1 // SPDX-License-Identifier: GPL-2.0-only 1 // SPDX-License-Identifier: GPL-2.0-only
2 /* 2 /*
3 * fs/dcache.c 3 * fs/dcache.c
4 * 4 *
5 * Complete reimplementation 5 * Complete reimplementation
6 * (C) 1997 Thomas Schoebel-Theuer, 6 * (C) 1997 Thomas Schoebel-Theuer,
7 * with heavy changes by Linus Torvalds 7 * with heavy changes by Linus Torvalds
8 */ 8 */
9 9
10 /* 10 /*
11 * Notes on the allocation strategy: 11 * Notes on the allocation strategy:
12 * 12 *
13 * The dcache is a master of the icache - when 13 * The dcache is a master of the icache - whenever a dcache entry
14 * exists, the inode will always exist. "iput( 14 * exists, the inode will always exist. "iput()" is done either when
15 * the dcache entry is deleted or garbage coll 15 * the dcache entry is deleted or garbage collected.
16 */ 16 */
17 17
18 #include <linux/ratelimit.h> 18 #include <linux/ratelimit.h>
19 #include <linux/string.h> 19 #include <linux/string.h>
20 #include <linux/mm.h> 20 #include <linux/mm.h>
21 #include <linux/fs.h> 21 #include <linux/fs.h>
22 #include <linux/fscrypt.h> 22 #include <linux/fscrypt.h>
23 #include <linux/fsnotify.h> 23 #include <linux/fsnotify.h>
24 #include <linux/slab.h> 24 #include <linux/slab.h>
25 #include <linux/init.h> 25 #include <linux/init.h>
26 #include <linux/hash.h> 26 #include <linux/hash.h>
27 #include <linux/cache.h> 27 #include <linux/cache.h>
28 #include <linux/export.h> 28 #include <linux/export.h>
29 #include <linux/security.h> 29 #include <linux/security.h>
30 #include <linux/seqlock.h> 30 #include <linux/seqlock.h>
31 #include <linux/memblock.h> 31 #include <linux/memblock.h>
32 #include <linux/bit_spinlock.h> 32 #include <linux/bit_spinlock.h>
33 #include <linux/rculist_bl.h> 33 #include <linux/rculist_bl.h>
34 #include <linux/list_lru.h> 34 #include <linux/list_lru.h>
35 #include "internal.h" 35 #include "internal.h"
36 #include "mount.h" 36 #include "mount.h"
37 37
38 #include <asm/runtime-const.h> <<
39 <<
40 /* 38 /*
41 * Usage: 39 * Usage:
42 * dcache->d_inode->i_lock protects: 40 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_u.d_alias, d_inode of alias 41 * - i_dentry, d_u.d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects: 42 * dcache_hash_bucket lock protects:
45 * - the dcache hash table 43 * - the dcache hash table
46 * s_roots bl list spinlock protects: 44 * s_roots bl list spinlock protects:
47 * - the s_roots list (see __d_drop) 45 * - the s_roots list (see __d_drop)
48 * dentry->d_sb->s_dentry_lru_lock protects: 46 * dentry->d_sb->s_dentry_lru_lock protects:
49 * - the dcache lru lists and counters 47 * - the dcache lru lists and counters
50 * d_lock protects: 48 * d_lock protects:
51 * - d_flags 49 * - d_flags
52 * - d_name 50 * - d_name
53 * - d_lru 51 * - d_lru
54 * - d_count 52 * - d_count
55 * - d_unhashed() 53 * - d_unhashed()
56 * - d_parent and d_chilren !! 54 * - d_parent and d_subdirs
57 * - childrens' d_sib and d_parent !! 55 * - childrens' d_child and d_parent
58 * - d_u.d_alias, d_inode 56 * - d_u.d_alias, d_inode
59 * 57 *
60 * Ordering: 58 * Ordering:
61 * dentry->d_inode->i_lock 59 * dentry->d_inode->i_lock
62 * dentry->d_lock 60 * dentry->d_lock
63 * dentry->d_sb->s_dentry_lru_lock 61 * dentry->d_sb->s_dentry_lru_lock
64 * dcache_hash_bucket lock 62 * dcache_hash_bucket lock
65 * s_roots lock 63 * s_roots lock
66 * 64 *
67 * If there is an ancestor relationship: 65 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock 66 * dentry->d_parent->...->d_parent->d_lock
69 * ... 67 * ...
70 * dentry->d_parent->d_lock 68 * dentry->d_parent->d_lock
71 * dentry->d_lock 69 * dentry->d_lock
72 * 70 *
73 * If no ancestor relationship: 71 * If no ancestor relationship:
74 * arbitrary, since it's serialized on rename_ 72 * arbitrary, since it's serialized on rename_lock
75 */ 73 */
76 int sysctl_vfs_cache_pressure __read_mostly = 74 int sysctl_vfs_cache_pressure __read_mostly = 100;
77 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); 75 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
78 76
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rena 77 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
80 78
81 EXPORT_SYMBOL(rename_lock); 79 EXPORT_SYMBOL(rename_lock);
82 80
83 static struct kmem_cache *dentry_cache __ro_af !! 81 static struct kmem_cache *dentry_cache __read_mostly;
84 82
85 const struct qstr empty_name = QSTR_INIT("", 0 83 const struct qstr empty_name = QSTR_INIT("", 0);
86 EXPORT_SYMBOL(empty_name); 84 EXPORT_SYMBOL(empty_name);
87 const struct qstr slash_name = QSTR_INIT("/", 85 const struct qstr slash_name = QSTR_INIT("/", 1);
88 EXPORT_SYMBOL(slash_name); 86 EXPORT_SYMBOL(slash_name);
89 const struct qstr dotdot_name = QSTR_INIT(".." 87 const struct qstr dotdot_name = QSTR_INIT("..", 2);
90 EXPORT_SYMBOL(dotdot_name); 88 EXPORT_SYMBOL(dotdot_name);
91 89
92 /* 90 /*
93 * This is the single most critical data struc 91 * This is the single most critical data structure when it comes
94 * to the dcache: the hashtable for lookups. S 92 * to the dcache: the hashtable for lookups. Somebody should try
95 * to make this good - I've just made it work. 93 * to make this good - I've just made it work.
96 * 94 *
97 * This hash-function tries to avoid losing to 95 * This hash-function tries to avoid losing too many bits of hash
98 * information, yet avoid using a prime hash-s 96 * information, yet avoid using a prime hash-size or similar.
99 * <<
100 * Marking the variables "used" ensures that t <<
101 * optimize them away completely on architectu <<
102 * constant infrastructure, this allows debugg <<
103 * values. But updating these values has no ef <<
104 */ 97 */
105 98
106 static unsigned int d_hash_shift __ro_after_in !! 99 static unsigned int d_hash_shift __read_mostly;
107 100
108 static struct hlist_bl_head *dentry_hashtable !! 101 static struct hlist_bl_head *dentry_hashtable __read_mostly;
109 102
110 static inline struct hlist_bl_head *d_hash(uns !! 103 static inline struct hlist_bl_head *d_hash(unsigned int hash)
111 { 104 {
112 return runtime_const_ptr(dentry_hashta !! 105 return dentry_hashtable + (hash >> d_hash_shift);
113 runtime_const_shift_right_32(h <<
114 } 106 }
115 107
116 #define IN_LOOKUP_SHIFT 10 108 #define IN_LOOKUP_SHIFT 10
117 static struct hlist_bl_head in_lookup_hashtabl 109 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
118 110
119 static inline struct hlist_bl_head *in_lookup_ 111 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
120 unsign 112 unsigned int hash)
121 { 113 {
122 hash += (unsigned long) parent / L1_CA 114 hash += (unsigned long) parent / L1_CACHE_BYTES;
123 return in_lookup_hashtable + hash_32(h 115 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
124 } 116 }
125 117
126 struct dentry_stat_t { !! 118
127 long nr_dentry; !! 119 /* Statistics gathering. */
128 long nr_unused; !! 120 struct dentry_stat_t dentry_stat = {
129 long age_limit; /* age in seco !! 121 .age_limit = 45,
130 long want_pages; /* pages reque <<
131 long nr_negative; /* # of unused <<
132 long dummy; /* Reserved fo <<
133 }; 122 };
134 123
135 static DEFINE_PER_CPU(long, nr_dentry); 124 static DEFINE_PER_CPU(long, nr_dentry);
136 static DEFINE_PER_CPU(long, nr_dentry_unused); 125 static DEFINE_PER_CPU(long, nr_dentry_unused);
137 static DEFINE_PER_CPU(long, nr_dentry_negative 126 static DEFINE_PER_CPU(long, nr_dentry_negative);
138 127
139 #if defined(CONFIG_SYSCTL) && defined(CONFIG_P 128 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
140 /* Statistics gathering. */ <<
141 static struct dentry_stat_t dentry_stat = { <<
142 .age_limit = 45, <<
143 }; <<
144 129
145 /* 130 /*
146 * Here we resort to our own counters instead 131 * Here we resort to our own counters instead of using generic per-cpu counters
147 * for consistency with what the vfs inode cod 132 * for consistency with what the vfs inode code does. We are expected to harvest
148 * better code and performance by having our o 133 * better code and performance by having our own specialized counters.
149 * 134 *
150 * Please note that the loop is done over all 135 * 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 136 * 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 137 * on and off. If one of them goes off, we will just keep their counters.
153 * 138 *
154 * glommer: See cffbc8a for details, and if yo 139 * glommer: See cffbc8a for details, and if you ever intend to change this,
155 * please update all vfs counters to match. 140 * please update all vfs counters to match.
156 */ 141 */
157 static long get_nr_dentry(void) 142 static long get_nr_dentry(void)
158 { 143 {
159 int i; 144 int i;
160 long sum = 0; 145 long sum = 0;
161 for_each_possible_cpu(i) 146 for_each_possible_cpu(i)
162 sum += per_cpu(nr_dentry, i); 147 sum += per_cpu(nr_dentry, i);
163 return sum < 0 ? 0 : sum; 148 return sum < 0 ? 0 : sum;
164 } 149 }
165 150
166 static long get_nr_dentry_unused(void) 151 static long get_nr_dentry_unused(void)
167 { 152 {
168 int i; 153 int i;
169 long sum = 0; 154 long sum = 0;
170 for_each_possible_cpu(i) 155 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry_unuse 156 sum += per_cpu(nr_dentry_unused, i);
172 return sum < 0 ? 0 : sum; 157 return sum < 0 ? 0 : sum;
173 } 158 }
174 159
175 static long get_nr_dentry_negative(void) 160 static long get_nr_dentry_negative(void)
176 { 161 {
177 int i; 162 int i;
178 long sum = 0; 163 long sum = 0;
179 164
180 for_each_possible_cpu(i) 165 for_each_possible_cpu(i)
181 sum += per_cpu(nr_dentry_negat 166 sum += per_cpu(nr_dentry_negative, i);
182 return sum < 0 ? 0 : sum; 167 return sum < 0 ? 0 : sum;
183 } 168 }
184 169
185 static int proc_nr_dentry(const struct ctl_tab !! 170 int proc_nr_dentry(struct ctl_table *table, int write, void *buffer,
186 size_t *lenp, loff_t !! 171 size_t *lenp, loff_t *ppos)
187 { 172 {
188 dentry_stat.nr_dentry = get_nr_dentry( 173 dentry_stat.nr_dentry = get_nr_dentry();
189 dentry_stat.nr_unused = get_nr_dentry_ 174 dentry_stat.nr_unused = get_nr_dentry_unused();
190 dentry_stat.nr_negative = get_nr_dentr 175 dentry_stat.nr_negative = get_nr_dentry_negative();
191 return proc_doulongvec_minmax(table, w 176 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
192 } 177 }
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 178 #endif
211 179
212 /* 180 /*
213 * Compare 2 name strings, return 0 if they ma 181 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
214 * The strings are both count bytes long, and 182 * The strings are both count bytes long, and count is non-zero.
215 */ 183 */
216 #ifdef CONFIG_DCACHE_WORD_ACCESS 184 #ifdef CONFIG_DCACHE_WORD_ACCESS
217 185
218 #include <asm/word-at-a-time.h> 186 #include <asm/word-at-a-time.h>
219 /* 187 /*
220 * NOTE! 'cs' and 'scount' come from a dentry, 188 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
221 * aligned allocation for this particular comp 189 * aligned allocation for this particular component. We don't
222 * strictly need the load_unaligned_zeropad() 190 * strictly need the load_unaligned_zeropad() safety, but it
223 * doesn't hurt either. 191 * doesn't hurt either.
224 * 192 *
225 * In contrast, 'ct' and 'tcount' can be from 193 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
226 * need the careful unaligned handling. 194 * need the careful unaligned handling.
227 */ 195 */
228 static inline int dentry_string_cmp(const unsi 196 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
229 { 197 {
230 unsigned long a,b,mask; 198 unsigned long a,b,mask;
231 199
232 for (;;) { 200 for (;;) {
233 a = read_word_at_a_time(cs); 201 a = read_word_at_a_time(cs);
234 b = load_unaligned_zeropad(ct) 202 b = load_unaligned_zeropad(ct);
235 if (tcount < sizeof(unsigned l 203 if (tcount < sizeof(unsigned long))
236 break; 204 break;
237 if (unlikely(a != b)) 205 if (unlikely(a != b))
238 return 1; 206 return 1;
239 cs += sizeof(unsigned long); 207 cs += sizeof(unsigned long);
240 ct += sizeof(unsigned long); 208 ct += sizeof(unsigned long);
241 tcount -= sizeof(unsigned long 209 tcount -= sizeof(unsigned long);
242 if (!tcount) 210 if (!tcount)
243 return 0; 211 return 0;
244 } 212 }
245 mask = bytemask_from_count(tcount); 213 mask = bytemask_from_count(tcount);
246 return unlikely(!!((a ^ b) & mask)); 214 return unlikely(!!((a ^ b) & mask));
247 } 215 }
248 216
249 #else 217 #else
250 218
251 static inline int dentry_string_cmp(const unsi 219 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
252 { 220 {
253 do { 221 do {
254 if (*cs != *ct) 222 if (*cs != *ct)
255 return 1; 223 return 1;
256 cs++; 224 cs++;
257 ct++; 225 ct++;
258 tcount--; 226 tcount--;
259 } while (tcount); 227 } while (tcount);
260 return 0; 228 return 0;
261 } 229 }
262 230
263 #endif 231 #endif
264 232
265 static inline int dentry_cmp(const struct dent 233 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
266 { 234 {
267 /* 235 /*
268 * Be careful about RCU walk racing wi 236 * Be careful about RCU walk racing with rename:
269 * use 'READ_ONCE' to fetch the name p 237 * use 'READ_ONCE' to fetch the name pointer.
270 * 238 *
271 * NOTE! Even if a rename will mean th 239 * NOTE! Even if a rename will mean that the length
272 * was not loaded atomically, we don't 240 * was not loaded atomically, we don't care. The
273 * RCU walk will check the sequence co 241 * RCU walk will check the sequence count eventually,
274 * and catch it. And we won't overrun 242 * and catch it. And we won't overrun the buffer,
275 * because we're reading the name poin 243 * because we're reading the name pointer atomically,
276 * and a dentry name is guaranteed to 244 * and a dentry name is guaranteed to be properly
277 * terminated with a NUL byte. 245 * terminated with a NUL byte.
278 * 246 *
279 * End result: even if 'len' is wrong, 247 * End result: even if 'len' is wrong, we'll exit
280 * early because the data cannot match 248 * early because the data cannot match (there can
281 * be no NUL in the ct/tcount data) 249 * be no NUL in the ct/tcount data)
282 */ 250 */
283 const unsigned char *cs = READ_ONCE(de 251 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
284 252
285 return dentry_string_cmp(cs, ct, tcoun 253 return dentry_string_cmp(cs, ct, tcount);
286 } 254 }
287 255
288 struct external_name { 256 struct external_name {
289 union { 257 union {
290 atomic_t count; 258 atomic_t count;
291 struct rcu_head head; 259 struct rcu_head head;
292 } u; 260 } u;
293 unsigned char name[]; 261 unsigned char name[];
294 }; 262 };
295 263
296 static inline struct external_name *external_n 264 static inline struct external_name *external_name(struct dentry *dentry)
297 { 265 {
298 return container_of(dentry->d_name.nam 266 return container_of(dentry->d_name.name, struct external_name, name[0]);
299 } 267 }
300 268
301 static void __d_free(struct rcu_head *head) 269 static void __d_free(struct rcu_head *head)
302 { 270 {
303 struct dentry *dentry = container_of(h 271 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
304 272
305 kmem_cache_free(dentry_cache, dentry); 273 kmem_cache_free(dentry_cache, dentry);
306 } 274 }
307 275
308 static void __d_free_external(struct rcu_head 276 static void __d_free_external(struct rcu_head *head)
309 { 277 {
310 struct dentry *dentry = container_of(h 278 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
311 kfree(external_name(dentry)); 279 kfree(external_name(dentry));
312 kmem_cache_free(dentry_cache, dentry); 280 kmem_cache_free(dentry_cache, dentry);
313 } 281 }
314 282
315 static inline int dname_external(const struct 283 static inline int dname_external(const struct dentry *dentry)
316 { 284 {
317 return dentry->d_name.name != dentry-> 285 return dentry->d_name.name != dentry->d_iname;
318 } 286 }
319 287
320 void take_dentry_name_snapshot(struct name_sna 288 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
321 { 289 {
322 spin_lock(&dentry->d_lock); 290 spin_lock(&dentry->d_lock);
323 name->name = dentry->d_name; 291 name->name = dentry->d_name;
324 if (unlikely(dname_external(dentry))) 292 if (unlikely(dname_external(dentry))) {
325 atomic_inc(&external_name(dent 293 atomic_inc(&external_name(dentry)->u.count);
326 } else { 294 } else {
327 memcpy(name->inline_name, dent 295 memcpy(name->inline_name, dentry->d_iname,
328 dentry->d_name.len + 1) 296 dentry->d_name.len + 1);
329 name->name.name = name->inline 297 name->name.name = name->inline_name;
330 } 298 }
331 spin_unlock(&dentry->d_lock); 299 spin_unlock(&dentry->d_lock);
332 } 300 }
333 EXPORT_SYMBOL(take_dentry_name_snapshot); 301 EXPORT_SYMBOL(take_dentry_name_snapshot);
334 302
335 void release_dentry_name_snapshot(struct name_ 303 void release_dentry_name_snapshot(struct name_snapshot *name)
336 { 304 {
337 if (unlikely(name->name.name != name-> 305 if (unlikely(name->name.name != name->inline_name)) {
338 struct external_name *p; 306 struct external_name *p;
339 p = container_of(name->name.na 307 p = container_of(name->name.name, struct external_name, name[0]);
340 if (unlikely(atomic_dec_and_te 308 if (unlikely(atomic_dec_and_test(&p->u.count)))
341 kfree_rcu(p, u.head); 309 kfree_rcu(p, u.head);
342 } 310 }
343 } 311 }
344 EXPORT_SYMBOL(release_dentry_name_snapshot); 312 EXPORT_SYMBOL(release_dentry_name_snapshot);
345 313
346 static inline void __d_set_inode_and_type(stru 314 static inline void __d_set_inode_and_type(struct dentry *dentry,
347 stru 315 struct inode *inode,
348 unsi 316 unsigned type_flags)
349 { 317 {
350 unsigned flags; 318 unsigned flags;
351 319
352 dentry->d_inode = inode; 320 dentry->d_inode = inode;
353 flags = READ_ONCE(dentry->d_flags); 321 flags = READ_ONCE(dentry->d_flags);
354 flags &= ~DCACHE_ENTRY_TYPE; !! 322 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
355 flags |= type_flags; 323 flags |= type_flags;
356 smp_store_release(&dentry->d_flags, fl 324 smp_store_release(&dentry->d_flags, flags);
357 } 325 }
358 326
359 static inline void __d_clear_type_and_inode(st 327 static inline void __d_clear_type_and_inode(struct dentry *dentry)
360 { 328 {
361 unsigned flags = READ_ONCE(dentry->d_f 329 unsigned flags = READ_ONCE(dentry->d_flags);
362 330
363 flags &= ~DCACHE_ENTRY_TYPE; !! 331 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
364 WRITE_ONCE(dentry->d_flags, flags); 332 WRITE_ONCE(dentry->d_flags, flags);
365 dentry->d_inode = NULL; 333 dentry->d_inode = NULL;
366 /* !! 334 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 335 this_cpu_inc(nr_dentry_negative);
372 } 336 }
373 337
374 static void dentry_free(struct dentry *dentry) 338 static void dentry_free(struct dentry *dentry)
375 { 339 {
376 WARN_ON(!hlist_unhashed(&dentry->d_u.d 340 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
377 if (unlikely(dname_external(dentry))) 341 if (unlikely(dname_external(dentry))) {
378 struct external_name *p = exte 342 struct external_name *p = external_name(dentry);
379 if (likely(atomic_dec_and_test 343 if (likely(atomic_dec_and_test(&p->u.count))) {
380 call_rcu(&dentry->d_u. 344 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
381 return; 345 return;
382 } 346 }
383 } 347 }
384 /* if dentry was never visible to RCU, 348 /* if dentry was never visible to RCU, immediate free is OK */
385 if (dentry->d_flags & DCACHE_NORCU) 349 if (dentry->d_flags & DCACHE_NORCU)
386 __d_free(&dentry->d_u.d_rcu); 350 __d_free(&dentry->d_u.d_rcu);
387 else 351 else
388 call_rcu(&dentry->d_u.d_rcu, _ 352 call_rcu(&dentry->d_u.d_rcu, __d_free);
389 } 353 }
390 354
391 /* 355 /*
392 * Release the dentry's inode, using the files 356 * Release the dentry's inode, using the filesystem
393 * d_iput() operation if defined. 357 * d_iput() operation if defined.
394 */ 358 */
395 static void dentry_unlink_inode(struct dentry 359 static void dentry_unlink_inode(struct dentry * dentry)
396 __releases(dentry->d_lock) 360 __releases(dentry->d_lock)
397 __releases(dentry->d_inode->i_lock) 361 __releases(dentry->d_inode->i_lock)
398 { 362 {
399 struct inode *inode = dentry->d_inode; 363 struct inode *inode = dentry->d_inode;
400 364
401 raw_write_seqcount_begin(&dentry->d_se 365 raw_write_seqcount_begin(&dentry->d_seq);
402 __d_clear_type_and_inode(dentry); 366 __d_clear_type_and_inode(dentry);
403 hlist_del_init(&dentry->d_u.d_alias); 367 hlist_del_init(&dentry->d_u.d_alias);
404 raw_write_seqcount_end(&dentry->d_seq) 368 raw_write_seqcount_end(&dentry->d_seq);
405 spin_unlock(&dentry->d_lock); 369 spin_unlock(&dentry->d_lock);
406 spin_unlock(&inode->i_lock); 370 spin_unlock(&inode->i_lock);
407 if (!inode->i_nlink) 371 if (!inode->i_nlink)
408 fsnotify_inoderemove(inode); 372 fsnotify_inoderemove(inode);
409 if (dentry->d_op && dentry->d_op->d_ip 373 if (dentry->d_op && dentry->d_op->d_iput)
410 dentry->d_op->d_iput(dentry, i 374 dentry->d_op->d_iput(dentry, inode);
411 else 375 else
412 iput(inode); 376 iput(inode);
413 } 377 }
414 378
415 /* 379 /*
416 * The DCACHE_LRU_LIST bit is set whenever the 380 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
417 * is in use - which includes both the "real" 381 * is in use - which includes both the "real" per-superblock
418 * LRU list _and_ the DCACHE_SHRINK_LIST use. 382 * LRU list _and_ the DCACHE_SHRINK_LIST use.
419 * 383 *
420 * The DCACHE_SHRINK_LIST bit is set whenever 384 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
421 * on the shrink list (ie not on the superbloc 385 * on the shrink list (ie not on the superblock LRU list).
422 * 386 *
423 * The per-cpu "nr_dentry_unused" counters are 387 * The per-cpu "nr_dentry_unused" counters are updated with
424 * the DCACHE_LRU_LIST bit. 388 * the DCACHE_LRU_LIST bit.
425 * 389 *
426 * The per-cpu "nr_dentry_negative" counters a 390 * The per-cpu "nr_dentry_negative" counters are only updated
427 * when deleted from or added to the per-super 391 * when deleted from or added to the per-superblock LRU list, not
428 * from/to the shrink list. That is to avoid a 392 * from/to the shrink list. That is to avoid an unneeded dec/inc
429 * pair when moving from LRU to shrink list in 393 * pair when moving from LRU to shrink list in select_collect().
430 * 394 *
431 * These helper functions make sure we always 395 * These helper functions make sure we always follow the
432 * rules. d_lock must be held by the caller. 396 * rules. d_lock must be held by the caller.
433 */ 397 */
434 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(( 398 #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) 399 static void d_lru_add(struct dentry *dentry)
436 { 400 {
437 D_FLAG_VERIFY(dentry, 0); 401 D_FLAG_VERIFY(dentry, 0);
438 dentry->d_flags |= DCACHE_LRU_LIST; 402 dentry->d_flags |= DCACHE_LRU_LIST;
439 this_cpu_inc(nr_dentry_unused); 403 this_cpu_inc(nr_dentry_unused);
440 if (d_is_negative(dentry)) 404 if (d_is_negative(dentry))
441 this_cpu_inc(nr_dentry_negativ 405 this_cpu_inc(nr_dentry_negative);
442 WARN_ON_ONCE(!list_lru_add_obj( !! 406 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
443 &dentry->d_sb->s_dentr <<
444 } 407 }
445 408
446 static void d_lru_del(struct dentry *dentry) 409 static void d_lru_del(struct dentry *dentry)
447 { 410 {
448 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 411 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
449 dentry->d_flags &= ~DCACHE_LRU_LIST; 412 dentry->d_flags &= ~DCACHE_LRU_LIST;
450 this_cpu_dec(nr_dentry_unused); 413 this_cpu_dec(nr_dentry_unused);
451 if (d_is_negative(dentry)) 414 if (d_is_negative(dentry))
452 this_cpu_dec(nr_dentry_negativ 415 this_cpu_dec(nr_dentry_negative);
453 WARN_ON_ONCE(!list_lru_del_obj( !! 416 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
454 &dentry->d_sb->s_dentr <<
455 } 417 }
456 418
457 static void d_shrink_del(struct dentry *dentry 419 static void d_shrink_del(struct dentry *dentry)
458 { 420 {
459 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LI 421 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
460 list_del_init(&dentry->d_lru); 422 list_del_init(&dentry->d_lru);
461 dentry->d_flags &= ~(DCACHE_SHRINK_LIS 423 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
462 this_cpu_dec(nr_dentry_unused); 424 this_cpu_dec(nr_dentry_unused);
463 } 425 }
464 426
465 static void d_shrink_add(struct dentry *dentry 427 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
466 { 428 {
467 D_FLAG_VERIFY(dentry, 0); 429 D_FLAG_VERIFY(dentry, 0);
468 list_add(&dentry->d_lru, list); 430 list_add(&dentry->d_lru, list);
469 dentry->d_flags |= DCACHE_SHRINK_LIST 431 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
470 this_cpu_inc(nr_dentry_unused); 432 this_cpu_inc(nr_dentry_unused);
471 } 433 }
472 434
473 /* 435 /*
474 * These can only be called under the global L 436 * These can only be called under the global LRU lock, ie during the
475 * callback for freeing the LRU list. "isolate 437 * callback for freeing the LRU list. "isolate" removes it from the
476 * LRU lists entirely, while shrink_move moves 438 * LRU lists entirely, while shrink_move moves it to the indicated
477 * private list. 439 * private list.
478 */ 440 */
479 static void d_lru_isolate(struct list_lru_one 441 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
480 { 442 {
481 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 443 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
482 dentry->d_flags &= ~DCACHE_LRU_LIST; 444 dentry->d_flags &= ~DCACHE_LRU_LIST;
483 this_cpu_dec(nr_dentry_unused); 445 this_cpu_dec(nr_dentry_unused);
484 if (d_is_negative(dentry)) 446 if (d_is_negative(dentry))
485 this_cpu_dec(nr_dentry_negativ 447 this_cpu_dec(nr_dentry_negative);
486 list_lru_isolate(lru, &dentry->d_lru); 448 list_lru_isolate(lru, &dentry->d_lru);
487 } 449 }
488 450
489 static void d_lru_shrink_move(struct list_lru_ 451 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
490 struct list_head 452 struct list_head *list)
491 { 453 {
492 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 454 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
493 dentry->d_flags |= DCACHE_SHRINK_LIST; 455 dentry->d_flags |= DCACHE_SHRINK_LIST;
494 if (d_is_negative(dentry)) 456 if (d_is_negative(dentry))
495 this_cpu_dec(nr_dentry_negativ 457 this_cpu_dec(nr_dentry_negative);
496 list_lru_isolate_move(lru, &dentry->d_ 458 list_lru_isolate_move(lru, &dentry->d_lru, list);
497 } 459 }
498 460
499 static void ___d_drop(struct dentry *dentry) 461 static void ___d_drop(struct dentry *dentry)
500 { 462 {
501 struct hlist_bl_head *b; 463 struct hlist_bl_head *b;
502 /* 464 /*
503 * Hashed dentries are normally on the 465 * Hashed dentries are normally on the dentry hashtable,
504 * with the exception of those newly a 466 * with the exception of those newly allocated by
505 * d_obtain_root, which are always IS_ 467 * d_obtain_root, which are always IS_ROOT:
506 */ 468 */
507 if (unlikely(IS_ROOT(dentry))) 469 if (unlikely(IS_ROOT(dentry)))
508 b = &dentry->d_sb->s_roots; 470 b = &dentry->d_sb->s_roots;
509 else 471 else
510 b = d_hash(dentry->d_name.hash 472 b = d_hash(dentry->d_name.hash);
511 473
512 hlist_bl_lock(b); 474 hlist_bl_lock(b);
513 __hlist_bl_del(&dentry->d_hash); 475 __hlist_bl_del(&dentry->d_hash);
514 hlist_bl_unlock(b); 476 hlist_bl_unlock(b);
515 } 477 }
516 478
517 void __d_drop(struct dentry *dentry) 479 void __d_drop(struct dentry *dentry)
518 { 480 {
519 if (!d_unhashed(dentry)) { 481 if (!d_unhashed(dentry)) {
520 ___d_drop(dentry); 482 ___d_drop(dentry);
521 dentry->d_hash.pprev = NULL; 483 dentry->d_hash.pprev = NULL;
522 write_seqcount_invalidate(&den 484 write_seqcount_invalidate(&dentry->d_seq);
523 } 485 }
524 } 486 }
525 EXPORT_SYMBOL(__d_drop); 487 EXPORT_SYMBOL(__d_drop);
526 488
527 /** 489 /**
528 * d_drop - drop a dentry 490 * d_drop - drop a dentry
529 * @dentry: dentry to drop 491 * @dentry: dentry to drop
530 * 492 *
531 * d_drop() unhashes the entry from the parent 493 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
532 * be found through a VFS lookup any more. Not 494 * be found through a VFS lookup any more. Note that this is different from
533 * deleting the dentry - d_delete will try to 495 * deleting the dentry - d_delete will try to mark the dentry negative if
534 * possible, giving a successful _negative_ lo 496 * possible, giving a successful _negative_ lookup, while d_drop will
535 * just make the cache lookup fail. 497 * just make the cache lookup fail.
536 * 498 *
537 * d_drop() is used mainly for stuff that want 499 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
538 * reason (NFS timeouts or autofs deletes). 500 * reason (NFS timeouts or autofs deletes).
539 * 501 *
540 * __d_drop requires dentry->d_lock 502 * __d_drop requires dentry->d_lock
541 * 503 *
542 * ___d_drop doesn't mark dentry as "unhashed" 504 * ___d_drop doesn't mark dentry as "unhashed"
543 * (dentry->d_hash.pprev will be LIST_POISON2, 505 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
544 */ 506 */
545 void d_drop(struct dentry *dentry) 507 void d_drop(struct dentry *dentry)
546 { 508 {
547 spin_lock(&dentry->d_lock); 509 spin_lock(&dentry->d_lock);
548 __d_drop(dentry); 510 __d_drop(dentry);
549 spin_unlock(&dentry->d_lock); 511 spin_unlock(&dentry->d_lock);
550 } 512 }
551 EXPORT_SYMBOL(d_drop); 513 EXPORT_SYMBOL(d_drop);
552 514
553 static inline void dentry_unlist(struct dentry !! 515 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
554 { 516 {
555 struct dentry *next; 517 struct dentry *next;
556 /* 518 /*
557 * Inform d_walk() and shrink_dentry_l 519 * Inform d_walk() and shrink_dentry_list() that we are no longer
558 * attached to the dentry tree 520 * attached to the dentry tree
559 */ 521 */
560 dentry->d_flags |= DCACHE_DENTRY_KILLE 522 dentry->d_flags |= DCACHE_DENTRY_KILLED;
561 if (unlikely(hlist_unhashed(&dentry->d !! 523 if (unlikely(list_empty(&dentry->d_child)))
562 return; 524 return;
563 __hlist_del(&dentry->d_sib); !! 525 __list_del_entry(&dentry->d_child);
564 /* 526 /*
565 * Cursors can move around the list of 527 * Cursors can move around the list of children. While we'd been
566 * a normal list member, it didn't mat !! 528 * a normal list member, it didn't matter - ->d_child.next would've
567 * been updated. However, from now on 529 * been updated. However, from now on it won't be and for the
568 * things like d_walk() it might end u 530 * things like d_walk() it might end up with a nasty surprise.
569 * Normally d_walk() doesn't care abou 531 * Normally d_walk() doesn't care about cursors moving around -
570 * ->d_lock on parent prevents that an 532 * ->d_lock on parent prevents that and since a cursor has no children
571 * of its own, we get through it witho 533 * of its own, we get through it without ever unlocking the parent.
572 * There is one exception, though - if 534 * There is one exception, though - if we ascend from a child that
573 * gets killed as soon as we unlock it 535 * gets killed as soon as we unlock it, the next sibling is found
574 * using the value left in its ->d_sib !! 536 * using the value left in its ->d_child.next. And if _that_
575 * pointed to a cursor, and cursor got 537 * pointed to a cursor, and cursor got moved (e.g. by lseek())
576 * before d_walk() regains parent->d_l 538 * before d_walk() regains parent->d_lock, we'll end up skipping
577 * everything the cursor had been move 539 * everything the cursor had been moved past.
578 * 540 *
579 * Solution: make sure that the pointe !! 541 * Solution: make sure that the pointer left behind in ->d_child.next
580 * points to something that won't be m 542 * points to something that won't be moving around. I.e. skip the
581 * cursors. 543 * cursors.
582 */ 544 */
583 while (dentry->d_sib.next) { !! 545 while (dentry->d_child.next != &parent->d_subdirs) {
584 next = hlist_entry(dentry->d_s !! 546 next = list_entry(dentry->d_child.next, struct dentry, d_child);
585 if (likely(!(next->d_flags & D 547 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
586 break; 548 break;
587 dentry->d_sib.next = next->d_s !! 549 dentry->d_child.next = next->d_child.next;
588 } 550 }
589 } 551 }
590 552
591 static struct dentry *__dentry_kill(struct den !! 553 static void __dentry_kill(struct dentry *dentry)
592 { 554 {
593 struct dentry *parent = NULL; 555 struct dentry *parent = NULL;
594 bool can_free = true; 556 bool can_free = true;
>> 557 if (!IS_ROOT(dentry))
>> 558 parent = dentry->d_parent;
595 559
596 /* 560 /*
597 * The dentry is now unrecoverably dea 561 * The dentry is now unrecoverably dead to the world.
598 */ 562 */
599 lockref_mark_dead(&dentry->d_lockref); 563 lockref_mark_dead(&dentry->d_lockref);
600 564
601 /* 565 /*
602 * inform the fs via d_prune that this 566 * inform the fs via d_prune that this dentry is about to be
603 * unhashed and destroyed. 567 * unhashed and destroyed.
604 */ 568 */
605 if (dentry->d_flags & DCACHE_OP_PRUNE) 569 if (dentry->d_flags & DCACHE_OP_PRUNE)
606 dentry->d_op->d_prune(dentry); 570 dentry->d_op->d_prune(dentry);
607 571
608 if (dentry->d_flags & DCACHE_LRU_LIST) 572 if (dentry->d_flags & DCACHE_LRU_LIST) {
609 if (!(dentry->d_flags & DCACHE 573 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
610 d_lru_del(dentry); 574 d_lru_del(dentry);
611 } 575 }
612 /* if it was on the hash then remove i 576 /* if it was on the hash then remove it */
613 __d_drop(dentry); 577 __d_drop(dentry);
>> 578 dentry_unlist(dentry, parent);
>> 579 if (parent)
>> 580 spin_unlock(&parent->d_lock);
614 if (dentry->d_inode) 581 if (dentry->d_inode)
615 dentry_unlink_inode(dentry); 582 dentry_unlink_inode(dentry);
616 else 583 else
617 spin_unlock(&dentry->d_lock); 584 spin_unlock(&dentry->d_lock);
618 this_cpu_dec(nr_dentry); 585 this_cpu_dec(nr_dentry);
619 if (dentry->d_op && dentry->d_op->d_re 586 if (dentry->d_op && dentry->d_op->d_release)
620 dentry->d_op->d_release(dentry 587 dentry->d_op->d_release(dentry);
621 588
622 cond_resched(); !! 589 spin_lock(&dentry->d_lock);
623 /* now that it's negative, ->d_parent !! 590 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
624 if (!IS_ROOT(dentry)) { !! 591 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; 592 can_free = false;
>> 593 }
632 spin_unlock(&dentry->d_lock); 594 spin_unlock(&dentry->d_lock);
633 if (likely(can_free)) 595 if (likely(can_free))
634 dentry_free(dentry); 596 dentry_free(dentry);
635 if (parent && --parent->d_lockref.coun !! 597 cond_resched();
>> 598 }
>> 599
>> 600 static struct dentry *__lock_parent(struct dentry *dentry)
>> 601 {
>> 602 struct dentry *parent;
>> 603 rcu_read_lock();
>> 604 spin_unlock(&dentry->d_lock);
>> 605 again:
>> 606 parent = READ_ONCE(dentry->d_parent);
>> 607 spin_lock(&parent->d_lock);
>> 608 /*
>> 609 * We can't blindly lock dentry until we are sure
>> 610 * that we won't violate the locking order.
>> 611 * Any changes of dentry->d_parent must have
>> 612 * been done with parent->d_lock held, so
>> 613 * spin_lock() above is enough of a barrier
>> 614 * for checking if it's still our child.
>> 615 */
>> 616 if (unlikely(parent != dentry->d_parent)) {
636 spin_unlock(&parent->d_lock); 617 spin_unlock(&parent->d_lock);
637 return NULL; !! 618 goto again;
638 } 619 }
>> 620 rcu_read_unlock();
>> 621 if (parent != dentry)
>> 622 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
>> 623 else
>> 624 parent = NULL;
639 return parent; 625 return parent;
640 } 626 }
641 627
642 /* !! 628 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 { 629 {
655 struct inode *inode = dentry->d_inode; !! 630 struct dentry *parent = dentry->d_parent;
656 !! 631 if (IS_ROOT(dentry))
657 if (unlikely(dentry->d_lockref.count)) !! 632 return NULL;
658 return false; !! 633 if (likely(spin_trylock(&parent->d_lock)))
659 !! 634 return parent;
660 if (!inode || likely(spin_trylock(&ino !! 635 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 } 636 }
678 637
679 /* !! 638 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 { 639 {
692 unsigned int d_flags; !! 640 WARN_ON(d_in_lookup(dentry));
693 <<
694 smp_rmb(); <<
695 d_flags = READ_ONCE(dentry->d_flags); <<
696 641
697 // Unreachable? Nobody would be able t !! 642 /* Unreachable? Get rid of it */
698 if (unlikely(d_unhashed(dentry))) 643 if (unlikely(d_unhashed(dentry)))
699 return false; 644 return false;
700 645
701 // Same if it's disconnected !! 646 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
702 if (unlikely(d_flags & DCACHE_DISCONNE <<
703 return false; 647 return false;
704 648
705 // ->d_delete() might tell us not to b !! 649 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
706 // ->d_lock; can't decide without it !! 650 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; 651 return false;
710 } 652 }
711 653
712 // Explicitly told not to bother !! 654 if (unlikely(dentry->d_flags & DCACHE_DONTCACHE))
713 if (unlikely(d_flags & DCACHE_DONTCACH <<
714 return false; 655 return false;
715 656
716 // At this point it looks like we ough !! 657 /* retain; LRU fodder */
717 // need to do something - put it on LR !! 658 dentry->d_lockref.count--;
718 // and mark it referenced if it was on !! 659 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); 660 d_lru_add(dentry);
725 } else if (unlikely(!(d_flags & DCACHE !! 661 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
726 if (!locked) <<
727 return false; <<
728 dentry->d_flags |= DCACHE_REFE 662 dentry->d_flags |= DCACHE_REFERENCED;
729 } <<
730 return true; 663 return true;
731 } 664 }
732 665
733 void d_mark_dontcache(struct inode *inode) 666 void d_mark_dontcache(struct inode *inode)
734 { 667 {
735 struct dentry *de; 668 struct dentry *de;
736 669
737 spin_lock(&inode->i_lock); 670 spin_lock(&inode->i_lock);
738 hlist_for_each_entry(de, &inode->i_den 671 hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
739 spin_lock(&de->d_lock); 672 spin_lock(&de->d_lock);
740 de->d_flags |= DCACHE_DONTCACH 673 de->d_flags |= DCACHE_DONTCACHE;
741 spin_unlock(&de->d_lock); 674 spin_unlock(&de->d_lock);
742 } 675 }
743 inode->i_state |= I_DONTCACHE; 676 inode->i_state |= I_DONTCACHE;
744 spin_unlock(&inode->i_lock); 677 spin_unlock(&inode->i_lock);
745 } 678 }
746 EXPORT_SYMBOL(d_mark_dontcache); 679 EXPORT_SYMBOL(d_mark_dontcache);
747 680
748 /* 681 /*
>> 682 * Finish off a dentry we've decided to kill.
>> 683 * dentry->d_lock must be held, returns with it unlocked.
>> 684 * Returns dentry requiring refcount drop, or NULL if we're done.
>> 685 */
>> 686 static struct dentry *dentry_kill(struct dentry *dentry)
>> 687 __releases(dentry->d_lock)
>> 688 {
>> 689 struct inode *inode = dentry->d_inode;
>> 690 struct dentry *parent = NULL;
>> 691
>> 692 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
>> 693 goto slow_positive;
>> 694
>> 695 if (!IS_ROOT(dentry)) {
>> 696 parent = dentry->d_parent;
>> 697 if (unlikely(!spin_trylock(&parent->d_lock))) {
>> 698 parent = __lock_parent(dentry);
>> 699 if (likely(inode || !dentry->d_inode))
>> 700 goto got_locks;
>> 701 /* negative that became positive */
>> 702 if (parent)
>> 703 spin_unlock(&parent->d_lock);
>> 704 inode = dentry->d_inode;
>> 705 goto slow_positive;
>> 706 }
>> 707 }
>> 708 __dentry_kill(dentry);
>> 709 return parent;
>> 710
>> 711 slow_positive:
>> 712 spin_unlock(&dentry->d_lock);
>> 713 spin_lock(&inode->i_lock);
>> 714 spin_lock(&dentry->d_lock);
>> 715 parent = lock_parent(dentry);
>> 716 got_locks:
>> 717 if (unlikely(dentry->d_lockref.count != 1)) {
>> 718 dentry->d_lockref.count--;
>> 719 } else if (likely(!retain_dentry(dentry))) {
>> 720 __dentry_kill(dentry);
>> 721 return parent;
>> 722 }
>> 723 /* we are keeping it, after all */
>> 724 if (inode)
>> 725 spin_unlock(&inode->i_lock);
>> 726 if (parent)
>> 727 spin_unlock(&parent->d_lock);
>> 728 spin_unlock(&dentry->d_lock);
>> 729 return NULL;
>> 730 }
>> 731
>> 732 /*
749 * Try to do a lockless dput(), and return whe 733 * Try to do a lockless dput(), and return whether that was successful.
750 * 734 *
751 * If unsuccessful, we return false, having al 735 * 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 * 736 *
755 * The caller needs to hold the RCU read lock, 737 * The caller needs to hold the RCU read lock, so that the dentry is
756 * guaranteed to stay around even if the refco 738 * guaranteed to stay around even if the refcount goes down to zero!
757 */ 739 */
758 static inline bool fast_dput(struct dentry *de 740 static inline bool fast_dput(struct dentry *dentry)
759 { 741 {
760 int ret; 742 int ret;
>> 743 unsigned int d_flags;
>> 744
>> 745 /*
>> 746 * If we have a d_op->d_delete() operation, we sould not
>> 747 * let the dentry count go to zero, so use "put_or_lock".
>> 748 */
>> 749 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
>> 750 return lockref_put_or_lock(&dentry->d_lockref);
761 751
762 /* 752 /*
763 * try to decrement the lockref optimi !! 753 * .. otherwise, we can try to just decrement the
>> 754 * lockref optimistically.
764 */ 755 */
765 ret = lockref_put_return(&dentry->d_lo 756 ret = lockref_put_return(&dentry->d_lockref);
766 757
767 /* 758 /*
768 * If the lockref_put_return() failed 759 * If the lockref_put_return() failed due to the lock being held
769 * by somebody else, the fast path has 760 * by somebody else, the fast path has failed. We will need to
770 * get the lock, and then check the co 761 * get the lock, and then check the count again.
771 */ 762 */
772 if (unlikely(ret < 0)) { 763 if (unlikely(ret < 0)) {
773 spin_lock(&dentry->d_lock); 764 spin_lock(&dentry->d_lock);
774 if (WARN_ON_ONCE(dentry->d_loc !! 765 if (dentry->d_lockref.count > 1) {
>> 766 dentry->d_lockref.count--;
775 spin_unlock(&dentry->d 767 spin_unlock(&dentry->d_lock);
776 return true; 768 return true;
777 } 769 }
778 dentry->d_lockref.count--; !! 770 return false;
779 goto locked; <<
780 } 771 }
781 772
782 /* 773 /*
783 * If we weren't the last ref, we're d 774 * If we weren't the last ref, we're done.
784 */ 775 */
785 if (ret) 776 if (ret)
786 return true; 777 return true;
787 778
788 /* 779 /*
789 * Can we decide that decrement of ref !! 780 * Careful, careful. The reference count went down
790 * taking the lock? There's a very co !! 781 * to zero, but we don't hold the dentry lock, so
791 * dentry looks like it ought to be re !! 782 * somebody else could get it again, and do another
792 * to do. !! 783 * dput(), and we need to not race with that.
>> 784 *
>> 785 * However, there is a very special and common case
>> 786 * where we don't care, because there is nothing to
>> 787 * do: the dentry is still hashed, it does not have
>> 788 * a 'delete' op, and it's referenced and already on
>> 789 * the LRU list.
>> 790 *
>> 791 * NOTE! Since we aren't locked, these values are
>> 792 * not "stable". However, it is sufficient that at
>> 793 * some point after we dropped the reference the
>> 794 * dentry was hashed and the flags had the proper
>> 795 * value. Other dentry users may have re-gotten
>> 796 * a reference to the dentry and change that, but
>> 797 * our work is done - we can leave the dentry
>> 798 * around with a zero refcount.
>> 799 *
>> 800 * Nevertheless, there are two cases that we should kill
>> 801 * the dentry anyway.
>> 802 * 1. free disconnected dentries as soon as their refcount
>> 803 * reached zero.
>> 804 * 2. free dentries if they should not be cached.
793 */ 805 */
794 if (retain_dentry(dentry, false)) !! 806 smp_rmb();
>> 807 d_flags = READ_ONCE(dentry->d_flags);
>> 808 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST |
>> 809 DCACHE_DISCONNECTED | DCACHE_DONTCACHE;
>> 810
>> 811 /* Nothing to do? Dropping the reference was all we needed? */
>> 812 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
795 return true; 813 return true;
796 814
797 /* 815 /*
798 * Either not worth retaining or we ca !! 816 * Not the fast normal case? Get the lock. We've already decremented
799 * Get the lock, then. We've already !! 817 * the refcount, but we'll need to re-check the situation after
800 * but we'll need to re-check the situ !! 818 * getting the lock.
801 */ 819 */
802 spin_lock(&dentry->d_lock); 820 spin_lock(&dentry->d_lock);
803 821
804 /* 822 /*
805 * Did somebody else grab a reference 823 * Did somebody else grab a reference to it in the meantime, and
806 * we're no longer the last user after 824 * we're no longer the last user after all? Alternatively, somebody
807 * else could have killed it and marke 825 * else could have killed it and marked it dead. Either way, we
808 * don't need to do anything else. 826 * don't need to do anything else.
809 */ 827 */
810 locked: !! 828 if (dentry->d_lockref.count) {
811 if (dentry->d_lockref.count || retain_ <<
812 spin_unlock(&dentry->d_lock); 829 spin_unlock(&dentry->d_lock);
813 return true; 830 return true;
814 } 831 }
>> 832
>> 833 /*
>> 834 * Re-get the reference we optimistically dropped. We hold the
>> 835 * lock, and we just tested that it was zero, so we can just
>> 836 * set it to 1.
>> 837 */
>> 838 dentry->d_lockref.count = 1;
815 return false; 839 return false;
816 } 840 }
817 841
818 842
819 /* 843 /*
820 * This is dput 844 * This is dput
821 * 845 *
822 * This is complicated by the fact that we do 846 * This is complicated by the fact that we do not want to put
823 * dentries that are no longer on any hash cha 847 * dentries that are no longer on any hash chain on the unused
824 * list: we'd much rather just get rid of them 848 * list: we'd much rather just get rid of them immediately.
825 * 849 *
826 * However, that implies that we have to trave 850 * However, that implies that we have to traverse the dentry
827 * tree upwards to the parents which might _al 851 * tree upwards to the parents which might _also_ now be
828 * scheduled for deletion (it may have been on 852 * scheduled for deletion (it may have been only waiting for
829 * its last child to go away). 853 * its last child to go away).
830 * 854 *
831 * This tail recursion is done by hand as we d 855 * This tail recursion is done by hand as we don't want to depend
832 * on the compiler to always get this right (g 856 * on the compiler to always get this right (gcc generally doesn't).
833 * Real recursion would eat up our stack space 857 * Real recursion would eat up our stack space.
834 */ 858 */
835 859
836 /* 860 /*
837 * dput - release a dentry 861 * dput - release a dentry
838 * @dentry: dentry to release 862 * @dentry: dentry to release
839 * 863 *
840 * Release a dentry. This will drop the usage 864 * Release a dentry. This will drop the usage count and if appropriate
841 * call the dentry unlink method as well as re 865 * call the dentry unlink method as well as removing it from the queues and
842 * releasing its resources. If the parent dent 866 * releasing its resources. If the parent dentries were scheduled for release
843 * they too may now get deleted. 867 * they too may now get deleted.
844 */ 868 */
845 void dput(struct dentry *dentry) 869 void dput(struct dentry *dentry)
846 { 870 {
847 if (!dentry) !! 871 while (dentry) {
848 return; !! 872 might_sleep();
849 might_sleep(); !! 873
850 rcu_read_lock(); !! 874 rcu_read_lock();
851 if (likely(fast_dput(dentry))) { !! 875 if (likely(fast_dput(dentry))) {
852 rcu_read_unlock(); !! 876 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; 877 return;
860 if (retain_dentry(dentry, true !! 878 }
>> 879
>> 880 /* Slow case: now with the dentry lock held */
>> 881 rcu_read_unlock();
>> 882
>> 883 if (likely(retain_dentry(dentry))) {
861 spin_unlock(&dentry->d 884 spin_unlock(&dentry->d_lock);
862 return; 885 return;
863 } 886 }
864 rcu_read_lock(); !! 887
>> 888 dentry = dentry_kill(dentry);
865 } 889 }
866 rcu_read_unlock(); <<
867 spin_unlock(&dentry->d_lock); <<
868 } 890 }
869 EXPORT_SYMBOL(dput); 891 EXPORT_SYMBOL(dput);
870 892
871 static void to_shrink_list(struct dentry *dent !! 893 static void __dput_to_list(struct dentry *dentry, struct list_head *list)
872 __must_hold(&dentry->d_lock) 894 __must_hold(&dentry->d_lock)
873 { 895 {
874 if (!(dentry->d_flags & DCACHE_SHRINK_ !! 896 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
>> 897 /* let the owner of the list it's on deal with it */
>> 898 --dentry->d_lockref.count;
>> 899 } else {
875 if (dentry->d_flags & DCACHE_L 900 if (dentry->d_flags & DCACHE_LRU_LIST)
876 d_lru_del(dentry); 901 d_lru_del(dentry);
877 d_shrink_add(dentry, list); !! 902 if (!--dentry->d_lockref.count)
>> 903 d_shrink_add(dentry, list);
878 } 904 }
879 } 905 }
880 906
881 void dput_to_list(struct dentry *dentry, struc 907 void dput_to_list(struct dentry *dentry, struct list_head *list)
882 { 908 {
883 rcu_read_lock(); 909 rcu_read_lock();
884 if (likely(fast_dput(dentry))) { 910 if (likely(fast_dput(dentry))) {
885 rcu_read_unlock(); 911 rcu_read_unlock();
886 return; 912 return;
887 } 913 }
888 rcu_read_unlock(); 914 rcu_read_unlock();
889 to_shrink_list(dentry, list); !! 915 if (!retain_dentry(dentry))
>> 916 __dput_to_list(dentry, list);
890 spin_unlock(&dentry->d_lock); 917 spin_unlock(&dentry->d_lock);
891 } 918 }
892 919
>> 920 /* This must be called with d_lock held */
>> 921 static inline void __dget_dlock(struct dentry *dentry)
>> 922 {
>> 923 dentry->d_lockref.count++;
>> 924 }
>> 925
>> 926 static inline void __dget(struct dentry *dentry)
>> 927 {
>> 928 lockref_get(&dentry->d_lockref);
>> 929 }
>> 930
893 struct dentry *dget_parent(struct dentry *dent 931 struct dentry *dget_parent(struct dentry *dentry)
894 { 932 {
895 int gotref; 933 int gotref;
896 struct dentry *ret; 934 struct dentry *ret;
897 unsigned seq; 935 unsigned seq;
898 936
899 /* 937 /*
900 * Do optimistic parent lookup without 938 * Do optimistic parent lookup without any
901 * locking. 939 * locking.
902 */ 940 */
903 rcu_read_lock(); 941 rcu_read_lock();
904 seq = raw_seqcount_begin(&dentry->d_se 942 seq = raw_seqcount_begin(&dentry->d_seq);
905 ret = READ_ONCE(dentry->d_parent); 943 ret = READ_ONCE(dentry->d_parent);
906 gotref = lockref_get_not_zero(&ret->d_ 944 gotref = lockref_get_not_zero(&ret->d_lockref);
907 rcu_read_unlock(); 945 rcu_read_unlock();
908 if (likely(gotref)) { 946 if (likely(gotref)) {
909 if (!read_seqcount_retry(&dent 947 if (!read_seqcount_retry(&dentry->d_seq, seq))
910 return ret; 948 return ret;
911 dput(ret); 949 dput(ret);
912 } 950 }
913 951
914 repeat: 952 repeat:
915 /* 953 /*
916 * Don't need rcu_dereference because 954 * Don't need rcu_dereference because we re-check it was correct under
917 * the lock. 955 * the lock.
918 */ 956 */
919 rcu_read_lock(); 957 rcu_read_lock();
920 ret = dentry->d_parent; 958 ret = dentry->d_parent;
921 spin_lock(&ret->d_lock); 959 spin_lock(&ret->d_lock);
922 if (unlikely(ret != dentry->d_parent)) 960 if (unlikely(ret != dentry->d_parent)) {
923 spin_unlock(&ret->d_lock); 961 spin_unlock(&ret->d_lock);
924 rcu_read_unlock(); 962 rcu_read_unlock();
925 goto repeat; 963 goto repeat;
926 } 964 }
927 rcu_read_unlock(); 965 rcu_read_unlock();
928 BUG_ON(!ret->d_lockref.count); 966 BUG_ON(!ret->d_lockref.count);
929 ret->d_lockref.count++; 967 ret->d_lockref.count++;
930 spin_unlock(&ret->d_lock); 968 spin_unlock(&ret->d_lock);
931 return ret; 969 return ret;
932 } 970 }
933 EXPORT_SYMBOL(dget_parent); 971 EXPORT_SYMBOL(dget_parent);
934 972
935 static struct dentry * __d_find_any_alias(stru 973 static struct dentry * __d_find_any_alias(struct inode *inode)
936 { 974 {
937 struct dentry *alias; 975 struct dentry *alias;
938 976
939 if (hlist_empty(&inode->i_dentry)) 977 if (hlist_empty(&inode->i_dentry))
940 return NULL; 978 return NULL;
941 alias = hlist_entry(inode->i_dentry.fi 979 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
942 lockref_get(&alias->d_lockref); !! 980 __dget(alias);
943 return alias; 981 return alias;
944 } 982 }
945 983
946 /** 984 /**
947 * d_find_any_alias - find any alias for a giv 985 * d_find_any_alias - find any alias for a given inode
948 * @inode: inode to find an alias for 986 * @inode: inode to find an alias for
949 * 987 *
950 * If any aliases exist for the given inode, t 988 * If any aliases exist for the given inode, take and return a
951 * reference for one of them. If no aliases e 989 * reference for one of them. If no aliases exist, return %NULL.
952 */ 990 */
953 struct dentry *d_find_any_alias(struct inode * 991 struct dentry *d_find_any_alias(struct inode *inode)
954 { 992 {
955 struct dentry *de; 993 struct dentry *de;
956 994
957 spin_lock(&inode->i_lock); 995 spin_lock(&inode->i_lock);
958 de = __d_find_any_alias(inode); 996 de = __d_find_any_alias(inode);
959 spin_unlock(&inode->i_lock); 997 spin_unlock(&inode->i_lock);
960 return de; 998 return de;
961 } 999 }
962 EXPORT_SYMBOL(d_find_any_alias); 1000 EXPORT_SYMBOL(d_find_any_alias);
963 1001
964 static struct dentry *__d_find_alias(struct in 1002 static struct dentry *__d_find_alias(struct inode *inode)
965 { 1003 {
966 struct dentry *alias; 1004 struct dentry *alias;
967 1005
968 if (S_ISDIR(inode->i_mode)) 1006 if (S_ISDIR(inode->i_mode))
969 return __d_find_any_alias(inod 1007 return __d_find_any_alias(inode);
970 1008
971 hlist_for_each_entry(alias, &inode->i_ 1009 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
972 spin_lock(&alias->d_lock); 1010 spin_lock(&alias->d_lock);
973 if (!d_unhashed(alias)) { 1011 if (!d_unhashed(alias)) {
974 dget_dlock(alias); !! 1012 __dget_dlock(alias);
975 spin_unlock(&alias->d_ 1013 spin_unlock(&alias->d_lock);
976 return alias; 1014 return alias;
977 } 1015 }
978 spin_unlock(&alias->d_lock); 1016 spin_unlock(&alias->d_lock);
979 } 1017 }
980 return NULL; 1018 return NULL;
981 } 1019 }
982 1020
983 /** 1021 /**
984 * d_find_alias - grab a hashed alias of inode 1022 * d_find_alias - grab a hashed alias of inode
985 * @inode: inode in question 1023 * @inode: inode in question
986 * 1024 *
987 * If inode has a hashed alias, or is a direct 1025 * If inode has a hashed alias, or is a directory and has any alias,
988 * acquire the reference to alias and return i 1026 * acquire the reference to alias and return it. Otherwise return NULL.
989 * Notice that if inode is a directory there c 1027 * Notice that if inode is a directory there can be only one alias and
990 * it can be unhashed only if it has no childr 1028 * it can be unhashed only if it has no children, or if it is the root
991 * of a filesystem, or if the directory was re 1029 * of a filesystem, or if the directory was renamed and d_revalidate
992 * was the first vfs operation to notice. 1030 * was the first vfs operation to notice.
993 * 1031 *
994 * If the inode has an IS_ROOT, DCACHE_DISCONN 1032 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
995 * any other hashed alias over that one. 1033 * any other hashed alias over that one.
996 */ 1034 */
997 struct dentry *d_find_alias(struct inode *inod 1035 struct dentry *d_find_alias(struct inode *inode)
998 { 1036 {
999 struct dentry *de = NULL; 1037 struct dentry *de = NULL;
1000 1038
1001 if (!hlist_empty(&inode->i_dentry)) { 1039 if (!hlist_empty(&inode->i_dentry)) {
1002 spin_lock(&inode->i_lock); 1040 spin_lock(&inode->i_lock);
1003 de = __d_find_alias(inode); 1041 de = __d_find_alias(inode);
1004 spin_unlock(&inode->i_lock); 1042 spin_unlock(&inode->i_lock);
1005 } 1043 }
1006 return de; 1044 return de;
1007 } 1045 }
1008 EXPORT_SYMBOL(d_find_alias); 1046 EXPORT_SYMBOL(d_find_alias);
1009 1047
1010 /* 1048 /*
1011 * Caller MUST be holding rcu_read_lock() an 1049 * Caller MUST be holding rcu_read_lock() and be guaranteed
1012 * that inode won't get freed until rcu_read 1050 * that inode won't get freed until rcu_read_unlock().
1013 */ 1051 */
1014 struct dentry *d_find_alias_rcu(struct inode 1052 struct dentry *d_find_alias_rcu(struct inode *inode)
1015 { 1053 {
1016 struct hlist_head *l = &inode->i_dent 1054 struct hlist_head *l = &inode->i_dentry;
1017 struct dentry *de = NULL; 1055 struct dentry *de = NULL;
1018 1056
1019 spin_lock(&inode->i_lock); 1057 spin_lock(&inode->i_lock);
1020 // ->i_dentry and ->i_rcu are colocat 1058 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1021 // used without having I_FREEING set, 1059 // used without having I_FREEING set, which means no aliases left
1022 if (likely(!(inode->i_state & I_FREEI 1060 if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1023 if (S_ISDIR(inode->i_mode)) { 1061 if (S_ISDIR(inode->i_mode)) {
1024 de = hlist_entry(l->f 1062 de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1025 } else { 1063 } else {
1026 hlist_for_each_entry( 1064 hlist_for_each_entry(de, l, d_u.d_alias)
1027 if (!d_unhash 1065 if (!d_unhashed(de))
1028 break 1066 break;
1029 } 1067 }
1030 } 1068 }
1031 spin_unlock(&inode->i_lock); 1069 spin_unlock(&inode->i_lock);
1032 return de; 1070 return de;
1033 } 1071 }
1034 1072
1035 /* 1073 /*
1036 * Try to kill dentries associated with 1074 * Try to kill dentries associated with this inode.
1037 * WARNING: you must own a reference to inode 1075 * WARNING: you must own a reference to inode.
1038 */ 1076 */
1039 void d_prune_aliases(struct inode *inode) 1077 void d_prune_aliases(struct inode *inode)
1040 { 1078 {
1041 LIST_HEAD(dispose); <<
1042 struct dentry *dentry; 1079 struct dentry *dentry;
1043 !! 1080 restart:
1044 spin_lock(&inode->i_lock); 1081 spin_lock(&inode->i_lock);
1045 hlist_for_each_entry(dentry, &inode-> 1082 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1046 spin_lock(&dentry->d_lock); 1083 spin_lock(&dentry->d_lock);
1047 if (!dentry->d_lockref.count) !! 1084 if (!dentry->d_lockref.count) {
1048 to_shrink_list(dentry !! 1085 struct dentry *parent = lock_parent(dentry);
>> 1086 if (likely(!dentry->d_lockref.count)) {
>> 1087 __dentry_kill(dentry);
>> 1088 dput(parent);
>> 1089 goto restart;
>> 1090 }
>> 1091 if (parent)
>> 1092 spin_unlock(&parent->d_lock);
>> 1093 }
1049 spin_unlock(&dentry->d_lock); 1094 spin_unlock(&dentry->d_lock);
1050 } 1095 }
1051 spin_unlock(&inode->i_lock); 1096 spin_unlock(&inode->i_lock);
1052 shrink_dentry_list(&dispose); <<
1053 } 1097 }
1054 EXPORT_SYMBOL(d_prune_aliases); 1098 EXPORT_SYMBOL(d_prune_aliases);
1055 1099
1056 static inline void shrink_kill(struct dentry !! 1100 /*
>> 1101 * Lock a dentry from shrink list.
>> 1102 * Called under rcu_read_lock() and dentry->d_lock; the former
>> 1103 * guarantees that nothing we access will be freed under us.
>> 1104 * Note that dentry is *not* protected from concurrent dentry_kill(),
>> 1105 * d_delete(), etc.
>> 1106 *
>> 1107 * Return false if dentry has been disrupted or grabbed, leaving
>> 1108 * the caller to kick it off-list. Otherwise, return true and have
>> 1109 * that dentry's inode and parent both locked.
>> 1110 */
>> 1111 static bool shrink_lock_dentry(struct dentry *dentry)
1057 { 1112 {
1058 do { !! 1113 struct inode *inode;
1059 rcu_read_unlock(); !! 1114 struct dentry *parent;
1060 victim = __dentry_kill(victim !! 1115
1061 rcu_read_lock(); !! 1116 if (dentry->d_lockref.count)
1062 } while (victim && lock_for_kill(vict !! 1117 return false;
1063 rcu_read_unlock(); !! 1118
1064 if (victim) !! 1119 inode = dentry->d_inode;
1065 spin_unlock(&victim->d_lock); !! 1120 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
>> 1121 spin_unlock(&dentry->d_lock);
>> 1122 spin_lock(&inode->i_lock);
>> 1123 spin_lock(&dentry->d_lock);
>> 1124 if (unlikely(dentry->d_lockref.count))
>> 1125 goto out;
>> 1126 /* changed inode means that somebody had grabbed it */
>> 1127 if (unlikely(inode != dentry->d_inode))
>> 1128 goto out;
>> 1129 }
>> 1130
>> 1131 parent = dentry->d_parent;
>> 1132 if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
>> 1133 return true;
>> 1134
>> 1135 spin_unlock(&dentry->d_lock);
>> 1136 spin_lock(&parent->d_lock);
>> 1137 if (unlikely(parent != dentry->d_parent)) {
>> 1138 spin_unlock(&parent->d_lock);
>> 1139 spin_lock(&dentry->d_lock);
>> 1140 goto out;
>> 1141 }
>> 1142 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
>> 1143 if (likely(!dentry->d_lockref.count))
>> 1144 return true;
>> 1145 spin_unlock(&parent->d_lock);
>> 1146 out:
>> 1147 if (inode)
>> 1148 spin_unlock(&inode->i_lock);
>> 1149 return false;
1066 } 1150 }
1067 1151
1068 void shrink_dentry_list(struct list_head *lis 1152 void shrink_dentry_list(struct list_head *list)
1069 { 1153 {
1070 while (!list_empty(list)) { 1154 while (!list_empty(list)) {
1071 struct dentry *dentry; !! 1155 struct dentry *dentry, *parent;
1072 1156
1073 dentry = list_entry(list->pre 1157 dentry = list_entry(list->prev, struct dentry, d_lru);
1074 spin_lock(&dentry->d_lock); 1158 spin_lock(&dentry->d_lock);
1075 rcu_read_lock(); 1159 rcu_read_lock();
1076 if (!lock_for_kill(dentry)) { !! 1160 if (!shrink_lock_dentry(dentry)) {
1077 bool can_free; !! 1161 bool can_free = false;
1078 rcu_read_unlock(); 1162 rcu_read_unlock();
1079 d_shrink_del(dentry); 1163 d_shrink_del(dentry);
1080 can_free = dentry->d_ !! 1164 if (dentry->d_lockref.count < 0)
>> 1165 can_free = dentry->d_flags & DCACHE_MAY_FREE;
1081 spin_unlock(&dentry-> 1166 spin_unlock(&dentry->d_lock);
1082 if (can_free) 1167 if (can_free)
1083 dentry_free(d 1168 dentry_free(dentry);
1084 continue; 1169 continue;
1085 } 1170 }
>> 1171 rcu_read_unlock();
1086 d_shrink_del(dentry); 1172 d_shrink_del(dentry);
1087 shrink_kill(dentry); !! 1173 parent = dentry->d_parent;
>> 1174 if (parent != dentry)
>> 1175 __dput_to_list(parent, list);
>> 1176 __dentry_kill(dentry);
1088 } 1177 }
1089 } 1178 }
1090 1179
1091 static enum lru_status dentry_lru_isolate(str 1180 static enum lru_status dentry_lru_isolate(struct list_head *item,
1092 struct list_lru_one *lru, spi 1181 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1093 { 1182 {
1094 struct list_head *freeable = arg; 1183 struct list_head *freeable = arg;
1095 struct dentry *dentry = container_o 1184 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1096 1185
1097 1186
1098 /* 1187 /*
1099 * we are inverting the lru lock/dent 1188 * we are inverting the lru lock/dentry->d_lock here,
1100 * so use a trylock. If we fail to ge 1189 * so use a trylock. If we fail to get the lock, just skip
1101 * it 1190 * it
1102 */ 1191 */
1103 if (!spin_trylock(&dentry->d_lock)) 1192 if (!spin_trylock(&dentry->d_lock))
1104 return LRU_SKIP; 1193 return LRU_SKIP;
1105 1194
1106 /* 1195 /*
1107 * Referenced dentries are still in u 1196 * Referenced dentries are still in use. If they have active
1108 * counts, just remove them from the 1197 * counts, just remove them from the LRU. Otherwise give them
1109 * another pass through the LRU. 1198 * another pass through the LRU.
1110 */ 1199 */
1111 if (dentry->d_lockref.count) { 1200 if (dentry->d_lockref.count) {
1112 d_lru_isolate(lru, dentry); 1201 d_lru_isolate(lru, dentry);
1113 spin_unlock(&dentry->d_lock); 1202 spin_unlock(&dentry->d_lock);
1114 return LRU_REMOVED; 1203 return LRU_REMOVED;
1115 } 1204 }
1116 1205
1117 if (dentry->d_flags & DCACHE_REFERENC 1206 if (dentry->d_flags & DCACHE_REFERENCED) {
1118 dentry->d_flags &= ~DCACHE_RE 1207 dentry->d_flags &= ~DCACHE_REFERENCED;
1119 spin_unlock(&dentry->d_lock); 1208 spin_unlock(&dentry->d_lock);
1120 1209
1121 /* 1210 /*
1122 * The list move itself will 1211 * The list move itself will be made by the common LRU code. At
1123 * this point, we've dropped 1212 * this point, we've dropped the dentry->d_lock but keep the
1124 * lru lock. This is safe to 1213 * lru lock. This is safe to do, since every list movement is
1125 * protected by the lru lock 1214 * protected by the lru lock even if both locks are held.
1126 * 1215 *
1127 * This is guaranteed by the 1216 * This is guaranteed by the fact that all LRU management
1128 * functions are intermediate 1217 * functions are intermediated by the LRU API calls like
1129 * list_lru_add_obj and list_ !! 1218 * list_lru_add and list_lru_del. List movement in this file
1130 * only ever occur through th 1219 * only ever occur through this functions or through callbacks
1131 * like this one, that are ca 1220 * like this one, that are called from the LRU API.
1132 * 1221 *
1133 * The only exceptions to thi 1222 * The only exceptions to this are functions like
1134 * shrink_dentry_list, and co 1223 * shrink_dentry_list, and code that first checks for the
1135 * DCACHE_SHRINK_LIST flag. 1224 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1136 * operating only with stack 1225 * operating only with stack provided lists after they are
1137 * properly isolated from the 1226 * properly isolated from the main list. It is thus, always a
1138 * local access. 1227 * local access.
1139 */ 1228 */
1140 return LRU_ROTATE; 1229 return LRU_ROTATE;
1141 } 1230 }
1142 1231
1143 d_lru_shrink_move(lru, dentry, freeab 1232 d_lru_shrink_move(lru, dentry, freeable);
1144 spin_unlock(&dentry->d_lock); 1233 spin_unlock(&dentry->d_lock);
1145 1234
1146 return LRU_REMOVED; 1235 return LRU_REMOVED;
1147 } 1236 }
1148 1237
1149 /** 1238 /**
1150 * prune_dcache_sb - shrink the dcache 1239 * prune_dcache_sb - shrink the dcache
1151 * @sb: superblock 1240 * @sb: superblock
1152 * @sc: shrink control, passed to list_lru_sh 1241 * @sc: shrink control, passed to list_lru_shrink_walk()
1153 * 1242 *
1154 * Attempt to shrink the superblock dcache LR 1243 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1155 * is done when we need more memory and calle 1244 * is done when we need more memory and called from the superblock shrinker
1156 * function. 1245 * function.
1157 * 1246 *
1158 * This function may fail to free any resourc 1247 * This function may fail to free any resources if all the dentries are in
1159 * use. 1248 * use.
1160 */ 1249 */
1161 long prune_dcache_sb(struct super_block *sb, 1250 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1162 { 1251 {
1163 LIST_HEAD(dispose); 1252 LIST_HEAD(dispose);
1164 long freed; 1253 long freed;
1165 1254
1166 freed = list_lru_shrink_walk(&sb->s_d 1255 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1167 dentry_l 1256 dentry_lru_isolate, &dispose);
1168 shrink_dentry_list(&dispose); 1257 shrink_dentry_list(&dispose);
1169 return freed; 1258 return freed;
1170 } 1259 }
1171 1260
1172 static enum lru_status dentry_lru_isolate_shr 1261 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1173 struct list_lru_one *lru, spi 1262 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1174 { 1263 {
1175 struct list_head *freeable = arg; 1264 struct list_head *freeable = arg;
1176 struct dentry *dentry = container_o 1265 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1177 1266
1178 /* 1267 /*
1179 * we are inverting the lru lock/dent 1268 * we are inverting the lru lock/dentry->d_lock here,
1180 * so use a trylock. If we fail to ge 1269 * so use a trylock. If we fail to get the lock, just skip
1181 * it 1270 * it
1182 */ 1271 */
1183 if (!spin_trylock(&dentry->d_lock)) 1272 if (!spin_trylock(&dentry->d_lock))
1184 return LRU_SKIP; 1273 return LRU_SKIP;
1185 1274
1186 d_lru_shrink_move(lru, dentry, freeab 1275 d_lru_shrink_move(lru, dentry, freeable);
1187 spin_unlock(&dentry->d_lock); 1276 spin_unlock(&dentry->d_lock);
1188 1277
1189 return LRU_REMOVED; 1278 return LRU_REMOVED;
1190 } 1279 }
1191 1280
1192 1281
1193 /** 1282 /**
1194 * shrink_dcache_sb - shrink dcache for a sup 1283 * shrink_dcache_sb - shrink dcache for a superblock
1195 * @sb: superblock 1284 * @sb: superblock
1196 * 1285 *
1197 * Shrink the dcache for the specified super 1286 * Shrink the dcache for the specified super block. This is used to free
1198 * the dcache before unmounting a file system 1287 * the dcache before unmounting a file system.
1199 */ 1288 */
1200 void shrink_dcache_sb(struct super_block *sb) 1289 void shrink_dcache_sb(struct super_block *sb)
1201 { 1290 {
1202 do { 1291 do {
1203 LIST_HEAD(dispose); 1292 LIST_HEAD(dispose);
1204 1293
1205 list_lru_walk(&sb->s_dentry_l 1294 list_lru_walk(&sb->s_dentry_lru,
1206 dentry_lru_isolate_sh 1295 dentry_lru_isolate_shrink, &dispose, 1024);
1207 shrink_dentry_list(&dispose); 1296 shrink_dentry_list(&dispose);
1208 } while (list_lru_count(&sb->s_dentry 1297 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1209 } 1298 }
1210 EXPORT_SYMBOL(shrink_dcache_sb); 1299 EXPORT_SYMBOL(shrink_dcache_sb);
1211 1300
1212 /** 1301 /**
1213 * enum d_walk_ret - action to talke during t 1302 * enum d_walk_ret - action to talke during tree walk
1214 * @D_WALK_CONTINUE: contrinue walk 1303 * @D_WALK_CONTINUE: contrinue walk
1215 * @D_WALK_QUIT: quit walk 1304 * @D_WALK_QUIT: quit walk
1216 * @D_WALK_NORETRY: quit when retry is ne 1305 * @D_WALK_NORETRY: quit when retry is needed
1217 * @D_WALK_SKIP: skip this dentry and 1306 * @D_WALK_SKIP: skip this dentry and its children
1218 */ 1307 */
1219 enum d_walk_ret { 1308 enum d_walk_ret {
1220 D_WALK_CONTINUE, 1309 D_WALK_CONTINUE,
1221 D_WALK_QUIT, 1310 D_WALK_QUIT,
1222 D_WALK_NORETRY, 1311 D_WALK_NORETRY,
1223 D_WALK_SKIP, 1312 D_WALK_SKIP,
1224 }; 1313 };
1225 1314
1226 /** 1315 /**
1227 * d_walk - walk the dentry tree 1316 * d_walk - walk the dentry tree
1228 * @parent: start of walk 1317 * @parent: start of walk
1229 * @data: data passed to @enter() and @ 1318 * @data: data passed to @enter() and @finish()
1230 * @enter: callback when first entering 1319 * @enter: callback when first entering the dentry
1231 * 1320 *
1232 * The @enter() callbacks are called with d_l 1321 * The @enter() callbacks are called with d_lock held.
1233 */ 1322 */
1234 static void d_walk(struct dentry *parent, voi 1323 static void d_walk(struct dentry *parent, void *data,
1235 enum d_walk_ret (*enter)(v 1324 enum d_walk_ret (*enter)(void *, struct dentry *))
1236 { 1325 {
1237 struct dentry *this_parent, *dentry; !! 1326 struct dentry *this_parent;
>> 1327 struct list_head *next;
1238 unsigned seq = 0; 1328 unsigned seq = 0;
1239 enum d_walk_ret ret; 1329 enum d_walk_ret ret;
1240 bool retry = true; 1330 bool retry = true;
1241 1331
1242 again: 1332 again:
1243 read_seqbegin_or_lock(&rename_lock, & 1333 read_seqbegin_or_lock(&rename_lock, &seq);
1244 this_parent = parent; 1334 this_parent = parent;
1245 spin_lock(&this_parent->d_lock); 1335 spin_lock(&this_parent->d_lock);
1246 1336
1247 ret = enter(data, this_parent); 1337 ret = enter(data, this_parent);
1248 switch (ret) { 1338 switch (ret) {
1249 case D_WALK_CONTINUE: 1339 case D_WALK_CONTINUE:
1250 break; 1340 break;
1251 case D_WALK_QUIT: 1341 case D_WALK_QUIT:
1252 case D_WALK_SKIP: 1342 case D_WALK_SKIP:
1253 goto out_unlock; 1343 goto out_unlock;
1254 case D_WALK_NORETRY: 1344 case D_WALK_NORETRY:
1255 retry = false; 1345 retry = false;
1256 break; 1346 break;
1257 } 1347 }
1258 repeat: 1348 repeat:
1259 dentry = d_first_child(this_parent); !! 1349 next = this_parent->d_subdirs.next;
1260 resume: 1350 resume:
1261 hlist_for_each_entry_from(dentry, d_s !! 1351 while (next != &this_parent->d_subdirs) {
>> 1352 struct list_head *tmp = next;
>> 1353 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
>> 1354 next = tmp->next;
>> 1355
1262 if (unlikely(dentry->d_flags 1356 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1263 continue; 1357 continue;
1264 1358
1265 spin_lock_nested(&dentry->d_l 1359 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1266 1360
1267 ret = enter(data, dentry); 1361 ret = enter(data, dentry);
1268 switch (ret) { 1362 switch (ret) {
1269 case D_WALK_CONTINUE: 1363 case D_WALK_CONTINUE:
1270 break; 1364 break;
1271 case D_WALK_QUIT: 1365 case D_WALK_QUIT:
1272 spin_unlock(&dentry-> 1366 spin_unlock(&dentry->d_lock);
1273 goto out_unlock; 1367 goto out_unlock;
1274 case D_WALK_NORETRY: 1368 case D_WALK_NORETRY:
1275 retry = false; 1369 retry = false;
1276 break; 1370 break;
1277 case D_WALK_SKIP: 1371 case D_WALK_SKIP:
1278 spin_unlock(&dentry-> 1372 spin_unlock(&dentry->d_lock);
1279 continue; 1373 continue;
1280 } 1374 }
1281 1375
1282 if (!hlist_empty(&dentry->d_c !! 1376 if (!list_empty(&dentry->d_subdirs)) {
1283 spin_unlock(&this_par 1377 spin_unlock(&this_parent->d_lock);
1284 spin_release(&dentry- 1378 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1285 this_parent = dentry; 1379 this_parent = dentry;
1286 spin_acquire(&this_pa 1380 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1287 goto repeat; 1381 goto repeat;
1288 } 1382 }
1289 spin_unlock(&dentry->d_lock); 1383 spin_unlock(&dentry->d_lock);
1290 } 1384 }
1291 /* 1385 /*
1292 * All done at this level ... ascend 1386 * All done at this level ... ascend and resume the search.
1293 */ 1387 */
1294 rcu_read_lock(); 1388 rcu_read_lock();
1295 ascend: 1389 ascend:
1296 if (this_parent != parent) { 1390 if (this_parent != parent) {
1297 dentry = this_parent; !! 1391 struct dentry *child = this_parent;
1298 this_parent = dentry->d_paren !! 1392 this_parent = child->d_parent;
1299 1393
1300 spin_unlock(&dentry->d_lock); !! 1394 spin_unlock(&child->d_lock);
1301 spin_lock(&this_parent->d_loc 1395 spin_lock(&this_parent->d_lock);
1302 1396
1303 /* might go back up the wrong 1397 /* might go back up the wrong parent if we have had a rename. */
1304 if (need_seqretry(&rename_loc 1398 if (need_seqretry(&rename_lock, seq))
1305 goto rename_retry; 1399 goto rename_retry;
1306 /* go into the first sibling 1400 /* go into the first sibling still alive */
1307 hlist_for_each_entry_continue !! 1401 do {
1308 if (likely(!(dentry-> !! 1402 next = child->d_child.next;
1309 rcu_read_unlo !! 1403 if (next == &this_parent->d_subdirs)
1310 goto resume; !! 1404 goto ascend;
1311 } !! 1405 child = list_entry(next, struct dentry, d_child);
1312 } !! 1406 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1313 goto ascend; !! 1407 rcu_read_unlock();
>> 1408 goto resume;
1314 } 1409 }
1315 if (need_seqretry(&rename_lock, seq)) 1410 if (need_seqretry(&rename_lock, seq))
1316 goto rename_retry; 1411 goto rename_retry;
1317 rcu_read_unlock(); 1412 rcu_read_unlock();
1318 1413
1319 out_unlock: 1414 out_unlock:
1320 spin_unlock(&this_parent->d_lock); 1415 spin_unlock(&this_parent->d_lock);
1321 done_seqretry(&rename_lock, seq); 1416 done_seqretry(&rename_lock, seq);
1322 return; 1417 return;
1323 1418
1324 rename_retry: 1419 rename_retry:
1325 spin_unlock(&this_parent->d_lock); 1420 spin_unlock(&this_parent->d_lock);
1326 rcu_read_unlock(); 1421 rcu_read_unlock();
1327 BUG_ON(seq & 1); 1422 BUG_ON(seq & 1);
1328 if (!retry) 1423 if (!retry)
1329 return; 1424 return;
1330 seq = 1; 1425 seq = 1;
1331 goto again; 1426 goto again;
1332 } 1427 }
1333 1428
1334 struct check_mount { 1429 struct check_mount {
1335 struct vfsmount *mnt; 1430 struct vfsmount *mnt;
1336 unsigned int mounted; 1431 unsigned int mounted;
1337 }; 1432 };
1338 1433
1339 static enum d_walk_ret path_check_mount(void 1434 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1340 { 1435 {
1341 struct check_mount *info = data; 1436 struct check_mount *info = data;
1342 struct path path = { .mnt = info->mnt 1437 struct path path = { .mnt = info->mnt, .dentry = dentry };
1343 1438
1344 if (likely(!d_mountpoint(dentry))) 1439 if (likely(!d_mountpoint(dentry)))
1345 return D_WALK_CONTINUE; 1440 return D_WALK_CONTINUE;
1346 if (__path_is_mountpoint(&path)) { 1441 if (__path_is_mountpoint(&path)) {
1347 info->mounted = 1; 1442 info->mounted = 1;
1348 return D_WALK_QUIT; 1443 return D_WALK_QUIT;
1349 } 1444 }
1350 return D_WALK_CONTINUE; 1445 return D_WALK_CONTINUE;
1351 } 1446 }
1352 1447
1353 /** 1448 /**
1354 * path_has_submounts - check for mounts over 1449 * path_has_submounts - check for mounts over a dentry in the
1355 * current namespace. 1450 * current namespace.
1356 * @parent: path to check. 1451 * @parent: path to check.
1357 * 1452 *
1358 * Return true if the parent or its subdirect 1453 * Return true if the parent or its subdirectories contain
1359 * a mount point in the current namespace. 1454 * a mount point in the current namespace.
1360 */ 1455 */
1361 int path_has_submounts(const struct path *par 1456 int path_has_submounts(const struct path *parent)
1362 { 1457 {
1363 struct check_mount data = { .mnt = pa 1458 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1364 1459
1365 read_seqlock_excl(&mount_lock); 1460 read_seqlock_excl(&mount_lock);
1366 d_walk(parent->dentry, &data, path_ch 1461 d_walk(parent->dentry, &data, path_check_mount);
1367 read_sequnlock_excl(&mount_lock); 1462 read_sequnlock_excl(&mount_lock);
1368 1463
1369 return data.mounted; 1464 return data.mounted;
1370 } 1465 }
1371 EXPORT_SYMBOL(path_has_submounts); 1466 EXPORT_SYMBOL(path_has_submounts);
1372 1467
1373 /* 1468 /*
1374 * Called by mount code to set a mountpoint a 1469 * Called by mount code to set a mountpoint and check if the mountpoint is
1375 * reachable (e.g. NFS can unhash a directory 1470 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1376 * subtree can become unreachable). 1471 * subtree can become unreachable).
1377 * 1472 *
1378 * Only one of d_invalidate() and d_set_mount 1473 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1379 * this reason take rename_lock and d_lock on 1474 * this reason take rename_lock and d_lock on dentry and ancestors.
1380 */ 1475 */
1381 int d_set_mounted(struct dentry *dentry) 1476 int d_set_mounted(struct dentry *dentry)
1382 { 1477 {
1383 struct dentry *p; 1478 struct dentry *p;
1384 int ret = -ENOENT; 1479 int ret = -ENOENT;
1385 write_seqlock(&rename_lock); 1480 write_seqlock(&rename_lock);
1386 for (p = dentry->d_parent; !IS_ROOT(p 1481 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1387 /* Need exclusion wrt. d_inva 1482 /* Need exclusion wrt. d_invalidate() */
1388 spin_lock(&p->d_lock); 1483 spin_lock(&p->d_lock);
1389 if (unlikely(d_unhashed(p))) 1484 if (unlikely(d_unhashed(p))) {
1390 spin_unlock(&p->d_loc 1485 spin_unlock(&p->d_lock);
1391 goto out; 1486 goto out;
1392 } 1487 }
1393 spin_unlock(&p->d_lock); 1488 spin_unlock(&p->d_lock);
1394 } 1489 }
1395 spin_lock(&dentry->d_lock); 1490 spin_lock(&dentry->d_lock);
1396 if (!d_unlinked(dentry)) { 1491 if (!d_unlinked(dentry)) {
1397 ret = -EBUSY; 1492 ret = -EBUSY;
1398 if (!d_mountpoint(dentry)) { 1493 if (!d_mountpoint(dentry)) {
1399 dentry->d_flags |= DC 1494 dentry->d_flags |= DCACHE_MOUNTED;
1400 ret = 0; 1495 ret = 0;
1401 } 1496 }
1402 } 1497 }
1403 spin_unlock(&dentry->d_lock); 1498 spin_unlock(&dentry->d_lock);
1404 out: 1499 out:
1405 write_sequnlock(&rename_lock); 1500 write_sequnlock(&rename_lock);
1406 return ret; 1501 return ret;
1407 } 1502 }
1408 1503
1409 /* 1504 /*
1410 * Search the dentry child list of the specif 1505 * Search the dentry child list of the specified parent,
1411 * and move any unused dentries to the end of 1506 * and move any unused dentries to the end of the unused
1412 * list for prune_dcache(). We descend to the 1507 * list for prune_dcache(). We descend to the next level
1413 * whenever the d_children list is non-empty !! 1508 * whenever the d_subdirs list is non-empty and continue
1414 * searching. 1509 * searching.
1415 * 1510 *
1416 * It returns zero iff there are no unused ch 1511 * It returns zero iff there are no unused children,
1417 * otherwise it returns the number of childr 1512 * otherwise it returns the number of children moved to
1418 * the end of the unused list. This may not b 1513 * the end of the unused list. This may not be the total
1419 * number of unused children, because select_ 1514 * number of unused children, because select_parent can
1420 * drop the lock and return early due to late 1515 * drop the lock and return early due to latency
1421 * constraints. 1516 * constraints.
1422 */ 1517 */
1423 1518
1424 struct select_data { 1519 struct select_data {
1425 struct dentry *start; 1520 struct dentry *start;
1426 union { 1521 union {
1427 long found; 1522 long found;
1428 struct dentry *victim; 1523 struct dentry *victim;
1429 }; 1524 };
1430 struct list_head dispose; 1525 struct list_head dispose;
1431 }; 1526 };
1432 1527
1433 static enum d_walk_ret select_collect(void *_ 1528 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1434 { 1529 {
1435 struct select_data *data = _data; 1530 struct select_data *data = _data;
1436 enum d_walk_ret ret = D_WALK_CONTINUE 1531 enum d_walk_ret ret = D_WALK_CONTINUE;
1437 1532
1438 if (data->start == dentry) 1533 if (data->start == dentry)
1439 goto out; 1534 goto out;
1440 1535
1441 if (dentry->d_flags & DCACHE_SHRINK_L 1536 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1442 data->found++; 1537 data->found++;
1443 } else if (!dentry->d_lockref.count) !! 1538 } else {
1444 to_shrink_list(dentry, &data- !! 1539 if (dentry->d_flags & DCACHE_LRU_LIST)
1445 data->found++; !! 1540 d_lru_del(dentry);
1446 } else if (dentry->d_lockref.count < !! 1541 if (!dentry->d_lockref.count) {
1447 data->found++; !! 1542 d_shrink_add(dentry, &data->dispose);
>> 1543 data->found++;
>> 1544 }
1448 } 1545 }
1449 /* 1546 /*
1450 * We can return to the caller if we 1547 * We can return to the caller if we have found some (this
1451 * ensures forward progress). We'll b 1548 * ensures forward progress). We'll be coming back to find
1452 * the rest. 1549 * the rest.
1453 */ 1550 */
1454 if (!list_empty(&data->dispose)) 1551 if (!list_empty(&data->dispose))
1455 ret = need_resched() ? D_WALK 1552 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1456 out: 1553 out:
1457 return ret; 1554 return ret;
1458 } 1555 }
1459 1556
1460 static enum d_walk_ret select_collect2(void * 1557 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1461 { 1558 {
1462 struct select_data *data = _data; 1559 struct select_data *data = _data;
1463 enum d_walk_ret ret = D_WALK_CONTINUE 1560 enum d_walk_ret ret = D_WALK_CONTINUE;
1464 1561
1465 if (data->start == dentry) 1562 if (data->start == dentry)
1466 goto out; 1563 goto out;
1467 1564
1468 if (!dentry->d_lockref.count) { !! 1565 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1469 if (dentry->d_flags & DCACHE_ !! 1566 if (!dentry->d_lockref.count) {
1470 rcu_read_lock(); 1567 rcu_read_lock();
1471 data->victim = dentry 1568 data->victim = dentry;
1472 return D_WALK_QUIT; 1569 return D_WALK_QUIT;
1473 } 1570 }
1474 to_shrink_list(dentry, &data- !! 1571 } else {
>> 1572 if (dentry->d_flags & DCACHE_LRU_LIST)
>> 1573 d_lru_del(dentry);
>> 1574 if (!dentry->d_lockref.count)
>> 1575 d_shrink_add(dentry, &data->dispose);
1475 } 1576 }
1476 /* 1577 /*
1477 * We can return to the caller if we 1578 * We can return to the caller if we have found some (this
1478 * ensures forward progress). We'll b 1579 * ensures forward progress). We'll be coming back to find
1479 * the rest. 1580 * the rest.
1480 */ 1581 */
1481 if (!list_empty(&data->dispose)) 1582 if (!list_empty(&data->dispose))
1482 ret = need_resched() ? D_WALK 1583 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1483 out: 1584 out:
1484 return ret; 1585 return ret;
1485 } 1586 }
1486 1587
1487 /** 1588 /**
1488 * shrink_dcache_parent - prune dcache 1589 * shrink_dcache_parent - prune dcache
1489 * @parent: parent of entries to prune 1590 * @parent: parent of entries to prune
1490 * 1591 *
1491 * Prune the dcache to remove unused children 1592 * Prune the dcache to remove unused children of the parent dentry.
1492 */ 1593 */
1493 void shrink_dcache_parent(struct dentry *pare 1594 void shrink_dcache_parent(struct dentry *parent)
1494 { 1595 {
1495 for (;;) { 1596 for (;;) {
1496 struct select_data data = {.s 1597 struct select_data data = {.start = parent};
1497 1598
1498 INIT_LIST_HEAD(&data.dispose) 1599 INIT_LIST_HEAD(&data.dispose);
1499 d_walk(parent, &data, select_ 1600 d_walk(parent, &data, select_collect);
1500 1601
1501 if (!list_empty(&data.dispose 1602 if (!list_empty(&data.dispose)) {
1502 shrink_dentry_list(&d 1603 shrink_dentry_list(&data.dispose);
1503 continue; 1604 continue;
1504 } 1605 }
1505 1606
1506 cond_resched(); 1607 cond_resched();
1507 if (!data.found) 1608 if (!data.found)
1508 break; 1609 break;
1509 data.victim = NULL; 1610 data.victim = NULL;
1510 d_walk(parent, &data, select_ 1611 d_walk(parent, &data, select_collect2);
1511 if (data.victim) { 1612 if (data.victim) {
>> 1613 struct dentry *parent;
1512 spin_lock(&data.victi 1614 spin_lock(&data.victim->d_lock);
1513 if (!lock_for_kill(da !! 1615 if (!shrink_lock_dentry(data.victim)) {
1514 spin_unlock(& 1616 spin_unlock(&data.victim->d_lock);
1515 rcu_read_unlo 1617 rcu_read_unlock();
1516 } else { 1618 } else {
1517 shrink_kill(d !! 1619 rcu_read_unlock();
>> 1620 parent = data.victim->d_parent;
>> 1621 if (parent != data.victim)
>> 1622 __dput_to_list(parent, &data.dispose);
>> 1623 __dentry_kill(data.victim);
1518 } 1624 }
1519 } 1625 }
1520 if (!list_empty(&data.dispose 1626 if (!list_empty(&data.dispose))
1521 shrink_dentry_list(&d 1627 shrink_dentry_list(&data.dispose);
1522 } 1628 }
1523 } 1629 }
1524 EXPORT_SYMBOL(shrink_dcache_parent); 1630 EXPORT_SYMBOL(shrink_dcache_parent);
1525 1631
1526 static enum d_walk_ret umount_check(void *_da 1632 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1527 { 1633 {
1528 /* it has busy descendents; complain 1634 /* it has busy descendents; complain about those instead */
1529 if (!hlist_empty(&dentry->d_children) !! 1635 if (!list_empty(&dentry->d_subdirs))
1530 return D_WALK_CONTINUE; 1636 return D_WALK_CONTINUE;
1531 1637
1532 /* root with refcount 1 is fine */ 1638 /* root with refcount 1 is fine */
1533 if (dentry == _data && dentry->d_lock 1639 if (dentry == _data && dentry->d_lockref.count == 1)
1534 return D_WALK_CONTINUE; 1640 return D_WALK_CONTINUE;
1535 1641
1536 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} !! 1642 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1537 " still in use (%d) [ 1643 " still in use (%d) [unmount of %s %s]\n",
1538 dentry, 1644 dentry,
1539 dentry->d_inode ? 1645 dentry->d_inode ?
1540 dentry->d_inode->i_ino 1646 dentry->d_inode->i_ino : 0UL,
1541 dentry, 1647 dentry,
1542 dentry->d_lockref.coun 1648 dentry->d_lockref.count,
1543 dentry->d_sb->s_type-> 1649 dentry->d_sb->s_type->name,
1544 dentry->d_sb->s_id); 1650 dentry->d_sb->s_id);
>> 1651 WARN_ON(1);
1545 return D_WALK_CONTINUE; 1652 return D_WALK_CONTINUE;
1546 } 1653 }
1547 1654
1548 static void do_one_tree(struct dentry *dentry 1655 static void do_one_tree(struct dentry *dentry)
1549 { 1656 {
1550 shrink_dcache_parent(dentry); 1657 shrink_dcache_parent(dentry);
1551 d_walk(dentry, dentry, umount_check); 1658 d_walk(dentry, dentry, umount_check);
1552 d_drop(dentry); 1659 d_drop(dentry);
1553 dput(dentry); 1660 dput(dentry);
1554 } 1661 }
1555 1662
1556 /* 1663 /*
1557 * destroy the dentries attached to a superbl 1664 * destroy the dentries attached to a superblock on unmounting
1558 */ 1665 */
1559 void shrink_dcache_for_umount(struct super_bl 1666 void shrink_dcache_for_umount(struct super_block *sb)
1560 { 1667 {
1561 struct dentry *dentry; 1668 struct dentry *dentry;
1562 1669
1563 rwsem_assert_held_write(&sb->s_umount !! 1670 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1564 1671
1565 dentry = sb->s_root; 1672 dentry = sb->s_root;
1566 sb->s_root = NULL; 1673 sb->s_root = NULL;
1567 do_one_tree(dentry); 1674 do_one_tree(dentry);
1568 1675
1569 while (!hlist_bl_empty(&sb->s_roots)) 1676 while (!hlist_bl_empty(&sb->s_roots)) {
1570 dentry = dget(hlist_bl_entry( 1677 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1571 do_one_tree(dentry); 1678 do_one_tree(dentry);
1572 } 1679 }
1573 } 1680 }
1574 1681
1575 static enum d_walk_ret find_submount(void *_d 1682 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1576 { 1683 {
1577 struct dentry **victim = _data; 1684 struct dentry **victim = _data;
1578 if (d_mountpoint(dentry)) { 1685 if (d_mountpoint(dentry)) {
1579 *victim = dget_dlock(dentry); !! 1686 __dget_dlock(dentry);
>> 1687 *victim = dentry;
1580 return D_WALK_QUIT; 1688 return D_WALK_QUIT;
1581 } 1689 }
1582 return D_WALK_CONTINUE; 1690 return D_WALK_CONTINUE;
1583 } 1691 }
1584 1692
1585 /** 1693 /**
1586 * d_invalidate - detach submounts, prune dca 1694 * d_invalidate - detach submounts, prune dcache, and drop
1587 * @dentry: dentry to invalidate (aka detach, 1695 * @dentry: dentry to invalidate (aka detach, prune and drop)
1588 */ 1696 */
1589 void d_invalidate(struct dentry *dentry) 1697 void d_invalidate(struct dentry *dentry)
1590 { 1698 {
1591 bool had_submounts = false; 1699 bool had_submounts = false;
1592 spin_lock(&dentry->d_lock); 1700 spin_lock(&dentry->d_lock);
1593 if (d_unhashed(dentry)) { 1701 if (d_unhashed(dentry)) {
1594 spin_unlock(&dentry->d_lock); 1702 spin_unlock(&dentry->d_lock);
1595 return; 1703 return;
1596 } 1704 }
1597 __d_drop(dentry); 1705 __d_drop(dentry);
1598 spin_unlock(&dentry->d_lock); 1706 spin_unlock(&dentry->d_lock);
1599 1707
1600 /* Negative dentries can be dropped w 1708 /* Negative dentries can be dropped without further checks */
1601 if (!dentry->d_inode) 1709 if (!dentry->d_inode)
1602 return; 1710 return;
1603 1711
1604 shrink_dcache_parent(dentry); 1712 shrink_dcache_parent(dentry);
1605 for (;;) { 1713 for (;;) {
1606 struct dentry *victim = NULL; 1714 struct dentry *victim = NULL;
1607 d_walk(dentry, &victim, find_ 1715 d_walk(dentry, &victim, find_submount);
1608 if (!victim) { 1716 if (!victim) {
1609 if (had_submounts) 1717 if (had_submounts)
1610 shrink_dcache 1718 shrink_dcache_parent(dentry);
1611 return; 1719 return;
1612 } 1720 }
1613 had_submounts = true; 1721 had_submounts = true;
1614 detach_mounts(victim); 1722 detach_mounts(victim);
1615 dput(victim); 1723 dput(victim);
1616 } 1724 }
1617 } 1725 }
1618 EXPORT_SYMBOL(d_invalidate); 1726 EXPORT_SYMBOL(d_invalidate);
1619 1727
1620 /** 1728 /**
1621 * __d_alloc - allocate a dcache ent 1729 * __d_alloc - allocate a dcache entry
1622 * @sb: filesystem it will belong to 1730 * @sb: filesystem it will belong to
1623 * @name: qstr of the name 1731 * @name: qstr of the name
1624 * 1732 *
1625 * Allocates a dentry. It returns %NULL if th 1733 * Allocates a dentry. It returns %NULL if there is insufficient memory
1626 * available. On a success the dentry is retu 1734 * available. On a success the dentry is returned. The name passed in is
1627 * copied and the copy passed in may be reuse 1735 * copied and the copy passed in may be reused after this call.
1628 */ 1736 */
1629 1737
1630 static struct dentry *__d_alloc(struct super_ 1738 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1631 { 1739 {
1632 struct dentry *dentry; 1740 struct dentry *dentry;
1633 char *dname; 1741 char *dname;
1634 int err; 1742 int err;
1635 1743
1636 dentry = kmem_cache_alloc_lru(dentry_ !! 1744 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1637 GFP_KER <<
1638 if (!dentry) 1745 if (!dentry)
1639 return NULL; 1746 return NULL;
1640 1747
1641 /* 1748 /*
1642 * We guarantee that the inline name 1749 * We guarantee that the inline name is always NUL-terminated.
1643 * This way the memcpy() done by the 1750 * This way the memcpy() done by the name switching in rename
1644 * will still always have a NUL at th 1751 * will still always have a NUL at the end, even if we might
1645 * be overwriting an internal NUL cha 1752 * be overwriting an internal NUL character
1646 */ 1753 */
1647 dentry->d_iname[DNAME_INLINE_LEN-1] = 1754 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1648 if (unlikely(!name)) { 1755 if (unlikely(!name)) {
1649 name = &slash_name; 1756 name = &slash_name;
1650 dname = dentry->d_iname; 1757 dname = dentry->d_iname;
1651 } else if (name->len > DNAME_INLINE_L 1758 } else if (name->len > DNAME_INLINE_LEN-1) {
1652 size_t size = offsetof(struct 1759 size_t size = offsetof(struct external_name, name[1]);
1653 struct external_name *p = kma 1760 struct external_name *p = kmalloc(size + name->len,
1654 1761 GFP_KERNEL_ACCOUNT |
1655 1762 __GFP_RECLAIMABLE);
1656 if (!p) { 1763 if (!p) {
1657 kmem_cache_free(dentr 1764 kmem_cache_free(dentry_cache, dentry);
1658 return NULL; 1765 return NULL;
1659 } 1766 }
1660 atomic_set(&p->u.count, 1); 1767 atomic_set(&p->u.count, 1);
1661 dname = p->name; 1768 dname = p->name;
1662 } else { 1769 } else {
1663 dname = dentry->d_iname; 1770 dname = dentry->d_iname;
1664 } 1771 }
1665 1772
1666 dentry->d_name.len = name->len; 1773 dentry->d_name.len = name->len;
1667 dentry->d_name.hash = name->hash; 1774 dentry->d_name.hash = name->hash;
1668 memcpy(dname, name->name, name->len); 1775 memcpy(dname, name->name, name->len);
1669 dname[name->len] = 0; 1776 dname[name->len] = 0;
1670 1777
1671 /* Make sure we always see the termin 1778 /* Make sure we always see the terminating NUL character */
1672 smp_store_release(&dentry->d_name.nam 1779 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1673 1780
1674 dentry->d_lockref.count = 1; 1781 dentry->d_lockref.count = 1;
1675 dentry->d_flags = 0; 1782 dentry->d_flags = 0;
1676 spin_lock_init(&dentry->d_lock); 1783 spin_lock_init(&dentry->d_lock);
1677 seqcount_spinlock_init(&dentry->d_seq 1784 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1678 dentry->d_inode = NULL; 1785 dentry->d_inode = NULL;
1679 dentry->d_parent = dentry; 1786 dentry->d_parent = dentry;
1680 dentry->d_sb = sb; 1787 dentry->d_sb = sb;
1681 dentry->d_op = NULL; 1788 dentry->d_op = NULL;
1682 dentry->d_fsdata = NULL; 1789 dentry->d_fsdata = NULL;
1683 INIT_HLIST_BL_NODE(&dentry->d_hash); 1790 INIT_HLIST_BL_NODE(&dentry->d_hash);
1684 INIT_LIST_HEAD(&dentry->d_lru); 1791 INIT_LIST_HEAD(&dentry->d_lru);
1685 INIT_HLIST_HEAD(&dentry->d_children); !! 1792 INIT_LIST_HEAD(&dentry->d_subdirs);
1686 INIT_HLIST_NODE(&dentry->d_u.d_alias) 1793 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1687 INIT_HLIST_NODE(&dentry->d_sib); !! 1794 INIT_LIST_HEAD(&dentry->d_child);
1688 d_set_d_op(dentry, dentry->d_sb->s_d_ 1795 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1689 1796
1690 if (dentry->d_op && dentry->d_op->d_i 1797 if (dentry->d_op && dentry->d_op->d_init) {
1691 err = dentry->d_op->d_init(de 1798 err = dentry->d_op->d_init(dentry);
1692 if (err) { 1799 if (err) {
1693 if (dname_external(de 1800 if (dname_external(dentry))
1694 kfree(externa 1801 kfree(external_name(dentry));
1695 kmem_cache_free(dentr 1802 kmem_cache_free(dentry_cache, dentry);
1696 return NULL; 1803 return NULL;
1697 } 1804 }
1698 } 1805 }
1699 1806
1700 this_cpu_inc(nr_dentry); 1807 this_cpu_inc(nr_dentry);
1701 1808
1702 return dentry; 1809 return dentry;
1703 } 1810 }
1704 1811
1705 /** 1812 /**
1706 * d_alloc - allocate a dcache ent 1813 * d_alloc - allocate a dcache entry
1707 * @parent: parent of entry to allocate 1814 * @parent: parent of entry to allocate
1708 * @name: qstr of the name 1815 * @name: qstr of the name
1709 * 1816 *
1710 * Allocates a dentry. It returns %NULL if th 1817 * Allocates a dentry. It returns %NULL if there is insufficient memory
1711 * available. On a success the dentry is retu 1818 * available. On a success the dentry is returned. The name passed in is
1712 * copied and the copy passed in may be reuse 1819 * copied and the copy passed in may be reused after this call.
1713 */ 1820 */
1714 struct dentry *d_alloc(struct dentry * parent 1821 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1715 { 1822 {
1716 struct dentry *dentry = __d_alloc(par 1823 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1717 if (!dentry) 1824 if (!dentry)
1718 return NULL; 1825 return NULL;
1719 spin_lock(&parent->d_lock); 1826 spin_lock(&parent->d_lock);
1720 /* 1827 /*
1721 * don't need child lock because it i 1828 * don't need child lock because it is not subject
1722 * to concurrency here 1829 * to concurrency here
1723 */ 1830 */
1724 dentry->d_parent = dget_dlock(parent) !! 1831 __dget_dlock(parent);
1725 hlist_add_head(&dentry->d_sib, &paren !! 1832 dentry->d_parent = parent;
>> 1833 list_add(&dentry->d_child, &parent->d_subdirs);
1726 spin_unlock(&parent->d_lock); 1834 spin_unlock(&parent->d_lock);
1727 1835
1728 return dentry; 1836 return dentry;
1729 } 1837 }
1730 EXPORT_SYMBOL(d_alloc); 1838 EXPORT_SYMBOL(d_alloc);
1731 1839
1732 struct dentry *d_alloc_anon(struct super_bloc 1840 struct dentry *d_alloc_anon(struct super_block *sb)
1733 { 1841 {
1734 return __d_alloc(sb, NULL); 1842 return __d_alloc(sb, NULL);
1735 } 1843 }
1736 EXPORT_SYMBOL(d_alloc_anon); 1844 EXPORT_SYMBOL(d_alloc_anon);
1737 1845
1738 struct dentry *d_alloc_cursor(struct dentry * 1846 struct dentry *d_alloc_cursor(struct dentry * parent)
1739 { 1847 {
1740 struct dentry *dentry = d_alloc_anon( 1848 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1741 if (dentry) { 1849 if (dentry) {
1742 dentry->d_flags |= DCACHE_DEN 1850 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1743 dentry->d_parent = dget(paren 1851 dentry->d_parent = dget(parent);
1744 } 1852 }
1745 return dentry; 1853 return dentry;
1746 } 1854 }
1747 1855
1748 /** 1856 /**
1749 * d_alloc_pseudo - allocate a dentry (for lo 1857 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1750 * @sb: the superblock 1858 * @sb: the superblock
1751 * @name: qstr of the name 1859 * @name: qstr of the name
1752 * 1860 *
1753 * For a filesystem that just pins its dentri 1861 * For a filesystem that just pins its dentries in memory and never
1754 * performs lookups at all, return an unhashe 1862 * performs lookups at all, return an unhashed IS_ROOT dentry.
1755 * This is used for pipes, sockets et.al. - t 1863 * This is used for pipes, sockets et.al. - the stuff that should
1756 * never be anyone's children or parents. Un 1864 * never be anyone's children or parents. Unlike all other
1757 * dentries, these will not have RCU delay be 1865 * dentries, these will not have RCU delay between dropping the
1758 * last reference and freeing them. 1866 * last reference and freeing them.
1759 * 1867 *
1760 * The only user is alloc_file_pseudo() and t 1868 * The only user is alloc_file_pseudo() and that's what should
1761 * be considered a public interface. Don't u 1869 * be considered a public interface. Don't use directly.
1762 */ 1870 */
1763 struct dentry *d_alloc_pseudo(struct super_bl 1871 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1764 { 1872 {
1765 static const struct dentry_operations <<
1766 .d_dname = simple_dname <<
1767 }; <<
1768 struct dentry *dentry = __d_alloc(sb, 1873 struct dentry *dentry = __d_alloc(sb, name);
1769 if (likely(dentry)) { !! 1874 if (likely(dentry))
1770 dentry->d_flags |= DCACHE_NOR 1875 dentry->d_flags |= DCACHE_NORCU;
1771 if (!sb->s_d_op) <<
1772 d_set_d_op(dentry, &a <<
1773 } <<
1774 return dentry; 1876 return dentry;
1775 } 1877 }
1776 1878
1777 struct dentry *d_alloc_name(struct dentry *pa 1879 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1778 { 1880 {
1779 struct qstr q; 1881 struct qstr q;
1780 1882
1781 q.name = name; 1883 q.name = name;
1782 q.hash_len = hashlen_string(parent, n 1884 q.hash_len = hashlen_string(parent, name);
1783 return d_alloc(parent, &q); 1885 return d_alloc(parent, &q);
1784 } 1886 }
1785 EXPORT_SYMBOL(d_alloc_name); 1887 EXPORT_SYMBOL(d_alloc_name);
1786 1888
1787 void d_set_d_op(struct dentry *dentry, const 1889 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1788 { 1890 {
1789 WARN_ON_ONCE(dentry->d_op); 1891 WARN_ON_ONCE(dentry->d_op);
1790 WARN_ON_ONCE(dentry->d_flags & (DCACH 1892 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1791 DCACHE_OP_COM 1893 DCACHE_OP_COMPARE |
1792 DCACHE_OP_REV 1894 DCACHE_OP_REVALIDATE |
1793 DCACHE_OP_WEA 1895 DCACHE_OP_WEAK_REVALIDATE |
1794 DCACHE_OP_DEL 1896 DCACHE_OP_DELETE |
1795 DCACHE_OP_REA 1897 DCACHE_OP_REAL));
1796 dentry->d_op = op; 1898 dentry->d_op = op;
1797 if (!op) 1899 if (!op)
1798 return; 1900 return;
1799 if (op->d_hash) 1901 if (op->d_hash)
1800 dentry->d_flags |= DCACHE_OP_ 1902 dentry->d_flags |= DCACHE_OP_HASH;
1801 if (op->d_compare) 1903 if (op->d_compare)
1802 dentry->d_flags |= DCACHE_OP_ 1904 dentry->d_flags |= DCACHE_OP_COMPARE;
1803 if (op->d_revalidate) 1905 if (op->d_revalidate)
1804 dentry->d_flags |= DCACHE_OP_ 1906 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1805 if (op->d_weak_revalidate) 1907 if (op->d_weak_revalidate)
1806 dentry->d_flags |= DCACHE_OP_ 1908 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1807 if (op->d_delete) 1909 if (op->d_delete)
1808 dentry->d_flags |= DCACHE_OP_ 1910 dentry->d_flags |= DCACHE_OP_DELETE;
1809 if (op->d_prune) 1911 if (op->d_prune)
1810 dentry->d_flags |= DCACHE_OP_ 1912 dentry->d_flags |= DCACHE_OP_PRUNE;
1811 if (op->d_real) 1913 if (op->d_real)
1812 dentry->d_flags |= DCACHE_OP_ 1914 dentry->d_flags |= DCACHE_OP_REAL;
1813 1915
1814 } 1916 }
1815 EXPORT_SYMBOL(d_set_d_op); 1917 EXPORT_SYMBOL(d_set_d_op);
1816 1918
>> 1919
>> 1920 /*
>> 1921 * d_set_fallthru - Mark a dentry as falling through to a lower layer
>> 1922 * @dentry - The dentry to mark
>> 1923 *
>> 1924 * Mark a dentry as falling through to the lower layer (as set with
>> 1925 * d_pin_lower()). This flag may be recorded on the medium.
>> 1926 */
>> 1927 void d_set_fallthru(struct dentry *dentry)
>> 1928 {
>> 1929 spin_lock(&dentry->d_lock);
>> 1930 dentry->d_flags |= DCACHE_FALLTHRU;
>> 1931 spin_unlock(&dentry->d_lock);
>> 1932 }
>> 1933 EXPORT_SYMBOL(d_set_fallthru);
>> 1934
1817 static unsigned d_flags_for_inode(struct inod 1935 static unsigned d_flags_for_inode(struct inode *inode)
1818 { 1936 {
1819 unsigned add_flags = DCACHE_REGULAR_T 1937 unsigned add_flags = DCACHE_REGULAR_TYPE;
1820 1938
1821 if (!inode) 1939 if (!inode)
1822 return DCACHE_MISS_TYPE; 1940 return DCACHE_MISS_TYPE;
1823 1941
1824 if (S_ISDIR(inode->i_mode)) { 1942 if (S_ISDIR(inode->i_mode)) {
1825 add_flags = DCACHE_DIRECTORY_ 1943 add_flags = DCACHE_DIRECTORY_TYPE;
1826 if (unlikely(!(inode->i_opfla 1944 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1827 if (unlikely(!inode-> 1945 if (unlikely(!inode->i_op->lookup))
1828 add_flags = D 1946 add_flags = DCACHE_AUTODIR_TYPE;
1829 else 1947 else
1830 inode->i_opfl 1948 inode->i_opflags |= IOP_LOOKUP;
1831 } 1949 }
1832 goto type_determined; 1950 goto type_determined;
1833 } 1951 }
1834 1952
1835 if (unlikely(!(inode->i_opflags & IOP 1953 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1836 if (unlikely(inode->i_op->get 1954 if (unlikely(inode->i_op->get_link)) {
1837 add_flags = DCACHE_SY 1955 add_flags = DCACHE_SYMLINK_TYPE;
1838 goto type_determined; 1956 goto type_determined;
1839 } 1957 }
1840 inode->i_opflags |= IOP_NOFOL 1958 inode->i_opflags |= IOP_NOFOLLOW;
1841 } 1959 }
1842 1960
1843 if (unlikely(!S_ISREG(inode->i_mode)) 1961 if (unlikely(!S_ISREG(inode->i_mode)))
1844 add_flags = DCACHE_SPECIAL_TY 1962 add_flags = DCACHE_SPECIAL_TYPE;
1845 1963
1846 type_determined: 1964 type_determined:
1847 if (unlikely(IS_AUTOMOUNT(inode))) 1965 if (unlikely(IS_AUTOMOUNT(inode)))
1848 add_flags |= DCACHE_NEED_AUTO 1966 add_flags |= DCACHE_NEED_AUTOMOUNT;
1849 return add_flags; 1967 return add_flags;
1850 } 1968 }
1851 1969
1852 static void __d_instantiate(struct dentry *de 1970 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1853 { 1971 {
1854 unsigned add_flags = d_flags_for_inod 1972 unsigned add_flags = d_flags_for_inode(inode);
1855 WARN_ON(d_in_lookup(dentry)); 1973 WARN_ON(d_in_lookup(dentry));
1856 1974
1857 spin_lock(&dentry->d_lock); 1975 spin_lock(&dentry->d_lock);
1858 /* 1976 /*
1859 * The negative counter only tracks d !! 1977 * Decrement negative dentry count if it was in the LRU list.
1860 * d_lru is on another list. <<
1861 */ 1978 */
1862 if ((dentry->d_flags & !! 1979 if (dentry->d_flags & DCACHE_LRU_LIST)
1863 (DCACHE_LRU_LIST|DCACHE_SHRINK_L <<
1864 this_cpu_dec(nr_dentry_negati 1980 this_cpu_dec(nr_dentry_negative);
1865 hlist_add_head(&dentry->d_u.d_alias, 1981 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1866 raw_write_seqcount_begin(&dentry->d_s 1982 raw_write_seqcount_begin(&dentry->d_seq);
1867 __d_set_inode_and_type(dentry, inode, 1983 __d_set_inode_and_type(dentry, inode, add_flags);
1868 raw_write_seqcount_end(&dentry->d_seq 1984 raw_write_seqcount_end(&dentry->d_seq);
1869 fsnotify_update_flags(dentry); 1985 fsnotify_update_flags(dentry);
1870 spin_unlock(&dentry->d_lock); 1986 spin_unlock(&dentry->d_lock);
1871 } 1987 }
1872 1988
1873 /** 1989 /**
1874 * d_instantiate - fill in inode information 1990 * d_instantiate - fill in inode information for a dentry
1875 * @entry: dentry to complete 1991 * @entry: dentry to complete
1876 * @inode: inode to attach to this dentry 1992 * @inode: inode to attach to this dentry
1877 * 1993 *
1878 * Fill in inode information in the entry. 1994 * Fill in inode information in the entry.
1879 * 1995 *
1880 * This turns negative dentries into producti 1996 * This turns negative dentries into productive full members
1881 * of society. 1997 * of society.
1882 * 1998 *
1883 * NOTE! This assumes that the inode count ha 1999 * NOTE! This assumes that the inode count has been incremented
1884 * (or otherwise set) by the caller to indica 2000 * (or otherwise set) by the caller to indicate that it is now
1885 * in use by the dcache. 2001 * in use by the dcache.
1886 */ 2002 */
1887 2003
1888 void d_instantiate(struct dentry *entry, stru 2004 void d_instantiate(struct dentry *entry, struct inode * inode)
1889 { 2005 {
1890 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 2006 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1891 if (inode) { 2007 if (inode) {
1892 security_d_instantiate(entry, 2008 security_d_instantiate(entry, inode);
1893 spin_lock(&inode->i_lock); 2009 spin_lock(&inode->i_lock);
1894 __d_instantiate(entry, inode) 2010 __d_instantiate(entry, inode);
1895 spin_unlock(&inode->i_lock); 2011 spin_unlock(&inode->i_lock);
1896 } 2012 }
1897 } 2013 }
1898 EXPORT_SYMBOL(d_instantiate); 2014 EXPORT_SYMBOL(d_instantiate);
1899 2015
1900 /* 2016 /*
1901 * This should be equivalent to d_instantiate 2017 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1902 * with lockdep-related part of unlock_new_in 2018 * with lockdep-related part of unlock_new_inode() done before
1903 * anything else. Use that instead of open-c 2019 * anything else. Use that instead of open-coding d_instantiate()/
1904 * unlock_new_inode() combinations. 2020 * unlock_new_inode() combinations.
1905 */ 2021 */
1906 void d_instantiate_new(struct dentry *entry, 2022 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1907 { 2023 {
1908 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 2024 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1909 BUG_ON(!inode); 2025 BUG_ON(!inode);
1910 lockdep_annotate_inode_mutex_key(inod 2026 lockdep_annotate_inode_mutex_key(inode);
1911 security_d_instantiate(entry, inode); 2027 security_d_instantiate(entry, inode);
1912 spin_lock(&inode->i_lock); 2028 spin_lock(&inode->i_lock);
1913 __d_instantiate(entry, inode); 2029 __d_instantiate(entry, inode);
1914 WARN_ON(!(inode->i_state & I_NEW)); 2030 WARN_ON(!(inode->i_state & I_NEW));
1915 inode->i_state &= ~I_NEW & ~I_CREATIN 2031 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(); 2032 smp_mb();
1922 inode_wake_up_bit(inode, __I_NEW); !! 2033 wake_up_bit(&inode->i_state, __I_NEW);
1923 spin_unlock(&inode->i_lock); 2034 spin_unlock(&inode->i_lock);
1924 } 2035 }
1925 EXPORT_SYMBOL(d_instantiate_new); 2036 EXPORT_SYMBOL(d_instantiate_new);
1926 2037
1927 struct dentry *d_make_root(struct inode *root 2038 struct dentry *d_make_root(struct inode *root_inode)
1928 { 2039 {
1929 struct dentry *res = NULL; 2040 struct dentry *res = NULL;
1930 2041
1931 if (root_inode) { 2042 if (root_inode) {
1932 res = d_alloc_anon(root_inode 2043 res = d_alloc_anon(root_inode->i_sb);
1933 if (res) 2044 if (res)
1934 d_instantiate(res, ro 2045 d_instantiate(res, root_inode);
1935 else 2046 else
1936 iput(root_inode); 2047 iput(root_inode);
1937 } 2048 }
1938 return res; 2049 return res;
1939 } 2050 }
1940 EXPORT_SYMBOL(d_make_root); 2051 EXPORT_SYMBOL(d_make_root);
1941 2052
>> 2053 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
>> 2054 struct inode *inode,
>> 2055 bool disconnected)
>> 2056 {
>> 2057 struct dentry *res;
>> 2058 unsigned add_flags;
>> 2059
>> 2060 security_d_instantiate(dentry, inode);
>> 2061 spin_lock(&inode->i_lock);
>> 2062 res = __d_find_any_alias(inode);
>> 2063 if (res) {
>> 2064 spin_unlock(&inode->i_lock);
>> 2065 dput(dentry);
>> 2066 goto out_iput;
>> 2067 }
>> 2068
>> 2069 /* attach a disconnected dentry */
>> 2070 add_flags = d_flags_for_inode(inode);
>> 2071
>> 2072 if (disconnected)
>> 2073 add_flags |= DCACHE_DISCONNECTED;
>> 2074
>> 2075 spin_lock(&dentry->d_lock);
>> 2076 __d_set_inode_and_type(dentry, inode, add_flags);
>> 2077 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
>> 2078 if (!disconnected) {
>> 2079 hlist_bl_lock(&dentry->d_sb->s_roots);
>> 2080 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
>> 2081 hlist_bl_unlock(&dentry->d_sb->s_roots);
>> 2082 }
>> 2083 spin_unlock(&dentry->d_lock);
>> 2084 spin_unlock(&inode->i_lock);
>> 2085
>> 2086 return dentry;
>> 2087
>> 2088 out_iput:
>> 2089 iput(inode);
>> 2090 return res;
>> 2091 }
>> 2092
>> 2093 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
>> 2094 {
>> 2095 return __d_instantiate_anon(dentry, inode, true);
>> 2096 }
>> 2097 EXPORT_SYMBOL(d_instantiate_anon);
>> 2098
1942 static struct dentry *__d_obtain_alias(struct 2099 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
1943 { 2100 {
1944 struct super_block *sb; !! 2101 struct dentry *tmp;
1945 struct dentry *new, *res; !! 2102 struct dentry *res;
1946 2103
1947 if (!inode) 2104 if (!inode)
1948 return ERR_PTR(-ESTALE); 2105 return ERR_PTR(-ESTALE);
1949 if (IS_ERR(inode)) 2106 if (IS_ERR(inode))
1950 return ERR_CAST(inode); 2107 return ERR_CAST(inode);
1951 2108
1952 sb = inode->i_sb; !! 2109 res = d_find_any_alias(inode);
1953 <<
1954 res = d_find_any_alias(inode); /* exi <<
1955 if (res) 2110 if (res)
1956 goto out; !! 2111 goto out_iput;
1957 2112
1958 new = d_alloc_anon(sb); !! 2113 tmp = d_alloc_anon(inode->i_sb);
1959 if (!new) { !! 2114 if (!tmp) {
1960 res = ERR_PTR(-ENOMEM); 2115 res = ERR_PTR(-ENOMEM);
1961 goto out; !! 2116 goto out_iput;
1962 } 2117 }
1963 2118
1964 security_d_instantiate(new, inode); !! 2119 return __d_instantiate_anon(tmp, inode, disconnected);
1965 spin_lock(&inode->i_lock); <<
1966 res = __d_find_any_alias(inode); /* r <<
1967 if (likely(!res)) { /* still no alias <<
1968 unsigned add_flags = d_flags_ <<
1969 <<
1970 if (disconnected) <<
1971 add_flags |= DCACHE_D <<
1972 2120
1973 spin_lock(&new->d_lock); !! 2121 out_iput:
1974 __d_set_inode_and_type(new, i <<
1975 hlist_add_head(&new->d_u.d_al <<
1976 if (!disconnected) { <<
1977 hlist_bl_lock(&sb->s_ <<
1978 hlist_bl_add_head(&ne <<
1979 hlist_bl_unlock(&sb-> <<
1980 } <<
1981 spin_unlock(&new->d_lock); <<
1982 spin_unlock(&inode->i_lock); <<
1983 inode = NULL; /* consumed by <<
1984 res = new; <<
1985 } else { <<
1986 spin_unlock(&inode->i_lock); <<
1987 dput(new); <<
1988 } <<
1989 <<
1990 out: <<
1991 iput(inode); 2122 iput(inode);
1992 return res; 2123 return res;
1993 } 2124 }
1994 2125
1995 /** 2126 /**
1996 * d_obtain_alias - find or allocate a DISCON 2127 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1997 * @inode: inode to allocate the dentry for 2128 * @inode: inode to allocate the dentry for
1998 * 2129 *
1999 * Obtain a dentry for an inode resulting fro 2130 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2000 * similar open by handle operations. The re 2131 * similar open by handle operations. The returned dentry may be anonymous,
2001 * or may have a full name (if the inode was 2132 * or may have a full name (if the inode was already in the cache).
2002 * 2133 *
2003 * When called on a directory inode, we must 2134 * When called on a directory inode, we must ensure that the inode only ever
2004 * has one dentry. If a dentry is found, tha 2135 * has one dentry. If a dentry is found, that is returned instead of
2005 * allocating a new one. 2136 * allocating a new one.
2006 * 2137 *
2007 * On successful return, the reference to the 2138 * On successful return, the reference to the inode has been transferred
2008 * to the dentry. In case of an error the re 2139 * 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 2140 * 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 2141 * be passed in and the error will be propagated to the return value,
2011 * with a %NULL @inode replaced by ERR_PTR(-E 2142 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2012 */ 2143 */
2013 struct dentry *d_obtain_alias(struct inode *i 2144 struct dentry *d_obtain_alias(struct inode *inode)
2014 { 2145 {
2015 return __d_obtain_alias(inode, true); 2146 return __d_obtain_alias(inode, true);
2016 } 2147 }
2017 EXPORT_SYMBOL(d_obtain_alias); 2148 EXPORT_SYMBOL(d_obtain_alias);
2018 2149
2019 /** 2150 /**
2020 * d_obtain_root - find or allocate a dentry 2151 * d_obtain_root - find or allocate a dentry for a given inode
2021 * @inode: inode to allocate the dentry for 2152 * @inode: inode to allocate the dentry for
2022 * 2153 *
2023 * Obtain an IS_ROOT dentry for the root of a 2154 * Obtain an IS_ROOT dentry for the root of a filesystem.
2024 * 2155 *
2025 * We must ensure that directory inodes only 2156 * We must ensure that directory inodes only ever have one dentry. If a
2026 * dentry is found, that is returned instead 2157 * dentry is found, that is returned instead of allocating a new one.
2027 * 2158 *
2028 * On successful return, the reference to the 2159 * On successful return, the reference to the inode has been transferred
2029 * to the dentry. In case of an error the re 2160 * to the dentry. In case of an error the reference on the inode is
2030 * released. A %NULL or IS_ERR inode may be 2161 * released. A %NULL or IS_ERR inode may be passed in and will be the
2031 * error will be propagate to the return valu 2162 * error will be propagate to the return value, with a %NULL @inode
2032 * replaced by ERR_PTR(-ESTALE). 2163 * replaced by ERR_PTR(-ESTALE).
2033 */ 2164 */
2034 struct dentry *d_obtain_root(struct inode *in 2165 struct dentry *d_obtain_root(struct inode *inode)
2035 { 2166 {
2036 return __d_obtain_alias(inode, false) 2167 return __d_obtain_alias(inode, false);
2037 } 2168 }
2038 EXPORT_SYMBOL(d_obtain_root); 2169 EXPORT_SYMBOL(d_obtain_root);
2039 2170
2040 /** 2171 /**
2041 * d_add_ci - lookup or allocate new dentry w 2172 * d_add_ci - lookup or allocate new dentry with case-exact name
2042 * @inode: the inode case-insensitive lookup 2173 * @inode: the inode case-insensitive lookup has found
2043 * @dentry: the negative dentry that was pass 2174 * @dentry: the negative dentry that was passed to the parent's lookup func
2044 * @name: the case-exact name to be associa 2175 * @name: the case-exact name to be associated with the returned dentry
2045 * 2176 *
2046 * This is to avoid filling the dcache with c 2177 * This is to avoid filling the dcache with case-insensitive names to the
2047 * same inode, only the actual correct case i 2178 * same inode, only the actual correct case is stored in the dcache for
2048 * case-insensitive filesystems. 2179 * case-insensitive filesystems.
2049 * 2180 *
2050 * For a case-insensitive lookup match and if 2181 * For a case-insensitive lookup match and if the case-exact dentry
2051 * already exists in the dcache, use it and r 2182 * already exists in the dcache, use it and return it.
2052 * 2183 *
2053 * If no entry exists with the exact case nam 2184 * If no entry exists with the exact case name, allocate new dentry with
2054 * the exact case, and return the spliced ent 2185 * the exact case, and return the spliced entry.
2055 */ 2186 */
2056 struct dentry *d_add_ci(struct dentry *dentry 2187 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2057 struct qstr *name) 2188 struct qstr *name)
2058 { 2189 {
2059 struct dentry *found, *res; 2190 struct dentry *found, *res;
2060 2191
2061 /* 2192 /*
2062 * First check if a dentry matching t 2193 * First check if a dentry matching the name already exists,
2063 * if not go ahead and create it now. 2194 * if not go ahead and create it now.
2064 */ 2195 */
2065 found = d_hash_and_lookup(dentry->d_p 2196 found = d_hash_and_lookup(dentry->d_parent, name);
2066 if (found) { 2197 if (found) {
2067 iput(inode); 2198 iput(inode);
2068 return found; 2199 return found;
2069 } 2200 }
2070 if (d_in_lookup(dentry)) { 2201 if (d_in_lookup(dentry)) {
2071 found = d_alloc_parallel(dent 2202 found = d_alloc_parallel(dentry->d_parent, name,
2072 dentr 2203 dentry->d_wait);
2073 if (IS_ERR(found) || !d_in_lo 2204 if (IS_ERR(found) || !d_in_lookup(found)) {
2074 iput(inode); 2205 iput(inode);
2075 return found; 2206 return found;
2076 } 2207 }
2077 } else { 2208 } else {
2078 found = d_alloc(dentry->d_par 2209 found = d_alloc(dentry->d_parent, name);
2079 if (!found) { 2210 if (!found) {
2080 iput(inode); 2211 iput(inode);
2081 return ERR_PTR(-ENOME 2212 return ERR_PTR(-ENOMEM);
2082 } 2213 }
2083 } 2214 }
2084 res = d_splice_alias(inode, found); 2215 res = d_splice_alias(inode, found);
2085 if (res) { 2216 if (res) {
2086 d_lookup_done(found); <<
2087 dput(found); 2217 dput(found);
2088 return res; 2218 return res;
2089 } 2219 }
2090 return found; 2220 return found;
2091 } 2221 }
2092 EXPORT_SYMBOL(d_add_ci); 2222 EXPORT_SYMBOL(d_add_ci);
2093 2223
2094 /** !! 2224
2095 * d_same_name - compare dentry name with cas !! 2225 static inline bool d_same_name(const struct dentry *dentry,
2096 * @parent: parent dentry !! 2226 const struct dentry *parent,
2097 * @dentry: the negative dentry that was pass !! 2227 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 { 2228 {
2105 if (likely(!(parent->d_flags & DCACHE 2229 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2106 if (dentry->d_name.len != nam 2230 if (dentry->d_name.len != name->len)
2107 return false; 2231 return false;
2108 return dentry_cmp(dentry, nam 2232 return dentry_cmp(dentry, name->name, name->len) == 0;
2109 } 2233 }
2110 return parent->d_op->d_compare(dentry 2234 return parent->d_op->d_compare(dentry,
2111 dentry 2235 dentry->d_name.len, dentry->d_name.name,
2112 name) 2236 name) == 0;
2113 } 2237 }
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 2238
2158 /** 2239 /**
2159 * __d_lookup_rcu - search for a dentry (racy 2240 * __d_lookup_rcu - search for a dentry (racy, store-free)
2160 * @parent: parent dentry 2241 * @parent: parent dentry
2161 * @name: qstr of name we wish to find 2242 * @name: qstr of name we wish to find
2162 * @seqp: returns d_seq value at the point wh 2243 * @seqp: returns d_seq value at the point where the dentry was found
2163 * Returns: dentry, or NULL 2244 * Returns: dentry, or NULL
2164 * 2245 *
2165 * __d_lookup_rcu is the dcache lookup functi 2246 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2166 * resolution (store-free path walking) desig 2247 * resolution (store-free path walking) design described in
2167 * Documentation/filesystems/path-lookup.txt. 2248 * Documentation/filesystems/path-lookup.txt.
2168 * 2249 *
2169 * This is not to be used outside core vfs. 2250 * This is not to be used outside core vfs.
2170 * 2251 *
2171 * __d_lookup_rcu must only be used in rcu-wa 2252 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2172 * held, and rcu_read_lock held. The returned 2253 * held, and rcu_read_lock held. The returned dentry must not be stored into
2173 * without taking d_lock and checking d_seq s 2254 * without taking d_lock and checking d_seq sequence count against @seq
2174 * returned here. 2255 * returned here.
2175 * 2256 *
>> 2257 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
>> 2258 * function.
>> 2259 *
2176 * Alternatively, __d_lookup_rcu may be calle 2260 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2177 * the returned dentry, so long as its parent 2261 * the returned dentry, so long as its parent's seqlock is checked after the
2178 * child is looked up. Thus, an interlocking 2262 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2179 * is formed, giving integrity down the path 2263 * is formed, giving integrity down the path walk.
2180 * 2264 *
2181 * NOTE! The caller *has* to check the result 2265 * NOTE! The caller *has* to check the resulting dentry against the sequence
2182 * number we've returned before using any of 2266 * number we've returned before using any of the resulting dentry state!
2183 */ 2267 */
2184 struct dentry *__d_lookup_rcu(const struct de 2268 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2185 const struct 2269 const struct qstr *name,
2186 unsigned *seq 2270 unsigned *seqp)
2187 { 2271 {
2188 u64 hashlen = name->hash_len; 2272 u64 hashlen = name->hash_len;
2189 const unsigned char *str = name->name 2273 const unsigned char *str = name->name;
2190 struct hlist_bl_head *b = d_hash(hash !! 2274 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2191 struct hlist_bl_node *node; 2275 struct hlist_bl_node *node;
2192 struct dentry *dentry; 2276 struct dentry *dentry;
2193 2277
2194 /* 2278 /*
2195 * Note: There is significant duplica 2279 * Note: There is significant duplication with __d_lookup_rcu which is
2196 * required to prevent single threade 2280 * required to prevent single threaded performance regressions
2197 * especially on architectures where 2281 * especially on architectures where smp_rmb (in seqcounts) are costly.
2198 * Keep the two functions in sync. 2282 * Keep the two functions in sync.
2199 */ 2283 */
2200 2284
2201 if (unlikely(parent->d_flags & DCACHE <<
2202 return __d_lookup_rcu_op_comp <<
2203 <<
2204 /* 2285 /*
2205 * The hash list is protected using R 2286 * The hash list is protected using RCU.
2206 * 2287 *
2207 * Carefully use d_seq when comparing 2288 * Carefully use d_seq when comparing a candidate dentry, to avoid
2208 * races with d_move(). 2289 * races with d_move().
2209 * 2290 *
2210 * It is possible that concurrent ren 2291 * It is possible that concurrent renames can mess up our list
2211 * walk here and result in missing ou 2292 * walk here and result in missing our dentry, resulting in the
2212 * false-negative result. d_lookup() 2293 * false-negative result. d_lookup() protects against concurrent
2213 * renames using rename_lock seqlock. 2294 * renames using rename_lock seqlock.
2214 * 2295 *
2215 * See Documentation/filesystems/path 2296 * See Documentation/filesystems/path-lookup.txt for more details.
2216 */ 2297 */
2217 hlist_bl_for_each_entry_rcu(dentry, n 2298 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2218 unsigned seq; 2299 unsigned seq;
2219 2300
>> 2301 seqretry:
2220 /* 2302 /*
2221 * The dentry sequence count 2303 * The dentry sequence count protects us from concurrent
2222 * renames, and thus protects 2304 * renames, and thus protects parent and name fields.
2223 * 2305 *
2224 * The caller must perform a 2306 * The caller must perform a seqcount check in order
2225 * to do anything useful with 2307 * to do anything useful with the returned dentry.
2226 * 2308 *
2227 * NOTE! We do a "raw" seqcou 2309 * NOTE! We do a "raw" seqcount_begin here. That means that
2228 * we don't wait for the sequ 2310 * we don't wait for the sequence count to stabilize if it
2229 * is in the middle of a sequ 2311 * is in the middle of a sequence change. If we do the slow
2230 * dentry compare, we will do 2312 * dentry compare, we will do seqretries until it is stable,
2231 * and if we end up with a su 2313 * and if we end up with a successful lookup, we actually
2232 * want to exit RCU lookup an 2314 * want to exit RCU lookup anyway.
2233 * 2315 *
2234 * Note that raw_seqcount_beg 2316 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2235 * we are still guaranteed NU 2317 * we are still guaranteed NUL-termination of ->d_name.name.
2236 */ 2318 */
2237 seq = raw_seqcount_begin(&den 2319 seq = raw_seqcount_begin(&dentry->d_seq);
2238 if (dentry->d_parent != paren 2320 if (dentry->d_parent != parent)
2239 continue; 2321 continue;
2240 if (d_unhashed(dentry)) 2322 if (d_unhashed(dentry))
2241 continue; 2323 continue;
2242 if (dentry->d_name.hash_len ! !! 2324
2243 continue; !! 2325 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2244 if (dentry_cmp(dentry, str, h !! 2326 int tlen;
2245 continue; !! 2327 const char *tname;
>> 2328 if (dentry->d_name.hash != hashlen_hash(hashlen))
>> 2329 continue;
>> 2330 tlen = dentry->d_name.len;
>> 2331 tname = dentry->d_name.name;
>> 2332 /* we want a consistent (name,len) pair */
>> 2333 if (read_seqcount_retry(&dentry->d_seq, seq)) {
>> 2334 cpu_relax();
>> 2335 goto seqretry;
>> 2336 }
>> 2337 if (parent->d_op->d_compare(dentry,
>> 2338 tlen, tname, name) != 0)
>> 2339 continue;
>> 2340 } else {
>> 2341 if (dentry->d_name.hash_len != hashlen)
>> 2342 continue;
>> 2343 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
>> 2344 continue;
>> 2345 }
2246 *seqp = seq; 2346 *seqp = seq;
2247 return dentry; 2347 return dentry;
2248 } 2348 }
2249 return NULL; 2349 return NULL;
2250 } 2350 }
2251 2351
2252 /** 2352 /**
2253 * d_lookup - search for a dentry 2353 * d_lookup - search for a dentry
2254 * @parent: parent dentry 2354 * @parent: parent dentry
2255 * @name: qstr of name we wish to find 2355 * @name: qstr of name we wish to find
2256 * Returns: dentry, or NULL 2356 * Returns: dentry, or NULL
2257 * 2357 *
2258 * d_lookup searches the children of the pare 2358 * d_lookup searches the children of the parent dentry for the name in
2259 * question. If the dentry is found its refer 2359 * question. If the dentry is found its reference count is incremented and the
2260 * dentry is returned. The caller must use dp 2360 * dentry is returned. The caller must use dput to free the entry when it has
2261 * finished using it. %NULL is returned if th 2361 * finished using it. %NULL is returned if the dentry does not exist.
2262 */ 2362 */
2263 struct dentry *d_lookup(const struct dentry * 2363 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2264 { 2364 {
2265 struct dentry *dentry; 2365 struct dentry *dentry;
2266 unsigned seq; 2366 unsigned seq;
2267 2367
2268 do { 2368 do {
2269 seq = read_seqbegin(&rename_l 2369 seq = read_seqbegin(&rename_lock);
2270 dentry = __d_lookup(parent, n 2370 dentry = __d_lookup(parent, name);
2271 if (dentry) 2371 if (dentry)
2272 break; 2372 break;
2273 } while (read_seqretry(&rename_lock, 2373 } while (read_seqretry(&rename_lock, seq));
2274 return dentry; 2374 return dentry;
2275 } 2375 }
2276 EXPORT_SYMBOL(d_lookup); 2376 EXPORT_SYMBOL(d_lookup);
2277 2377
2278 /** 2378 /**
2279 * __d_lookup - search for a dentry (racy) 2379 * __d_lookup - search for a dentry (racy)
2280 * @parent: parent dentry 2380 * @parent: parent dentry
2281 * @name: qstr of name we wish to find 2381 * @name: qstr of name we wish to find
2282 * Returns: dentry, or NULL 2382 * Returns: dentry, or NULL
2283 * 2383 *
2284 * __d_lookup is like d_lookup, however it ma 2384 * __d_lookup is like d_lookup, however it may (rarely) return a
2285 * false-negative result due to unrelated ren 2385 * false-negative result due to unrelated rename activity.
2286 * 2386 *
2287 * __d_lookup is slightly faster by avoiding 2387 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2288 * however it must be used carefully, eg. wit 2388 * however it must be used carefully, eg. with a following d_lookup in
2289 * the case of failure. 2389 * the case of failure.
2290 * 2390 *
2291 * __d_lookup callers must be commented. 2391 * __d_lookup callers must be commented.
2292 */ 2392 */
2293 struct dentry *__d_lookup(const struct dentry 2393 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2294 { 2394 {
2295 unsigned int hash = name->hash; 2395 unsigned int hash = name->hash;
2296 struct hlist_bl_head *b = d_hash(hash 2396 struct hlist_bl_head *b = d_hash(hash);
2297 struct hlist_bl_node *node; 2397 struct hlist_bl_node *node;
2298 struct dentry *found = NULL; 2398 struct dentry *found = NULL;
2299 struct dentry *dentry; 2399 struct dentry *dentry;
2300 2400
2301 /* 2401 /*
2302 * Note: There is significant duplica 2402 * Note: There is significant duplication with __d_lookup_rcu which is
2303 * required to prevent single threade 2403 * required to prevent single threaded performance regressions
2304 * especially on architectures where 2404 * especially on architectures where smp_rmb (in seqcounts) are costly.
2305 * Keep the two functions in sync. 2405 * Keep the two functions in sync.
2306 */ 2406 */
2307 2407
2308 /* 2408 /*
2309 * The hash list is protected using R 2409 * The hash list is protected using RCU.
2310 * 2410 *
2311 * Take d_lock when comparing a candi 2411 * Take d_lock when comparing a candidate dentry, to avoid races
2312 * with d_move(). 2412 * with d_move().
2313 * 2413 *
2314 * It is possible that concurrent ren 2414 * It is possible that concurrent renames can mess up our list
2315 * walk here and result in missing ou 2415 * walk here and result in missing our dentry, resulting in the
2316 * false-negative result. d_lookup() 2416 * false-negative result. d_lookup() protects against concurrent
2317 * renames using rename_lock seqlock. 2417 * renames using rename_lock seqlock.
2318 * 2418 *
2319 * See Documentation/filesystems/path 2419 * See Documentation/filesystems/path-lookup.txt for more details.
2320 */ 2420 */
2321 rcu_read_lock(); 2421 rcu_read_lock();
2322 2422
2323 hlist_bl_for_each_entry_rcu(dentry, n 2423 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2324 2424
2325 if (dentry->d_name.hash != ha 2425 if (dentry->d_name.hash != hash)
2326 continue; 2426 continue;
2327 2427
2328 spin_lock(&dentry->d_lock); 2428 spin_lock(&dentry->d_lock);
2329 if (dentry->d_parent != paren 2429 if (dentry->d_parent != parent)
2330 goto next; 2430 goto next;
2331 if (d_unhashed(dentry)) 2431 if (d_unhashed(dentry))
2332 goto next; 2432 goto next;
2333 2433
2334 if (!d_same_name(dentry, pare 2434 if (!d_same_name(dentry, parent, name))
2335 goto next; 2435 goto next;
2336 2436
2337 dentry->d_lockref.count++; 2437 dentry->d_lockref.count++;
2338 found = dentry; 2438 found = dentry;
2339 spin_unlock(&dentry->d_lock); 2439 spin_unlock(&dentry->d_lock);
2340 break; 2440 break;
2341 next: 2441 next:
2342 spin_unlock(&dentry->d_lock); 2442 spin_unlock(&dentry->d_lock);
2343 } 2443 }
2344 rcu_read_unlock(); 2444 rcu_read_unlock();
2345 2445
2346 return found; 2446 return found;
2347 } 2447 }
2348 2448
2349 /** 2449 /**
2350 * d_hash_and_lookup - hash the qstr then sea 2450 * d_hash_and_lookup - hash the qstr then search for a dentry
2351 * @dir: Directory to search in 2451 * @dir: Directory to search in
2352 * @name: qstr of name we wish to find 2452 * @name: qstr of name we wish to find
2353 * 2453 *
2354 * On lookup failure NULL is returned; on bad 2454 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2355 */ 2455 */
2356 struct dentry *d_hash_and_lookup(struct dentr 2456 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2357 { 2457 {
2358 /* 2458 /*
2359 * Check for a fs-specific hash funct 2459 * Check for a fs-specific hash function. Note that we must
2360 * calculate the standard hash first, 2460 * calculate the standard hash first, as the d_op->d_hash()
2361 * routine may choose to leave the ha 2461 * routine may choose to leave the hash value unchanged.
2362 */ 2462 */
2363 name->hash = full_name_hash(dir, name 2463 name->hash = full_name_hash(dir, name->name, name->len);
2364 if (dir->d_flags & DCACHE_OP_HASH) { 2464 if (dir->d_flags & DCACHE_OP_HASH) {
2365 int err = dir->d_op->d_hash(d 2465 int err = dir->d_op->d_hash(dir, name);
2366 if (unlikely(err < 0)) 2466 if (unlikely(err < 0))
2367 return ERR_PTR(err); 2467 return ERR_PTR(err);
2368 } 2468 }
2369 return d_lookup(dir, name); 2469 return d_lookup(dir, name);
2370 } 2470 }
2371 EXPORT_SYMBOL(d_hash_and_lookup); 2471 EXPORT_SYMBOL(d_hash_and_lookup);
2372 2472
2373 /* 2473 /*
2374 * When a file is deleted, we have two option 2474 * When a file is deleted, we have two options:
2375 * - turn this dentry into a negative dentry 2475 * - turn this dentry into a negative dentry
2376 * - unhash this dentry and free it. 2476 * - unhash this dentry and free it.
2377 * 2477 *
2378 * Usually, we want to just turn this into 2478 * Usually, we want to just turn this into
2379 * a negative dentry, but if anybody else is 2479 * a negative dentry, but if anybody else is
2380 * currently using the dentry or the inode 2480 * currently using the dentry or the inode
2381 * we can't do that and we fall back on remov 2481 * we can't do that and we fall back on removing
2382 * it from the hash queues and waiting for 2482 * it from the hash queues and waiting for
2383 * it to be deleted later when it has no user 2483 * it to be deleted later when it has no users
2384 */ 2484 */
2385 2485
2386 /** 2486 /**
2387 * d_delete - delete a dentry 2487 * d_delete - delete a dentry
2388 * @dentry: The dentry to delete 2488 * @dentry: The dentry to delete
2389 * 2489 *
2390 * Turn the dentry into a negative dentry if 2490 * Turn the dentry into a negative dentry if possible, otherwise
2391 * remove it from the hash queues so it can b 2491 * remove it from the hash queues so it can be deleted later
2392 */ 2492 */
2393 2493
2394 void d_delete(struct dentry * dentry) 2494 void d_delete(struct dentry * dentry)
2395 { 2495 {
2396 struct inode *inode = dentry->d_inode 2496 struct inode *inode = dentry->d_inode;
2397 2497
2398 spin_lock(&inode->i_lock); 2498 spin_lock(&inode->i_lock);
2399 spin_lock(&dentry->d_lock); 2499 spin_lock(&dentry->d_lock);
2400 /* 2500 /*
2401 * Are we the only user? 2501 * Are we the only user?
2402 */ 2502 */
2403 if (dentry->d_lockref.count == 1) { 2503 if (dentry->d_lockref.count == 1) {
2404 dentry->d_flags &= ~DCACHE_CA 2504 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2405 dentry_unlink_inode(dentry); 2505 dentry_unlink_inode(dentry);
2406 } else { 2506 } else {
2407 __d_drop(dentry); 2507 __d_drop(dentry);
2408 spin_unlock(&dentry->d_lock); 2508 spin_unlock(&dentry->d_lock);
2409 spin_unlock(&inode->i_lock); 2509 spin_unlock(&inode->i_lock);
2410 } 2510 }
2411 } 2511 }
2412 EXPORT_SYMBOL(d_delete); 2512 EXPORT_SYMBOL(d_delete);
2413 2513
2414 static void __d_rehash(struct dentry *entry) 2514 static void __d_rehash(struct dentry *entry)
2415 { 2515 {
2416 struct hlist_bl_head *b = d_hash(entr 2516 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2417 2517
2418 hlist_bl_lock(b); 2518 hlist_bl_lock(b);
2419 hlist_bl_add_head_rcu(&entry->d_hash, 2519 hlist_bl_add_head_rcu(&entry->d_hash, b);
2420 hlist_bl_unlock(b); 2520 hlist_bl_unlock(b);
2421 } 2521 }
2422 2522
2423 /** 2523 /**
2424 * d_rehash - add an entry back to the ha 2524 * d_rehash - add an entry back to the hash
2425 * @entry: dentry to add to the hash 2525 * @entry: dentry to add to the hash
2426 * 2526 *
2427 * Adds a dentry to the hash according to its 2527 * Adds a dentry to the hash according to its name.
2428 */ 2528 */
2429 2529
2430 void d_rehash(struct dentry * entry) 2530 void d_rehash(struct dentry * entry)
2431 { 2531 {
2432 spin_lock(&entry->d_lock); 2532 spin_lock(&entry->d_lock);
2433 __d_rehash(entry); 2533 __d_rehash(entry);
2434 spin_unlock(&entry->d_lock); 2534 spin_unlock(&entry->d_lock);
2435 } 2535 }
2436 EXPORT_SYMBOL(d_rehash); 2536 EXPORT_SYMBOL(d_rehash);
2437 2537
2438 static inline unsigned start_dir_add(struct i 2538 static inline unsigned start_dir_add(struct inode *dir)
2439 { 2539 {
2440 preempt_disable_nested(); !! 2540
2441 for (;;) { 2541 for (;;) {
2442 unsigned n = dir->i_dir_seq; 2542 unsigned n = dir->i_dir_seq;
2443 if (!(n & 1) && cmpxchg(&dir- 2543 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2444 return n; 2544 return n;
2445 cpu_relax(); 2545 cpu_relax();
2446 } 2546 }
2447 } 2547 }
2448 2548
2449 static inline void end_dir_add(struct inode * !! 2549 static inline void end_dir_add(struct inode *dir, unsigned n)
2450 wait_queue_hea <<
2451 { 2550 {
2452 smp_store_release(&dir->i_dir_seq, n 2551 smp_store_release(&dir->i_dir_seq, n + 2);
2453 preempt_enable_nested(); <<
2454 wake_up_all(d_wait); <<
2455 } 2552 }
2456 2553
2457 static void d_wait_lookup(struct dentry *dent 2554 static void d_wait_lookup(struct dentry *dentry)
2458 { 2555 {
2459 if (d_in_lookup(dentry)) { 2556 if (d_in_lookup(dentry)) {
2460 DECLARE_WAITQUEUE(wait, curre 2557 DECLARE_WAITQUEUE(wait, current);
2461 add_wait_queue(dentry->d_wait 2558 add_wait_queue(dentry->d_wait, &wait);
2462 do { 2559 do {
2463 set_current_state(TAS 2560 set_current_state(TASK_UNINTERRUPTIBLE);
2464 spin_unlock(&dentry-> 2561 spin_unlock(&dentry->d_lock);
2465 schedule(); 2562 schedule();
2466 spin_lock(&dentry->d_ 2563 spin_lock(&dentry->d_lock);
2467 } while (d_in_lookup(dentry)) 2564 } while (d_in_lookup(dentry));
2468 } 2565 }
2469 } 2566 }
2470 2567
2471 struct dentry *d_alloc_parallel(struct dentry 2568 struct dentry *d_alloc_parallel(struct dentry *parent,
2472 const struct 2569 const struct qstr *name,
2473 wait_queue_he 2570 wait_queue_head_t *wq)
2474 { 2571 {
2475 unsigned int hash = name->hash; 2572 unsigned int hash = name->hash;
2476 struct hlist_bl_head *b = in_lookup_h 2573 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2477 struct hlist_bl_node *node; 2574 struct hlist_bl_node *node;
2478 struct dentry *new = d_alloc(parent, 2575 struct dentry *new = d_alloc(parent, name);
2479 struct dentry *dentry; 2576 struct dentry *dentry;
2480 unsigned seq, r_seq, d_seq; 2577 unsigned seq, r_seq, d_seq;
2481 2578
2482 if (unlikely(!new)) 2579 if (unlikely(!new))
2483 return ERR_PTR(-ENOMEM); 2580 return ERR_PTR(-ENOMEM);
2484 2581
2485 retry: 2582 retry:
2486 rcu_read_lock(); 2583 rcu_read_lock();
2487 seq = smp_load_acquire(&parent->d_ino 2584 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2488 r_seq = read_seqbegin(&rename_lock); 2585 r_seq = read_seqbegin(&rename_lock);
2489 dentry = __d_lookup_rcu(parent, name, 2586 dentry = __d_lookup_rcu(parent, name, &d_seq);
2490 if (unlikely(dentry)) { 2587 if (unlikely(dentry)) {
2491 if (!lockref_get_not_dead(&de 2588 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2492 rcu_read_unlock(); 2589 rcu_read_unlock();
2493 goto retry; 2590 goto retry;
2494 } 2591 }
2495 if (read_seqcount_retry(&dent 2592 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2496 rcu_read_unlock(); 2593 rcu_read_unlock();
2497 dput(dentry); 2594 dput(dentry);
2498 goto retry; 2595 goto retry;
2499 } 2596 }
2500 rcu_read_unlock(); 2597 rcu_read_unlock();
2501 dput(new); 2598 dput(new);
2502 return dentry; 2599 return dentry;
2503 } 2600 }
2504 if (unlikely(read_seqretry(&rename_lo 2601 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2505 rcu_read_unlock(); 2602 rcu_read_unlock();
2506 goto retry; 2603 goto retry;
2507 } 2604 }
2508 2605
2509 if (unlikely(seq & 1)) { 2606 if (unlikely(seq & 1)) {
2510 rcu_read_unlock(); 2607 rcu_read_unlock();
2511 goto retry; 2608 goto retry;
2512 } 2609 }
2513 2610
2514 hlist_bl_lock(b); 2611 hlist_bl_lock(b);
2515 if (unlikely(READ_ONCE(parent->d_inod 2612 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2516 hlist_bl_unlock(b); 2613 hlist_bl_unlock(b);
2517 rcu_read_unlock(); 2614 rcu_read_unlock();
2518 goto retry; 2615 goto retry;
2519 } 2616 }
2520 /* 2617 /*
2521 * No changes for the parent since th 2618 * No changes for the parent since the beginning of d_lookup().
2522 * Since all removals from the chain 2619 * Since all removals from the chain happen with hlist_bl_lock(),
2523 * any potential in-lookup matches ar 2620 * any potential in-lookup matches are going to stay here until
2524 * we unlock the chain. All fields a 2621 * we unlock the chain. All fields are stable in everything
2525 * we encounter. 2622 * we encounter.
2526 */ 2623 */
2527 hlist_bl_for_each_entry(dentry, node, 2624 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2528 if (dentry->d_name.hash != ha 2625 if (dentry->d_name.hash != hash)
2529 continue; 2626 continue;
2530 if (dentry->d_parent != paren 2627 if (dentry->d_parent != parent)
2531 continue; 2628 continue;
2532 if (!d_same_name(dentry, pare 2629 if (!d_same_name(dentry, parent, name))
2533 continue; 2630 continue;
2534 hlist_bl_unlock(b); 2631 hlist_bl_unlock(b);
2535 /* now we can try to grab a r 2632 /* now we can try to grab a reference */
2536 if (!lockref_get_not_dead(&de 2633 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2537 rcu_read_unlock(); 2634 rcu_read_unlock();
2538 goto retry; 2635 goto retry;
2539 } 2636 }
2540 2637
2541 rcu_read_unlock(); 2638 rcu_read_unlock();
2542 /* 2639 /*
2543 * somebody is likely to be s 2640 * somebody is likely to be still doing lookup for it;
2544 * wait for them to finish 2641 * wait for them to finish
2545 */ 2642 */
2546 spin_lock(&dentry->d_lock); 2643 spin_lock(&dentry->d_lock);
2547 d_wait_lookup(dentry); 2644 d_wait_lookup(dentry);
2548 /* 2645 /*
2549 * it's not in-lookup anymore 2646 * it's not in-lookup anymore; in principle we should repeat
2550 * everything from dcache loo 2647 * everything from dcache lookup, but it's likely to be what
2551 * d_lookup() would've found 2648 * d_lookup() would've found anyway. If it is, just return it;
2552 * otherwise we really have t 2649 * otherwise we really have to repeat the whole thing.
2553 */ 2650 */
2554 if (unlikely(dentry->d_name.h 2651 if (unlikely(dentry->d_name.hash != hash))
2555 goto mismatch; 2652 goto mismatch;
2556 if (unlikely(dentry->d_parent 2653 if (unlikely(dentry->d_parent != parent))
2557 goto mismatch; 2654 goto mismatch;
2558 if (unlikely(d_unhashed(dentr 2655 if (unlikely(d_unhashed(dentry)))
2559 goto mismatch; 2656 goto mismatch;
2560 if (unlikely(!d_same_name(den 2657 if (unlikely(!d_same_name(dentry, parent, name)))
2561 goto mismatch; 2658 goto mismatch;
2562 /* OK, it *is* a hashed match 2659 /* OK, it *is* a hashed match; return it */
2563 spin_unlock(&dentry->d_lock); 2660 spin_unlock(&dentry->d_lock);
2564 dput(new); 2661 dput(new);
2565 return dentry; 2662 return dentry;
2566 } 2663 }
2567 rcu_read_unlock(); 2664 rcu_read_unlock();
2568 /* we can't take ->d_lock here; it's 2665 /* we can't take ->d_lock here; it's OK, though. */
2569 new->d_flags |= DCACHE_PAR_LOOKUP; 2666 new->d_flags |= DCACHE_PAR_LOOKUP;
2570 new->d_wait = wq; 2667 new->d_wait = wq;
2571 hlist_bl_add_head(&new->d_u.d_in_look !! 2668 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2572 hlist_bl_unlock(b); 2669 hlist_bl_unlock(b);
2573 return new; 2670 return new;
2574 mismatch: 2671 mismatch:
2575 spin_unlock(&dentry->d_lock); 2672 spin_unlock(&dentry->d_lock);
2576 dput(dentry); 2673 dput(dentry);
2577 goto retry; 2674 goto retry;
2578 } 2675 }
2579 EXPORT_SYMBOL(d_alloc_parallel); 2676 EXPORT_SYMBOL(d_alloc_parallel);
2580 2677
2581 /* !! 2678 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 { 2679 {
2588 wait_queue_head_t *d_wait; !! 2680 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2589 struct hlist_bl_head *b; !! 2681 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); 2682 hlist_bl_lock(b);
2595 dentry->d_flags &= ~DCACHE_PAR_LOOKUP 2683 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2596 __hlist_bl_del(&dentry->d_u.d_in_look 2684 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2597 d_wait = dentry->d_wait; !! 2685 wake_up_all(dentry->d_wait);
2598 dentry->d_wait = NULL; 2686 dentry->d_wait = NULL;
2599 hlist_bl_unlock(b); 2687 hlist_bl_unlock(b);
2600 INIT_HLIST_NODE(&dentry->d_u.d_alias) 2688 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2601 INIT_LIST_HEAD(&dentry->d_lru); 2689 INIT_LIST_HEAD(&dentry->d_lru);
2602 return d_wait; <<
2603 } 2690 }
2604 !! 2691 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 2692
2613 /* inode->i_lock held if inode is non-NULL */ 2693 /* inode->i_lock held if inode is non-NULL */
2614 2694
2615 static inline void __d_add(struct dentry *den 2695 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2616 { 2696 {
2617 wait_queue_head_t *d_wait; <<
2618 struct inode *dir = NULL; 2697 struct inode *dir = NULL;
2619 unsigned n; 2698 unsigned n;
2620 spin_lock(&dentry->d_lock); 2699 spin_lock(&dentry->d_lock);
2621 if (unlikely(d_in_lookup(dentry))) { 2700 if (unlikely(d_in_lookup(dentry))) {
2622 dir = dentry->d_parent->d_ino 2701 dir = dentry->d_parent->d_inode;
2623 n = start_dir_add(dir); 2702 n = start_dir_add(dir);
2624 d_wait = __d_lookup_unhash(de !! 2703 __d_lookup_done(dentry);
2625 } 2704 }
2626 if (inode) { 2705 if (inode) {
2627 unsigned add_flags = d_flags_ 2706 unsigned add_flags = d_flags_for_inode(inode);
2628 hlist_add_head(&dentry->d_u.d 2707 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2629 raw_write_seqcount_begin(&den 2708 raw_write_seqcount_begin(&dentry->d_seq);
2630 __d_set_inode_and_type(dentry 2709 __d_set_inode_and_type(dentry, inode, add_flags);
2631 raw_write_seqcount_end(&dentr 2710 raw_write_seqcount_end(&dentry->d_seq);
2632 fsnotify_update_flags(dentry) 2711 fsnotify_update_flags(dentry);
2633 } 2712 }
2634 __d_rehash(dentry); 2713 __d_rehash(dentry);
2635 if (dir) 2714 if (dir)
2636 end_dir_add(dir, n, d_wait); !! 2715 end_dir_add(dir, n);
2637 spin_unlock(&dentry->d_lock); 2716 spin_unlock(&dentry->d_lock);
2638 if (inode) 2717 if (inode)
2639 spin_unlock(&inode->i_lock); 2718 spin_unlock(&inode->i_lock);
2640 } 2719 }
2641 2720
2642 /** 2721 /**
2643 * d_add - add dentry to hash queues 2722 * d_add - add dentry to hash queues
2644 * @entry: dentry to add 2723 * @entry: dentry to add
2645 * @inode: The inode to attach to this dentry 2724 * @inode: The inode to attach to this dentry
2646 * 2725 *
2647 * This adds the entry to the hash queues and 2726 * This adds the entry to the hash queues and initializes @inode.
2648 * The entry was actually filled in earlier d 2727 * The entry was actually filled in earlier during d_alloc().
2649 */ 2728 */
2650 2729
2651 void d_add(struct dentry *entry, struct inode 2730 void d_add(struct dentry *entry, struct inode *inode)
2652 { 2731 {
2653 if (inode) { 2732 if (inode) {
2654 security_d_instantiate(entry, 2733 security_d_instantiate(entry, inode);
2655 spin_lock(&inode->i_lock); 2734 spin_lock(&inode->i_lock);
2656 } 2735 }
2657 __d_add(entry, inode); 2736 __d_add(entry, inode);
2658 } 2737 }
2659 EXPORT_SYMBOL(d_add); 2738 EXPORT_SYMBOL(d_add);
2660 2739
2661 /** 2740 /**
2662 * d_exact_alias - find and hash an exact unh 2741 * d_exact_alias - find and hash an exact unhashed alias
2663 * @entry: dentry to add 2742 * @entry: dentry to add
2664 * @inode: The inode to go with this dentry 2743 * @inode: The inode to go with this dentry
2665 * 2744 *
2666 * If an unhashed dentry with the same name/p 2745 * If an unhashed dentry with the same name/parent and desired
2667 * inode already exists, hash and return it. 2746 * inode already exists, hash and return it. Otherwise, return
2668 * NULL. 2747 * NULL.
2669 * 2748 *
2670 * Parent directory should be locked. 2749 * Parent directory should be locked.
2671 */ 2750 */
2672 struct dentry *d_exact_alias(struct dentry *e 2751 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2673 { 2752 {
2674 struct dentry *alias; 2753 struct dentry *alias;
2675 unsigned int hash = entry->d_name.has 2754 unsigned int hash = entry->d_name.hash;
2676 2755
2677 spin_lock(&inode->i_lock); 2756 spin_lock(&inode->i_lock);
2678 hlist_for_each_entry(alias, &inode->i 2757 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2679 /* 2758 /*
2680 * Don't need alias->d_lock h 2759 * Don't need alias->d_lock here, because aliases with
2681 * d_parent == entry->d_paren 2760 * d_parent == entry->d_parent are not subject to name or
2682 * parent changes, because th 2761 * parent changes, because the parent inode i_mutex is held.
2683 */ 2762 */
2684 if (alias->d_name.hash != has 2763 if (alias->d_name.hash != hash)
2685 continue; 2764 continue;
2686 if (alias->d_parent != entry- 2765 if (alias->d_parent != entry->d_parent)
2687 continue; 2766 continue;
2688 if (!d_same_name(alias, entry 2767 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2689 continue; 2768 continue;
2690 spin_lock(&alias->d_lock); 2769 spin_lock(&alias->d_lock);
2691 if (!d_unhashed(alias)) { 2770 if (!d_unhashed(alias)) {
2692 spin_unlock(&alias->d 2771 spin_unlock(&alias->d_lock);
2693 alias = NULL; 2772 alias = NULL;
2694 } else { 2773 } else {
2695 dget_dlock(alias); !! 2774 __dget_dlock(alias);
2696 __d_rehash(alias); 2775 __d_rehash(alias);
2697 spin_unlock(&alias->d 2776 spin_unlock(&alias->d_lock);
2698 } 2777 }
2699 spin_unlock(&inode->i_lock); 2778 spin_unlock(&inode->i_lock);
2700 return alias; 2779 return alias;
2701 } 2780 }
2702 spin_unlock(&inode->i_lock); 2781 spin_unlock(&inode->i_lock);
2703 return NULL; 2782 return NULL;
2704 } 2783 }
2705 EXPORT_SYMBOL(d_exact_alias); 2784 EXPORT_SYMBOL(d_exact_alias);
2706 2785
2707 static void swap_names(struct dentry *dentry, 2786 static void swap_names(struct dentry *dentry, struct dentry *target)
2708 { 2787 {
2709 if (unlikely(dname_external(target))) 2788 if (unlikely(dname_external(target))) {
2710 if (unlikely(dname_external(d 2789 if (unlikely(dname_external(dentry))) {
2711 /* 2790 /*
2712 * Both external: swa 2791 * Both external: swap the pointers
2713 */ 2792 */
2714 swap(target->d_name.n 2793 swap(target->d_name.name, dentry->d_name.name);
2715 } else { 2794 } else {
2716 /* 2795 /*
2717 * dentry:internal, t 2796 * dentry:internal, target:external. Steal target's
2718 * storage and make t 2797 * storage and make target internal.
2719 */ 2798 */
2720 memcpy(target->d_inam 2799 memcpy(target->d_iname, dentry->d_name.name,
2721 dentr 2800 dentry->d_name.len + 1);
2722 dentry->d_name.name = 2801 dentry->d_name.name = target->d_name.name;
2723 target->d_name.name = 2802 target->d_name.name = target->d_iname;
2724 } 2803 }
2725 } else { 2804 } else {
2726 if (unlikely(dname_external(d 2805 if (unlikely(dname_external(dentry))) {
2727 /* 2806 /*
2728 * dentry:external, t 2807 * dentry:external, target:internal. Give dentry's
2729 * storage to target 2808 * storage to target and make dentry internal
2730 */ 2809 */
2731 memcpy(dentry->d_inam 2810 memcpy(dentry->d_iname, target->d_name.name,
2732 targe 2811 target->d_name.len + 1);
2733 target->d_name.name = 2812 target->d_name.name = dentry->d_name.name;
2734 dentry->d_name.name = 2813 dentry->d_name.name = dentry->d_iname;
2735 } else { 2814 } else {
2736 /* 2815 /*
2737 * Both are internal. 2816 * Both are internal.
2738 */ 2817 */
2739 unsigned int i; 2818 unsigned int i;
2740 BUILD_BUG_ON(!IS_ALIG 2819 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2741 for (i = 0; i < DNAME 2820 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2742 swap(((long * 2821 swap(((long *) &dentry->d_iname)[i],
2743 ((long * 2822 ((long *) &target->d_iname)[i]);
2744 } 2823 }
2745 } 2824 }
2746 } 2825 }
2747 swap(dentry->d_name.hash_len, target- 2826 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2748 } 2827 }
2749 2828
2750 static void copy_name(struct dentry *dentry, 2829 static void copy_name(struct dentry *dentry, struct dentry *target)
2751 { 2830 {
2752 struct external_name *old_name = NULL 2831 struct external_name *old_name = NULL;
2753 if (unlikely(dname_external(dentry))) 2832 if (unlikely(dname_external(dentry)))
2754 old_name = external_name(dent 2833 old_name = external_name(dentry);
2755 if (unlikely(dname_external(target))) 2834 if (unlikely(dname_external(target))) {
2756 atomic_inc(&external_name(tar 2835 atomic_inc(&external_name(target)->u.count);
2757 dentry->d_name = target->d_na 2836 dentry->d_name = target->d_name;
2758 } else { 2837 } else {
2759 memcpy(dentry->d_iname, targe 2838 memcpy(dentry->d_iname, target->d_name.name,
2760 target->d_nam 2839 target->d_name.len + 1);
2761 dentry->d_name.name = dentry- 2840 dentry->d_name.name = dentry->d_iname;
2762 dentry->d_name.hash_len = tar 2841 dentry->d_name.hash_len = target->d_name.hash_len;
2763 } 2842 }
2764 if (old_name && likely(atomic_dec_and 2843 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2765 kfree_rcu(old_name, u.head); 2844 kfree_rcu(old_name, u.head);
2766 } 2845 }
2767 2846
2768 /* 2847 /*
2769 * __d_move - move a dentry 2848 * __d_move - move a dentry
2770 * @dentry: entry to move 2849 * @dentry: entry to move
2771 * @target: new dentry 2850 * @target: new dentry
2772 * @exchange: exchange the two dentries 2851 * @exchange: exchange the two dentries
2773 * 2852 *
2774 * Update the dcache to reflect the move of a 2853 * Update the dcache to reflect the move of a file name. Negative
2775 * dcache entries should not be moved in this 2854 * dcache entries should not be moved in this way. Caller must hold
2776 * rename_lock, the i_mutex of the source and 2855 * rename_lock, the i_mutex of the source and target directories,
2777 * and the sb->s_vfs_rename_mutex if they dif 2856 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2778 */ 2857 */
2779 static void __d_move(struct dentry *dentry, s 2858 static void __d_move(struct dentry *dentry, struct dentry *target,
2780 bool exchange) 2859 bool exchange)
2781 { 2860 {
2782 struct dentry *old_parent, *p; 2861 struct dentry *old_parent, *p;
2783 wait_queue_head_t *d_wait; <<
2784 struct inode *dir = NULL; 2862 struct inode *dir = NULL;
2785 unsigned n; 2863 unsigned n;
2786 2864
2787 WARN_ON(!dentry->d_inode); 2865 WARN_ON(!dentry->d_inode);
2788 if (WARN_ON(dentry == target)) 2866 if (WARN_ON(dentry == target))
2789 return; 2867 return;
2790 2868
2791 BUG_ON(d_ancestor(target, dentry)); 2869 BUG_ON(d_ancestor(target, dentry));
2792 old_parent = dentry->d_parent; 2870 old_parent = dentry->d_parent;
2793 p = d_ancestor(old_parent, target); 2871 p = d_ancestor(old_parent, target);
2794 if (IS_ROOT(dentry)) { 2872 if (IS_ROOT(dentry)) {
2795 BUG_ON(p); 2873 BUG_ON(p);
2796 spin_lock(&target->d_parent-> 2874 spin_lock(&target->d_parent->d_lock);
2797 } else if (!p) { 2875 } else if (!p) {
2798 /* target is not a descendent 2876 /* target is not a descendent of dentry->d_parent */
2799 spin_lock(&target->d_parent-> 2877 spin_lock(&target->d_parent->d_lock);
2800 spin_lock_nested(&old_parent- 2878 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2801 } else { 2879 } else {
2802 BUG_ON(p == dentry); 2880 BUG_ON(p == dentry);
2803 spin_lock(&old_parent->d_lock 2881 spin_lock(&old_parent->d_lock);
2804 if (p != target) 2882 if (p != target)
2805 spin_lock_nested(&tar 2883 spin_lock_nested(&target->d_parent->d_lock,
2806 DENTR 2884 DENTRY_D_LOCK_NESTED);
2807 } 2885 }
2808 spin_lock_nested(&dentry->d_lock, 2); 2886 spin_lock_nested(&dentry->d_lock, 2);
2809 spin_lock_nested(&target->d_lock, 3); 2887 spin_lock_nested(&target->d_lock, 3);
2810 2888
2811 if (unlikely(d_in_lookup(target))) { 2889 if (unlikely(d_in_lookup(target))) {
2812 dir = target->d_parent->d_ino 2890 dir = target->d_parent->d_inode;
2813 n = start_dir_add(dir); 2891 n = start_dir_add(dir);
2814 d_wait = __d_lookup_unhash(ta !! 2892 __d_lookup_done(target);
2815 } 2893 }
2816 2894
2817 write_seqcount_begin(&dentry->d_seq); 2895 write_seqcount_begin(&dentry->d_seq);
2818 write_seqcount_begin_nested(&target-> 2896 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2819 2897
2820 /* unhash both */ 2898 /* unhash both */
2821 if (!d_unhashed(dentry)) 2899 if (!d_unhashed(dentry))
2822 ___d_drop(dentry); 2900 ___d_drop(dentry);
2823 if (!d_unhashed(target)) 2901 if (!d_unhashed(target))
2824 ___d_drop(target); 2902 ___d_drop(target);
2825 2903
2826 /* ... and switch them in the tree */ 2904 /* ... and switch them in the tree */
2827 dentry->d_parent = target->d_parent; 2905 dentry->d_parent = target->d_parent;
2828 if (!exchange) { 2906 if (!exchange) {
2829 copy_name(dentry, target); 2907 copy_name(dentry, target);
2830 target->d_hash.pprev = NULL; 2908 target->d_hash.pprev = NULL;
2831 dentry->d_parent->d_lockref.c 2909 dentry->d_parent->d_lockref.count++;
2832 if (dentry != old_parent) /* 2910 if (dentry != old_parent) /* wasn't IS_ROOT */
2833 WARN_ON(!--old_parent 2911 WARN_ON(!--old_parent->d_lockref.count);
2834 } else { 2912 } else {
2835 target->d_parent = old_parent 2913 target->d_parent = old_parent;
2836 swap_names(dentry, target); 2914 swap_names(dentry, target);
2837 if (!hlist_unhashed(&target-> !! 2915 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); 2916 __d_rehash(target);
2841 fsnotify_update_flags(target) 2917 fsnotify_update_flags(target);
2842 } 2918 }
2843 if (!hlist_unhashed(&dentry->d_sib)) !! 2919 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); 2920 __d_rehash(dentry);
2847 fsnotify_update_flags(dentry); 2921 fsnotify_update_flags(dentry);
2848 fscrypt_handle_d_move(dentry); 2922 fscrypt_handle_d_move(dentry);
2849 2923
2850 write_seqcount_end(&target->d_seq); 2924 write_seqcount_end(&target->d_seq);
2851 write_seqcount_end(&dentry->d_seq); 2925 write_seqcount_end(&dentry->d_seq);
2852 2926
2853 if (dir) 2927 if (dir)
2854 end_dir_add(dir, n, d_wait); !! 2928 end_dir_add(dir, n);
2855 2929
2856 if (dentry->d_parent != old_parent) 2930 if (dentry->d_parent != old_parent)
2857 spin_unlock(&dentry->d_parent 2931 spin_unlock(&dentry->d_parent->d_lock);
2858 if (dentry != old_parent) 2932 if (dentry != old_parent)
2859 spin_unlock(&old_parent->d_lo 2933 spin_unlock(&old_parent->d_lock);
2860 spin_unlock(&target->d_lock); 2934 spin_unlock(&target->d_lock);
2861 spin_unlock(&dentry->d_lock); 2935 spin_unlock(&dentry->d_lock);
2862 } 2936 }
2863 2937
2864 /* 2938 /*
2865 * d_move - move a dentry 2939 * d_move - move a dentry
2866 * @dentry: entry to move 2940 * @dentry: entry to move
2867 * @target: new dentry 2941 * @target: new dentry
2868 * 2942 *
2869 * Update the dcache to reflect the move of a 2943 * Update the dcache to reflect the move of a file name. Negative
2870 * dcache entries should not be moved in this 2944 * dcache entries should not be moved in this way. See the locking
2871 * requirements for __d_move. 2945 * requirements for __d_move.
2872 */ 2946 */
2873 void d_move(struct dentry *dentry, struct den 2947 void d_move(struct dentry *dentry, struct dentry *target)
2874 { 2948 {
2875 write_seqlock(&rename_lock); 2949 write_seqlock(&rename_lock);
2876 __d_move(dentry, target, false); 2950 __d_move(dentry, target, false);
2877 write_sequnlock(&rename_lock); 2951 write_sequnlock(&rename_lock);
2878 } 2952 }
2879 EXPORT_SYMBOL(d_move); 2953 EXPORT_SYMBOL(d_move);
2880 2954
2881 /* 2955 /*
2882 * d_exchange - exchange two dentries 2956 * d_exchange - exchange two dentries
2883 * @dentry1: first dentry 2957 * @dentry1: first dentry
2884 * @dentry2: second dentry 2958 * @dentry2: second dentry
2885 */ 2959 */
2886 void d_exchange(struct dentry *dentry1, struc 2960 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2887 { 2961 {
2888 write_seqlock(&rename_lock); 2962 write_seqlock(&rename_lock);
2889 2963
2890 WARN_ON(!dentry1->d_inode); 2964 WARN_ON(!dentry1->d_inode);
2891 WARN_ON(!dentry2->d_inode); 2965 WARN_ON(!dentry2->d_inode);
2892 WARN_ON(IS_ROOT(dentry1)); 2966 WARN_ON(IS_ROOT(dentry1));
2893 WARN_ON(IS_ROOT(dentry2)); 2967 WARN_ON(IS_ROOT(dentry2));
2894 2968
2895 __d_move(dentry1, dentry2, true); 2969 __d_move(dentry1, dentry2, true);
2896 2970
2897 write_sequnlock(&rename_lock); 2971 write_sequnlock(&rename_lock);
2898 } 2972 }
2899 2973
2900 /** 2974 /**
2901 * d_ancestor - search for an ancestor 2975 * d_ancestor - search for an ancestor
2902 * @p1: ancestor dentry 2976 * @p1: ancestor dentry
2903 * @p2: child dentry 2977 * @p2: child dentry
2904 * 2978 *
2905 * Returns the ancestor dentry of p2 which is 2979 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2906 * an ancestor of p2, else NULL. 2980 * an ancestor of p2, else NULL.
2907 */ 2981 */
2908 struct dentry *d_ancestor(struct dentry *p1, 2982 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2909 { 2983 {
2910 struct dentry *p; 2984 struct dentry *p;
2911 2985
2912 for (p = p2; !IS_ROOT(p); p = p->d_pa 2986 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2913 if (p->d_parent == p1) 2987 if (p->d_parent == p1)
2914 return p; 2988 return p;
2915 } 2989 }
2916 return NULL; 2990 return NULL;
2917 } 2991 }
2918 2992
2919 /* 2993 /*
2920 * This helper attempts to cope with remotely 2994 * This helper attempts to cope with remotely renamed directories
2921 * 2995 *
2922 * It assumes that the caller is already hold 2996 * It assumes that the caller is already holding
2923 * dentry->d_parent->d_inode->i_mutex, and re 2997 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2924 * 2998 *
2925 * Note: If ever the locking in lock_rename() 2999 * Note: If ever the locking in lock_rename() changes, then please
2926 * remember to update this too... 3000 * remember to update this too...
2927 */ 3001 */
2928 static int __d_unalias(struct dentry *dentry, !! 3002 static int __d_unalias(struct inode *inode,
>> 3003 struct dentry *dentry, struct dentry *alias)
2929 { 3004 {
2930 struct mutex *m1 = NULL; 3005 struct mutex *m1 = NULL;
2931 struct rw_semaphore *m2 = NULL; 3006 struct rw_semaphore *m2 = NULL;
2932 int ret = -ESTALE; 3007 int ret = -ESTALE;
2933 3008
2934 /* If alias and dentry share a parent 3009 /* If alias and dentry share a parent, then no extra locks required */
2935 if (alias->d_parent == dentry->d_pare 3010 if (alias->d_parent == dentry->d_parent)
2936 goto out_unalias; 3011 goto out_unalias;
2937 3012
2938 /* See lock_rename() */ 3013 /* See lock_rename() */
2939 if (!mutex_trylock(&dentry->d_sb->s_v 3014 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2940 goto out_err; 3015 goto out_err;
2941 m1 = &dentry->d_sb->s_vfs_rename_mute 3016 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2942 if (!inode_trylock_shared(alias->d_pa 3017 if (!inode_trylock_shared(alias->d_parent->d_inode))
2943 goto out_err; 3018 goto out_err;
2944 m2 = &alias->d_parent->d_inode->i_rws 3019 m2 = &alias->d_parent->d_inode->i_rwsem;
2945 out_unalias: 3020 out_unalias:
2946 __d_move(alias, dentry, false); 3021 __d_move(alias, dentry, false);
2947 ret = 0; 3022 ret = 0;
2948 out_err: 3023 out_err:
2949 if (m2) 3024 if (m2)
2950 up_read(m2); 3025 up_read(m2);
2951 if (m1) 3026 if (m1)
2952 mutex_unlock(m1); 3027 mutex_unlock(m1);
2953 return ret; 3028 return ret;
2954 } 3029 }
2955 3030
2956 /** 3031 /**
2957 * d_splice_alias - splice a disconnected den 3032 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2958 * @inode: the inode which may have a discon 3033 * @inode: the inode which may have a disconnected dentry
2959 * @dentry: a negative dentry which we want t 3034 * @dentry: a negative dentry which we want to point to the inode.
2960 * 3035 *
2961 * If inode is a directory and has an IS_ROOT 3036 * 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 3037 * place of the given dentry and return it, else simply d_add the inode
2963 * to the dentry and return NULL. 3038 * to the dentry and return NULL.
2964 * 3039 *
2965 * If a non-IS_ROOT directory is found, the f 3040 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2966 * we should error out: directories can't hav 3041 * we should error out: directories can't have multiple aliases.
2967 * 3042 *
2968 * This is needed in the lookup routine of an 3043 * This is needed in the lookup routine of any filesystem that is exportable
2969 * (via knfsd) so that we can build dcache pa 3044 * (via knfsd) so that we can build dcache paths to directories effectively.
2970 * 3045 *
2971 * If a dentry was found and moved, then it i 3046 * If a dentry was found and moved, then it is returned. Otherwise NULL
2972 * is returned. This matches the expected re 3047 * is returned. This matches the expected return value of ->lookup.
2973 * 3048 *
2974 * Cluster filesystems may call this function 3049 * Cluster filesystems may call this function with a negative, hashed dentry.
2975 * In that case, we know that the inode will 3050 * 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 3051 * will only occur during atomic_open. So we need to check for the dentry
2977 * being already hashed only in the final cas 3052 * being already hashed only in the final case.
2978 */ 3053 */
2979 struct dentry *d_splice_alias(struct inode *i 3054 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2980 { 3055 {
2981 if (IS_ERR(inode)) 3056 if (IS_ERR(inode))
2982 return ERR_CAST(inode); 3057 return ERR_CAST(inode);
2983 3058
2984 BUG_ON(!d_unhashed(dentry)); 3059 BUG_ON(!d_unhashed(dentry));
2985 3060
2986 if (!inode) 3061 if (!inode)
2987 goto out; 3062 goto out;
2988 3063
2989 security_d_instantiate(dentry, inode) 3064 security_d_instantiate(dentry, inode);
2990 spin_lock(&inode->i_lock); 3065 spin_lock(&inode->i_lock);
2991 if (S_ISDIR(inode->i_mode)) { 3066 if (S_ISDIR(inode->i_mode)) {
2992 struct dentry *new = __d_find 3067 struct dentry *new = __d_find_any_alias(inode);
2993 if (unlikely(new)) { 3068 if (unlikely(new)) {
2994 /* The reference to n 3069 /* The reference to new ensures it remains an alias */
2995 spin_unlock(&inode->i 3070 spin_unlock(&inode->i_lock);
2996 write_seqlock(&rename 3071 write_seqlock(&rename_lock);
2997 if (unlikely(d_ancest 3072 if (unlikely(d_ancestor(new, dentry))) {
2998 write_sequnlo 3073 write_sequnlock(&rename_lock);
2999 dput(new); 3074 dput(new);
3000 new = ERR_PTR 3075 new = ERR_PTR(-ELOOP);
3001 pr_warn_ratel 3076 pr_warn_ratelimited(
3002 "VFS: 3077 "VFS: Lookup of '%s' in %s %s"
3003 " wou 3078 " would have caused loop\n",
3004 dentr 3079 dentry->d_name.name,
3005 inode 3080 inode->i_sb->s_type->name,
3006 inode 3081 inode->i_sb->s_id);
3007 } else if (!IS_ROOT(n 3082 } else if (!IS_ROOT(new)) {
3008 struct dentry 3083 struct dentry *old_parent = dget(new->d_parent);
3009 int err = __d !! 3084 int err = __d_unalias(inode, dentry, new);
3010 write_sequnlo 3085 write_sequnlock(&rename_lock);
3011 if (err) { 3086 if (err) {
3012 dput( 3087 dput(new);
3013 new = 3088 new = ERR_PTR(err);
3014 } 3089 }
3015 dput(old_pare 3090 dput(old_parent);
3016 } else { 3091 } else {
3017 __d_move(new, 3092 __d_move(new, dentry, false);
3018 write_sequnlo 3093 write_sequnlock(&rename_lock);
3019 } 3094 }
3020 iput(inode); 3095 iput(inode);
3021 return new; 3096 return new;
3022 } 3097 }
3023 } 3098 }
3024 out: 3099 out:
3025 __d_add(dentry, inode); 3100 __d_add(dentry, inode);
3026 return NULL; 3101 return NULL;
3027 } 3102 }
3028 EXPORT_SYMBOL(d_splice_alias); 3103 EXPORT_SYMBOL(d_splice_alias);
3029 3104
3030 /* 3105 /*
3031 * Test whether new_dentry is a subdirectory 3106 * Test whether new_dentry is a subdirectory of old_dentry.
3032 * 3107 *
3033 * Trivially implemented using the dcache str 3108 * Trivially implemented using the dcache structure
3034 */ 3109 */
3035 3110
3036 /** 3111 /**
3037 * is_subdir - is new dentry a subdirectory o 3112 * is_subdir - is new dentry a subdirectory of old_dentry
3038 * @new_dentry: new dentry 3113 * @new_dentry: new dentry
3039 * @old_dentry: old dentry 3114 * @old_dentry: old dentry
3040 * 3115 *
3041 * Returns true if new_dentry is a subdirecto 3116 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3042 * Returns false otherwise. 3117 * Returns false otherwise.
3043 * Caller must ensure that "new_dentry" is pi 3118 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3044 */ 3119 */
3045 3120
3046 bool is_subdir(struct dentry *new_dentry, str 3121 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3047 { 3122 {
3048 bool subdir; !! 3123 bool result;
3049 unsigned seq; 3124 unsigned seq;
3050 3125
3051 if (new_dentry == old_dentry) 3126 if (new_dentry == old_dentry)
3052 return true; 3127 return true;
3053 3128
3054 /* Access d_parent under rcu as d_mov !! 3129 do {
3055 rcu_read_lock(); !! 3130 /* for restarting inner loop in case of seq retry */
3056 seq = read_seqbegin(&rename_lock); !! 3131 seq = read_seqbegin(&rename_lock);
3057 subdir = d_ancestor(old_dentry, new_d !! 3132 /*
3058 /* Try lockless once... */ !! 3133 * Need rcu_readlock to protect against the d_parent trashing
3059 if (read_seqretry(&rename_lock, seq)) !! 3134 * due to d_move
3060 /* ...else acquire lock for p !! 3135 */
3061 read_seqlock_excl(&rename_loc !! 3136 rcu_read_lock();
3062 subdir = d_ancestor(old_dentr !! 3137 if (d_ancestor(old_dentry, new_dentry))
3063 read_sequnlock_excl(&rename_l !! 3138 result = true;
3064 } !! 3139 else
3065 rcu_read_unlock(); !! 3140 result = false;
3066 return subdir; !! 3141 rcu_read_unlock();
>> 3142 } while (read_seqretry(&rename_lock, seq));
>> 3143
>> 3144 return result;
3067 } 3145 }
3068 EXPORT_SYMBOL(is_subdir); 3146 EXPORT_SYMBOL(is_subdir);
3069 3147
3070 static enum d_walk_ret d_genocide_kill(void * 3148 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3071 { 3149 {
3072 struct dentry *root = data; 3150 struct dentry *root = data;
3073 if (dentry != root) { 3151 if (dentry != root) {
3074 if (d_unhashed(dentry) || !de 3152 if (d_unhashed(dentry) || !dentry->d_inode)
3075 return D_WALK_SKIP; 3153 return D_WALK_SKIP;
3076 3154
3077 if (!(dentry->d_flags & DCACH 3155 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3078 dentry->d_flags |= DC 3156 dentry->d_flags |= DCACHE_GENOCIDE;
3079 dentry->d_lockref.cou 3157 dentry->d_lockref.count--;
3080 } 3158 }
3081 } 3159 }
3082 return D_WALK_CONTINUE; 3160 return D_WALK_CONTINUE;
3083 } 3161 }
3084 3162
3085 void d_genocide(struct dentry *parent) 3163 void d_genocide(struct dentry *parent)
3086 { 3164 {
3087 d_walk(parent, parent, d_genocide_kil 3165 d_walk(parent, parent, d_genocide_kill);
3088 } 3166 }
3089 3167
3090 void d_mark_tmpfile(struct file *file, struct !! 3168 EXPORT_SYMBOL(d_genocide);
3091 { <<
3092 struct dentry *dentry = file->f_path. <<
3093 3169
>> 3170 void d_tmpfile(struct dentry *dentry, struct inode *inode)
>> 3171 {
>> 3172 inode_dec_link_count(inode);
3094 BUG_ON(dentry->d_name.name != dentry- 3173 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3095 !hlist_unhashed(&dentry->d_u. 3174 !hlist_unhashed(&dentry->d_u.d_alias) ||
3096 !d_unlinked(dentry)); 3175 !d_unlinked(dentry));
3097 spin_lock(&dentry->d_parent->d_lock); 3176 spin_lock(&dentry->d_parent->d_lock);
3098 spin_lock_nested(&dentry->d_lock, DEN 3177 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3099 dentry->d_name.len = sprintf(dentry-> 3178 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3100 (unsigned lon 3179 (unsigned long long)inode->i_ino);
3101 spin_unlock(&dentry->d_lock); 3180 spin_unlock(&dentry->d_lock);
3102 spin_unlock(&dentry->d_parent->d_lock 3181 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); 3182 d_instantiate(dentry, inode);
3113 } 3183 }
3114 EXPORT_SYMBOL(d_tmpfile); 3184 EXPORT_SYMBOL(d_tmpfile);
3115 3185
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 3186 static __initdata unsigned long dhash_entries;
3145 static int __init set_dhash_entries(char *str 3187 static int __init set_dhash_entries(char *str)
3146 { 3188 {
3147 if (!str) 3189 if (!str)
3148 return 0; 3190 return 0;
3149 dhash_entries = simple_strtoul(str, & 3191 dhash_entries = simple_strtoul(str, &str, 0);
3150 return 1; 3192 return 1;
3151 } 3193 }
3152 __setup("dhash_entries=", set_dhash_entries); 3194 __setup("dhash_entries=", set_dhash_entries);
3153 3195
3154 static void __init dcache_init_early(void) 3196 static void __init dcache_init_early(void)
3155 { 3197 {
3156 /* If hashes are distributed across N 3198 /* If hashes are distributed across NUMA nodes, defer
3157 * hash allocation until vmalloc spac 3199 * hash allocation until vmalloc space is available.
3158 */ 3200 */
3159 if (hashdist) 3201 if (hashdist)
3160 return; 3202 return;
3161 3203
3162 dentry_hashtable = 3204 dentry_hashtable =
3163 alloc_large_system_hash("Dent 3205 alloc_large_system_hash("Dentry cache",
3164 sizeo 3206 sizeof(struct hlist_bl_head),
3165 dhash 3207 dhash_entries,
3166 13, 3208 13,
3167 HASH_ 3209 HASH_EARLY | HASH_ZERO,
3168 &d_ha 3210 &d_hash_shift,
3169 NULL, 3211 NULL,
3170 0, 3212 0,
3171 0); 3213 0);
3172 d_hash_shift = 32 - d_hash_shift; 3214 d_hash_shift = 32 - d_hash_shift;
3173 <<
3174 runtime_const_init(shift, d_hash_shif <<
3175 runtime_const_init(ptr, dentry_hashta <<
3176 } 3215 }
3177 3216
3178 static void __init dcache_init(void) 3217 static void __init dcache_init(void)
3179 { 3218 {
3180 /* 3219 /*
3181 * A constructor could be added for s 3220 * A constructor could be added for stable state like the lists,
3182 * but it is probably not worth it be 3221 * but it is probably not worth it because of the cache nature
3183 * of the dcache. 3222 * of the dcache.
3184 */ 3223 */
3185 dentry_cache = KMEM_CACHE_USERCOPY(de 3224 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3186 SLAB_RECLAIM_ACCOUNT|SLAB_PAN !! 3225 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3187 d_iname); 3226 d_iname);
3188 3227
3189 /* Hash may have been set up in dcach 3228 /* Hash may have been set up in dcache_init_early */
3190 if (!hashdist) 3229 if (!hashdist)
3191 return; 3230 return;
3192 3231
3193 dentry_hashtable = 3232 dentry_hashtable =
3194 alloc_large_system_hash("Dent 3233 alloc_large_system_hash("Dentry cache",
3195 sizeo 3234 sizeof(struct hlist_bl_head),
3196 dhash 3235 dhash_entries,
3197 13, 3236 13,
3198 HASH_ 3237 HASH_ZERO,
3199 &d_ha 3238 &d_hash_shift,
3200 NULL, 3239 NULL,
3201 0, 3240 0,
3202 0); 3241 0);
3203 d_hash_shift = 32 - d_hash_shift; 3242 d_hash_shift = 32 - d_hash_shift;
3204 <<
3205 runtime_const_init(shift, d_hash_shif <<
3206 runtime_const_init(ptr, dentry_hashta <<
3207 } 3243 }
3208 3244
3209 /* SLAB cache for __getname() consumers */ 3245 /* SLAB cache for __getname() consumers */
3210 struct kmem_cache *names_cachep __ro_after_in !! 3246 struct kmem_cache *names_cachep __read_mostly;
3211 EXPORT_SYMBOL(names_cachep); 3247 EXPORT_SYMBOL(names_cachep);
3212 3248
3213 void __init vfs_caches_init_early(void) 3249 void __init vfs_caches_init_early(void)
3214 { 3250 {
3215 int i; 3251 int i;
3216 3252
3217 for (i = 0; i < ARRAY_SIZE(in_lookup_ 3253 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3218 INIT_HLIST_BL_HEAD(&in_lookup 3254 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3219 3255
3220 dcache_init_early(); 3256 dcache_init_early();
3221 inode_init_early(); 3257 inode_init_early();
3222 } 3258 }
3223 3259
3224 void __init vfs_caches_init(void) 3260 void __init vfs_caches_init(void)
3225 { 3261 {
3226 names_cachep = kmem_cache_create_user 3262 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3227 SLAB_HWCACHE_ALIGN|SL 3263 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3228 3264
3229 dcache_init(); 3265 dcache_init();
3230 inode_init(); 3266 inode_init();
3231 files_init(); 3267 files_init();
3232 files_maxfiles_init(); 3268 files_maxfiles_init();
3233 mnt_init(); 3269 mnt_init();
3234 bdev_cache_init(); 3270 bdev_cache_init();
3235 chrdev_init(); 3271 chrdev_init();
3236 } 3272 }
3237 3273
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.