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