READ_ONCE(parent->d_inod 2515 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2516 hlist_bl_unlock(b); 2516 hlist_bl_unlock(b);
2517 rcu_read_unlock(); 2517 rcu_read_unlock();
2518 goto retry; 2518 goto retry;
2519 } 2519 }
2520 /* 2520 /*
2521 * No changes for the parent since th 2521 * No changes for the parent since the beginning of d_lookup().
2522 * Since all removals from the chain 2522 * Since all removals from the chain happen with hlist_bl_lock(),
2523 * any potential in-lookup matches ar 2523 * any potential in-lookup matches are going to stay here until
2524 * we unlock the chain. All fields a 2524 * we unlock the chain. All fields are stable in everything
2525 * we encounter. 2525 * we encounter.
2526 */ 2526 */
2527 hlist_bl_for_each_entry(dentry, node, 2527 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2528 if (dentry->d_name.hash != ha 2528 if (dentry->d_name.hash != hash)
2529 continue; 2529 continue;
2530 if (dentry->d_parent != paren 2530 if (dentry->d_parent != parent)
2531 continue; 2531 continue;
2532 if (!d_same_name(dentry, pare 2532 if (!d_same_name(dentry, parent, name))
2533 continue; 2533 continue;
2534 hlist_bl_unlock(b); 2534 hlist_bl_unlock(b);
2535 /* now we can try to grab a r 2535 /* now we can try to grab a reference */
2536 if (!lockref_get_not_dead(&de 2536 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2537 rcu_read_unlock(); 2537 rcu_read_unlock();
2538 goto retry; 2538 goto retry;
2539 } 2539 }
2540 2540
2541 rcu_read_unlock(); 2541 rcu_read_unlock();
2542 /* 2542 /*
2543 * somebody is likely to be s 2543 * somebody is likely to be still doing lookup for it;
2544 * wait for them to finish 2544 * wait for them to finish
2545 */ 2545 */
2546 spin_lock(&dentry->d_lock); 2546 spin_lock(&dentry->d_lock);
2547 d_wait_lookup(dentry); 2547 d_wait_lookup(dentry);
2548 /* 2548 /*
2549 * it's not in-lookup anymore 2549 * it's not in-lookup anymore; in principle we should repeat
2550 * everything from dcache loo 2550 * everything from dcache lookup, but it's likely to be what
2551 * d_lookup() would've found 2551 * d_lookup() would've found anyway. If it is, just return it;
2552 * otherwise we really have t 2552 * otherwise we really have to repeat the whole thing.
2553 */ 2553 */
2554 if (unlikely(dentry->d_name.h 2554 if (unlikely(dentry->d_name.hash != hash))
2555 goto mismatch; 2555 goto mismatch;
2556 if (unlikely(dentry->d_parent 2556 if (unlikely(dentry->d_parent != parent))
2557 goto mismatch; 2557 goto mismatch;
2558 if (unlikely(d_unhashed(dentr 2558 if (unlikely(d_unhashed(dentry)))
2559 goto mismatch; 2559 goto mismatch;
2560 if (unlikely(!d_same_name(den 2560 if (unlikely(!d_same_name(dentry, parent, name)))
2561 goto mismatch; 2561 goto mismatch;
2562 /* OK, it *is* a hashed match 2562 /* OK, it *is* a hashed match; return it */
2563 spin_unlock(&dentry->d_lock); 2563 spin_unlock(&dentry->d_lock);
2564 dput(new); 2564 dput(new);
2565 return dentry; 2565 return dentry;
2566 } 2566 }
2567 rcu_read_unlock(); 2567 rcu_read_unlock();
2568 /* we can't take ->d_lock here; it's 2568 /* we can't take ->d_lock here; it's OK, though. */
2569 new->d_flags |= DCACHE_PAR_LOOKUP; 2569 new->d_flags |= DCACHE_PAR_LOOKUP;
2570 new->d_wait = wq; 2570 new->d_wait = wq;
2571 hlist_bl_add_head(&new->d_u.d_in_look 2571 hlist_bl_add_head(&new->d_u.d_in_lookup_hash, b);
2572 hlist_bl_unlock(b); 2572 hlist_bl_unlock(b);
2573 return new; 2573 return new;
2574 mismatch: 2574 mismatch:
2575 spin_unlock(&dentry->d_lock); 2575 spin_unlock(&dentry->d_lock);
2576 dput(dentry); 2576 dput(dentry);
2577 goto retry; 2577 goto retry;
2578 } 2578 }
2579 EXPORT_SYMBOL(d_alloc_parallel); 2579 EXPORT_SYMBOL(d_alloc_parallel);
2580 2580
2581 /* 2581 /*
2582 * - Unhash the dentry 2582 * - Unhash the dentry
2583 * - Retrieve and clear the waitqueue head in 2583 * - Retrieve and clear the waitqueue head in dentry
2584 * - Return the waitqueue head 2584 * - Return the waitqueue head
2585 */ 2585 */
2586 static wait_queue_head_t *__d_lookup_unhash(s 2586 static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry)
2587 { 2587 {
2588 wait_queue_head_t *d_wait; 2588 wait_queue_head_t *d_wait;
2589 struct hlist_bl_head *b; 2589 struct hlist_bl_head *b;
2590 2590
2591 lockdep_assert_held(&dentry->d_lock); 2591 lockdep_assert_held(&dentry->d_lock);
2592 2592
2593 b = in_lookup_hash(dentry->d_parent, 2593 b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash);
2594 hlist_bl_lock(b); 2594 hlist_bl_lock(b);
2595 dentry->d_flags &= ~DCACHE_PAR_LOOKUP 2595 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2596 __hlist_bl_del(&dentry->d_u.d_in_look 2596 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2597 d_wait = dentry->d_wait; 2597 d_wait = dentry->d_wait;
2598 dentry->d_wait = NULL; 2598 dentry->d_wait = NULL;
2599 hlist_bl_unlock(b); 2599 hlist_bl_unlock(b);
2600 INIT_HLIST_NODE(&dentry->d_u.d_alias) 2600 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2601 INIT_LIST_HEAD(&dentry->d_lru); 2601 INIT_LIST_HEAD(&dentry->d_lru);
2602 return d_wait; 2602 return d_wait;
2603 } 2603 }
2604 2604
2605 void __d_lookup_unhash_wake(struct dentry *de 2605 void __d_lookup_unhash_wake(struct dentry *dentry)
2606 { 2606 {
2607 spin_lock(&dentry->d_lock); 2607 spin_lock(&dentry->d_lock);
2608 wake_up_all(__d_lookup_unhash(dentry) 2608 wake_up_all(__d_lookup_unhash(dentry));
2609 spin_unlock(&dentry->d_lock); 2609 spin_unlock(&dentry->d_lock);
2610 } 2610 }
2611 EXPORT_SYMBOL(__d_lookup_unhash_wake); 2611 EXPORT_SYMBOL(__d_lookup_unhash_wake);
2612 2612
2613 /* inode->i_lock held if inode is non-NULL */ 2613 /* inode->i_lock held if inode is non-NULL */
2614 2614
2615 static inline void __d_add(struct dentry *den 2615 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2616 { 2616 {
2617 wait_queue_head_t *d_wait; 2617 wait_queue_head_t *d_wait;
2618 struct inode *dir = NULL; 2618 struct inode *dir = NULL;
2619 unsigned n; 2619 unsigned n;
2620 spin_lock(&dentry->d_lock); 2620 spin_lock(&dentry->d_lock);
2621 if (unlikely(d_in_lookup(dentry))) { 2621 if (unlikely(d_in_lookup(dentry))) {
2622 dir = dentry->d_parent->d_ino 2622 dir = dentry->d_parent->d_inode;
2623 n = start_dir_add(dir); 2623 n = start_dir_add(dir);
2624 d_wait = __d_lookup_unhash(de 2624 d_wait = __d_lookup_unhash(dentry);
2625 } 2625 }
2626 if (inode) { 2626 if (inode) {
2627 unsigned add_flags = d_flags_ 2627 unsigned add_flags = d_flags_for_inode(inode);
2628 hlist_add_head(&dentry->d_u.d 2628 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2629 raw_write_seqcount_begin(&den 2629 raw_write_seqcount_begin(&dentry->d_seq);
2630 __d_set_inode_and_type(dentry 2630 __d_set_inode_and_type(dentry, inode, add_flags);
2631 raw_write_seqcount_end(&dentr 2631 raw_write_seqcount_end(&dentry->d_seq);
2632 fsnotify_update_flags(dentry) 2632 fsnotify_update_flags(dentry);
2633 } 2633 }
2634 __d_rehash(dentry); 2634 __d_rehash(dentry);
2635 if (dir) 2635 if (dir)
2636 end_dir_add(dir, n, d_wait); 2636 end_dir_add(dir, n, d_wait);
2637 spin_unlock(&dentry->d_lock); 2637 spin_unlock(&dentry->d_lock);
2638 if (inode) 2638 if (inode)
2639 spin_unlock(&inode->i_lock); 2639 spin_unlock(&inode->i_lock);
2640 } 2640 }
2641 2641
2642 /** 2642 /**
2643 * d_add - add dentry to hash queues 2643 * d_add - add dentry to hash queues
2644 * @entry: dentry to add 2644 * @entry: dentry to add
2645 * @inode: The inode to attach to this dentry 2645 * @inode: The inode to attach to this dentry
2646 * 2646 *
2647 * This adds the entry to the hash queues and 2647 * This adds the entry to the hash queues and initializes @inode.
2648 * The entry was actually filled in earlier d 2648 * The entry was actually filled in earlier during d_alloc().
2649 */ 2649 */
2650 2650
2651 void d_add(struct dentry *entry, struct inode 2651 void d_add(struct dentry *entry, struct inode *inode)
2652 { 2652 {
2653 if (inode) { 2653 if (inode) {
2654 security_d_instantiate(entry, 2654 security_d_instantiate(entry, inode);
2655 spin_lock(&inode->i_lock); 2655 spin_lock(&inode->i_lock);
2656 } 2656 }
2657 __d_add(entry, inode); 2657 __d_add(entry, inode);
2658 } 2658 }
2659 EXPORT_SYMBOL(d_add); 2659 EXPORT_SYMBOL(d_add);
2660 2660
2661 /** 2661 /**
2662 * d_exact_alias - find and hash an exact unh 2662 * d_exact_alias - find and hash an exact unhashed alias
2663 * @entry: dentry to add 2663 * @entry: dentry to add
2664 * @inode: The inode to go with this dentry 2664 * @inode: The inode to go with this dentry
2665 * 2665 *
2666 * If an unhashed dentry with the same name/p 2666 * If an unhashed dentry with the same name/parent and desired
2667 * inode already exists, hash and return it. 2667 * inode already exists, hash and return it. Otherwise, return
2668 * NULL. 2668 * NULL.
2669 * 2669 *
2670 * Parent directory should be locked. 2670 * Parent directory should be locked.
2671 */ 2671 */
2672 struct dentry *d_exact_alias(struct dentry *e 2672 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2673 { 2673 {
2674 struct dentry *alias; 2674 struct dentry *alias;
2675 unsigned int hash = entry->d_name.has 2675 unsigned int hash = entry->d_name.hash;
2676 2676
2677 spin_lock(&inode->i_lock); 2677 spin_lock(&inode->i_lock);
2678 hlist_for_each_entry(alias, &inode->i 2678 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2679 /* 2679 /*
2680 * Don't need alias->d_lock h 2680 * Don't need alias->d_lock here, because aliases with
2681 * d_parent == entry->d_paren 2681 * d_parent == entry->d_parent are not subject to name or
2682 * parent changes, because th 2682 * parent changes, because the parent inode i_mutex is held.
2683 */ 2683 */
2684 if (alias->d_name.hash != has 2684 if (alias->d_name.hash != hash)
2685 continue; 2685 continue;
2686 if (alias->d_parent != entry- 2686 if (alias->d_parent != entry->d_parent)
2687 continue; 2687 continue;
2688 if (!d_same_name(alias, entry 2688 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2689 continue; 2689 continue;
2690 spin_lock(&alias->d_lock); 2690 spin_lock(&alias->d_lock);
2691 if (!d_unhashed(alias)) { 2691 if (!d_unhashed(alias)) {
2692 spin_unlock(&alias->d 2692 spin_unlock(&alias->d_lock);
2693 alias = NULL; 2693 alias = NULL;
2694 } else { 2694 } else {
2695 dget_dlock(alias); 2695 dget_dlock(alias);
2696 __d_rehash(alias); 2696 __d_rehash(alias);
2697 spin_unlock(&alias->d 2697 spin_unlock(&alias->d_lock);
2698 } 2698 }
2699 spin_unlock(&inode->i_lock); 2699 spin_unlock(&inode->i_lock);
2700 return alias; 2700 return alias;
2701 } 2701 }
2702 spin_unlock(&inode->i_lock); 2702 spin_unlock(&inode->i_lock);
2703 return NULL; 2703 return NULL;
2704 } 2704 }
2705 EXPORT_SYMBOL(d_exact_alias); 2705 EXPORT_SYMBOL(d_exact_alias);
2706 2706
2707 static void swap_names(struct dentry *dentry, 2707 static void swap_names(struct dentry *dentry, struct dentry *target)
2708 { 2708 {
2709 if (unlikely(dname_external(target))) 2709 if (unlikely(dname_external(target))) {
2710 if (unlikely(dname_external(d 2710 if (unlikely(dname_external(dentry))) {
2711 /* 2711 /*
2712 * Both external: swa 2712 * Both external: swap the pointers
2713 */ 2713 */
2714 swap(target->d_name.n 2714 swap(target->d_name.name, dentry->d_name.name);
2715 } else { 2715 } else {
2716 /* 2716 /*
2717 * dentry:internal, t 2717 * dentry:internal, target:external. Steal target's
2718 * storage and make t 2718 * storage and make target internal.
2719 */ 2719 */
2720 memcpy(target->d_inam 2720 memcpy(target->d_iname, dentry->d_name.name,
2721 dentr 2721 dentry->d_name.len + 1);
2722 dentry->d_name.name = 2722 dentry->d_name.name = target->d_name.name;
2723 target->d_name.name = 2723 target->d_name.name = target->d_iname;
2724 } 2724 }
2725 } else { 2725 } else {
2726 if (unlikely(dname_external(d 2726 if (unlikely(dname_external(dentry))) {
2727 /* 2727 /*
2728 * dentry:external, t 2728 * dentry:external, target:internal. Give dentry's
2729 * storage to target 2729 * storage to target and make dentry internal
2730 */ 2730 */
2731 memcpy(dentry->d_inam 2731 memcpy(dentry->d_iname, target->d_name.name,
2732 targe 2732 target->d_name.len + 1);
2733 target->d_name.name = 2733 target->d_name.name = dentry->d_name.name;
2734 dentry->d_name.name = 2734 dentry->d_name.name = dentry->d_iname;
2735 } else { 2735 } else {
2736 /* 2736 /*
2737 * Both are internal. 2737 * Both are internal.
2738 */ 2738 */
2739 unsigned int i; 2739 unsigned int i;
2740 BUILD_BUG_ON(!IS_ALIG 2740 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2741 for (i = 0; i < DNAME 2741 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2742 swap(((long * 2742 swap(((long *) &dentry->d_iname)[i],
2743 ((long * 2743 ((long *) &target->d_iname)[i]);
2744 } 2744 }
2745 } 2745 }
2746 } 2746 }
2747 swap(dentry->d_name.hash_len, target- 2747 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2748 } 2748 }
2749 2749
2750 static void copy_name(struct dentry *dentry, 2750 static void copy_name(struct dentry *dentry, struct dentry *target)
2751 { 2751 {
2752 struct external_name *old_name = NULL 2752 struct external_name *old_name = NULL;
2753 if (unlikely(dname_external(dentry))) 2753 if (unlikely(dname_external(dentry)))
2754 old_name = external_name(dent 2754 old_name = external_name(dentry);
2755 if (unlikely(dname_external(target))) 2755 if (unlikely(dname_external(target))) {
2756 atomic_inc(&external_name(tar 2756 atomic_inc(&external_name(target)->u.count);
2757 dentry->d_name = target->d_na 2757 dentry->d_name = target->d_name;
2758 } else { 2758 } else {
2759 memcpy(dentry->d_iname, targe 2759 memcpy(dentry->d_iname, target->d_name.name,
2760 target->d_nam 2760 target->d_name.len + 1);
2761 dentry->d_name.name = dentry- 2761 dentry->d_name.name = dentry->d_iname;
2762 dentry->d_name.hash_len = tar 2762 dentry->d_name.hash_len = target->d_name.hash_len;
2763 } 2763 }
2764 if (old_name && likely(atomic_dec_and 2764 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2765 kfree_rcu(old_name, u.head); 2765 kfree_rcu(old_name, u.head);
2766 } 2766 }
2767 2767
2768 /* 2768 /*
2769 * __d_move - move a dentry 2769 * __d_move - move a dentry
2770 * @dentry: entry to move 2770 * @dentry: entry to move
2771 * @target: new dentry 2771 * @target: new dentry
2772 * @exchange: exchange the two dentries 2772 * @exchange: exchange the two dentries
2773 * 2773 *
2774 * Update the dcache to reflect the move of a 2774 * Update the dcache to reflect the move of a file name. Negative
2775 * dcache entries should not be moved in this 2775 * dcache entries should not be moved in this way. Caller must hold
2776 * rename_lock, the i_mutex of the source and 2776 * rename_lock, the i_mutex of the source and target directories,
2777 * and the sb->s_vfs_rename_mutex if they dif 2777 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2778 */ 2778 */
2779 static void __d_move(struct dentry *dentry, s 2779 static void __d_move(struct dentry *dentry, struct dentry *target,
2780 bool exchange) 2780 bool exchange)
2781 { 2781 {
2782 struct dentry *old_parent, *p; 2782 struct dentry *old_parent, *p;
2783 wait_queue_head_t *d_wait; 2783 wait_queue_head_t *d_wait;
2784 struct inode *dir = NULL; 2784 struct inode *dir = NULL;
2785 unsigned n; 2785 unsigned n;
2786 2786
2787 WARN_ON(!dentry->d_inode); 2787 WARN_ON(!dentry->d_inode);
2788 if (WARN_ON(dentry == target)) 2788 if (WARN_ON(dentry == target))
2789 return; 2789 return;
2790 2790
2791 BUG_ON(d_ancestor(target, dentry)); 2791 BUG_ON(d_ancestor(target, dentry));
2792 old_parent = dentry->d_parent; 2792 old_parent = dentry->d_parent;
2793 p = d_ancestor(old_parent, target); 2793 p = d_ancestor(old_parent, target);
2794 if (IS_ROOT(dentry)) { 2794 if (IS_ROOT(dentry)) {
2795 BUG_ON(p); 2795 BUG_ON(p);
2796 spin_lock(&target->d_parent-> 2796 spin_lock(&target->d_parent->d_lock);
2797 } else if (!p) { 2797 } else if (!p) {
2798 /* target is not a descendent 2798 /* target is not a descendent of dentry->d_parent */
2799 spin_lock(&target->d_parent-> 2799 spin_lock(&target->d_parent->d_lock);
2800 spin_lock_nested(&old_parent- 2800 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2801 } else { 2801 } else {
2802 BUG_ON(p == dentry); 2802 BUG_ON(p == dentry);
2803 spin_lock(&old_parent->d_lock 2803 spin_lock(&old_parent->d_lock);
2804 if (p != target) 2804 if (p != target)
2805 spin_lock_nested(&tar 2805 spin_lock_nested(&target->d_parent->d_lock,
2806 DENTR 2806 DENTRY_D_LOCK_NESTED);
2807 } 2807 }
2808 spin_lock_nested(&dentry->d_lock, 2); 2808 spin_lock_nested(&dentry->d_lock, 2);
2809 spin_lock_nested(&target->d_lock, 3); 2809 spin_lock_nested(&target->d_lock, 3);
2810 2810
2811 if (unlikely(d_in_lookup(target))) { 2811 if (unlikely(d_in_lookup(target))) {
2812 dir = target->d_parent->d_ino 2812 dir = target->d_parent->d_inode;
2813 n = start_dir_add(dir); 2813 n = start_dir_add(dir);
2814 d_wait = __d_lookup_unhash(ta 2814 d_wait = __d_lookup_unhash(target);
2815 } 2815 }
2816 2816
2817 write_seqcount_begin(&dentry->d_seq); 2817 write_seqcount_begin(&dentry->d_seq);
2818 write_seqcount_begin_nested(&target-> 2818 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2819 2819
2820 /* unhash both */ 2820 /* unhash both */
2821 if (!d_unhashed(dentry)) 2821 if (!d_unhashed(dentry))
2822 ___d_drop(dentry); 2822 ___d_drop(dentry);
2823 if (!d_unhashed(target)) 2823 if (!d_unhashed(target))
2824 ___d_drop(target); 2824 ___d_drop(target);
2825 2825
2826 /* ... and switch them in the tree */ 2826 /* ... and switch them in the tree */
2827 dentry->d_parent = target->d_parent; 2827 dentry->d_parent = target->d_parent;
2828 if (!exchange) { 2828 if (!exchange) {
2829 copy_name(dentry, target); 2829 copy_name(dentry, target);
2830 target->d_hash.pprev = NULL; 2830 target->d_hash.pprev = NULL;
2831 dentry->d_parent->d_lockref.c 2831 dentry->d_parent->d_lockref.count++;
2832 if (dentry != old_parent) /* 2832 if (dentry != old_parent) /* wasn't IS_ROOT */
2833 WARN_ON(!--old_parent 2833 WARN_ON(!--old_parent->d_lockref.count);
2834 } else { 2834 } else {
2835 target->d_parent = old_parent 2835 target->d_parent = old_parent;
2836 swap_names(dentry, target); 2836 swap_names(dentry, target);
2837 if (!hlist_unhashed(&target-> 2837 if (!hlist_unhashed(&target->d_sib))
2838 __hlist_del(&target-> 2838 __hlist_del(&target->d_sib);
2839 hlist_add_head(&target->d_sib 2839 hlist_add_head(&target->d_sib, &target->d_parent->d_children);
2840 __d_rehash(target); 2840 __d_rehash(target);
2841 fsnotify_update_flags(target) 2841 fsnotify_update_flags(target);
2842 } 2842 }
2843 if (!hlist_unhashed(&dentry->d_sib)) 2843 if (!hlist_unhashed(&dentry->d_sib))
2844 __hlist_del(&dentry->d_sib); 2844 __hlist_del(&dentry->d_sib);
2845 hlist_add_head(&dentry->d_sib, &dentr 2845 hlist_add_head(&dentry->d_sib, &dentry->d_parent->d_children);
2846 __d_rehash(dentry); 2846 __d_rehash(dentry);
2847 fsnotify_update_flags(dentry); 2847 fsnotify_update_flags(dentry);
2848 fscrypt_handle_d_move(dentry); 2848 fscrypt_handle_d_move(dentry);
2849 2849
2850 write_seqcount_end(&target->d_seq); 2850 write_seqcount_end(&target->d_seq);
2851 write_seqcount_end(&dentry->d_seq); 2851 write_seqcount_end(&dentry->d_seq);
2852 2852
2853 if (dir) 2853 if (dir)
2854 end_dir_add(dir, n, d_wait); 2854 end_dir_add(dir, n, d_wait);
2855 2855
2856 if (dentry->d_parent != old_parent) 2856 if (dentry->d_parent != old_parent)
2857 spin_unlock(&dentry->d_parent 2857 spin_unlock(&dentry->d_parent->d_lock);
2858 if (dentry != old_parent) 2858 if (dentry != old_parent)
2859 spin_unlock(&old_parent->d_lo 2859 spin_unlock(&old_parent->d_lock);
2860 spin_unlock(&target->d_lock); 2860 spin_unlock(&target->d_lock);
2861 spin_unlock(&dentry->d_lock); 2861 spin_unlock(&dentry->d_lock);
2862 } 2862 }
2863 2863
2864 /* 2864 /*
2865 * d_move - move a dentry 2865 * d_move - move a dentry
2866 * @dentry: entry to move 2866 * @dentry: entry to move
2867 * @target: new dentry 2867 * @target: new dentry
2868 * 2868 *
2869 * Update the dcache to reflect the move of a 2869 * Update the dcache to reflect the move of a file name. Negative
2870 * dcache entries should not be moved in this 2870 * dcache entries should not be moved in this way. See the locking
2871 * requirements for __d_move. 2871 * requirements for __d_move.
2872 */ 2872 */
2873 void d_move(struct dentry *dentry, struct den 2873 void d_move(struct dentry *dentry, struct dentry *target)
2874 { 2874 {
2875 write_seqlock(&rename_lock); 2875 write_seqlock(&rename_lock);
2876 __d_move(dentry, target, false); 2876 __d_move(dentry, target, false);
2877 write_sequnlock(&rename_lock); 2877 write_sequnlock(&rename_lock);
2878 } 2878 }
2879 EXPORT_SYMBOL(d_move); 2879 EXPORT_SYMBOL(d_move);
2880 2880
2881 /* 2881 /*
2882 * d_exchange - exchange two dentries 2882 * d_exchange - exchange two dentries
2883 * @dentry1: first dentry 2883 * @dentry1: first dentry
2884 * @dentry2: second dentry 2884 * @dentry2: second dentry
2885 */ 2885 */
2886 void d_exchange(struct dentry *dentry1, struc 2886 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2887 { 2887 {
2888 write_seqlock(&rename_lock); 2888 write_seqlock(&rename_lock);
2889 2889
2890 WARN_ON(!dentry1->d_inode); 2890 WARN_ON(!dentry1->d_inode);
2891 WARN_ON(!dentry2->d_inode); 2891 WARN_ON(!dentry2->d_inode);
2892 WARN_ON(IS_ROOT(dentry1)); 2892 WARN_ON(IS_ROOT(dentry1));
2893 WARN_ON(IS_ROOT(dentry2)); 2893 WARN_ON(IS_ROOT(dentry2));
2894 2894
2895 __d_move(dentry1, dentry2, true); 2895 __d_move(dentry1, dentry2, true);
2896 2896
2897 write_sequnlock(&rename_lock); 2897 write_sequnlock(&rename_lock);
2898 } 2898 }
2899 2899
2900 /** 2900 /**
2901 * d_ancestor - search for an ancestor 2901 * d_ancestor - search for an ancestor
2902 * @p1: ancestor dentry 2902 * @p1: ancestor dentry
2903 * @p2: child dentry 2903 * @p2: child dentry
2904 * 2904 *
2905 * Returns the ancestor dentry of p2 which is 2905 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2906 * an ancestor of p2, else NULL. 2906 * an ancestor of p2, else NULL.
2907 */ 2907 */
2908 struct dentry *d_ancestor(struct dentry *p1, 2908 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2909 { 2909 {
2910 struct dentry *p; 2910 struct dentry *p;
2911 2911
2912 for (p = p2; !IS_ROOT(p); p = p->d_pa 2912 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2913 if (p->d_parent == p1) 2913 if (p->d_parent == p1)
2914 return p; 2914 return p;
2915 } 2915 }
2916 return NULL; 2916 return NULL;
2917 } 2917 }
2918 2918
2919 /* 2919 /*
2920 * This helper attempts to cope with remotely 2920 * This helper attempts to cope with remotely renamed directories
2921 * 2921 *
2922 * It assumes that the caller is already hold 2922 * It assumes that the caller is already holding
2923 * dentry->d_parent->d_inode->i_mutex, and re 2923 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2924 * 2924 *
2925 * Note: If ever the locking in lock_rename() 2925 * Note: If ever the locking in lock_rename() changes, then please
2926 * remember to update this too... 2926 * remember to update this too...
2927 */ 2927 */
2928 static int __d_unalias(struct dentry *dentry, 2928 static int __d_unalias(struct dentry *dentry, struct dentry *alias)
2929 { 2929 {
2930 struct mutex *m1 = NULL; 2930 struct mutex *m1 = NULL;
2931 struct rw_semaphore *m2 = NULL; 2931 struct rw_semaphore *m2 = NULL;
2932 int ret = -ESTALE; 2932 int ret = -ESTALE;
2933 2933
2934 /* If alias and dentry share a parent 2934 /* If alias and dentry share a parent, then no extra locks required */
2935 if (alias->d_parent == dentry->d_pare 2935 if (alias->d_parent == dentry->d_parent)
2936 goto out_unalias; 2936 goto out_unalias;
2937 2937
2938 /* See lock_rename() */ 2938 /* See lock_rename() */
2939 if (!mutex_trylock(&dentry->d_sb->s_v 2939 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2940 goto out_err; 2940 goto out_err;
2941 m1 = &dentry->d_sb->s_vfs_rename_mute 2941 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2942 if (!inode_trylock_shared(alias->d_pa 2942 if (!inode_trylock_shared(alias->d_parent->d_inode))
2943 goto out_err; 2943 goto out_err;
2944 m2 = &alias->d_parent->d_inode->i_rws 2944 m2 = &alias->d_parent->d_inode->i_rwsem;
2945 out_unalias: 2945 out_unalias:
2946 __d_move(alias, dentry, false); 2946 __d_move(alias, dentry, false);
2947 ret = 0; 2947 ret = 0;
2948 out_err: 2948 out_err:
2949 if (m2) 2949 if (m2)
2950 up_read(m2); 2950 up_read(m2);
2951 if (m1) 2951 if (m1)
2952 mutex_unlock(m1); 2952 mutex_unlock(m1);
2953 return ret; 2953 return ret;
2954 } 2954 }
2955 2955
2956 /** 2956 /**
2957 * d_splice_alias - splice a disconnected den 2957 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2958 * @inode: the inode which may have a discon 2958 * @inode: the inode which may have a disconnected dentry
2959 * @dentry: a negative dentry which we want t 2959 * @dentry: a negative dentry which we want to point to the inode.
2960 * 2960 *
2961 * If inode is a directory and has an IS_ROOT 2961 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2962 * place of the given dentry and return it, e 2962 * place of the given dentry and return it, else simply d_add the inode
2963 * to the dentry and return NULL. 2963 * to the dentry and return NULL.
2964 * 2964 *
2965 * If a non-IS_ROOT directory is found, the f 2965 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2966 * we should error out: directories can't hav 2966 * we should error out: directories can't have multiple aliases.
2967 * 2967 *
2968 * This is needed in the lookup routine of an 2968 * This is needed in the lookup routine of any filesystem that is exportable
2969 * (via knfsd) so that we can build dcache pa 2969 * (via knfsd) so that we can build dcache paths to directories effectively.
2970 * 2970 *
2971 * If a dentry was found and moved, then it i 2971 * If a dentry was found and moved, then it is returned. Otherwise NULL
2972 * is returned. This matches the expected re 2972 * is returned. This matches the expected return value of ->lookup.
2973 * 2973 *
2974 * Cluster filesystems may call this function 2974 * Cluster filesystems may call this function with a negative, hashed dentry.
2975 * In that case, we know that the inode will 2975 * In that case, we know that the inode will be a regular file, and also this
2976 * will only occur during atomic_open. So we 2976 * will only occur during atomic_open. So we need to check for the dentry
2977 * being already hashed only in the final cas 2977 * being already hashed only in the final case.
2978 */ 2978 */
2979 struct dentry *d_splice_alias(struct inode *i 2979 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2980 { 2980 {
2981 if (IS_ERR(inode)) 2981 if (IS_ERR(inode))
2982 return ERR_CAST(inode); 2982 return ERR_CAST(inode);
2983 2983
2984 BUG_ON(!d_unhashed(dentry)); 2984 BUG_ON(!d_unhashed(dentry));
2985 2985
2986 if (!inode) 2986 if (!inode)
2987 goto out; 2987 goto out;
2988 2988
2989 security_d_instantiate(dentry, inode) 2989 security_d_instantiate(dentry, inode);
2990 spin_lock(&inode->i_lock); 2990 spin_lock(&inode->i_lock);
2991 if (S_ISDIR(inode->i_mode)) { 2991 if (S_ISDIR(inode->i_mode)) {
2992 struct dentry *new = __d_find 2992 struct dentry *new = __d_find_any_alias(inode);
2993 if (unlikely(new)) { 2993 if (unlikely(new)) {
2994 /* The reference to n 2994 /* The reference to new ensures it remains an alias */
2995 spin_unlock(&inode->i 2995 spin_unlock(&inode->i_lock);
2996 write_seqlock(&rename 2996 write_seqlock(&rename_lock);
2997 if (unlikely(d_ancest 2997 if (unlikely(d_ancestor(new, dentry))) {
2998 write_sequnlo 2998 write_sequnlock(&rename_lock);
2999 dput(new); 2999 dput(new);
3000 new = ERR_PTR 3000 new = ERR_PTR(-ELOOP);
3001 pr_warn_ratel 3001 pr_warn_ratelimited(
3002 "VFS: 3002 "VFS: Lookup of '%s' in %s %s"
3003 " wou 3003 " would have caused loop\n",
3004 dentr 3004 dentry->d_name.name,
3005 inode 3005 inode->i_sb->s_type->name,
3006 inode 3006 inode->i_sb->s_id);
3007 } else if (!IS_ROOT(n 3007 } else if (!IS_ROOT(new)) {
3008 struct dentry 3008 struct dentry *old_parent = dget(new->d_parent);
3009 int err = __d 3009 int err = __d_unalias(dentry, new);
3010 write_sequnlo 3010 write_sequnlock(&rename_lock);
3011 if (err) { 3011 if (err) {
3012 dput( 3012 dput(new);
3013 new = 3013 new = ERR_PTR(err);
3014 } 3014 }
3015 dput(old_pare 3015 dput(old_parent);
3016 } else { 3016 } else {
3017 __d_move(new, 3017 __d_move(new, dentry, false);
3018 write_sequnlo 3018 write_sequnlock(&rename_lock);
3019 } 3019 }
3020 iput(inode); 3020 iput(inode);
3021 return new; 3021 return new;
3022 } 3022 }
3023 } 3023 }
3024 out: 3024 out:
3025 __d_add(dentry, inode); 3025 __d_add(dentry, inode);
3026 return NULL; 3026 return NULL;
3027 } 3027 }
3028 EXPORT_SYMBOL(d_splice_alias); 3028 EXPORT_SYMBOL(d_splice_alias);
3029 3029
3030 /* 3030 /*
3031 * Test whether new_dentry is a subdirectory 3031 * Test whether new_dentry is a subdirectory of old_dentry.
3032 * 3032 *
3033 * Trivially implemented using the dcache str 3033 * Trivially implemented using the dcache structure
3034 */ 3034 */
3035 3035
3036 /** 3036 /**
3037 * is_subdir - is new dentry a subdirectory o 3037 * is_subdir - is new dentry a subdirectory of old_dentry
3038 * @new_dentry: new dentry 3038 * @new_dentry: new dentry
3039 * @old_dentry: old dentry 3039 * @old_dentry: old dentry
3040 * 3040 *
3041 * Returns true if new_dentry is a subdirecto 3041 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3042 * Returns false otherwise. 3042 * Returns false otherwise.
3043 * Caller must ensure that "new_dentry" is pi 3043 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3044 */ 3044 */
3045 3045
3046 bool is_subdir(struct dentry *new_dentry, str 3046 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3047 { 3047 {
3048 bool subdir; 3048 bool subdir;
3049 unsigned seq; 3049 unsigned seq;
3050 3050
3051 if (new_dentry == old_dentry) 3051 if (new_dentry == old_dentry)
3052 return true; 3052 return true;
3053 3053
3054 /* Access d_parent under rcu as d_mov 3054 /* Access d_parent under rcu as d_move() may change it. */
3055 rcu_read_lock(); 3055 rcu_read_lock();
3056 seq = read_seqbegin(&rename_lock); 3056 seq = read_seqbegin(&rename_lock);
3057 subdir = d_ancestor(old_dentry, new_d 3057 subdir = d_ancestor(old_dentry, new_dentry);
3058 /* Try lockless once... */ 3058 /* Try lockless once... */
3059 if (read_seqretry(&rename_lock, seq)) 3059 if (read_seqretry(&rename_lock, seq)) {
3060 /* ...else acquire lock for p 3060 /* ...else acquire lock for progress even on deep chains. */
3061 read_seqlock_excl(&rename_loc 3061 read_seqlock_excl(&rename_lock);
3062 subdir = d_ancestor(old_dentr 3062 subdir = d_ancestor(old_dentry, new_dentry);
3063 read_sequnlock_excl(&rename_l 3063 read_sequnlock_excl(&rename_lock);
3064 } 3064 }
3065 rcu_read_unlock(); 3065 rcu_read_unlock();
3066 return subdir; 3066 return subdir;
3067 } 3067 }
3068 EXPORT_SYMBOL(is_subdir); 3068 EXPORT_SYMBOL(is_subdir);
3069 3069
3070 static enum d_walk_ret d_genocide_kill(void * 3070 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3071 { 3071 {
3072 struct dentry *root = data; 3072 struct dentry *root = data;
3073 if (dentry != root) { 3073 if (dentry != root) {
3074 if (d_unhashed(dentry) || !de 3074 if (d_unhashed(dentry) || !dentry->d_inode)
3075 return D_WALK_SKIP; 3075 return D_WALK_SKIP;
3076 3076
3077 if (!(dentry->d_flags & DCACH 3077 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3078 dentry->d_flags |= DC 3078 dentry->d_flags |= DCACHE_GENOCIDE;
3079 dentry->d_lockref.cou 3079 dentry->d_lockref.count--;
3080 } 3080 }
3081 } 3081 }
3082 return D_WALK_CONTINUE; 3082 return D_WALK_CONTINUE;
3083 } 3083 }
3084 3084
3085 void d_genocide(struct dentry *parent) 3085 void d_genocide(struct dentry *parent)
3086 { 3086 {
3087 d_walk(parent, parent, d_genocide_kil 3087 d_walk(parent, parent, d_genocide_kill);
3088 } 3088 }
3089 3089
3090 void d_mark_tmpfile(struct file *file, struct 3090 void d_mark_tmpfile(struct file *file, struct inode *inode)
3091 { 3091 {
3092 struct dentry *dentry = file->f_path. 3092 struct dentry *dentry = file->f_path.dentry;
3093 3093
3094 BUG_ON(dentry->d_name.name != dentry- 3094 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3095 !hlist_unhashed(&dentry->d_u. 3095 !hlist_unhashed(&dentry->d_u.d_alias) ||
3096 !d_unlinked(dentry)); 3096 !d_unlinked(dentry));
3097 spin_lock(&dentry->d_parent->d_lock); 3097 spin_lock(&dentry->d_parent->d_lock);
3098 spin_lock_nested(&dentry->d_lock, DEN 3098 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3099 dentry->d_name.len = sprintf(dentry-> 3099 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3100 (unsigned lon 3100 (unsigned long long)inode->i_ino);
3101 spin_unlock(&dentry->d_lock); 3101 spin_unlock(&dentry->d_lock);
3102 spin_unlock(&dentry->d_parent->d_lock 3102 spin_unlock(&dentry->d_parent->d_lock);
3103 } 3103 }
3104 EXPORT_SYMBOL(d_mark_tmpfile); 3104 EXPORT_SYMBOL(d_mark_tmpfile);
3105 3105
3106 void d_tmpfile(struct file *file, struct inod 3106 void d_tmpfile(struct file *file, struct inode *inode)
3107 { 3107 {
3108 struct dentry *dentry = file->f_path. 3108 struct dentry *dentry = file->f_path.dentry;
3109 3109
3110 inode_dec_link_count(inode); 3110 inode_dec_link_count(inode);
3111 d_mark_tmpfile(file, inode); 3111 d_mark_tmpfile(file, inode);
3112 d_instantiate(dentry, inode); 3112 d_instantiate(dentry, inode);
3113 } 3113 }
3114 EXPORT_SYMBOL(d_tmpfile); 3114 EXPORT_SYMBOL(d_tmpfile);
3115 3115
3116 /* 3116 /*
3117 * Obtain inode number of the parent dentry. 3117 * Obtain inode number of the parent dentry.
3118 */ 3118 */
3119 ino_t d_parent_ino(struct dentry *dentry) 3119 ino_t d_parent_ino(struct dentry *dentry)
3120 { 3120 {
3121 struct dentry *parent; 3121 struct dentry *parent;
3122 struct inode *iparent; 3122 struct inode *iparent;
3123 unsigned seq; 3123 unsigned seq;
3124 ino_t ret; 3124 ino_t ret;
3125 3125
3126 scoped_guard(rcu) { 3126 scoped_guard(rcu) {
3127 seq = raw_seqcount_begin(&den 3127 seq = raw_seqcount_begin(&dentry->d_seq);
3128 parent = READ_ONCE(dentry->d_ 3128 parent = READ_ONCE(dentry->d_parent);
3129 iparent = d_inode_rcu(parent) 3129 iparent = d_inode_rcu(parent);
3130 if (likely(iparent)) { 3130 if (likely(iparent)) {
3131 ret = iparent->i_ino; 3131 ret = iparent->i_ino;
3132 if (!read_seqcount_re 3132 if (!read_seqcount_retry(&dentry->d_seq, seq))
3133 return ret; 3133 return ret;
3134 } 3134 }
3135 } 3135 }
3136 3136
3137 spin_lock(&dentry->d_lock); 3137 spin_lock(&dentry->d_lock);
3138 ret = dentry->d_parent->d_inode->i_in 3138 ret = dentry->d_parent->d_inode->i_ino;
3139 spin_unlock(&dentry->d_lock); 3139 spin_unlock(&dentry->d_lock);
3140 return ret; 3140 return ret;
3141 } 3141 }
3142 EXPORT_SYMBOL(d_parent_ino); 3142 EXPORT_SYMBOL(d_parent_ino);
3143 3143
3144 static __initdata unsigned long dhash_entries 3144 static __initdata unsigned long dhash_entries;
3145 static int __init set_dhash_entries(char *str 3145 static int __init set_dhash_entries(char *str)
3146 { 3146 {
3147 if (!str) 3147 if (!str)
3148 return 0; 3148 return 0;
3149 dhash_entries = simple_strtoul(str, & 3149 dhash_entries = simple_strtoul(str, &str, 0);
3150 return 1; 3150 return 1;
3151 } 3151 }
3152 __setup("dhash_entries=", set_dhash_entries); 3152 __setup("dhash_entries=", set_dhash_entries);
3153 3153
3154 static void __init dcache_init_early(void) 3154 static void __init dcache_init_early(void)
3155 { 3155 {
3156 /* If hashes are distributed across N 3156 /* If hashes are distributed across NUMA nodes, defer
3157 * hash allocation until vmalloc spac 3157 * hash allocation until vmalloc space is available.
3158 */ 3158 */
3159 if (hashdist) 3159 if (hashdist)
3160 return; 3160 return;
3161 3161
3162 dentry_hashtable = 3162 dentry_hashtable =
3163 alloc_large_system_hash("Dent 3163 alloc_large_system_hash("Dentry cache",
3164 sizeo 3164 sizeof(struct hlist_bl_head),
3165 dhash 3165 dhash_entries,
3166 13, 3166 13,
3167 HASH_ 3167 HASH_EARLY | HASH_ZERO,
3168 &d_ha 3168 &d_hash_shift,
3169 NULL, 3169 NULL,
3170 0, 3170 0,
3171 0); 3171 0);
3172 d_hash_shift = 32 - d_hash_shift; 3172 d_hash_shift = 32 - d_hash_shift;
3173 3173
3174 runtime_const_init(shift, d_hash_shif 3174 runtime_const_init(shift, d_hash_shift);
3175 runtime_const_init(ptr, dentry_hashta 3175 runtime_const_init(ptr, dentry_hashtable);
3176 } 3176 }
3177 3177
3178 static void __init dcache_init(void) 3178 static void __init dcache_init(void)
3179 { 3179 {
3180 /* 3180 /*
3181 * A constructor could be added for s 3181 * A constructor could be added for stable state like the lists,
3182 * but it is probably not worth it be 3182 * but it is probably not worth it because of the cache nature
3183 * of the dcache. 3183 * of the dcache.
3184 */ 3184 */
3185 dentry_cache = KMEM_CACHE_USERCOPY(de 3185 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3186 SLAB_RECLAIM_ACCOUNT|SLAB_PAN 3186 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_ACCOUNT,
3187 d_iname); 3187 d_iname);
3188 3188
3189 /* Hash may have been set up in dcach 3189 /* Hash may have been set up in dcache_init_early */
3190 if (!hashdist) 3190 if (!hashdist)
3191 return; 3191 return;
3192 3192
3193 dentry_hashtable = 3193 dentry_hashtable =
3194 alloc_large_system_hash("Dent 3194 alloc_large_system_hash("Dentry cache",
3195 sizeo 3195 sizeof(struct hlist_bl_head),
3196 dhash 3196 dhash_entries,
3197 13, 3197 13,
3198 HASH_ 3198 HASH_ZERO,
3199 &d_ha 3199 &d_hash_shift,
3200 NULL, 3200 NULL,
3201 0, 3201 0,
3202 0); 3202 0);
3203 d_hash_shift = 32 - d_hash_shift; 3203 d_hash_shift = 32 - d_hash_shift;
3204 3204
3205 runtime_const_init(shift, d_hash_shif 3205 runtime_const_init(shift, d_hash_shift);
3206 runtime_const_init(ptr, dentry_hashta 3206 runtime_const_init(ptr, dentry_hashtable);
3207 } 3207 }
3208 3208
3209 /* SLAB cache for __getname() consumers */ 3209 /* SLAB cache for __getname() consumers */
3210 struct kmem_cache *names_cachep __ro_after_in 3210 struct kmem_cache *names_cachep __ro_after_init;
3211 EXPORT_SYMBOL(names_cachep); 3211 EXPORT_SYMBOL(names_cachep);
3212 3212
3213 void __init vfs_caches_init_early(void) 3213 void __init vfs_caches_init_early(void)
3214 { 3214 {
3215 int i; 3215 int i;
3216 3216
3217 for (i = 0; i < ARRAY_SIZE(in_lookup_ 3217 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3218 INIT_HLIST_BL_HEAD(&in_lookup 3218 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3219 3219
3220 dcache_init_early(); 3220 dcache_init_early();
3221 inode_init_early(); 3221 inode_init_early();
3222 } 3222 }
3223 3223
3224 void __init vfs_caches_init(void) 3224 void __init vfs_caches_init(void)
3225 { 3225 {
3226 names_cachep = kmem_cache_create_user 3226 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3227 SLAB_HWCACHE_ALIGN|SL 3227 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3228 3228
3229 dcache_init(); 3229 dcache_init();
3230 inode_init(); 3230 inode_init();
3231 files_init(); 3231 files_init();
3232 files_maxfiles_init(); 3232 files_maxfiles_init();
3233 mnt_init(); 3233 mnt_init();
3234 bdev_cache_init(); 3234 bdev_cache_init();
3235 chrdev_init(); 3235 chrdev_init();
3236 } 3236 }
3237 3237
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.