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> <<
39 <<
40 /* 38 /*
41 * Usage: 39 * Usage:
42 * dcache->d_inode->i_lock protects: 40 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_u.d_alias, d_inode of alias 41 * - i_dentry, d_u.d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects: 42 * dcache_hash_bucket lock protects:
45 * - the dcache hash table 43 * - the dcache hash table
46 * s_roots bl list spinlock protects: 44 * s_roots bl list spinlock protects:
47 * - the s_roots list (see __d_drop) 45 * - the s_roots list (see __d_drop)
48 * dentry->d_sb->s_dentry_lru_lock protects: 46 * dentry->d_sb->s_dentry_lru_lock protects:
49 * - the dcache lru lists and counters 47 * - the dcache lru lists and counters
50 * d_lock protects: 48 * d_lock protects:
51 * - d_flags 49 * - d_flags
52 * - d_name 50 * - d_name
53 * - d_lru 51 * - d_lru
54 * - d_count 52 * - d_count
55 * - d_unhashed() 53 * - d_unhashed()
56 * - d_parent and d_chilren !! 54 * - d_parent and d_subdirs
57 * - childrens' d_sib and d_parent !! 55 * - childrens' d_child and d_parent
58 * - d_u.d_alias, d_inode 56 * - d_u.d_alias, d_inode
59 * 57 *
60 * Ordering: 58 * Ordering:
61 * dentry->d_inode->i_lock 59 * dentry->d_inode->i_lock
62 * dentry->d_lock 60 * dentry->d_lock
63 * dentry->d_sb->s_dentry_lru_lock 61 * dentry->d_sb->s_dentry_lru_lock
64 * dcache_hash_bucket lock 62 * dcache_hash_bucket lock
65 * s_roots lock 63 * s_roots lock
66 * 64 *
67 * If there is an ancestor relationship: 65 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock 66 * dentry->d_parent->...->d_parent->d_lock
69 * ... 67 * ...
70 * dentry->d_parent->d_lock 68 * dentry->d_parent->d_lock
71 * dentry->d_lock 69 * dentry->d_lock
72 * 70 *
73 * If no ancestor relationship: 71 * If no ancestor relationship:
74 * arbitrary, since it's serialized on rename_ 72 * arbitrary, since it's serialized on rename_lock
75 */ 73 */
76 int sysctl_vfs_cache_pressure __read_mostly = 74 int sysctl_vfs_cache_pressure __read_mostly = 100;
77 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); 75 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
78 76
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rena 77 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
80 78
81 EXPORT_SYMBOL(rename_lock); 79 EXPORT_SYMBOL(rename_lock);
82 80
83 static struct kmem_cache *dentry_cache __ro_af !! 81 static struct kmem_cache *dentry_cache __read_mostly;
84 82
85 const struct qstr empty_name = QSTR_INIT("", 0 83 const struct qstr empty_name = QSTR_INIT("", 0);
86 EXPORT_SYMBOL(empty_name); 84 EXPORT_SYMBOL(empty_name);
87 const struct qstr slash_name = QSTR_INIT("/", 85 const struct qstr slash_name = QSTR_INIT("/", 1);
88 EXPORT_SYMBOL(slash_name); 86 EXPORT_SYMBOL(slash_name);
89 const struct qstr dotdot_name = QSTR_INIT(".." 87 const struct qstr dotdot_name = QSTR_INIT("..", 2);
90 EXPORT_SYMBOL(dotdot_name); 88 EXPORT_SYMBOL(dotdot_name);
91 89
92 /* 90 /*
93 * This is the single most critical data struc 91 * This is the single most critical data structure when it comes
94 * to the dcache: the hashtable for lookups. S 92 * to the dcache: the hashtable for lookups. Somebody should try
95 * to make this good - I've just made it work. 93 * to make this good - I've just made it work.
96 * 94 *
97 * This hash-function tries to avoid losing to 95 * This hash-function tries to avoid losing too many bits of hash
98 * information, yet avoid using a prime hash-s 96 * information, yet avoid using a prime hash-size or similar.
99 * <<
100 * Marking the variables "used" ensures that t <<
101 * optimize them away completely on architectu <<
102 * constant infrastructure, this allows debugg <<
103 * values. But updating these values has no ef <<
104 */ 97 */
105 98
106 static unsigned int d_hash_shift __ro_after_in !! 99 static unsigned int d_hash_shift __read_mostly;
107 100
108 static struct hlist_bl_head *dentry_hashtable !! 101 static struct hlist_bl_head *dentry_hashtable __read_mostly;
109 102
110 static inline struct hlist_bl_head *d_hash(uns !! 103 static inline struct hlist_bl_head *d_hash(unsigned int hash)
111 { 104 {
112 return runtime_const_ptr(dentry_hashta !! 105 return dentry_hashtable + (hash >> d_hash_shift);
113 runtime_const_shift_right_32(h <<
114 } 106 }
115 107
116 #define IN_LOOKUP_SHIFT 10 108 #define IN_LOOKUP_SHIFT 10
117 static struct hlist_bl_head in_lookup_hashtabl 109 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
118 110
119 static inline struct hlist_bl_head *in_lookup_ 111 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
120 unsign 112 unsigned int hash)
121 { 113 {
122 hash += (unsigned long) parent / L1_CA 114 hash += (unsigned long) parent / L1_CACHE_BYTES;
123 return in_lookup_hashtable + hash_32(h 115 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
124 } 116 }
125 117
126 struct dentry_stat_t { 118 struct dentry_stat_t {
127 long nr_dentry; 119 long nr_dentry;
128 long nr_unused; 120 long nr_unused;
129 long age_limit; /* age in seco 121 long age_limit; /* age in seconds */
130 long want_pages; /* pages reque 122 long want_pages; /* pages requested by system */
131 long nr_negative; /* # of unused 123 long nr_negative; /* # of unused negative dentries */
132 long dummy; /* Reserved fo 124 long dummy; /* Reserved for future use */
133 }; 125 };
134 126
135 static DEFINE_PER_CPU(long, nr_dentry); 127 static DEFINE_PER_CPU(long, nr_dentry);
136 static DEFINE_PER_CPU(long, nr_dentry_unused); 128 static DEFINE_PER_CPU(long, nr_dentry_unused);
137 static DEFINE_PER_CPU(long, nr_dentry_negative 129 static DEFINE_PER_CPU(long, nr_dentry_negative);
138 130
139 #if defined(CONFIG_SYSCTL) && defined(CONFIG_P 131 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
140 /* Statistics gathering. */ 132 /* Statistics gathering. */
141 static struct dentry_stat_t dentry_stat = { 133 static struct dentry_stat_t dentry_stat = {
142 .age_limit = 45, 134 .age_limit = 45,
143 }; 135 };
144 136
145 /* 137 /*
146 * Here we resort to our own counters instead 138 * Here we resort to our own counters instead of using generic per-cpu counters
147 * for consistency with what the vfs inode cod 139 * for consistency with what the vfs inode code does. We are expected to harvest
148 * better code and performance by having our o 140 * better code and performance by having our own specialized counters.
149 * 141 *
150 * Please note that the loop is done over all 142 * 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 143 * 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 144 * on and off. If one of them goes off, we will just keep their counters.
153 * 145 *
154 * glommer: See cffbc8a for details, and if yo 146 * glommer: See cffbc8a for details, and if you ever intend to change this,
155 * please update all vfs counters to match. 147 * please update all vfs counters to match.
156 */ 148 */
157 static long get_nr_dentry(void) 149 static long get_nr_dentry(void)
158 { 150 {
159 int i; 151 int i;
160 long sum = 0; 152 long sum = 0;
161 for_each_possible_cpu(i) 153 for_each_possible_cpu(i)
162 sum += per_cpu(nr_dentry, i); 154 sum += per_cpu(nr_dentry, i);
163 return sum < 0 ? 0 : sum; 155 return sum < 0 ? 0 : sum;
164 } 156 }
165 157
166 static long get_nr_dentry_unused(void) 158 static long get_nr_dentry_unused(void)
167 { 159 {
168 int i; 160 int i;
169 long sum = 0; 161 long sum = 0;
170 for_each_possible_cpu(i) 162 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry_unuse 163 sum += per_cpu(nr_dentry_unused, i);
172 return sum < 0 ? 0 : sum; 164 return sum < 0 ? 0 : sum;
173 } 165 }
174 166
175 static long get_nr_dentry_negative(void) 167 static long get_nr_dentry_negative(void)
176 { 168 {
177 int i; 169 int i;
178 long sum = 0; 170 long sum = 0;
179 171
180 for_each_possible_cpu(i) 172 for_each_possible_cpu(i)
181 sum += per_cpu(nr_dentry_negat 173 sum += per_cpu(nr_dentry_negative, i);
182 return sum < 0 ? 0 : sum; 174 return sum < 0 ? 0 : sum;
183 } 175 }
184 176
185 static int proc_nr_dentry(const struct ctl_tab !! 177 static int proc_nr_dentry(struct ctl_table *table, int write, void *buffer,
186 size_t *lenp, loff_t 178 size_t *lenp, loff_t *ppos)
187 { 179 {
188 dentry_stat.nr_dentry = get_nr_dentry( 180 dentry_stat.nr_dentry = get_nr_dentry();
189 dentry_stat.nr_unused = get_nr_dentry_ 181 dentry_stat.nr_unused = get_nr_dentry_unused();
190 dentry_stat.nr_negative = get_nr_dentr 182 dentry_stat.nr_negative = get_nr_dentry_negative();
191 return proc_doulongvec_minmax(table, w 183 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
192 } 184 }
193 185
194 static struct ctl_table fs_dcache_sysctls[] = 186 static struct ctl_table fs_dcache_sysctls[] = {
195 { 187 {
196 .procname = "dentry-stat 188 .procname = "dentry-state",
197 .data = &dentry_stat 189 .data = &dentry_stat,
198 .maxlen = 6*sizeof(lon 190 .maxlen = 6*sizeof(long),
199 .mode = 0444, 191 .mode = 0444,
200 .proc_handler = proc_nr_dent 192 .proc_handler = proc_nr_dentry,
201 }, 193 },
>> 194 { }
202 }; 195 };
203 196
204 static int __init init_fs_dcache_sysctls(void) 197 static int __init init_fs_dcache_sysctls(void)
205 { 198 {
206 register_sysctl_init("fs", fs_dcache_s 199 register_sysctl_init("fs", fs_dcache_sysctls);
207 return 0; 200 return 0;
208 } 201 }
209 fs_initcall(init_fs_dcache_sysctls); 202 fs_initcall(init_fs_dcache_sysctls);
210 #endif 203 #endif
211 204
212 /* 205 /*
213 * Compare 2 name strings, return 0 if they ma 206 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
214 * The strings are both count bytes long, and 207 * The strings are both count bytes long, and count is non-zero.
215 */ 208 */
216 #ifdef CONFIG_DCACHE_WORD_ACCESS 209 #ifdef CONFIG_DCACHE_WORD_ACCESS
217 210
218 #include <asm/word-at-a-time.h> 211 #include <asm/word-at-a-time.h>
219 /* 212 /*
220 * NOTE! 'cs' and 'scount' come from a dentry, 213 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
221 * aligned allocation for this particular comp 214 * aligned allocation for this particular component. We don't
222 * strictly need the load_unaligned_zeropad() 215 * strictly need the load_unaligned_zeropad() safety, but it
223 * doesn't hurt either. 216 * doesn't hurt either.
224 * 217 *
225 * In contrast, 'ct' and 'tcount' can be from 218 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
226 * need the careful unaligned handling. 219 * need the careful unaligned handling.
227 */ 220 */
228 static inline int dentry_string_cmp(const unsi 221 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
229 { 222 {
230 unsigned long a,b,mask; 223 unsigned long a,b,mask;
231 224
232 for (;;) { 225 for (;;) {
233 a = read_word_at_a_time(cs); 226 a = read_word_at_a_time(cs);
234 b = load_unaligned_zeropad(ct) 227 b = load_unaligned_zeropad(ct);
235 if (tcount < sizeof(unsigned l 228 if (tcount < sizeof(unsigned long))
236 break; 229 break;
237 if (unlikely(a != b)) 230 if (unlikely(a != b))
238 return 1; 231 return 1;
239 cs += sizeof(unsigned long); 232 cs += sizeof(unsigned long);
240 ct += sizeof(unsigned long); 233 ct += sizeof(unsigned long);
241 tcount -= sizeof(unsigned long 234 tcount -= sizeof(unsigned long);
242 if (!tcount) 235 if (!tcount)
243 return 0; 236 return 0;
244 } 237 }
245 mask = bytemask_from_count(tcount); 238 mask = bytemask_from_count(tcount);
246 return unlikely(!!((a ^ b) & mask)); 239 return unlikely(!!((a ^ b) & mask));
247 } 240 }
248 241
249 #else 242 #else
250 243
251 static inline int dentry_string_cmp(const unsi 244 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
252 { 245 {
253 do { 246 do {
254 if (*cs != *ct) 247 if (*cs != *ct)
255 return 1; 248 return 1;
256 cs++; 249 cs++;
257 ct++; 250 ct++;
258 tcount--; 251 tcount--;
259 } while (tcount); 252 } while (tcount);
260 return 0; 253 return 0;
261 } 254 }
262 255
263 #endif 256 #endif
264 257
265 static inline int dentry_cmp(const struct dent 258 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
266 { 259 {
267 /* 260 /*
268 * Be careful about RCU walk racing wi 261 * Be careful about RCU walk racing with rename:
269 * use 'READ_ONCE' to fetch the name p 262 * use 'READ_ONCE' to fetch the name pointer.
270 * 263 *
271 * NOTE! Even if a rename will mean th 264 * NOTE! Even if a rename will mean that the length
272 * was not loaded atomically, we don't 265 * was not loaded atomically, we don't care. The
273 * RCU walk will check the sequence co 266 * RCU walk will check the sequence count eventually,
274 * and catch it. And we won't overrun 267 * and catch it. And we won't overrun the buffer,
275 * because we're reading the name poin 268 * because we're reading the name pointer atomically,
276 * and a dentry name is guaranteed to 269 * and a dentry name is guaranteed to be properly
277 * terminated with a NUL byte. 270 * terminated with a NUL byte.
278 * 271 *
279 * End result: even if 'len' is wrong, 272 * End result: even if 'len' is wrong, we'll exit
280 * early because the data cannot match 273 * early because the data cannot match (there can
281 * be no NUL in the ct/tcount data) 274 * be no NUL in the ct/tcount data)
282 */ 275 */
283 const unsigned char *cs = READ_ONCE(de 276 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
284 277
285 return dentry_string_cmp(cs, ct, tcoun 278 return dentry_string_cmp(cs, ct, tcount);
286 } 279 }
287 280
288 struct external_name { 281 struct external_name {
289 union { 282 union {
290 atomic_t count; 283 atomic_t count;
291 struct rcu_head head; 284 struct rcu_head head;
292 } u; 285 } u;
293 unsigned char name[]; 286 unsigned char name[];
294 }; 287 };
295 288
296 static inline struct external_name *external_n 289 static inline struct external_name *external_name(struct dentry *dentry)
297 { 290 {
298 return container_of(dentry->d_name.nam 291 return container_of(dentry->d_name.name, struct external_name, name[0]);
299 } 292 }
300 293
301 static void __d_free(struct rcu_head *head) 294 static void __d_free(struct rcu_head *head)
302 { 295 {
303 struct dentry *dentry = container_of(h 296 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
304 297
305 kmem_cache_free(dentry_cache, dentry); 298 kmem_cache_free(dentry_cache, dentry);
306 } 299 }
307 300
308 static void __d_free_external(struct rcu_head 301 static void __d_free_external(struct rcu_head *head)
309 { 302 {
310 struct dentry *dentry = container_of(h 303 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
311 kfree(external_name(dentry)); 304 kfree(external_name(dentry));
312 kmem_cache_free(dentry_cache, dentry); 305 kmem_cache_free(dentry_cache, dentry);
313 } 306 }
314 307
315 static inline int dname_external(const struct 308 static inline int dname_external(const struct dentry *dentry)
316 { 309 {
317 return dentry->d_name.name != dentry-> 310 return dentry->d_name.name != dentry->d_iname;
318 } 311 }
319 312
320 void take_dentry_name_snapshot(struct name_sna 313 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
321 { 314 {
322 spin_lock(&dentry->d_lock); 315 spin_lock(&dentry->d_lock);
323 name->name = dentry->d_name; 316 name->name = dentry->d_name;
324 if (unlikely(dname_external(dentry))) 317 if (unlikely(dname_external(dentry))) {
325 atomic_inc(&external_name(dent 318 atomic_inc(&external_name(dentry)->u.count);
326 } else { 319 } else {
327 memcpy(name->inline_name, dent 320 memcpy(name->inline_name, dentry->d_iname,
328 dentry->d_name.len + 1) 321 dentry->d_name.len + 1);
329 name->name.name = name->inline 322 name->name.name = name->inline_name;
330 } 323 }
331 spin_unlock(&dentry->d_lock); 324 spin_unlock(&dentry->d_lock);
332 } 325 }
333 EXPORT_SYMBOL(take_dentry_name_snapshot); 326 EXPORT_SYMBOL(take_dentry_name_snapshot);
334 327
335 void release_dentry_name_snapshot(struct name_ 328 void release_dentry_name_snapshot(struct name_snapshot *name)
336 { 329 {
337 if (unlikely(name->name.name != name-> 330 if (unlikely(name->name.name != name->inline_name)) {
338 struct external_name *p; 331 struct external_name *p;
339 p = container_of(name->name.na 332 p = container_of(name->name.name, struct external_name, name[0]);
340 if (unlikely(atomic_dec_and_te 333 if (unlikely(atomic_dec_and_test(&p->u.count)))
341 kfree_rcu(p, u.head); 334 kfree_rcu(p, u.head);
342 } 335 }
343 } 336 }
344 EXPORT_SYMBOL(release_dentry_name_snapshot); 337 EXPORT_SYMBOL(release_dentry_name_snapshot);
345 338
346 static inline void __d_set_inode_and_type(stru 339 static inline void __d_set_inode_and_type(struct dentry *dentry,
347 stru 340 struct inode *inode,
348 unsi 341 unsigned type_flags)
349 { 342 {
350 unsigned flags; 343 unsigned flags;
351 344
352 dentry->d_inode = inode; 345 dentry->d_inode = inode;
353 flags = READ_ONCE(dentry->d_flags); 346 flags = READ_ONCE(dentry->d_flags);
354 flags &= ~DCACHE_ENTRY_TYPE; !! 347 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
355 flags |= type_flags; 348 flags |= type_flags;
356 smp_store_release(&dentry->d_flags, fl 349 smp_store_release(&dentry->d_flags, flags);
357 } 350 }
358 351
359 static inline void __d_clear_type_and_inode(st 352 static inline void __d_clear_type_and_inode(struct dentry *dentry)
360 { 353 {
361 unsigned flags = READ_ONCE(dentry->d_f 354 unsigned flags = READ_ONCE(dentry->d_flags);
362 355
363 flags &= ~DCACHE_ENTRY_TYPE; !! 356 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
364 WRITE_ONCE(dentry->d_flags, flags); 357 WRITE_ONCE(dentry->d_flags, flags);
365 dentry->d_inode = NULL; 358 dentry->d_inode = NULL;
366 /* 359 /*
367 * The negative counter only tracks de 360 * The negative counter only tracks dentries on the LRU. Don't inc if
368 * d_lru is on another list. 361 * d_lru is on another list.
369 */ 362 */
370 if ((flags & (DCACHE_LRU_LIST|DCACHE_S 363 if ((flags & (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
371 this_cpu_inc(nr_dentry_negativ 364 this_cpu_inc(nr_dentry_negative);
372 } 365 }
373 366
374 static void dentry_free(struct dentry *dentry) 367 static void dentry_free(struct dentry *dentry)
375 { 368 {
376 WARN_ON(!hlist_unhashed(&dentry->d_u.d 369 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
377 if (unlikely(dname_external(dentry))) 370 if (unlikely(dname_external(dentry))) {
378 struct external_name *p = exte 371 struct external_name *p = external_name(dentry);
379 if (likely(atomic_dec_and_test 372 if (likely(atomic_dec_and_test(&p->u.count))) {
380 call_rcu(&dentry->d_u. 373 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
381 return; 374 return;
382 } 375 }
383 } 376 }
384 /* if dentry was never visible to RCU, 377 /* if dentry was never visible to RCU, immediate free is OK */
385 if (dentry->d_flags & DCACHE_NORCU) 378 if (dentry->d_flags & DCACHE_NORCU)
386 __d_free(&dentry->d_u.d_rcu); 379 __d_free(&dentry->d_u.d_rcu);
387 else 380 else
388 call_rcu(&dentry->d_u.d_rcu, _ 381 call_rcu(&dentry->d_u.d_rcu, __d_free);
389 } 382 }
390 383
391 /* 384 /*
392 * Release the dentry's inode, using the files 385 * Release the dentry's inode, using the filesystem
393 * d_iput() operation if defined. 386 * d_iput() operation if defined.
394 */ 387 */
395 static void dentry_unlink_inode(struct dentry 388 static void dentry_unlink_inode(struct dentry * dentry)
396 __releases(dentry->d_lock) 389 __releases(dentry->d_lock)
397 __releases(dentry->d_inode->i_lock) 390 __releases(dentry->d_inode->i_lock)
398 { 391 {
399 struct inode *inode = dentry->d_inode; 392 struct inode *inode = dentry->d_inode;
400 393
401 raw_write_seqcount_begin(&dentry->d_se 394 raw_write_seqcount_begin(&dentry->d_seq);
402 __d_clear_type_and_inode(dentry); 395 __d_clear_type_and_inode(dentry);
403 hlist_del_init(&dentry->d_u.d_alias); 396 hlist_del_init(&dentry->d_u.d_alias);
404 raw_write_seqcount_end(&dentry->d_seq) 397 raw_write_seqcount_end(&dentry->d_seq);
405 spin_unlock(&dentry->d_lock); 398 spin_unlock(&dentry->d_lock);
406 spin_unlock(&inode->i_lock); 399 spin_unlock(&inode->i_lock);
407 if (!inode->i_nlink) 400 if (!inode->i_nlink)
408 fsnotify_inoderemove(inode); 401 fsnotify_inoderemove(inode);
409 if (dentry->d_op && dentry->d_op->d_ip 402 if (dentry->d_op && dentry->d_op->d_iput)
410 dentry->d_op->d_iput(dentry, i 403 dentry->d_op->d_iput(dentry, inode);
411 else 404 else
412 iput(inode); 405 iput(inode);
413 } 406 }
414 407
415 /* 408 /*
416 * The DCACHE_LRU_LIST bit is set whenever the 409 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
417 * is in use - which includes both the "real" 410 * is in use - which includes both the "real" per-superblock
418 * LRU list _and_ the DCACHE_SHRINK_LIST use. 411 * LRU list _and_ the DCACHE_SHRINK_LIST use.
419 * 412 *
420 * The DCACHE_SHRINK_LIST bit is set whenever 413 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
421 * on the shrink list (ie not on the superbloc 414 * on the shrink list (ie not on the superblock LRU list).
422 * 415 *
423 * The per-cpu "nr_dentry_unused" counters are 416 * The per-cpu "nr_dentry_unused" counters are updated with
424 * the DCACHE_LRU_LIST bit. 417 * the DCACHE_LRU_LIST bit.
425 * 418 *
426 * The per-cpu "nr_dentry_negative" counters a 419 * The per-cpu "nr_dentry_negative" counters are only updated
427 * when deleted from or added to the per-super 420 * when deleted from or added to the per-superblock LRU list, not
428 * from/to the shrink list. That is to avoid a 421 * from/to the shrink list. That is to avoid an unneeded dec/inc
429 * pair when moving from LRU to shrink list in 422 * pair when moving from LRU to shrink list in select_collect().
430 * 423 *
431 * These helper functions make sure we always 424 * These helper functions make sure we always follow the
432 * rules. d_lock must be held by the caller. 425 * rules. d_lock must be held by the caller.
433 */ 426 */
434 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(( 427 #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) 428 static void d_lru_add(struct dentry *dentry)
436 { 429 {
437 D_FLAG_VERIFY(dentry, 0); 430 D_FLAG_VERIFY(dentry, 0);
438 dentry->d_flags |= DCACHE_LRU_LIST; 431 dentry->d_flags |= DCACHE_LRU_LIST;
439 this_cpu_inc(nr_dentry_unused); 432 this_cpu_inc(nr_dentry_unused);
440 if (d_is_negative(dentry)) 433 if (d_is_negative(dentry))
441 this_cpu_inc(nr_dentry_negativ 434 this_cpu_inc(nr_dentry_negative);
442 WARN_ON_ONCE(!list_lru_add_obj( !! 435 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
443 &dentry->d_sb->s_dentr <<
444 } 436 }
445 437
446 static void d_lru_del(struct dentry *dentry) 438 static void d_lru_del(struct dentry *dentry)
447 { 439 {
448 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 440 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
449 dentry->d_flags &= ~DCACHE_LRU_LIST; 441 dentry->d_flags &= ~DCACHE_LRU_LIST;
450 this_cpu_dec(nr_dentry_unused); 442 this_cpu_dec(nr_dentry_unused);
451 if (d_is_negative(dentry)) 443 if (d_is_negative(dentry))
452 this_cpu_dec(nr_dentry_negativ 444 this_cpu_dec(nr_dentry_negative);
453 WARN_ON_ONCE(!list_lru_del_obj( !! 445 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
454 &dentry->d_sb->s_dentr <<
455 } 446 }
456 447
457 static void d_shrink_del(struct dentry *dentry 448 static void d_shrink_del(struct dentry *dentry)
458 { 449 {
459 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LI 450 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
460 list_del_init(&dentry->d_lru); 451 list_del_init(&dentry->d_lru);
461 dentry->d_flags &= ~(DCACHE_SHRINK_LIS 452 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
462 this_cpu_dec(nr_dentry_unused); 453 this_cpu_dec(nr_dentry_unused);
463 } 454 }
464 455
465 static void d_shrink_add(struct dentry *dentry 456 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
466 { 457 {
467 D_FLAG_VERIFY(dentry, 0); 458 D_FLAG_VERIFY(dentry, 0);
468 list_add(&dentry->d_lru, list); 459 list_add(&dentry->d_lru, list);
469 dentry->d_flags |= DCACHE_SHRINK_LIST 460 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
470 this_cpu_inc(nr_dentry_unused); 461 this_cpu_inc(nr_dentry_unused);
471 } 462 }
472 463
473 /* 464 /*
474 * These can only be called under the global L 465 * These can only be called under the global LRU lock, ie during the
475 * callback for freeing the LRU list. "isolate 466 * callback for freeing the LRU list. "isolate" removes it from the
476 * LRU lists entirely, while shrink_move moves 467 * LRU lists entirely, while shrink_move moves it to the indicated
477 * private list. 468 * private list.
478 */ 469 */
479 static void d_lru_isolate(struct list_lru_one 470 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
480 { 471 {
481 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 472 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
482 dentry->d_flags &= ~DCACHE_LRU_LIST; 473 dentry->d_flags &= ~DCACHE_LRU_LIST;
483 this_cpu_dec(nr_dentry_unused); 474 this_cpu_dec(nr_dentry_unused);
484 if (d_is_negative(dentry)) 475 if (d_is_negative(dentry))
485 this_cpu_dec(nr_dentry_negativ 476 this_cpu_dec(nr_dentry_negative);
486 list_lru_isolate(lru, &dentry->d_lru); 477 list_lru_isolate(lru, &dentry->d_lru);
487 } 478 }
488 479
489 static void d_lru_shrink_move(struct list_lru_ 480 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
490 struct list_head 481 struct list_head *list)
491 { 482 {
492 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 483 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
493 dentry->d_flags |= DCACHE_SHRINK_LIST; 484 dentry->d_flags |= DCACHE_SHRINK_LIST;
494 if (d_is_negative(dentry)) 485 if (d_is_negative(dentry))
495 this_cpu_dec(nr_dentry_negativ 486 this_cpu_dec(nr_dentry_negative);
496 list_lru_isolate_move(lru, &dentry->d_ 487 list_lru_isolate_move(lru, &dentry->d_lru, list);
497 } 488 }
498 489
499 static void ___d_drop(struct dentry *dentry) 490 static void ___d_drop(struct dentry *dentry)
500 { 491 {
501 struct hlist_bl_head *b; 492 struct hlist_bl_head *b;
502 /* 493 /*
503 * Hashed dentries are normally on the 494 * Hashed dentries are normally on the dentry hashtable,
504 * with the exception of those newly a 495 * with the exception of those newly allocated by
505 * d_obtain_root, which are always IS_ 496 * d_obtain_root, which are always IS_ROOT:
506 */ 497 */
507 if (unlikely(IS_ROOT(dentry))) 498 if (unlikely(IS_ROOT(dentry)))
508 b = &dentry->d_sb->s_roots; 499 b = &dentry->d_sb->s_roots;
509 else 500 else
510 b = d_hash(dentry->d_name.hash 501 b = d_hash(dentry->d_name.hash);
511 502
512 hlist_bl_lock(b); 503 hlist_bl_lock(b);
513 __hlist_bl_del(&dentry->d_hash); 504 __hlist_bl_del(&dentry->d_hash);
514 hlist_bl_unlock(b); 505 hlist_bl_unlock(b);
515 } 506 }
516 507
517 void __d_drop(struct dentry *dentry) 508 void __d_drop(struct dentry *dentry)
518 { 509 {
519 if (!d_unhashed(dentry)) { 510 if (!d_unhashed(dentry)) {
520 ___d_drop(dentry); 511 ___d_drop(dentry);
521 dentry->d_hash.pprev = NULL; 512 dentry->d_hash.pprev = NULL;
522 write_seqcount_invalidate(&den 513 write_seqcount_invalidate(&dentry->d_seq);
523 } 514 }
524 } 515 }
525 EXPORT_SYMBOL(__d_drop); 516 EXPORT_SYMBOL(__d_drop);
526 517
527 /** 518 /**
528 * d_drop - drop a dentry 519 * d_drop - drop a dentry
529 * @dentry: dentry to drop 520 * @dentry: dentry to drop
530 * 521 *
531 * d_drop() unhashes the entry from the parent 522 * 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 523 * be found through a VFS lookup any more. Note that this is different from
533 * deleting the dentry - d_delete will try to 524 * deleting the dentry - d_delete will try to mark the dentry negative if
534 * possible, giving a successful _negative_ lo 525 * possible, giving a successful _negative_ lookup, while d_drop will
535 * just make the cache lookup fail. 526 * just make the cache lookup fail.
536 * 527 *
537 * d_drop() is used mainly for stuff that want 528 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
538 * reason (NFS timeouts or autofs deletes). 529 * reason (NFS timeouts or autofs deletes).
539 * 530 *
540 * __d_drop requires dentry->d_lock 531 * __d_drop requires dentry->d_lock
541 * 532 *
542 * ___d_drop doesn't mark dentry as "unhashed" 533 * ___d_drop doesn't mark dentry as "unhashed"
543 * (dentry->d_hash.pprev will be LIST_POISON2, 534 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
544 */ 535 */
545 void d_drop(struct dentry *dentry) 536 void d_drop(struct dentry *dentry)
546 { 537 {
547 spin_lock(&dentry->d_lock); 538 spin_lock(&dentry->d_lock);
548 __d_drop(dentry); 539 __d_drop(dentry);
549 spin_unlock(&dentry->d_lock); 540 spin_unlock(&dentry->d_lock);
550 } 541 }
551 EXPORT_SYMBOL(d_drop); 542 EXPORT_SYMBOL(d_drop);
552 543
553 static inline void dentry_unlist(struct dentry !! 544 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
554 { 545 {
555 struct dentry *next; 546 struct dentry *next;
556 /* 547 /*
557 * Inform d_walk() and shrink_dentry_l 548 * Inform d_walk() and shrink_dentry_list() that we are no longer
558 * attached to the dentry tree 549 * attached to the dentry tree
559 */ 550 */
560 dentry->d_flags |= DCACHE_DENTRY_KILLE 551 dentry->d_flags |= DCACHE_DENTRY_KILLED;
561 if (unlikely(hlist_unhashed(&dentry->d !! 552 if (unlikely(list_empty(&dentry->d_child)))
562 return; 553 return;
563 __hlist_del(&dentry->d_sib); !! 554 __list_del_entry(&dentry->d_child);
564 /* 555 /*
565 * Cursors can move around the list of 556 * Cursors can move around the list of children. While we'd been
566 * a normal list member, it didn't mat !! 557 * a normal list member, it didn't matter - ->d_child.next would've
567 * been updated. However, from now on 558 * been updated. However, from now on it won't be and for the
568 * things like d_walk() it might end u 559 * things like d_walk() it might end up with a nasty surprise.
569 * Normally d_walk() doesn't care abou 560 * Normally d_walk() doesn't care about cursors moving around -
570 * ->d_lock on parent prevents that an 561 * ->d_lock on parent prevents that and since a cursor has no children
571 * of its own, we get through it witho 562 * of its own, we get through it without ever unlocking the parent.
572 * There is one exception, though - if 563 * There is one exception, though - if we ascend from a child that
573 * gets killed as soon as we unlock it 564 * gets killed as soon as we unlock it, the next sibling is found
574 * using the value left in its ->d_sib !! 565 * using the value left in its ->d_child.next. And if _that_
575 * pointed to a cursor, and cursor got 566 * pointed to a cursor, and cursor got moved (e.g. by lseek())
576 * before d_walk() regains parent->d_l 567 * before d_walk() regains parent->d_lock, we'll end up skipping
577 * everything the cursor had been move 568 * everything the cursor had been moved past.
578 * 569 *
579 * Solution: make sure that the pointe !! 570 * Solution: make sure that the pointer left behind in ->d_child.next
580 * points to something that won't be m 571 * points to something that won't be moving around. I.e. skip the
581 * cursors. 572 * cursors.
582 */ 573 */
583 while (dentry->d_sib.next) { !! 574 while (dentry->d_child.next != &parent->d_subdirs) {
584 next = hlist_entry(dentry->d_s !! 575 next = list_entry(dentry->d_child.next, struct dentry, d_child);
585 if (likely(!(next->d_flags & D 576 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
586 break; 577 break;
587 dentry->d_sib.next = next->d_s !! 578 dentry->d_child.next = next->d_child.next;
588 } 579 }
589 } 580 }
590 581
591 static struct dentry *__dentry_kill(struct den !! 582 static void __dentry_kill(struct dentry *dentry)
592 { 583 {
593 struct dentry *parent = NULL; 584 struct dentry *parent = NULL;
594 bool can_free = true; 585 bool can_free = true;
>> 586 if (!IS_ROOT(dentry))
>> 587 parent = dentry->d_parent;
595 588
596 /* 589 /*
597 * The dentry is now unrecoverably dea 590 * The dentry is now unrecoverably dead to the world.
598 */ 591 */
599 lockref_mark_dead(&dentry->d_lockref); 592 lockref_mark_dead(&dentry->d_lockref);
600 593
601 /* 594 /*
602 * inform the fs via d_prune that this 595 * inform the fs via d_prune that this dentry is about to be
603 * unhashed and destroyed. 596 * unhashed and destroyed.
604 */ 597 */
605 if (dentry->d_flags & DCACHE_OP_PRUNE) 598 if (dentry->d_flags & DCACHE_OP_PRUNE)
606 dentry->d_op->d_prune(dentry); 599 dentry->d_op->d_prune(dentry);
607 600
608 if (dentry->d_flags & DCACHE_LRU_LIST) 601 if (dentry->d_flags & DCACHE_LRU_LIST) {
609 if (!(dentry->d_flags & DCACHE 602 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
610 d_lru_del(dentry); 603 d_lru_del(dentry);
611 } 604 }
612 /* if it was on the hash then remove i 605 /* if it was on the hash then remove it */
613 __d_drop(dentry); 606 __d_drop(dentry);
>> 607 dentry_unlist(dentry, parent);
>> 608 if (parent)
>> 609 spin_unlock(&parent->d_lock);
614 if (dentry->d_inode) 610 if (dentry->d_inode)
615 dentry_unlink_inode(dentry); 611 dentry_unlink_inode(dentry);
616 else 612 else
617 spin_unlock(&dentry->d_lock); 613 spin_unlock(&dentry->d_lock);
618 this_cpu_dec(nr_dentry); 614 this_cpu_dec(nr_dentry);
619 if (dentry->d_op && dentry->d_op->d_re 615 if (dentry->d_op && dentry->d_op->d_release)
620 dentry->d_op->d_release(dentry 616 dentry->d_op->d_release(dentry);
621 617
622 cond_resched(); !! 618 spin_lock(&dentry->d_lock);
623 /* now that it's negative, ->d_parent !! 619 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
624 if (!IS_ROOT(dentry)) { !! 620 dentry->d_flags |= DCACHE_MAY_FREE;
625 parent = dentry->d_parent; <<
626 spin_lock(&parent->d_lock); <<
627 } <<
628 spin_lock_nested(&dentry->d_lock, DENT <<
629 dentry_unlist(dentry); <<
630 if (dentry->d_flags & DCACHE_SHRINK_LI <<
631 can_free = false; 621 can_free = false;
>> 622 }
632 spin_unlock(&dentry->d_lock); 623 spin_unlock(&dentry->d_lock);
633 if (likely(can_free)) 624 if (likely(can_free))
634 dentry_free(dentry); 625 dentry_free(dentry);
635 if (parent && --parent->d_lockref.coun !! 626 cond_resched();
>> 627 }
>> 628
>> 629 static struct dentry *__lock_parent(struct dentry *dentry)
>> 630 {
>> 631 struct dentry *parent;
>> 632 rcu_read_lock();
>> 633 spin_unlock(&dentry->d_lock);
>> 634 again:
>> 635 parent = READ_ONCE(dentry->d_parent);
>> 636 spin_lock(&parent->d_lock);
>> 637 /*
>> 638 * We can't blindly lock dentry until we are sure
>> 639 * that we won't violate the locking order.
>> 640 * Any changes of dentry->d_parent must have
>> 641 * been done with parent->d_lock held, so
>> 642 * spin_lock() above is enough of a barrier
>> 643 * for checking if it's still our child.
>> 644 */
>> 645 if (unlikely(parent != dentry->d_parent)) {
636 spin_unlock(&parent->d_lock); 646 spin_unlock(&parent->d_lock);
637 return NULL; !! 647 goto again;
638 } 648 }
>> 649 rcu_read_unlock();
>> 650 if (parent != dentry)
>> 651 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
>> 652 else
>> 653 parent = NULL;
639 return parent; 654 return parent;
640 } 655 }
641 656
642 /* !! 657 static inline struct dentry *lock_parent(struct dentry *dentry)
643 * Lock a dentry for feeding it to __dentry_ki <<
644 * Called under rcu_read_lock() and dentry->d_ <<
645 * guarantees that nothing we access will be f <<
646 * Note that dentry is *not* protected from co <<
647 * d_delete(), etc. <<
648 * <<
649 * Return false if dentry is busy. Otherwise, <<
650 * that dentry's inode locked. <<
651 */ <<
652 <<
653 static bool lock_for_kill(struct dentry *dentr <<
654 { 658 {
655 struct inode *inode = dentry->d_inode; !! 659 struct dentry *parent = dentry->d_parent;
656 !! 660 if (IS_ROOT(dentry))
657 if (unlikely(dentry->d_lockref.count)) !! 661 return NULL;
658 return false; !! 662 if (likely(spin_trylock(&parent->d_lock)))
659 !! 663 return parent;
660 if (!inode || likely(spin_trylock(&ino !! 664 return __lock_parent(dentry);
661 return true; <<
662 <<
663 do { <<
664 spin_unlock(&dentry->d_lock); <<
665 spin_lock(&inode->i_lock); <<
666 spin_lock(&dentry->d_lock); <<
667 if (likely(inode == dentry->d_ <<
668 break; <<
669 spin_unlock(&inode->i_lock); <<
670 inode = dentry->d_inode; <<
671 } while (inode); <<
672 if (likely(!dentry->d_lockref.count)) <<
673 return true; <<
674 if (inode) <<
675 spin_unlock(&inode->i_lock); <<
676 return false; <<
677 } 665 }
678 666
679 /* !! 667 static inline bool retain_dentry(struct dentry *dentry)
680 * Decide if dentry is worth retaining. Usual <<
681 * locked; if not locked, we are more limited <<
682 * without a lock. False in this case means " <<
683 * <<
684 * In case we aren't locked, these predicates <<
685 * sufficient that at some point after we drop <<
686 * hashed and the flags had the proper value. <<
687 * re-gotten a reference to the dentry and cha <<
688 * we can leave the dentry around with a zero <<
689 */ <<
690 static inline bool retain_dentry(struct dentry <<
691 { 668 {
692 unsigned int d_flags; !! 669 WARN_ON(d_in_lookup(dentry));
693 <<
694 smp_rmb(); <<
695 d_flags = READ_ONCE(dentry->d_flags); <<
696 670
697 // Unreachable? Nobody would be able t !! 671 /* Unreachable? Get rid of it */
698 if (unlikely(d_unhashed(dentry))) 672 if (unlikely(d_unhashed(dentry)))
699 return false; 673 return false;
700 674
701 // Same if it's disconnected !! 675 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
702 if (unlikely(d_flags & DCACHE_DISCONNE <<
703 return false; 676 return false;
704 677
705 // ->d_delete() might tell us not to b !! 678 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
706 // ->d_lock; can't decide without it !! 679 if (dentry->d_op->d_delete(dentry))
707 if (unlikely(d_flags & DCACHE_OP_DELET <<
708 if (!locked || dentry->d_op->d <<
709 return false; 680 return false;
710 } 681 }
711 682
712 // Explicitly told not to bother !! 683 if (unlikely(dentry->d_flags & DCACHE_DONTCACHE))
713 if (unlikely(d_flags & DCACHE_DONTCACH <<
714 return false; 684 return false;
715 685
716 // At this point it looks like we ough !! 686 /* retain; LRU fodder */
717 // need to do something - put it on LR !! 687 dentry->d_lockref.count--;
718 // and mark it referenced if it was on !! 688 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
719 // Unfortunately, both actions require <<
720 // case we'd have to punt rather than <<
721 if (unlikely(!(d_flags & DCACHE_LRU_LI <<
722 if (!locked) <<
723 return false; <<
724 d_lru_add(dentry); 689 d_lru_add(dentry);
725 } else if (unlikely(!(d_flags & DCACHE !! 690 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
726 if (!locked) <<
727 return false; <<
728 dentry->d_flags |= DCACHE_REFE 691 dentry->d_flags |= DCACHE_REFERENCED;
729 } <<
730 return true; 692 return true;
731 } 693 }
732 694
733 void d_mark_dontcache(struct inode *inode) 695 void d_mark_dontcache(struct inode *inode)
734 { 696 {
735 struct dentry *de; 697 struct dentry *de;
736 698
737 spin_lock(&inode->i_lock); 699 spin_lock(&inode->i_lock);
738 hlist_for_each_entry(de, &inode->i_den 700 hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
739 spin_lock(&de->d_lock); 701 spin_lock(&de->d_lock);
740 de->d_flags |= DCACHE_DONTCACH 702 de->d_flags |= DCACHE_DONTCACHE;
741 spin_unlock(&de->d_lock); 703 spin_unlock(&de->d_lock);
742 } 704 }
743 inode->i_state |= I_DONTCACHE; 705 inode->i_state |= I_DONTCACHE;
744 spin_unlock(&inode->i_lock); 706 spin_unlock(&inode->i_lock);
745 } 707 }
746 EXPORT_SYMBOL(d_mark_dontcache); 708 EXPORT_SYMBOL(d_mark_dontcache);
747 709
748 /* 710 /*
>> 711 * Finish off a dentry we've decided to kill.
>> 712 * dentry->d_lock must be held, returns with it unlocked.
>> 713 * Returns dentry requiring refcount drop, or NULL if we're done.
>> 714 */
>> 715 static struct dentry *dentry_kill(struct dentry *dentry)
>> 716 __releases(dentry->d_lock)
>> 717 {
>> 718 struct inode *inode = dentry->d_inode;
>> 719 struct dentry *parent = NULL;
>> 720
>> 721 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
>> 722 goto slow_positive;
>> 723
>> 724 if (!IS_ROOT(dentry)) {
>> 725 parent = dentry->d_parent;
>> 726 if (unlikely(!spin_trylock(&parent->d_lock))) {
>> 727 parent = __lock_parent(dentry);
>> 728 if (likely(inode || !dentry->d_inode))
>> 729 goto got_locks;
>> 730 /* negative that became positive */
>> 731 if (parent)
>> 732 spin_unlock(&parent->d_lock);
>> 733 inode = dentry->d_inode;
>> 734 goto slow_positive;
>> 735 }
>> 736 }
>> 737 __dentry_kill(dentry);
>> 738 return parent;
>> 739
>> 740 slow_positive:
>> 741 spin_unlock(&dentry->d_lock);
>> 742 spin_lock(&inode->i_lock);
>> 743 spin_lock(&dentry->d_lock);
>> 744 parent = lock_parent(dentry);
>> 745 got_locks:
>> 746 if (unlikely(dentry->d_lockref.count != 1)) {
>> 747 dentry->d_lockref.count--;
>> 748 } else if (likely(!retain_dentry(dentry))) {
>> 749 __dentry_kill(dentry);
>> 750 return parent;
>> 751 }
>> 752 /* we are keeping it, after all */
>> 753 if (inode)
>> 754 spin_unlock(&inode->i_lock);
>> 755 if (parent)
>> 756 spin_unlock(&parent->d_lock);
>> 757 spin_unlock(&dentry->d_lock);
>> 758 return NULL;
>> 759 }
>> 760
>> 761 /*
749 * Try to do a lockless dput(), and return whe 762 * Try to do a lockless dput(), and return whether that was successful.
750 * 763 *
751 * If unsuccessful, we return false, having al 764 * If unsuccessful, we return false, having already taken the dentry lock.
752 * In that case refcount is guaranteed to be z <<
753 * decided that it's not worth keeping around. <<
754 * 765 *
755 * The caller needs to hold the RCU read lock, 766 * The caller needs to hold the RCU read lock, so that the dentry is
756 * guaranteed to stay around even if the refco 767 * guaranteed to stay around even if the refcount goes down to zero!
757 */ 768 */
758 static inline bool fast_dput(struct dentry *de 769 static inline bool fast_dput(struct dentry *dentry)
759 { 770 {
760 int ret; 771 int ret;
>> 772 unsigned int d_flags;
>> 773
>> 774 /*
>> 775 * If we have a d_op->d_delete() operation, we sould not
>> 776 * let the dentry count go to zero, so use "put_or_lock".
>> 777 */
>> 778 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
>> 779 return lockref_put_or_lock(&dentry->d_lockref);
761 780
762 /* 781 /*
763 * try to decrement the lockref optimi !! 782 * .. otherwise, we can try to just decrement the
>> 783 * lockref optimistically.
764 */ 784 */
765 ret = lockref_put_return(&dentry->d_lo 785 ret = lockref_put_return(&dentry->d_lockref);
766 786
767 /* 787 /*
768 * If the lockref_put_return() failed 788 * If the lockref_put_return() failed due to the lock being held
769 * by somebody else, the fast path has 789 * by somebody else, the fast path has failed. We will need to
770 * get the lock, and then check the co 790 * get the lock, and then check the count again.
771 */ 791 */
772 if (unlikely(ret < 0)) { 792 if (unlikely(ret < 0)) {
773 spin_lock(&dentry->d_lock); 793 spin_lock(&dentry->d_lock);
774 if (WARN_ON_ONCE(dentry->d_loc 794 if (WARN_ON_ONCE(dentry->d_lockref.count <= 0)) {
775 spin_unlock(&dentry->d 795 spin_unlock(&dentry->d_lock);
776 return true; 796 return true;
777 } 797 }
778 dentry->d_lockref.count--; 798 dentry->d_lockref.count--;
779 goto locked; 799 goto locked;
780 } 800 }
781 801
782 /* 802 /*
783 * If we weren't the last ref, we're d 803 * If we weren't the last ref, we're done.
784 */ 804 */
785 if (ret) 805 if (ret)
786 return true; 806 return true;
787 807
788 /* 808 /*
789 * Can we decide that decrement of ref !! 809 * Careful, careful. The reference count went down
790 * taking the lock? There's a very co !! 810 * to zero, but we don't hold the dentry lock, so
791 * dentry looks like it ought to be re !! 811 * somebody else could get it again, and do another
792 * to do. !! 812 * dput(), and we need to not race with that.
>> 813 *
>> 814 * However, there is a very special and common case
>> 815 * where we don't care, because there is nothing to
>> 816 * do: the dentry is still hashed, it does not have
>> 817 * a 'delete' op, and it's referenced and already on
>> 818 * the LRU list.
>> 819 *
>> 820 * NOTE! Since we aren't locked, these values are
>> 821 * not "stable". However, it is sufficient that at
>> 822 * some point after we dropped the reference the
>> 823 * dentry was hashed and the flags had the proper
>> 824 * value. Other dentry users may have re-gotten
>> 825 * a reference to the dentry and change that, but
>> 826 * our work is done - we can leave the dentry
>> 827 * around with a zero refcount.
>> 828 *
>> 829 * Nevertheless, there are two cases that we should kill
>> 830 * the dentry anyway.
>> 831 * 1. free disconnected dentries as soon as their refcount
>> 832 * reached zero.
>> 833 * 2. free dentries if they should not be cached.
793 */ 834 */
794 if (retain_dentry(dentry, false)) !! 835 smp_rmb();
>> 836 d_flags = READ_ONCE(dentry->d_flags);
>> 837 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
>> 838 DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
>> 839
>> 840 /* Nothing to do? Dropping the reference was all we needed? */
>> 841 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
795 return true; 842 return true;
796 843
797 /* 844 /*
798 * Either not worth retaining or we ca !! 845 * Not the fast normal case? Get the lock. We've already decremented
799 * Get the lock, then. We've already !! 846 * the refcount, but we'll need to re-check the situation after
800 * but we'll need to re-check the situ !! 847 * getting the lock.
801 */ 848 */
802 spin_lock(&dentry->d_lock); 849 spin_lock(&dentry->d_lock);
803 850
804 /* 851 /*
805 * Did somebody else grab a reference 852 * Did somebody else grab a reference to it in the meantime, and
806 * we're no longer the last user after 853 * we're no longer the last user after all? Alternatively, somebody
807 * else could have killed it and marke 854 * else could have killed it and marked it dead. Either way, we
808 * don't need to do anything else. 855 * don't need to do anything else.
809 */ 856 */
810 locked: 857 locked:
811 if (dentry->d_lockref.count || retain_ !! 858 if (dentry->d_lockref.count) {
812 spin_unlock(&dentry->d_lock); 859 spin_unlock(&dentry->d_lock);
813 return true; 860 return true;
814 } 861 }
>> 862
>> 863 /*
>> 864 * Re-get the reference we optimistically dropped. We hold the
>> 865 * lock, and we just tested that it was zero, so we can just
>> 866 * set it to 1.
>> 867 */
>> 868 dentry->d_lockref.count = 1;
815 return false; 869 return false;
816 } 870 }
817 871
818 872
819 /* 873 /*
820 * This is dput 874 * This is dput
821 * 875 *
822 * This is complicated by the fact that we do 876 * This is complicated by the fact that we do not want to put
823 * dentries that are no longer on any hash cha 877 * dentries that are no longer on any hash chain on the unused
824 * list: we'd much rather just get rid of them 878 * list: we'd much rather just get rid of them immediately.
825 * 879 *
826 * However, that implies that we have to trave 880 * However, that implies that we have to traverse the dentry
827 * tree upwards to the parents which might _al 881 * tree upwards to the parents which might _also_ now be
828 * scheduled for deletion (it may have been on 882 * scheduled for deletion (it may have been only waiting for
829 * its last child to go away). 883 * its last child to go away).
830 * 884 *
831 * This tail recursion is done by hand as we d 885 * This tail recursion is done by hand as we don't want to depend
832 * on the compiler to always get this right (g 886 * on the compiler to always get this right (gcc generally doesn't).
833 * Real recursion would eat up our stack space 887 * Real recursion would eat up our stack space.
834 */ 888 */
835 889
836 /* 890 /*
837 * dput - release a dentry 891 * dput - release a dentry
838 * @dentry: dentry to release 892 * @dentry: dentry to release
839 * 893 *
840 * Release a dentry. This will drop the usage 894 * Release a dentry. This will drop the usage count and if appropriate
841 * call the dentry unlink method as well as re 895 * call the dentry unlink method as well as removing it from the queues and
842 * releasing its resources. If the parent dent 896 * releasing its resources. If the parent dentries were scheduled for release
843 * they too may now get deleted. 897 * they too may now get deleted.
844 */ 898 */
845 void dput(struct dentry *dentry) 899 void dput(struct dentry *dentry)
846 { 900 {
847 if (!dentry) !! 901 while (dentry) {
848 return; !! 902 might_sleep();
849 might_sleep(); !! 903
850 rcu_read_lock(); !! 904 rcu_read_lock();
851 if (likely(fast_dput(dentry))) { !! 905 if (likely(fast_dput(dentry))) {
852 rcu_read_unlock(); !! 906 rcu_read_unlock();
853 return; <<
854 } <<
855 while (lock_for_kill(dentry)) { <<
856 rcu_read_unlock(); <<
857 dentry = __dentry_kill(dentry) <<
858 if (!dentry) <<
859 return; 907 return;
860 if (retain_dentry(dentry, true !! 908 }
>> 909
>> 910 /* Slow case: now with the dentry lock held */
>> 911 rcu_read_unlock();
>> 912
>> 913 if (likely(retain_dentry(dentry))) {
861 spin_unlock(&dentry->d 914 spin_unlock(&dentry->d_lock);
862 return; 915 return;
863 } 916 }
864 rcu_read_lock(); !! 917
>> 918 dentry = dentry_kill(dentry);
865 } 919 }
866 rcu_read_unlock(); <<
867 spin_unlock(&dentry->d_lock); <<
868 } 920 }
869 EXPORT_SYMBOL(dput); 921 EXPORT_SYMBOL(dput);
870 922
871 static void to_shrink_list(struct dentry *dent !! 923 static void __dput_to_list(struct dentry *dentry, struct list_head *list)
872 __must_hold(&dentry->d_lock) 924 __must_hold(&dentry->d_lock)
873 { 925 {
874 if (!(dentry->d_flags & DCACHE_SHRINK_ !! 926 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
>> 927 /* let the owner of the list it's on deal with it */
>> 928 --dentry->d_lockref.count;
>> 929 } else {
875 if (dentry->d_flags & DCACHE_L 930 if (dentry->d_flags & DCACHE_LRU_LIST)
876 d_lru_del(dentry); 931 d_lru_del(dentry);
877 d_shrink_add(dentry, list); !! 932 if (!--dentry->d_lockref.count)
>> 933 d_shrink_add(dentry, list);
878 } 934 }
879 } 935 }
880 936
881 void dput_to_list(struct dentry *dentry, struc 937 void dput_to_list(struct dentry *dentry, struct list_head *list)
882 { 938 {
883 rcu_read_lock(); 939 rcu_read_lock();
884 if (likely(fast_dput(dentry))) { 940 if (likely(fast_dput(dentry))) {
885 rcu_read_unlock(); 941 rcu_read_unlock();
886 return; 942 return;
887 } 943 }
888 rcu_read_unlock(); 944 rcu_read_unlock();
889 to_shrink_list(dentry, list); !! 945 if (!retain_dentry(dentry))
>> 946 __dput_to_list(dentry, list);
890 spin_unlock(&dentry->d_lock); 947 spin_unlock(&dentry->d_lock);
891 } 948 }
892 949
>> 950 /* This must be called with d_lock held */
>> 951 static inline void __dget_dlock(struct dentry *dentry)
>> 952 {
>> 953 dentry->d_lockref.count++;
>> 954 }
>> 955
>> 956 static inline void __dget(struct dentry *dentry)
>> 957 {
>> 958 lockref_get(&dentry->d_lockref);
>> 959 }
>> 960
893 struct dentry *dget_parent(struct dentry *dent 961 struct dentry *dget_parent(struct dentry *dentry)
894 { 962 {
895 int gotref; 963 int gotref;
896 struct dentry *ret; 964 struct dentry *ret;
897 unsigned seq; 965 unsigned seq;
898 966
899 /* 967 /*
900 * Do optimistic parent lookup without 968 * Do optimistic parent lookup without any
901 * locking. 969 * locking.
902 */ 970 */
903 rcu_read_lock(); 971 rcu_read_lock();
904 seq = raw_seqcount_begin(&dentry->d_se 972 seq = raw_seqcount_begin(&dentry->d_seq);
905 ret = READ_ONCE(dentry->d_parent); 973 ret = READ_ONCE(dentry->d_parent);
906 gotref = lockref_get_not_zero(&ret->d_ 974 gotref = lockref_get_not_zero(&ret->d_lockref);
907 rcu_read_unlock(); 975 rcu_read_unlock();
908 if (likely(gotref)) { 976 if (likely(gotref)) {
909 if (!read_seqcount_retry(&dent 977 if (!read_seqcount_retry(&dentry->d_seq, seq))
910 return ret; 978 return ret;
911 dput(ret); 979 dput(ret);
912 } 980 }
913 981
914 repeat: 982 repeat:
915 /* 983 /*
916 * Don't need rcu_dereference because 984 * Don't need rcu_dereference because we re-check it was correct under
917 * the lock. 985 * the lock.
918 */ 986 */
919 rcu_read_lock(); 987 rcu_read_lock();
920 ret = dentry->d_parent; 988 ret = dentry->d_parent;
921 spin_lock(&ret->d_lock); 989 spin_lock(&ret->d_lock);
922 if (unlikely(ret != dentry->d_parent)) 990 if (unlikely(ret != dentry->d_parent)) {
923 spin_unlock(&ret->d_lock); 991 spin_unlock(&ret->d_lock);
924 rcu_read_unlock(); 992 rcu_read_unlock();
925 goto repeat; 993 goto repeat;
926 } 994 }
927 rcu_read_unlock(); 995 rcu_read_unlock();
928 BUG_ON(!ret->d_lockref.count); 996 BUG_ON(!ret->d_lockref.count);
929 ret->d_lockref.count++; 997 ret->d_lockref.count++;
930 spin_unlock(&ret->d_lock); 998 spin_unlock(&ret->d_lock);
931 return ret; 999 return ret;
932 } 1000 }
933 EXPORT_SYMBOL(dget_parent); 1001 EXPORT_SYMBOL(dget_parent);
934 1002
935 static struct dentry * __d_find_any_alias(stru 1003 static struct dentry * __d_find_any_alias(struct inode *inode)
936 { 1004 {
937 struct dentry *alias; 1005 struct dentry *alias;
938 1006
939 if (hlist_empty(&inode->i_dentry)) 1007 if (hlist_empty(&inode->i_dentry))
940 return NULL; 1008 return NULL;
941 alias = hlist_entry(inode->i_dentry.fi 1009 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
942 lockref_get(&alias->d_lockref); !! 1010 __dget(alias);
943 return alias; 1011 return alias;
944 } 1012 }
945 1013
946 /** 1014 /**
947 * d_find_any_alias - find any alias for a giv 1015 * d_find_any_alias - find any alias for a given inode
948 * @inode: inode to find an alias for 1016 * @inode: inode to find an alias for
949 * 1017 *
950 * If any aliases exist for the given inode, t 1018 * If any aliases exist for the given inode, take and return a
951 * reference for one of them. If no aliases e 1019 * reference for one of them. If no aliases exist, return %NULL.
952 */ 1020 */
953 struct dentry *d_find_any_alias(struct inode * 1021 struct dentry *d_find_any_alias(struct inode *inode)
954 { 1022 {
955 struct dentry *de; 1023 struct dentry *de;
956 1024
957 spin_lock(&inode->i_lock); 1025 spin_lock(&inode->i_lock);
958 de = __d_find_any_alias(inode); 1026 de = __d_find_any_alias(inode);
959 spin_unlock(&inode->i_lock); 1027 spin_unlock(&inode->i_lock);
960 return de; 1028 return de;
961 } 1029 }
962 EXPORT_SYMBOL(d_find_any_alias); 1030 EXPORT_SYMBOL(d_find_any_alias);
963 1031
964 static struct dentry *__d_find_alias(struct in 1032 static struct dentry *__d_find_alias(struct inode *inode)
965 { 1033 {
966 struct dentry *alias; 1034 struct dentry *alias;
967 1035
968 if (S_ISDIR(inode->i_mode)) 1036 if (S_ISDIR(inode->i_mode))
969 return __d_find_any_alias(inod 1037 return __d_find_any_alias(inode);
970 1038
971 hlist_for_each_entry(alias, &inode->i_ 1039 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
972 spin_lock(&alias->d_lock); 1040 spin_lock(&alias->d_lock);
973 if (!d_unhashed(alias)) { 1041 if (!d_unhashed(alias)) {
974 dget_dlock(alias); !! 1042 __dget_dlock(alias);
975 spin_unlock(&alias->d_ 1043 spin_unlock(&alias->d_lock);
976 return alias; 1044 return alias;
977 } 1045 }
978 spin_unlock(&alias->d_lock); 1046 spin_unlock(&alias->d_lock);
979 } 1047 }
980 return NULL; 1048 return NULL;
981 } 1049 }
982 1050
983 /** 1051 /**
984 * d_find_alias - grab a hashed alias of inode 1052 * d_find_alias - grab a hashed alias of inode
985 * @inode: inode in question 1053 * @inode: inode in question
986 * 1054 *
987 * If inode has a hashed alias, or is a direct 1055 * If inode has a hashed alias, or is a directory and has any alias,
988 * acquire the reference to alias and return i 1056 * acquire the reference to alias and return it. Otherwise return NULL.
989 * Notice that if inode is a directory there c 1057 * 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 1058 * 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 1059 * of a filesystem, or if the directory was renamed and d_revalidate
992 * was the first vfs operation to notice. 1060 * was the first vfs operation to notice.
993 * 1061 *
994 * If the inode has an IS_ROOT, DCACHE_DISCONN 1062 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
995 * any other hashed alias over that one. 1063 * any other hashed alias over that one.
996 */ 1064 */
997 struct dentry *d_find_alias(struct inode *inod 1065 struct dentry *d_find_alias(struct inode *inode)
998 { 1066 {
999 struct dentry *de = NULL; 1067 struct dentry *de = NULL;
1000 1068
1001 if (!hlist_empty(&inode->i_dentry)) { 1069 if (!hlist_empty(&inode->i_dentry)) {
1002 spin_lock(&inode->i_lock); 1070 spin_lock(&inode->i_lock);
1003 de = __d_find_alias(inode); 1071 de = __d_find_alias(inode);
1004 spin_unlock(&inode->i_lock); 1072 spin_unlock(&inode->i_lock);
1005 } 1073 }
1006 return de; 1074 return de;
1007 } 1075 }
1008 EXPORT_SYMBOL(d_find_alias); 1076 EXPORT_SYMBOL(d_find_alias);
1009 1077
1010 /* 1078 /*
1011 * Caller MUST be holding rcu_read_lock() an 1079 * Caller MUST be holding rcu_read_lock() and be guaranteed
1012 * that inode won't get freed until rcu_read 1080 * that inode won't get freed until rcu_read_unlock().
1013 */ 1081 */
1014 struct dentry *d_find_alias_rcu(struct inode 1082 struct dentry *d_find_alias_rcu(struct inode *inode)
1015 { 1083 {
1016 struct hlist_head *l = &inode->i_dent 1084 struct hlist_head *l = &inode->i_dentry;
1017 struct dentry *de = NULL; 1085 struct dentry *de = NULL;
1018 1086
1019 spin_lock(&inode->i_lock); 1087 spin_lock(&inode->i_lock);
1020 // ->i_dentry and ->i_rcu are colocat 1088 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1021 // used without having I_FREEING set, 1089 // used without having I_FREEING set, which means no aliases left
1022 if (likely(!(inode->i_state & I_FREEI 1090 if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1023 if (S_ISDIR(inode->i_mode)) { 1091 if (S_ISDIR(inode->i_mode)) {
1024 de = hlist_entry(l->f 1092 de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1025 } else { 1093 } else {
1026 hlist_for_each_entry( 1094 hlist_for_each_entry(de, l, d_u.d_alias)
1027 if (!d_unhash 1095 if (!d_unhashed(de))
1028 break 1096 break;
1029 } 1097 }
1030 } 1098 }
1031 spin_unlock(&inode->i_lock); 1099 spin_unlock(&inode->i_lock);
1032 return de; 1100 return de;
1033 } 1101 }
1034 1102
1035 /* 1103 /*
1036 * Try to kill dentries associated with 1104 * Try to kill dentries associated with this inode.
1037 * WARNING: you must own a reference to inode 1105 * WARNING: you must own a reference to inode.
1038 */ 1106 */
1039 void d_prune_aliases(struct inode *inode) 1107 void d_prune_aliases(struct inode *inode)
1040 { 1108 {
1041 LIST_HEAD(dispose); <<
1042 struct dentry *dentry; 1109 struct dentry *dentry;
1043 !! 1110 restart:
1044 spin_lock(&inode->i_lock); 1111 spin_lock(&inode->i_lock);
1045 hlist_for_each_entry(dentry, &inode-> 1112 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1046 spin_lock(&dentry->d_lock); 1113 spin_lock(&dentry->d_lock);
1047 if (!dentry->d_lockref.count) !! 1114 if (!dentry->d_lockref.count) {
1048 to_shrink_list(dentry !! 1115 struct dentry *parent = lock_parent(dentry);
>> 1116 if (likely(!dentry->d_lockref.count)) {
>> 1117 __dentry_kill(dentry);
>> 1118 dput(parent);
>> 1119 goto restart;
>> 1120 }
>> 1121 if (parent)
>> 1122 spin_unlock(&parent->d_lock);
>> 1123 }
1049 spin_unlock(&dentry->d_lock); 1124 spin_unlock(&dentry->d_lock);
1050 } 1125 }
1051 spin_unlock(&inode->i_lock); 1126 spin_unlock(&inode->i_lock);
1052 shrink_dentry_list(&dispose); <<
1053 } 1127 }
1054 EXPORT_SYMBOL(d_prune_aliases); 1128 EXPORT_SYMBOL(d_prune_aliases);
1055 1129
1056 static inline void shrink_kill(struct dentry !! 1130 /*
>> 1131 * Lock a dentry from shrink list.
>> 1132 * Called under rcu_read_lock() and dentry->d_lock; the former
>> 1133 * guarantees that nothing we access will be freed under us.
>> 1134 * Note that dentry is *not* protected from concurrent dentry_kill(),
>> 1135 * d_delete(), etc.
>> 1136 *
>> 1137 * Return false if dentry has been disrupted or grabbed, leaving
>> 1138 * the caller to kick it off-list. Otherwise, return true and have
>> 1139 * that dentry's inode and parent both locked.
>> 1140 */
>> 1141 static bool shrink_lock_dentry(struct dentry *dentry)
1057 { 1142 {
1058 do { !! 1143 struct inode *inode;
1059 rcu_read_unlock(); !! 1144 struct dentry *parent;
1060 victim = __dentry_kill(victim !! 1145
1061 rcu_read_lock(); !! 1146 if (dentry->d_lockref.count)
1062 } while (victim && lock_for_kill(vict !! 1147 return false;
1063 rcu_read_unlock(); !! 1148
1064 if (victim) !! 1149 inode = dentry->d_inode;
1065 spin_unlock(&victim->d_lock); !! 1150 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
>> 1151 spin_unlock(&dentry->d_lock);
>> 1152 spin_lock(&inode->i_lock);
>> 1153 spin_lock(&dentry->d_lock);
>> 1154 if (unlikely(dentry->d_lockref.count))
>> 1155 goto out;
>> 1156 /* changed inode means that somebody had grabbed it */
>> 1157 if (unlikely(inode != dentry->d_inode))
>> 1158 goto out;
>> 1159 }
>> 1160
>> 1161 parent = dentry->d_parent;
>> 1162 if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
>> 1163 return true;
>> 1164
>> 1165 spin_unlock(&dentry->d_lock);
>> 1166 spin_lock(&parent->d_lock);
>> 1167 if (unlikely(parent != dentry->d_parent)) {
>> 1168 spin_unlock(&parent->d_lock);
>> 1169 spin_lock(&dentry->d_lock);
>> 1170 goto out;
>> 1171 }
>> 1172 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
>> 1173 if (likely(!dentry->d_lockref.count))
>> 1174 return true;
>> 1175 spin_unlock(&parent->d_lock);
>> 1176 out:
>> 1177 if (inode)
>> 1178 spin_unlock(&inode->i_lock);
>> 1179 return false;
1066 } 1180 }
1067 1181
1068 void shrink_dentry_list(struct list_head *lis 1182 void shrink_dentry_list(struct list_head *list)
1069 { 1183 {
1070 while (!list_empty(list)) { 1184 while (!list_empty(list)) {
1071 struct dentry *dentry; !! 1185 struct dentry *dentry, *parent;
1072 1186
1073 dentry = list_entry(list->pre 1187 dentry = list_entry(list->prev, struct dentry, d_lru);
1074 spin_lock(&dentry->d_lock); 1188 spin_lock(&dentry->d_lock);
1075 rcu_read_lock(); 1189 rcu_read_lock();
1076 if (!lock_for_kill(dentry)) { !! 1190 if (!shrink_lock_dentry(dentry)) {
1077 bool can_free; !! 1191 bool can_free = false;
1078 rcu_read_unlock(); 1192 rcu_read_unlock();
1079 d_shrink_del(dentry); 1193 d_shrink_del(dentry);
1080 can_free = dentry->d_ !! 1194 if (dentry->d_lockref.count < 0)
>> 1195 can_free = dentry->d_flags & DCACHE_MAY_FREE;
1081 spin_unlock(&dentry-> 1196 spin_unlock(&dentry->d_lock);
1082 if (can_free) 1197 if (can_free)
1083 dentry_free(d 1198 dentry_free(dentry);
1084 continue; 1199 continue;
1085 } 1200 }
>> 1201 rcu_read_unlock();
1086 d_shrink_del(dentry); 1202 d_shrink_del(dentry);
1087 shrink_kill(dentry); !! 1203 parent = dentry->d_parent;
>> 1204 if (parent != dentry)
>> 1205 __dput_to_list(parent, list);
>> 1206 __dentry_kill(dentry);
1088 } 1207 }
1089 } 1208 }
1090 1209
1091 static enum lru_status dentry_lru_isolate(str 1210 static enum lru_status dentry_lru_isolate(struct list_head *item,
1092 struct list_lru_one *lru, spi 1211 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1093 { 1212 {
1094 struct list_head *freeable = arg; 1213 struct list_head *freeable = arg;
1095 struct dentry *dentry = container_o 1214 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1096 1215
1097 1216
1098 /* 1217 /*
1099 * we are inverting the lru lock/dent 1218 * we are inverting the lru lock/dentry->d_lock here,
1100 * so use a trylock. If we fail to ge 1219 * so use a trylock. If we fail to get the lock, just skip
1101 * it 1220 * it
1102 */ 1221 */
1103 if (!spin_trylock(&dentry->d_lock)) 1222 if (!spin_trylock(&dentry->d_lock))
1104 return LRU_SKIP; 1223 return LRU_SKIP;
1105 1224
1106 /* 1225 /*
1107 * Referenced dentries are still in u 1226 * Referenced dentries are still in use. If they have active
1108 * counts, just remove them from the 1227 * counts, just remove them from the LRU. Otherwise give them
1109 * another pass through the LRU. 1228 * another pass through the LRU.
1110 */ 1229 */
1111 if (dentry->d_lockref.count) { 1230 if (dentry->d_lockref.count) {
1112 d_lru_isolate(lru, dentry); 1231 d_lru_isolate(lru, dentry);
1113 spin_unlock(&dentry->d_lock); 1232 spin_unlock(&dentry->d_lock);
1114 return LRU_REMOVED; 1233 return LRU_REMOVED;
1115 } 1234 }
1116 1235
1117 if (dentry->d_flags & DCACHE_REFERENC 1236 if (dentry->d_flags & DCACHE_REFERENCED) {
1118 dentry->d_flags &= ~DCACHE_RE 1237 dentry->d_flags &= ~DCACHE_REFERENCED;
1119 spin_unlock(&dentry->d_lock); 1238 spin_unlock(&dentry->d_lock);
1120 1239
1121 /* 1240 /*
1122 * The list move itself will 1241 * The list move itself will be made by the common LRU code. At
1123 * this point, we've dropped 1242 * this point, we've dropped the dentry->d_lock but keep the
1124 * lru lock. This is safe to 1243 * lru lock. This is safe to do, since every list movement is
1125 * protected by the lru lock 1244 * protected by the lru lock even if both locks are held.
1126 * 1245 *
1127 * This is guaranteed by the 1246 * This is guaranteed by the fact that all LRU management
1128 * functions are intermediate 1247 * functions are intermediated by the LRU API calls like
1129 * list_lru_add_obj and list_ !! 1248 * list_lru_add and list_lru_del. List movement in this file
1130 * only ever occur through th 1249 * only ever occur through this functions or through callbacks
1131 * like this one, that are ca 1250 * like this one, that are called from the LRU API.
1132 * 1251 *
1133 * The only exceptions to thi 1252 * The only exceptions to this are functions like
1134 * shrink_dentry_list, and co 1253 * shrink_dentry_list, and code that first checks for the
1135 * DCACHE_SHRINK_LIST flag. 1254 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1136 * operating only with stack 1255 * operating only with stack provided lists after they are
1137 * properly isolated from the 1256 * properly isolated from the main list. It is thus, always a
1138 * local access. 1257 * local access.
1139 */ 1258 */
1140 return LRU_ROTATE; 1259 return LRU_ROTATE;
1141 } 1260 }
1142 1261
1143 d_lru_shrink_move(lru, dentry, freeab 1262 d_lru_shrink_move(lru, dentry, freeable);
1144 spin_unlock(&dentry->d_lock); 1263 spin_unlock(&dentry->d_lock);
1145 1264
1146 return LRU_REMOVED; 1265 return LRU_REMOVED;
1147 } 1266 }
1148 1267
1149 /** 1268 /**
1150 * prune_dcache_sb - shrink the dcache 1269 * prune_dcache_sb - shrink the dcache
1151 * @sb: superblock 1270 * @sb: superblock
1152 * @sc: shrink control, passed to list_lru_sh 1271 * @sc: shrink control, passed to list_lru_shrink_walk()
1153 * 1272 *
1154 * Attempt to shrink the superblock dcache LR 1273 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1155 * is done when we need more memory and calle 1274 * is done when we need more memory and called from the superblock shrinker
1156 * function. 1275 * function.
1157 * 1276 *
1158 * This function may fail to free any resourc 1277 * This function may fail to free any resources if all the dentries are in
1159 * use. 1278 * use.
1160 */ 1279 */
1161 long prune_dcache_sb(struct super_block *sb, 1280 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1162 { 1281 {
1163 LIST_HEAD(dispose); 1282 LIST_HEAD(dispose);
1164 long freed; 1283 long freed;
1165 1284
1166 freed = list_lru_shrink_walk(&sb->s_d 1285 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1167 dentry_l 1286 dentry_lru_isolate, &dispose);
1168 shrink_dentry_list(&dispose); 1287 shrink_dentry_list(&dispose);
1169 return freed; 1288 return freed;
1170 } 1289 }
1171 1290
1172 static enum lru_status dentry_lru_isolate_shr 1291 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1173 struct list_lru_one *lru, spi 1292 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1174 { 1293 {
1175 struct list_head *freeable = arg; 1294 struct list_head *freeable = arg;
1176 struct dentry *dentry = container_o 1295 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1177 1296
1178 /* 1297 /*
1179 * we are inverting the lru lock/dent 1298 * we are inverting the lru lock/dentry->d_lock here,
1180 * so use a trylock. If we fail to ge 1299 * so use a trylock. If we fail to get the lock, just skip
1181 * it 1300 * it
1182 */ 1301 */
1183 if (!spin_trylock(&dentry->d_lock)) 1302 if (!spin_trylock(&dentry->d_lock))
1184 return LRU_SKIP; 1303 return LRU_SKIP;
1185 1304
1186 d_lru_shrink_move(lru, dentry, freeab 1305 d_lru_shrink_move(lru, dentry, freeable);
1187 spin_unlock(&dentry->d_lock); 1306 spin_unlock(&dentry->d_lock);
1188 1307
1189 return LRU_REMOVED; 1308 return LRU_REMOVED;
1190 } 1309 }
1191 1310
1192 1311
1193 /** 1312 /**
1194 * shrink_dcache_sb - shrink dcache for a sup 1313 * shrink_dcache_sb - shrink dcache for a superblock
1195 * @sb: superblock 1314 * @sb: superblock
1196 * 1315 *
1197 * Shrink the dcache for the specified super 1316 * Shrink the dcache for the specified super block. This is used to free
1198 * the dcache before unmounting a file system 1317 * the dcache before unmounting a file system.
1199 */ 1318 */
1200 void shrink_dcache_sb(struct super_block *sb) 1319 void shrink_dcache_sb(struct super_block *sb)
1201 { 1320 {
1202 do { 1321 do {
1203 LIST_HEAD(dispose); 1322 LIST_HEAD(dispose);
1204 1323
1205 list_lru_walk(&sb->s_dentry_l 1324 list_lru_walk(&sb->s_dentry_lru,
1206 dentry_lru_isolate_sh 1325 dentry_lru_isolate_shrink, &dispose, 1024);
1207 shrink_dentry_list(&dispose); 1326 shrink_dentry_list(&dispose);
1208 } while (list_lru_count(&sb->s_dentry 1327 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1209 } 1328 }
1210 EXPORT_SYMBOL(shrink_dcache_sb); 1329 EXPORT_SYMBOL(shrink_dcache_sb);
1211 1330
1212 /** 1331 /**
1213 * enum d_walk_ret - action to talke during t 1332 * enum d_walk_ret - action to talke during tree walk
1214 * @D_WALK_CONTINUE: contrinue walk 1333 * @D_WALK_CONTINUE: contrinue walk
1215 * @D_WALK_QUIT: quit walk 1334 * @D_WALK_QUIT: quit walk
1216 * @D_WALK_NORETRY: quit when retry is ne 1335 * @D_WALK_NORETRY: quit when retry is needed
1217 * @D_WALK_SKIP: skip this dentry and 1336 * @D_WALK_SKIP: skip this dentry and its children
1218 */ 1337 */
1219 enum d_walk_ret { 1338 enum d_walk_ret {
1220 D_WALK_CONTINUE, 1339 D_WALK_CONTINUE,
1221 D_WALK_QUIT, 1340 D_WALK_QUIT,
1222 D_WALK_NORETRY, 1341 D_WALK_NORETRY,
1223 D_WALK_SKIP, 1342 D_WALK_SKIP,
1224 }; 1343 };
1225 1344
1226 /** 1345 /**
1227 * d_walk - walk the dentry tree 1346 * d_walk - walk the dentry tree
1228 * @parent: start of walk 1347 * @parent: start of walk
1229 * @data: data passed to @enter() and @ 1348 * @data: data passed to @enter() and @finish()
1230 * @enter: callback when first entering 1349 * @enter: callback when first entering the dentry
1231 * 1350 *
1232 * The @enter() callbacks are called with d_l 1351 * The @enter() callbacks are called with d_lock held.
1233 */ 1352 */
1234 static void d_walk(struct dentry *parent, voi 1353 static void d_walk(struct dentry *parent, void *data,
1235 enum d_walk_ret (*enter)(v 1354 enum d_walk_ret (*enter)(void *, struct dentry *))
1236 { 1355 {
1237 struct dentry *this_parent, *dentry; !! 1356 struct dentry *this_parent;
>> 1357 struct list_head *next;
1238 unsigned seq = 0; 1358 unsigned seq = 0;
1239 enum d_walk_ret ret; 1359 enum d_walk_ret ret;
1240 bool retry = true; 1360 bool retry = true;
1241 1361
1242 again: 1362 again:
1243 read_seqbegin_or_lock(&rename_lock, & 1363 read_seqbegin_or_lock(&rename_lock, &seq);
1244 this_parent = parent; 1364 this_parent = parent;
1245 spin_lock(&this_parent->d_lock); 1365 spin_lock(&this_parent->d_lock);
1246 1366
1247 ret = enter(data, this_parent); 1367 ret = enter(data, this_parent);
1248 switch (ret) { 1368 switch (ret) {
1249 case D_WALK_CONTINUE: 1369 case D_WALK_CONTINUE:
1250 break; 1370 break;
1251 case D_WALK_QUIT: 1371 case D_WALK_QUIT:
1252 case D_WALK_SKIP: 1372 case D_WALK_SKIP:
1253 goto out_unlock; 1373 goto out_unlock;
1254 case D_WALK_NORETRY: 1374 case D_WALK_NORETRY:
1255 retry = false; 1375 retry = false;
1256 break; 1376 break;
1257 } 1377 }
1258 repeat: 1378 repeat:
1259 dentry = d_first_child(this_parent); !! 1379 next = this_parent->d_subdirs.next;
1260 resume: 1380 resume:
1261 hlist_for_each_entry_from(dentry, d_s !! 1381 while (next != &this_parent->d_subdirs) {
>> 1382 struct list_head *tmp = next;
>> 1383 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
>> 1384 next = tmp->next;
>> 1385
1262 if (unlikely(dentry->d_flags 1386 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1263 continue; 1387 continue;
1264 1388
1265 spin_lock_nested(&dentry->d_l 1389 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1266 1390
1267 ret = enter(data, dentry); 1391 ret = enter(data, dentry);
1268 switch (ret) { 1392 switch (ret) {
1269 case D_WALK_CONTINUE: 1393 case D_WALK_CONTINUE:
1270 break; 1394 break;
1271 case D_WALK_QUIT: 1395 case D_WALK_QUIT:
1272 spin_unlock(&dentry-> 1396 spin_unlock(&dentry->d_lock);
1273 goto out_unlock; 1397 goto out_unlock;
1274 case D_WALK_NORETRY: 1398 case D_WALK_NORETRY:
1275 retry = false; 1399 retry = false;
1276 break; 1400 break;
1277 case D_WALK_SKIP: 1401 case D_WALK_SKIP:
1278 spin_unlock(&dentry-> 1402 spin_unlock(&dentry->d_lock);
1279 continue; 1403 continue;
1280 } 1404 }
1281 1405
1282 if (!hlist_empty(&dentry->d_c !! 1406 if (!list_empty(&dentry->d_subdirs)) {
1283 spin_unlock(&this_par 1407 spin_unlock(&this_parent->d_lock);
1284 spin_release(&dentry- 1408 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1285 this_parent = dentry; 1409 this_parent = dentry;
1286 spin_acquire(&this_pa 1410 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1287 goto repeat; 1411 goto repeat;
1288 } 1412 }
1289 spin_unlock(&dentry->d_lock); 1413 spin_unlock(&dentry->d_lock);
1290 } 1414 }
1291 /* 1415 /*
1292 * All done at this level ... ascend 1416 * All done at this level ... ascend and resume the search.
1293 */ 1417 */
1294 rcu_read_lock(); 1418 rcu_read_lock();
1295 ascend: 1419 ascend:
1296 if (this_parent != parent) { 1420 if (this_parent != parent) {
1297 dentry = this_parent; !! 1421 struct dentry *child = this_parent;
1298 this_parent = dentry->d_paren !! 1422 this_parent = child->d_parent;
1299 1423
1300 spin_unlock(&dentry->d_lock); !! 1424 spin_unlock(&child->d_lock);
1301 spin_lock(&this_parent->d_loc 1425 spin_lock(&this_parent->d_lock);
1302 1426
1303 /* might go back up the wrong 1427 /* might go back up the wrong parent if we have had a rename. */
1304 if (need_seqretry(&rename_loc 1428 if (need_seqretry(&rename_lock, seq))
1305 goto rename_retry; 1429 goto rename_retry;
1306 /* go into the first sibling 1430 /* go into the first sibling still alive */
1307 hlist_for_each_entry_continue !! 1431 do {
1308 if (likely(!(dentry-> !! 1432 next = child->d_child.next;
1309 rcu_read_unlo !! 1433 if (next == &this_parent->d_subdirs)
1310 goto resume; !! 1434 goto ascend;
1311 } !! 1435 child = list_entry(next, struct dentry, d_child);
1312 } !! 1436 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1313 goto ascend; !! 1437 rcu_read_unlock();
>> 1438 goto resume;
1314 } 1439 }
1315 if (need_seqretry(&rename_lock, seq)) 1440 if (need_seqretry(&rename_lock, seq))
1316 goto rename_retry; 1441 goto rename_retry;
1317 rcu_read_unlock(); 1442 rcu_read_unlock();
1318 1443
1319 out_unlock: 1444 out_unlock:
1320 spin_unlock(&this_parent->d_lock); 1445 spin_unlock(&this_parent->d_lock);
1321 done_seqretry(&rename_lock, seq); 1446 done_seqretry(&rename_lock, seq);
1322 return; 1447 return;
1323 1448
1324 rename_retry: 1449 rename_retry:
1325 spin_unlock(&this_parent->d_lock); 1450 spin_unlock(&this_parent->d_lock);
1326 rcu_read_unlock(); 1451 rcu_read_unlock();
1327 BUG_ON(seq & 1); 1452 BUG_ON(seq & 1);
1328 if (!retry) 1453 if (!retry)
1329 return; 1454 return;
1330 seq = 1; 1455 seq = 1;
1331 goto again; 1456 goto again;
1332 } 1457 }
1333 1458
1334 struct check_mount { 1459 struct check_mount {
1335 struct vfsmount *mnt; 1460 struct vfsmount *mnt;
1336 unsigned int mounted; 1461 unsigned int mounted;
1337 }; 1462 };
1338 1463
1339 static enum d_walk_ret path_check_mount(void 1464 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1340 { 1465 {
1341 struct check_mount *info = data; 1466 struct check_mount *info = data;
1342 struct path path = { .mnt = info->mnt 1467 struct path path = { .mnt = info->mnt, .dentry = dentry };
1343 1468
1344 if (likely(!d_mountpoint(dentry))) 1469 if (likely(!d_mountpoint(dentry)))
1345 return D_WALK_CONTINUE; 1470 return D_WALK_CONTINUE;
1346 if (__path_is_mountpoint(&path)) { 1471 if (__path_is_mountpoint(&path)) {
1347 info->mounted = 1; 1472 info->mounted = 1;
1348 return D_WALK_QUIT; 1473 return D_WALK_QUIT;
1349 } 1474 }
1350 return D_WALK_CONTINUE; 1475 return D_WALK_CONTINUE;
1351 } 1476 }
1352 1477
1353 /** 1478 /**
1354 * path_has_submounts - check for mounts over 1479 * path_has_submounts - check for mounts over a dentry in the
1355 * current namespace. 1480 * current namespace.
1356 * @parent: path to check. 1481 * @parent: path to check.
1357 * 1482 *
1358 * Return true if the parent or its subdirect 1483 * Return true if the parent or its subdirectories contain
1359 * a mount point in the current namespace. 1484 * a mount point in the current namespace.
1360 */ 1485 */
1361 int path_has_submounts(const struct path *par 1486 int path_has_submounts(const struct path *parent)
1362 { 1487 {
1363 struct check_mount data = { .mnt = pa 1488 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1364 1489
1365 read_seqlock_excl(&mount_lock); 1490 read_seqlock_excl(&mount_lock);
1366 d_walk(parent->dentry, &data, path_ch 1491 d_walk(parent->dentry, &data, path_check_mount);
1367 read_sequnlock_excl(&mount_lock); 1492 read_sequnlock_excl(&mount_lock);
1368 1493
1369 return data.mounted; 1494 return data.mounted;
1370 } 1495 }
1371 EXPORT_SYMBOL(path_has_submounts); 1496 EXPORT_SYMBOL(path_has_submounts);
1372 1497
1373 /* 1498 /*
1374 * Called by mount code to set a mountpoint a 1499 * Called by mount code to set a mountpoint and check if the mountpoint is
1375 * reachable (e.g. NFS can unhash a directory 1500 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1376 * subtree can become unreachable). 1501 * subtree can become unreachable).
1377 * 1502 *
1378 * Only one of d_invalidate() and d_set_mount 1503 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1379 * this reason take rename_lock and d_lock on 1504 * this reason take rename_lock and d_lock on dentry and ancestors.
1380 */ 1505 */
1381 int d_set_mounted(struct dentry *dentry) 1506 int d_set_mounted(struct dentry *dentry)
1382 { 1507 {
1383 struct dentry *p; 1508 struct dentry *p;
1384 int ret = -ENOENT; 1509 int ret = -ENOENT;
1385 write_seqlock(&rename_lock); 1510 write_seqlock(&rename_lock);
1386 for (p = dentry->d_parent; !IS_ROOT(p 1511 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1387 /* Need exclusion wrt. d_inva 1512 /* Need exclusion wrt. d_invalidate() */
1388 spin_lock(&p->d_lock); 1513 spin_lock(&p->d_lock);
1389 if (unlikely(d_unhashed(p))) 1514 if (unlikely(d_unhashed(p))) {
1390 spin_unlock(&p->d_loc 1515 spin_unlock(&p->d_lock);
1391 goto out; 1516 goto out;
1392 } 1517 }
1393 spin_unlock(&p->d_lock); 1518 spin_unlock(&p->d_lock);
1394 } 1519 }
1395 spin_lock(&dentry->d_lock); 1520 spin_lock(&dentry->d_lock);
1396 if (!d_unlinked(dentry)) { 1521 if (!d_unlinked(dentry)) {
1397 ret = -EBUSY; 1522 ret = -EBUSY;
1398 if (!d_mountpoint(dentry)) { 1523 if (!d_mountpoint(dentry)) {
1399 dentry->d_flags |= DC 1524 dentry->d_flags |= DCACHE_MOUNTED;
1400 ret = 0; 1525 ret = 0;
1401 } 1526 }
1402 } 1527 }
1403 spin_unlock(&dentry->d_lock); 1528 spin_unlock(&dentry->d_lock);
1404 out: 1529 out:
1405 write_sequnlock(&rename_lock); 1530 write_sequnlock(&rename_lock);
1406 return ret; 1531 return ret;
1407 } 1532 }
1408 1533
1409 /* 1534 /*
1410 * Search the dentry child list of the specif 1535 * Search the dentry child list of the specified parent,
1411 * and move any unused dentries to the end of 1536 * and move any unused dentries to the end of the unused
1412 * list for prune_dcache(). We descend to the 1537 * list for prune_dcache(). We descend to the next level
1413 * whenever the d_children list is non-empty !! 1538 * whenever the d_subdirs list is non-empty and continue
1414 * searching. 1539 * searching.
1415 * 1540 *
1416 * It returns zero iff there are no unused ch 1541 * It returns zero iff there are no unused children,
1417 * otherwise it returns the number of childr 1542 * otherwise it returns the number of children moved to
1418 * the end of the unused list. This may not b 1543 * the end of the unused list. This may not be the total
1419 * number of unused children, because select_ 1544 * number of unused children, because select_parent can
1420 * drop the lock and return early due to late 1545 * drop the lock and return early due to latency
1421 * constraints. 1546 * constraints.
1422 */ 1547 */
1423 1548
1424 struct select_data { 1549 struct select_data {
1425 struct dentry *start; 1550 struct dentry *start;
1426 union { 1551 union {
1427 long found; 1552 long found;
1428 struct dentry *victim; 1553 struct dentry *victim;
1429 }; 1554 };
1430 struct list_head dispose; 1555 struct list_head dispose;
1431 }; 1556 };
1432 1557
1433 static enum d_walk_ret select_collect(void *_ 1558 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1434 { 1559 {
1435 struct select_data *data = _data; 1560 struct select_data *data = _data;
1436 enum d_walk_ret ret = D_WALK_CONTINUE 1561 enum d_walk_ret ret = D_WALK_CONTINUE;
1437 1562
1438 if (data->start == dentry) 1563 if (data->start == dentry)
1439 goto out; 1564 goto out;
1440 1565
1441 if (dentry->d_flags & DCACHE_SHRINK_L 1566 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1442 data->found++; 1567 data->found++;
1443 } else if (!dentry->d_lockref.count) !! 1568 } else {
1444 to_shrink_list(dentry, &data- !! 1569 if (dentry->d_flags & DCACHE_LRU_LIST)
1445 data->found++; !! 1570 d_lru_del(dentry);
1446 } else if (dentry->d_lockref.count < !! 1571 if (!dentry->d_lockref.count) {
1447 data->found++; !! 1572 d_shrink_add(dentry, &data->dispose);
>> 1573 data->found++;
>> 1574 }
1448 } 1575 }
1449 /* 1576 /*
1450 * We can return to the caller if we 1577 * We can return to the caller if we have found some (this
1451 * ensures forward progress). We'll b 1578 * ensures forward progress). We'll be coming back to find
1452 * the rest. 1579 * the rest.
1453 */ 1580 */
1454 if (!list_empty(&data->dispose)) 1581 if (!list_empty(&data->dispose))
1455 ret = need_resched() ? D_WALK 1582 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1456 out: 1583 out:
1457 return ret; 1584 return ret;
1458 } 1585 }
1459 1586
1460 static enum d_walk_ret select_collect2(void * 1587 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1461 { 1588 {
1462 struct select_data *data = _data; 1589 struct select_data *data = _data;
1463 enum d_walk_ret ret = D_WALK_CONTINUE 1590 enum d_walk_ret ret = D_WALK_CONTINUE;
1464 1591
1465 if (data->start == dentry) 1592 if (data->start == dentry)
1466 goto out; 1593 goto out;
1467 1594
1468 if (!dentry->d_lockref.count) { !! 1595 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1469 if (dentry->d_flags & DCACHE_ !! 1596 if (!dentry->d_lockref.count) {
1470 rcu_read_lock(); 1597 rcu_read_lock();
1471 data->victim = dentry 1598 data->victim = dentry;
1472 return D_WALK_QUIT; 1599 return D_WALK_QUIT;
1473 } 1600 }
1474 to_shrink_list(dentry, &data- !! 1601 } else {
>> 1602 if (dentry->d_flags & DCACHE_LRU_LIST)
>> 1603 d_lru_del(dentry);
>> 1604 if (!dentry->d_lockref.count)
>> 1605 d_shrink_add(dentry, &data->dispose);
1475 } 1606 }
1476 /* 1607 /*
1477 * We can return to the caller if we 1608 * We can return to the caller if we have found some (this
1478 * ensures forward progress). We'll b 1609 * ensures forward progress). We'll be coming back to find
1479 * the rest. 1610 * the rest.
1480 */ 1611 */
1481 if (!list_empty(&data->dispose)) 1612 if (!list_empty(&data->dispose))
1482 ret = need_resched() ? D_WALK 1613 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1483 out: 1614 out:
1484 return ret; 1615 return ret;
1485 } 1616 }
1486 1617
1487 /** 1618 /**
1488 * shrink_dcache_parent - prune dcache 1619 * shrink_dcache_parent - prune dcache
1489 * @parent: parent of entries to prune 1620 * @parent: parent of entries to prune
1490 * 1621 *
1491 * Prune the dcache to remove unused children 1622 * Prune the dcache to remove unused children of the parent dentry.
1492 */ 1623 */
1493 void shrink_dcache_parent(struct dentry *pare 1624 void shrink_dcache_parent(struct dentry *parent)
1494 { 1625 {
1495 for (;;) { 1626 for (;;) {
1496 struct select_data data = {.s 1627 struct select_data data = {.start = parent};
1497 1628
1498 INIT_LIST_HEAD(&data.dispose) 1629 INIT_LIST_HEAD(&data.dispose);
1499 d_walk(parent, &data, select_ 1630 d_walk(parent, &data, select_collect);
1500 1631
1501 if (!list_empty(&data.dispose 1632 if (!list_empty(&data.dispose)) {
1502 shrink_dentry_list(&d 1633 shrink_dentry_list(&data.dispose);
1503 continue; 1634 continue;
1504 } 1635 }
1505 1636
1506 cond_resched(); 1637 cond_resched();
1507 if (!data.found) 1638 if (!data.found)
1508 break; 1639 break;
1509 data.victim = NULL; 1640 data.victim = NULL;
1510 d_walk(parent, &data, select_ 1641 d_walk(parent, &data, select_collect2);
1511 if (data.victim) { 1642 if (data.victim) {
>> 1643 struct dentry *parent;
1512 spin_lock(&data.victi 1644 spin_lock(&data.victim->d_lock);
1513 if (!lock_for_kill(da !! 1645 if (!shrink_lock_dentry(data.victim)) {
1514 spin_unlock(& 1646 spin_unlock(&data.victim->d_lock);
1515 rcu_read_unlo 1647 rcu_read_unlock();
1516 } else { 1648 } else {
1517 shrink_kill(d !! 1649 rcu_read_unlock();
>> 1650 parent = data.victim->d_parent;
>> 1651 if (parent != data.victim)
>> 1652 __dput_to_list(parent, &data.dispose);
>> 1653 __dentry_kill(data.victim);
1518 } 1654 }
1519 } 1655 }
1520 if (!list_empty(&data.dispose 1656 if (!list_empty(&data.dispose))
1521 shrink_dentry_list(&d 1657 shrink_dentry_list(&data.dispose);
1522 } 1658 }
1523 } 1659 }
1524 EXPORT_SYMBOL(shrink_dcache_parent); 1660 EXPORT_SYMBOL(shrink_dcache_parent);
1525 1661
1526 static enum d_walk_ret umount_check(void *_da 1662 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1527 { 1663 {
1528 /* it has busy descendents; complain 1664 /* it has busy descendents; complain about those instead */
1529 if (!hlist_empty(&dentry->d_children) !! 1665 if (!list_empty(&dentry->d_subdirs))
1530 return D_WALK_CONTINUE; 1666 return D_WALK_CONTINUE;
1531 1667
1532 /* root with refcount 1 is fine */ 1668 /* root with refcount 1 is fine */
1533 if (dentry == _data && dentry->d_lock 1669 if (dentry == _data && dentry->d_lockref.count == 1)
1534 return D_WALK_CONTINUE; 1670 return D_WALK_CONTINUE;
1535 1671
1536 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} !! 1672 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1537 " still in use (%d) [ 1673 " still in use (%d) [unmount of %s %s]\n",
1538 dentry, 1674 dentry,
1539 dentry->d_inode ? 1675 dentry->d_inode ?
1540 dentry->d_inode->i_ino 1676 dentry->d_inode->i_ino : 0UL,
1541 dentry, 1677 dentry,
1542 dentry->d_lockref.coun 1678 dentry->d_lockref.count,
1543 dentry->d_sb->s_type-> 1679 dentry->d_sb->s_type->name,
1544 dentry->d_sb->s_id); 1680 dentry->d_sb->s_id);
>> 1681 WARN_ON(1);
1545 return D_WALK_CONTINUE; 1682 return D_WALK_CONTINUE;
1546 } 1683 }
1547 1684
1548 static void do_one_tree(struct dentry *dentry 1685 static void do_one_tree(struct dentry *dentry)
1549 { 1686 {
1550 shrink_dcache_parent(dentry); 1687 shrink_dcache_parent(dentry);
1551 d_walk(dentry, dentry, umount_check); 1688 d_walk(dentry, dentry, umount_check);
1552 d_drop(dentry); 1689 d_drop(dentry);
1553 dput(dentry); 1690 dput(dentry);
1554 } 1691 }
1555 1692
1556 /* 1693 /*
1557 * destroy the dentries attached to a superbl 1694 * destroy the dentries attached to a superblock on unmounting
1558 */ 1695 */
1559 void shrink_dcache_for_umount(struct super_bl 1696 void shrink_dcache_for_umount(struct super_block *sb)
1560 { 1697 {
1561 struct dentry *dentry; 1698 struct dentry *dentry;
1562 1699
1563 rwsem_assert_held_write(&sb->s_umount !! 1700 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1564 1701
1565 dentry = sb->s_root; 1702 dentry = sb->s_root;
1566 sb->s_root = NULL; 1703 sb->s_root = NULL;
1567 do_one_tree(dentry); 1704 do_one_tree(dentry);
1568 1705
1569 while (!hlist_bl_empty(&sb->s_roots)) 1706 while (!hlist_bl_empty(&sb->s_roots)) {
1570 dentry = dget(hlist_bl_entry( 1707 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1571 do_one_tree(dentry); 1708 do_one_tree(dentry);
1572 } 1709 }
1573 } 1710 }
1574 1711
1575 static enum d_walk_ret find_submount(void *_d 1712 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1576 { 1713 {
1577 struct dentry **victim = _data; 1714 struct dentry **victim = _data;
1578 if (d_mountpoint(dentry)) { 1715 if (d_mountpoint(dentry)) {
1579 *victim = dget_dlock(dentry); !! 1716 __dget_dlock(dentry);
>> 1717 *victim = dentry;
1580 return D_WALK_QUIT; 1718 return D_WALK_QUIT;
1581 } 1719 }
1582 return D_WALK_CONTINUE; 1720 return D_WALK_CONTINUE;
1583 } 1721 }
1584 1722
1585 /** 1723 /**
1586 * d_invalidate - detach submounts, prune dca 1724 * d_invalidate - detach submounts, prune dcache, and drop
1587 * @dentry: dentry to invalidate (aka detach, 1725 * @dentry: dentry to invalidate (aka detach, prune and drop)
1588 */ 1726 */
1589 void d_invalidate(struct dentry *dentry) 1727 void d_invalidate(struct dentry *dentry)
1590 { 1728 {
1591 bool had_submounts = false; 1729 bool had_submounts = false;
1592 spin_lock(&dentry->d_lock); 1730 spin_lock(&dentry->d_lock);
1593 if (d_unhashed(dentry)) { 1731 if (d_unhashed(dentry)) {
1594 spin_unlock(&dentry->d_lock); 1732 spin_unlock(&dentry->d_lock);
1595 return; 1733 return;
1596 } 1734 }
1597 __d_drop(dentry); 1735 __d_drop(dentry);
1598 spin_unlock(&dentry->d_lock); 1736 spin_unlock(&dentry->d_lock);
1599 1737
1600 /* Negative dentries can be dropped w 1738 /* Negative dentries can be dropped without further checks */
1601 if (!dentry->d_inode) 1739 if (!dentry->d_inode)
1602 return; 1740 return;
1603 1741
1604 shrink_dcache_parent(dentry); 1742 shrink_dcache_parent(dentry);
1605 for (;;) { 1743 for (;;) {
1606 struct dentry *victim = NULL; 1744 struct dentry *victim = NULL;
1607 d_walk(dentry, &victim, find_ 1745 d_walk(dentry, &victim, find_submount);
1608 if (!victim) { 1746 if (!victim) {
1609 if (had_submounts) 1747 if (had_submounts)
1610 shrink_dcache 1748 shrink_dcache_parent(dentry);
1611 return; 1749 return;
1612 } 1750 }
1613 had_submounts = true; 1751 had_submounts = true;
1614 detach_mounts(victim); 1752 detach_mounts(victim);
1615 dput(victim); 1753 dput(victim);
1616 } 1754 }
1617 } 1755 }
1618 EXPORT_SYMBOL(d_invalidate); 1756 EXPORT_SYMBOL(d_invalidate);
1619 1757
1620 /** 1758 /**
1621 * __d_alloc - allocate a dcache ent 1759 * __d_alloc - allocate a dcache entry
1622 * @sb: filesystem it will belong to 1760 * @sb: filesystem it will belong to
1623 * @name: qstr of the name 1761 * @name: qstr of the name
1624 * 1762 *
1625 * Allocates a dentry. It returns %NULL if th 1763 * Allocates a dentry. It returns %NULL if there is insufficient memory
1626 * available. On a success the dentry is retu 1764 * available. On a success the dentry is returned. The name passed in is
1627 * copied and the copy passed in may be reuse 1765 * copied and the copy passed in may be reused after this call.
1628 */ 1766 */
1629 1767
1630 static struct dentry *__d_alloc(struct super_ 1768 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1631 { 1769 {
1632 struct dentry *dentry; 1770 struct dentry *dentry;
1633 char *dname; 1771 char *dname;
1634 int err; 1772 int err;
1635 1773
1636 dentry = kmem_cache_alloc_lru(dentry_ 1774 dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru,
1637 GFP_KER 1775 GFP_KERNEL);
1638 if (!dentry) 1776 if (!dentry)
1639 return NULL; 1777 return NULL;
1640 1778
1641 /* 1779 /*
1642 * We guarantee that the inline name 1780 * We guarantee that the inline name is always NUL-terminated.
1643 * This way the memcpy() done by the 1781 * This way the memcpy() done by the name switching in rename
1644 * will still always have a NUL at th 1782 * will still always have a NUL at the end, even if we might
1645 * be overwriting an internal NUL cha 1783 * be overwriting an internal NUL character
1646 */ 1784 */
1647 dentry->d_iname[DNAME_INLINE_LEN-1] = 1785 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1648 if (unlikely(!name)) { 1786 if (unlikely(!name)) {
1649 name = &slash_name; 1787 name = &slash_name;
1650 dname = dentry->d_iname; 1788 dname = dentry->d_iname;
1651 } else if (name->len > DNAME_INLINE_L 1789 } else if (name->len > DNAME_INLINE_LEN-1) {
1652 size_t size = offsetof(struct 1790 size_t size = offsetof(struct external_name, name[1]);
1653 struct external_name *p = kma 1791 struct external_name *p = kmalloc(size + name->len,
1654 1792 GFP_KERNEL_ACCOUNT |
1655 1793 __GFP_RECLAIMABLE);
1656 if (!p) { 1794 if (!p) {
1657 kmem_cache_free(dentr 1795 kmem_cache_free(dentry_cache, dentry);
1658 return NULL; 1796 return NULL;
1659 } 1797 }
1660 atomic_set(&p->u.count, 1); 1798 atomic_set(&p->u.count, 1);
1661 dname = p->name; 1799 dname = p->name;
1662 } else { 1800 } else {
1663 dname = dentry->d_iname; 1801 dname = dentry->d_iname;
1664 } 1802 }
1665 1803
1666 dentry->d_name.len = name->len; 1804 dentry->d_name.len = name->len;
1667 dentry->d_name.hash = name->hash; 1805 dentry->d_name.hash = name->hash;
1668 memcpy(dname, name->name, name->len); 1806 memcpy(dname, name->name, name->len);
1669 dname[name->len] = 0; 1807 dname[name->len] = 0;
1670 1808
1671 /* Make sure we always see the termin 1809 /* Make sure we always see the terminating NUL character */
1672 smp_store_release(&dentry->d_name.nam 1810 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1673 1811
1674 dentry->d_lockref.count = 1; 1812 dentry->d_lockref.count = 1;
1675 dentry->d_flags = 0; 1813 dentry->d_flags = 0;
1676 spin_lock_init(&dentry->d_lock); 1814 spin_lock_init(&dentry->d_lock);
1677 seqcount_spinlock_init(&dentry->d_seq 1815 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1678 dentry->d_inode = NULL; 1816 dentry->d_inode = NULL;
1679 dentry->d_parent = dentry; 1817 dentry->d_parent = dentry;
1680 dentry->d_sb = sb; 1818 dentry->d_sb = sb;
1681 dentry->d_op = NULL; 1819 dentry->d_op = NULL;
1682 dentry->d_fsdata = NULL; 1820 dentry->d_fsdata = NULL;
1683 INIT_HLIST_BL_NODE(&dentry->d_hash); 1821 INIT_HLIST_BL_NODE(&dentry->d_hash);
1684 INIT_LIST_HEAD(&dentry->d_lru); 1822 INIT_LIST_HEAD(&dentry->d_lru);
1685 INIT_HLIST_HEAD(&dentry->d_children); !! 1823 INIT_LIST_HEAD(&dentry->d_subdirs);
1686 INIT_HLIST_NODE(&dentry->d_u.d_alias) 1824 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1687 INIT_HLIST_NODE(&dentry->d_sib); !! 1825 INIT_LIST_HEAD(&dentry->d_child);
1688 d_set_d_op(dentry, dentry->d_sb->s_d_ 1826 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1689 1827
1690 if (dentry->d_op && dentry->d_op->d_i 1828 if (dentry->d_op && dentry->d_op->d_init) {
1691 err = dentry->d_op->d_init(de 1829 err = dentry->d_op->d_init(dentry);
1692 if (err) { 1830 if (err) {
1693 if (dname_external(de 1831 if (dname_external(dentry))
1694 kfree(externa 1832 kfree(external_name(dentry));
1695 kmem_cache_free(dentr 1833 kmem_cache_free(dentry_cache, dentry);
1696 return NULL; 1834 return NULL;
1697 } 1835 }
1698 } 1836 }
1699 1837
1700 this_cpu_inc(nr_dentry); 1838 this_cpu_inc(nr_dentry);
1701 1839
1702 return dentry; 1840 return dentry;
1703 } 1841 }
1704 1842
1705 /** 1843 /**
1706 * d_alloc - allocate a dcache ent 1844 * d_alloc - allocate a dcache entry
1707 * @parent: parent of entry to allocate 1845 * @parent: parent of entry to allocate
1708 * @name: qstr of the name 1846 * @name: qstr of the name
1709 * 1847 *
1710 * Allocates a dentry. It returns %NULL if th 1848 * Allocates a dentry. It returns %NULL if there is insufficient memory
1711 * available. On a success the dentry is retu 1849 * available. On a success the dentry is returned. The name passed in is
1712 * copied and the copy passed in may be reuse 1850 * copied and the copy passed in may be reused after this call.
1713 */ 1851 */
1714 struct dentry *d_alloc(struct dentry * parent 1852 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1715 { 1853 {
1716 struct dentry *dentry = __d_alloc(par 1854 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1717 if (!dentry) 1855 if (!dentry)
1718 return NULL; 1856 return NULL;
1719 spin_lock(&parent->d_lock); 1857 spin_lock(&parent->d_lock);
1720 /* 1858 /*
1721 * don't need child lock because it i 1859 * don't need child lock because it is not subject
1722 * to concurrency here 1860 * to concurrency here
1723 */ 1861 */
1724 dentry->d_parent = dget_dlock(parent) !! 1862 __dget_dlock(parent);
1725 hlist_add_head(&dentry->d_sib, &paren !! 1863 dentry->d_parent = parent;
>> 1864 list_add(&dentry->d_child, &parent->d_subdirs);
1726 spin_unlock(&parent->d_lock); 1865 spin_unlock(&parent->d_lock);
1727 1866
1728 return dentry; 1867 return dentry;
1729 } 1868 }
1730 EXPORT_SYMBOL(d_alloc); 1869 EXPORT_SYMBOL(d_alloc);
1731 1870
1732 struct dentry *d_alloc_anon(struct super_bloc 1871 struct dentry *d_alloc_anon(struct super_block *sb)
1733 { 1872 {
1734 return __d_alloc(sb, NULL); 1873 return __d_alloc(sb, NULL);
1735 } 1874 }
1736 EXPORT_SYMBOL(d_alloc_anon); 1875 EXPORT_SYMBOL(d_alloc_anon);
1737 1876
1738 struct dentry *d_alloc_cursor(struct dentry * 1877 struct dentry *d_alloc_cursor(struct dentry * parent)
1739 { 1878 {
1740 struct dentry *dentry = d_alloc_anon( 1879 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1741 if (dentry) { 1880 if (dentry) {
1742 dentry->d_flags |= DCACHE_DEN 1881 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1743 dentry->d_parent = dget(paren 1882 dentry->d_parent = dget(parent);
1744 } 1883 }
1745 return dentry; 1884 return dentry;
1746 } 1885 }
1747 1886
1748 /** 1887 /**
1749 * d_alloc_pseudo - allocate a dentry (for lo 1888 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1750 * @sb: the superblock 1889 * @sb: the superblock
1751 * @name: qstr of the name 1890 * @name: qstr of the name
1752 * 1891 *
1753 * For a filesystem that just pins its dentri 1892 * For a filesystem that just pins its dentries in memory and never
1754 * performs lookups at all, return an unhashe 1893 * performs lookups at all, return an unhashed IS_ROOT dentry.
1755 * This is used for pipes, sockets et.al. - t 1894 * This is used for pipes, sockets et.al. - the stuff that should
1756 * never be anyone's children or parents. Un 1895 * never be anyone's children or parents. Unlike all other
1757 * dentries, these will not have RCU delay be 1896 * dentries, these will not have RCU delay between dropping the
1758 * last reference and freeing them. 1897 * last reference and freeing them.
1759 * 1898 *
1760 * The only user is alloc_file_pseudo() and t 1899 * The only user is alloc_file_pseudo() and that's what should
1761 * be considered a public interface. Don't u 1900 * be considered a public interface. Don't use directly.
1762 */ 1901 */
1763 struct dentry *d_alloc_pseudo(struct super_bl 1902 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1764 { 1903 {
1765 static const struct dentry_operations <<
1766 .d_dname = simple_dname <<
1767 }; <<
1768 struct dentry *dentry = __d_alloc(sb, 1904 struct dentry *dentry = __d_alloc(sb, name);
1769 if (likely(dentry)) { !! 1905 if (likely(dentry))
1770 dentry->d_flags |= DCACHE_NOR 1906 dentry->d_flags |= DCACHE_NORCU;
1771 if (!sb->s_d_op) <<
1772 d_set_d_op(dentry, &a <<
1773 } <<
1774 return dentry; 1907 return dentry;
1775 } 1908 }
1776 1909
1777 struct dentry *d_alloc_name(struct dentry *pa 1910 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1778 { 1911 {
1779 struct qstr q; 1912 struct qstr q;
1780 1913
1781 q.name = name; 1914 q.name = name;
1782 q.hash_len = hashlen_string(parent, n 1915 q.hash_len = hashlen_string(parent, name);
1783 return d_alloc(parent, &q); 1916 return d_alloc(parent, &q);
1784 } 1917 }
1785 EXPORT_SYMBOL(d_alloc_name); 1918 EXPORT_SYMBOL(d_alloc_name);
1786 1919
1787 void d_set_d_op(struct dentry *dentry, const 1920 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1788 { 1921 {
1789 WARN_ON_ONCE(dentry->d_op); 1922 WARN_ON_ONCE(dentry->d_op);
1790 WARN_ON_ONCE(dentry->d_flags & (DCACH 1923 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1791 DCACHE_OP_COM 1924 DCACHE_OP_COMPARE |
1792 DCACHE_OP_REV 1925 DCACHE_OP_REVALIDATE |
1793 DCACHE_OP_WEA 1926 DCACHE_OP_WEAK_REVALIDATE |
1794 DCACHE_OP_DEL 1927 DCACHE_OP_DELETE |
1795 DCACHE_OP_REA 1928 DCACHE_OP_REAL));
1796 dentry->d_op = op; 1929 dentry->d_op = op;
1797 if (!op) 1930 if (!op)
1798 return; 1931 return;
1799 if (op->d_hash) 1932 if (op->d_hash)
1800 dentry->d_flags |= DCACHE_OP_ 1933 dentry->d_flags |= DCACHE_OP_HASH;
1801 if (op->d_compare) 1934 if (op->d_compare)
1802 dentry->d_flags |= DCACHE_OP_ 1935 dentry->d_flags |= DCACHE_OP_COMPARE;
1803 if (op->d_revalidate) 1936 if (op->d_revalidate)
1804 dentry->d_flags |= DCACHE_OP_ 1937 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1805 if (op->d_weak_revalidate) 1938 if (op->d_weak_revalidate)
1806 dentry->d_flags |= DCACHE_OP_ 1939 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1807 if (op->d_delete) 1940 if (op->d_delete)
1808 dentry->d_flags |= DCACHE_OP_ 1941 dentry->d_flags |= DCACHE_OP_DELETE;
1809 if (op->d_prune) 1942 if (op->d_prune)
1810 dentry->d_flags |= DCACHE_OP_ 1943 dentry->d_flags |= DCACHE_OP_PRUNE;
1811 if (op->d_real) 1944 if (op->d_real)
1812 dentry->d_flags |= DCACHE_OP_ 1945 dentry->d_flags |= DCACHE_OP_REAL;
1813 1946
1814 } 1947 }
1815 EXPORT_SYMBOL(d_set_d_op); 1948 EXPORT_SYMBOL(d_set_d_op);
1816 1949
>> 1950
>> 1951 /*
>> 1952 * d_set_fallthru - Mark a dentry as falling through to a lower layer
>> 1953 * @dentry - The dentry to mark
>> 1954 *
>> 1955 * Mark a dentry as falling through to the lower layer (as set with
>> 1956 * d_pin_lower()). This flag may be recorded on the medium.
>> 1957 */
>> 1958 void d_set_fallthru(struct dentry *dentry)
>> 1959 {
>> 1960 spin_lock(&dentry->d_lock);
>> 1961 dentry->d_flags |= DCACHE_FALLTHRU;
>> 1962 spin_unlock(&dentry->d_lock);
>> 1963 }
>> 1964 EXPORT_SYMBOL(d_set_fallthru);
>> 1965
1817 static unsigned d_flags_for_inode(struct inod 1966 static unsigned d_flags_for_inode(struct inode *inode)
1818 { 1967 {
1819 unsigned add_flags = DCACHE_REGULAR_T 1968 unsigned add_flags = DCACHE_REGULAR_TYPE;
1820 1969
1821 if (!inode) 1970 if (!inode)
1822 return DCACHE_MISS_TYPE; 1971 return DCACHE_MISS_TYPE;
1823 1972
1824 if (S_ISDIR(inode->i_mode)) { 1973 if (S_ISDIR(inode->i_mode)) {
1825 add_flags = DCACHE_DIRECTORY_ 1974 add_flags = DCACHE_DIRECTORY_TYPE;
1826 if (unlikely(!(inode->i_opfla 1975 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1827 if (unlikely(!inode-> 1976 if (unlikely(!inode->i_op->lookup))
1828 add_flags = D 1977 add_flags = DCACHE_AUTODIR_TYPE;
1829 else 1978 else
1830 inode->i_opfl 1979 inode->i_opflags |= IOP_LOOKUP;
1831 } 1980 }
1832 goto type_determined; 1981 goto type_determined;
1833 } 1982 }
1834 1983
1835 if (unlikely(!(inode->i_opflags & IOP 1984 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1836 if (unlikely(inode->i_op->get 1985 if (unlikely(inode->i_op->get_link)) {
1837 add_flags = DCACHE_SY 1986 add_flags = DCACHE_SYMLINK_TYPE;
1838 goto type_determined; 1987 goto type_determined;
1839 } 1988 }
1840 inode->i_opflags |= IOP_NOFOL 1989 inode->i_opflags |= IOP_NOFOLLOW;
1841 } 1990 }
1842 1991
1843 if (unlikely(!S_ISREG(inode->i_mode)) 1992 if (unlikely(!S_ISREG(inode->i_mode)))
1844 add_flags = DCACHE_SPECIAL_TY 1993 add_flags = DCACHE_SPECIAL_TYPE;
1845 1994
1846 type_determined: 1995 type_determined:
1847 if (unlikely(IS_AUTOMOUNT(inode))) 1996 if (unlikely(IS_AUTOMOUNT(inode)))
1848 add_flags |= DCACHE_NEED_AUTO 1997 add_flags |= DCACHE_NEED_AUTOMOUNT;
1849 return add_flags; 1998 return add_flags;
1850 } 1999 }
1851 2000
1852 static void __d_instantiate(struct dentry *de 2001 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1853 { 2002 {
1854 unsigned add_flags = d_flags_for_inod 2003 unsigned add_flags = d_flags_for_inode(inode);
1855 WARN_ON(d_in_lookup(dentry)); 2004 WARN_ON(d_in_lookup(dentry));
1856 2005
1857 spin_lock(&dentry->d_lock); 2006 spin_lock(&dentry->d_lock);
1858 /* 2007 /*
1859 * The negative counter only tracks d 2008 * The negative counter only tracks dentries on the LRU. Don't dec if
1860 * d_lru is on another list. 2009 * d_lru is on another list.
1861 */ 2010 */
1862 if ((dentry->d_flags & 2011 if ((dentry->d_flags &
1863 (DCACHE_LRU_LIST|DCACHE_SHRINK_L 2012 (DCACHE_LRU_LIST|DCACHE_SHRINK_LIST)) == DCACHE_LRU_LIST)
1864 this_cpu_dec(nr_dentry_negati 2013 this_cpu_dec(nr_dentry_negative);
1865 hlist_add_head(&dentry->d_u.d_alias, 2014 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1866 raw_write_seqcount_begin(&dentry->d_s 2015 raw_write_seqcount_begin(&dentry->d_seq);
1867 __d_set_inode_and_type(dentry, inode, 2016 __d_set_inode_and_type(dentry, inode, add_flags);
1868 raw_write_seqcount_end(&dentry->d_seq 2017 raw_write_seqcount_end(&dentry->d_seq);
1869 fsnotify_update_flags(dentry); 2018 fsnotify_update_flags(dentry);
1870 spin_unlock(&dentry->d_lock); 2019 spin_unlock(&dentry->d_lock);
1871 } 2020 }
1872 2021
1873 /** 2022 /**
1874 * d_instantiate - fill in inode information 2023 * d_instantiate - fill in inode information for a dentry
1875 * @entry: dentry to complete 2024 * @entry: dentry to complete
1876 * @inode: inode to attach to this dentry 2025 * @inode: inode to attach to this dentry
1877 * 2026 *
1878 * Fill in inode information in the entry. 2027 * Fill in inode information in the entry.
1879 * 2028 *
1880 * This turns negative dentries into producti 2029 * This turns negative dentries into productive full members
1881 * of society. 2030 * of society.
1882 * 2031 *
1883 * NOTE! This assumes that the inode count ha 2032 * NOTE! This assumes that the inode count has been incremented
1884 * (or otherwise set) by the caller to indica 2033 * (or otherwise set) by the caller to indicate that it is now
1885 * in use by the dcache. 2034 * in use by the dcache.
1886 */ 2035 */
1887 2036
1888 void d_instantiate(struct dentry *entry, stru 2037 void d_instantiate(struct dentry *entry, struct inode * inode)
1889 { 2038 {
1890 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 2039 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1891 if (inode) { 2040 if (inode) {
1892 security_d_instantiate(entry, 2041 security_d_instantiate(entry, inode);
1893 spin_lock(&inode->i_lock); 2042 spin_lock(&inode->i_lock);
1894 __d_instantiate(entry, inode) 2043 __d_instantiate(entry, inode);
1895 spin_unlock(&inode->i_lock); 2044 spin_unlock(&inode->i_lock);
1896 } 2045 }
1897 } 2046 }
1898 EXPORT_SYMBOL(d_instantiate); 2047 EXPORT_SYMBOL(d_instantiate);
1899 2048
1900 /* 2049 /*
1901 * This should be equivalent to d_instantiate 2050 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1902 * with lockdep-related part of unlock_new_in 2051 * with lockdep-related part of unlock_new_inode() done before
1903 * anything else. Use that instead of open-c 2052 * anything else. Use that instead of open-coding d_instantiate()/
1904 * unlock_new_inode() combinations. 2053 * unlock_new_inode() combinations.
1905 */ 2054 */
1906 void d_instantiate_new(struct dentry *entry, 2055 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1907 { 2056 {
1908 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 2057 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1909 BUG_ON(!inode); 2058 BUG_ON(!inode);
1910 lockdep_annotate_inode_mutex_key(inod 2059 lockdep_annotate_inode_mutex_key(inode);
1911 security_d_instantiate(entry, inode); 2060 security_d_instantiate(entry, inode);
1912 spin_lock(&inode->i_lock); 2061 spin_lock(&inode->i_lock);
1913 __d_instantiate(entry, inode); 2062 __d_instantiate(entry, inode);
1914 WARN_ON(!(inode->i_state & I_NEW)); 2063 WARN_ON(!(inode->i_state & I_NEW));
1915 inode->i_state &= ~I_NEW & ~I_CREATIN 2064 inode->i_state &= ~I_NEW & ~I_CREATING;
1916 /* <<
1917 * Pairs with the barrier in prepare_ <<
1918 * ___wait_var_event() either sees th <<
1919 * waitqueue_active() check in wake_u <<
1920 */ <<
1921 smp_mb(); 2065 smp_mb();
1922 inode_wake_up_bit(inode, __I_NEW); !! 2066 wake_up_bit(&inode->i_state, __I_NEW);
1923 spin_unlock(&inode->i_lock); 2067 spin_unlock(&inode->i_lock);
1924 } 2068 }
1925 EXPORT_SYMBOL(d_instantiate_new); 2069 EXPORT_SYMBOL(d_instantiate_new);
1926 2070
1927 struct dentry *d_make_root(struct inode *root 2071 struct dentry *d_make_root(struct inode *root_inode)
1928 { 2072 {
1929 struct dentry *res = NULL; 2073 struct dentry *res = NULL;
1930 2074
1931 if (root_inode) { 2075 if (root_inode) {
1932 res = d_alloc_anon(root_inode 2076 res = d_alloc_anon(root_inode->i_sb);
1933 if (res) 2077 if (res)
1934 d_instantiate(res, ro 2078 d_instantiate(res, root_inode);
1935 else 2079 else
1936 iput(root_inode); 2080 iput(root_inode);
1937 } 2081 }
1938 return res; 2082 return res;
1939 } 2083 }
1940 EXPORT_SYMBOL(d_make_root); 2084 EXPORT_SYMBOL(d_make_root);
1941 2085
>> 2086 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
>> 2087 struct inode *inode,
>> 2088 bool disconnected)
>> 2089 {
>> 2090 struct dentry *res;
>> 2091 unsigned add_flags;
>> 2092
>> 2093 security_d_instantiate(dentry, inode);
>> 2094 spin_lock(&inode->i_lock);
>> 2095 res = __d_find_any_alias(inode);
>> 2096 if (res) {
>> 2097 spin_unlock(&inode->i_lock);
>> 2098 dput(dentry);
>> 2099 goto out_iput;
>> 2100 }
>> 2101
>> 2102 /* attach a disconnected dentry */
>> 2103 add_flags = d_flags_for_inode(inode);
>> 2104
>> 2105 if (disconnected)
>> 2106 add_flags |= DCACHE_DISCONNECTED;
>> 2107
>> 2108 spin_lock(&dentry->d_lock);
>> 2109 __d_set_inode_and_type(dentry, inode, add_flags);
>> 2110 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
>> 2111 if (!disconnected) {
>> 2112 hlist_bl_lock(&dentry->d_sb->s_roots);
>> 2113 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
>> 2114 hlist_bl_unlock(&dentry->d_sb->s_roots);
>> 2115 }
>> 2116 spin_unlock(&dentry->d_lock);
>> 2117 spin_unlock(&inode->i_lock);
>> 2118
>> 2119 return dentry;
>> 2120
>> 2121 out_iput:
>> 2122 iput(inode);
>> 2123 return res;
>> 2124 }
>> 2125
>> 2126 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
>> 2127 {
>> 2128 return __d_instantiate_anon(dentry, inode, true);
>> 2129 }
>> 2130 EXPORT_SYMBOL(d_instantiate_anon);
>> 2131
1942 static struct dentry *__d_obtain_alias(struct 2132 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
1943 { 2133 {
1944 struct super_block *sb; !! 2134 struct dentry *tmp;
1945 struct dentry *new, *res; !! 2135 struct dentry *res;
1946 2136
1947 if (!inode) 2137 if (!inode)
1948 return ERR_PTR(-ESTALE); 2138 return ERR_PTR(-ESTALE);
1949 if (IS_ERR(inode)) 2139 if (IS_ERR(inode))
1950 return ERR_CAST(inode); 2140 return ERR_CAST(inode);
1951 2141
1952 sb = inode->i_sb; !! 2142 res = d_find_any_alias(inode);
1953 <<
1954 res = d_find_any_alias(inode); /* exi <<
1955 if (res) 2143 if (res)
1956 goto out; !! 2144 goto out_iput;
1957 2145
1958 new = d_alloc_anon(sb); !! 2146 tmp = d_alloc_anon(inode->i_sb);
1959 if (!new) { !! 2147 if (!tmp) {
1960 res = ERR_PTR(-ENOMEM); 2148 res = ERR_PTR(-ENOMEM);
1961 goto out; !! 2149 goto out_iput;
1962 } 2150 }
1963 2151
1964 security_d_instantiate(new, inode); !! 2152 return __d_instantiate_anon(tmp, inode, disconnected);
1965 spin_lock(&inode->i_lock); <<
1966 res = __d_find_any_alias(inode); /* r <<
1967 if (likely(!res)) { /* still no alias <<
1968 unsigned add_flags = d_flags_ <<
1969 <<
1970 if (disconnected) <<
1971 add_flags |= DCACHE_D <<
1972 2153
1973 spin_lock(&new->d_lock); !! 2154 out_iput:
1974 __d_set_inode_and_type(new, i <<
1975 hlist_add_head(&new->d_u.d_al <<
1976 if (!disconnected) { <<
1977 hlist_bl_lock(&sb->s_ <<
1978 hlist_bl_add_head(&ne <<
1979 hlist_bl_unlock(&sb-> <<
1980 } <<
1981 spin_unlock(&new->d_lock); <<
1982 spin_unlock(&inode->i_lock); <<
1983 inode = NULL; /* consumed by <<
1984 res = new; <<
1985 } else { <<
1986 spin_unlock(&inode->i_lock); <<
1987 dput(new); <<
1988 } <<
1989 <<
1990 out: <<
1991 iput(inode); 2155 iput(inode);
1992 return res; 2156 return res;
1993 } 2157 }
1994 2158
1995 /** 2159 /**
1996 * d_obtain_alias - find or allocate a DISCON 2160 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1997 * @inode: inode to allocate the dentry for 2161 * @inode: inode to allocate the dentry for
1998 * 2162 *
1999 * Obtain a dentry for an inode resulting fro 2163 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2000 * similar open by handle operations. The re 2164 * similar open by handle operations. The returned dentry may be anonymous,
2001 * or may have a full name (if the inode was 2165 * or may have a full name (if the inode was already in the cache).
2002 * 2166 *
2003 * When called on a directory inode, we must 2167 * When called on a directory inode, we must ensure that the inode only ever
2004 * has one dentry. If a dentry is found, tha 2168 * has one dentry. If a dentry is found, that is returned instead of
2005 * allocating a new one. 2169 * allocating a new one.
2006 * 2170 *
2007 * On successful return, the reference to the 2171 * On successful return, the reference to the inode has been transferred
2008 * to the dentry. In case of an error the re 2172 * 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 2173 * 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 2174 * be passed in and the error will be propagated to the return value,
2011 * with a %NULL @inode replaced by ERR_PTR(-E 2175 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2012 */ 2176 */
2013 struct dentry *d_obtain_alias(struct inode *i 2177 struct dentry *d_obtain_alias(struct inode *inode)
2014 { 2178 {
2015 return __d_obtain_alias(inode, true); 2179 return __d_obtain_alias(inode, true);
2016 } 2180 }
2017 EXPORT_SYMBOL(d_obtain_alias); 2181 EXPORT_SYMBOL(d_obtain_alias);
2018 2182
2019 /** 2183 /**
2020 * d_obtain_root - find or allocate a dentry 2184 * d_obtain_root - find or allocate a dentry for a given inode
2021 * @inode: inode to allocate the dentry for 2185 * @inode: inode to allocate the dentry for
2022 * 2186 *
2023 * Obtain an IS_ROOT dentry for the root of a 2187 * Obtain an IS_ROOT dentry for the root of a filesystem.
2024 * 2188 *
2025 * We must ensure that directory inodes only 2189 * We must ensure that directory inodes only ever have one dentry. If a
2026 * dentry is found, that is returned instead 2190 * dentry is found, that is returned instead of allocating a new one.
2027 * 2191 *
2028 * On successful return, the reference to the 2192 * On successful return, the reference to the inode has been transferred
2029 * to the dentry. In case of an error the re 2193 * to the dentry. In case of an error the reference on the inode is
2030 * released. A %NULL or IS_ERR inode may be 2194 * released. A %NULL or IS_ERR inode may be passed in and will be the
2031 * error will be propagate to the return valu 2195 * error will be propagate to the return value, with a %NULL @inode
2032 * replaced by ERR_PTR(-ESTALE). 2196 * replaced by ERR_PTR(-ESTALE).
2033 */ 2197 */
2034 struct dentry *d_obtain_root(struct inode *in 2198 struct dentry *d_obtain_root(struct inode *inode)
2035 { 2199 {
2036 return __d_obtain_alias(inode, false) 2200 return __d_obtain_alias(inode, false);
2037 } 2201 }
2038 EXPORT_SYMBOL(d_obtain_root); 2202 EXPORT_SYMBOL(d_obtain_root);
2039 2203
2040 /** 2204 /**
2041 * d_add_ci - lookup or allocate new dentry w 2205 * d_add_ci - lookup or allocate new dentry with case-exact name
2042 * @inode: the inode case-insensitive lookup 2206 * @inode: the inode case-insensitive lookup has found
2043 * @dentry: the negative dentry that was pass 2207 * @dentry: the negative dentry that was passed to the parent's lookup func
2044 * @name: the case-exact name to be associa 2208 * @name: the case-exact name to be associated with the returned dentry
2045 * 2209 *
2046 * This is to avoid filling the dcache with c 2210 * This is to avoid filling the dcache with case-insensitive names to the
2047 * same inode, only the actual correct case i 2211 * same inode, only the actual correct case is stored in the dcache for
2048 * case-insensitive filesystems. 2212 * case-insensitive filesystems.
2049 * 2213 *
2050 * For a case-insensitive lookup match and if 2214 * For a case-insensitive lookup match and if the case-exact dentry
2051 * already exists in the dcache, use it and r 2215 * already exists in the dcache, use it and return it.
2052 * 2216 *
2053 * If no entry exists with the exact case nam 2217 * If no entry exists with the exact case name, allocate new dentry with
2054 * the exact case, and return the spliced ent 2218 * the exact case, and return the spliced entry.
2055 */ 2219 */
2056 struct dentry *d_add_ci(struct dentry *dentry 2220 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2057 struct qstr *name) 2221 struct qstr *name)
2058 { 2222 {
2059 struct dentry *found, *res; 2223 struct dentry *found, *res;
2060 2224
2061 /* 2225 /*
2062 * First check if a dentry matching t 2226 * First check if a dentry matching the name already exists,
2063 * if not go ahead and create it now. 2227 * if not go ahead and create it now.
2064 */ 2228 */
2065 found = d_hash_and_lookup(dentry->d_p 2229 found = d_hash_and_lookup(dentry->d_parent, name);
2066 if (found) { 2230 if (found) {
2067 iput(inode); 2231 iput(inode);
2068 return found; 2232 return found;
2069 } 2233 }
2070 if (d_in_lookup(dentry)) { 2234 if (d_in_lookup(dentry)) {
2071 found = d_alloc_parallel(dent 2235 found = d_alloc_parallel(dentry->d_parent, name,
2072 dentr 2236 dentry->d_wait);
2073 if (IS_ERR(found) || !d_in_lo 2237 if (IS_ERR(found) || !d_in_lookup(found)) {
2074 iput(inode); 2238 iput(inode);
2075 return found; 2239 return found;
2076 } 2240 }
2077 } else { 2241 } else {
2078 found = d_alloc(dentry->d_par 2242 found = d_alloc(dentry->d_parent, name);
2079 if (!found) { 2243 if (!found) {
2080 iput(inode); 2244 iput(inode);
2081 return ERR_PTR(-ENOME 2245 return ERR_PTR(-ENOMEM);
2082 } 2246 }
2083 } 2247 }
2084 res = d_splice_alias(inode, found); 2248 res = d_splice_alias(inode, found);
2085 if (res) { 2249 if (res) {
2086 d_lookup_done(found); 2250 d_lookup_done(found);
2087 dput(found); 2251 dput(found);
2088 return res; 2252 return res;
2089 } 2253 }
2090 return found; 2254 return found;
2091 } 2255 }
2092 EXPORT_SYMBOL(d_add_ci); 2256 EXPORT_SYMBOL(d_add_ci);
2093 2257
2094 /** 2258 /**
2095 * d_same_name - compare dentry name with cas 2259 * d_same_name - compare dentry name with case-exact name
2096 * @parent: parent dentry 2260 * @parent: parent dentry
2097 * @dentry: the negative dentry that was pass 2261 * @dentry: the negative dentry that was passed to the parent's lookup func
2098 * @name: the case-exact name to be associa 2262 * @name: the case-exact name to be associated with the returned dentry
2099 * 2263 *
2100 * Return: true if names are same, or false 2264 * Return: true if names are same, or false
2101 */ 2265 */
2102 bool d_same_name(const struct dentry *dentry, 2266 bool d_same_name(const struct dentry *dentry, const struct dentry *parent,
2103 const struct qstr *name) 2267 const struct qstr *name)
2104 { 2268 {
2105 if (likely(!(parent->d_flags & DCACHE 2269 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2106 if (dentry->d_name.len != nam 2270 if (dentry->d_name.len != name->len)
2107 return false; 2271 return false;
2108 return dentry_cmp(dentry, nam 2272 return dentry_cmp(dentry, name->name, name->len) == 0;
2109 } 2273 }
2110 return parent->d_op->d_compare(dentry 2274 return parent->d_op->d_compare(dentry,
2111 dentry 2275 dentry->d_name.len, dentry->d_name.name,
2112 name) 2276 name) == 0;
2113 } 2277 }
2114 EXPORT_SYMBOL_GPL(d_same_name); 2278 EXPORT_SYMBOL_GPL(d_same_name);
2115 2279
2116 /* 2280 /*
2117 * This is __d_lookup_rcu() when the parent d 2281 * This is __d_lookup_rcu() when the parent dentry has
2118 * DCACHE_OP_COMPARE, which makes things much 2282 * DCACHE_OP_COMPARE, which makes things much nastier.
2119 */ 2283 */
2120 static noinline struct dentry *__d_lookup_rcu 2284 static noinline struct dentry *__d_lookup_rcu_op_compare(
2121 const struct dentry *parent, 2285 const struct dentry *parent,
2122 const struct qstr *name, 2286 const struct qstr *name,
2123 unsigned *seqp) 2287 unsigned *seqp)
2124 { 2288 {
2125 u64 hashlen = name->hash_len; 2289 u64 hashlen = name->hash_len;
2126 struct hlist_bl_head *b = d_hash(hash !! 2290 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2127 struct hlist_bl_node *node; 2291 struct hlist_bl_node *node;
2128 struct dentry *dentry; 2292 struct dentry *dentry;
2129 2293
2130 hlist_bl_for_each_entry_rcu(dentry, n 2294 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2131 int tlen; 2295 int tlen;
2132 const char *tname; 2296 const char *tname;
2133 unsigned seq; 2297 unsigned seq;
2134 2298
2135 seqretry: 2299 seqretry:
2136 seq = raw_seqcount_begin(&den 2300 seq = raw_seqcount_begin(&dentry->d_seq);
2137 if (dentry->d_parent != paren 2301 if (dentry->d_parent != parent)
2138 continue; 2302 continue;
2139 if (d_unhashed(dentry)) 2303 if (d_unhashed(dentry))
2140 continue; 2304 continue;
2141 if (dentry->d_name.hash != ha 2305 if (dentry->d_name.hash != hashlen_hash(hashlen))
2142 continue; 2306 continue;
2143 tlen = dentry->d_name.len; 2307 tlen = dentry->d_name.len;
2144 tname = dentry->d_name.name; 2308 tname = dentry->d_name.name;
2145 /* we want a consistent (name 2309 /* we want a consistent (name,len) pair */
2146 if (read_seqcount_retry(&dent 2310 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2147 cpu_relax(); 2311 cpu_relax();
2148 goto seqretry; 2312 goto seqretry;
2149 } 2313 }
2150 if (parent->d_op->d_compare(d 2314 if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0)
2151 continue; 2315 continue;
2152 *seqp = seq; 2316 *seqp = seq;
2153 return dentry; 2317 return dentry;
2154 } 2318 }
2155 return NULL; 2319 return NULL;
2156 } 2320 }
2157 2321
2158 /** 2322 /**
2159 * __d_lookup_rcu - search for a dentry (racy 2323 * __d_lookup_rcu - search for a dentry (racy, store-free)
2160 * @parent: parent dentry 2324 * @parent: parent dentry
2161 * @name: qstr of name we wish to find 2325 * @name: qstr of name we wish to find
2162 * @seqp: returns d_seq value at the point wh 2326 * @seqp: returns d_seq value at the point where the dentry was found
2163 * Returns: dentry, or NULL 2327 * Returns: dentry, or NULL
2164 * 2328 *
2165 * __d_lookup_rcu is the dcache lookup functi 2329 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2166 * resolution (store-free path walking) desig 2330 * resolution (store-free path walking) design described in
2167 * Documentation/filesystems/path-lookup.txt. 2331 * Documentation/filesystems/path-lookup.txt.
2168 * 2332 *
2169 * This is not to be used outside core vfs. 2333 * This is not to be used outside core vfs.
2170 * 2334 *
2171 * __d_lookup_rcu must only be used in rcu-wa 2335 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2172 * held, and rcu_read_lock held. The returned 2336 * held, and rcu_read_lock held. The returned dentry must not be stored into
2173 * without taking d_lock and checking d_seq s 2337 * without taking d_lock and checking d_seq sequence count against @seq
2174 * returned here. 2338 * returned here.
2175 * 2339 *
>> 2340 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
>> 2341 * function.
>> 2342 *
2176 * Alternatively, __d_lookup_rcu may be calle 2343 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2177 * the returned dentry, so long as its parent 2344 * the returned dentry, so long as its parent's seqlock is checked after the
2178 * child is looked up. Thus, an interlocking 2345 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2179 * is formed, giving integrity down the path 2346 * is formed, giving integrity down the path walk.
2180 * 2347 *
2181 * NOTE! The caller *has* to check the result 2348 * NOTE! The caller *has* to check the resulting dentry against the sequence
2182 * number we've returned before using any of 2349 * number we've returned before using any of the resulting dentry state!
2183 */ 2350 */
2184 struct dentry *__d_lookup_rcu(const struct de 2351 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2185 const struct 2352 const struct qstr *name,
2186 unsigned *seq 2353 unsigned *seqp)
2187 { 2354 {
2188 u64 hashlen = name->hash_len; 2355 u64 hashlen = name->hash_len;
2189 const unsigned char *str = name->name 2356 const unsigned char *str = name->name;
2190 struct hlist_bl_head *b = d_hash(hash !! 2357 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2191 struct hlist_bl_node *node; 2358 struct hlist_bl_node *node;
2192 struct dentry *dentry; 2359 struct dentry *dentry;
2193 2360
2194 /* 2361 /*
2195 * Note: There is significant duplica 2362 * Note: There is significant duplication with __d_lookup_rcu which is
2196 * required to prevent single threade 2363 * required to prevent single threaded performance regressions
2197 * especially on architectures where 2364 * especially on architectures where smp_rmb (in seqcounts) are costly.
2198 * Keep the two functions in sync. 2365 * Keep the two functions in sync.
2199 */ 2366 */
2200 2367
2201 if (unlikely(parent->d_flags & DCACHE 2368 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE))
2202 return __d_lookup_rcu_op_comp 2369 return __d_lookup_rcu_op_compare(parent, name, seqp);
2203 2370
2204 /* 2371 /*
2205 * The hash list is protected using R 2372 * The hash list is protected using RCU.
2206 * 2373 *
2207 * Carefully use d_seq when comparing 2374 * Carefully use d_seq when comparing a candidate dentry, to avoid
2208 * races with d_move(). 2375 * races with d_move().
2209 * 2376 *
2210 * It is possible that concurrent ren 2377 * It is possible that concurrent renames can mess up our list
2211 * walk here and result in missing ou 2378 * walk here and result in missing our dentry, resulting in the
2212 * false-negative result. d_lookup() 2379 * false-negative result. d_lookup() protects against concurrent
2213 * renames using rename_lock seqlock. 2380 * renames using rename_lock seqlock.
2214 * 2381 *
2215 * See Documentation/filesystems/path 2382 * See Documentation/filesystems/path-lookup.txt for more details.
2216 */ 2383 */
2217 hlist_bl_for_each_entry_rcu(dentry, n 2384 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2218 unsigned seq; 2385 unsigned seq;
2219 2386
2220 /* 2387 /*
2221 * The dentry sequence count 2388 * The dentry sequence count protects us from concurrent
2222 * renames, and thus protects 2389 * renames, and thus protects parent and name fields.
2223 * 2390 *
2224 * The caller must perform a 2391 * The caller must perform a seqcount check in order
2225 * to do anything useful with 2392 * to do anything useful with the returned dentry.
2226 * 2393 *
2227 * NOTE! We do a "raw" seqcou 2394 * NOTE! We do a "raw" seqcount_begin here. That means that
2228 * we don't wait for the sequ 2395 * we don't wait for the sequence count to stabilize if it
2229 * is in the middle of a sequ 2396 * is in the middle of a sequence change. If we do the slow
2230 * dentry compare, we will do 2397 * dentry compare, we will do seqretries until it is stable,
2231 * and if we end up with a su 2398 * and if we end up with a successful lookup, we actually
2232 * want to exit RCU lookup an 2399 * want to exit RCU lookup anyway.
2233 * 2400 *
2234 * Note that raw_seqcount_beg 2401 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2235 * we are still guaranteed NU 2402 * we are still guaranteed NUL-termination of ->d_name.name.
2236 */ 2403 */
2237 seq = raw_seqcount_begin(&den 2404 seq = raw_seqcount_begin(&dentry->d_seq);
2238 if (dentry->d_parent != paren 2405 if (dentry->d_parent != parent)
2239 continue; 2406 continue;
2240 if (d_unhashed(dentry)) 2407 if (d_unhashed(dentry))
2241 continue; 2408 continue;
2242 if (dentry->d_name.hash_len ! 2409 if (dentry->d_name.hash_len != hashlen)
2243 continue; 2410 continue;
2244 if (dentry_cmp(dentry, str, h 2411 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2245 continue; 2412 continue;
2246 *seqp = seq; 2413 *seqp = seq;
2247 return dentry; 2414 return dentry;
2248 } 2415 }
2249 return NULL; 2416 return NULL;
2250 } 2417 }
2251 2418
2252 /** 2419 /**
2253 * d_lookup - search for a dentry 2420 * d_lookup - search for a dentry
2254 * @parent: parent dentry 2421 * @parent: parent dentry
2255 * @name: qstr of name we wish to find 2422 * @name: qstr of name we wish to find
2256 * Returns: dentry, or NULL 2423 * Returns: dentry, or NULL
2257 * 2424 *
2258 * d_lookup searches the children of the pare 2425 * d_lookup searches the children of the parent dentry for the name in
2259 * question. If the dentry is found its refer 2426 * question. If the dentry is found its reference count is incremented and the
2260 * dentry is returned. The caller must use dp 2427 * dentry is returned. The caller must use dput to free the entry when it has
2261 * finished using it. %NULL is returned if th 2428 * finished using it. %NULL is returned if the dentry does not exist.
2262 */ 2429 */
2263 struct dentry *d_lookup(const struct dentry * 2430 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2264 { 2431 {
2265 struct dentry *dentry; 2432 struct dentry *dentry;
2266 unsigned seq; 2433 unsigned seq;
2267 2434
2268 do { 2435 do {
2269 seq = read_seqbegin(&rename_l 2436 seq = read_seqbegin(&rename_lock);
2270 dentry = __d_lookup(parent, n 2437 dentry = __d_lookup(parent, name);
2271 if (dentry) 2438 if (dentry)
2272 break; 2439 break;
2273 } while (read_seqretry(&rename_lock, 2440 } while (read_seqretry(&rename_lock, seq));
2274 return dentry; 2441 return dentry;
2275 } 2442 }
2276 EXPORT_SYMBOL(d_lookup); 2443 EXPORT_SYMBOL(d_lookup);
2277 2444
2278 /** 2445 /**
2279 * __d_lookup - search for a dentry (racy) 2446 * __d_lookup - search for a dentry (racy)
2280 * @parent: parent dentry 2447 * @parent: parent dentry
2281 * @name: qstr of name we wish to find 2448 * @name: qstr of name we wish to find
2282 * Returns: dentry, or NULL 2449 * Returns: dentry, or NULL
2283 * 2450 *
2284 * __d_lookup is like d_lookup, however it ma 2451 * __d_lookup is like d_lookup, however it may (rarely) return a
2285 * false-negative result due to unrelated ren 2452 * false-negative result due to unrelated rename activity.
2286 * 2453 *
2287 * __d_lookup is slightly faster by avoiding 2454 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2288 * however it must be used carefully, eg. wit 2455 * however it must be used carefully, eg. with a following d_lookup in
2289 * the case of failure. 2456 * the case of failure.
2290 * 2457 *
2291 * __d_lookup callers must be commented. 2458 * __d_lookup callers must be commented.
2292 */ 2459 */
2293 struct dentry *__d_lookup(const struct dentry 2460 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2294 { 2461 {
2295 unsigned int hash = name->hash; 2462 unsigned int hash = name->hash;
2296 struct hlist_bl_head *b = d_hash(hash 2463 struct hlist_bl_head *b = d_hash(hash);
2297 struct hlist_bl_node *node; 2464 struct hlist_bl_node *node;
2298 struct dentry *found = NULL; 2465 struct dentry *found = NULL;
2299 struct dentry *dentry; 2466 struct dentry *dentry;
2300 2467
2301 /* 2468 /*
2302 * Note: There is significant duplica 2469 * Note: There is significant duplication with __d_lookup_rcu which is
2303 * required to prevent single threade 2470 * required to prevent single threaded performance regressions
2304 * especially on architectures where 2471 * especially on architectures where smp_rmb (in seqcounts) are costly.
2305 * Keep the two functions in sync. 2472 * Keep the two functions in sync.
2306 */ 2473 */
2307 2474
2308 /* 2475 /*
2309 * The hash list is protected using R 2476 * The hash list is protected using RCU.
2310 * 2477 *
2311 * Take d_lock when comparing a candi 2478 * Take d_lock when comparing a candidate dentry, to avoid races
2312 * with d_move(). 2479 * with d_move().
2313 * 2480 *
2314 * It is possible that concurrent ren 2481 * It is possible that concurrent renames can mess up our list
2315 * walk here and result in missing ou 2482 * walk here and result in missing our dentry, resulting in the
2316 * false-negative result. d_lookup() 2483 * false-negative result. d_lookup() protects against concurrent
2317 * renames using rename_lock seqlock. 2484 * renames using rename_lock seqlock.
2318 * 2485 *
2319 * See Documentation/filesystems/path 2486 * See Documentation/filesystems/path-lookup.txt for more details.
2320 */ 2487 */
2321 rcu_read_lock(); 2488 rcu_read_lock();
2322 2489
2323 hlist_bl_for_each_entry_rcu(dentry, n 2490 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2324 2491
2325 if (dentry->d_name.hash != ha 2492 if (dentry->d_name.hash != hash)
2326 continue; 2493 continue;
2327 2494
2328 spin_lock(&dentry->d_lock); 2495 spin_lock(&dentry->d_lock);
2329 if (dentry->d_parent != paren 2496 if (dentry->d_parent != parent)
2330 goto next; 2497 goto next;
2331 if (d_unhashed(dentry)) 2498 if (d_unhashed(dentry))
2332 goto next; 2499 goto next;
2333 2500
2334 if (!d_same_name(dentry, pare 2501 if (!d_same_name(dentry, parent, name))
2335 goto next; 2502 goto next;
2336 2503
2337 dentry->d_lockref.count++; 2504 dentry->d_lockref.count++;
2338 found = dentry; 2505 found = dentry;
2339 spin_unlock(&dentry->d_lock); 2506 spin_unlock(&dentry->d_lock);
2340 break; 2507 break;
2341 next: 2508 next:
2342 spin_unlock(&dentry->d_lock); 2509 spin_unlock(&dentry->d_lock);
2343 } 2510 }
2344 rcu_read_unlock(); 2511 rcu_read_unlock();
2345 2512
2346 return found; 2513 return found;
2347 } 2514 }
2348 2515
2349 /** 2516 /**
2350 * d_hash_and_lookup - hash the qstr then sea 2517 * d_hash_and_lookup - hash the qstr then search for a dentry
2351 * @dir: Directory to search in 2518 * @dir: Directory to search in
2352 * @name: qstr of name we wish to find 2519 * @name: qstr of name we wish to find
2353 * 2520 *
2354 * On lookup failure NULL is returned; on bad 2521 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2355 */ 2522 */
2356 struct dentry *d_hash_and_lookup(struct dentr 2523 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2357 { 2524 {
2358 /* 2525 /*
2359 * Check for a fs-specific hash funct 2526 * Check for a fs-specific hash function. Note that we must
2360 * calculate the standard hash first, 2527 * calculate the standard hash first, as the d_op->d_hash()
2361 * routine may choose to leave the ha 2528 * routine may choose to leave the hash value unchanged.
2362 */ 2529 */
2363 name->hash = full_name_hash(dir, name 2530 name->hash = full_name_hash(dir, name->name, name->len);
2364 if (dir->d_flags & DCACHE_OP_HASH) { 2531 if (dir->d_flags & DCACHE_OP_HASH) {
2365 int err = dir->d_op->d_hash(d 2532 int err = dir->d_op->d_hash(dir, name);
2366 if (unlikely(err < 0)) 2533 if (unlikely(err < 0))
2367 return ERR_PTR(err); 2534 return ERR_PTR(err);
2368 } 2535 }
2369 return d_lookup(dir, name); 2536 return d_lookup(dir, name);
2370 } 2537 }
2371 EXPORT_SYMBOL(d_hash_and_lookup); 2538 EXPORT_SYMBOL(d_hash_and_lookup);
2372 2539
2373 /* 2540 /*
2374 * When a file is deleted, we have two option 2541 * When a file is deleted, we have two options:
2375 * - turn this dentry into a negative dentry 2542 * - turn this dentry into a negative dentry
2376 * - unhash this dentry and free it. 2543 * - unhash this dentry and free it.
2377 * 2544 *
2378 * Usually, we want to just turn this into 2545 * Usually, we want to just turn this into
2379 * a negative dentry, but if anybody else is 2546 * a negative dentry, but if anybody else is
2380 * currently using the dentry or the inode 2547 * currently using the dentry or the inode
2381 * we can't do that and we fall back on remov 2548 * we can't do that and we fall back on removing
2382 * it from the hash queues and waiting for 2549 * it from the hash queues and waiting for
2383 * it to be deleted later when it has no user 2550 * it to be deleted later when it has no users
2384 */ 2551 */
2385 2552
2386 /** 2553 /**
2387 * d_delete - delete a dentry 2554 * d_delete - delete a dentry
2388 * @dentry: The dentry to delete 2555 * @dentry: The dentry to delete
2389 * 2556 *
2390 * Turn the dentry into a negative dentry if 2557 * Turn the dentry into a negative dentry if possible, otherwise
2391 * remove it from the hash queues so it can b 2558 * remove it from the hash queues so it can be deleted later
2392 */ 2559 */
2393 2560
2394 void d_delete(struct dentry * dentry) 2561 void d_delete(struct dentry * dentry)
2395 { 2562 {
2396 struct inode *inode = dentry->d_inode 2563 struct inode *inode = dentry->d_inode;
2397 2564
2398 spin_lock(&inode->i_lock); 2565 spin_lock(&inode->i_lock);
2399 spin_lock(&dentry->d_lock); 2566 spin_lock(&dentry->d_lock);
2400 /* 2567 /*
2401 * Are we the only user? 2568 * Are we the only user?
2402 */ 2569 */
2403 if (dentry->d_lockref.count == 1) { 2570 if (dentry->d_lockref.count == 1) {
2404 dentry->d_flags &= ~DCACHE_CA 2571 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2405 dentry_unlink_inode(dentry); 2572 dentry_unlink_inode(dentry);
2406 } else { 2573 } else {
2407 __d_drop(dentry); 2574 __d_drop(dentry);
2408 spin_unlock(&dentry->d_lock); 2575 spin_unlock(&dentry->d_lock);
2409 spin_unlock(&inode->i_lock); 2576 spin_unlock(&inode->i_lock);
2410 } 2577 }
2411 } 2578 }
2412 EXPORT_SYMBOL(d_delete); 2579 EXPORT_SYMBOL(d_delete);
2413 2580
2414 static void __d_rehash(struct dentry *entry) 2581 static void __d_rehash(struct dentry *entry)
2415 { 2582 {
2416 struct hlist_bl_head *b = d_hash(entr 2583 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2417 2584
2418 hlist_bl_lock(b); 2585 hlist_bl_lock(b);
2419 hlist_bl_add_head_rcu(&entry->d_hash, 2586 hlist_bl_add_head_rcu(&entry->d_hash, b);
2420 hlist_bl_unlock(b); 2587 hlist_bl_unlock(b);
2421 } 2588 }
2422 2589
2423 /** 2590 /**
2424 * d_rehash - add an entry back to the ha 2591 * d_rehash - add an entry back to the hash
2425 * @entry: dentry to add to the hash 2592 * @entry: dentry to add to the hash
2426 * 2593 *
2427 * Adds a dentry to the hash according to its 2594 * Adds a dentry to the hash according to its name.
2428 */ 2595 */
2429 2596
2430 void d_rehash(struct dentry * entry) 2597 void d_rehash(struct dentry * entry)
2431 { 2598 {
2432 spin_lock(&entry->d_lock); 2599 spin_lock(&entry->d_lock);
2433 __d_rehash(entry); 2600 __d_rehash(entry);
2434 spin_unlock(&entry->d_lock); 2601 spin_unlock(&entry->d_lock);
2435 } 2602 }
2436 EXPORT_SYMBOL(d_rehash); 2603 EXPORT_SYMBOL(d_rehash);
2437 2604
2438 static inline unsigned start_dir_add(struct i 2605 static inline unsigned start_dir_add(struct inode *dir)
2439 { 2606 {
2440 preempt_disable_nested(); 2607 preempt_disable_nested();
2441 for (;;) { 2608 for (;;) {
2442 unsigned n = dir->i_dir_seq; 2609 unsigned n = dir->i_dir_seq;
2443 if (!(n & 1) && cmpxchg(&dir- 2610 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2444 return n; 2611 return n;
2445 cpu_relax(); 2612 cpu_relax();
2446 } 2613 }
2447 } 2614 }
2448 2615
2449 static inline void end_dir_add(struct inode * 2616 static inline void end_dir_add(struct inode *dir, unsigned int n,
2450 wait_queue_hea 2617 wait_queue_head_t *d_wait)
2451 { 2618 {
2452 smp_store_release(&dir->i_dir_seq, n 2619 smp_store_release(&dir->i_dir_seq, n + 2);
2453 preempt_enable_nested(); 2620 preempt_enable_nested();
2454 wake_up_all(d_wait); 2621 wake_up_all(d_wait);
2455 } 2622 }
2456 2623
2457 static void d_wait_lookup(struct dentry *dent 2624 static void d_wait_lookup(struct dentry *dentry)
2458 { 2625 {
2459 if (d_in_lookup(dentry)) { 2626 if (d_in_lookup(dentry)) {
2460 DECLARE_WAITQUEUE(wait, curre 2627 DECLARE_WAITQUEUE(wait, current);
2461 add_wait_queue(dentry->d_wait 2628 add_wait_queue(dentry->d_wait, &wait);
2462 do { 2629 do {
2463 set_current_state(TAS 2630 set_current_state(TASK_UNINTERRUPTIBLE);
2464 spin_unlock(&dentry-> 2631 spin_unlock(&dentry->d_lock);
2465 schedule(); 2632 schedule();
2466 spin_lock(&dentry->d_ 2633 spin_lock(&dentry->d_lock);
2467 } while (d_in_lookup(dentry)) 2634 } while (d_in_lookup(dentry));
2468 } 2635 }
2469 } 2636 }
2470 2637
2471 struct dentry *d_alloc_parallel(struct dentry 2638 struct dentry *d_alloc_parallel(struct dentry *parent,
2472 const struct 2639 const struct qstr *name,
2473 wait_queue_he 2640 wait_queue_head_t *wq)
2474 { 2641 {
2475 unsigned int hash = name->hash; 2642 unsigned int hash = name->hash;
2476 struct hlist_bl_head *b = in_lookup_h 2643 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2477 struct hlist_bl_node *node; 2644 struct hlist_bl_node *node;
2478 struct dentry *new = d_alloc(parent, 2645 struct dentry *new = d_alloc(parent, name);
2479 struct dentry *dentry; 2646 struct dentry *dentry;
2480 unsigned seq, r_seq, d_seq; 2647 unsigned seq, r_seq, d_seq;
2481 2648
2482 if (unlikely(!new)) 2649 if (unlikely(!new))
2483 return ERR_PTR(-ENOMEM); 2650 return ERR_PTR(-ENOMEM);
2484 2651
2485 retry: 2652 retry:
2486 rcu_read_lock(); 2653 rcu_read_lock();
2487 seq = smp_load_acquire(&parent->d_ino 2654 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2488 r_seq = read_seqbegin(&rename_lock); 2655 r_seq = read_seqbegin(&rename_lock);
2489 dentry = __d_lookup_rcu(parent, name, 2656 dentry = __d_lookup_rcu(parent, name, &d_seq);
2490 if (unlikely(dentry)) { 2657 if (unlikely(dentry)) {
2491 if (!lockref_get_not_dead(&de 2658 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2492 rcu_read_unlock(); 2659 rcu_read_unlock();
2493 goto retry; 2660 goto retry;
2494 } 2661 }
2495 if (read_seqcount_retry(&dent 2662 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2496 rcu_read_unlock(); 2663 rcu_read_unlock();
2497 dput(dentry); 2664 dput(dentry);
2498 goto retry; 2665 goto retry;
2499 } 2666 }
2500 rcu_read_unlock(); 2667 rcu_read_unlock();
2501 dput(new); 2668 dput(new);
2502 return dentry; 2669 return dentry;
2503 } 2670 }
2504 if (unlikely(read_seqretry(&rename_lo 2671 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2505 rcu_read_unlock(); 2672 rcu_read_unlock();
2506 goto retry; 2673 goto retry;
2507 } 2674 }
2508 2675
2509 if (unlikely(seq & 1)) { 2676 if (unlikely(seq & 1)) {
2510 rcu_read_unlock(); 2677 rcu_read_unlock();
2511 goto retry; 2678 goto retry;
2512 } 2679 }
2513 2680
2514 hlist_bl_lock(b); 2681 hlist_bl_lock(b);
2515 if (unlikely(READ_ONCE(parent->d_inod 2682 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2516 hlist_bl_unlock(b); 2683 hlist_bl_unlock(b);
2517 rcu_read_unlock(); 2684 rcu_read_unlock();
2518 goto retry; 2685 goto retry;
2519 } 2686 }
2520 /* 2687 /*
2521 * No changes for the parent since th 2688 * No changes for the parent since the beginning of d_lookup().
2522 * Since all removals from the chain 2689 * Since all removals from the chain happen with hlist_bl_lock(),
2523 * any potential in-lookup matches ar 2690 * any potential in-lookup matches are going to stay here until
2524 * we unlock the chain. All fields a 2691 * we unlock the chain. All fields are stable in everything
2525 * we encounter. 2692 * we encounter.
2526 */ 2693 */
2527 hlist_bl_for_each_entry(dentry, node, 2694 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2528 if (dentry->d_name.hash != ha 2695 if (dentry->d_name.hash != hash)
2529 continue; 2696 continue;
2530 if (dentry->d_parent != paren 2697 if (dentry->d_parent != parent)
2531 continue; 2698 continue;
2532 if (!d_same_name(dentry, pare 2699 if (!d_same_name(dentry, parent, name))
2533 continue; 2700 continue;
2534 hlist_bl_unlock(b); 2701 hlist_bl_unlock(b);
2535 /* now we can try to grab a r 2702 /* now we can try to grab a reference */
2536 if (!lockref_get_not_dead(&de 2703 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2537 rcu_read_unlock(); 2704 rcu_read_unlock();
2538 goto retry; 2705 goto retry;
2539 } 2706 }
2540 2707
2541 rcu_read_unlock(); 2708 rcu_read_unlock();
2542 /* 2709 /*
2543 * somebody is likely to be s 2710 * somebody is likely to be still doing lookup for it;
2544 * wait for them to finish 2711 * wait for them to finish
2545 */ 2712 */
2546 spin_lock(&dentry->d_lock); 2713 spin_lock(&dentry->d_lock);
2547 d_wait_lookup(dentry); 2714 d_wait_lookup(dentry);
2548 /* 2715 /*
2549 * it's not in-lookup anymore 2716 * it's not in-lookup anymore; in principle we should repeat
2550 * everything from dcache loo 2717 * everything from dcache lookup, but it's likely to be what
2551 * d_lookup() would've found 2718 * d_lookup() would've found anyway. If it is, just return it;
2552 * otherwise we really have t 2719 * otherwise we really have to repeat the whole thing.
2553 */ 2720 */
2554 if (unlikely(dentry->d_name.h 2721 if (unlikely(dentry->d_name.hash != hash))
2555 goto mismatch; 2722 goto mismatch;
2556 if (unlikely(dentry->d_parent 2723 if (unlikely(dentry->d_parent != parent))
2557 goto mismatch; 2724 goto mismatch;
2558 if (unlikely(d_unhashed(dentr 2725 if (unlikely(d_unhashed(dentry)))
2559 goto mismatch; 2726 goto mismatch;
2560 if (unlikely(!d_same_name(den 2727 if (unlikely(!d_same_name(dentry, parent, name)))
2561 goto mismatch; 2728 goto mismatch;
2562 /* OK, it *is* a hashed match 2729 /* OK, it *is* a hashed match; return it */
2563 spin_unlock(&dentry->d_lock); 2730 spin_unlock(&dentry->d_lock);
2564 dput(new); 2731 dput(new);
2565 return dentry; 2732 return dentry;
2566 } 2733 }
2567 rcu_read_unlock(); 2734 rcu_read_unlock();
2568 /* we can't take ->d_lock here; it's 2735 /* we can't take ->d_lock here; it's OK, though. */
2569 new->d_flags |= DCACHE_PAR_LOOKUP; 2736 new->d_flags |= DCACHE_PAR_LOOKUP;
2570 new->d_wait = wq; 2737 new->d_wait = wq;
2571 hlist_bl_add_head(&new->d_u.d_in_look !! 2738 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2572 hlist_bl_unlock(b); 2739 hlist_bl_unlock(b);
2573 return new; 2740 return new;
2574 mismatch: 2741 mismatch:
2575 spin_unlock(&dentry->d_lock); 2742 spin_unlock(&dentry->d_lock);
2576 dput(dentry); 2743 dput(dentry);
2577 goto retry; 2744 goto retry;
2578 } 2745 }
2579 EXPORT_SYMBOL(d_alloc_parallel); 2746 EXPORT_SYMBOL(d_alloc_parallel);
2580 2747
2581 /* 2748 /*
2582 * - Unhash the dentry 2749 * - Unhash the dentry
2583 * - Retrieve and clear the waitqueue head in 2750 * - Retrieve and clear the waitqueue head in dentry
2584 * - Return the waitqueue head 2751 * - Return the waitqueue head
2585 */ 2752 */
2586 static wait_queue_head_t *__d_lookup_unhash(s 2753 static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry)
2587 { 2754 {
2588 wait_queue_head_t *d_wait; 2755 wait_queue_head_t *d_wait;
2589 struct hlist_bl_head *b; 2756 struct hlist_bl_head *b;
2590 2757
2591 lockdep_assert_held(&dentry->d_lock); 2758 lockdep_assert_held(&dentry->d_lock);
2592 2759
2593 b = in_lookup_hash(dentry->d_parent, 2760 b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash);
2594 hlist_bl_lock(b); 2761 hlist_bl_lock(b);
2595 dentry->d_flags &= ~DCACHE_PAR_LOOKUP 2762 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2596 __hlist_bl_del(&dentry->d_u.d_in_look 2763 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2597 d_wait = dentry->d_wait; 2764 d_wait = dentry->d_wait;
2598 dentry->d_wait = NULL; 2765 dentry->d_wait = NULL;
2599 hlist_bl_unlock(b); 2766 hlist_bl_unlock(b);
2600 INIT_HLIST_NODE(&dentry->d_u.d_alias) 2767 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2601 INIT_LIST_HEAD(&dentry->d_lru); 2768 INIT_LIST_HEAD(&dentry->d_lru);
2602 return d_wait; 2769 return d_wait;
2603 } 2770 }
2604 2771
2605 void __d_lookup_unhash_wake(struct dentry *de 2772 void __d_lookup_unhash_wake(struct dentry *dentry)
2606 { 2773 {
2607 spin_lock(&dentry->d_lock); 2774 spin_lock(&dentry->d_lock);
2608 wake_up_all(__d_lookup_unhash(dentry) 2775 wake_up_all(__d_lookup_unhash(dentry));
2609 spin_unlock(&dentry->d_lock); 2776 spin_unlock(&dentry->d_lock);
2610 } 2777 }
2611 EXPORT_SYMBOL(__d_lookup_unhash_wake); 2778 EXPORT_SYMBOL(__d_lookup_unhash_wake);
2612 2779
2613 /* inode->i_lock held if inode is non-NULL */ 2780 /* inode->i_lock held if inode is non-NULL */
2614 2781
2615 static inline void __d_add(struct dentry *den 2782 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2616 { 2783 {
2617 wait_queue_head_t *d_wait; 2784 wait_queue_head_t *d_wait;
2618 struct inode *dir = NULL; 2785 struct inode *dir = NULL;
2619 unsigned n; 2786 unsigned n;
2620 spin_lock(&dentry->d_lock); 2787 spin_lock(&dentry->d_lock);
2621 if (unlikely(d_in_lookup(dentry))) { 2788 if (unlikely(d_in_lookup(dentry))) {
2622 dir = dentry->d_parent->d_ino 2789 dir = dentry->d_parent->d_inode;
2623 n = start_dir_add(dir); 2790 n = start_dir_add(dir);
2624 d_wait = __d_lookup_unhash(de 2791 d_wait = __d_lookup_unhash(dentry);
2625 } 2792 }
2626 if (inode) { 2793 if (inode) {
2627 unsigned add_flags = d_flags_ 2794 unsigned add_flags = d_flags_for_inode(inode);
2628 hlist_add_head(&dentry->d_u.d 2795 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2629 raw_write_seqcount_begin(&den 2796 raw_write_seqcount_begin(&dentry->d_seq);
2630 __d_set_inode_and_type(dentry 2797 __d_set_inode_and_type(dentry, inode, add_flags);
2631 raw_write_seqcount_end(&dentr 2798 raw_write_seqcount_end(&dentry->d_seq);
2632 fsnotify_update_flags(dentry) 2799 fsnotify_update_flags(dentry);
2633 } 2800 }
2634 __d_rehash(dentry); 2801 __d_rehash(dentry);
2635 if (dir) 2802 if (dir)
2636 end_dir_add(dir, n, d_wait); 2803 end_dir_add(dir, n, d_wait);
2637 spin_unlock(&dentry->d_lock); 2804 spin_unlock(&dentry->d_lock);
2638 if (inode) 2805 if (inode)
2639 spin_unlock(&inode->i_lock); 2806 spin_unlock(&inode->i_lock);
2640 } 2807 }
2641 2808
2642 /** 2809 /**
2643 * d_add - add dentry to hash queues 2810 * d_add - add dentry to hash queues
2644 * @entry: dentry to add 2811 * @entry: dentry to add
2645 * @inode: The inode to attach to this dentry 2812 * @inode: The inode to attach to this dentry
2646 * 2813 *
2647 * This adds the entry to the hash queues and 2814 * This adds the entry to the hash queues and initializes @inode.
2648 * The entry was actually filled in earlier d 2815 * The entry was actually filled in earlier during d_alloc().
2649 */ 2816 */
2650 2817
2651 void d_add(struct dentry *entry, struct inode 2818 void d_add(struct dentry *entry, struct inode *inode)
2652 { 2819 {
2653 if (inode) { 2820 if (inode) {
2654 security_d_instantiate(entry, 2821 security_d_instantiate(entry, inode);
2655 spin_lock(&inode->i_lock); 2822 spin_lock(&inode->i_lock);
2656 } 2823 }
2657 __d_add(entry, inode); 2824 __d_add(entry, inode);
2658 } 2825 }
2659 EXPORT_SYMBOL(d_add); 2826 EXPORT_SYMBOL(d_add);
2660 2827
2661 /** 2828 /**
2662 * d_exact_alias - find and hash an exact unh 2829 * d_exact_alias - find and hash an exact unhashed alias
2663 * @entry: dentry to add 2830 * @entry: dentry to add
2664 * @inode: The inode to go with this dentry 2831 * @inode: The inode to go with this dentry
2665 * 2832 *
2666 * If an unhashed dentry with the same name/p 2833 * If an unhashed dentry with the same name/parent and desired
2667 * inode already exists, hash and return it. 2834 * inode already exists, hash and return it. Otherwise, return
2668 * NULL. 2835 * NULL.
2669 * 2836 *
2670 * Parent directory should be locked. 2837 * Parent directory should be locked.
2671 */ 2838 */
2672 struct dentry *d_exact_alias(struct dentry *e 2839 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2673 { 2840 {
2674 struct dentry *alias; 2841 struct dentry *alias;
2675 unsigned int hash = entry->d_name.has 2842 unsigned int hash = entry->d_name.hash;
2676 2843
2677 spin_lock(&inode->i_lock); 2844 spin_lock(&inode->i_lock);
2678 hlist_for_each_entry(alias, &inode->i 2845 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2679 /* 2846 /*
2680 * Don't need alias->d_lock h 2847 * Don't need alias->d_lock here, because aliases with
2681 * d_parent == entry->d_paren 2848 * d_parent == entry->d_parent are not subject to name or
2682 * parent changes, because th 2849 * parent changes, because the parent inode i_mutex is held.
2683 */ 2850 */
2684 if (alias->d_name.hash != has 2851 if (alias->d_name.hash != hash)
2685 continue; 2852 continue;
2686 if (alias->d_parent != entry- 2853 if (alias->d_parent != entry->d_parent)
2687 continue; 2854 continue;
2688 if (!d_same_name(alias, entry 2855 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2689 continue; 2856 continue;
2690 spin_lock(&alias->d_lock); 2857 spin_lock(&alias->d_lock);
2691 if (!d_unhashed(alias)) { 2858 if (!d_unhashed(alias)) {
2692 spin_unlock(&alias->d 2859 spin_unlock(&alias->d_lock);
2693 alias = NULL; 2860 alias = NULL;
2694 } else { 2861 } else {
2695 dget_dlock(alias); !! 2862 __dget_dlock(alias);
2696 __d_rehash(alias); 2863 __d_rehash(alias);
2697 spin_unlock(&alias->d 2864 spin_unlock(&alias->d_lock);
2698 } 2865 }
2699 spin_unlock(&inode->i_lock); 2866 spin_unlock(&inode->i_lock);
2700 return alias; 2867 return alias;
2701 } 2868 }
2702 spin_unlock(&inode->i_lock); 2869 spin_unlock(&inode->i_lock);
2703 return NULL; 2870 return NULL;
2704 } 2871 }
2705 EXPORT_SYMBOL(d_exact_alias); 2872 EXPORT_SYMBOL(d_exact_alias);
2706 2873
2707 static void swap_names(struct dentry *dentry, 2874 static void swap_names(struct dentry *dentry, struct dentry *target)
2708 { 2875 {
2709 if (unlikely(dname_external(target))) 2876 if (unlikely(dname_external(target))) {
2710 if (unlikely(dname_external(d 2877 if (unlikely(dname_external(dentry))) {
2711 /* 2878 /*
2712 * Both external: swa 2879 * Both external: swap the pointers
2713 */ 2880 */
2714 swap(target->d_name.n 2881 swap(target->d_name.name, dentry->d_name.name);
2715 } else { 2882 } else {
2716 /* 2883 /*
2717 * dentry:internal, t 2884 * dentry:internal, target:external. Steal target's
2718 * storage and make t 2885 * storage and make target internal.
2719 */ 2886 */
2720 memcpy(target->d_inam 2887 memcpy(target->d_iname, dentry->d_name.name,
2721 dentr 2888 dentry->d_name.len + 1);
2722 dentry->d_name.name = 2889 dentry->d_name.name = target->d_name.name;
2723 target->d_name.name = 2890 target->d_name.name = target->d_iname;
2724 } 2891 }
2725 } else { 2892 } else {
2726 if (unlikely(dname_external(d 2893 if (unlikely(dname_external(dentry))) {
2727 /* 2894 /*
2728 * dentry:external, t 2895 * dentry:external, target:internal. Give dentry's
2729 * storage to target 2896 * storage to target and make dentry internal
2730 */ 2897 */
2731 memcpy(dentry->d_inam 2898 memcpy(dentry->d_iname, target->d_name.name,
2732 targe 2899 target->d_name.len + 1);
2733 target->d_name.name = 2900 target->d_name.name = dentry->d_name.name;
2734 dentry->d_name.name = 2901 dentry->d_name.name = dentry->d_iname;
2735 } else { 2902 } else {
2736 /* 2903 /*
2737 * Both are internal. 2904 * Both are internal.
2738 */ 2905 */
2739 unsigned int i; 2906 unsigned int i;
2740 BUILD_BUG_ON(!IS_ALIG 2907 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2741 for (i = 0; i < DNAME 2908 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2742 swap(((long * 2909 swap(((long *) &dentry->d_iname)[i],
2743 ((long * 2910 ((long *) &target->d_iname)[i]);
2744 } 2911 }
2745 } 2912 }
2746 } 2913 }
2747 swap(dentry->d_name.hash_len, target- 2914 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2748 } 2915 }
2749 2916
2750 static void copy_name(struct dentry *dentry, 2917 static void copy_name(struct dentry *dentry, struct dentry *target)
2751 { 2918 {
2752 struct external_name *old_name = NULL 2919 struct external_name *old_name = NULL;
2753 if (unlikely(dname_external(dentry))) 2920 if (unlikely(dname_external(dentry)))
2754 old_name = external_name(dent 2921 old_name = external_name(dentry);
2755 if (unlikely(dname_external(target))) 2922 if (unlikely(dname_external(target))) {
2756 atomic_inc(&external_name(tar 2923 atomic_inc(&external_name(target)->u.count);
2757 dentry->d_name = target->d_na 2924 dentry->d_name = target->d_name;
2758 } else { 2925 } else {
2759 memcpy(dentry->d_iname, targe 2926 memcpy(dentry->d_iname, target->d_name.name,
2760 target->d_nam 2927 target->d_name.len + 1);
2761 dentry->d_name.name = dentry- 2928 dentry->d_name.name = dentry->d_iname;
2762 dentry->d_name.hash_len = tar 2929 dentry->d_name.hash_len = target->d_name.hash_len;
2763 } 2930 }
2764 if (old_name && likely(atomic_dec_and 2931 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2765 kfree_rcu(old_name, u.head); 2932 kfree_rcu(old_name, u.head);
2766 } 2933 }
2767 2934
2768 /* 2935 /*
2769 * __d_move - move a dentry 2936 * __d_move - move a dentry
2770 * @dentry: entry to move 2937 * @dentry: entry to move
2771 * @target: new dentry 2938 * @target: new dentry
2772 * @exchange: exchange the two dentries 2939 * @exchange: exchange the two dentries
2773 * 2940 *
2774 * Update the dcache to reflect the move of a 2941 * Update the dcache to reflect the move of a file name. Negative
2775 * dcache entries should not be moved in this 2942 * dcache entries should not be moved in this way. Caller must hold
2776 * rename_lock, the i_mutex of the source and 2943 * rename_lock, the i_mutex of the source and target directories,
2777 * and the sb->s_vfs_rename_mutex if they dif 2944 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2778 */ 2945 */
2779 static void __d_move(struct dentry *dentry, s 2946 static void __d_move(struct dentry *dentry, struct dentry *target,
2780 bool exchange) 2947 bool exchange)
2781 { 2948 {
2782 struct dentry *old_parent, *p; 2949 struct dentry *old_parent, *p;
2783 wait_queue_head_t *d_wait; 2950 wait_queue_head_t *d_wait;
2784 struct inode *dir = NULL; 2951 struct inode *dir = NULL;
2785 unsigned n; 2952 unsigned n;
2786 2953
2787 WARN_ON(!dentry->d_inode); 2954 WARN_ON(!dentry->d_inode);
2788 if (WARN_ON(dentry == target)) 2955 if (WARN_ON(dentry == target))
2789 return; 2956 return;
2790 2957
2791 BUG_ON(d_ancestor(target, dentry)); 2958 BUG_ON(d_ancestor(target, dentry));
2792 old_parent = dentry->d_parent; 2959 old_parent = dentry->d_parent;
2793 p = d_ancestor(old_parent, target); 2960 p = d_ancestor(old_parent, target);
2794 if (IS_ROOT(dentry)) { 2961 if (IS_ROOT(dentry)) {
2795 BUG_ON(p); 2962 BUG_ON(p);
2796 spin_lock(&target->d_parent-> 2963 spin_lock(&target->d_parent->d_lock);
2797 } else if (!p) { 2964 } else if (!p) {
2798 /* target is not a descendent 2965 /* target is not a descendent of dentry->d_parent */
2799 spin_lock(&target->d_parent-> 2966 spin_lock(&target->d_parent->d_lock);
2800 spin_lock_nested(&old_parent- 2967 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2801 } else { 2968 } else {
2802 BUG_ON(p == dentry); 2969 BUG_ON(p == dentry);
2803 spin_lock(&old_parent->d_lock 2970 spin_lock(&old_parent->d_lock);
2804 if (p != target) 2971 if (p != target)
2805 spin_lock_nested(&tar 2972 spin_lock_nested(&target->d_parent->d_lock,
2806 DENTR 2973 DENTRY_D_LOCK_NESTED);
2807 } 2974 }
2808 spin_lock_nested(&dentry->d_lock, 2); 2975 spin_lock_nested(&dentry->d_lock, 2);
2809 spin_lock_nested(&target->d_lock, 3); 2976 spin_lock_nested(&target->d_lock, 3);
2810 2977
2811 if (unlikely(d_in_lookup(target))) { 2978 if (unlikely(d_in_lookup(target))) {
2812 dir = target->d_parent->d_ino 2979 dir = target->d_parent->d_inode;
2813 n = start_dir_add(dir); 2980 n = start_dir_add(dir);
2814 d_wait = __d_lookup_unhash(ta 2981 d_wait = __d_lookup_unhash(target);
2815 } 2982 }
2816 2983
2817 write_seqcount_begin(&dentry->d_seq); 2984 write_seqcount_begin(&dentry->d_seq);
2818 write_seqcount_begin_nested(&target-> 2985 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2819 2986
2820 /* unhash both */ 2987 /* unhash both */
2821 if (!d_unhashed(dentry)) 2988 if (!d_unhashed(dentry))
2822 ___d_drop(dentry); 2989 ___d_drop(dentry);
2823 if (!d_unhashed(target)) 2990 if (!d_unhashed(target))
2824 ___d_drop(target); 2991 ___d_drop(target);
2825 2992
2826 /* ... and switch them in the tree */ 2993 /* ... and switch them in the tree */
2827 dentry->d_parent = target->d_parent; 2994 dentry->d_parent = target->d_parent;
2828 if (!exchange) { 2995 if (!exchange) {
2829 copy_name(dentry, target); 2996 copy_name(dentry, target);
2830 target->d_hash.pprev = NULL; 2997 target->d_hash.pprev = NULL;
2831 dentry->d_parent->d_lockref.c 2998 dentry->d_parent->d_lockref.count++;
2832 if (dentry != old_parent) /* 2999 if (dentry != old_parent) /* wasn't IS_ROOT */
2833 WARN_ON(!--old_parent 3000 WARN_ON(!--old_parent->d_lockref.count);
2834 } else { 3001 } else {
2835 target->d_parent = old_parent 3002 target->d_parent = old_parent;
2836 swap_names(dentry, target); 3003 swap_names(dentry, target);
2837 if (!hlist_unhashed(&target-> !! 3004 list_move(&target->d_child, &target->d_parent->d_subdirs);
2838 __hlist_del(&target-> <<
2839 hlist_add_head(&target->d_sib <<
2840 __d_rehash(target); 3005 __d_rehash(target);
2841 fsnotify_update_flags(target) 3006 fsnotify_update_flags(target);
2842 } 3007 }
2843 if (!hlist_unhashed(&dentry->d_sib)) !! 3008 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2844 __hlist_del(&dentry->d_sib); <<
2845 hlist_add_head(&dentry->d_sib, &dentr <<
2846 __d_rehash(dentry); 3009 __d_rehash(dentry);
2847 fsnotify_update_flags(dentry); 3010 fsnotify_update_flags(dentry);
2848 fscrypt_handle_d_move(dentry); 3011 fscrypt_handle_d_move(dentry);
2849 3012
2850 write_seqcount_end(&target->d_seq); 3013 write_seqcount_end(&target->d_seq);
2851 write_seqcount_end(&dentry->d_seq); 3014 write_seqcount_end(&dentry->d_seq);
2852 3015
2853 if (dir) 3016 if (dir)
2854 end_dir_add(dir, n, d_wait); 3017 end_dir_add(dir, n, d_wait);
2855 3018
2856 if (dentry->d_parent != old_parent) 3019 if (dentry->d_parent != old_parent)
2857 spin_unlock(&dentry->d_parent 3020 spin_unlock(&dentry->d_parent->d_lock);
2858 if (dentry != old_parent) 3021 if (dentry != old_parent)
2859 spin_unlock(&old_parent->d_lo 3022 spin_unlock(&old_parent->d_lock);
2860 spin_unlock(&target->d_lock); 3023 spin_unlock(&target->d_lock);
2861 spin_unlock(&dentry->d_lock); 3024 spin_unlock(&dentry->d_lock);
2862 } 3025 }
2863 3026
2864 /* 3027 /*
2865 * d_move - move a dentry 3028 * d_move - move a dentry
2866 * @dentry: entry to move 3029 * @dentry: entry to move
2867 * @target: new dentry 3030 * @target: new dentry
2868 * 3031 *
2869 * Update the dcache to reflect the move of a 3032 * Update the dcache to reflect the move of a file name. Negative
2870 * dcache entries should not be moved in this 3033 * dcache entries should not be moved in this way. See the locking
2871 * requirements for __d_move. 3034 * requirements for __d_move.
2872 */ 3035 */
2873 void d_move(struct dentry *dentry, struct den 3036 void d_move(struct dentry *dentry, struct dentry *target)
2874 { 3037 {
2875 write_seqlock(&rename_lock); 3038 write_seqlock(&rename_lock);
2876 __d_move(dentry, target, false); 3039 __d_move(dentry, target, false);
2877 write_sequnlock(&rename_lock); 3040 write_sequnlock(&rename_lock);
2878 } 3041 }
2879 EXPORT_SYMBOL(d_move); 3042 EXPORT_SYMBOL(d_move);
2880 3043
2881 /* 3044 /*
2882 * d_exchange - exchange two dentries 3045 * d_exchange - exchange two dentries
2883 * @dentry1: first dentry 3046 * @dentry1: first dentry
2884 * @dentry2: second dentry 3047 * @dentry2: second dentry
2885 */ 3048 */
2886 void d_exchange(struct dentry *dentry1, struc 3049 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2887 { 3050 {
2888 write_seqlock(&rename_lock); 3051 write_seqlock(&rename_lock);
2889 3052
2890 WARN_ON(!dentry1->d_inode); 3053 WARN_ON(!dentry1->d_inode);
2891 WARN_ON(!dentry2->d_inode); 3054 WARN_ON(!dentry2->d_inode);
2892 WARN_ON(IS_ROOT(dentry1)); 3055 WARN_ON(IS_ROOT(dentry1));
2893 WARN_ON(IS_ROOT(dentry2)); 3056 WARN_ON(IS_ROOT(dentry2));
2894 3057
2895 __d_move(dentry1, dentry2, true); 3058 __d_move(dentry1, dentry2, true);
2896 3059
2897 write_sequnlock(&rename_lock); 3060 write_sequnlock(&rename_lock);
2898 } 3061 }
2899 3062
2900 /** 3063 /**
2901 * d_ancestor - search for an ancestor 3064 * d_ancestor - search for an ancestor
2902 * @p1: ancestor dentry 3065 * @p1: ancestor dentry
2903 * @p2: child dentry 3066 * @p2: child dentry
2904 * 3067 *
2905 * Returns the ancestor dentry of p2 which is 3068 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2906 * an ancestor of p2, else NULL. 3069 * an ancestor of p2, else NULL.
2907 */ 3070 */
2908 struct dentry *d_ancestor(struct dentry *p1, 3071 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2909 { 3072 {
2910 struct dentry *p; 3073 struct dentry *p;
2911 3074
2912 for (p = p2; !IS_ROOT(p); p = p->d_pa 3075 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2913 if (p->d_parent == p1) 3076 if (p->d_parent == p1)
2914 return p; 3077 return p;
2915 } 3078 }
2916 return NULL; 3079 return NULL;
2917 } 3080 }
2918 3081
2919 /* 3082 /*
2920 * This helper attempts to cope with remotely 3083 * This helper attempts to cope with remotely renamed directories
2921 * 3084 *
2922 * It assumes that the caller is already hold 3085 * It assumes that the caller is already holding
2923 * dentry->d_parent->d_inode->i_mutex, and re 3086 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2924 * 3087 *
2925 * Note: If ever the locking in lock_rename() 3088 * Note: If ever the locking in lock_rename() changes, then please
2926 * remember to update this too... 3089 * remember to update this too...
2927 */ 3090 */
2928 static int __d_unalias(struct dentry *dentry, !! 3091 static int __d_unalias(struct inode *inode,
>> 3092 struct dentry *dentry, struct dentry *alias)
2929 { 3093 {
2930 struct mutex *m1 = NULL; 3094 struct mutex *m1 = NULL;
2931 struct rw_semaphore *m2 = NULL; 3095 struct rw_semaphore *m2 = NULL;
2932 int ret = -ESTALE; 3096 int ret = -ESTALE;
2933 3097
2934 /* If alias and dentry share a parent 3098 /* If alias and dentry share a parent, then no extra locks required */
2935 if (alias->d_parent == dentry->d_pare 3099 if (alias->d_parent == dentry->d_parent)
2936 goto out_unalias; 3100 goto out_unalias;
2937 3101
2938 /* See lock_rename() */ 3102 /* See lock_rename() */
2939 if (!mutex_trylock(&dentry->d_sb->s_v 3103 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2940 goto out_err; 3104 goto out_err;
2941 m1 = &dentry->d_sb->s_vfs_rename_mute 3105 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2942 if (!inode_trylock_shared(alias->d_pa 3106 if (!inode_trylock_shared(alias->d_parent->d_inode))
2943 goto out_err; 3107 goto out_err;
2944 m2 = &alias->d_parent->d_inode->i_rws 3108 m2 = &alias->d_parent->d_inode->i_rwsem;
2945 out_unalias: 3109 out_unalias:
2946 __d_move(alias, dentry, false); 3110 __d_move(alias, dentry, false);
2947 ret = 0; 3111 ret = 0;
2948 out_err: 3112 out_err:
2949 if (m2) 3113 if (m2)
2950 up_read(m2); 3114 up_read(m2);
2951 if (m1) 3115 if (m1)
2952 mutex_unlock(m1); 3116 mutex_unlock(m1);
2953 return ret; 3117 return ret;
2954 } 3118 }
2955 3119
2956 /** 3120 /**
2957 * d_splice_alias - splice a disconnected den 3121 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2958 * @inode: the inode which may have a discon 3122 * @inode: the inode which may have a disconnected dentry
2959 * @dentry: a negative dentry which we want t 3123 * @dentry: a negative dentry which we want to point to the inode.
2960 * 3124 *
2961 * If inode is a directory and has an IS_ROOT 3125 * 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 3126 * place of the given dentry and return it, else simply d_add the inode
2963 * to the dentry and return NULL. 3127 * to the dentry and return NULL.
2964 * 3128 *
2965 * If a non-IS_ROOT directory is found, the f 3129 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2966 * we should error out: directories can't hav 3130 * we should error out: directories can't have multiple aliases.
2967 * 3131 *
2968 * This is needed in the lookup routine of an 3132 * This is needed in the lookup routine of any filesystem that is exportable
2969 * (via knfsd) so that we can build dcache pa 3133 * (via knfsd) so that we can build dcache paths to directories effectively.
2970 * 3134 *
2971 * If a dentry was found and moved, then it i 3135 * If a dentry was found and moved, then it is returned. Otherwise NULL
2972 * is returned. This matches the expected re 3136 * is returned. This matches the expected return value of ->lookup.
2973 * 3137 *
2974 * Cluster filesystems may call this function 3138 * Cluster filesystems may call this function with a negative, hashed dentry.
2975 * In that case, we know that the inode will 3139 * 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 3140 * will only occur during atomic_open. So we need to check for the dentry
2977 * being already hashed only in the final cas 3141 * being already hashed only in the final case.
2978 */ 3142 */
2979 struct dentry *d_splice_alias(struct inode *i 3143 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2980 { 3144 {
2981 if (IS_ERR(inode)) 3145 if (IS_ERR(inode))
2982 return ERR_CAST(inode); 3146 return ERR_CAST(inode);
2983 3147
2984 BUG_ON(!d_unhashed(dentry)); 3148 BUG_ON(!d_unhashed(dentry));
2985 3149
2986 if (!inode) 3150 if (!inode)
2987 goto out; 3151 goto out;
2988 3152
2989 security_d_instantiate(dentry, inode) 3153 security_d_instantiate(dentry, inode);
2990 spin_lock(&inode->i_lock); 3154 spin_lock(&inode->i_lock);
2991 if (S_ISDIR(inode->i_mode)) { 3155 if (S_ISDIR(inode->i_mode)) {
2992 struct dentry *new = __d_find 3156 struct dentry *new = __d_find_any_alias(inode);
2993 if (unlikely(new)) { 3157 if (unlikely(new)) {
2994 /* The reference to n 3158 /* The reference to new ensures it remains an alias */
2995 spin_unlock(&inode->i 3159 spin_unlock(&inode->i_lock);
2996 write_seqlock(&rename 3160 write_seqlock(&rename_lock);
2997 if (unlikely(d_ancest 3161 if (unlikely(d_ancestor(new, dentry))) {
2998 write_sequnlo 3162 write_sequnlock(&rename_lock);
2999 dput(new); 3163 dput(new);
3000 new = ERR_PTR 3164 new = ERR_PTR(-ELOOP);
3001 pr_warn_ratel 3165 pr_warn_ratelimited(
3002 "VFS: 3166 "VFS: Lookup of '%s' in %s %s"
3003 " wou 3167 " would have caused loop\n",
3004 dentr 3168 dentry->d_name.name,
3005 inode 3169 inode->i_sb->s_type->name,
3006 inode 3170 inode->i_sb->s_id);
3007 } else if (!IS_ROOT(n 3171 } else if (!IS_ROOT(new)) {
3008 struct dentry 3172 struct dentry *old_parent = dget(new->d_parent);
3009 int err = __d !! 3173 int err = __d_unalias(inode, dentry, new);
3010 write_sequnlo 3174 write_sequnlock(&rename_lock);
3011 if (err) { 3175 if (err) {
3012 dput( 3176 dput(new);
3013 new = 3177 new = ERR_PTR(err);
3014 } 3178 }
3015 dput(old_pare 3179 dput(old_parent);
3016 } else { 3180 } else {
3017 __d_move(new, 3181 __d_move(new, dentry, false);
3018 write_sequnlo 3182 write_sequnlock(&rename_lock);
3019 } 3183 }
3020 iput(inode); 3184 iput(inode);
3021 return new; 3185 return new;
3022 } 3186 }
3023 } 3187 }
3024 out: 3188 out:
3025 __d_add(dentry, inode); 3189 __d_add(dentry, inode);
3026 return NULL; 3190 return NULL;
3027 } 3191 }
3028 EXPORT_SYMBOL(d_splice_alias); 3192 EXPORT_SYMBOL(d_splice_alias);
3029 3193
3030 /* 3194 /*
3031 * Test whether new_dentry is a subdirectory 3195 * Test whether new_dentry is a subdirectory of old_dentry.
3032 * 3196 *
3033 * Trivially implemented using the dcache str 3197 * Trivially implemented using the dcache structure
3034 */ 3198 */
3035 3199
3036 /** 3200 /**
3037 * is_subdir - is new dentry a subdirectory o 3201 * is_subdir - is new dentry a subdirectory of old_dentry
3038 * @new_dentry: new dentry 3202 * @new_dentry: new dentry
3039 * @old_dentry: old dentry 3203 * @old_dentry: old dentry
3040 * 3204 *
3041 * Returns true if new_dentry is a subdirecto 3205 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3042 * Returns false otherwise. 3206 * Returns false otherwise.
3043 * Caller must ensure that "new_dentry" is pi 3207 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3044 */ 3208 */
3045 3209
3046 bool is_subdir(struct dentry *new_dentry, str 3210 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3047 { 3211 {
3048 bool subdir; 3212 bool subdir;
3049 unsigned seq; 3213 unsigned seq;
3050 3214
3051 if (new_dentry == old_dentry) 3215 if (new_dentry == old_dentry)
3052 return true; 3216 return true;
3053 3217
3054 /* Access d_parent under rcu as d_mov 3218 /* Access d_parent under rcu as d_move() may change it. */
3055 rcu_read_lock(); 3219 rcu_read_lock();
3056 seq = read_seqbegin(&rename_lock); 3220 seq = read_seqbegin(&rename_lock);
3057 subdir = d_ancestor(old_dentry, new_d 3221 subdir = d_ancestor(old_dentry, new_dentry);
3058 /* Try lockless once... */ 3222 /* Try lockless once... */
3059 if (read_seqretry(&rename_lock, seq)) 3223 if (read_seqretry(&rename_lock, seq)) {
3060 /* ...else acquire lock for p 3224 /* ...else acquire lock for progress even on deep chains. */
3061 read_seqlock_excl(&rename_loc 3225 read_seqlock_excl(&rename_lock);
3062 subdir = d_ancestor(old_dentr 3226 subdir = d_ancestor(old_dentry, new_dentry);
3063 read_sequnlock_excl(&rename_l 3227 read_sequnlock_excl(&rename_lock);
3064 } 3228 }
3065 rcu_read_unlock(); 3229 rcu_read_unlock();
3066 return subdir; 3230 return subdir;
3067 } 3231 }
3068 EXPORT_SYMBOL(is_subdir); 3232 EXPORT_SYMBOL(is_subdir);
3069 3233
3070 static enum d_walk_ret d_genocide_kill(void * 3234 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3071 { 3235 {
3072 struct dentry *root = data; 3236 struct dentry *root = data;
3073 if (dentry != root) { 3237 if (dentry != root) {
3074 if (d_unhashed(dentry) || !de 3238 if (d_unhashed(dentry) || !dentry->d_inode)
3075 return D_WALK_SKIP; 3239 return D_WALK_SKIP;
3076 3240
3077 if (!(dentry->d_flags & DCACH 3241 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3078 dentry->d_flags |= DC 3242 dentry->d_flags |= DCACHE_GENOCIDE;
3079 dentry->d_lockref.cou 3243 dentry->d_lockref.count--;
3080 } 3244 }
3081 } 3245 }
3082 return D_WALK_CONTINUE; 3246 return D_WALK_CONTINUE;
3083 } 3247 }
3084 3248
3085 void d_genocide(struct dentry *parent) 3249 void d_genocide(struct dentry *parent)
3086 { 3250 {
3087 d_walk(parent, parent, d_genocide_kil 3251 d_walk(parent, parent, d_genocide_kill);
3088 } 3252 }
3089 3253
3090 void d_mark_tmpfile(struct file *file, struct !! 3254 EXPORT_SYMBOL(d_genocide);
>> 3255
>> 3256 void d_tmpfile(struct file *file, struct inode *inode)
3091 { 3257 {
3092 struct dentry *dentry = file->f_path. 3258 struct dentry *dentry = file->f_path.dentry;
3093 3259
>> 3260 inode_dec_link_count(inode);
3094 BUG_ON(dentry->d_name.name != dentry- 3261 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3095 !hlist_unhashed(&dentry->d_u. 3262 !hlist_unhashed(&dentry->d_u.d_alias) ||
3096 !d_unlinked(dentry)); 3263 !d_unlinked(dentry));
3097 spin_lock(&dentry->d_parent->d_lock); 3264 spin_lock(&dentry->d_parent->d_lock);
3098 spin_lock_nested(&dentry->d_lock, DEN 3265 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3099 dentry->d_name.len = sprintf(dentry-> 3266 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3100 (unsigned lon 3267 (unsigned long long)inode->i_ino);
3101 spin_unlock(&dentry->d_lock); 3268 spin_unlock(&dentry->d_lock);
3102 spin_unlock(&dentry->d_parent->d_lock 3269 spin_unlock(&dentry->d_parent->d_lock);
3103 } <<
3104 EXPORT_SYMBOL(d_mark_tmpfile); <<
3105 <<
3106 void d_tmpfile(struct file *file, struct inod <<
3107 { <<
3108 struct dentry *dentry = file->f_path. <<
3109 <<
3110 inode_dec_link_count(inode); <<
3111 d_mark_tmpfile(file, inode); <<
3112 d_instantiate(dentry, inode); 3270 d_instantiate(dentry, inode);
3113 } 3271 }
3114 EXPORT_SYMBOL(d_tmpfile); 3272 EXPORT_SYMBOL(d_tmpfile);
3115 3273
3116 /* <<
3117 * Obtain inode number of the parent dentry. <<
3118 */ <<
3119 ino_t d_parent_ino(struct dentry *dentry) <<
3120 { <<
3121 struct dentry *parent; <<
3122 struct inode *iparent; <<
3123 unsigned seq; <<
3124 ino_t ret; <<
3125 <<
3126 scoped_guard(rcu) { <<
3127 seq = raw_seqcount_begin(&den <<
3128 parent = READ_ONCE(dentry->d_ <<
3129 iparent = d_inode_rcu(parent) <<
3130 if (likely(iparent)) { <<
3131 ret = iparent->i_ino; <<
3132 if (!read_seqcount_re <<
3133 return ret; <<
3134 } <<
3135 } <<
3136 <<
3137 spin_lock(&dentry->d_lock); <<
3138 ret = dentry->d_parent->d_inode->i_in <<
3139 spin_unlock(&dentry->d_lock); <<
3140 return ret; <<
3141 } <<
3142 EXPORT_SYMBOL(d_parent_ino); <<
3143 <<
3144 static __initdata unsigned long dhash_entries 3274 static __initdata unsigned long dhash_entries;
3145 static int __init set_dhash_entries(char *str 3275 static int __init set_dhash_entries(char *str)
3146 { 3276 {
3147 if (!str) 3277 if (!str)
3148 return 0; 3278 return 0;
3149 dhash_entries = simple_strtoul(str, & 3279 dhash_entries = simple_strtoul(str, &str, 0);
3150 return 1; 3280 return 1;
3151 } 3281 }
3152 __setup("dhash_entries=", set_dhash_entries); 3282 __setup("dhash_entries=", set_dhash_entries);
3153 3283
3154 static void __init dcache_init_early(void) 3284 static void __init dcache_init_early(void)
3155 { 3285 {
3156 /* If hashes are distributed across N 3286 /* If hashes are distributed across NUMA nodes, defer
3157 * hash allocation until vmalloc spac 3287 * hash allocation until vmalloc space is available.
3158 */ 3288 */
3159 if (hashdist) 3289 if (hashdist)
3160 return; 3290 return;
3161 3291
3162 dentry_hashtable = 3292 dentry_hashtable =
3163 alloc_large_system_hash("Dent 3293 alloc_large_system_hash("Dentry cache",
3164 sizeo 3294 sizeof(struct hlist_bl_head),
3165 dhash 3295 dhash_entries,
3166 13, 3296 13,
3167 HASH_ 3297 HASH_EARLY | HASH_ZERO,
3168 &d_ha 3298 &d_hash_shift,
3169 NULL, 3299 NULL,
3170 0, 3300 0,
3171 0); 3301 0);
3172 d_hash_shift = 32 - d_hash_shift; 3302 d_hash_shift = 32 - d_hash_shift;
3173 <<
3174 runtime_const_init(shift, d_hash_shif <<
3175 runtime_const_init(ptr, dentry_hashta <<
3176 } 3303 }
3177 3304
3178 static void __init dcache_init(void) 3305 static void __init dcache_init(void)
3179 { 3306 {
3180 /* 3307 /*
3181 * A constructor could be added for s 3308 * A constructor could be added for stable state like the lists,
3182 * but it is probably not worth it be 3309 * but it is probably not worth it because of the cache nature
3183 * of the dcache. 3310 * of the dcache.
3184 */ 3311 */
3185 dentry_cache = KMEM_CACHE_USERCOPY(de 3312 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3186 SLAB_RECLAIM_ACCOUNT|SLAB_PAN !! 3313 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3187 d_iname); 3314 d_iname);
3188 3315
3189 /* Hash may have been set up in dcach 3316 /* Hash may have been set up in dcache_init_early */
3190 if (!hashdist) 3317 if (!hashdist)
3191 return; 3318 return;
3192 3319
3193 dentry_hashtable = 3320 dentry_hashtable =
3194 alloc_large_system_hash("Dent 3321 alloc_large_system_hash("Dentry cache",
3195 sizeo 3322 sizeof(struct hlist_bl_head),
3196 dhash 3323 dhash_entries,
3197 13, 3324 13,
3198 HASH_ 3325 HASH_ZERO,
3199 &d_ha 3326 &d_hash_shift,
3200 NULL, 3327 NULL,
3201 0, 3328 0,
3202 0); 3329 0);
3203 d_hash_shift = 32 - d_hash_shift; 3330 d_hash_shift = 32 - d_hash_shift;
3204 <<
3205 runtime_const_init(shift, d_hash_shif <<
3206 runtime_const_init(ptr, dentry_hashta <<
3207 } 3331 }
3208 3332
3209 /* SLAB cache for __getname() consumers */ 3333 /* SLAB cache for __getname() consumers */
3210 struct kmem_cache *names_cachep __ro_after_in !! 3334 struct kmem_cache *names_cachep __read_mostly;
3211 EXPORT_SYMBOL(names_cachep); 3335 EXPORT_SYMBOL(names_cachep);
3212 3336
3213 void __init vfs_caches_init_early(void) 3337 void __init vfs_caches_init_early(void)
3214 { 3338 {
3215 int i; 3339 int i;
3216 3340
3217 for (i = 0; i < ARRAY_SIZE(in_lookup_ 3341 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3218 INIT_HLIST_BL_HEAD(&in_lookup 3342 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3219 3343
3220 dcache_init_early(); 3344 dcache_init_early();
3221 inode_init_early(); 3345 inode_init_early();
3222 } 3346 }
3223 3347
3224 void __init vfs_caches_init(void) 3348 void __init vfs_caches_init(void)
3225 { 3349 {
3226 names_cachep = kmem_cache_create_user 3350 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3227 SLAB_HWCACHE_ALIGN|SL 3351 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3228 3352
3229 dcache_init(); 3353 dcache_init();
3230 inode_init(); 3354 inode_init();
3231 files_init(); 3355 files_init();
3232 files_maxfiles_init(); 3356 files_maxfiles_init();
3233 mnt_init(); 3357 mnt_init();
3234 bdev_cache_init(); 3358 bdev_cache_init();
3235 chrdev_init(); 3359 chrdev_init();
3236 } 3360 }
3237 3361
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.