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