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/syscalls.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>
>> 27 #include <linux/mount.h>
>> 28 #include <linux/file.h>
>> 29 #include <linux/uaccess.h>
29 #include <linux/security.h> 30 #include <linux/security.h>
30 #include <linux/seqlock.h> 31 #include <linux/seqlock.h>
31 #include <linux/memblock.h> !! 32 #include <linux/swap.h>
>> 33 #include <linux/bootmem.h>
>> 34 #include <linux/fs_struct.h>
>> 35 #include <linux/hardirq.h>
32 #include <linux/bit_spinlock.h> 36 #include <linux/bit_spinlock.h>
33 #include <linux/rculist_bl.h> 37 #include <linux/rculist_bl.h>
>> 38 #include <linux/prefetch.h>
>> 39 #include <linux/ratelimit.h>
34 #include <linux/list_lru.h> 40 #include <linux/list_lru.h>
>> 41 #include <linux/kasan.h>
>> 42
35 #include "internal.h" 43 #include "internal.h"
36 #include "mount.h" 44 #include "mount.h"
37 45
38 #include <asm/runtime-const.h> <<
39 <<
40 /* 46 /*
41 * Usage: 47 * Usage:
42 * dcache->d_inode->i_lock protects: 48 * dcache->d_inode->i_lock protects:
43 * - i_dentry, d_u.d_alias, d_inode of alias 49 * - i_dentry, d_u.d_alias, d_inode of aliases
44 * dcache_hash_bucket lock protects: 50 * dcache_hash_bucket lock protects:
45 * - the dcache hash table 51 * - the dcache hash table
46 * s_roots bl list spinlock protects: !! 52 * s_anon bl list spinlock protects:
47 * - the s_roots list (see __d_drop) !! 53 * - the s_anon list (see __d_drop)
48 * dentry->d_sb->s_dentry_lru_lock protects: 54 * dentry->d_sb->s_dentry_lru_lock protects:
49 * - the dcache lru lists and counters 55 * - the dcache lru lists and counters
50 * d_lock protects: 56 * d_lock protects:
51 * - d_flags 57 * - d_flags
52 * - d_name 58 * - d_name
53 * - d_lru 59 * - d_lru
54 * - d_count 60 * - d_count
55 * - d_unhashed() 61 * - d_unhashed()
56 * - d_parent and d_chilren !! 62 * - d_parent and d_subdirs
57 * - childrens' d_sib and d_parent !! 63 * - childrens' d_child and d_parent
58 * - d_u.d_alias, d_inode 64 * - d_u.d_alias, d_inode
59 * 65 *
60 * Ordering: 66 * Ordering:
61 * dentry->d_inode->i_lock 67 * dentry->d_inode->i_lock
62 * dentry->d_lock 68 * dentry->d_lock
63 * dentry->d_sb->s_dentry_lru_lock 69 * dentry->d_sb->s_dentry_lru_lock
64 * dcache_hash_bucket lock 70 * dcache_hash_bucket lock
65 * s_roots lock !! 71 * s_anon lock
66 * 72 *
67 * If there is an ancestor relationship: 73 * If there is an ancestor relationship:
68 * dentry->d_parent->...->d_parent->d_lock 74 * dentry->d_parent->...->d_parent->d_lock
69 * ... 75 * ...
70 * dentry->d_parent->d_lock 76 * dentry->d_parent->d_lock
71 * dentry->d_lock 77 * dentry->d_lock
72 * 78 *
73 * If no ancestor relationship: 79 * If no ancestor relationship:
74 * arbitrary, since it's serialized on rename_ !! 80 * if (dentry1 < dentry2)
>> 81 * dentry1->d_lock
>> 82 * dentry2->d_lock
75 */ 83 */
76 int sysctl_vfs_cache_pressure __read_mostly = 84 int sysctl_vfs_cache_pressure __read_mostly = 100;
77 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); 85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
78 86
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rena 87 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
80 88
81 EXPORT_SYMBOL(rename_lock); 89 EXPORT_SYMBOL(rename_lock);
82 90
83 static struct kmem_cache *dentry_cache __ro_af !! 91 static struct kmem_cache *dentry_cache __read_mostly;
84 92
85 const struct qstr empty_name = QSTR_INIT("", 0 93 const struct qstr empty_name = QSTR_INIT("", 0);
86 EXPORT_SYMBOL(empty_name); 94 EXPORT_SYMBOL(empty_name);
87 const struct qstr slash_name = QSTR_INIT("/", 95 const struct qstr slash_name = QSTR_INIT("/", 1);
88 EXPORT_SYMBOL(slash_name); 96 EXPORT_SYMBOL(slash_name);
89 const struct qstr dotdot_name = QSTR_INIT(".." <<
90 EXPORT_SYMBOL(dotdot_name); <<
91 97
92 /* 98 /*
93 * This is the single most critical data struc 99 * This is the single most critical data structure when it comes
94 * to the dcache: the hashtable for lookups. S 100 * to the dcache: the hashtable for lookups. Somebody should try
95 * to make this good - I've just made it work. 101 * to make this good - I've just made it work.
96 * 102 *
97 * This hash-function tries to avoid losing to 103 * This hash-function tries to avoid losing too many bits of hash
98 * information, yet avoid using a prime hash-s 104 * 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 */ 105 */
105 106
106 static unsigned int d_hash_shift __ro_after_in !! 107 static unsigned int d_hash_mask __read_mostly;
>> 108 static unsigned int d_hash_shift __read_mostly;
107 109
108 static struct hlist_bl_head *dentry_hashtable !! 110 static struct hlist_bl_head *dentry_hashtable __read_mostly;
109 111
110 static inline struct hlist_bl_head *d_hash(uns !! 112 static inline struct hlist_bl_head *d_hash(unsigned int hash)
111 { 113 {
112 return runtime_const_ptr(dentry_hashta !! 114 return dentry_hashtable + (hash >> (32 - d_hash_shift));
113 runtime_const_shift_right_32(h <<
114 } 115 }
115 116
116 #define IN_LOOKUP_SHIFT 10 117 #define IN_LOOKUP_SHIFT 10
117 static struct hlist_bl_head in_lookup_hashtabl 118 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
118 119
119 static inline struct hlist_bl_head *in_lookup_ 120 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
120 unsign 121 unsigned int hash)
121 { 122 {
122 hash += (unsigned long) parent / L1_CA 123 hash += (unsigned long) parent / L1_CACHE_BYTES;
123 return in_lookup_hashtable + hash_32(h 124 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
124 } 125 }
125 126
126 struct dentry_stat_t { !! 127
127 long nr_dentry; !! 128 /* Statistics gathering. */
128 long nr_unused; !! 129 struct dentry_stat_t dentry_stat = {
129 long age_limit; /* age in seco !! 130 .age_limit = 45,
130 long want_pages; /* pages reque <<
131 long nr_negative; /* # of unused <<
132 long dummy; /* Reserved fo <<
133 }; 131 };
134 132
135 static DEFINE_PER_CPU(long, nr_dentry); 133 static DEFINE_PER_CPU(long, nr_dentry);
136 static DEFINE_PER_CPU(long, nr_dentry_unused); 134 static DEFINE_PER_CPU(long, nr_dentry_unused);
137 static DEFINE_PER_CPU(long, nr_dentry_negative <<
138 135
139 #if defined(CONFIG_SYSCTL) && defined(CONFIG_P 136 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
140 /* Statistics gathering. */ <<
141 static struct dentry_stat_t dentry_stat = { <<
142 .age_limit = 45, <<
143 }; <<
144 137
145 /* 138 /*
146 * Here we resort to our own counters instead 139 * Here we resort to our own counters instead of using generic per-cpu counters
147 * for consistency with what the vfs inode cod 140 * for consistency with what the vfs inode code does. We are expected to harvest
148 * better code and performance by having our o 141 * better code and performance by having our own specialized counters.
149 * 142 *
150 * Please note that the loop is done over all 143 * 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 144 * 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 145 * on and off. If one of them goes off, we will just keep their counters.
153 * 146 *
154 * glommer: See cffbc8a for details, and if yo 147 * glommer: See cffbc8a for details, and if you ever intend to change this,
155 * please update all vfs counters to match. 148 * please update all vfs counters to match.
156 */ 149 */
157 static long get_nr_dentry(void) 150 static long get_nr_dentry(void)
158 { 151 {
159 int i; 152 int i;
160 long sum = 0; 153 long sum = 0;
161 for_each_possible_cpu(i) 154 for_each_possible_cpu(i)
162 sum += per_cpu(nr_dentry, i); 155 sum += per_cpu(nr_dentry, i);
163 return sum < 0 ? 0 : sum; 156 return sum < 0 ? 0 : sum;
164 } 157 }
165 158
166 static long get_nr_dentry_unused(void) 159 static long get_nr_dentry_unused(void)
167 { 160 {
168 int i; 161 int i;
169 long sum = 0; 162 long sum = 0;
170 for_each_possible_cpu(i) 163 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry_unuse 164 sum += per_cpu(nr_dentry_unused, i);
172 return sum < 0 ? 0 : sum; 165 return sum < 0 ? 0 : sum;
173 } 166 }
174 167
175 static long get_nr_dentry_negative(void) !! 168 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
176 { !! 169 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 { 170 {
188 dentry_stat.nr_dentry = get_nr_dentry( 171 dentry_stat.nr_dentry = get_nr_dentry();
189 dentry_stat.nr_unused = get_nr_dentry_ 172 dentry_stat.nr_unused = get_nr_dentry_unused();
190 dentry_stat.nr_negative = get_nr_dentr <<
191 return proc_doulongvec_minmax(table, w 173 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
192 } 174 }
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 175 #endif
211 176
212 /* 177 /*
213 * Compare 2 name strings, return 0 if they ma 178 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
214 * The strings are both count bytes long, and 179 * The strings are both count bytes long, and count is non-zero.
215 */ 180 */
216 #ifdef CONFIG_DCACHE_WORD_ACCESS 181 #ifdef CONFIG_DCACHE_WORD_ACCESS
217 182
218 #include <asm/word-at-a-time.h> 183 #include <asm/word-at-a-time.h>
219 /* 184 /*
220 * NOTE! 'cs' and 'scount' come from a dentry, 185 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
221 * aligned allocation for this particular comp 186 * aligned allocation for this particular component. We don't
222 * strictly need the load_unaligned_zeropad() 187 * strictly need the load_unaligned_zeropad() safety, but it
223 * doesn't hurt either. 188 * doesn't hurt either.
224 * 189 *
225 * In contrast, 'ct' and 'tcount' can be from 190 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
226 * need the careful unaligned handling. 191 * need the careful unaligned handling.
227 */ 192 */
228 static inline int dentry_string_cmp(const unsi 193 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
229 { 194 {
230 unsigned long a,b,mask; 195 unsigned long a,b,mask;
231 196
232 for (;;) { 197 for (;;) {
233 a = read_word_at_a_time(cs); !! 198 a = *(unsigned long *)cs;
234 b = load_unaligned_zeropad(ct) 199 b = load_unaligned_zeropad(ct);
235 if (tcount < sizeof(unsigned l 200 if (tcount < sizeof(unsigned long))
236 break; 201 break;
237 if (unlikely(a != b)) 202 if (unlikely(a != b))
238 return 1; 203 return 1;
239 cs += sizeof(unsigned long); 204 cs += sizeof(unsigned long);
240 ct += sizeof(unsigned long); 205 ct += sizeof(unsigned long);
241 tcount -= sizeof(unsigned long 206 tcount -= sizeof(unsigned long);
242 if (!tcount) 207 if (!tcount)
243 return 0; 208 return 0;
244 } 209 }
245 mask = bytemask_from_count(tcount); 210 mask = bytemask_from_count(tcount);
246 return unlikely(!!((a ^ b) & mask)); 211 return unlikely(!!((a ^ b) & mask));
247 } 212 }
248 213
249 #else 214 #else
250 215
251 static inline int dentry_string_cmp(const unsi 216 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
252 { 217 {
253 do { 218 do {
254 if (*cs != *ct) 219 if (*cs != *ct)
255 return 1; 220 return 1;
256 cs++; 221 cs++;
257 ct++; 222 ct++;
258 tcount--; 223 tcount--;
259 } while (tcount); 224 } while (tcount);
260 return 0; 225 return 0;
261 } 226 }
262 227
263 #endif 228 #endif
264 229
265 static inline int dentry_cmp(const struct dent 230 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
266 { 231 {
267 /* 232 /*
268 * Be careful about RCU walk racing wi 233 * Be careful about RCU walk racing with rename:
269 * use 'READ_ONCE' to fetch the name p 234 * use 'READ_ONCE' to fetch the name pointer.
270 * 235 *
271 * NOTE! Even if a rename will mean th 236 * NOTE! Even if a rename will mean that the length
272 * was not loaded atomically, we don't 237 * was not loaded atomically, we don't care. The
273 * RCU walk will check the sequence co 238 * RCU walk will check the sequence count eventually,
274 * and catch it. And we won't overrun 239 * and catch it. And we won't overrun the buffer,
275 * because we're reading the name poin 240 * because we're reading the name pointer atomically,
276 * and a dentry name is guaranteed to 241 * and a dentry name is guaranteed to be properly
277 * terminated with a NUL byte. 242 * terminated with a NUL byte.
278 * 243 *
279 * End result: even if 'len' is wrong, 244 * End result: even if 'len' is wrong, we'll exit
280 * early because the data cannot match 245 * early because the data cannot match (there can
281 * be no NUL in the ct/tcount data) 246 * be no NUL in the ct/tcount data)
282 */ 247 */
283 const unsigned char *cs = READ_ONCE(de 248 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
284 249
285 return dentry_string_cmp(cs, ct, tcoun 250 return dentry_string_cmp(cs, ct, tcount);
286 } 251 }
287 252
288 struct external_name { 253 struct external_name {
289 union { 254 union {
290 atomic_t count; 255 atomic_t count;
291 struct rcu_head head; 256 struct rcu_head head;
292 } u; 257 } u;
293 unsigned char name[]; 258 unsigned char name[];
294 }; 259 };
295 260
296 static inline struct external_name *external_n 261 static inline struct external_name *external_name(struct dentry *dentry)
297 { 262 {
298 return container_of(dentry->d_name.nam 263 return container_of(dentry->d_name.name, struct external_name, name[0]);
299 } 264 }
300 265
301 static void __d_free(struct rcu_head *head) 266 static void __d_free(struct rcu_head *head)
302 { 267 {
303 struct dentry *dentry = container_of(h 268 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
304 269
305 kmem_cache_free(dentry_cache, dentry); 270 kmem_cache_free(dentry_cache, dentry);
306 } 271 }
307 272
308 static void __d_free_external(struct rcu_head 273 static void __d_free_external(struct rcu_head *head)
309 { 274 {
310 struct dentry *dentry = container_of(h 275 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
311 kfree(external_name(dentry)); 276 kfree(external_name(dentry));
312 kmem_cache_free(dentry_cache, dentry); !! 277 kmem_cache_free(dentry_cache, dentry);
313 } 278 }
314 279
315 static inline int dname_external(const struct 280 static inline int dname_external(const struct dentry *dentry)
316 { 281 {
317 return dentry->d_name.name != dentry-> 282 return dentry->d_name.name != dentry->d_iname;
318 } 283 }
319 284
320 void take_dentry_name_snapshot(struct name_sna 285 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
321 { 286 {
322 spin_lock(&dentry->d_lock); 287 spin_lock(&dentry->d_lock);
323 name->name = dentry->d_name; <<
324 if (unlikely(dname_external(dentry))) 288 if (unlikely(dname_external(dentry))) {
325 atomic_inc(&external_name(dent !! 289 struct external_name *p = external_name(dentry);
>> 290 atomic_inc(&p->u.count);
>> 291 spin_unlock(&dentry->d_lock);
>> 292 name->name = p->name;
326 } else { 293 } else {
327 memcpy(name->inline_name, dent !! 294 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
328 dentry->d_name.len + 1) !! 295 spin_unlock(&dentry->d_lock);
329 name->name.name = name->inline !! 296 name->name = name->inline_name;
330 } 297 }
331 spin_unlock(&dentry->d_lock); <<
332 } 298 }
333 EXPORT_SYMBOL(take_dentry_name_snapshot); 299 EXPORT_SYMBOL(take_dentry_name_snapshot);
334 300
335 void release_dentry_name_snapshot(struct name_ 301 void release_dentry_name_snapshot(struct name_snapshot *name)
336 { 302 {
337 if (unlikely(name->name.name != name-> !! 303 if (unlikely(name->name != name->inline_name)) {
338 struct external_name *p; 304 struct external_name *p;
339 p = container_of(name->name.na !! 305 p = container_of(name->name, struct external_name, name[0]);
340 if (unlikely(atomic_dec_and_te 306 if (unlikely(atomic_dec_and_test(&p->u.count)))
341 kfree_rcu(p, u.head); 307 kfree_rcu(p, u.head);
342 } 308 }
343 } 309 }
344 EXPORT_SYMBOL(release_dentry_name_snapshot); 310 EXPORT_SYMBOL(release_dentry_name_snapshot);
345 311
346 static inline void __d_set_inode_and_type(stru 312 static inline void __d_set_inode_and_type(struct dentry *dentry,
347 stru 313 struct inode *inode,
348 unsi 314 unsigned type_flags)
349 { 315 {
350 unsigned flags; 316 unsigned flags;
351 317
352 dentry->d_inode = inode; 318 dentry->d_inode = inode;
353 flags = READ_ONCE(dentry->d_flags); 319 flags = READ_ONCE(dentry->d_flags);
354 flags &= ~DCACHE_ENTRY_TYPE; !! 320 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
355 flags |= type_flags; 321 flags |= type_flags;
356 smp_store_release(&dentry->d_flags, fl !! 322 WRITE_ONCE(dentry->d_flags, flags);
357 } 323 }
358 324
359 static inline void __d_clear_type_and_inode(st 325 static inline void __d_clear_type_and_inode(struct dentry *dentry)
360 { 326 {
361 unsigned flags = READ_ONCE(dentry->d_f 327 unsigned flags = READ_ONCE(dentry->d_flags);
362 328
363 flags &= ~DCACHE_ENTRY_TYPE; !! 329 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
364 WRITE_ONCE(dentry->d_flags, flags); 330 WRITE_ONCE(dentry->d_flags, flags);
365 dentry->d_inode = NULL; 331 dentry->d_inode = NULL;
366 /* <<
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 } 332 }
373 333
374 static void dentry_free(struct dentry *dentry) 334 static void dentry_free(struct dentry *dentry)
375 { 335 {
376 WARN_ON(!hlist_unhashed(&dentry->d_u.d 336 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
377 if (unlikely(dname_external(dentry))) 337 if (unlikely(dname_external(dentry))) {
378 struct external_name *p = exte 338 struct external_name *p = external_name(dentry);
379 if (likely(atomic_dec_and_test 339 if (likely(atomic_dec_and_test(&p->u.count))) {
380 call_rcu(&dentry->d_u. 340 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
381 return; 341 return;
382 } 342 }
383 } 343 }
384 /* if dentry was never visible to RCU, 344 /* if dentry was never visible to RCU, immediate free is OK */
385 if (dentry->d_flags & DCACHE_NORCU) !! 345 if (!(dentry->d_flags & DCACHE_RCUACCESS))
386 __d_free(&dentry->d_u.d_rcu); 346 __d_free(&dentry->d_u.d_rcu);
387 else 347 else
388 call_rcu(&dentry->d_u.d_rcu, _ 348 call_rcu(&dentry->d_u.d_rcu, __d_free);
389 } 349 }
390 350
391 /* 351 /*
392 * Release the dentry's inode, using the files 352 * Release the dentry's inode, using the filesystem
393 * d_iput() operation if defined. 353 * d_iput() operation if defined.
394 */ 354 */
395 static void dentry_unlink_inode(struct dentry 355 static void dentry_unlink_inode(struct dentry * dentry)
396 __releases(dentry->d_lock) 356 __releases(dentry->d_lock)
397 __releases(dentry->d_inode->i_lock) 357 __releases(dentry->d_inode->i_lock)
398 { 358 {
399 struct inode *inode = dentry->d_inode; 359 struct inode *inode = dentry->d_inode;
>> 360 bool hashed = !d_unhashed(dentry);
400 361
401 raw_write_seqcount_begin(&dentry->d_se !! 362 if (hashed)
>> 363 raw_write_seqcount_begin(&dentry->d_seq);
402 __d_clear_type_and_inode(dentry); 364 __d_clear_type_and_inode(dentry);
403 hlist_del_init(&dentry->d_u.d_alias); 365 hlist_del_init(&dentry->d_u.d_alias);
404 raw_write_seqcount_end(&dentry->d_seq) !! 366 if (hashed)
>> 367 raw_write_seqcount_end(&dentry->d_seq);
405 spin_unlock(&dentry->d_lock); 368 spin_unlock(&dentry->d_lock);
406 spin_unlock(&inode->i_lock); 369 spin_unlock(&inode->i_lock);
407 if (!inode->i_nlink) 370 if (!inode->i_nlink)
408 fsnotify_inoderemove(inode); 371 fsnotify_inoderemove(inode);
409 if (dentry->d_op && dentry->d_op->d_ip 372 if (dentry->d_op && dentry->d_op->d_iput)
410 dentry->d_op->d_iput(dentry, i 373 dentry->d_op->d_iput(dentry, inode);
411 else 374 else
412 iput(inode); 375 iput(inode);
413 } 376 }
414 377
415 /* 378 /*
416 * The DCACHE_LRU_LIST bit is set whenever the 379 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
417 * is in use - which includes both the "real" 380 * is in use - which includes both the "real" per-superblock
418 * LRU list _and_ the DCACHE_SHRINK_LIST use. 381 * LRU list _and_ the DCACHE_SHRINK_LIST use.
419 * 382 *
420 * The DCACHE_SHRINK_LIST bit is set whenever 383 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
421 * on the shrink list (ie not on the superbloc 384 * on the shrink list (ie not on the superblock LRU list).
422 * 385 *
423 * The per-cpu "nr_dentry_unused" counters are 386 * The per-cpu "nr_dentry_unused" counters are updated with
424 * the DCACHE_LRU_LIST bit. 387 * the DCACHE_LRU_LIST bit.
425 * 388 *
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 389 * These helper functions make sure we always follow the
432 * rules. d_lock must be held by the caller. 390 * rules. d_lock must be held by the caller.
433 */ 391 */
434 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(( 392 #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) 393 static void d_lru_add(struct dentry *dentry)
436 { 394 {
437 D_FLAG_VERIFY(dentry, 0); 395 D_FLAG_VERIFY(dentry, 0);
438 dentry->d_flags |= DCACHE_LRU_LIST; 396 dentry->d_flags |= DCACHE_LRU_LIST;
439 this_cpu_inc(nr_dentry_unused); 397 this_cpu_inc(nr_dentry_unused);
440 if (d_is_negative(dentry)) !! 398 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 } 399 }
445 400
446 static void d_lru_del(struct dentry *dentry) 401 static void d_lru_del(struct dentry *dentry)
447 { 402 {
448 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 403 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
449 dentry->d_flags &= ~DCACHE_LRU_LIST; 404 dentry->d_flags &= ~DCACHE_LRU_LIST;
450 this_cpu_dec(nr_dentry_unused); 405 this_cpu_dec(nr_dentry_unused);
451 if (d_is_negative(dentry)) !! 406 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 } 407 }
456 408
457 static void d_shrink_del(struct dentry *dentry 409 static void d_shrink_del(struct dentry *dentry)
458 { 410 {
459 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LI 411 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
460 list_del_init(&dentry->d_lru); 412 list_del_init(&dentry->d_lru);
461 dentry->d_flags &= ~(DCACHE_SHRINK_LIS 413 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
462 this_cpu_dec(nr_dentry_unused); 414 this_cpu_dec(nr_dentry_unused);
463 } 415 }
464 416
465 static void d_shrink_add(struct dentry *dentry 417 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
466 { 418 {
467 D_FLAG_VERIFY(dentry, 0); 419 D_FLAG_VERIFY(dentry, 0);
468 list_add(&dentry->d_lru, list); 420 list_add(&dentry->d_lru, list);
469 dentry->d_flags |= DCACHE_SHRINK_LIST 421 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
470 this_cpu_inc(nr_dentry_unused); 422 this_cpu_inc(nr_dentry_unused);
471 } 423 }
472 424
473 /* 425 /*
474 * These can only be called under the global L 426 * These can only be called under the global LRU lock, ie during the
475 * callback for freeing the LRU list. "isolate 427 * callback for freeing the LRU list. "isolate" removes it from the
476 * LRU lists entirely, while shrink_move moves 428 * LRU lists entirely, while shrink_move moves it to the indicated
477 * private list. 429 * private list.
478 */ 430 */
479 static void d_lru_isolate(struct list_lru_one 431 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
480 { 432 {
481 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 433 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
482 dentry->d_flags &= ~DCACHE_LRU_LIST; 434 dentry->d_flags &= ~DCACHE_LRU_LIST;
483 this_cpu_dec(nr_dentry_unused); 435 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); 436 list_lru_isolate(lru, &dentry->d_lru);
487 } 437 }
488 438
489 static void d_lru_shrink_move(struct list_lru_ 439 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
490 struct list_head 440 struct list_head *list)
491 { 441 {
492 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST) 442 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
493 dentry->d_flags |= DCACHE_SHRINK_LIST; 443 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_ 444 list_lru_isolate_move(lru, &dentry->d_lru, list);
497 } 445 }
498 446
499 static void ___d_drop(struct dentry *dentry) !! 447 /*
500 { !! 448 * dentry_lru_(add|del)_list) must be called with d_lock held.
501 struct hlist_bl_head *b; !! 449 */
502 /* !! 450 static void dentry_lru_add(struct dentry *dentry)
503 * Hashed dentries are normally on the <<
504 * with the exception of those newly a <<
505 * d_obtain_root, which are always IS_ <<
506 */ <<
507 if (unlikely(IS_ROOT(dentry))) <<
508 b = &dentry->d_sb->s_roots; <<
509 else <<
510 b = d_hash(dentry->d_name.hash <<
511 <<
512 hlist_bl_lock(b); <<
513 __hlist_bl_del(&dentry->d_hash); <<
514 hlist_bl_unlock(b); <<
515 } <<
516 <<
517 void __d_drop(struct dentry *dentry) <<
518 { 451 {
519 if (!d_unhashed(dentry)) { !! 452 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
520 ___d_drop(dentry); !! 453 d_lru_add(dentry);
521 dentry->d_hash.pprev = NULL; !! 454 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
522 write_seqcount_invalidate(&den !! 455 dentry->d_flags |= DCACHE_REFERENCED;
523 } <<
524 } 456 }
525 EXPORT_SYMBOL(__d_drop); <<
526 457
527 /** 458 /**
528 * d_drop - drop a dentry 459 * d_drop - drop a dentry
529 * @dentry: dentry to drop 460 * @dentry: dentry to drop
530 * 461 *
531 * d_drop() unhashes the entry from the parent 462 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
532 * be found through a VFS lookup any more. Not 463 * be found through a VFS lookup any more. Note that this is different from
533 * deleting the dentry - d_delete will try to 464 * deleting the dentry - d_delete will try to mark the dentry negative if
534 * possible, giving a successful _negative_ lo 465 * possible, giving a successful _negative_ lookup, while d_drop will
535 * just make the cache lookup fail. 466 * just make the cache lookup fail.
536 * 467 *
537 * d_drop() is used mainly for stuff that want 468 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
538 * reason (NFS timeouts or autofs deletes). 469 * reason (NFS timeouts or autofs deletes).
539 * 470 *
540 * __d_drop requires dentry->d_lock 471 * __d_drop requires dentry->d_lock
541 * <<
542 * ___d_drop doesn't mark dentry as "unhashed" 472 * ___d_drop doesn't mark dentry as "unhashed"
543 * (dentry->d_hash.pprev will be LIST_POISON2, !! 473 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
544 */ 474 */
>> 475 static void ___d_drop(struct dentry *dentry)
>> 476 {
>> 477 if (!d_unhashed(dentry)) {
>> 478 struct hlist_bl_head *b;
>> 479 /*
>> 480 * Hashed dentries are normally on the dentry hashtable,
>> 481 * with the exception of those newly allocated by
>> 482 * d_obtain_alias, which are always IS_ROOT:
>> 483 */
>> 484 if (unlikely(IS_ROOT(dentry)))
>> 485 b = &dentry->d_sb->s_anon;
>> 486 else
>> 487 b = d_hash(dentry->d_name.hash);
>> 488
>> 489 hlist_bl_lock(b);
>> 490 __hlist_bl_del(&dentry->d_hash);
>> 491 hlist_bl_unlock(b);
>> 492 /* After this call, in-progress rcu-walk path lookup will fail. */
>> 493 write_seqcount_invalidate(&dentry->d_seq);
>> 494 }
>> 495 }
>> 496
>> 497 void __d_drop(struct dentry *dentry)
>> 498 {
>> 499 ___d_drop(dentry);
>> 500 dentry->d_hash.pprev = NULL;
>> 501 }
>> 502 EXPORT_SYMBOL(__d_drop);
>> 503
545 void d_drop(struct dentry *dentry) 504 void d_drop(struct dentry *dentry)
546 { 505 {
547 spin_lock(&dentry->d_lock); 506 spin_lock(&dentry->d_lock);
548 __d_drop(dentry); 507 __d_drop(dentry);
549 spin_unlock(&dentry->d_lock); 508 spin_unlock(&dentry->d_lock);
550 } 509 }
551 EXPORT_SYMBOL(d_drop); 510 EXPORT_SYMBOL(d_drop);
552 511
553 static inline void dentry_unlist(struct dentry !! 512 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
554 { 513 {
555 struct dentry *next; 514 struct dentry *next;
556 /* 515 /*
557 * Inform d_walk() and shrink_dentry_l 516 * Inform d_walk() and shrink_dentry_list() that we are no longer
558 * attached to the dentry tree 517 * attached to the dentry tree
559 */ 518 */
560 dentry->d_flags |= DCACHE_DENTRY_KILLE 519 dentry->d_flags |= DCACHE_DENTRY_KILLED;
561 if (unlikely(hlist_unhashed(&dentry->d !! 520 if (unlikely(list_empty(&dentry->d_child)))
562 return; 521 return;
563 __hlist_del(&dentry->d_sib); !! 522 __list_del_entry(&dentry->d_child);
564 /* 523 /*
565 * Cursors can move around the list of 524 * Cursors can move around the list of children. While we'd been
566 * a normal list member, it didn't mat !! 525 * a normal list member, it didn't matter - ->d_child.next would've
567 * been updated. However, from now on 526 * been updated. However, from now on it won't be and for the
568 * things like d_walk() it might end u 527 * things like d_walk() it might end up with a nasty surprise.
569 * Normally d_walk() doesn't care abou 528 * Normally d_walk() doesn't care about cursors moving around -
570 * ->d_lock on parent prevents that an 529 * ->d_lock on parent prevents that and since a cursor has no children
571 * of its own, we get through it witho 530 * of its own, we get through it without ever unlocking the parent.
572 * There is one exception, though - if 531 * There is one exception, though - if we ascend from a child that
573 * gets killed as soon as we unlock it 532 * gets killed as soon as we unlock it, the next sibling is found
574 * using the value left in its ->d_sib !! 533 * using the value left in its ->d_child.next. And if _that_
575 * pointed to a cursor, and cursor got 534 * pointed to a cursor, and cursor got moved (e.g. by lseek())
576 * before d_walk() regains parent->d_l 535 * before d_walk() regains parent->d_lock, we'll end up skipping
577 * everything the cursor had been move 536 * everything the cursor had been moved past.
578 * 537 *
579 * Solution: make sure that the pointe !! 538 * Solution: make sure that the pointer left behind in ->d_child.next
580 * points to something that won't be m 539 * points to something that won't be moving around. I.e. skip the
581 * cursors. 540 * cursors.
582 */ 541 */
583 while (dentry->d_sib.next) { !! 542 while (dentry->d_child.next != &parent->d_subdirs) {
584 next = hlist_entry(dentry->d_s !! 543 next = list_entry(dentry->d_child.next, struct dentry, d_child);
585 if (likely(!(next->d_flags & D 544 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
586 break; 545 break;
587 dentry->d_sib.next = next->d_s !! 546 dentry->d_child.next = next->d_child.next;
588 } 547 }
589 } 548 }
590 549
591 static struct dentry *__dentry_kill(struct den !! 550 static void __dentry_kill(struct dentry *dentry)
592 { 551 {
593 struct dentry *parent = NULL; 552 struct dentry *parent = NULL;
594 bool can_free = true; 553 bool can_free = true;
>> 554 if (!IS_ROOT(dentry))
>> 555 parent = dentry->d_parent;
595 556
596 /* 557 /*
597 * The dentry is now unrecoverably dea 558 * The dentry is now unrecoverably dead to the world.
598 */ 559 */
599 lockref_mark_dead(&dentry->d_lockref); 560 lockref_mark_dead(&dentry->d_lockref);
600 561
601 /* 562 /*
602 * inform the fs via d_prune that this 563 * inform the fs via d_prune that this dentry is about to be
603 * unhashed and destroyed. 564 * unhashed and destroyed.
604 */ 565 */
605 if (dentry->d_flags & DCACHE_OP_PRUNE) 566 if (dentry->d_flags & DCACHE_OP_PRUNE)
606 dentry->d_op->d_prune(dentry); 567 dentry->d_op->d_prune(dentry);
607 568
608 if (dentry->d_flags & DCACHE_LRU_LIST) 569 if (dentry->d_flags & DCACHE_LRU_LIST) {
609 if (!(dentry->d_flags & DCACHE 570 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
610 d_lru_del(dentry); 571 d_lru_del(dentry);
611 } 572 }
612 /* if it was on the hash then remove i 573 /* if it was on the hash then remove it */
613 __d_drop(dentry); 574 __d_drop(dentry);
>> 575 dentry_unlist(dentry, parent);
>> 576 if (parent)
>> 577 spin_unlock(&parent->d_lock);
614 if (dentry->d_inode) 578 if (dentry->d_inode)
615 dentry_unlink_inode(dentry); 579 dentry_unlink_inode(dentry);
616 else 580 else
617 spin_unlock(&dentry->d_lock); 581 spin_unlock(&dentry->d_lock);
618 this_cpu_dec(nr_dentry); 582 this_cpu_dec(nr_dentry);
619 if (dentry->d_op && dentry->d_op->d_re 583 if (dentry->d_op && dentry->d_op->d_release)
620 dentry->d_op->d_release(dentry 584 dentry->d_op->d_release(dentry);
621 585
622 cond_resched(); !! 586 spin_lock(&dentry->d_lock);
623 /* now that it's negative, ->d_parent !! 587 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
624 if (!IS_ROOT(dentry)) { !! 588 dentry->d_flags |= DCACHE_MAY_FREE;
625 parent = dentry->d_parent; <<
626 spin_lock(&parent->d_lock); <<
627 } <<
628 spin_lock_nested(&dentry->d_lock, DENT <<
629 dentry_unlist(dentry); <<
630 if (dentry->d_flags & DCACHE_SHRINK_LI <<
631 can_free = false; 589 can_free = false;
>> 590 }
632 spin_unlock(&dentry->d_lock); 591 spin_unlock(&dentry->d_lock);
633 if (likely(can_free)) 592 if (likely(can_free))
634 dentry_free(dentry); 593 dentry_free(dentry);
635 if (parent && --parent->d_lockref.coun <<
636 spin_unlock(&parent->d_lock); <<
637 return NULL; <<
638 } <<
639 return parent; <<
640 } 594 }
641 595
642 /* 596 /*
643 * Lock a dentry for feeding it to __dentry_ki !! 597 * Finish off a dentry we've decided to kill.
644 * Called under rcu_read_lock() and dentry->d_ !! 598 * dentry->d_lock must be held, returns with it unlocked.
645 * guarantees that nothing we access will be f !! 599 * If ref is non-zero, then decrement the refcount too.
646 * Note that dentry is *not* protected from co !! 600 * Returns dentry requiring refcount drop, or NULL if we're done.
647 * d_delete(), etc. <<
648 * <<
649 * Return false if dentry is busy. Otherwise, <<
650 * that dentry's inode locked. <<
651 */ 601 */
652 !! 602 static struct dentry *dentry_kill(struct dentry *dentry)
653 static bool lock_for_kill(struct dentry *dentr !! 603 __releases(dentry->d_lock)
654 { 604 {
655 struct inode *inode = dentry->d_inode; 605 struct inode *inode = dentry->d_inode;
>> 606 struct dentry *parent = NULL;
656 607
657 if (unlikely(dentry->d_lockref.count)) !! 608 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
658 return false; !! 609 goto failed;
659 <<
660 if (!inode || likely(spin_trylock(&ino <<
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 } <<
678 <<
679 /* <<
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 { <<
692 unsigned int d_flags; <<
693 <<
694 smp_rmb(); <<
695 d_flags = READ_ONCE(dentry->d_flags); <<
696 <<
697 // Unreachable? Nobody would be able t <<
698 if (unlikely(d_unhashed(dentry))) <<
699 return false; <<
700 610
701 // Same if it's disconnected !! 611 if (!IS_ROOT(dentry)) {
702 if (unlikely(d_flags & DCACHE_DISCONNE !! 612 parent = dentry->d_parent;
703 return false; !! 613 if (unlikely(!spin_trylock(&parent->d_lock))) {
704 !! 614 if (inode)
705 // ->d_delete() might tell us not to b !! 615 spin_unlock(&inode->i_lock);
706 // ->d_lock; can't decide without it !! 616 goto failed;
707 if (unlikely(d_flags & DCACHE_OP_DELET !! 617 }
708 if (!locked || dentry->d_op->d <<
709 return false; <<
710 } 618 }
711 619
712 // Explicitly told not to bother !! 620 __dentry_kill(dentry);
713 if (unlikely(d_flags & DCACHE_DONTCACH !! 621 return parent;
714 return false; !! 622
715 !! 623 failed:
716 // At this point it looks like we ough !! 624 spin_unlock(&dentry->d_lock);
717 // need to do something - put it on LR !! 625 return dentry; /* try again with same dentry */
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); <<
725 } else if (unlikely(!(d_flags & DCACHE <<
726 if (!locked) <<
727 return false; <<
728 dentry->d_flags |= DCACHE_REFE <<
729 } <<
730 return true; <<
731 } 626 }
732 627
733 void d_mark_dontcache(struct inode *inode) !! 628 static inline struct dentry *lock_parent(struct dentry *dentry)
734 { 629 {
735 struct dentry *de; !! 630 struct dentry *parent = dentry->d_parent;
736 !! 631 if (IS_ROOT(dentry))
737 spin_lock(&inode->i_lock); !! 632 return NULL;
738 hlist_for_each_entry(de, &inode->i_den !! 633 if (unlikely(dentry->d_lockref.count < 0))
739 spin_lock(&de->d_lock); !! 634 return NULL;
740 de->d_flags |= DCACHE_DONTCACH !! 635 if (likely(spin_trylock(&parent->d_lock)))
741 spin_unlock(&de->d_lock); !! 636 return parent;
>> 637 rcu_read_lock();
>> 638 spin_unlock(&dentry->d_lock);
>> 639 again:
>> 640 parent = READ_ONCE(dentry->d_parent);
>> 641 spin_lock(&parent->d_lock);
>> 642 /*
>> 643 * We can't blindly lock dentry until we are sure
>> 644 * that we won't violate the locking order.
>> 645 * Any changes of dentry->d_parent must have
>> 646 * been done with parent->d_lock held, so
>> 647 * spin_lock() above is enough of a barrier
>> 648 * for checking if it's still our child.
>> 649 */
>> 650 if (unlikely(parent != dentry->d_parent)) {
>> 651 spin_unlock(&parent->d_lock);
>> 652 goto again;
742 } 653 }
743 inode->i_state |= I_DONTCACHE; !! 654 if (parent != dentry) {
744 spin_unlock(&inode->i_lock); !! 655 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
>> 656 if (unlikely(dentry->d_lockref.count < 0)) {
>> 657 spin_unlock(&parent->d_lock);
>> 658 parent = NULL;
>> 659 }
>> 660 } else {
>> 661 parent = NULL;
>> 662 }
>> 663 rcu_read_unlock();
>> 664 return parent;
745 } 665 }
746 EXPORT_SYMBOL(d_mark_dontcache); <<
747 666
748 /* 667 /*
749 * Try to do a lockless dput(), and return whe 668 * Try to do a lockless dput(), and return whether that was successful.
750 * 669 *
751 * If unsuccessful, we return false, having al 670 * 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 * 671 *
755 * The caller needs to hold the RCU read lock, 672 * The caller needs to hold the RCU read lock, so that the dentry is
756 * guaranteed to stay around even if the refco 673 * guaranteed to stay around even if the refcount goes down to zero!
757 */ 674 */
758 static inline bool fast_dput(struct dentry *de 675 static inline bool fast_dput(struct dentry *dentry)
759 { 676 {
760 int ret; 677 int ret;
>> 678 unsigned int d_flags;
>> 679
>> 680 /*
>> 681 * If we have a d_op->d_delete() operation, we sould not
>> 682 * let the dentry count go to zero, so use "put_or_lock".
>> 683 */
>> 684 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
>> 685 return lockref_put_or_lock(&dentry->d_lockref);
761 686
762 /* 687 /*
763 * try to decrement the lockref optimi !! 688 * .. otherwise, we can try to just decrement the
>> 689 * lockref optimistically.
764 */ 690 */
765 ret = lockref_put_return(&dentry->d_lo 691 ret = lockref_put_return(&dentry->d_lockref);
766 692
767 /* 693 /*
768 * If the lockref_put_return() failed 694 * If the lockref_put_return() failed due to the lock being held
769 * by somebody else, the fast path has 695 * by somebody else, the fast path has failed. We will need to
770 * get the lock, and then check the co 696 * get the lock, and then check the count again.
771 */ 697 */
772 if (unlikely(ret < 0)) { 698 if (unlikely(ret < 0)) {
773 spin_lock(&dentry->d_lock); 699 spin_lock(&dentry->d_lock);
774 if (WARN_ON_ONCE(dentry->d_loc !! 700 if (dentry->d_lockref.count > 1) {
>> 701 dentry->d_lockref.count--;
775 spin_unlock(&dentry->d 702 spin_unlock(&dentry->d_lock);
776 return true; !! 703 return 1;
777 } 704 }
778 dentry->d_lockref.count--; !! 705 return 0;
779 goto locked; <<
780 } 706 }
781 707
782 /* 708 /*
783 * If we weren't the last ref, we're d 709 * If we weren't the last ref, we're done.
784 */ 710 */
785 if (ret) 711 if (ret)
786 return true; !! 712 return 1;
787 713
788 /* 714 /*
789 * Can we decide that decrement of ref !! 715 * Careful, careful. The reference count went down
790 * taking the lock? There's a very co !! 716 * to zero, but we don't hold the dentry lock, so
791 * dentry looks like it ought to be re !! 717 * somebody else could get it again, and do another
792 * to do. !! 718 * dput(), and we need to not race with that.
>> 719 *
>> 720 * However, there is a very special and common case
>> 721 * where we don't care, because there is nothing to
>> 722 * do: the dentry is still hashed, it does not have
>> 723 * a 'delete' op, and it's referenced and already on
>> 724 * the LRU list.
>> 725 *
>> 726 * NOTE! Since we aren't locked, these values are
>> 727 * not "stable". However, it is sufficient that at
>> 728 * some point after we dropped the reference the
>> 729 * dentry was hashed and the flags had the proper
>> 730 * value. Other dentry users may have re-gotten
>> 731 * a reference to the dentry and change that, but
>> 732 * our work is done - we can leave the dentry
>> 733 * around with a zero refcount.
793 */ 734 */
794 if (retain_dentry(dentry, false)) !! 735 smp_rmb();
795 return true; !! 736 d_flags = READ_ONCE(dentry->d_flags);
>> 737 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
>> 738
>> 739 /* Nothing to do? Dropping the reference was all we needed? */
>> 740 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
>> 741 return 1;
796 742
797 /* 743 /*
798 * Either not worth retaining or we ca !! 744 * Not the fast normal case? Get the lock. We've already decremented
799 * Get the lock, then. We've already !! 745 * the refcount, but we'll need to re-check the situation after
800 * but we'll need to re-check the situ !! 746 * getting the lock.
801 */ 747 */
802 spin_lock(&dentry->d_lock); 748 spin_lock(&dentry->d_lock);
803 749
804 /* 750 /*
805 * Did somebody else grab a reference 751 * Did somebody else grab a reference to it in the meantime, and
806 * we're no longer the last user after 752 * we're no longer the last user after all? Alternatively, somebody
807 * else could have killed it and marke 753 * else could have killed it and marked it dead. Either way, we
808 * don't need to do anything else. 754 * don't need to do anything else.
809 */ 755 */
810 locked: !! 756 if (dentry->d_lockref.count) {
811 if (dentry->d_lockref.count || retain_ <<
812 spin_unlock(&dentry->d_lock); 757 spin_unlock(&dentry->d_lock);
813 return true; !! 758 return 1;
814 } 759 }
815 return false; !! 760
>> 761 /*
>> 762 * Re-get the reference we optimistically dropped. We hold the
>> 763 * lock, and we just tested that it was zero, so we can just
>> 764 * set it to 1.
>> 765 */
>> 766 dentry->d_lockref.count = 1;
>> 767 return 0;
816 } 768 }
817 769
818 770
819 /* 771 /*
820 * This is dput 772 * This is dput
821 * 773 *
822 * This is complicated by the fact that we do 774 * This is complicated by the fact that we do not want to put
823 * dentries that are no longer on any hash cha 775 * dentries that are no longer on any hash chain on the unused
824 * list: we'd much rather just get rid of them 776 * list: we'd much rather just get rid of them immediately.
825 * 777 *
826 * However, that implies that we have to trave 778 * However, that implies that we have to traverse the dentry
827 * tree upwards to the parents which might _al 779 * tree upwards to the parents which might _also_ now be
828 * scheduled for deletion (it may have been on 780 * scheduled for deletion (it may have been only waiting for
829 * its last child to go away). 781 * its last child to go away).
830 * 782 *
831 * This tail recursion is done by hand as we d 783 * This tail recursion is done by hand as we don't want to depend
832 * on the compiler to always get this right (g 784 * on the compiler to always get this right (gcc generally doesn't).
833 * Real recursion would eat up our stack space 785 * Real recursion would eat up our stack space.
834 */ 786 */
835 787
836 /* 788 /*
837 * dput - release a dentry 789 * dput - release a dentry
838 * @dentry: dentry to release 790 * @dentry: dentry to release
839 * 791 *
840 * Release a dentry. This will drop the usage 792 * Release a dentry. This will drop the usage count and if appropriate
841 * call the dentry unlink method as well as re 793 * call the dentry unlink method as well as removing it from the queues and
842 * releasing its resources. If the parent dent 794 * releasing its resources. If the parent dentries were scheduled for release
843 * they too may now get deleted. 795 * they too may now get deleted.
844 */ 796 */
845 void dput(struct dentry *dentry) 797 void dput(struct dentry *dentry)
846 { 798 {
847 if (!dentry) !! 799 if (unlikely(!dentry))
848 return; 800 return;
>> 801
>> 802 repeat:
849 might_sleep(); 803 might_sleep();
>> 804
850 rcu_read_lock(); 805 rcu_read_lock();
851 if (likely(fast_dput(dentry))) { 806 if (likely(fast_dput(dentry))) {
852 rcu_read_unlock(); 807 rcu_read_unlock();
853 return; 808 return;
854 } 809 }
855 while (lock_for_kill(dentry)) { !! 810
856 rcu_read_unlock(); !! 811 /* Slow case: now with the dentry lock held */
857 dentry = __dentry_kill(dentry) <<
858 if (!dentry) <<
859 return; <<
860 if (retain_dentry(dentry, true <<
861 spin_unlock(&dentry->d <<
862 return; <<
863 } <<
864 rcu_read_lock(); <<
865 } <<
866 rcu_read_unlock(); 812 rcu_read_unlock();
>> 813
>> 814 WARN_ON(d_in_lookup(dentry));
>> 815
>> 816 /* Unreachable? Get rid of it */
>> 817 if (unlikely(d_unhashed(dentry)))
>> 818 goto kill_it;
>> 819
>> 820 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
>> 821 goto kill_it;
>> 822
>> 823 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
>> 824 if (dentry->d_op->d_delete(dentry))
>> 825 goto kill_it;
>> 826 }
>> 827
>> 828 dentry_lru_add(dentry);
>> 829
>> 830 dentry->d_lockref.count--;
867 spin_unlock(&dentry->d_lock); 831 spin_unlock(&dentry->d_lock);
>> 832 return;
>> 833
>> 834 kill_it:
>> 835 dentry = dentry_kill(dentry);
>> 836 if (dentry) {
>> 837 cond_resched();
>> 838 goto repeat;
>> 839 }
868 } 840 }
869 EXPORT_SYMBOL(dput); 841 EXPORT_SYMBOL(dput);
870 842
871 static void to_shrink_list(struct dentry *dent !! 843
872 __must_hold(&dentry->d_lock) !! 844 /* This must be called with d_lock held */
>> 845 static inline void __dget_dlock(struct dentry *dentry)
873 { 846 {
874 if (!(dentry->d_flags & DCACHE_SHRINK_ !! 847 dentry->d_lockref.count++;
875 if (dentry->d_flags & DCACHE_L <<
876 d_lru_del(dentry); <<
877 d_shrink_add(dentry, list); <<
878 } <<
879 } 848 }
880 849
881 void dput_to_list(struct dentry *dentry, struc !! 850 static inline void __dget(struct dentry *dentry)
882 { 851 {
883 rcu_read_lock(); !! 852 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 } 853 }
892 854
893 struct dentry *dget_parent(struct dentry *dent 855 struct dentry *dget_parent(struct dentry *dentry)
894 { 856 {
895 int gotref; 857 int gotref;
896 struct dentry *ret; 858 struct dentry *ret;
897 unsigned seq; <<
898 859
899 /* 860 /*
900 * Do optimistic parent lookup without 861 * Do optimistic parent lookup without any
901 * locking. 862 * locking.
902 */ 863 */
903 rcu_read_lock(); 864 rcu_read_lock();
904 seq = raw_seqcount_begin(&dentry->d_se <<
905 ret = READ_ONCE(dentry->d_parent); 865 ret = READ_ONCE(dentry->d_parent);
906 gotref = lockref_get_not_zero(&ret->d_ 866 gotref = lockref_get_not_zero(&ret->d_lockref);
907 rcu_read_unlock(); 867 rcu_read_unlock();
908 if (likely(gotref)) { 868 if (likely(gotref)) {
909 if (!read_seqcount_retry(&dent !! 869 if (likely(ret == READ_ONCE(dentry->d_parent)))
910 return ret; 870 return ret;
911 dput(ret); 871 dput(ret);
912 } 872 }
913 873
914 repeat: 874 repeat:
915 /* 875 /*
916 * Don't need rcu_dereference because 876 * Don't need rcu_dereference because we re-check it was correct under
917 * the lock. 877 * the lock.
918 */ 878 */
919 rcu_read_lock(); 879 rcu_read_lock();
920 ret = dentry->d_parent; 880 ret = dentry->d_parent;
921 spin_lock(&ret->d_lock); 881 spin_lock(&ret->d_lock);
922 if (unlikely(ret != dentry->d_parent)) 882 if (unlikely(ret != dentry->d_parent)) {
923 spin_unlock(&ret->d_lock); 883 spin_unlock(&ret->d_lock);
924 rcu_read_unlock(); 884 rcu_read_unlock();
925 goto repeat; 885 goto repeat;
926 } 886 }
927 rcu_read_unlock(); 887 rcu_read_unlock();
928 BUG_ON(!ret->d_lockref.count); 888 BUG_ON(!ret->d_lockref.count);
929 ret->d_lockref.count++; 889 ret->d_lockref.count++;
930 spin_unlock(&ret->d_lock); 890 spin_unlock(&ret->d_lock);
931 return ret; 891 return ret;
932 } 892 }
933 EXPORT_SYMBOL(dget_parent); 893 EXPORT_SYMBOL(dget_parent);
934 894
935 static struct dentry * __d_find_any_alias(stru <<
936 { <<
937 struct dentry *alias; <<
938 <<
939 if (hlist_empty(&inode->i_dentry)) <<
940 return NULL; <<
941 alias = hlist_entry(inode->i_dentry.fi <<
942 lockref_get(&alias->d_lockref); <<
943 return alias; <<
944 } <<
945 <<
946 /** 895 /**
947 * d_find_any_alias - find any alias for a giv !! 896 * d_find_alias - grab a hashed alias of inode
948 * @inode: inode to find an alias for !! 897 * @inode: inode in question
949 * 898 *
950 * If any aliases exist for the given inode, t !! 899 * If inode has a hashed alias, or is a directory and has any alias,
951 * reference for one of them. If no aliases e !! 900 * acquire the reference to alias and return it. Otherwise return NULL.
>> 901 * Notice that if inode is a directory there can be only one alias and
>> 902 * it can be unhashed only if it has no children, or if it is the root
>> 903 * of a filesystem, or if the directory was renamed and d_revalidate
>> 904 * was the first vfs operation to notice.
>> 905 *
>> 906 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
>> 907 * any other hashed alias over that one.
952 */ 908 */
953 struct dentry *d_find_any_alias(struct inode * <<
954 { <<
955 struct dentry *de; <<
956 <<
957 spin_lock(&inode->i_lock); <<
958 de = __d_find_any_alias(inode); <<
959 spin_unlock(&inode->i_lock); <<
960 return de; <<
961 } <<
962 EXPORT_SYMBOL(d_find_any_alias); <<
963 <<
964 static struct dentry *__d_find_alias(struct in 909 static struct dentry *__d_find_alias(struct inode *inode)
965 { 910 {
966 struct dentry *alias; !! 911 struct dentry *alias, *discon_alias;
967 <<
968 if (S_ISDIR(inode->i_mode)) <<
969 return __d_find_any_alias(inod <<
970 912
>> 913 again:
>> 914 discon_alias = NULL;
971 hlist_for_each_entry(alias, &inode->i_ 915 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
972 spin_lock(&alias->d_lock); 916 spin_lock(&alias->d_lock);
973 if (!d_unhashed(alias)) { !! 917 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
974 dget_dlock(alias); !! 918 if (IS_ROOT(alias) &&
>> 919 (alias->d_flags & DCACHE_DISCONNECTED)) {
>> 920 discon_alias = alias;
>> 921 } else {
>> 922 __dget_dlock(alias);
>> 923 spin_unlock(&alias->d_lock);
>> 924 return alias;
>> 925 }
>> 926 }
>> 927 spin_unlock(&alias->d_lock);
>> 928 }
>> 929 if (discon_alias) {
>> 930 alias = discon_alias;
>> 931 spin_lock(&alias->d_lock);
>> 932 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
>> 933 __dget_dlock(alias);
975 spin_unlock(&alias->d_ 934 spin_unlock(&alias->d_lock);
976 return alias; 935 return alias;
977 } 936 }
978 spin_unlock(&alias->d_lock); 937 spin_unlock(&alias->d_lock);
>> 938 goto again;
979 } 939 }
980 return NULL; 940 return NULL;
981 } 941 }
982 942
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 943 struct dentry *d_find_alias(struct inode *inode)
998 { 944 {
999 struct dentry *de = NULL; 945 struct dentry *de = NULL;
1000 946
1001 if (!hlist_empty(&inode->i_dentry)) { 947 if (!hlist_empty(&inode->i_dentry)) {
1002 spin_lock(&inode->i_lock); 948 spin_lock(&inode->i_lock);
1003 de = __d_find_alias(inode); 949 de = __d_find_alias(inode);
1004 spin_unlock(&inode->i_lock); 950 spin_unlock(&inode->i_lock);
1005 } 951 }
1006 return de; 952 return de;
1007 } 953 }
1008 EXPORT_SYMBOL(d_find_alias); 954 EXPORT_SYMBOL(d_find_alias);
1009 955
1010 /* 956 /*
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 957 * Try to kill dentries associated with this inode.
1037 * WARNING: you must own a reference to inode 958 * WARNING: you must own a reference to inode.
1038 */ 959 */
1039 void d_prune_aliases(struct inode *inode) 960 void d_prune_aliases(struct inode *inode)
1040 { 961 {
1041 LIST_HEAD(dispose); <<
1042 struct dentry *dentry; 962 struct dentry *dentry;
1043 !! 963 restart:
1044 spin_lock(&inode->i_lock); 964 spin_lock(&inode->i_lock);
1045 hlist_for_each_entry(dentry, &inode-> 965 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1046 spin_lock(&dentry->d_lock); 966 spin_lock(&dentry->d_lock);
1047 if (!dentry->d_lockref.count) !! 967 if (!dentry->d_lockref.count) {
1048 to_shrink_list(dentry !! 968 struct dentry *parent = lock_parent(dentry);
>> 969 if (likely(!dentry->d_lockref.count)) {
>> 970 __dentry_kill(dentry);
>> 971 dput(parent);
>> 972 goto restart;
>> 973 }
>> 974 if (parent)
>> 975 spin_unlock(&parent->d_lock);
>> 976 }
1049 spin_unlock(&dentry->d_lock); 977 spin_unlock(&dentry->d_lock);
1050 } 978 }
1051 spin_unlock(&inode->i_lock); 979 spin_unlock(&inode->i_lock);
1052 shrink_dentry_list(&dispose); <<
1053 } 980 }
1054 EXPORT_SYMBOL(d_prune_aliases); 981 EXPORT_SYMBOL(d_prune_aliases);
1055 982
1056 static inline void shrink_kill(struct dentry !! 983 static void shrink_dentry_list(struct list_head *list)
1057 { 984 {
1058 do { !! 985 struct dentry *dentry, *parent;
1059 rcu_read_unlock(); <<
1060 victim = __dentry_kill(victim <<
1061 rcu_read_lock(); <<
1062 } while (victim && lock_for_kill(vict <<
1063 rcu_read_unlock(); <<
1064 if (victim) <<
1065 spin_unlock(&victim->d_lock); <<
1066 } <<
1067 986
1068 void shrink_dentry_list(struct list_head *lis <<
1069 { <<
1070 while (!list_empty(list)) { 987 while (!list_empty(list)) {
1071 struct dentry *dentry; !! 988 struct inode *inode;
1072 <<
1073 dentry = list_entry(list->pre 989 dentry = list_entry(list->prev, struct dentry, d_lru);
1074 spin_lock(&dentry->d_lock); 990 spin_lock(&dentry->d_lock);
1075 rcu_read_lock(); !! 991 parent = lock_parent(dentry);
1076 if (!lock_for_kill(dentry)) { !! 992
1077 bool can_free; !! 993 /*
1078 rcu_read_unlock(); !! 994 * The dispose list is isolated and dentries are not accounted
1079 d_shrink_del(dentry); !! 995 * to the LRU here, so we can simply remove it from the list
1080 can_free = dentry->d_ !! 996 * here regardless of whether it is referenced or not.
>> 997 */
>> 998 d_shrink_del(dentry);
>> 999
>> 1000 /*
>> 1001 * We found an inuse dentry which was not removed from
>> 1002 * the LRU because of laziness during lookup. Do not free it.
>> 1003 */
>> 1004 if (dentry->d_lockref.count > 0) {
1081 spin_unlock(&dentry-> 1005 spin_unlock(&dentry->d_lock);
>> 1006 if (parent)
>> 1007 spin_unlock(&parent->d_lock);
>> 1008 continue;
>> 1009 }
>> 1010
>> 1011
>> 1012 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
>> 1013 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
>> 1014 spin_unlock(&dentry->d_lock);
>> 1015 if (parent)
>> 1016 spin_unlock(&parent->d_lock);
1082 if (can_free) 1017 if (can_free)
1083 dentry_free(d 1018 dentry_free(dentry);
1084 continue; 1019 continue;
1085 } 1020 }
1086 d_shrink_del(dentry); !! 1021
1087 shrink_kill(dentry); !! 1022 inode = dentry->d_inode;
>> 1023 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
>> 1024 d_shrink_add(dentry, list);
>> 1025 spin_unlock(&dentry->d_lock);
>> 1026 if (parent)
>> 1027 spin_unlock(&parent->d_lock);
>> 1028 continue;
>> 1029 }
>> 1030
>> 1031 __dentry_kill(dentry);
>> 1032
>> 1033 /*
>> 1034 * We need to prune ancestors too. This is necessary to prevent
>> 1035 * quadratic behavior of shrink_dcache_parent(), but is also
>> 1036 * expected to be beneficial in reducing dentry cache
>> 1037 * fragmentation.
>> 1038 */
>> 1039 dentry = parent;
>> 1040 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
>> 1041 parent = lock_parent(dentry);
>> 1042 if (dentry->d_lockref.count != 1) {
>> 1043 dentry->d_lockref.count--;
>> 1044 spin_unlock(&dentry->d_lock);
>> 1045 if (parent)
>> 1046 spin_unlock(&parent->d_lock);
>> 1047 break;
>> 1048 }
>> 1049 inode = dentry->d_inode; /* can't be NULL */
>> 1050 if (unlikely(!spin_trylock(&inode->i_lock))) {
>> 1051 spin_unlock(&dentry->d_lock);
>> 1052 if (parent)
>> 1053 spin_unlock(&parent->d_lock);
>> 1054 cpu_relax();
>> 1055 continue;
>> 1056 }
>> 1057 __dentry_kill(dentry);
>> 1058 dentry = parent;
>> 1059 }
1088 } 1060 }
1089 } 1061 }
1090 1062
1091 static enum lru_status dentry_lru_isolate(str 1063 static enum lru_status dentry_lru_isolate(struct list_head *item,
1092 struct list_lru_one *lru, spi 1064 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1093 { 1065 {
1094 struct list_head *freeable = arg; 1066 struct list_head *freeable = arg;
1095 struct dentry *dentry = container_o 1067 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1096 1068
1097 1069
1098 /* 1070 /*
1099 * we are inverting the lru lock/dent 1071 * we are inverting the lru lock/dentry->d_lock here,
1100 * so use a trylock. If we fail to ge 1072 * so use a trylock. If we fail to get the lock, just skip
1101 * it 1073 * it
1102 */ 1074 */
1103 if (!spin_trylock(&dentry->d_lock)) 1075 if (!spin_trylock(&dentry->d_lock))
1104 return LRU_SKIP; 1076 return LRU_SKIP;
1105 1077
1106 /* 1078 /*
1107 * Referenced dentries are still in u 1079 * Referenced dentries are still in use. If they have active
1108 * counts, just remove them from the 1080 * counts, just remove them from the LRU. Otherwise give them
1109 * another pass through the LRU. 1081 * another pass through the LRU.
1110 */ 1082 */
1111 if (dentry->d_lockref.count) { 1083 if (dentry->d_lockref.count) {
1112 d_lru_isolate(lru, dentry); 1084 d_lru_isolate(lru, dentry);
1113 spin_unlock(&dentry->d_lock); 1085 spin_unlock(&dentry->d_lock);
1114 return LRU_REMOVED; 1086 return LRU_REMOVED;
1115 } 1087 }
1116 1088
1117 if (dentry->d_flags & DCACHE_REFERENC 1089 if (dentry->d_flags & DCACHE_REFERENCED) {
1118 dentry->d_flags &= ~DCACHE_RE 1090 dentry->d_flags &= ~DCACHE_REFERENCED;
1119 spin_unlock(&dentry->d_lock); 1091 spin_unlock(&dentry->d_lock);
1120 1092
1121 /* 1093 /*
1122 * The list move itself will 1094 * The list move itself will be made by the common LRU code. At
1123 * this point, we've dropped 1095 * this point, we've dropped the dentry->d_lock but keep the
1124 * lru lock. This is safe to 1096 * lru lock. This is safe to do, since every list movement is
1125 * protected by the lru lock 1097 * protected by the lru lock even if both locks are held.
1126 * 1098 *
1127 * This is guaranteed by the 1099 * This is guaranteed by the fact that all LRU management
1128 * functions are intermediate 1100 * functions are intermediated by the LRU API calls like
1129 * list_lru_add_obj and list_ !! 1101 * list_lru_add and list_lru_del. List movement in this file
1130 * only ever occur through th 1102 * only ever occur through this functions or through callbacks
1131 * like this one, that are ca 1103 * like this one, that are called from the LRU API.
1132 * 1104 *
1133 * The only exceptions to thi 1105 * The only exceptions to this are functions like
1134 * shrink_dentry_list, and co 1106 * shrink_dentry_list, and code that first checks for the
1135 * DCACHE_SHRINK_LIST flag. 1107 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1136 * operating only with stack 1108 * operating only with stack provided lists after they are
1137 * properly isolated from the 1109 * properly isolated from the main list. It is thus, always a
1138 * local access. 1110 * local access.
1139 */ 1111 */
1140 return LRU_ROTATE; 1112 return LRU_ROTATE;
1141 } 1113 }
1142 1114
1143 d_lru_shrink_move(lru, dentry, freeab 1115 d_lru_shrink_move(lru, dentry, freeable);
1144 spin_unlock(&dentry->d_lock); 1116 spin_unlock(&dentry->d_lock);
1145 1117
1146 return LRU_REMOVED; 1118 return LRU_REMOVED;
1147 } 1119 }
1148 1120
1149 /** 1121 /**
1150 * prune_dcache_sb - shrink the dcache 1122 * prune_dcache_sb - shrink the dcache
1151 * @sb: superblock 1123 * @sb: superblock
1152 * @sc: shrink control, passed to list_lru_sh 1124 * @sc: shrink control, passed to list_lru_shrink_walk()
1153 * 1125 *
1154 * Attempt to shrink the superblock dcache LR 1126 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1155 * is done when we need more memory and calle 1127 * is done when we need more memory and called from the superblock shrinker
1156 * function. 1128 * function.
1157 * 1129 *
1158 * This function may fail to free any resourc 1130 * This function may fail to free any resources if all the dentries are in
1159 * use. 1131 * use.
1160 */ 1132 */
1161 long prune_dcache_sb(struct super_block *sb, 1133 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1162 { 1134 {
1163 LIST_HEAD(dispose); 1135 LIST_HEAD(dispose);
1164 long freed; 1136 long freed;
1165 1137
1166 freed = list_lru_shrink_walk(&sb->s_d 1138 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1167 dentry_l 1139 dentry_lru_isolate, &dispose);
1168 shrink_dentry_list(&dispose); 1140 shrink_dentry_list(&dispose);
1169 return freed; 1141 return freed;
1170 } 1142 }
1171 1143
1172 static enum lru_status dentry_lru_isolate_shr 1144 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1173 struct list_lru_one *lru, spi 1145 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1174 { 1146 {
1175 struct list_head *freeable = arg; 1147 struct list_head *freeable = arg;
1176 struct dentry *dentry = container_o 1148 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1177 1149
1178 /* 1150 /*
1179 * we are inverting the lru lock/dent 1151 * we are inverting the lru lock/dentry->d_lock here,
1180 * so use a trylock. If we fail to ge 1152 * so use a trylock. If we fail to get the lock, just skip
1181 * it 1153 * it
1182 */ 1154 */
1183 if (!spin_trylock(&dentry->d_lock)) 1155 if (!spin_trylock(&dentry->d_lock))
1184 return LRU_SKIP; 1156 return LRU_SKIP;
1185 1157
1186 d_lru_shrink_move(lru, dentry, freeab 1158 d_lru_shrink_move(lru, dentry, freeable);
1187 spin_unlock(&dentry->d_lock); 1159 spin_unlock(&dentry->d_lock);
1188 1160
1189 return LRU_REMOVED; 1161 return LRU_REMOVED;
1190 } 1162 }
1191 1163
1192 1164
1193 /** 1165 /**
1194 * shrink_dcache_sb - shrink dcache for a sup 1166 * shrink_dcache_sb - shrink dcache for a superblock
1195 * @sb: superblock 1167 * @sb: superblock
1196 * 1168 *
1197 * Shrink the dcache for the specified super 1169 * Shrink the dcache for the specified super block. This is used to free
1198 * the dcache before unmounting a file system 1170 * the dcache before unmounting a file system.
1199 */ 1171 */
1200 void shrink_dcache_sb(struct super_block *sb) 1172 void shrink_dcache_sb(struct super_block *sb)
1201 { 1173 {
>> 1174 long freed;
>> 1175
1202 do { 1176 do {
1203 LIST_HEAD(dispose); 1177 LIST_HEAD(dispose);
1204 1178
1205 list_lru_walk(&sb->s_dentry_l !! 1179 freed = list_lru_walk(&sb->s_dentry_lru,
1206 dentry_lru_isolate_sh 1180 dentry_lru_isolate_shrink, &dispose, 1024);
>> 1181
>> 1182 this_cpu_sub(nr_dentry_unused, freed);
1207 shrink_dentry_list(&dispose); 1183 shrink_dentry_list(&dispose);
>> 1184 cond_resched();
1208 } while (list_lru_count(&sb->s_dentry 1185 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1209 } 1186 }
1210 EXPORT_SYMBOL(shrink_dcache_sb); 1187 EXPORT_SYMBOL(shrink_dcache_sb);
1211 1188
1212 /** 1189 /**
1213 * enum d_walk_ret - action to talke during t 1190 * enum d_walk_ret - action to talke during tree walk
1214 * @D_WALK_CONTINUE: contrinue walk 1191 * @D_WALK_CONTINUE: contrinue walk
1215 * @D_WALK_QUIT: quit walk 1192 * @D_WALK_QUIT: quit walk
1216 * @D_WALK_NORETRY: quit when retry is ne 1193 * @D_WALK_NORETRY: quit when retry is needed
1217 * @D_WALK_SKIP: skip this dentry and 1194 * @D_WALK_SKIP: skip this dentry and its children
1218 */ 1195 */
1219 enum d_walk_ret { 1196 enum d_walk_ret {
1220 D_WALK_CONTINUE, 1197 D_WALK_CONTINUE,
1221 D_WALK_QUIT, 1198 D_WALK_QUIT,
1222 D_WALK_NORETRY, 1199 D_WALK_NORETRY,
1223 D_WALK_SKIP, 1200 D_WALK_SKIP,
1224 }; 1201 };
1225 1202
1226 /** 1203 /**
1227 * d_walk - walk the dentry tree 1204 * d_walk - walk the dentry tree
1228 * @parent: start of walk 1205 * @parent: start of walk
1229 * @data: data passed to @enter() and @ 1206 * @data: data passed to @enter() and @finish()
1230 * @enter: callback when first entering 1207 * @enter: callback when first entering the dentry
>> 1208 * @finish: callback when successfully finished the walk
1231 * 1209 *
1232 * The @enter() callbacks are called with d_l !! 1210 * The @enter() and @finish() callbacks are called with d_lock held.
1233 */ 1211 */
1234 static void d_walk(struct dentry *parent, voi 1212 static void d_walk(struct dentry *parent, void *data,
1235 enum d_walk_ret (*enter)(v !! 1213 enum d_walk_ret (*enter)(void *, struct dentry *),
>> 1214 void (*finish)(void *))
1236 { 1215 {
1237 struct dentry *this_parent, *dentry; !! 1216 struct dentry *this_parent;
>> 1217 struct list_head *next;
1238 unsigned seq = 0; 1218 unsigned seq = 0;
1239 enum d_walk_ret ret; 1219 enum d_walk_ret ret;
1240 bool retry = true; 1220 bool retry = true;
1241 1221
1242 again: 1222 again:
1243 read_seqbegin_or_lock(&rename_lock, & 1223 read_seqbegin_or_lock(&rename_lock, &seq);
1244 this_parent = parent; 1224 this_parent = parent;
1245 spin_lock(&this_parent->d_lock); 1225 spin_lock(&this_parent->d_lock);
1246 1226
1247 ret = enter(data, this_parent); 1227 ret = enter(data, this_parent);
1248 switch (ret) { 1228 switch (ret) {
1249 case D_WALK_CONTINUE: 1229 case D_WALK_CONTINUE:
1250 break; 1230 break;
1251 case D_WALK_QUIT: 1231 case D_WALK_QUIT:
1252 case D_WALK_SKIP: 1232 case D_WALK_SKIP:
1253 goto out_unlock; 1233 goto out_unlock;
1254 case D_WALK_NORETRY: 1234 case D_WALK_NORETRY:
1255 retry = false; 1235 retry = false;
1256 break; 1236 break;
1257 } 1237 }
1258 repeat: 1238 repeat:
1259 dentry = d_first_child(this_parent); !! 1239 next = this_parent->d_subdirs.next;
1260 resume: 1240 resume:
1261 hlist_for_each_entry_from(dentry, d_s !! 1241 while (next != &this_parent->d_subdirs) {
>> 1242 struct list_head *tmp = next;
>> 1243 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
>> 1244 next = tmp->next;
>> 1245
1262 if (unlikely(dentry->d_flags 1246 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1263 continue; 1247 continue;
1264 1248
1265 spin_lock_nested(&dentry->d_l 1249 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1266 1250
1267 ret = enter(data, dentry); 1251 ret = enter(data, dentry);
1268 switch (ret) { 1252 switch (ret) {
1269 case D_WALK_CONTINUE: 1253 case D_WALK_CONTINUE:
1270 break; 1254 break;
1271 case D_WALK_QUIT: 1255 case D_WALK_QUIT:
1272 spin_unlock(&dentry-> 1256 spin_unlock(&dentry->d_lock);
1273 goto out_unlock; 1257 goto out_unlock;
1274 case D_WALK_NORETRY: 1258 case D_WALK_NORETRY:
1275 retry = false; 1259 retry = false;
1276 break; 1260 break;
1277 case D_WALK_SKIP: 1261 case D_WALK_SKIP:
1278 spin_unlock(&dentry-> 1262 spin_unlock(&dentry->d_lock);
1279 continue; 1263 continue;
1280 } 1264 }
1281 1265
1282 if (!hlist_empty(&dentry->d_c !! 1266 if (!list_empty(&dentry->d_subdirs)) {
1283 spin_unlock(&this_par 1267 spin_unlock(&this_parent->d_lock);
1284 spin_release(&dentry- !! 1268 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1285 this_parent = dentry; 1269 this_parent = dentry;
1286 spin_acquire(&this_pa 1270 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1287 goto repeat; 1271 goto repeat;
1288 } 1272 }
1289 spin_unlock(&dentry->d_lock); 1273 spin_unlock(&dentry->d_lock);
1290 } 1274 }
1291 /* 1275 /*
1292 * All done at this level ... ascend 1276 * All done at this level ... ascend and resume the search.
1293 */ 1277 */
1294 rcu_read_lock(); 1278 rcu_read_lock();
1295 ascend: 1279 ascend:
1296 if (this_parent != parent) { 1280 if (this_parent != parent) {
1297 dentry = this_parent; !! 1281 struct dentry *child = this_parent;
1298 this_parent = dentry->d_paren !! 1282 this_parent = child->d_parent;
1299 1283
1300 spin_unlock(&dentry->d_lock); !! 1284 spin_unlock(&child->d_lock);
1301 spin_lock(&this_parent->d_loc 1285 spin_lock(&this_parent->d_lock);
1302 1286
1303 /* might go back up the wrong 1287 /* might go back up the wrong parent if we have had a rename. */
1304 if (need_seqretry(&rename_loc 1288 if (need_seqretry(&rename_lock, seq))
1305 goto rename_retry; 1289 goto rename_retry;
1306 /* go into the first sibling 1290 /* go into the first sibling still alive */
1307 hlist_for_each_entry_continue !! 1291 do {
1308 if (likely(!(dentry-> !! 1292 next = child->d_child.next;
1309 rcu_read_unlo !! 1293 if (next == &this_parent->d_subdirs)
1310 goto resume; !! 1294 goto ascend;
1311 } !! 1295 child = list_entry(next, struct dentry, d_child);
1312 } !! 1296 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1313 goto ascend; !! 1297 rcu_read_unlock();
>> 1298 goto resume;
1314 } 1299 }
1315 if (need_seqretry(&rename_lock, seq)) 1300 if (need_seqretry(&rename_lock, seq))
1316 goto rename_retry; 1301 goto rename_retry;
1317 rcu_read_unlock(); 1302 rcu_read_unlock();
>> 1303 if (finish)
>> 1304 finish(data);
1318 1305
1319 out_unlock: 1306 out_unlock:
1320 spin_unlock(&this_parent->d_lock); 1307 spin_unlock(&this_parent->d_lock);
1321 done_seqretry(&rename_lock, seq); 1308 done_seqretry(&rename_lock, seq);
1322 return; 1309 return;
1323 1310
1324 rename_retry: 1311 rename_retry:
1325 spin_unlock(&this_parent->d_lock); 1312 spin_unlock(&this_parent->d_lock);
1326 rcu_read_unlock(); 1313 rcu_read_unlock();
1327 BUG_ON(seq & 1); 1314 BUG_ON(seq & 1);
1328 if (!retry) 1315 if (!retry)
1329 return; 1316 return;
1330 seq = 1; 1317 seq = 1;
1331 goto again; 1318 goto again;
1332 } 1319 }
1333 1320
1334 struct check_mount { 1321 struct check_mount {
1335 struct vfsmount *mnt; 1322 struct vfsmount *mnt;
1336 unsigned int mounted; 1323 unsigned int mounted;
1337 }; 1324 };
1338 1325
1339 static enum d_walk_ret path_check_mount(void 1326 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1340 { 1327 {
1341 struct check_mount *info = data; 1328 struct check_mount *info = data;
1342 struct path path = { .mnt = info->mnt 1329 struct path path = { .mnt = info->mnt, .dentry = dentry };
1343 1330
1344 if (likely(!d_mountpoint(dentry))) 1331 if (likely(!d_mountpoint(dentry)))
1345 return D_WALK_CONTINUE; 1332 return D_WALK_CONTINUE;
1346 if (__path_is_mountpoint(&path)) { 1333 if (__path_is_mountpoint(&path)) {
1347 info->mounted = 1; 1334 info->mounted = 1;
1348 return D_WALK_QUIT; 1335 return D_WALK_QUIT;
1349 } 1336 }
1350 return D_WALK_CONTINUE; 1337 return D_WALK_CONTINUE;
1351 } 1338 }
1352 1339
1353 /** 1340 /**
1354 * path_has_submounts - check for mounts over 1341 * path_has_submounts - check for mounts over a dentry in the
1355 * current namespace. 1342 * current namespace.
1356 * @parent: path to check. 1343 * @parent: path to check.
1357 * 1344 *
1358 * Return true if the parent or its subdirect 1345 * Return true if the parent or its subdirectories contain
1359 * a mount point in the current namespace. 1346 * a mount point in the current namespace.
1360 */ 1347 */
1361 int path_has_submounts(const struct path *par 1348 int path_has_submounts(const struct path *parent)
1362 { 1349 {
1363 struct check_mount data = { .mnt = pa 1350 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1364 1351
1365 read_seqlock_excl(&mount_lock); 1352 read_seqlock_excl(&mount_lock);
1366 d_walk(parent->dentry, &data, path_ch !! 1353 d_walk(parent->dentry, &data, path_check_mount, NULL);
1367 read_sequnlock_excl(&mount_lock); 1354 read_sequnlock_excl(&mount_lock);
1368 1355
1369 return data.mounted; 1356 return data.mounted;
1370 } 1357 }
1371 EXPORT_SYMBOL(path_has_submounts); 1358 EXPORT_SYMBOL(path_has_submounts);
1372 1359
1373 /* 1360 /*
1374 * Called by mount code to set a mountpoint a 1361 * Called by mount code to set a mountpoint and check if the mountpoint is
1375 * reachable (e.g. NFS can unhash a directory 1362 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1376 * subtree can become unreachable). 1363 * subtree can become unreachable).
1377 * 1364 *
1378 * Only one of d_invalidate() and d_set_mount 1365 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1379 * this reason take rename_lock and d_lock on 1366 * this reason take rename_lock and d_lock on dentry and ancestors.
1380 */ 1367 */
1381 int d_set_mounted(struct dentry *dentry) 1368 int d_set_mounted(struct dentry *dentry)
1382 { 1369 {
1383 struct dentry *p; 1370 struct dentry *p;
1384 int ret = -ENOENT; 1371 int ret = -ENOENT;
1385 write_seqlock(&rename_lock); 1372 write_seqlock(&rename_lock);
1386 for (p = dentry->d_parent; !IS_ROOT(p 1373 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1387 /* Need exclusion wrt. d_inva 1374 /* Need exclusion wrt. d_invalidate() */
1388 spin_lock(&p->d_lock); 1375 spin_lock(&p->d_lock);
1389 if (unlikely(d_unhashed(p))) 1376 if (unlikely(d_unhashed(p))) {
1390 spin_unlock(&p->d_loc 1377 spin_unlock(&p->d_lock);
1391 goto out; 1378 goto out;
1392 } 1379 }
1393 spin_unlock(&p->d_lock); 1380 spin_unlock(&p->d_lock);
1394 } 1381 }
1395 spin_lock(&dentry->d_lock); 1382 spin_lock(&dentry->d_lock);
1396 if (!d_unlinked(dentry)) { 1383 if (!d_unlinked(dentry)) {
1397 ret = -EBUSY; 1384 ret = -EBUSY;
1398 if (!d_mountpoint(dentry)) { 1385 if (!d_mountpoint(dentry)) {
1399 dentry->d_flags |= DC 1386 dentry->d_flags |= DCACHE_MOUNTED;
1400 ret = 0; 1387 ret = 0;
1401 } 1388 }
1402 } 1389 }
1403 spin_unlock(&dentry->d_lock); 1390 spin_unlock(&dentry->d_lock);
1404 out: 1391 out:
1405 write_sequnlock(&rename_lock); 1392 write_sequnlock(&rename_lock);
1406 return ret; 1393 return ret;
1407 } 1394 }
1408 1395
1409 /* 1396 /*
1410 * Search the dentry child list of the specif 1397 * Search the dentry child list of the specified parent,
1411 * and move any unused dentries to the end of 1398 * and move any unused dentries to the end of the unused
1412 * list for prune_dcache(). We descend to the 1399 * list for prune_dcache(). We descend to the next level
1413 * whenever the d_children list is non-empty !! 1400 * whenever the d_subdirs list is non-empty and continue
1414 * searching. 1401 * searching.
1415 * 1402 *
1416 * It returns zero iff there are no unused ch 1403 * It returns zero iff there are no unused children,
1417 * otherwise it returns the number of childr 1404 * otherwise it returns the number of children moved to
1418 * the end of the unused list. This may not b 1405 * the end of the unused list. This may not be the total
1419 * number of unused children, because select_ 1406 * number of unused children, because select_parent can
1420 * drop the lock and return early due to late 1407 * drop the lock and return early due to latency
1421 * constraints. 1408 * constraints.
1422 */ 1409 */
1423 1410
1424 struct select_data { 1411 struct select_data {
1425 struct dentry *start; 1412 struct dentry *start;
1426 union { <<
1427 long found; <<
1428 struct dentry *victim; <<
1429 }; <<
1430 struct list_head dispose; 1413 struct list_head dispose;
>> 1414 int found;
1431 }; 1415 };
1432 1416
1433 static enum d_walk_ret select_collect(void *_ 1417 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1434 { 1418 {
1435 struct select_data *data = _data; 1419 struct select_data *data = _data;
1436 enum d_walk_ret ret = D_WALK_CONTINUE 1420 enum d_walk_ret ret = D_WALK_CONTINUE;
1437 1421
1438 if (data->start == dentry) 1422 if (data->start == dentry)
1439 goto out; 1423 goto out;
1440 1424
1441 if (dentry->d_flags & DCACHE_SHRINK_L 1425 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1442 data->found++; 1426 data->found++;
1443 } else if (!dentry->d_lockref.count) !! 1427 } else {
1444 to_shrink_list(dentry, &data- !! 1428 if (dentry->d_flags & DCACHE_LRU_LIST)
1445 data->found++; !! 1429 d_lru_del(dentry);
1446 } else if (dentry->d_lockref.count < !! 1430 if (!dentry->d_lockref.count) {
1447 data->found++; !! 1431 d_shrink_add(dentry, &data->dispose);
1448 } !! 1432 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 } 1433 }
1474 to_shrink_list(dentry, &data- <<
1475 } 1434 }
1476 /* 1435 /*
1477 * We can return to the caller if we 1436 * We can return to the caller if we have found some (this
1478 * ensures forward progress). We'll b 1437 * ensures forward progress). We'll be coming back to find
1479 * the rest. 1438 * the rest.
1480 */ 1439 */
1481 if (!list_empty(&data->dispose)) 1440 if (!list_empty(&data->dispose))
1482 ret = need_resched() ? D_WALK 1441 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1483 out: 1442 out:
1484 return ret; 1443 return ret;
1485 } 1444 }
1486 1445
1487 /** 1446 /**
1488 * shrink_dcache_parent - prune dcache 1447 * shrink_dcache_parent - prune dcache
1489 * @parent: parent of entries to prune 1448 * @parent: parent of entries to prune
1490 * 1449 *
1491 * Prune the dcache to remove unused children 1450 * Prune the dcache to remove unused children of the parent dentry.
1492 */ 1451 */
1493 void shrink_dcache_parent(struct dentry *pare 1452 void shrink_dcache_parent(struct dentry *parent)
1494 { 1453 {
1495 for (;;) { 1454 for (;;) {
1496 struct select_data data = {.s !! 1455 struct select_data data;
1497 1456
1498 INIT_LIST_HEAD(&data.dispose) 1457 INIT_LIST_HEAD(&data.dispose);
1499 d_walk(parent, &data, select_ !! 1458 data.start = parent;
1500 !! 1459 data.found = 0;
1501 if (!list_empty(&data.dispose <<
1502 shrink_dentry_list(&d <<
1503 continue; <<
1504 } <<
1505 1460
1506 cond_resched(); !! 1461 d_walk(parent, &data, select_collect, NULL);
1507 if (!data.found) 1462 if (!data.found)
1508 break; 1463 break;
1509 data.victim = NULL; !! 1464
1510 d_walk(parent, &data, select_ !! 1465 shrink_dentry_list(&data.dispose);
1511 if (data.victim) { !! 1466 cond_resched();
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 } 1467 }
1523 } 1468 }
1524 EXPORT_SYMBOL(shrink_dcache_parent); 1469 EXPORT_SYMBOL(shrink_dcache_parent);
1525 1470
1526 static enum d_walk_ret umount_check(void *_da 1471 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1527 { 1472 {
1528 /* it has busy descendents; complain 1473 /* it has busy descendents; complain about those instead */
1529 if (!hlist_empty(&dentry->d_children) !! 1474 if (!list_empty(&dentry->d_subdirs))
1530 return D_WALK_CONTINUE; 1475 return D_WALK_CONTINUE;
1531 1476
1532 /* root with refcount 1 is fine */ 1477 /* root with refcount 1 is fine */
1533 if (dentry == _data && dentry->d_lock 1478 if (dentry == _data && dentry->d_lockref.count == 1)
1534 return D_WALK_CONTINUE; 1479 return D_WALK_CONTINUE;
1535 1480
1536 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} !! 1481 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1537 " still in use (%d) [ 1482 " still in use (%d) [unmount of %s %s]\n",
1538 dentry, 1483 dentry,
1539 dentry->d_inode ? 1484 dentry->d_inode ?
1540 dentry->d_inode->i_ino 1485 dentry->d_inode->i_ino : 0UL,
1541 dentry, 1486 dentry,
1542 dentry->d_lockref.coun 1487 dentry->d_lockref.count,
1543 dentry->d_sb->s_type-> 1488 dentry->d_sb->s_type->name,
1544 dentry->d_sb->s_id); 1489 dentry->d_sb->s_id);
>> 1490 WARN_ON(1);
1545 return D_WALK_CONTINUE; 1491 return D_WALK_CONTINUE;
1546 } 1492 }
1547 1493
1548 static void do_one_tree(struct dentry *dentry 1494 static void do_one_tree(struct dentry *dentry)
1549 { 1495 {
1550 shrink_dcache_parent(dentry); 1496 shrink_dcache_parent(dentry);
1551 d_walk(dentry, dentry, umount_check); !! 1497 d_walk(dentry, dentry, umount_check, NULL);
1552 d_drop(dentry); 1498 d_drop(dentry);
1553 dput(dentry); 1499 dput(dentry);
1554 } 1500 }
1555 1501
1556 /* 1502 /*
1557 * destroy the dentries attached to a superbl 1503 * destroy the dentries attached to a superblock on unmounting
1558 */ 1504 */
1559 void shrink_dcache_for_umount(struct super_bl 1505 void shrink_dcache_for_umount(struct super_block *sb)
1560 { 1506 {
1561 struct dentry *dentry; 1507 struct dentry *dentry;
1562 1508
1563 rwsem_assert_held_write(&sb->s_umount !! 1509 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1564 1510
1565 dentry = sb->s_root; 1511 dentry = sb->s_root;
1566 sb->s_root = NULL; 1512 sb->s_root = NULL;
1567 do_one_tree(dentry); 1513 do_one_tree(dentry);
1568 1514
1569 while (!hlist_bl_empty(&sb->s_roots)) !! 1515 while (!hlist_bl_empty(&sb->s_anon)) {
1570 dentry = dget(hlist_bl_entry( !! 1516 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1571 do_one_tree(dentry); 1517 do_one_tree(dentry);
1572 } 1518 }
1573 } 1519 }
1574 1520
1575 static enum d_walk_ret find_submount(void *_d !! 1521 struct detach_data {
>> 1522 struct select_data select;
>> 1523 struct dentry *mountpoint;
>> 1524 };
>> 1525 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1576 { 1526 {
1577 struct dentry **victim = _data; !! 1527 struct detach_data *data = _data;
>> 1528
1578 if (d_mountpoint(dentry)) { 1529 if (d_mountpoint(dentry)) {
1579 *victim = dget_dlock(dentry); !! 1530 __dget_dlock(dentry);
>> 1531 data->mountpoint = dentry;
1580 return D_WALK_QUIT; 1532 return D_WALK_QUIT;
1581 } 1533 }
1582 return D_WALK_CONTINUE; !! 1534
>> 1535 return select_collect(&data->select, dentry);
>> 1536 }
>> 1537
>> 1538 static void check_and_drop(void *_data)
>> 1539 {
>> 1540 struct detach_data *data = _data;
>> 1541
>> 1542 if (!data->mountpoint && list_empty(&data->select.dispose))
>> 1543 __d_drop(data->select.start);
1583 } 1544 }
1584 1545
1585 /** 1546 /**
1586 * d_invalidate - detach submounts, prune dca 1547 * d_invalidate - detach submounts, prune dcache, and drop
1587 * @dentry: dentry to invalidate (aka detach, 1548 * @dentry: dentry to invalidate (aka detach, prune and drop)
>> 1549 *
>> 1550 * no dcache lock.
>> 1551 *
>> 1552 * The final d_drop is done as an atomic operation relative to
>> 1553 * rename_lock ensuring there are no races with d_set_mounted. This
>> 1554 * ensures there are no unhashed dentries on the path to a mountpoint.
1588 */ 1555 */
1589 void d_invalidate(struct dentry *dentry) 1556 void d_invalidate(struct dentry *dentry)
1590 { 1557 {
1591 bool had_submounts = false; !! 1558 /*
>> 1559 * If it's already been dropped, return OK.
>> 1560 */
1592 spin_lock(&dentry->d_lock); 1561 spin_lock(&dentry->d_lock);
1593 if (d_unhashed(dentry)) { 1562 if (d_unhashed(dentry)) {
1594 spin_unlock(&dentry->d_lock); 1563 spin_unlock(&dentry->d_lock);
1595 return; 1564 return;
1596 } 1565 }
1597 __d_drop(dentry); <<
1598 spin_unlock(&dentry->d_lock); 1566 spin_unlock(&dentry->d_lock);
1599 1567
1600 /* Negative dentries can be dropped w 1568 /* Negative dentries can be dropped without further checks */
1601 if (!dentry->d_inode) !! 1569 if (!dentry->d_inode) {
>> 1570 d_drop(dentry);
1602 return; 1571 return;
>> 1572 }
1603 1573
1604 shrink_dcache_parent(dentry); <<
1605 for (;;) { 1574 for (;;) {
1606 struct dentry *victim = NULL; !! 1575 struct detach_data data;
1607 d_walk(dentry, &victim, find_ !! 1576
1608 if (!victim) { !! 1577 data.mountpoint = NULL;
1609 if (had_submounts) !! 1578 INIT_LIST_HEAD(&data.select.dispose);
1610 shrink_dcache !! 1579 data.select.start = dentry;
>> 1580 data.select.found = 0;
>> 1581
>> 1582 d_walk(dentry, &data, detach_and_collect, check_and_drop);
>> 1583
>> 1584 if (!list_empty(&data.select.dispose))
>> 1585 shrink_dentry_list(&data.select.dispose);
>> 1586 else if (!data.mountpoint)
1611 return; 1587 return;
>> 1588
>> 1589 if (data.mountpoint) {
>> 1590 detach_mounts(data.mountpoint);
>> 1591 dput(data.mountpoint);
1612 } 1592 }
1613 had_submounts = true; !! 1593 cond_resched();
1614 detach_mounts(victim); <<
1615 dput(victim); <<
1616 } 1594 }
1617 } 1595 }
1618 EXPORT_SYMBOL(d_invalidate); 1596 EXPORT_SYMBOL(d_invalidate);
1619 1597
1620 /** 1598 /**
1621 * __d_alloc - allocate a dcache ent 1599 * __d_alloc - allocate a dcache entry
1622 * @sb: filesystem it will belong to 1600 * @sb: filesystem it will belong to
1623 * @name: qstr of the name 1601 * @name: qstr of the name
1624 * 1602 *
1625 * Allocates a dentry. It returns %NULL if th 1603 * Allocates a dentry. It returns %NULL if there is insufficient memory
1626 * available. On a success the dentry is retu 1604 * available. On a success the dentry is returned. The name passed in is
1627 * copied and the copy passed in may be reuse 1605 * copied and the copy passed in may be reused after this call.
1628 */ 1606 */
1629 1607
1630 static struct dentry *__d_alloc(struct super_ !! 1608 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1631 { 1609 {
1632 struct dentry *dentry; 1610 struct dentry *dentry;
1633 char *dname; 1611 char *dname;
1634 int err; 1612 int err;
1635 1613
1636 dentry = kmem_cache_alloc_lru(dentry_ !! 1614 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1637 GFP_KER <<
1638 if (!dentry) 1615 if (!dentry)
1639 return NULL; 1616 return NULL;
1640 1617
1641 /* 1618 /*
1642 * We guarantee that the inline name 1619 * We guarantee that the inline name is always NUL-terminated.
1643 * This way the memcpy() done by the 1620 * This way the memcpy() done by the name switching in rename
1644 * will still always have a NUL at th 1621 * will still always have a NUL at the end, even if we might
1645 * be overwriting an internal NUL cha 1622 * be overwriting an internal NUL character
1646 */ 1623 */
1647 dentry->d_iname[DNAME_INLINE_LEN-1] = 1624 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1648 if (unlikely(!name)) { 1625 if (unlikely(!name)) {
1649 name = &slash_name; 1626 name = &slash_name;
1650 dname = dentry->d_iname; 1627 dname = dentry->d_iname;
1651 } else if (name->len > DNAME_INLINE_L 1628 } else if (name->len > DNAME_INLINE_LEN-1) {
1652 size_t size = offsetof(struct 1629 size_t size = offsetof(struct external_name, name[1]);
1653 struct external_name *p = kma 1630 struct external_name *p = kmalloc(size + name->len,
1654 !! 1631 GFP_KERNEL_ACCOUNT);
1655 <<
1656 if (!p) { 1632 if (!p) {
1657 kmem_cache_free(dentr 1633 kmem_cache_free(dentry_cache, dentry);
1658 return NULL; 1634 return NULL;
1659 } 1635 }
1660 atomic_set(&p->u.count, 1); 1636 atomic_set(&p->u.count, 1);
1661 dname = p->name; 1637 dname = p->name;
>> 1638 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
>> 1639 kasan_unpoison_shadow(dname,
>> 1640 round_up(name->len + 1, sizeof(unsigned long)));
1662 } else { 1641 } else {
1663 dname = dentry->d_iname; 1642 dname = dentry->d_iname;
1664 } 1643 }
1665 1644
1666 dentry->d_name.len = name->len; 1645 dentry->d_name.len = name->len;
1667 dentry->d_name.hash = name->hash; 1646 dentry->d_name.hash = name->hash;
1668 memcpy(dname, name->name, name->len); 1647 memcpy(dname, name->name, name->len);
1669 dname[name->len] = 0; 1648 dname[name->len] = 0;
1670 1649
1671 /* Make sure we always see the termin 1650 /* Make sure we always see the terminating NUL character */
1672 smp_store_release(&dentry->d_name.nam !! 1651 smp_wmb();
>> 1652 dentry->d_name.name = dname;
1673 1653
1674 dentry->d_lockref.count = 1; 1654 dentry->d_lockref.count = 1;
1675 dentry->d_flags = 0; 1655 dentry->d_flags = 0;
1676 spin_lock_init(&dentry->d_lock); 1656 spin_lock_init(&dentry->d_lock);
1677 seqcount_spinlock_init(&dentry->d_seq !! 1657 seqcount_init(&dentry->d_seq);
1678 dentry->d_inode = NULL; 1658 dentry->d_inode = NULL;
1679 dentry->d_parent = dentry; 1659 dentry->d_parent = dentry;
1680 dentry->d_sb = sb; 1660 dentry->d_sb = sb;
1681 dentry->d_op = NULL; 1661 dentry->d_op = NULL;
1682 dentry->d_fsdata = NULL; 1662 dentry->d_fsdata = NULL;
1683 INIT_HLIST_BL_NODE(&dentry->d_hash); 1663 INIT_HLIST_BL_NODE(&dentry->d_hash);
1684 INIT_LIST_HEAD(&dentry->d_lru); 1664 INIT_LIST_HEAD(&dentry->d_lru);
1685 INIT_HLIST_HEAD(&dentry->d_children); !! 1665 INIT_LIST_HEAD(&dentry->d_subdirs);
1686 INIT_HLIST_NODE(&dentry->d_u.d_alias) 1666 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1687 INIT_HLIST_NODE(&dentry->d_sib); !! 1667 INIT_LIST_HEAD(&dentry->d_child);
1688 d_set_d_op(dentry, dentry->d_sb->s_d_ 1668 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1689 1669
1690 if (dentry->d_op && dentry->d_op->d_i 1670 if (dentry->d_op && dentry->d_op->d_init) {
1691 err = dentry->d_op->d_init(de 1671 err = dentry->d_op->d_init(dentry);
1692 if (err) { 1672 if (err) {
1693 if (dname_external(de 1673 if (dname_external(dentry))
1694 kfree(externa 1674 kfree(external_name(dentry));
1695 kmem_cache_free(dentr 1675 kmem_cache_free(dentry_cache, dentry);
1696 return NULL; 1676 return NULL;
1697 } 1677 }
1698 } 1678 }
1699 1679
1700 this_cpu_inc(nr_dentry); 1680 this_cpu_inc(nr_dentry);
1701 1681
1702 return dentry; 1682 return dentry;
1703 } 1683 }
1704 1684
1705 /** 1685 /**
1706 * d_alloc - allocate a dcache ent 1686 * d_alloc - allocate a dcache entry
1707 * @parent: parent of entry to allocate 1687 * @parent: parent of entry to allocate
1708 * @name: qstr of the name 1688 * @name: qstr of the name
1709 * 1689 *
1710 * Allocates a dentry. It returns %NULL if th 1690 * Allocates a dentry. It returns %NULL if there is insufficient memory
1711 * available. On a success the dentry is retu 1691 * available. On a success the dentry is returned. The name passed in is
1712 * copied and the copy passed in may be reuse 1692 * copied and the copy passed in may be reused after this call.
1713 */ 1693 */
1714 struct dentry *d_alloc(struct dentry * parent 1694 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1715 { 1695 {
1716 struct dentry *dentry = __d_alloc(par 1696 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1717 if (!dentry) 1697 if (!dentry)
1718 return NULL; 1698 return NULL;
>> 1699 dentry->d_flags |= DCACHE_RCUACCESS;
1719 spin_lock(&parent->d_lock); 1700 spin_lock(&parent->d_lock);
1720 /* 1701 /*
1721 * don't need child lock because it i 1702 * don't need child lock because it is not subject
1722 * to concurrency here 1703 * to concurrency here
1723 */ 1704 */
1724 dentry->d_parent = dget_dlock(parent) !! 1705 __dget_dlock(parent);
1725 hlist_add_head(&dentry->d_sib, &paren !! 1706 dentry->d_parent = parent;
>> 1707 list_add(&dentry->d_child, &parent->d_subdirs);
1726 spin_unlock(&parent->d_lock); 1708 spin_unlock(&parent->d_lock);
1727 1709
1728 return dentry; 1710 return dentry;
1729 } 1711 }
1730 EXPORT_SYMBOL(d_alloc); 1712 EXPORT_SYMBOL(d_alloc);
1731 1713
1732 struct dentry *d_alloc_anon(struct super_bloc <<
1733 { <<
1734 return __d_alloc(sb, NULL); <<
1735 } <<
1736 EXPORT_SYMBOL(d_alloc_anon); <<
1737 <<
1738 struct dentry *d_alloc_cursor(struct dentry * 1714 struct dentry *d_alloc_cursor(struct dentry * parent)
1739 { 1715 {
1740 struct dentry *dentry = d_alloc_anon( !! 1716 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1741 if (dentry) { 1717 if (dentry) {
1742 dentry->d_flags |= DCACHE_DEN !! 1718 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1743 dentry->d_parent = dget(paren 1719 dentry->d_parent = dget(parent);
1744 } 1720 }
1745 return dentry; 1721 return dentry;
1746 } 1722 }
1747 1723
1748 /** 1724 /**
1749 * d_alloc_pseudo - allocate a dentry (for lo 1725 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1750 * @sb: the superblock 1726 * @sb: the superblock
1751 * @name: qstr of the name 1727 * @name: qstr of the name
1752 * 1728 *
1753 * For a filesystem that just pins its dentri 1729 * For a filesystem that just pins its dentries in memory and never
1754 * performs lookups at all, return an unhashe 1730 * 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 */ 1731 */
1763 struct dentry *d_alloc_pseudo(struct super_bl 1732 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1764 { 1733 {
1765 static const struct dentry_operations !! 1734 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 } 1735 }
>> 1736 EXPORT_SYMBOL(d_alloc_pseudo);
1776 1737
1777 struct dentry *d_alloc_name(struct dentry *pa 1738 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1778 { 1739 {
1779 struct qstr q; 1740 struct qstr q;
1780 1741
1781 q.name = name; 1742 q.name = name;
1782 q.hash_len = hashlen_string(parent, n 1743 q.hash_len = hashlen_string(parent, name);
1783 return d_alloc(parent, &q); 1744 return d_alloc(parent, &q);
1784 } 1745 }
1785 EXPORT_SYMBOL(d_alloc_name); 1746 EXPORT_SYMBOL(d_alloc_name);
1786 1747
1787 void d_set_d_op(struct dentry *dentry, const 1748 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1788 { 1749 {
1789 WARN_ON_ONCE(dentry->d_op); 1750 WARN_ON_ONCE(dentry->d_op);
1790 WARN_ON_ONCE(dentry->d_flags & (DCACH 1751 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1791 DCACHE_OP_COM 1752 DCACHE_OP_COMPARE |
1792 DCACHE_OP_REV 1753 DCACHE_OP_REVALIDATE |
1793 DCACHE_OP_WEA 1754 DCACHE_OP_WEAK_REVALIDATE |
1794 DCACHE_OP_DEL 1755 DCACHE_OP_DELETE |
1795 DCACHE_OP_REA 1756 DCACHE_OP_REAL));
1796 dentry->d_op = op; 1757 dentry->d_op = op;
1797 if (!op) 1758 if (!op)
1798 return; 1759 return;
1799 if (op->d_hash) 1760 if (op->d_hash)
1800 dentry->d_flags |= DCACHE_OP_ 1761 dentry->d_flags |= DCACHE_OP_HASH;
1801 if (op->d_compare) 1762 if (op->d_compare)
1802 dentry->d_flags |= DCACHE_OP_ 1763 dentry->d_flags |= DCACHE_OP_COMPARE;
1803 if (op->d_revalidate) 1764 if (op->d_revalidate)
1804 dentry->d_flags |= DCACHE_OP_ 1765 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1805 if (op->d_weak_revalidate) 1766 if (op->d_weak_revalidate)
1806 dentry->d_flags |= DCACHE_OP_ 1767 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1807 if (op->d_delete) 1768 if (op->d_delete)
1808 dentry->d_flags |= DCACHE_OP_ 1769 dentry->d_flags |= DCACHE_OP_DELETE;
1809 if (op->d_prune) 1770 if (op->d_prune)
1810 dentry->d_flags |= DCACHE_OP_ 1771 dentry->d_flags |= DCACHE_OP_PRUNE;
1811 if (op->d_real) 1772 if (op->d_real)
1812 dentry->d_flags |= DCACHE_OP_ 1773 dentry->d_flags |= DCACHE_OP_REAL;
1813 1774
1814 } 1775 }
1815 EXPORT_SYMBOL(d_set_d_op); 1776 EXPORT_SYMBOL(d_set_d_op);
1816 1777
>> 1778
>> 1779 /*
>> 1780 * d_set_fallthru - Mark a dentry as falling through to a lower layer
>> 1781 * @dentry - The dentry to mark
>> 1782 *
>> 1783 * Mark a dentry as falling through to the lower layer (as set with
>> 1784 * d_pin_lower()). This flag may be recorded on the medium.
>> 1785 */
>> 1786 void d_set_fallthru(struct dentry *dentry)
>> 1787 {
>> 1788 spin_lock(&dentry->d_lock);
>> 1789 dentry->d_flags |= DCACHE_FALLTHRU;
>> 1790 spin_unlock(&dentry->d_lock);
>> 1791 }
>> 1792 EXPORT_SYMBOL(d_set_fallthru);
>> 1793
1817 static unsigned d_flags_for_inode(struct inod 1794 static unsigned d_flags_for_inode(struct inode *inode)
1818 { 1795 {
1819 unsigned add_flags = DCACHE_REGULAR_T 1796 unsigned add_flags = DCACHE_REGULAR_TYPE;
1820 1797
1821 if (!inode) 1798 if (!inode)
1822 return DCACHE_MISS_TYPE; 1799 return DCACHE_MISS_TYPE;
1823 1800
1824 if (S_ISDIR(inode->i_mode)) { 1801 if (S_ISDIR(inode->i_mode)) {
1825 add_flags = DCACHE_DIRECTORY_ 1802 add_flags = DCACHE_DIRECTORY_TYPE;
1826 if (unlikely(!(inode->i_opfla 1803 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1827 if (unlikely(!inode-> 1804 if (unlikely(!inode->i_op->lookup))
1828 add_flags = D 1805 add_flags = DCACHE_AUTODIR_TYPE;
1829 else 1806 else
1830 inode->i_opfl 1807 inode->i_opflags |= IOP_LOOKUP;
1831 } 1808 }
1832 goto type_determined; 1809 goto type_determined;
1833 } 1810 }
1834 1811
1835 if (unlikely(!(inode->i_opflags & IOP 1812 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1836 if (unlikely(inode->i_op->get 1813 if (unlikely(inode->i_op->get_link)) {
1837 add_flags = DCACHE_SY 1814 add_flags = DCACHE_SYMLINK_TYPE;
1838 goto type_determined; 1815 goto type_determined;
1839 } 1816 }
1840 inode->i_opflags |= IOP_NOFOL 1817 inode->i_opflags |= IOP_NOFOLLOW;
1841 } 1818 }
1842 1819
1843 if (unlikely(!S_ISREG(inode->i_mode)) 1820 if (unlikely(!S_ISREG(inode->i_mode)))
1844 add_flags = DCACHE_SPECIAL_TY 1821 add_flags = DCACHE_SPECIAL_TYPE;
1845 1822
1846 type_determined: 1823 type_determined:
1847 if (unlikely(IS_AUTOMOUNT(inode))) 1824 if (unlikely(IS_AUTOMOUNT(inode)))
1848 add_flags |= DCACHE_NEED_AUTO 1825 add_flags |= DCACHE_NEED_AUTOMOUNT;
1849 return add_flags; 1826 return add_flags;
1850 } 1827 }
1851 1828
1852 static void __d_instantiate(struct dentry *de 1829 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1853 { 1830 {
1854 unsigned add_flags = d_flags_for_inod 1831 unsigned add_flags = d_flags_for_inode(inode);
1855 WARN_ON(d_in_lookup(dentry)); 1832 WARN_ON(d_in_lookup(dentry));
1856 1833
1857 spin_lock(&dentry->d_lock); 1834 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, 1835 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1866 raw_write_seqcount_begin(&dentry->d_s 1836 raw_write_seqcount_begin(&dentry->d_seq);
1867 __d_set_inode_and_type(dentry, inode, 1837 __d_set_inode_and_type(dentry, inode, add_flags);
1868 raw_write_seqcount_end(&dentry->d_seq 1838 raw_write_seqcount_end(&dentry->d_seq);
1869 fsnotify_update_flags(dentry); 1839 fsnotify_update_flags(dentry);
1870 spin_unlock(&dentry->d_lock); 1840 spin_unlock(&dentry->d_lock);
1871 } 1841 }
1872 1842
1873 /** 1843 /**
1874 * d_instantiate - fill in inode information 1844 * d_instantiate - fill in inode information for a dentry
1875 * @entry: dentry to complete 1845 * @entry: dentry to complete
1876 * @inode: inode to attach to this dentry 1846 * @inode: inode to attach to this dentry
1877 * 1847 *
1878 * Fill in inode information in the entry. 1848 * Fill in inode information in the entry.
1879 * 1849 *
1880 * This turns negative dentries into producti 1850 * This turns negative dentries into productive full members
1881 * of society. 1851 * of society.
1882 * 1852 *
1883 * NOTE! This assumes that the inode count ha 1853 * NOTE! This assumes that the inode count has been incremented
1884 * (or otherwise set) by the caller to indica 1854 * (or otherwise set) by the caller to indicate that it is now
1885 * in use by the dcache. 1855 * in use by the dcache.
1886 */ 1856 */
1887 1857
1888 void d_instantiate(struct dentry *entry, stru 1858 void d_instantiate(struct dentry *entry, struct inode * inode)
1889 { 1859 {
1890 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 1860 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1891 if (inode) { 1861 if (inode) {
1892 security_d_instantiate(entry, 1862 security_d_instantiate(entry, inode);
1893 spin_lock(&inode->i_lock); 1863 spin_lock(&inode->i_lock);
1894 __d_instantiate(entry, inode) 1864 __d_instantiate(entry, inode);
1895 spin_unlock(&inode->i_lock); 1865 spin_unlock(&inode->i_lock);
1896 } 1866 }
1897 } 1867 }
1898 EXPORT_SYMBOL(d_instantiate); 1868 EXPORT_SYMBOL(d_instantiate);
1899 1869
1900 /* !! 1870 /**
1901 * This should be equivalent to d_instantiate !! 1871 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1902 * with lockdep-related part of unlock_new_in !! 1872 * @entry: dentry to complete
1903 * anything else. Use that instead of open-c !! 1873 * @inode: inode to attach to this dentry
1904 * unlock_new_inode() combinations. !! 1874 *
>> 1875 * Fill in inode information in the entry. If a directory alias is found, then
>> 1876 * return an error (and drop inode). Together with d_materialise_unique() this
>> 1877 * guarantees that a directory inode may never have more than one alias.
1905 */ 1878 */
1906 void d_instantiate_new(struct dentry *entry, !! 1879 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1907 { 1880 {
1908 BUG_ON(!hlist_unhashed(&entry->d_u.d_ 1881 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1909 BUG_ON(!inode); !! 1882
1910 lockdep_annotate_inode_mutex_key(inod <<
1911 security_d_instantiate(entry, inode); 1883 security_d_instantiate(entry, inode);
1912 spin_lock(&inode->i_lock); 1884 spin_lock(&inode->i_lock);
>> 1885 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
>> 1886 spin_unlock(&inode->i_lock);
>> 1887 iput(inode);
>> 1888 return -EBUSY;
>> 1889 }
1913 __d_instantiate(entry, inode); 1890 __d_instantiate(entry, inode);
1914 WARN_ON(!(inode->i_state & I_NEW)); <<
1915 inode->i_state &= ~I_NEW & ~I_CREATIN <<
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(); <<
1922 inode_wake_up_bit(inode, __I_NEW); <<
1923 spin_unlock(&inode->i_lock); 1891 spin_unlock(&inode->i_lock);
>> 1892
>> 1893 return 0;
1924 } 1894 }
1925 EXPORT_SYMBOL(d_instantiate_new); !! 1895 EXPORT_SYMBOL(d_instantiate_no_diralias);
1926 1896
1927 struct dentry *d_make_root(struct inode *root 1897 struct dentry *d_make_root(struct inode *root_inode)
1928 { 1898 {
1929 struct dentry *res = NULL; 1899 struct dentry *res = NULL;
1930 1900
1931 if (root_inode) { 1901 if (root_inode) {
1932 res = d_alloc_anon(root_inode !! 1902 res = __d_alloc(root_inode->i_sb, NULL);
1933 if (res) 1903 if (res)
1934 d_instantiate(res, ro 1904 d_instantiate(res, root_inode);
1935 else 1905 else
1936 iput(root_inode); 1906 iput(root_inode);
1937 } 1907 }
1938 return res; 1908 return res;
1939 } 1909 }
1940 EXPORT_SYMBOL(d_make_root); 1910 EXPORT_SYMBOL(d_make_root);
1941 1911
1942 static struct dentry *__d_obtain_alias(struct !! 1912 static struct dentry * __d_find_any_alias(struct inode *inode)
1943 { 1913 {
1944 struct super_block *sb; !! 1914 struct dentry *alias;
1945 struct dentry *new, *res; !! 1915
>> 1916 if (hlist_empty(&inode->i_dentry))
>> 1917 return NULL;
>> 1918 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
>> 1919 __dget(alias);
>> 1920 return alias;
>> 1921 }
>> 1922
>> 1923 /**
>> 1924 * d_find_any_alias - find any alias for a given inode
>> 1925 * @inode: inode to find an alias for
>> 1926 *
>> 1927 * If any aliases exist for the given inode, take and return a
>> 1928 * reference for one of them. If no aliases exist, return %NULL.
>> 1929 */
>> 1930 struct dentry *d_find_any_alias(struct inode *inode)
>> 1931 {
>> 1932 struct dentry *de;
>> 1933
>> 1934 spin_lock(&inode->i_lock);
>> 1935 de = __d_find_any_alias(inode);
>> 1936 spin_unlock(&inode->i_lock);
>> 1937 return de;
>> 1938 }
>> 1939 EXPORT_SYMBOL(d_find_any_alias);
>> 1940
>> 1941 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
>> 1942 {
>> 1943 struct dentry *tmp;
>> 1944 struct dentry *res;
>> 1945 unsigned add_flags;
1946 1946
1947 if (!inode) 1947 if (!inode)
1948 return ERR_PTR(-ESTALE); 1948 return ERR_PTR(-ESTALE);
1949 if (IS_ERR(inode)) 1949 if (IS_ERR(inode))
1950 return ERR_CAST(inode); 1950 return ERR_CAST(inode);
1951 1951
1952 sb = inode->i_sb; !! 1952 res = d_find_any_alias(inode);
1953 <<
1954 res = d_find_any_alias(inode); /* exi <<
1955 if (res) 1953 if (res)
1956 goto out; !! 1954 goto out_iput;
1957 1955
1958 new = d_alloc_anon(sb); !! 1956 tmp = __d_alloc(inode->i_sb, NULL);
1959 if (!new) { !! 1957 if (!tmp) {
1960 res = ERR_PTR(-ENOMEM); 1958 res = ERR_PTR(-ENOMEM);
1961 goto out; !! 1959 goto out_iput;
1962 } 1960 }
1963 1961
1964 security_d_instantiate(new, inode); !! 1962 security_d_instantiate(tmp, inode);
1965 spin_lock(&inode->i_lock); 1963 spin_lock(&inode->i_lock);
1966 res = __d_find_any_alias(inode); /* r !! 1964 res = __d_find_any_alias(inode);
1967 if (likely(!res)) { /* still no alias !! 1965 if (res) {
1968 unsigned add_flags = d_flags_ <<
1969 <<
1970 if (disconnected) <<
1971 add_flags |= DCACHE_D <<
1972 <<
1973 spin_lock(&new->d_lock); <<
1974 __d_set_inode_and_type(new, i <<
1975 hlist_add_head(&new->d_u.d_al <<
1976 if (!disconnected) { <<
1977 hlist_bl_lock(&sb->s_ <<
1978 hlist_bl_add_head(&ne <<
1979 hlist_bl_unlock(&sb-> <<
1980 } <<
1981 spin_unlock(&new->d_lock); <<
1982 spin_unlock(&inode->i_lock); <<
1983 inode = NULL; /* consumed by <<
1984 res = new; <<
1985 } else { <<
1986 spin_unlock(&inode->i_lock); 1966 spin_unlock(&inode->i_lock);
1987 dput(new); !! 1967 dput(tmp);
>> 1968 goto out_iput;
1988 } 1969 }
1989 1970
1990 out: !! 1971 /* attach a disconnected dentry */
>> 1972 add_flags = d_flags_for_inode(inode);
>> 1973
>> 1974 if (disconnected)
>> 1975 add_flags |= DCACHE_DISCONNECTED;
>> 1976
>> 1977 spin_lock(&tmp->d_lock);
>> 1978 __d_set_inode_and_type(tmp, inode, add_flags);
>> 1979 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
>> 1980 hlist_bl_lock(&tmp->d_sb->s_anon);
>> 1981 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
>> 1982 hlist_bl_unlock(&tmp->d_sb->s_anon);
>> 1983 spin_unlock(&tmp->d_lock);
>> 1984 spin_unlock(&inode->i_lock);
>> 1985
>> 1986 return tmp;
>> 1987
>> 1988 out_iput:
1991 iput(inode); 1989 iput(inode);
1992 return res; 1990 return res;
1993 } 1991 }
1994 1992
1995 /** 1993 /**
1996 * d_obtain_alias - find or allocate a DISCON 1994 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1997 * @inode: inode to allocate the dentry for 1995 * @inode: inode to allocate the dentry for
1998 * 1996 *
1999 * Obtain a dentry for an inode resulting fro 1997 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2000 * similar open by handle operations. The re 1998 * similar open by handle operations. The returned dentry may be anonymous,
2001 * or may have a full name (if the inode was 1999 * or may have a full name (if the inode was already in the cache).
2002 * 2000 *
2003 * When called on a directory inode, we must 2001 * When called on a directory inode, we must ensure that the inode only ever
2004 * has one dentry. If a dentry is found, tha 2002 * has one dentry. If a dentry is found, that is returned instead of
2005 * allocating a new one. 2003 * allocating a new one.
2006 * 2004 *
2007 * On successful return, the reference to the 2005 * On successful return, the reference to the inode has been transferred
2008 * to the dentry. In case of an error the re 2006 * 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 2007 * 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 2008 * be passed in and the error will be propagated to the return value,
2011 * with a %NULL @inode replaced by ERR_PTR(-E 2009 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2012 */ 2010 */
2013 struct dentry *d_obtain_alias(struct inode *i 2011 struct dentry *d_obtain_alias(struct inode *inode)
2014 { 2012 {
2015 return __d_obtain_alias(inode, true); !! 2013 return __d_obtain_alias(inode, 1);
2016 } 2014 }
2017 EXPORT_SYMBOL(d_obtain_alias); 2015 EXPORT_SYMBOL(d_obtain_alias);
2018 2016
2019 /** 2017 /**
2020 * d_obtain_root - find or allocate a dentry 2018 * d_obtain_root - find or allocate a dentry for a given inode
2021 * @inode: inode to allocate the dentry for 2019 * @inode: inode to allocate the dentry for
2022 * 2020 *
2023 * Obtain an IS_ROOT dentry for the root of a 2021 * Obtain an IS_ROOT dentry for the root of a filesystem.
2024 * 2022 *
2025 * We must ensure that directory inodes only 2023 * We must ensure that directory inodes only ever have one dentry. If a
2026 * dentry is found, that is returned instead 2024 * dentry is found, that is returned instead of allocating a new one.
2027 * 2025 *
2028 * On successful return, the reference to the 2026 * On successful return, the reference to the inode has been transferred
2029 * to the dentry. In case of an error the re 2027 * to the dentry. In case of an error the reference on the inode is
2030 * released. A %NULL or IS_ERR inode may be 2028 * released. A %NULL or IS_ERR inode may be passed in and will be the
2031 * error will be propagate to the return valu 2029 * error will be propagate to the return value, with a %NULL @inode
2032 * replaced by ERR_PTR(-ESTALE). 2030 * replaced by ERR_PTR(-ESTALE).
2033 */ 2031 */
2034 struct dentry *d_obtain_root(struct inode *in 2032 struct dentry *d_obtain_root(struct inode *inode)
2035 { 2033 {
2036 return __d_obtain_alias(inode, false) !! 2034 return __d_obtain_alias(inode, 0);
2037 } 2035 }
2038 EXPORT_SYMBOL(d_obtain_root); 2036 EXPORT_SYMBOL(d_obtain_root);
2039 2037
2040 /** 2038 /**
2041 * d_add_ci - lookup or allocate new dentry w 2039 * d_add_ci - lookup or allocate new dentry with case-exact name
2042 * @inode: the inode case-insensitive lookup 2040 * @inode: the inode case-insensitive lookup has found
2043 * @dentry: the negative dentry that was pass 2041 * @dentry: the negative dentry that was passed to the parent's lookup func
2044 * @name: the case-exact name to be associa 2042 * @name: the case-exact name to be associated with the returned dentry
2045 * 2043 *
2046 * This is to avoid filling the dcache with c 2044 * This is to avoid filling the dcache with case-insensitive names to the
2047 * same inode, only the actual correct case i 2045 * same inode, only the actual correct case is stored in the dcache for
2048 * case-insensitive filesystems. 2046 * case-insensitive filesystems.
2049 * 2047 *
2050 * For a case-insensitive lookup match and if !! 2048 * For a case-insensitive lookup match and if the the case-exact dentry
2051 * already exists in the dcache, use it and r !! 2049 * already exists in in the dcache, use it and return it.
2052 * 2050 *
2053 * If no entry exists with the exact case nam 2051 * If no entry exists with the exact case name, allocate new dentry with
2054 * the exact case, and return the spliced ent 2052 * the exact case, and return the spliced entry.
2055 */ 2053 */
2056 struct dentry *d_add_ci(struct dentry *dentry 2054 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2057 struct qstr *name) 2055 struct qstr *name)
2058 { 2056 {
2059 struct dentry *found, *res; 2057 struct dentry *found, *res;
2060 2058
2061 /* 2059 /*
2062 * First check if a dentry matching t 2060 * First check if a dentry matching the name already exists,
2063 * if not go ahead and create it now. 2061 * if not go ahead and create it now.
2064 */ 2062 */
2065 found = d_hash_and_lookup(dentry->d_p 2063 found = d_hash_and_lookup(dentry->d_parent, name);
2066 if (found) { 2064 if (found) {
2067 iput(inode); 2065 iput(inode);
2068 return found; 2066 return found;
2069 } 2067 }
2070 if (d_in_lookup(dentry)) { 2068 if (d_in_lookup(dentry)) {
2071 found = d_alloc_parallel(dent 2069 found = d_alloc_parallel(dentry->d_parent, name,
2072 dentr 2070 dentry->d_wait);
2073 if (IS_ERR(found) || !d_in_lo 2071 if (IS_ERR(found) || !d_in_lookup(found)) {
2074 iput(inode); 2072 iput(inode);
2075 return found; 2073 return found;
2076 } 2074 }
2077 } else { 2075 } else {
2078 found = d_alloc(dentry->d_par 2076 found = d_alloc(dentry->d_parent, name);
2079 if (!found) { 2077 if (!found) {
2080 iput(inode); 2078 iput(inode);
2081 return ERR_PTR(-ENOME 2079 return ERR_PTR(-ENOMEM);
2082 } 2080 }
2083 } 2081 }
2084 res = d_splice_alias(inode, found); 2082 res = d_splice_alias(inode, found);
2085 if (res) { 2083 if (res) {
2086 d_lookup_done(found); <<
2087 dput(found); 2084 dput(found);
2088 return res; 2085 return res;
2089 } 2086 }
2090 return found; 2087 return found;
2091 } 2088 }
2092 EXPORT_SYMBOL(d_add_ci); 2089 EXPORT_SYMBOL(d_add_ci);
2093 2090
2094 /** !! 2091
2095 * d_same_name - compare dentry name with cas !! 2092 static inline bool d_same_name(const struct dentry *dentry,
2096 * @parent: parent dentry !! 2093 const struct dentry *parent,
2097 * @dentry: the negative dentry that was pass !! 2094 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 { 2095 {
2105 if (likely(!(parent->d_flags & DCACHE 2096 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2106 if (dentry->d_name.len != nam 2097 if (dentry->d_name.len != name->len)
2107 return false; 2098 return false;
2108 return dentry_cmp(dentry, nam 2099 return dentry_cmp(dentry, name->name, name->len) == 0;
2109 } 2100 }
2110 return parent->d_op->d_compare(dentry 2101 return parent->d_op->d_compare(dentry,
2111 dentry 2102 dentry->d_name.len, dentry->d_name.name,
2112 name) 2103 name) == 0;
2113 } 2104 }
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 2105
2158 /** 2106 /**
2159 * __d_lookup_rcu - search for a dentry (racy 2107 * __d_lookup_rcu - search for a dentry (racy, store-free)
2160 * @parent: parent dentry 2108 * @parent: parent dentry
2161 * @name: qstr of name we wish to find 2109 * @name: qstr of name we wish to find
2162 * @seqp: returns d_seq value at the point wh 2110 * @seqp: returns d_seq value at the point where the dentry was found
2163 * Returns: dentry, or NULL 2111 * Returns: dentry, or NULL
2164 * 2112 *
2165 * __d_lookup_rcu is the dcache lookup functi 2113 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2166 * resolution (store-free path walking) desig 2114 * resolution (store-free path walking) design described in
2167 * Documentation/filesystems/path-lookup.txt. 2115 * Documentation/filesystems/path-lookup.txt.
2168 * 2116 *
2169 * This is not to be used outside core vfs. 2117 * This is not to be used outside core vfs.
2170 * 2118 *
2171 * __d_lookup_rcu must only be used in rcu-wa 2119 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2172 * held, and rcu_read_lock held. The returned 2120 * held, and rcu_read_lock held. The returned dentry must not be stored into
2173 * without taking d_lock and checking d_seq s 2121 * without taking d_lock and checking d_seq sequence count against @seq
2174 * returned here. 2122 * returned here.
2175 * 2123 *
>> 2124 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
>> 2125 * function.
>> 2126 *
2176 * Alternatively, __d_lookup_rcu may be calle 2127 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2177 * the returned dentry, so long as its parent 2128 * the returned dentry, so long as its parent's seqlock is checked after the
2178 * child is looked up. Thus, an interlocking 2129 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2179 * is formed, giving integrity down the path 2130 * is formed, giving integrity down the path walk.
2180 * 2131 *
2181 * NOTE! The caller *has* to check the result 2132 * NOTE! The caller *has* to check the resulting dentry against the sequence
2182 * number we've returned before using any of 2133 * number we've returned before using any of the resulting dentry state!
2183 */ 2134 */
2184 struct dentry *__d_lookup_rcu(const struct de 2135 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2185 const struct 2136 const struct qstr *name,
2186 unsigned *seq 2137 unsigned *seqp)
2187 { 2138 {
2188 u64 hashlen = name->hash_len; 2139 u64 hashlen = name->hash_len;
2189 const unsigned char *str = name->name 2140 const unsigned char *str = name->name;
2190 struct hlist_bl_head *b = d_hash(hash !! 2141 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2191 struct hlist_bl_node *node; 2142 struct hlist_bl_node *node;
2192 struct dentry *dentry; 2143 struct dentry *dentry;
2193 2144
2194 /* 2145 /*
2195 * Note: There is significant duplica 2146 * Note: There is significant duplication with __d_lookup_rcu which is
2196 * required to prevent single threade 2147 * required to prevent single threaded performance regressions
2197 * especially on architectures where 2148 * especially on architectures where smp_rmb (in seqcounts) are costly.
2198 * Keep the two functions in sync. 2149 * Keep the two functions in sync.
2199 */ 2150 */
2200 2151
2201 if (unlikely(parent->d_flags & DCACHE <<
2202 return __d_lookup_rcu_op_comp <<
2203 <<
2204 /* 2152 /*
2205 * The hash list is protected using R 2153 * The hash list is protected using RCU.
2206 * 2154 *
2207 * Carefully use d_seq when comparing 2155 * Carefully use d_seq when comparing a candidate dentry, to avoid
2208 * races with d_move(). 2156 * races with d_move().
2209 * 2157 *
2210 * It is possible that concurrent ren 2158 * It is possible that concurrent renames can mess up our list
2211 * walk here and result in missing ou 2159 * walk here and result in missing our dentry, resulting in the
2212 * false-negative result. d_lookup() 2160 * false-negative result. d_lookup() protects against concurrent
2213 * renames using rename_lock seqlock. 2161 * renames using rename_lock seqlock.
2214 * 2162 *
2215 * See Documentation/filesystems/path 2163 * See Documentation/filesystems/path-lookup.txt for more details.
2216 */ 2164 */
2217 hlist_bl_for_each_entry_rcu(dentry, n 2165 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2218 unsigned seq; 2166 unsigned seq;
2219 2167
>> 2168 seqretry:
2220 /* 2169 /*
2221 * The dentry sequence count 2170 * The dentry sequence count protects us from concurrent
2222 * renames, and thus protects 2171 * renames, and thus protects parent and name fields.
2223 * 2172 *
2224 * The caller must perform a 2173 * The caller must perform a seqcount check in order
2225 * to do anything useful with 2174 * to do anything useful with the returned dentry.
2226 * 2175 *
2227 * NOTE! We do a "raw" seqcou 2176 * NOTE! We do a "raw" seqcount_begin here. That means that
2228 * we don't wait for the sequ 2177 * we don't wait for the sequence count to stabilize if it
2229 * is in the middle of a sequ 2178 * is in the middle of a sequence change. If we do the slow
2230 * dentry compare, we will do 2179 * dentry compare, we will do seqretries until it is stable,
2231 * and if we end up with a su 2180 * and if we end up with a successful lookup, we actually
2232 * want to exit RCU lookup an 2181 * want to exit RCU lookup anyway.
2233 * 2182 *
2234 * Note that raw_seqcount_beg 2183 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2235 * we are still guaranteed NU 2184 * we are still guaranteed NUL-termination of ->d_name.name.
2236 */ 2185 */
2237 seq = raw_seqcount_begin(&den 2186 seq = raw_seqcount_begin(&dentry->d_seq);
2238 if (dentry->d_parent != paren 2187 if (dentry->d_parent != parent)
2239 continue; 2188 continue;
2240 if (d_unhashed(dentry)) 2189 if (d_unhashed(dentry))
2241 continue; 2190 continue;
2242 if (dentry->d_name.hash_len ! !! 2191
2243 continue; !! 2192 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2244 if (dentry_cmp(dentry, str, h !! 2193 int tlen;
2245 continue; !! 2194 const char *tname;
>> 2195 if (dentry->d_name.hash != hashlen_hash(hashlen))
>> 2196 continue;
>> 2197 tlen = dentry->d_name.len;
>> 2198 tname = dentry->d_name.name;
>> 2199 /* we want a consistent (name,len) pair */
>> 2200 if (read_seqcount_retry(&dentry->d_seq, seq)) {
>> 2201 cpu_relax();
>> 2202 goto seqretry;
>> 2203 }
>> 2204 if (parent->d_op->d_compare(dentry,
>> 2205 tlen, tname, name) != 0)
>> 2206 continue;
>> 2207 } else {
>> 2208 if (dentry->d_name.hash_len != hashlen)
>> 2209 continue;
>> 2210 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
>> 2211 continue;
>> 2212 }
2246 *seqp = seq; 2213 *seqp = seq;
2247 return dentry; 2214 return dentry;
2248 } 2215 }
2249 return NULL; 2216 return NULL;
2250 } 2217 }
2251 2218
2252 /** 2219 /**
2253 * d_lookup - search for a dentry 2220 * d_lookup - search for a dentry
2254 * @parent: parent dentry 2221 * @parent: parent dentry
2255 * @name: qstr of name we wish to find 2222 * @name: qstr of name we wish to find
2256 * Returns: dentry, or NULL 2223 * Returns: dentry, or NULL
2257 * 2224 *
2258 * d_lookup searches the children of the pare 2225 * d_lookup searches the children of the parent dentry for the name in
2259 * question. If the dentry is found its refer 2226 * question. If the dentry is found its reference count is incremented and the
2260 * dentry is returned. The caller must use dp 2227 * dentry is returned. The caller must use dput to free the entry when it has
2261 * finished using it. %NULL is returned if th 2228 * finished using it. %NULL is returned if the dentry does not exist.
2262 */ 2229 */
2263 struct dentry *d_lookup(const struct dentry * 2230 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2264 { 2231 {
2265 struct dentry *dentry; 2232 struct dentry *dentry;
2266 unsigned seq; 2233 unsigned seq;
2267 2234
2268 do { 2235 do {
2269 seq = read_seqbegin(&rename_l 2236 seq = read_seqbegin(&rename_lock);
2270 dentry = __d_lookup(parent, n 2237 dentry = __d_lookup(parent, name);
2271 if (dentry) 2238 if (dentry)
2272 break; 2239 break;
2273 } while (read_seqretry(&rename_lock, 2240 } while (read_seqretry(&rename_lock, seq));
2274 return dentry; 2241 return dentry;
2275 } 2242 }
2276 EXPORT_SYMBOL(d_lookup); 2243 EXPORT_SYMBOL(d_lookup);
2277 2244
2278 /** 2245 /**
2279 * __d_lookup - search for a dentry (racy) 2246 * __d_lookup - search for a dentry (racy)
2280 * @parent: parent dentry 2247 * @parent: parent dentry
2281 * @name: qstr of name we wish to find 2248 * @name: qstr of name we wish to find
2282 * Returns: dentry, or NULL 2249 * Returns: dentry, or NULL
2283 * 2250 *
2284 * __d_lookup is like d_lookup, however it ma 2251 * __d_lookup is like d_lookup, however it may (rarely) return a
2285 * false-negative result due to unrelated ren 2252 * false-negative result due to unrelated rename activity.
2286 * 2253 *
2287 * __d_lookup is slightly faster by avoiding 2254 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2288 * however it must be used carefully, eg. wit 2255 * however it must be used carefully, eg. with a following d_lookup in
2289 * the case of failure. 2256 * the case of failure.
2290 * 2257 *
2291 * __d_lookup callers must be commented. 2258 * __d_lookup callers must be commented.
2292 */ 2259 */
2293 struct dentry *__d_lookup(const struct dentry 2260 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2294 { 2261 {
2295 unsigned int hash = name->hash; 2262 unsigned int hash = name->hash;
2296 struct hlist_bl_head *b = d_hash(hash 2263 struct hlist_bl_head *b = d_hash(hash);
2297 struct hlist_bl_node *node; 2264 struct hlist_bl_node *node;
2298 struct dentry *found = NULL; 2265 struct dentry *found = NULL;
2299 struct dentry *dentry; 2266 struct dentry *dentry;
2300 2267
2301 /* 2268 /*
2302 * Note: There is significant duplica 2269 * Note: There is significant duplication with __d_lookup_rcu which is
2303 * required to prevent single threade 2270 * required to prevent single threaded performance regressions
2304 * especially on architectures where 2271 * especially on architectures where smp_rmb (in seqcounts) are costly.
2305 * Keep the two functions in sync. 2272 * Keep the two functions in sync.
2306 */ 2273 */
2307 2274
2308 /* 2275 /*
2309 * The hash list is protected using R 2276 * The hash list is protected using RCU.
2310 * 2277 *
2311 * Take d_lock when comparing a candi 2278 * Take d_lock when comparing a candidate dentry, to avoid races
2312 * with d_move(). 2279 * with d_move().
2313 * 2280 *
2314 * It is possible that concurrent ren 2281 * It is possible that concurrent renames can mess up our list
2315 * walk here and result in missing ou 2282 * walk here and result in missing our dentry, resulting in the
2316 * false-negative result. d_lookup() 2283 * false-negative result. d_lookup() protects against concurrent
2317 * renames using rename_lock seqlock. 2284 * renames using rename_lock seqlock.
2318 * 2285 *
2319 * See Documentation/filesystems/path 2286 * See Documentation/filesystems/path-lookup.txt for more details.
2320 */ 2287 */
2321 rcu_read_lock(); 2288 rcu_read_lock();
2322 2289
2323 hlist_bl_for_each_entry_rcu(dentry, n 2290 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2324 2291
2325 if (dentry->d_name.hash != ha 2292 if (dentry->d_name.hash != hash)
2326 continue; 2293 continue;
2327 2294
2328 spin_lock(&dentry->d_lock); 2295 spin_lock(&dentry->d_lock);
2329 if (dentry->d_parent != paren 2296 if (dentry->d_parent != parent)
2330 goto next; 2297 goto next;
2331 if (d_unhashed(dentry)) 2298 if (d_unhashed(dentry))
2332 goto next; 2299 goto next;
2333 2300
2334 if (!d_same_name(dentry, pare 2301 if (!d_same_name(dentry, parent, name))
2335 goto next; 2302 goto next;
2336 2303
2337 dentry->d_lockref.count++; 2304 dentry->d_lockref.count++;
2338 found = dentry; 2305 found = dentry;
2339 spin_unlock(&dentry->d_lock); 2306 spin_unlock(&dentry->d_lock);
2340 break; 2307 break;
2341 next: 2308 next:
2342 spin_unlock(&dentry->d_lock); 2309 spin_unlock(&dentry->d_lock);
2343 } 2310 }
2344 rcu_read_unlock(); 2311 rcu_read_unlock();
2345 2312
2346 return found; 2313 return found;
2347 } 2314 }
2348 2315
2349 /** 2316 /**
2350 * d_hash_and_lookup - hash the qstr then sea 2317 * d_hash_and_lookup - hash the qstr then search for a dentry
2351 * @dir: Directory to search in 2318 * @dir: Directory to search in
2352 * @name: qstr of name we wish to find 2319 * @name: qstr of name we wish to find
2353 * 2320 *
2354 * On lookup failure NULL is returned; on bad 2321 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2355 */ 2322 */
2356 struct dentry *d_hash_and_lookup(struct dentr 2323 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2357 { 2324 {
2358 /* 2325 /*
2359 * Check for a fs-specific hash funct 2326 * Check for a fs-specific hash function. Note that we must
2360 * calculate the standard hash first, 2327 * calculate the standard hash first, as the d_op->d_hash()
2361 * routine may choose to leave the ha 2328 * routine may choose to leave the hash value unchanged.
2362 */ 2329 */
2363 name->hash = full_name_hash(dir, name 2330 name->hash = full_name_hash(dir, name->name, name->len);
2364 if (dir->d_flags & DCACHE_OP_HASH) { 2331 if (dir->d_flags & DCACHE_OP_HASH) {
2365 int err = dir->d_op->d_hash(d 2332 int err = dir->d_op->d_hash(dir, name);
2366 if (unlikely(err < 0)) 2333 if (unlikely(err < 0))
2367 return ERR_PTR(err); 2334 return ERR_PTR(err);
2368 } 2335 }
2369 return d_lookup(dir, name); 2336 return d_lookup(dir, name);
2370 } 2337 }
2371 EXPORT_SYMBOL(d_hash_and_lookup); 2338 EXPORT_SYMBOL(d_hash_and_lookup);
2372 2339
2373 /* 2340 /*
2374 * When a file is deleted, we have two option 2341 * When a file is deleted, we have two options:
2375 * - turn this dentry into a negative dentry 2342 * - turn this dentry into a negative dentry
2376 * - unhash this dentry and free it. 2343 * - unhash this dentry and free it.
2377 * 2344 *
2378 * Usually, we want to just turn this into 2345 * Usually, we want to just turn this into
2379 * a negative dentry, but if anybody else is 2346 * a negative dentry, but if anybody else is
2380 * currently using the dentry or the inode 2347 * currently using the dentry or the inode
2381 * we can't do that and we fall back on remov 2348 * we can't do that and we fall back on removing
2382 * it from the hash queues and waiting for 2349 * it from the hash queues and waiting for
2383 * it to be deleted later when it has no user 2350 * it to be deleted later when it has no users
2384 */ 2351 */
2385 2352
2386 /** 2353 /**
2387 * d_delete - delete a dentry 2354 * d_delete - delete a dentry
2388 * @dentry: The dentry to delete 2355 * @dentry: The dentry to delete
2389 * 2356 *
2390 * Turn the dentry into a negative dentry if 2357 * Turn the dentry into a negative dentry if possible, otherwise
2391 * remove it from the hash queues so it can b 2358 * remove it from the hash queues so it can be deleted later
2392 */ 2359 */
2393 2360
2394 void d_delete(struct dentry * dentry) 2361 void d_delete(struct dentry * dentry)
2395 { 2362 {
2396 struct inode *inode = dentry->d_inode !! 2363 struct inode *inode;
2397 !! 2364 int isdir = 0;
2398 spin_lock(&inode->i_lock); <<
2399 spin_lock(&dentry->d_lock); <<
2400 /* 2365 /*
2401 * Are we the only user? 2366 * Are we the only user?
2402 */ 2367 */
>> 2368 again:
>> 2369 spin_lock(&dentry->d_lock);
>> 2370 inode = dentry->d_inode;
>> 2371 isdir = S_ISDIR(inode->i_mode);
2403 if (dentry->d_lockref.count == 1) { 2372 if (dentry->d_lockref.count == 1) {
>> 2373 if (!spin_trylock(&inode->i_lock)) {
>> 2374 spin_unlock(&dentry->d_lock);
>> 2375 cpu_relax();
>> 2376 goto again;
>> 2377 }
2404 dentry->d_flags &= ~DCACHE_CA 2378 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2405 dentry_unlink_inode(dentry); 2379 dentry_unlink_inode(dentry);
2406 } else { !! 2380 fsnotify_nameremove(dentry, isdir);
2407 __d_drop(dentry); !! 2381 return;
2408 spin_unlock(&dentry->d_lock); <<
2409 spin_unlock(&inode->i_lock); <<
2410 } 2382 }
>> 2383
>> 2384 if (!d_unhashed(dentry))
>> 2385 __d_drop(dentry);
>> 2386
>> 2387 spin_unlock(&dentry->d_lock);
>> 2388
>> 2389 fsnotify_nameremove(dentry, isdir);
2411 } 2390 }
2412 EXPORT_SYMBOL(d_delete); 2391 EXPORT_SYMBOL(d_delete);
2413 2392
2414 static void __d_rehash(struct dentry *entry) 2393 static void __d_rehash(struct dentry *entry)
2415 { 2394 {
2416 struct hlist_bl_head *b = d_hash(entr 2395 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2417 2396
2418 hlist_bl_lock(b); 2397 hlist_bl_lock(b);
2419 hlist_bl_add_head_rcu(&entry->d_hash, 2398 hlist_bl_add_head_rcu(&entry->d_hash, b);
2420 hlist_bl_unlock(b); 2399 hlist_bl_unlock(b);
2421 } 2400 }
2422 2401
2423 /** 2402 /**
2424 * d_rehash - add an entry back to the ha 2403 * d_rehash - add an entry back to the hash
2425 * @entry: dentry to add to the hash 2404 * @entry: dentry to add to the hash
2426 * 2405 *
2427 * Adds a dentry to the hash according to its 2406 * Adds a dentry to the hash according to its name.
2428 */ 2407 */
2429 2408
2430 void d_rehash(struct dentry * entry) 2409 void d_rehash(struct dentry * entry)
2431 { 2410 {
2432 spin_lock(&entry->d_lock); 2411 spin_lock(&entry->d_lock);
2433 __d_rehash(entry); 2412 __d_rehash(entry);
2434 spin_unlock(&entry->d_lock); 2413 spin_unlock(&entry->d_lock);
2435 } 2414 }
2436 EXPORT_SYMBOL(d_rehash); 2415 EXPORT_SYMBOL(d_rehash);
2437 2416
2438 static inline unsigned start_dir_add(struct i 2417 static inline unsigned start_dir_add(struct inode *dir)
2439 { 2418 {
2440 preempt_disable_nested(); !! 2419
2441 for (;;) { 2420 for (;;) {
2442 unsigned n = dir->i_dir_seq; 2421 unsigned n = dir->i_dir_seq;
2443 if (!(n & 1) && cmpxchg(&dir- 2422 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2444 return n; 2423 return n;
2445 cpu_relax(); 2424 cpu_relax();
2446 } 2425 }
2447 } 2426 }
2448 2427
2449 static inline void end_dir_add(struct inode * !! 2428 static inline void end_dir_add(struct inode *dir, unsigned n)
2450 wait_queue_hea <<
2451 { 2429 {
2452 smp_store_release(&dir->i_dir_seq, n 2430 smp_store_release(&dir->i_dir_seq, n + 2);
2453 preempt_enable_nested(); <<
2454 wake_up_all(d_wait); <<
2455 } 2431 }
2456 2432
2457 static void d_wait_lookup(struct dentry *dent 2433 static void d_wait_lookup(struct dentry *dentry)
2458 { 2434 {
2459 if (d_in_lookup(dentry)) { 2435 if (d_in_lookup(dentry)) {
2460 DECLARE_WAITQUEUE(wait, curre 2436 DECLARE_WAITQUEUE(wait, current);
2461 add_wait_queue(dentry->d_wait 2437 add_wait_queue(dentry->d_wait, &wait);
2462 do { 2438 do {
2463 set_current_state(TAS 2439 set_current_state(TASK_UNINTERRUPTIBLE);
2464 spin_unlock(&dentry-> 2440 spin_unlock(&dentry->d_lock);
2465 schedule(); 2441 schedule();
2466 spin_lock(&dentry->d_ 2442 spin_lock(&dentry->d_lock);
2467 } while (d_in_lookup(dentry)) 2443 } while (d_in_lookup(dentry));
2468 } 2444 }
2469 } 2445 }
2470 2446
2471 struct dentry *d_alloc_parallel(struct dentry 2447 struct dentry *d_alloc_parallel(struct dentry *parent,
2472 const struct 2448 const struct qstr *name,
2473 wait_queue_he 2449 wait_queue_head_t *wq)
2474 { 2450 {
2475 unsigned int hash = name->hash; 2451 unsigned int hash = name->hash;
2476 struct hlist_bl_head *b = in_lookup_h 2452 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2477 struct hlist_bl_node *node; 2453 struct hlist_bl_node *node;
2478 struct dentry *new = d_alloc(parent, 2454 struct dentry *new = d_alloc(parent, name);
2479 struct dentry *dentry; 2455 struct dentry *dentry;
2480 unsigned seq, r_seq, d_seq; 2456 unsigned seq, r_seq, d_seq;
2481 2457
2482 if (unlikely(!new)) 2458 if (unlikely(!new))
2483 return ERR_PTR(-ENOMEM); 2459 return ERR_PTR(-ENOMEM);
2484 2460
2485 retry: 2461 retry:
2486 rcu_read_lock(); 2462 rcu_read_lock();
2487 seq = smp_load_acquire(&parent->d_ino !! 2463 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2488 r_seq = read_seqbegin(&rename_lock); 2464 r_seq = read_seqbegin(&rename_lock);
2489 dentry = __d_lookup_rcu(parent, name, 2465 dentry = __d_lookup_rcu(parent, name, &d_seq);
2490 if (unlikely(dentry)) { 2466 if (unlikely(dentry)) {
2491 if (!lockref_get_not_dead(&de 2467 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2492 rcu_read_unlock(); 2468 rcu_read_unlock();
2493 goto retry; 2469 goto retry;
2494 } 2470 }
2495 if (read_seqcount_retry(&dent 2471 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2496 rcu_read_unlock(); 2472 rcu_read_unlock();
2497 dput(dentry); 2473 dput(dentry);
2498 goto retry; 2474 goto retry;
2499 } 2475 }
2500 rcu_read_unlock(); 2476 rcu_read_unlock();
2501 dput(new); 2477 dput(new);
2502 return dentry; 2478 return dentry;
2503 } 2479 }
2504 if (unlikely(read_seqretry(&rename_lo 2480 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2505 rcu_read_unlock(); 2481 rcu_read_unlock();
2506 goto retry; 2482 goto retry;
2507 } 2483 }
2508 <<
2509 if (unlikely(seq & 1)) { <<
2510 rcu_read_unlock(); <<
2511 goto retry; <<
2512 } <<
2513 <<
2514 hlist_bl_lock(b); 2484 hlist_bl_lock(b);
2515 if (unlikely(READ_ONCE(parent->d_inod !! 2485 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2516 hlist_bl_unlock(b); 2486 hlist_bl_unlock(b);
2517 rcu_read_unlock(); 2487 rcu_read_unlock();
2518 goto retry; 2488 goto retry;
2519 } 2489 }
2520 /* 2490 /*
2521 * No changes for the parent since th 2491 * No changes for the parent since the beginning of d_lookup().
2522 * Since all removals from the chain 2492 * Since all removals from the chain happen with hlist_bl_lock(),
2523 * any potential in-lookup matches ar 2493 * any potential in-lookup matches are going to stay here until
2524 * we unlock the chain. All fields a 2494 * we unlock the chain. All fields are stable in everything
2525 * we encounter. 2495 * we encounter.
2526 */ 2496 */
2527 hlist_bl_for_each_entry(dentry, node, 2497 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2528 if (dentry->d_name.hash != ha 2498 if (dentry->d_name.hash != hash)
2529 continue; 2499 continue;
2530 if (dentry->d_parent != paren 2500 if (dentry->d_parent != parent)
2531 continue; 2501 continue;
2532 if (!d_same_name(dentry, pare 2502 if (!d_same_name(dentry, parent, name))
2533 continue; 2503 continue;
2534 hlist_bl_unlock(b); 2504 hlist_bl_unlock(b);
2535 /* now we can try to grab a r 2505 /* now we can try to grab a reference */
2536 if (!lockref_get_not_dead(&de 2506 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2537 rcu_read_unlock(); 2507 rcu_read_unlock();
2538 goto retry; 2508 goto retry;
2539 } 2509 }
2540 2510
2541 rcu_read_unlock(); 2511 rcu_read_unlock();
2542 /* 2512 /*
2543 * somebody is likely to be s 2513 * somebody is likely to be still doing lookup for it;
2544 * wait for them to finish 2514 * wait for them to finish
2545 */ 2515 */
2546 spin_lock(&dentry->d_lock); 2516 spin_lock(&dentry->d_lock);
2547 d_wait_lookup(dentry); 2517 d_wait_lookup(dentry);
2548 /* 2518 /*
2549 * it's not in-lookup anymore 2519 * it's not in-lookup anymore; in principle we should repeat
2550 * everything from dcache loo 2520 * everything from dcache lookup, but it's likely to be what
2551 * d_lookup() would've found 2521 * d_lookup() would've found anyway. If it is, just return it;
2552 * otherwise we really have t 2522 * otherwise we really have to repeat the whole thing.
2553 */ 2523 */
2554 if (unlikely(dentry->d_name.h 2524 if (unlikely(dentry->d_name.hash != hash))
2555 goto mismatch; 2525 goto mismatch;
2556 if (unlikely(dentry->d_parent 2526 if (unlikely(dentry->d_parent != parent))
2557 goto mismatch; 2527 goto mismatch;
2558 if (unlikely(d_unhashed(dentr 2528 if (unlikely(d_unhashed(dentry)))
2559 goto mismatch; 2529 goto mismatch;
2560 if (unlikely(!d_same_name(den 2530 if (unlikely(!d_same_name(dentry, parent, name)))
2561 goto mismatch; 2531 goto mismatch;
2562 /* OK, it *is* a hashed match 2532 /* OK, it *is* a hashed match; return it */
2563 spin_unlock(&dentry->d_lock); 2533 spin_unlock(&dentry->d_lock);
2564 dput(new); 2534 dput(new);
2565 return dentry; 2535 return dentry;
2566 } 2536 }
2567 rcu_read_unlock(); 2537 rcu_read_unlock();
2568 /* we can't take ->d_lock here; it's 2538 /* we can't take ->d_lock here; it's OK, though. */
2569 new->d_flags |= DCACHE_PAR_LOOKUP; 2539 new->d_flags |= DCACHE_PAR_LOOKUP;
2570 new->d_wait = wq; 2540 new->d_wait = wq;
2571 hlist_bl_add_head(&new->d_u.d_in_look !! 2541 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2572 hlist_bl_unlock(b); 2542 hlist_bl_unlock(b);
2573 return new; 2543 return new;
2574 mismatch: 2544 mismatch:
2575 spin_unlock(&dentry->d_lock); 2545 spin_unlock(&dentry->d_lock);
2576 dput(dentry); 2546 dput(dentry);
2577 goto retry; 2547 goto retry;
2578 } 2548 }
2579 EXPORT_SYMBOL(d_alloc_parallel); 2549 EXPORT_SYMBOL(d_alloc_parallel);
2580 2550
2581 /* !! 2551 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 { 2552 {
2588 wait_queue_head_t *d_wait; !! 2553 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2589 struct hlist_bl_head *b; !! 2554 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); 2555 hlist_bl_lock(b);
2595 dentry->d_flags &= ~DCACHE_PAR_LOOKUP 2556 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2596 __hlist_bl_del(&dentry->d_u.d_in_look 2557 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2597 d_wait = dentry->d_wait; !! 2558 wake_up_all(dentry->d_wait);
2598 dentry->d_wait = NULL; 2559 dentry->d_wait = NULL;
2599 hlist_bl_unlock(b); 2560 hlist_bl_unlock(b);
2600 INIT_HLIST_NODE(&dentry->d_u.d_alias) 2561 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2601 INIT_LIST_HEAD(&dentry->d_lru); 2562 INIT_LIST_HEAD(&dentry->d_lru);
2602 return d_wait; <<
2603 } <<
2604 <<
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 } 2563 }
2611 EXPORT_SYMBOL(__d_lookup_unhash_wake); !! 2564 EXPORT_SYMBOL(__d_lookup_done);
2612 2565
2613 /* inode->i_lock held if inode is non-NULL */ 2566 /* inode->i_lock held if inode is non-NULL */
2614 2567
2615 static inline void __d_add(struct dentry *den 2568 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2616 { 2569 {
2617 wait_queue_head_t *d_wait; <<
2618 struct inode *dir = NULL; 2570 struct inode *dir = NULL;
2619 unsigned n; 2571 unsigned n;
2620 spin_lock(&dentry->d_lock); 2572 spin_lock(&dentry->d_lock);
2621 if (unlikely(d_in_lookup(dentry))) { 2573 if (unlikely(d_in_lookup(dentry))) {
2622 dir = dentry->d_parent->d_ino 2574 dir = dentry->d_parent->d_inode;
2623 n = start_dir_add(dir); 2575 n = start_dir_add(dir);
2624 d_wait = __d_lookup_unhash(de !! 2576 __d_lookup_done(dentry);
2625 } 2577 }
2626 if (inode) { 2578 if (inode) {
2627 unsigned add_flags = d_flags_ 2579 unsigned add_flags = d_flags_for_inode(inode);
2628 hlist_add_head(&dentry->d_u.d 2580 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2629 raw_write_seqcount_begin(&den 2581 raw_write_seqcount_begin(&dentry->d_seq);
2630 __d_set_inode_and_type(dentry 2582 __d_set_inode_and_type(dentry, inode, add_flags);
2631 raw_write_seqcount_end(&dentr 2583 raw_write_seqcount_end(&dentry->d_seq);
2632 fsnotify_update_flags(dentry) 2584 fsnotify_update_flags(dentry);
2633 } 2585 }
2634 __d_rehash(dentry); 2586 __d_rehash(dentry);
2635 if (dir) 2587 if (dir)
2636 end_dir_add(dir, n, d_wait); !! 2588 end_dir_add(dir, n);
2637 spin_unlock(&dentry->d_lock); 2589 spin_unlock(&dentry->d_lock);
2638 if (inode) 2590 if (inode)
2639 spin_unlock(&inode->i_lock); 2591 spin_unlock(&inode->i_lock);
2640 } 2592 }
2641 2593
2642 /** 2594 /**
2643 * d_add - add dentry to hash queues 2595 * d_add - add dentry to hash queues
2644 * @entry: dentry to add 2596 * @entry: dentry to add
2645 * @inode: The inode to attach to this dentry 2597 * @inode: The inode to attach to this dentry
2646 * 2598 *
2647 * This adds the entry to the hash queues and 2599 * This adds the entry to the hash queues and initializes @inode.
2648 * The entry was actually filled in earlier d 2600 * The entry was actually filled in earlier during d_alloc().
2649 */ 2601 */
2650 2602
2651 void d_add(struct dentry *entry, struct inode 2603 void d_add(struct dentry *entry, struct inode *inode)
2652 { 2604 {
2653 if (inode) { 2605 if (inode) {
2654 security_d_instantiate(entry, 2606 security_d_instantiate(entry, inode);
2655 spin_lock(&inode->i_lock); 2607 spin_lock(&inode->i_lock);
2656 } 2608 }
2657 __d_add(entry, inode); 2609 __d_add(entry, inode);
2658 } 2610 }
2659 EXPORT_SYMBOL(d_add); 2611 EXPORT_SYMBOL(d_add);
2660 2612
2661 /** 2613 /**
2662 * d_exact_alias - find and hash an exact unh 2614 * d_exact_alias - find and hash an exact unhashed alias
2663 * @entry: dentry to add 2615 * @entry: dentry to add
2664 * @inode: The inode to go with this dentry 2616 * @inode: The inode to go with this dentry
2665 * 2617 *
2666 * If an unhashed dentry with the same name/p 2618 * If an unhashed dentry with the same name/parent and desired
2667 * inode already exists, hash and return it. 2619 * inode already exists, hash and return it. Otherwise, return
2668 * NULL. 2620 * NULL.
2669 * 2621 *
2670 * Parent directory should be locked. 2622 * Parent directory should be locked.
2671 */ 2623 */
2672 struct dentry *d_exact_alias(struct dentry *e 2624 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2673 { 2625 {
2674 struct dentry *alias; 2626 struct dentry *alias;
2675 unsigned int hash = entry->d_name.has 2627 unsigned int hash = entry->d_name.hash;
2676 2628
2677 spin_lock(&inode->i_lock); 2629 spin_lock(&inode->i_lock);
2678 hlist_for_each_entry(alias, &inode->i 2630 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2679 /* 2631 /*
2680 * Don't need alias->d_lock h 2632 * Don't need alias->d_lock here, because aliases with
2681 * d_parent == entry->d_paren 2633 * d_parent == entry->d_parent are not subject to name or
2682 * parent changes, because th 2634 * parent changes, because the parent inode i_mutex is held.
2683 */ 2635 */
2684 if (alias->d_name.hash != has 2636 if (alias->d_name.hash != hash)
2685 continue; 2637 continue;
2686 if (alias->d_parent != entry- 2638 if (alias->d_parent != entry->d_parent)
2687 continue; 2639 continue;
2688 if (!d_same_name(alias, entry 2640 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2689 continue; 2641 continue;
2690 spin_lock(&alias->d_lock); 2642 spin_lock(&alias->d_lock);
2691 if (!d_unhashed(alias)) { 2643 if (!d_unhashed(alias)) {
2692 spin_unlock(&alias->d 2644 spin_unlock(&alias->d_lock);
2693 alias = NULL; 2645 alias = NULL;
2694 } else { 2646 } else {
2695 dget_dlock(alias); !! 2647 __dget_dlock(alias);
2696 __d_rehash(alias); 2648 __d_rehash(alias);
2697 spin_unlock(&alias->d 2649 spin_unlock(&alias->d_lock);
2698 } 2650 }
2699 spin_unlock(&inode->i_lock); 2651 spin_unlock(&inode->i_lock);
2700 return alias; 2652 return alias;
2701 } 2653 }
2702 spin_unlock(&inode->i_lock); 2654 spin_unlock(&inode->i_lock);
2703 return NULL; 2655 return NULL;
2704 } 2656 }
2705 EXPORT_SYMBOL(d_exact_alias); 2657 EXPORT_SYMBOL(d_exact_alias);
2706 2658
>> 2659 /**
>> 2660 * dentry_update_name_case - update case insensitive dentry with a new name
>> 2661 * @dentry: dentry to be updated
>> 2662 * @name: new name
>> 2663 *
>> 2664 * Update a case insensitive dentry with new case of name.
>> 2665 *
>> 2666 * dentry must have been returned by d_lookup with name @name. Old and new
>> 2667 * name lengths must match (ie. no d_compare which allows mismatched name
>> 2668 * lengths).
>> 2669 *
>> 2670 * Parent inode i_mutex must be held over d_lookup and into this call (to
>> 2671 * keep renames and concurrent inserts, and readdir(2) away).
>> 2672 */
>> 2673 void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
>> 2674 {
>> 2675 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
>> 2676 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
>> 2677
>> 2678 spin_lock(&dentry->d_lock);
>> 2679 write_seqcount_begin(&dentry->d_seq);
>> 2680 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
>> 2681 write_seqcount_end(&dentry->d_seq);
>> 2682 spin_unlock(&dentry->d_lock);
>> 2683 }
>> 2684 EXPORT_SYMBOL(dentry_update_name_case);
>> 2685
2707 static void swap_names(struct dentry *dentry, 2686 static void swap_names(struct dentry *dentry, struct dentry *target)
2708 { 2687 {
2709 if (unlikely(dname_external(target))) 2688 if (unlikely(dname_external(target))) {
2710 if (unlikely(dname_external(d 2689 if (unlikely(dname_external(dentry))) {
2711 /* 2690 /*
2712 * Both external: swa 2691 * Both external: swap the pointers
2713 */ 2692 */
2714 swap(target->d_name.n 2693 swap(target->d_name.name, dentry->d_name.name);
2715 } else { 2694 } else {
2716 /* 2695 /*
2717 * dentry:internal, t 2696 * dentry:internal, target:external. Steal target's
2718 * storage and make t 2697 * storage and make target internal.
2719 */ 2698 */
2720 memcpy(target->d_inam 2699 memcpy(target->d_iname, dentry->d_name.name,
2721 dentr 2700 dentry->d_name.len + 1);
2722 dentry->d_name.name = 2701 dentry->d_name.name = target->d_name.name;
2723 target->d_name.name = 2702 target->d_name.name = target->d_iname;
2724 } 2703 }
2725 } else { 2704 } else {
2726 if (unlikely(dname_external(d 2705 if (unlikely(dname_external(dentry))) {
2727 /* 2706 /*
2728 * dentry:external, t 2707 * dentry:external, target:internal. Give dentry's
2729 * storage to target 2708 * storage to target and make dentry internal
2730 */ 2709 */
2731 memcpy(dentry->d_inam 2710 memcpy(dentry->d_iname, target->d_name.name,
2732 targe 2711 target->d_name.len + 1);
2733 target->d_name.name = 2712 target->d_name.name = dentry->d_name.name;
2734 dentry->d_name.name = 2713 dentry->d_name.name = dentry->d_iname;
2735 } else { 2714 } else {
2736 /* 2715 /*
2737 * Both are internal. 2716 * Both are internal.
2738 */ 2717 */
2739 unsigned int i; 2718 unsigned int i;
2740 BUILD_BUG_ON(!IS_ALIG 2719 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2741 for (i = 0; i < DNAME 2720 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2742 swap(((long * 2721 swap(((long *) &dentry->d_iname)[i],
2743 ((long * 2722 ((long *) &target->d_iname)[i]);
2744 } 2723 }
2745 } 2724 }
2746 } 2725 }
2747 swap(dentry->d_name.hash_len, target- 2726 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2748 } 2727 }
2749 2728
2750 static void copy_name(struct dentry *dentry, 2729 static void copy_name(struct dentry *dentry, struct dentry *target)
2751 { 2730 {
2752 struct external_name *old_name = NULL 2731 struct external_name *old_name = NULL;
2753 if (unlikely(dname_external(dentry))) 2732 if (unlikely(dname_external(dentry)))
2754 old_name = external_name(dent 2733 old_name = external_name(dentry);
2755 if (unlikely(dname_external(target))) 2734 if (unlikely(dname_external(target))) {
2756 atomic_inc(&external_name(tar 2735 atomic_inc(&external_name(target)->u.count);
2757 dentry->d_name = target->d_na 2736 dentry->d_name = target->d_name;
2758 } else { 2737 } else {
2759 memcpy(dentry->d_iname, targe 2738 memcpy(dentry->d_iname, target->d_name.name,
2760 target->d_nam 2739 target->d_name.len + 1);
2761 dentry->d_name.name = dentry- 2740 dentry->d_name.name = dentry->d_iname;
2762 dentry->d_name.hash_len = tar 2741 dentry->d_name.hash_len = target->d_name.hash_len;
2763 } 2742 }
2764 if (old_name && likely(atomic_dec_and 2743 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2765 kfree_rcu(old_name, u.head); 2744 kfree_rcu(old_name, u.head);
2766 } 2745 }
2767 2746
>> 2747 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
>> 2748 {
>> 2749 /*
>> 2750 * XXXX: do we really need to take target->d_lock?
>> 2751 */
>> 2752 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
>> 2753 spin_lock(&target->d_parent->d_lock);
>> 2754 else {
>> 2755 if (d_ancestor(dentry->d_parent, target->d_parent)) {
>> 2756 spin_lock(&dentry->d_parent->d_lock);
>> 2757 spin_lock_nested(&target->d_parent->d_lock,
>> 2758 DENTRY_D_LOCK_NESTED);
>> 2759 } else {
>> 2760 spin_lock(&target->d_parent->d_lock);
>> 2761 spin_lock_nested(&dentry->d_parent->d_lock,
>> 2762 DENTRY_D_LOCK_NESTED);
>> 2763 }
>> 2764 }
>> 2765 if (target < dentry) {
>> 2766 spin_lock_nested(&target->d_lock, 2);
>> 2767 spin_lock_nested(&dentry->d_lock, 3);
>> 2768 } else {
>> 2769 spin_lock_nested(&dentry->d_lock, 2);
>> 2770 spin_lock_nested(&target->d_lock, 3);
>> 2771 }
>> 2772 }
>> 2773
>> 2774 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
>> 2775 {
>> 2776 if (target->d_parent != dentry->d_parent)
>> 2777 spin_unlock(&dentry->d_parent->d_lock);
>> 2778 if (target->d_parent != target)
>> 2779 spin_unlock(&target->d_parent->d_lock);
>> 2780 spin_unlock(&target->d_lock);
>> 2781 spin_unlock(&dentry->d_lock);
>> 2782 }
>> 2783
>> 2784 /*
>> 2785 * When switching names, the actual string doesn't strictly have to
>> 2786 * be preserved in the target - because we're dropping the target
>> 2787 * anyway. As such, we can just do a simple memcpy() to copy over
>> 2788 * the new name before we switch, unless we are going to rehash
>> 2789 * it. Note that if we *do* unhash the target, we are not allowed
>> 2790 * to rehash it without giving it a new name/hash key - whether
>> 2791 * we swap or overwrite the names here, resulting name won't match
>> 2792 * the reality in filesystem; it's only there for d_path() purposes.
>> 2793 * Note that all of this is happening under rename_lock, so the
>> 2794 * any hash lookup seeing it in the middle of manipulations will
>> 2795 * be discarded anyway. So we do not care what happens to the hash
>> 2796 * key in that case.
>> 2797 */
2768 /* 2798 /*
2769 * __d_move - move a dentry 2799 * __d_move - move a dentry
2770 * @dentry: entry to move 2800 * @dentry: entry to move
2771 * @target: new dentry 2801 * @target: new dentry
2772 * @exchange: exchange the two dentries 2802 * @exchange: exchange the two dentries
2773 * 2803 *
2774 * Update the dcache to reflect the move of a 2804 * Update the dcache to reflect the move of a file name. Negative
2775 * dcache entries should not be moved in this 2805 * dcache entries should not be moved in this way. Caller must hold
2776 * rename_lock, the i_mutex of the source and 2806 * rename_lock, the i_mutex of the source and target directories,
2777 * and the sb->s_vfs_rename_mutex if they dif 2807 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2778 */ 2808 */
2779 static void __d_move(struct dentry *dentry, s 2809 static void __d_move(struct dentry *dentry, struct dentry *target,
2780 bool exchange) 2810 bool exchange)
2781 { 2811 {
2782 struct dentry *old_parent, *p; <<
2783 wait_queue_head_t *d_wait; <<
2784 struct inode *dir = NULL; 2812 struct inode *dir = NULL;
2785 unsigned n; 2813 unsigned n;
>> 2814 if (!dentry->d_inode)
>> 2815 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2786 2816
2787 WARN_ON(!dentry->d_inode); !! 2817 BUG_ON(d_ancestor(dentry, target));
2788 if (WARN_ON(dentry == target)) <<
2789 return; <<
2790 <<
2791 BUG_ON(d_ancestor(target, dentry)); 2818 BUG_ON(d_ancestor(target, dentry));
2792 old_parent = dentry->d_parent; <<
2793 p = d_ancestor(old_parent, target); <<
2794 if (IS_ROOT(dentry)) { <<
2795 BUG_ON(p); <<
2796 spin_lock(&target->d_parent-> <<
2797 } else if (!p) { <<
2798 /* target is not a descendent <<
2799 spin_lock(&target->d_parent-> <<
2800 spin_lock_nested(&old_parent- <<
2801 } else { <<
2802 BUG_ON(p == dentry); <<
2803 spin_lock(&old_parent->d_lock <<
2804 if (p != target) <<
2805 spin_lock_nested(&tar <<
2806 DENTR <<
2807 } <<
2808 spin_lock_nested(&dentry->d_lock, 2); <<
2809 spin_lock_nested(&target->d_lock, 3); <<
2810 2819
>> 2820 dentry_lock_for_move(dentry, target);
2811 if (unlikely(d_in_lookup(target))) { 2821 if (unlikely(d_in_lookup(target))) {
2812 dir = target->d_parent->d_ino 2822 dir = target->d_parent->d_inode;
2813 n = start_dir_add(dir); 2823 n = start_dir_add(dir);
2814 d_wait = __d_lookup_unhash(ta !! 2824 __d_lookup_done(target);
2815 } 2825 }
2816 2826
2817 write_seqcount_begin(&dentry->d_seq); 2827 write_seqcount_begin(&dentry->d_seq);
2818 write_seqcount_begin_nested(&target-> 2828 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2819 2829
2820 /* unhash both */ 2830 /* unhash both */
2821 if (!d_unhashed(dentry)) !! 2831 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2822 ___d_drop(dentry); !! 2832 ___d_drop(dentry);
2823 if (!d_unhashed(target)) !! 2833 ___d_drop(target);
2824 ___d_drop(target); <<
2825 2834
2826 /* ... and switch them in the tree */ !! 2835 /* Switch the names.. */
2827 dentry->d_parent = target->d_parent; !! 2836 if (exchange)
2828 if (!exchange) { !! 2837 swap_names(dentry, target);
>> 2838 else
2829 copy_name(dentry, target); 2839 copy_name(dentry, target);
>> 2840
>> 2841 /* rehash in new place(s) */
>> 2842 __d_rehash(dentry);
>> 2843 if (exchange)
>> 2844 __d_rehash(target);
>> 2845 else
2830 target->d_hash.pprev = NULL; 2846 target->d_hash.pprev = NULL;
2831 dentry->d_parent->d_lockref.c !! 2847
2832 if (dentry != old_parent) /* !! 2848 /* ... and switch them in the tree */
2833 WARN_ON(!--old_parent !! 2849 if (IS_ROOT(dentry)) {
>> 2850 /* splicing a tree */
>> 2851 dentry->d_flags |= DCACHE_RCUACCESS;
>> 2852 dentry->d_parent = target->d_parent;
>> 2853 target->d_parent = target;
>> 2854 list_del_init(&target->d_child);
>> 2855 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2834 } else { 2856 } else {
2835 target->d_parent = old_parent !! 2857 /* swapping two dentries */
2836 swap_names(dentry, target); !! 2858 swap(dentry->d_parent, target->d_parent);
2837 if (!hlist_unhashed(&target-> !! 2859 list_move(&target->d_child, &target->d_parent->d_subdirs);
2838 __hlist_del(&target-> !! 2860 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2839 hlist_add_head(&target->d_sib !! 2861 if (exchange)
2840 __d_rehash(target); !! 2862 fsnotify_update_flags(target);
2841 fsnotify_update_flags(target) !! 2863 fsnotify_update_flags(dentry);
2842 } 2864 }
2843 if (!hlist_unhashed(&dentry->d_sib)) <<
2844 __hlist_del(&dentry->d_sib); <<
2845 hlist_add_head(&dentry->d_sib, &dentr <<
2846 __d_rehash(dentry); <<
2847 fsnotify_update_flags(dentry); <<
2848 fscrypt_handle_d_move(dentry); <<
2849 2865
2850 write_seqcount_end(&target->d_seq); 2866 write_seqcount_end(&target->d_seq);
2851 write_seqcount_end(&dentry->d_seq); 2867 write_seqcount_end(&dentry->d_seq);
2852 2868
2853 if (dir) 2869 if (dir)
2854 end_dir_add(dir, n, d_wait); !! 2870 end_dir_add(dir, n);
2855 !! 2871 dentry_unlock_for_move(dentry, target);
2856 if (dentry->d_parent != old_parent) <<
2857 spin_unlock(&dentry->d_parent <<
2858 if (dentry != old_parent) <<
2859 spin_unlock(&old_parent->d_lo <<
2860 spin_unlock(&target->d_lock); <<
2861 spin_unlock(&dentry->d_lock); <<
2862 } 2872 }
2863 2873
2864 /* 2874 /*
2865 * d_move - move a dentry 2875 * d_move - move a dentry
2866 * @dentry: entry to move 2876 * @dentry: entry to move
2867 * @target: new dentry 2877 * @target: new dentry
2868 * 2878 *
2869 * Update the dcache to reflect the move of a 2879 * Update the dcache to reflect the move of a file name. Negative
2870 * dcache entries should not be moved in this 2880 * dcache entries should not be moved in this way. See the locking
2871 * requirements for __d_move. 2881 * requirements for __d_move.
2872 */ 2882 */
2873 void d_move(struct dentry *dentry, struct den 2883 void d_move(struct dentry *dentry, struct dentry *target)
2874 { 2884 {
2875 write_seqlock(&rename_lock); 2885 write_seqlock(&rename_lock);
2876 __d_move(dentry, target, false); 2886 __d_move(dentry, target, false);
2877 write_sequnlock(&rename_lock); 2887 write_sequnlock(&rename_lock);
2878 } 2888 }
2879 EXPORT_SYMBOL(d_move); 2889 EXPORT_SYMBOL(d_move);
2880 2890
2881 /* 2891 /*
2882 * d_exchange - exchange two dentries 2892 * d_exchange - exchange two dentries
2883 * @dentry1: first dentry 2893 * @dentry1: first dentry
2884 * @dentry2: second dentry 2894 * @dentry2: second dentry
2885 */ 2895 */
2886 void d_exchange(struct dentry *dentry1, struc 2896 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2887 { 2897 {
2888 write_seqlock(&rename_lock); 2898 write_seqlock(&rename_lock);
2889 2899
2890 WARN_ON(!dentry1->d_inode); 2900 WARN_ON(!dentry1->d_inode);
2891 WARN_ON(!dentry2->d_inode); 2901 WARN_ON(!dentry2->d_inode);
2892 WARN_ON(IS_ROOT(dentry1)); 2902 WARN_ON(IS_ROOT(dentry1));
2893 WARN_ON(IS_ROOT(dentry2)); 2903 WARN_ON(IS_ROOT(dentry2));
2894 2904
2895 __d_move(dentry1, dentry2, true); 2905 __d_move(dentry1, dentry2, true);
2896 2906
2897 write_sequnlock(&rename_lock); 2907 write_sequnlock(&rename_lock);
2898 } 2908 }
2899 2909
2900 /** 2910 /**
2901 * d_ancestor - search for an ancestor 2911 * d_ancestor - search for an ancestor
2902 * @p1: ancestor dentry 2912 * @p1: ancestor dentry
2903 * @p2: child dentry 2913 * @p2: child dentry
2904 * 2914 *
2905 * Returns the ancestor dentry of p2 which is 2915 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2906 * an ancestor of p2, else NULL. 2916 * an ancestor of p2, else NULL.
2907 */ 2917 */
2908 struct dentry *d_ancestor(struct dentry *p1, 2918 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2909 { 2919 {
2910 struct dentry *p; 2920 struct dentry *p;
2911 2921
2912 for (p = p2; !IS_ROOT(p); p = p->d_pa 2922 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2913 if (p->d_parent == p1) 2923 if (p->d_parent == p1)
2914 return p; 2924 return p;
2915 } 2925 }
2916 return NULL; 2926 return NULL;
2917 } 2927 }
2918 2928
2919 /* 2929 /*
2920 * This helper attempts to cope with remotely 2930 * This helper attempts to cope with remotely renamed directories
2921 * 2931 *
2922 * It assumes that the caller is already hold 2932 * It assumes that the caller is already holding
2923 * dentry->d_parent->d_inode->i_mutex, and re 2933 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2924 * 2934 *
2925 * Note: If ever the locking in lock_rename() 2935 * Note: If ever the locking in lock_rename() changes, then please
2926 * remember to update this too... 2936 * remember to update this too...
2927 */ 2937 */
2928 static int __d_unalias(struct dentry *dentry, !! 2938 static int __d_unalias(struct inode *inode,
>> 2939 struct dentry *dentry, struct dentry *alias)
2929 { 2940 {
2930 struct mutex *m1 = NULL; 2941 struct mutex *m1 = NULL;
2931 struct rw_semaphore *m2 = NULL; 2942 struct rw_semaphore *m2 = NULL;
2932 int ret = -ESTALE; 2943 int ret = -ESTALE;
2933 2944
2934 /* If alias and dentry share a parent 2945 /* If alias and dentry share a parent, then no extra locks required */
2935 if (alias->d_parent == dentry->d_pare 2946 if (alias->d_parent == dentry->d_parent)
2936 goto out_unalias; 2947 goto out_unalias;
2937 2948
2938 /* See lock_rename() */ 2949 /* See lock_rename() */
2939 if (!mutex_trylock(&dentry->d_sb->s_v 2950 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2940 goto out_err; 2951 goto out_err;
2941 m1 = &dentry->d_sb->s_vfs_rename_mute 2952 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2942 if (!inode_trylock_shared(alias->d_pa 2953 if (!inode_trylock_shared(alias->d_parent->d_inode))
2943 goto out_err; 2954 goto out_err;
2944 m2 = &alias->d_parent->d_inode->i_rws 2955 m2 = &alias->d_parent->d_inode->i_rwsem;
2945 out_unalias: 2956 out_unalias:
2946 __d_move(alias, dentry, false); 2957 __d_move(alias, dentry, false);
2947 ret = 0; 2958 ret = 0;
2948 out_err: 2959 out_err:
2949 if (m2) 2960 if (m2)
2950 up_read(m2); 2961 up_read(m2);
2951 if (m1) 2962 if (m1)
2952 mutex_unlock(m1); 2963 mutex_unlock(m1);
2953 return ret; 2964 return ret;
2954 } 2965 }
2955 2966
2956 /** 2967 /**
2957 * d_splice_alias - splice a disconnected den 2968 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2958 * @inode: the inode which may have a discon 2969 * @inode: the inode which may have a disconnected dentry
2959 * @dentry: a negative dentry which we want t 2970 * @dentry: a negative dentry which we want to point to the inode.
2960 * 2971 *
2961 * If inode is a directory and has an IS_ROOT 2972 * 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 2973 * place of the given dentry and return it, else simply d_add the inode
2963 * to the dentry and return NULL. 2974 * to the dentry and return NULL.
2964 * 2975 *
2965 * If a non-IS_ROOT directory is found, the f 2976 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2966 * we should error out: directories can't hav 2977 * we should error out: directories can't have multiple aliases.
2967 * 2978 *
2968 * This is needed in the lookup routine of an 2979 * This is needed in the lookup routine of any filesystem that is exportable
2969 * (via knfsd) so that we can build dcache pa 2980 * (via knfsd) so that we can build dcache paths to directories effectively.
2970 * 2981 *
2971 * If a dentry was found and moved, then it i 2982 * If a dentry was found and moved, then it is returned. Otherwise NULL
2972 * is returned. This matches the expected re 2983 * is returned. This matches the expected return value of ->lookup.
2973 * 2984 *
2974 * Cluster filesystems may call this function 2985 * Cluster filesystems may call this function with a negative, hashed dentry.
2975 * In that case, we know that the inode will 2986 * 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 2987 * will only occur during atomic_open. So we need to check for the dentry
2977 * being already hashed only in the final cas 2988 * being already hashed only in the final case.
2978 */ 2989 */
2979 struct dentry *d_splice_alias(struct inode *i 2990 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2980 { 2991 {
2981 if (IS_ERR(inode)) 2992 if (IS_ERR(inode))
2982 return ERR_CAST(inode); 2993 return ERR_CAST(inode);
2983 2994
2984 BUG_ON(!d_unhashed(dentry)); 2995 BUG_ON(!d_unhashed(dentry));
2985 2996
2986 if (!inode) 2997 if (!inode)
2987 goto out; 2998 goto out;
2988 2999
2989 security_d_instantiate(dentry, inode) 3000 security_d_instantiate(dentry, inode);
2990 spin_lock(&inode->i_lock); 3001 spin_lock(&inode->i_lock);
2991 if (S_ISDIR(inode->i_mode)) { 3002 if (S_ISDIR(inode->i_mode)) {
2992 struct dentry *new = __d_find 3003 struct dentry *new = __d_find_any_alias(inode);
2993 if (unlikely(new)) { 3004 if (unlikely(new)) {
2994 /* The reference to n 3005 /* The reference to new ensures it remains an alias */
2995 spin_unlock(&inode->i 3006 spin_unlock(&inode->i_lock);
2996 write_seqlock(&rename 3007 write_seqlock(&rename_lock);
2997 if (unlikely(d_ancest 3008 if (unlikely(d_ancestor(new, dentry))) {
2998 write_sequnlo 3009 write_sequnlock(&rename_lock);
2999 dput(new); 3010 dput(new);
3000 new = ERR_PTR 3011 new = ERR_PTR(-ELOOP);
3001 pr_warn_ratel 3012 pr_warn_ratelimited(
3002 "VFS: 3013 "VFS: Lookup of '%s' in %s %s"
3003 " wou 3014 " would have caused loop\n",
3004 dentr 3015 dentry->d_name.name,
3005 inode 3016 inode->i_sb->s_type->name,
3006 inode 3017 inode->i_sb->s_id);
3007 } else if (!IS_ROOT(n 3018 } else if (!IS_ROOT(new)) {
3008 struct dentry !! 3019 int err = __d_unalias(inode, dentry, new);
3009 int err = __d <<
3010 write_sequnlo 3020 write_sequnlock(&rename_lock);
3011 if (err) { 3021 if (err) {
3012 dput( 3022 dput(new);
3013 new = 3023 new = ERR_PTR(err);
3014 } 3024 }
3015 dput(old_pare <<
3016 } else { 3025 } else {
3017 __d_move(new, 3026 __d_move(new, dentry, false);
3018 write_sequnlo 3027 write_sequnlock(&rename_lock);
3019 } 3028 }
3020 iput(inode); 3029 iput(inode);
3021 return new; 3030 return new;
3022 } 3031 }
3023 } 3032 }
3024 out: 3033 out:
3025 __d_add(dentry, inode); 3034 __d_add(dentry, inode);
3026 return NULL; 3035 return NULL;
3027 } 3036 }
3028 EXPORT_SYMBOL(d_splice_alias); 3037 EXPORT_SYMBOL(d_splice_alias);
3029 3038
>> 3039 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
>> 3040 {
>> 3041 *buflen -= namelen;
>> 3042 if (*buflen < 0)
>> 3043 return -ENAMETOOLONG;
>> 3044 *buffer -= namelen;
>> 3045 memcpy(*buffer, str, namelen);
>> 3046 return 0;
>> 3047 }
>> 3048
>> 3049 /**
>> 3050 * prepend_name - prepend a pathname in front of current buffer pointer
>> 3051 * @buffer: buffer pointer
>> 3052 * @buflen: allocated length of the buffer
>> 3053 * @name: name string and length qstr structure
>> 3054 *
>> 3055 * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
>> 3056 * make sure that either the old or the new name pointer and length are
>> 3057 * fetched. However, there may be mismatch between length and pointer.
>> 3058 * The length cannot be trusted, we need to copy it byte-by-byte until
>> 3059 * the length is reached or a null byte is found. It also prepends "/" at
>> 3060 * the beginning of the name. The sequence number check at the caller will
>> 3061 * retry it again when a d_move() does happen. So any garbage in the buffer
>> 3062 * due to mismatched pointer and length will be discarded.
>> 3063 *
>> 3064 * Data dependency barrier is needed to make sure that we see that terminating
>> 3065 * NUL. Alpha strikes again, film at 11...
>> 3066 */
>> 3067 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
>> 3068 {
>> 3069 const char *dname = READ_ONCE(name->name);
>> 3070 u32 dlen = READ_ONCE(name->len);
>> 3071 char *p;
>> 3072
>> 3073 smp_read_barrier_depends();
>> 3074
>> 3075 *buflen -= dlen + 1;
>> 3076 if (*buflen < 0)
>> 3077 return -ENAMETOOLONG;
>> 3078 p = *buffer -= dlen + 1;
>> 3079 *p++ = '/';
>> 3080 while (dlen--) {
>> 3081 char c = *dname++;
>> 3082 if (!c)
>> 3083 break;
>> 3084 *p++ = c;
>> 3085 }
>> 3086 return 0;
>> 3087 }
>> 3088
>> 3089 /**
>> 3090 * prepend_path - Prepend path string to a buffer
>> 3091 * @path: the dentry/vfsmount to report
>> 3092 * @root: root vfsmnt/dentry
>> 3093 * @buffer: pointer to the end of the buffer
>> 3094 * @buflen: pointer to buffer length
>> 3095 *
>> 3096 * The function will first try to write out the pathname without taking any
>> 3097 * lock other than the RCU read lock to make sure that dentries won't go away.
>> 3098 * It only checks the sequence number of the global rename_lock as any change
>> 3099 * in the dentry's d_seq will be preceded by changes in the rename_lock
>> 3100 * sequence number. If the sequence number had been changed, it will restart
>> 3101 * the whole pathname back-tracing sequence again by taking the rename_lock.
>> 3102 * In this case, there is no need to take the RCU read lock as the recursive
>> 3103 * parent pointer references will keep the dentry chain alive as long as no
>> 3104 * rename operation is performed.
>> 3105 */
>> 3106 static int prepend_path(const struct path *path,
>> 3107 const struct path *root,
>> 3108 char **buffer, int *buflen)
>> 3109 {
>> 3110 struct dentry *dentry;
>> 3111 struct vfsmount *vfsmnt;
>> 3112 struct mount *mnt;
>> 3113 int error = 0;
>> 3114 unsigned seq, m_seq = 0;
>> 3115 char *bptr;
>> 3116 int blen;
>> 3117
>> 3118 rcu_read_lock();
>> 3119 restart_mnt:
>> 3120 read_seqbegin_or_lock(&mount_lock, &m_seq);
>> 3121 seq = 0;
>> 3122 rcu_read_lock();
>> 3123 restart:
>> 3124 bptr = *buffer;
>> 3125 blen = *buflen;
>> 3126 error = 0;
>> 3127 dentry = path->dentry;
>> 3128 vfsmnt = path->mnt;
>> 3129 mnt = real_mount(vfsmnt);
>> 3130 read_seqbegin_or_lock(&rename_lock, &seq);
>> 3131 while (dentry != root->dentry || vfsmnt != root->mnt) {
>> 3132 struct dentry * parent;
>> 3133
>> 3134 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
>> 3135 struct mount *parent = READ_ONCE(mnt->mnt_parent);
>> 3136 /* Escaped? */
>> 3137 if (dentry != vfsmnt->mnt_root) {
>> 3138 bptr = *buffer;
>> 3139 blen = *buflen;
>> 3140 error = 3;
>> 3141 break;
>> 3142 }
>> 3143 /* Global root? */
>> 3144 if (mnt != parent) {
>> 3145 dentry = READ_ONCE(mnt->mnt_mountpoint);
>> 3146 mnt = parent;
>> 3147 vfsmnt = &mnt->mnt;
>> 3148 continue;
>> 3149 }
>> 3150 if (!error)
>> 3151 error = is_mounted(vfsmnt) ? 1 : 2;
>> 3152 break;
>> 3153 }
>> 3154 parent = dentry->d_parent;
>> 3155 prefetch(parent);
>> 3156 error = prepend_name(&bptr, &blen, &dentry->d_name);
>> 3157 if (error)
>> 3158 break;
>> 3159
>> 3160 dentry = parent;
>> 3161 }
>> 3162 if (!(seq & 1))
>> 3163 rcu_read_unlock();
>> 3164 if (need_seqretry(&rename_lock, seq)) {
>> 3165 seq = 1;
>> 3166 goto restart;
>> 3167 }
>> 3168 done_seqretry(&rename_lock, seq);
>> 3169
>> 3170 if (!(m_seq & 1))
>> 3171 rcu_read_unlock();
>> 3172 if (need_seqretry(&mount_lock, m_seq)) {
>> 3173 m_seq = 1;
>> 3174 goto restart_mnt;
>> 3175 }
>> 3176 done_seqretry(&mount_lock, m_seq);
>> 3177
>> 3178 if (error >= 0 && bptr == *buffer) {
>> 3179 if (--blen < 0)
>> 3180 error = -ENAMETOOLONG;
>> 3181 else
>> 3182 *--bptr = '/';
>> 3183 }
>> 3184 *buffer = bptr;
>> 3185 *buflen = blen;
>> 3186 return error;
>> 3187 }
>> 3188
>> 3189 /**
>> 3190 * __d_path - return the path of a dentry
>> 3191 * @path: the dentry/vfsmount to report
>> 3192 * @root: root vfsmnt/dentry
>> 3193 * @buf: buffer to return value in
>> 3194 * @buflen: buffer length
>> 3195 *
>> 3196 * Convert a dentry into an ASCII path name.
>> 3197 *
>> 3198 * Returns a pointer into the buffer or an error code if the
>> 3199 * path was too long.
>> 3200 *
>> 3201 * "buflen" should be positive.
>> 3202 *
>> 3203 * If the path is not reachable from the supplied root, return %NULL.
>> 3204 */
>> 3205 char *__d_path(const struct path *path,
>> 3206 const struct path *root,
>> 3207 char *buf, int buflen)
>> 3208 {
>> 3209 char *res = buf + buflen;
>> 3210 int error;
>> 3211
>> 3212 prepend(&res, &buflen, "\0", 1);
>> 3213 error = prepend_path(path, root, &res, &buflen);
>> 3214
>> 3215 if (error < 0)
>> 3216 return ERR_PTR(error);
>> 3217 if (error > 0)
>> 3218 return NULL;
>> 3219 return res;
>> 3220 }
>> 3221
>> 3222 char *d_absolute_path(const struct path *path,
>> 3223 char *buf, int buflen)
>> 3224 {
>> 3225 struct path root = {};
>> 3226 char *res = buf + buflen;
>> 3227 int error;
>> 3228
>> 3229 prepend(&res, &buflen, "\0", 1);
>> 3230 error = prepend_path(path, &root, &res, &buflen);
>> 3231
>> 3232 if (error > 1)
>> 3233 error = -EINVAL;
>> 3234 if (error < 0)
>> 3235 return ERR_PTR(error);
>> 3236 return res;
>> 3237 }
>> 3238
>> 3239 /*
>> 3240 * same as __d_path but appends "(deleted)" for unlinked files.
>> 3241 */
>> 3242 static int path_with_deleted(const struct path *path,
>> 3243 const struct path *root,
>> 3244 char **buf, int *buflen)
>> 3245 {
>> 3246 prepend(buf, buflen, "\0", 1);
>> 3247 if (d_unlinked(path->dentry)) {
>> 3248 int error = prepend(buf, buflen, " (deleted)", 10);
>> 3249 if (error)
>> 3250 return error;
>> 3251 }
>> 3252
>> 3253 return prepend_path(path, root, buf, buflen);
>> 3254 }
>> 3255
>> 3256 static int prepend_unreachable(char **buffer, int *buflen)
>> 3257 {
>> 3258 return prepend(buffer, buflen, "(unreachable)", 13);
>> 3259 }
>> 3260
>> 3261 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
>> 3262 {
>> 3263 unsigned seq;
>> 3264
>> 3265 do {
>> 3266 seq = read_seqcount_begin(&fs->seq);
>> 3267 *root = fs->root;
>> 3268 } while (read_seqcount_retry(&fs->seq, seq));
>> 3269 }
>> 3270
>> 3271 /**
>> 3272 * d_path - return the path of a dentry
>> 3273 * @path: path to report
>> 3274 * @buf: buffer to return value in
>> 3275 * @buflen: buffer length
>> 3276 *
>> 3277 * Convert a dentry into an ASCII path name. If the entry has been deleted
>> 3278 * the string " (deleted)" is appended. Note that this is ambiguous.
>> 3279 *
>> 3280 * Returns a pointer into the buffer or an error code if the path was
>> 3281 * too long. Note: Callers should use the returned pointer, not the passed
>> 3282 * in buffer, to use the name! The implementation often starts at an offset
>> 3283 * into the buffer, and may leave 0 bytes at the start.
>> 3284 *
>> 3285 * "buflen" should be positive.
>> 3286 */
>> 3287 char *d_path(const struct path *path, char *buf, int buflen)
>> 3288 {
>> 3289 char *res = buf + buflen;
>> 3290 struct path root;
>> 3291 int error;
>> 3292
>> 3293 /*
>> 3294 * We have various synthetic filesystems that never get mounted. On
>> 3295 * these filesystems dentries are never used for lookup purposes, and
>> 3296 * thus don't need to be hashed. They also don't need a name until a
>> 3297 * user wants to identify the object in /proc/pid/fd/. The little hack
>> 3298 * below allows us to generate a name for these objects on demand:
>> 3299 *
>> 3300 * Some pseudo inodes are mountable. When they are mounted
>> 3301 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
>> 3302 * and instead have d_path return the mounted path.
>> 3303 */
>> 3304 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
>> 3305 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
>> 3306 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
>> 3307
>> 3308 rcu_read_lock();
>> 3309 get_fs_root_rcu(current->fs, &root);
>> 3310 error = path_with_deleted(path, &root, &res, &buflen);
>> 3311 rcu_read_unlock();
>> 3312
>> 3313 if (error < 0)
>> 3314 res = ERR_PTR(error);
>> 3315 return res;
>> 3316 }
>> 3317 EXPORT_SYMBOL(d_path);
>> 3318
>> 3319 /*
>> 3320 * Helper function for dentry_operations.d_dname() members
>> 3321 */
>> 3322 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
>> 3323 const char *fmt, ...)
>> 3324 {
>> 3325 va_list args;
>> 3326 char temp[64];
>> 3327 int sz;
>> 3328
>> 3329 va_start(args, fmt);
>> 3330 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
>> 3331 va_end(args);
>> 3332
>> 3333 if (sz > sizeof(temp) || sz > buflen)
>> 3334 return ERR_PTR(-ENAMETOOLONG);
>> 3335
>> 3336 buffer += buflen - sz;
>> 3337 return memcpy(buffer, temp, sz);
>> 3338 }
>> 3339
>> 3340 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
>> 3341 {
>> 3342 char *end = buffer + buflen;
>> 3343 /* these dentries are never renamed, so d_lock is not needed */
>> 3344 if (prepend(&end, &buflen, " (deleted)", 11) ||
>> 3345 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
>> 3346 prepend(&end, &buflen, "/", 1))
>> 3347 end = ERR_PTR(-ENAMETOOLONG);
>> 3348 return end;
>> 3349 }
>> 3350 EXPORT_SYMBOL(simple_dname);
>> 3351
>> 3352 /*
>> 3353 * Write full pathname from the root of the filesystem into the buffer.
>> 3354 */
>> 3355 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
>> 3356 {
>> 3357 struct dentry *dentry;
>> 3358 char *end, *retval;
>> 3359 int len, seq = 0;
>> 3360 int error = 0;
>> 3361
>> 3362 if (buflen < 2)
>> 3363 goto Elong;
>> 3364
>> 3365 rcu_read_lock();
>> 3366 restart:
>> 3367 dentry = d;
>> 3368 end = buf + buflen;
>> 3369 len = buflen;
>> 3370 prepend(&end, &len, "\0", 1);
>> 3371 /* Get '/' right */
>> 3372 retval = end-1;
>> 3373 *retval = '/';
>> 3374 read_seqbegin_or_lock(&rename_lock, &seq);
>> 3375 while (!IS_ROOT(dentry)) {
>> 3376 struct dentry *parent = dentry->d_parent;
>> 3377
>> 3378 prefetch(parent);
>> 3379 error = prepend_name(&end, &len, &dentry->d_name);
>> 3380 if (error)
>> 3381 break;
>> 3382
>> 3383 retval = end;
>> 3384 dentry = parent;
>> 3385 }
>> 3386 if (!(seq & 1))
>> 3387 rcu_read_unlock();
>> 3388 if (need_seqretry(&rename_lock, seq)) {
>> 3389 seq = 1;
>> 3390 goto restart;
>> 3391 }
>> 3392 done_seqretry(&rename_lock, seq);
>> 3393 if (error)
>> 3394 goto Elong;
>> 3395 return retval;
>> 3396 Elong:
>> 3397 return ERR_PTR(-ENAMETOOLONG);
>> 3398 }
>> 3399
>> 3400 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
>> 3401 {
>> 3402 return __dentry_path(dentry, buf, buflen);
>> 3403 }
>> 3404 EXPORT_SYMBOL(dentry_path_raw);
>> 3405
>> 3406 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
>> 3407 {
>> 3408 char *p = NULL;
>> 3409 char *retval;
>> 3410
>> 3411 if (d_unlinked(dentry)) {
>> 3412 p = buf + buflen;
>> 3413 if (prepend(&p, &buflen, "//deleted", 10) != 0)
>> 3414 goto Elong;
>> 3415 buflen++;
>> 3416 }
>> 3417 retval = __dentry_path(dentry, buf, buflen);
>> 3418 if (!IS_ERR(retval) && p)
>> 3419 *p = '/'; /* restore '/' overriden with '\0' */
>> 3420 return retval;
>> 3421 Elong:
>> 3422 return ERR_PTR(-ENAMETOOLONG);
>> 3423 }
>> 3424
>> 3425 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
>> 3426 struct path *pwd)
>> 3427 {
>> 3428 unsigned seq;
>> 3429
>> 3430 do {
>> 3431 seq = read_seqcount_begin(&fs->seq);
>> 3432 *root = fs->root;
>> 3433 *pwd = fs->pwd;
>> 3434 } while (read_seqcount_retry(&fs->seq, seq));
>> 3435 }
>> 3436
>> 3437 /*
>> 3438 * NOTE! The user-level library version returns a
>> 3439 * character pointer. The kernel system call just
>> 3440 * returns the length of the buffer filled (which
>> 3441 * includes the ending '\0' character), or a negative
>> 3442 * error value. So libc would do something like
>> 3443 *
>> 3444 * char *getcwd(char * buf, size_t size)
>> 3445 * {
>> 3446 * int retval;
>> 3447 *
>> 3448 * retval = sys_getcwd(buf, size);
>> 3449 * if (retval >= 0)
>> 3450 * return buf;
>> 3451 * errno = -retval;
>> 3452 * return NULL;
>> 3453 * }
>> 3454 */
>> 3455 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
>> 3456 {
>> 3457 int error;
>> 3458 struct path pwd, root;
>> 3459 char *page = __getname();
>> 3460
>> 3461 if (!page)
>> 3462 return -ENOMEM;
>> 3463
>> 3464 rcu_read_lock();
>> 3465 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
>> 3466
>> 3467 error = -ENOENT;
>> 3468 if (!d_unlinked(pwd.dentry)) {
>> 3469 unsigned long len;
>> 3470 char *cwd = page + PATH_MAX;
>> 3471 int buflen = PATH_MAX;
>> 3472
>> 3473 prepend(&cwd, &buflen, "\0", 1);
>> 3474 error = prepend_path(&pwd, &root, &cwd, &buflen);
>> 3475 rcu_read_unlock();
>> 3476
>> 3477 if (error < 0)
>> 3478 goto out;
>> 3479
>> 3480 /* Unreachable from current root */
>> 3481 if (error > 0) {
>> 3482 error = prepend_unreachable(&cwd, &buflen);
>> 3483 if (error)
>> 3484 goto out;
>> 3485 }
>> 3486
>> 3487 error = -ERANGE;
>> 3488 len = PATH_MAX + page - cwd;
>> 3489 if (len <= size) {
>> 3490 error = len;
>> 3491 if (copy_to_user(buf, cwd, len))
>> 3492 error = -EFAULT;
>> 3493 }
>> 3494 } else {
>> 3495 rcu_read_unlock();
>> 3496 }
>> 3497
>> 3498 out:
>> 3499 __putname(page);
>> 3500 return error;
>> 3501 }
>> 3502
3030 /* 3503 /*
3031 * Test whether new_dentry is a subdirectory 3504 * Test whether new_dentry is a subdirectory of old_dentry.
3032 * 3505 *
3033 * Trivially implemented using the dcache str 3506 * Trivially implemented using the dcache structure
3034 */ 3507 */
3035 3508
3036 /** 3509 /**
3037 * is_subdir - is new dentry a subdirectory o 3510 * is_subdir - is new dentry a subdirectory of old_dentry
3038 * @new_dentry: new dentry 3511 * @new_dentry: new dentry
3039 * @old_dentry: old dentry 3512 * @old_dentry: old dentry
3040 * 3513 *
3041 * Returns true if new_dentry is a subdirecto 3514 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3042 * Returns false otherwise. 3515 * Returns false otherwise.
3043 * Caller must ensure that "new_dentry" is pi 3516 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3044 */ 3517 */
3045 3518
3046 bool is_subdir(struct dentry *new_dentry, str 3519 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3047 { 3520 {
3048 bool subdir; !! 3521 bool result;
3049 unsigned seq; 3522 unsigned seq;
3050 3523
3051 if (new_dentry == old_dentry) 3524 if (new_dentry == old_dentry)
3052 return true; 3525 return true;
3053 3526
3054 /* Access d_parent under rcu as d_mov !! 3527 do {
3055 rcu_read_lock(); !! 3528 /* for restarting inner loop in case of seq retry */
3056 seq = read_seqbegin(&rename_lock); !! 3529 seq = read_seqbegin(&rename_lock);
3057 subdir = d_ancestor(old_dentry, new_d !! 3530 /*
3058 /* Try lockless once... */ !! 3531 * Need rcu_readlock to protect against the d_parent trashing
3059 if (read_seqretry(&rename_lock, seq)) !! 3532 * due to d_move
3060 /* ...else acquire lock for p !! 3533 */
3061 read_seqlock_excl(&rename_loc !! 3534 rcu_read_lock();
3062 subdir = d_ancestor(old_dentr !! 3535 if (d_ancestor(old_dentry, new_dentry))
3063 read_sequnlock_excl(&rename_l !! 3536 result = true;
3064 } !! 3537 else
3065 rcu_read_unlock(); !! 3538 result = false;
3066 return subdir; !! 3539 rcu_read_unlock();
>> 3540 } while (read_seqretry(&rename_lock, seq));
>> 3541
>> 3542 return result;
3067 } 3543 }
3068 EXPORT_SYMBOL(is_subdir); <<
3069 3544
3070 static enum d_walk_ret d_genocide_kill(void * 3545 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3071 { 3546 {
3072 struct dentry *root = data; 3547 struct dentry *root = data;
3073 if (dentry != root) { 3548 if (dentry != root) {
3074 if (d_unhashed(dentry) || !de 3549 if (d_unhashed(dentry) || !dentry->d_inode)
3075 return D_WALK_SKIP; 3550 return D_WALK_SKIP;
3076 3551
3077 if (!(dentry->d_flags & DCACH 3552 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3078 dentry->d_flags |= DC 3553 dentry->d_flags |= DCACHE_GENOCIDE;
3079 dentry->d_lockref.cou 3554 dentry->d_lockref.count--;
3080 } 3555 }
3081 } 3556 }
3082 return D_WALK_CONTINUE; 3557 return D_WALK_CONTINUE;
3083 } 3558 }
3084 3559
3085 void d_genocide(struct dentry *parent) 3560 void d_genocide(struct dentry *parent)
3086 { 3561 {
3087 d_walk(parent, parent, d_genocide_kil !! 3562 d_walk(parent, parent, d_genocide_kill, NULL);
3088 } 3563 }
3089 3564
3090 void d_mark_tmpfile(struct file *file, struct !! 3565 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3091 { 3566 {
3092 struct dentry *dentry = file->f_path. !! 3567 inode_dec_link_count(inode);
3093 <<
3094 BUG_ON(dentry->d_name.name != dentry- 3568 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3095 !hlist_unhashed(&dentry->d_u. 3569 !hlist_unhashed(&dentry->d_u.d_alias) ||
3096 !d_unlinked(dentry)); 3570 !d_unlinked(dentry));
3097 spin_lock(&dentry->d_parent->d_lock); 3571 spin_lock(&dentry->d_parent->d_lock);
3098 spin_lock_nested(&dentry->d_lock, DEN 3572 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3099 dentry->d_name.len = sprintf(dentry-> 3573 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3100 (unsigned lon 3574 (unsigned long long)inode->i_ino);
3101 spin_unlock(&dentry->d_lock); 3575 spin_unlock(&dentry->d_lock);
3102 spin_unlock(&dentry->d_parent->d_lock 3576 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); 3577 d_instantiate(dentry, inode);
3113 } 3578 }
3114 EXPORT_SYMBOL(d_tmpfile); 3579 EXPORT_SYMBOL(d_tmpfile);
3115 3580
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 3581 static __initdata unsigned long dhash_entries;
3145 static int __init set_dhash_entries(char *str 3582 static int __init set_dhash_entries(char *str)
3146 { 3583 {
3147 if (!str) 3584 if (!str)
3148 return 0; 3585 return 0;
3149 dhash_entries = simple_strtoul(str, & 3586 dhash_entries = simple_strtoul(str, &str, 0);
3150 return 1; 3587 return 1;
3151 } 3588 }
3152 __setup("dhash_entries=", set_dhash_entries); 3589 __setup("dhash_entries=", set_dhash_entries);
3153 3590
3154 static void __init dcache_init_early(void) 3591 static void __init dcache_init_early(void)
3155 { 3592 {
3156 /* If hashes are distributed across N 3593 /* If hashes are distributed across NUMA nodes, defer
3157 * hash allocation until vmalloc spac 3594 * hash allocation until vmalloc space is available.
3158 */ 3595 */
3159 if (hashdist) 3596 if (hashdist)
3160 return; 3597 return;
3161 3598
3162 dentry_hashtable = 3599 dentry_hashtable =
3163 alloc_large_system_hash("Dent 3600 alloc_large_system_hash("Dentry cache",
3164 sizeo 3601 sizeof(struct hlist_bl_head),
3165 dhash 3602 dhash_entries,
3166 13, 3603 13,
3167 HASH_ 3604 HASH_EARLY | HASH_ZERO,
3168 &d_ha 3605 &d_hash_shift,
3169 NULL, !! 3606 &d_hash_mask,
3170 0, 3607 0,
3171 0); 3608 0);
3172 d_hash_shift = 32 - d_hash_shift; <<
3173 <<
3174 runtime_const_init(shift, d_hash_shif <<
3175 runtime_const_init(ptr, dentry_hashta <<
3176 } 3609 }
3177 3610
3178 static void __init dcache_init(void) 3611 static void __init dcache_init(void)
3179 { 3612 {
3180 /* 3613 /*
3181 * A constructor could be added for s 3614 * A constructor could be added for stable state like the lists,
3182 * but it is probably not worth it be 3615 * but it is probably not worth it because of the cache nature
3183 * of the dcache. 3616 * of the dcache.
3184 */ 3617 */
3185 dentry_cache = KMEM_CACHE_USERCOPY(de !! 3618 dentry_cache = KMEM_CACHE(dentry,
3186 SLAB_RECLAIM_ACCOUNT|SLAB_PAN !! 3619 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
3187 d_iname); <<
3188 3620
3189 /* Hash may have been set up in dcach 3621 /* Hash may have been set up in dcache_init_early */
3190 if (!hashdist) 3622 if (!hashdist)
3191 return; 3623 return;
3192 3624
3193 dentry_hashtable = 3625 dentry_hashtable =
3194 alloc_large_system_hash("Dent 3626 alloc_large_system_hash("Dentry cache",
3195 sizeo 3627 sizeof(struct hlist_bl_head),
3196 dhash 3628 dhash_entries,
3197 13, 3629 13,
3198 HASH_ 3630 HASH_ZERO,
3199 &d_ha 3631 &d_hash_shift,
3200 NULL, !! 3632 &d_hash_mask,
3201 0, 3633 0,
3202 0); 3634 0);
3203 d_hash_shift = 32 - d_hash_shift; <<
3204 <<
3205 runtime_const_init(shift, d_hash_shif <<
3206 runtime_const_init(ptr, dentry_hashta <<
3207 } 3635 }
3208 3636
3209 /* SLAB cache for __getname() consumers */ 3637 /* SLAB cache for __getname() consumers */
3210 struct kmem_cache *names_cachep __ro_after_in !! 3638 struct kmem_cache *names_cachep __read_mostly;
3211 EXPORT_SYMBOL(names_cachep); 3639 EXPORT_SYMBOL(names_cachep);
3212 3640
>> 3641 EXPORT_SYMBOL(d_genocide);
>> 3642
3213 void __init vfs_caches_init_early(void) 3643 void __init vfs_caches_init_early(void)
3214 { 3644 {
3215 int i; 3645 int i;
3216 3646
3217 for (i = 0; i < ARRAY_SIZE(in_lookup_ 3647 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3218 INIT_HLIST_BL_HEAD(&in_lookup 3648 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3219 3649
3220 dcache_init_early(); 3650 dcache_init_early();
3221 inode_init_early(); 3651 inode_init_early();
3222 } 3652 }
3223 3653
3224 void __init vfs_caches_init(void) 3654 void __init vfs_caches_init(void)
3225 { 3655 {
3226 names_cachep = kmem_cache_create_user !! 3656 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3227 SLAB_HWCACHE_ALIGN|SL !! 3657 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3228 3658
3229 dcache_init(); 3659 dcache_init();
3230 inode_init(); 3660 inode_init();
3231 files_init(); 3661 files_init();
3232 files_maxfiles_init(); 3662 files_maxfiles_init();
3233 mnt_init(); 3663 mnt_init();
3234 bdev_cache_init(); 3664 bdev_cache_init();
3235 chrdev_init(); 3665 chrdev_init();
3236 } 3666 }
3237 3667
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.