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