]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * fs/dcache.c | |
4 | * | |
5 | * Complete reimplementation | |
6 | * (C) 1997 Thomas Schoebel-Theuer, | |
7 | * with heavy changes by Linus Torvalds | |
8 | */ | |
9 | ||
10 | /* | |
11 | * Notes on the allocation strategy: | |
12 | * | |
13 | * The dcache is a master of the icache - whenever a dcache entry | |
14 | * exists, the inode will always exist. "iput()" is done either when | |
15 | * the dcache entry is deleted or garbage collected. | |
16 | */ | |
17 | ||
7a5cf791 | 18 | #include <linux/ratelimit.h> |
1da177e4 LT |
19 | #include <linux/string.h> |
20 | #include <linux/mm.h> | |
21 | #include <linux/fs.h> | |
0bf3d5c1 | 22 | #include <linux/fscrypt.h> |
7a91bf7f | 23 | #include <linux/fsnotify.h> |
1da177e4 LT |
24 | #include <linux/slab.h> |
25 | #include <linux/init.h> | |
1da177e4 LT |
26 | #include <linux/hash.h> |
27 | #include <linux/cache.h> | |
630d9c47 | 28 | #include <linux/export.h> |
1da177e4 LT |
29 | #include <linux/security.h> |
30 | #include <linux/seqlock.h> | |
57c8a661 | 31 | #include <linux/memblock.h> |
ceb5bdc2 NP |
32 | #include <linux/bit_spinlock.h> |
33 | #include <linux/rculist_bl.h> | |
f6041567 | 34 | #include <linux/list_lru.h> |
07f3f05c | 35 | #include "internal.h" |
b2dba1af | 36 | #include "mount.h" |
1da177e4 | 37 | |
789680d1 NP |
38 | /* |
39 | * Usage: | |
873feea0 | 40 | * dcache->d_inode->i_lock protects: |
946e51f2 | 41 | * - i_dentry, d_u.d_alias, d_inode of aliases |
ceb5bdc2 NP |
42 | * dcache_hash_bucket lock protects: |
43 | * - the dcache hash table | |
f1ee6162 N |
44 | * s_roots bl list spinlock protects: |
45 | * - the s_roots list (see __d_drop) | |
19156840 | 46 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
47 | * - the dcache lru lists and counters |
48 | * d_lock protects: | |
49 | * - d_flags | |
50 | * - d_name | |
51 | * - d_lru | |
b7ab39f6 | 52 | * - d_count |
da502956 | 53 | * - d_unhashed() |
2fd6b7f5 NP |
54 | * - d_parent and d_subdirs |
55 | * - childrens' d_child and d_parent | |
946e51f2 | 56 | * - d_u.d_alias, d_inode |
789680d1 NP |
57 | * |
58 | * Ordering: | |
873feea0 | 59 | * dentry->d_inode->i_lock |
b5c84bf6 | 60 | * dentry->d_lock |
19156840 | 61 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 | 62 | * dcache_hash_bucket lock |
f1ee6162 | 63 | * s_roots lock |
789680d1 | 64 | * |
da502956 NP |
65 | * If there is an ancestor relationship: |
66 | * dentry->d_parent->...->d_parent->d_lock | |
67 | * ... | |
68 | * dentry->d_parent->d_lock | |
69 | * dentry->d_lock | |
70 | * | |
71 | * If no ancestor relationship: | |
076515fc | 72 | * arbitrary, since it's serialized on rename_lock |
789680d1 | 73 | */ |
fa3536cc | 74 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
75 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
76 | ||
74c3cbe3 | 77 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 78 | |
949854d0 | 79 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 80 | |
e18b890b | 81 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 82 | |
cdf01226 DH |
83 | const struct qstr empty_name = QSTR_INIT("", 0); |
84 | EXPORT_SYMBOL(empty_name); | |
85 | const struct qstr slash_name = QSTR_INIT("/", 1); | |
86 | EXPORT_SYMBOL(slash_name); | |
87 | ||
1da177e4 LT |
88 | /* |
89 | * This is the single most critical data structure when it comes | |
90 | * to the dcache: the hashtable for lookups. Somebody should try | |
91 | * to make this good - I've just made it work. | |
92 | * | |
93 | * This hash-function tries to avoid losing too many bits of hash | |
94 | * information, yet avoid using a prime hash-size or similar. | |
95 | */ | |
1da177e4 | 96 | |
fa3536cc | 97 | static unsigned int d_hash_shift __read_mostly; |
ceb5bdc2 | 98 | |
b07ad996 | 99 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 100 | |
8387ff25 | 101 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 102 | { |
854d3e63 | 103 | return dentry_hashtable + (hash >> d_hash_shift); |
ceb5bdc2 NP |
104 | } |
105 | ||
94bdd655 AV |
106 | #define IN_LOOKUP_SHIFT 10 |
107 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
108 | ||
109 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
110 | unsigned int hash) | |
111 | { | |
112 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
113 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
114 | } | |
115 | ||
116 | ||
1da177e4 LT |
117 | /* Statistics gathering. */ |
118 | struct dentry_stat_t dentry_stat = { | |
119 | .age_limit = 45, | |
120 | }; | |
121 | ||
3942c07c | 122 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 123 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
af0c9af1 | 124 | static DEFINE_PER_CPU(long, nr_dentry_negative); |
312d3ca8 CH |
125 | |
126 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
127 | |
128 | /* | |
129 | * Here we resort to our own counters instead of using generic per-cpu counters | |
130 | * for consistency with what the vfs inode code does. We are expected to harvest | |
131 | * better code and performance by having our own specialized counters. | |
132 | * | |
133 | * Please note that the loop is done over all possible CPUs, not over all online | |
134 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
135 | * on and off. If one of them goes off, we will just keep their counters. | |
136 | * | |
137 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
138 | * please update all vfs counters to match. | |
139 | */ | |
3942c07c | 140 | static long get_nr_dentry(void) |
3e880fb5 NP |
141 | { |
142 | int i; | |
3942c07c | 143 | long sum = 0; |
3e880fb5 NP |
144 | for_each_possible_cpu(i) |
145 | sum += per_cpu(nr_dentry, i); | |
146 | return sum < 0 ? 0 : sum; | |
147 | } | |
148 | ||
62d36c77 DC |
149 | static long get_nr_dentry_unused(void) |
150 | { | |
151 | int i; | |
152 | long sum = 0; | |
153 | for_each_possible_cpu(i) | |
154 | sum += per_cpu(nr_dentry_unused, i); | |
155 | return sum < 0 ? 0 : sum; | |
156 | } | |
157 | ||
af0c9af1 WL |
158 | static long get_nr_dentry_negative(void) |
159 | { | |
160 | int i; | |
161 | long sum = 0; | |
162 | ||
163 | for_each_possible_cpu(i) | |
164 | sum += per_cpu(nr_dentry_negative, i); | |
165 | return sum < 0 ? 0 : sum; | |
166 | } | |
167 | ||
1f7e0616 | 168 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
169 | size_t *lenp, loff_t *ppos) |
170 | { | |
3e880fb5 | 171 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 172 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
af0c9af1 | 173 | dentry_stat.nr_negative = get_nr_dentry_negative(); |
3942c07c | 174 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
175 | } |
176 | #endif | |
177 | ||
5483f18e LT |
178 | /* |
179 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
180 | * The strings are both count bytes long, and count is non-zero. | |
181 | */ | |
e419b4cc LT |
182 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
183 | ||
184 | #include <asm/word-at-a-time.h> | |
185 | /* | |
186 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
187 | * aligned allocation for this particular component. We don't | |
188 | * strictly need the load_unaligned_zeropad() safety, but it | |
189 | * doesn't hurt either. | |
190 | * | |
191 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
192 | * need the careful unaligned handling. | |
193 | */ | |
94753db5 | 194 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 195 | { |
bfcfaa77 | 196 | unsigned long a,b,mask; |
bfcfaa77 LT |
197 | |
198 | for (;;) { | |
bfe7aa6c | 199 | a = read_word_at_a_time(cs); |
e419b4cc | 200 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
201 | if (tcount < sizeof(unsigned long)) |
202 | break; | |
203 | if (unlikely(a != b)) | |
204 | return 1; | |
205 | cs += sizeof(unsigned long); | |
206 | ct += sizeof(unsigned long); | |
207 | tcount -= sizeof(unsigned long); | |
208 | if (!tcount) | |
209 | return 0; | |
210 | } | |
a5c21dce | 211 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 212 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
213 | } |
214 | ||
bfcfaa77 | 215 | #else |
e419b4cc | 216 | |
94753db5 | 217 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 218 | { |
5483f18e LT |
219 | do { |
220 | if (*cs != *ct) | |
221 | return 1; | |
222 | cs++; | |
223 | ct++; | |
224 | tcount--; | |
225 | } while (tcount); | |
226 | return 0; | |
227 | } | |
228 | ||
e419b4cc LT |
229 | #endif |
230 | ||
94753db5 LT |
231 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
232 | { | |
94753db5 LT |
233 | /* |
234 | * Be careful about RCU walk racing with rename: | |
506458ef | 235 | * use 'READ_ONCE' to fetch the name pointer. |
94753db5 LT |
236 | * |
237 | * NOTE! Even if a rename will mean that the length | |
238 | * was not loaded atomically, we don't care. The | |
239 | * RCU walk will check the sequence count eventually, | |
240 | * and catch it. And we won't overrun the buffer, | |
241 | * because we're reading the name pointer atomically, | |
242 | * and a dentry name is guaranteed to be properly | |
243 | * terminated with a NUL byte. | |
244 | * | |
245 | * End result: even if 'len' is wrong, we'll exit | |
246 | * early because the data cannot match (there can | |
247 | * be no NUL in the ct/tcount data) | |
248 | */ | |
506458ef | 249 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
ae0a843c | 250 | |
6326c71f | 251 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
252 | } |
253 | ||
8d85b484 AV |
254 | struct external_name { |
255 | union { | |
256 | atomic_t count; | |
257 | struct rcu_head head; | |
258 | } u; | |
259 | unsigned char name[]; | |
260 | }; | |
261 | ||
262 | static inline struct external_name *external_name(struct dentry *dentry) | |
263 | { | |
264 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
265 | } | |
266 | ||
9c82ab9c | 267 | static void __d_free(struct rcu_head *head) |
1da177e4 | 268 | { |
9c82ab9c CH |
269 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
270 | ||
8d85b484 AV |
271 | kmem_cache_free(dentry_cache, dentry); |
272 | } | |
273 | ||
274 | static void __d_free_external(struct rcu_head *head) | |
275 | { | |
276 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
2e03b4bc | 277 | kfree(external_name(dentry)); |
f1782c9b | 278 | kmem_cache_free(dentry_cache, dentry); |
1da177e4 LT |
279 | } |
280 | ||
810bb172 AV |
281 | static inline int dname_external(const struct dentry *dentry) |
282 | { | |
283 | return dentry->d_name.name != dentry->d_iname; | |
284 | } | |
285 | ||
49d31c2f AV |
286 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
287 | { | |
288 | spin_lock(&dentry->d_lock); | |
230c6402 | 289 | name->name = dentry->d_name; |
49d31c2f | 290 | if (unlikely(dname_external(dentry))) { |
230c6402 | 291 | atomic_inc(&external_name(dentry)->u.count); |
49d31c2f | 292 | } else { |
6cd00a01 TH |
293 | memcpy(name->inline_name, dentry->d_iname, |
294 | dentry->d_name.len + 1); | |
230c6402 | 295 | name->name.name = name->inline_name; |
49d31c2f | 296 | } |
230c6402 | 297 | spin_unlock(&dentry->d_lock); |
49d31c2f AV |
298 | } |
299 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
300 | ||
301 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
302 | { | |
230c6402 | 303 | if (unlikely(name->name.name != name->inline_name)) { |
49d31c2f | 304 | struct external_name *p; |
230c6402 | 305 | p = container_of(name->name.name, struct external_name, name[0]); |
49d31c2f | 306 | if (unlikely(atomic_dec_and_test(&p->u.count))) |
2e03b4bc | 307 | kfree_rcu(p, u.head); |
49d31c2f AV |
308 | } |
309 | } | |
310 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
311 | ||
4bf46a27 DH |
312 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
313 | struct inode *inode, | |
314 | unsigned type_flags) | |
315 | { | |
316 | unsigned flags; | |
317 | ||
318 | dentry->d_inode = inode; | |
4bf46a27 DH |
319 | flags = READ_ONCE(dentry->d_flags); |
320 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
321 | flags |= type_flags; | |
322 | WRITE_ONCE(dentry->d_flags, flags); | |
323 | } | |
324 | ||
4bf46a27 DH |
325 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
326 | { | |
327 | unsigned flags = READ_ONCE(dentry->d_flags); | |
328 | ||
329 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
330 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 | 331 | dentry->d_inode = NULL; |
af0c9af1 WL |
332 | if (dentry->d_flags & DCACHE_LRU_LIST) |
333 | this_cpu_inc(nr_dentry_negative); | |
4bf46a27 DH |
334 | } |
335 | ||
b4f0354e AV |
336 | static void dentry_free(struct dentry *dentry) |
337 | { | |
946e51f2 | 338 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
339 | if (unlikely(dname_external(dentry))) { |
340 | struct external_name *p = external_name(dentry); | |
341 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
342 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
343 | return; | |
344 | } | |
345 | } | |
b4f0354e | 346 | /* if dentry was never visible to RCU, immediate free is OK */ |
5467a68c | 347 | if (dentry->d_flags & DCACHE_NORCU) |
b4f0354e AV |
348 | __d_free(&dentry->d_u.d_rcu); |
349 | else | |
350 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
351 | } | |
352 | ||
1da177e4 LT |
353 | /* |
354 | * Release the dentry's inode, using the filesystem | |
550dce01 | 355 | * d_iput() operation if defined. |
31e6b01f NP |
356 | */ |
357 | static void dentry_unlink_inode(struct dentry * dentry) | |
358 | __releases(dentry->d_lock) | |
873feea0 | 359 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
360 | { |
361 | struct inode *inode = dentry->d_inode; | |
a528aca7 | 362 | |
4c0d7cd5 | 363 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 364 | __d_clear_type_and_inode(dentry); |
946e51f2 | 365 | hlist_del_init(&dentry->d_u.d_alias); |
4c0d7cd5 | 366 | raw_write_seqcount_end(&dentry->d_seq); |
31e6b01f | 367 | spin_unlock(&dentry->d_lock); |
873feea0 | 368 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
369 | if (!inode->i_nlink) |
370 | fsnotify_inoderemove(inode); | |
371 | if (dentry->d_op && dentry->d_op->d_iput) | |
372 | dentry->d_op->d_iput(dentry, inode); | |
373 | else | |
374 | iput(inode); | |
375 | } | |
376 | ||
89dc77bc LT |
377 | /* |
378 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
379 | * is in use - which includes both the "real" per-superblock | |
380 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
381 | * | |
382 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
383 | * on the shrink list (ie not on the superblock LRU list). | |
384 | * | |
385 | * The per-cpu "nr_dentry_unused" counters are updated with | |
386 | * the DCACHE_LRU_LIST bit. | |
387 | * | |
af0c9af1 WL |
388 | * The per-cpu "nr_dentry_negative" counters are only updated |
389 | * when deleted from or added to the per-superblock LRU list, not | |
390 | * from/to the shrink list. That is to avoid an unneeded dec/inc | |
391 | * pair when moving from LRU to shrink list in select_collect(). | |
392 | * | |
89dc77bc LT |
393 | * These helper functions make sure we always follow the |
394 | * rules. d_lock must be held by the caller. | |
395 | */ | |
396 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
397 | static void d_lru_add(struct dentry *dentry) | |
398 | { | |
399 | D_FLAG_VERIFY(dentry, 0); | |
400 | dentry->d_flags |= DCACHE_LRU_LIST; | |
401 | this_cpu_inc(nr_dentry_unused); | |
af0c9af1 WL |
402 | if (d_is_negative(dentry)) |
403 | this_cpu_inc(nr_dentry_negative); | |
89dc77bc LT |
404 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
405 | } | |
406 | ||
407 | static void d_lru_del(struct dentry *dentry) | |
408 | { | |
409 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
410 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
411 | this_cpu_dec(nr_dentry_unused); | |
af0c9af1 WL |
412 | if (d_is_negative(dentry)) |
413 | this_cpu_dec(nr_dentry_negative); | |
89dc77bc LT |
414 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
415 | } | |
416 | ||
417 | static void d_shrink_del(struct dentry *dentry) | |
418 | { | |
419 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
420 | list_del_init(&dentry->d_lru); | |
421 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
422 | this_cpu_dec(nr_dentry_unused); | |
423 | } | |
424 | ||
425 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
426 | { | |
427 | D_FLAG_VERIFY(dentry, 0); | |
428 | list_add(&dentry->d_lru, list); | |
429 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
430 | this_cpu_inc(nr_dentry_unused); | |
431 | } | |
432 | ||
433 | /* | |
434 | * These can only be called under the global LRU lock, ie during the | |
435 | * callback for freeing the LRU list. "isolate" removes it from the | |
436 | * LRU lists entirely, while shrink_move moves it to the indicated | |
437 | * private list. | |
438 | */ | |
3f97b163 | 439 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
440 | { |
441 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
442 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
443 | this_cpu_dec(nr_dentry_unused); | |
af0c9af1 WL |
444 | if (d_is_negative(dentry)) |
445 | this_cpu_dec(nr_dentry_negative); | |
3f97b163 | 446 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
447 | } |
448 | ||
3f97b163 VD |
449 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
450 | struct list_head *list) | |
89dc77bc LT |
451 | { |
452 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
453 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
af0c9af1 WL |
454 | if (d_is_negative(dentry)) |
455 | this_cpu_dec(nr_dentry_negative); | |
3f97b163 | 456 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
457 | } |
458 | ||
789680d1 NP |
459 | /** |
460 | * d_drop - drop a dentry | |
461 | * @dentry: dentry to drop | |
462 | * | |
463 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
464 | * be found through a VFS lookup any more. Note that this is different from | |
465 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
466 | * possible, giving a successful _negative_ lookup, while d_drop will | |
467 | * just make the cache lookup fail. | |
468 | * | |
469 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
470 | * reason (NFS timeouts or autofs deletes). | |
471 | * | |
61647823 N |
472 | * __d_drop requires dentry->d_lock |
473 | * ___d_drop doesn't mark dentry as "unhashed" | |
474 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
789680d1 | 475 | */ |
61647823 | 476 | static void ___d_drop(struct dentry *dentry) |
789680d1 | 477 | { |
0632a9ac AV |
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_root, which are always IS_ROOT: | |
483 | */ | |
484 | if (unlikely(IS_ROOT(dentry))) | |
485 | b = &dentry->d_sb->s_roots; | |
486 | else | |
487 | b = d_hash(dentry->d_name.hash); | |
b61625d2 | 488 | |
0632a9ac AV |
489 | hlist_bl_lock(b); |
490 | __hlist_bl_del(&dentry->d_hash); | |
491 | hlist_bl_unlock(b); | |
789680d1 | 492 | } |
61647823 N |
493 | |
494 | void __d_drop(struct dentry *dentry) | |
495 | { | |
0632a9ac AV |
496 | if (!d_unhashed(dentry)) { |
497 | ___d_drop(dentry); | |
498 | dentry->d_hash.pprev = NULL; | |
499 | write_seqcount_invalidate(&dentry->d_seq); | |
500 | } | |
61647823 | 501 | } |
789680d1 NP |
502 | EXPORT_SYMBOL(__d_drop); |
503 | ||
504 | void d_drop(struct dentry *dentry) | |
505 | { | |
789680d1 NP |
506 | spin_lock(&dentry->d_lock); |
507 | __d_drop(dentry); | |
508 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
509 | } |
510 | EXPORT_SYMBOL(d_drop); | |
511 | ||
ba65dc5e AV |
512 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
513 | { | |
514 | struct dentry *next; | |
515 | /* | |
516 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
517 | * attached to the dentry tree | |
518 | */ | |
519 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
520 | if (unlikely(list_empty(&dentry->d_child))) | |
521 | return; | |
522 | __list_del_entry(&dentry->d_child); | |
523 | /* | |
524 | * Cursors can move around the list of children. While we'd been | |
525 | * a normal list member, it didn't matter - ->d_child.next would've | |
526 | * been updated. However, from now on it won't be and for the | |
527 | * things like d_walk() it might end up with a nasty surprise. | |
528 | * Normally d_walk() doesn't care about cursors moving around - | |
529 | * ->d_lock on parent prevents that and since a cursor has no children | |
530 | * of its own, we get through it without ever unlocking the parent. | |
531 | * There is one exception, though - if we ascend from a child that | |
532 | * gets killed as soon as we unlock it, the next sibling is found | |
533 | * using the value left in its ->d_child.next. And if _that_ | |
534 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
535 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
536 | * everything the cursor had been moved past. | |
537 | * | |
538 | * Solution: make sure that the pointer left behind in ->d_child.next | |
539 | * points to something that won't be moving around. I.e. skip the | |
540 | * cursors. | |
541 | */ | |
542 | while (dentry->d_child.next != &parent->d_subdirs) { | |
543 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
544 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
545 | break; | |
546 | dentry->d_child.next = next->d_child.next; | |
547 | } | |
548 | } | |
549 | ||
e55fd011 | 550 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 551 | { |
41edf278 AV |
552 | struct dentry *parent = NULL; |
553 | bool can_free = true; | |
41edf278 | 554 | if (!IS_ROOT(dentry)) |
77812a1e | 555 | parent = dentry->d_parent; |
31e6b01f | 556 | |
0d98439e LT |
557 | /* |
558 | * The dentry is now unrecoverably dead to the world. | |
559 | */ | |
560 | lockref_mark_dead(&dentry->d_lockref); | |
561 | ||
f0023bc6 | 562 | /* |
f0023bc6 SW |
563 | * inform the fs via d_prune that this dentry is about to be |
564 | * unhashed and destroyed. | |
565 | */ | |
29266201 | 566 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
567 | dentry->d_op->d_prune(dentry); |
568 | ||
01b60351 AV |
569 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
570 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
571 | d_lru_del(dentry); | |
01b60351 | 572 | } |
77812a1e NP |
573 | /* if it was on the hash then remove it */ |
574 | __d_drop(dentry); | |
ba65dc5e | 575 | dentry_unlist(dentry, parent); |
03b3b889 AV |
576 | if (parent) |
577 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
578 | if (dentry->d_inode) |
579 | dentry_unlink_inode(dentry); | |
580 | else | |
581 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
582 | this_cpu_dec(nr_dentry); |
583 | if (dentry->d_op && dentry->d_op->d_release) | |
584 | dentry->d_op->d_release(dentry); | |
585 | ||
41edf278 AV |
586 | spin_lock(&dentry->d_lock); |
587 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
588 | dentry->d_flags |= DCACHE_MAY_FREE; | |
589 | can_free = false; | |
590 | } | |
591 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
592 | if (likely(can_free)) |
593 | dentry_free(dentry); | |
9c5f1d30 | 594 | cond_resched(); |
e55fd011 AV |
595 | } |
596 | ||
8b987a46 | 597 | static struct dentry *__lock_parent(struct dentry *dentry) |
046b961b | 598 | { |
8b987a46 | 599 | struct dentry *parent; |
046b961b | 600 | rcu_read_lock(); |
c2338f2d | 601 | spin_unlock(&dentry->d_lock); |
046b961b | 602 | again: |
66702eb5 | 603 | parent = READ_ONCE(dentry->d_parent); |
046b961b AV |
604 | spin_lock(&parent->d_lock); |
605 | /* | |
606 | * We can't blindly lock dentry until we are sure | |
607 | * that we won't violate the locking order. | |
608 | * Any changes of dentry->d_parent must have | |
609 | * been done with parent->d_lock held, so | |
610 | * spin_lock() above is enough of a barrier | |
611 | * for checking if it's still our child. | |
612 | */ | |
613 | if (unlikely(parent != dentry->d_parent)) { | |
614 | spin_unlock(&parent->d_lock); | |
615 | goto again; | |
616 | } | |
65d8eb5a AV |
617 | rcu_read_unlock(); |
618 | if (parent != dentry) | |
9f12600f | 619 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
65d8eb5a | 620 | else |
046b961b AV |
621 | parent = NULL; |
622 | return parent; | |
623 | } | |
624 | ||
8b987a46 AV |
625 | static inline struct dentry *lock_parent(struct dentry *dentry) |
626 | { | |
627 | struct dentry *parent = dentry->d_parent; | |
628 | if (IS_ROOT(dentry)) | |
629 | return NULL; | |
630 | if (likely(spin_trylock(&parent->d_lock))) | |
631 | return parent; | |
632 | return __lock_parent(dentry); | |
633 | } | |
634 | ||
a338579f AV |
635 | static inline bool retain_dentry(struct dentry *dentry) |
636 | { | |
637 | WARN_ON(d_in_lookup(dentry)); | |
638 | ||
639 | /* Unreachable? Get rid of it */ | |
640 | if (unlikely(d_unhashed(dentry))) | |
641 | return false; | |
642 | ||
643 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
644 | return false; | |
645 | ||
646 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
647 | if (dentry->d_op->d_delete(dentry)) | |
648 | return false; | |
649 | } | |
62d9956c AV |
650 | /* retain; LRU fodder */ |
651 | dentry->d_lockref.count--; | |
652 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | |
653 | d_lru_add(dentry); | |
654 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) | |
655 | dentry->d_flags |= DCACHE_REFERENCED; | |
a338579f AV |
656 | return true; |
657 | } | |
658 | ||
c1d0c1a2 JO |
659 | /* |
660 | * Finish off a dentry we've decided to kill. | |
661 | * dentry->d_lock must be held, returns with it unlocked. | |
662 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
663 | */ | |
664 | static struct dentry *dentry_kill(struct dentry *dentry) | |
665 | __releases(dentry->d_lock) | |
666 | { | |
667 | struct inode *inode = dentry->d_inode; | |
668 | struct dentry *parent = NULL; | |
669 | ||
670 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
f657a666 | 671 | goto slow_positive; |
c1d0c1a2 JO |
672 | |
673 | if (!IS_ROOT(dentry)) { | |
674 | parent = dentry->d_parent; | |
675 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
f657a666 AV |
676 | parent = __lock_parent(dentry); |
677 | if (likely(inode || !dentry->d_inode)) | |
678 | goto got_locks; | |
679 | /* negative that became positive */ | |
680 | if (parent) | |
681 | spin_unlock(&parent->d_lock); | |
682 | inode = dentry->d_inode; | |
683 | goto slow_positive; | |
c1d0c1a2 JO |
684 | } |
685 | } | |
c1d0c1a2 JO |
686 | __dentry_kill(dentry); |
687 | return parent; | |
688 | ||
f657a666 AV |
689 | slow_positive: |
690 | spin_unlock(&dentry->d_lock); | |
691 | spin_lock(&inode->i_lock); | |
692 | spin_lock(&dentry->d_lock); | |
693 | parent = lock_parent(dentry); | |
694 | got_locks: | |
695 | if (unlikely(dentry->d_lockref.count != 1)) { | |
696 | dentry->d_lockref.count--; | |
697 | } else if (likely(!retain_dentry(dentry))) { | |
698 | __dentry_kill(dentry); | |
699 | return parent; | |
700 | } | |
701 | /* we are keeping it, after all */ | |
702 | if (inode) | |
703 | spin_unlock(&inode->i_lock); | |
704 | if (parent) | |
705 | spin_unlock(&parent->d_lock); | |
c1d0c1a2 | 706 | spin_unlock(&dentry->d_lock); |
f657a666 | 707 | return NULL; |
c1d0c1a2 JO |
708 | } |
709 | ||
360f5479 LT |
710 | /* |
711 | * Try to do a lockless dput(), and return whether that was successful. | |
712 | * | |
713 | * If unsuccessful, we return false, having already taken the dentry lock. | |
714 | * | |
715 | * The caller needs to hold the RCU read lock, so that the dentry is | |
716 | * guaranteed to stay around even if the refcount goes down to zero! | |
717 | */ | |
718 | static inline bool fast_dput(struct dentry *dentry) | |
719 | { | |
720 | int ret; | |
721 | unsigned int d_flags; | |
722 | ||
723 | /* | |
724 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 725 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
726 | */ |
727 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
728 | return lockref_put_or_lock(&dentry->d_lockref); | |
729 | ||
730 | /* | |
731 | * .. otherwise, we can try to just decrement the | |
732 | * lockref optimistically. | |
733 | */ | |
734 | ret = lockref_put_return(&dentry->d_lockref); | |
735 | ||
736 | /* | |
737 | * If the lockref_put_return() failed due to the lock being held | |
738 | * by somebody else, the fast path has failed. We will need to | |
739 | * get the lock, and then check the count again. | |
740 | */ | |
741 | if (unlikely(ret < 0)) { | |
742 | spin_lock(&dentry->d_lock); | |
743 | if (dentry->d_lockref.count > 1) { | |
744 | dentry->d_lockref.count--; | |
745 | spin_unlock(&dentry->d_lock); | |
7964410f | 746 | return true; |
360f5479 | 747 | } |
7964410f | 748 | return false; |
360f5479 LT |
749 | } |
750 | ||
751 | /* | |
752 | * If we weren't the last ref, we're done. | |
753 | */ | |
754 | if (ret) | |
7964410f | 755 | return true; |
360f5479 LT |
756 | |
757 | /* | |
758 | * Careful, careful. The reference count went down | |
759 | * to zero, but we don't hold the dentry lock, so | |
760 | * somebody else could get it again, and do another | |
761 | * dput(), and we need to not race with that. | |
762 | * | |
763 | * However, there is a very special and common case | |
764 | * where we don't care, because there is nothing to | |
765 | * do: the dentry is still hashed, it does not have | |
766 | * a 'delete' op, and it's referenced and already on | |
767 | * the LRU list. | |
768 | * | |
769 | * NOTE! Since we aren't locked, these values are | |
770 | * not "stable". However, it is sufficient that at | |
771 | * some point after we dropped the reference the | |
772 | * dentry was hashed and the flags had the proper | |
773 | * value. Other dentry users may have re-gotten | |
774 | * a reference to the dentry and change that, but | |
775 | * our work is done - we can leave the dentry | |
776 | * around with a zero refcount. | |
777 | */ | |
778 | smp_rmb(); | |
66702eb5 | 779 | d_flags = READ_ONCE(dentry->d_flags); |
75a6f82a | 780 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
781 | |
782 | /* Nothing to do? Dropping the reference was all we needed? */ | |
783 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
7964410f | 784 | return true; |
360f5479 LT |
785 | |
786 | /* | |
787 | * Not the fast normal case? Get the lock. We've already decremented | |
788 | * the refcount, but we'll need to re-check the situation after | |
789 | * getting the lock. | |
790 | */ | |
791 | spin_lock(&dentry->d_lock); | |
792 | ||
793 | /* | |
794 | * Did somebody else grab a reference to it in the meantime, and | |
795 | * we're no longer the last user after all? Alternatively, somebody | |
796 | * else could have killed it and marked it dead. Either way, we | |
797 | * don't need to do anything else. | |
798 | */ | |
799 | if (dentry->d_lockref.count) { | |
800 | spin_unlock(&dentry->d_lock); | |
7964410f | 801 | return true; |
360f5479 LT |
802 | } |
803 | ||
804 | /* | |
805 | * Re-get the reference we optimistically dropped. We hold the | |
806 | * lock, and we just tested that it was zero, so we can just | |
807 | * set it to 1. | |
808 | */ | |
809 | dentry->d_lockref.count = 1; | |
7964410f | 810 | return false; |
360f5479 LT |
811 | } |
812 | ||
813 | ||
1da177e4 LT |
814 | /* |
815 | * This is dput | |
816 | * | |
817 | * This is complicated by the fact that we do not want to put | |
818 | * dentries that are no longer on any hash chain on the unused | |
819 | * list: we'd much rather just get rid of them immediately. | |
820 | * | |
821 | * However, that implies that we have to traverse the dentry | |
822 | * tree upwards to the parents which might _also_ now be | |
823 | * scheduled for deletion (it may have been only waiting for | |
824 | * its last child to go away). | |
825 | * | |
826 | * This tail recursion is done by hand as we don't want to depend | |
827 | * on the compiler to always get this right (gcc generally doesn't). | |
828 | * Real recursion would eat up our stack space. | |
829 | */ | |
830 | ||
831 | /* | |
832 | * dput - release a dentry | |
833 | * @dentry: dentry to release | |
834 | * | |
835 | * Release a dentry. This will drop the usage count and if appropriate | |
836 | * call the dentry unlink method as well as removing it from the queues and | |
837 | * releasing its resources. If the parent dentries were scheduled for release | |
838 | * they too may now get deleted. | |
1da177e4 | 839 | */ |
1da177e4 LT |
840 | void dput(struct dentry *dentry) |
841 | { | |
1088a640 AV |
842 | while (dentry) { |
843 | might_sleep(); | |
1da177e4 | 844 | |
1088a640 AV |
845 | rcu_read_lock(); |
846 | if (likely(fast_dput(dentry))) { | |
847 | rcu_read_unlock(); | |
848 | return; | |
849 | } | |
47be6184 | 850 | |
1088a640 | 851 | /* Slow case: now with the dentry lock held */ |
360f5479 | 852 | rcu_read_unlock(); |
360f5479 | 853 | |
1088a640 AV |
854 | if (likely(retain_dentry(dentry))) { |
855 | spin_unlock(&dentry->d_lock); | |
856 | return; | |
857 | } | |
265ac902 | 858 | |
1088a640 | 859 | dentry = dentry_kill(dentry); |
47be6184 | 860 | } |
1da177e4 | 861 | } |
ec4f8605 | 862 | EXPORT_SYMBOL(dput); |
1da177e4 | 863 | |
9bdebc2b AV |
864 | static void __dput_to_list(struct dentry *dentry, struct list_head *list) |
865 | __must_hold(&dentry->d_lock) | |
866 | { | |
867 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
868 | /* let the owner of the list it's on deal with it */ | |
869 | --dentry->d_lockref.count; | |
870 | } else { | |
871 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
872 | d_lru_del(dentry); | |
873 | if (!--dentry->d_lockref.count) | |
874 | d_shrink_add(dentry, list); | |
875 | } | |
876 | } | |
877 | ||
878 | void dput_to_list(struct dentry *dentry, struct list_head *list) | |
879 | { | |
880 | rcu_read_lock(); | |
881 | if (likely(fast_dput(dentry))) { | |
882 | rcu_read_unlock(); | |
883 | return; | |
884 | } | |
885 | rcu_read_unlock(); | |
886 | if (!retain_dentry(dentry)) | |
887 | __dput_to_list(dentry, list); | |
888 | spin_unlock(&dentry->d_lock); | |
889 | } | |
1da177e4 | 890 | |
b5c84bf6 | 891 | /* This must be called with d_lock held */ |
dc0474be | 892 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 893 | { |
98474236 | 894 | dentry->d_lockref.count++; |
23044507 NP |
895 | } |
896 | ||
dc0474be | 897 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 898 | { |
98474236 | 899 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
900 | } |
901 | ||
b7ab39f6 NP |
902 | struct dentry *dget_parent(struct dentry *dentry) |
903 | { | |
df3d0bbc | 904 | int gotref; |
b7ab39f6 NP |
905 | struct dentry *ret; |
906 | ||
df3d0bbc WL |
907 | /* |
908 | * Do optimistic parent lookup without any | |
909 | * locking. | |
910 | */ | |
911 | rcu_read_lock(); | |
66702eb5 | 912 | ret = READ_ONCE(dentry->d_parent); |
df3d0bbc WL |
913 | gotref = lockref_get_not_zero(&ret->d_lockref); |
914 | rcu_read_unlock(); | |
915 | if (likely(gotref)) { | |
66702eb5 | 916 | if (likely(ret == READ_ONCE(dentry->d_parent))) |
df3d0bbc WL |
917 | return ret; |
918 | dput(ret); | |
919 | } | |
920 | ||
b7ab39f6 | 921 | repeat: |
a734eb45 NP |
922 | /* |
923 | * Don't need rcu_dereference because we re-check it was correct under | |
924 | * the lock. | |
925 | */ | |
926 | rcu_read_lock(); | |
b7ab39f6 | 927 | ret = dentry->d_parent; |
a734eb45 NP |
928 | spin_lock(&ret->d_lock); |
929 | if (unlikely(ret != dentry->d_parent)) { | |
930 | spin_unlock(&ret->d_lock); | |
931 | rcu_read_unlock(); | |
b7ab39f6 NP |
932 | goto repeat; |
933 | } | |
a734eb45 | 934 | rcu_read_unlock(); |
98474236 WL |
935 | BUG_ON(!ret->d_lockref.count); |
936 | ret->d_lockref.count++; | |
b7ab39f6 | 937 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
938 | return ret; |
939 | } | |
940 | EXPORT_SYMBOL(dget_parent); | |
941 | ||
61fec493 AV |
942 | static struct dentry * __d_find_any_alias(struct inode *inode) |
943 | { | |
944 | struct dentry *alias; | |
945 | ||
946 | if (hlist_empty(&inode->i_dentry)) | |
947 | return NULL; | |
948 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); | |
949 | __dget(alias); | |
950 | return alias; | |
951 | } | |
952 | ||
953 | /** | |
954 | * d_find_any_alias - find any alias for a given inode | |
955 | * @inode: inode to find an alias for | |
956 | * | |
957 | * If any aliases exist for the given inode, take and return a | |
958 | * reference for one of them. If no aliases exist, return %NULL. | |
959 | */ | |
960 | struct dentry *d_find_any_alias(struct inode *inode) | |
961 | { | |
962 | struct dentry *de; | |
963 | ||
964 | spin_lock(&inode->i_lock); | |
965 | de = __d_find_any_alias(inode); | |
966 | spin_unlock(&inode->i_lock); | |
967 | return de; | |
968 | } | |
969 | EXPORT_SYMBOL(d_find_any_alias); | |
970 | ||
1da177e4 LT |
971 | /** |
972 | * d_find_alias - grab a hashed alias of inode | |
973 | * @inode: inode in question | |
1da177e4 LT |
974 | * |
975 | * If inode has a hashed alias, or is a directory and has any alias, | |
976 | * acquire the reference to alias and return it. Otherwise return NULL. | |
977 | * Notice that if inode is a directory there can be only one alias and | |
978 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
979 | * of a filesystem, or if the directory was renamed and d_revalidate |
980 | * was the first vfs operation to notice. | |
1da177e4 | 981 | * |
21c0d8fd | 982 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 983 | * any other hashed alias over that one. |
1da177e4 | 984 | */ |
52ed46f0 | 985 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 986 | { |
61fec493 AV |
987 | struct dentry *alias; |
988 | ||
989 | if (S_ISDIR(inode->i_mode)) | |
990 | return __d_find_any_alias(inode); | |
1da177e4 | 991 | |
946e51f2 | 992 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 993 | spin_lock(&alias->d_lock); |
61fec493 | 994 | if (!d_unhashed(alias)) { |
8d80d7da BF |
995 | __dget_dlock(alias); |
996 | spin_unlock(&alias->d_lock); | |
997 | return alias; | |
1da177e4 | 998 | } |
da502956 | 999 | spin_unlock(&alias->d_lock); |
1da177e4 | 1000 | } |
da502956 | 1001 | return NULL; |
1da177e4 LT |
1002 | } |
1003 | ||
da502956 | 1004 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 1005 | { |
214fda1f DH |
1006 | struct dentry *de = NULL; |
1007 | ||
b3d9b7a3 | 1008 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 1009 | spin_lock(&inode->i_lock); |
52ed46f0 | 1010 | de = __d_find_alias(inode); |
873feea0 | 1011 | spin_unlock(&inode->i_lock); |
214fda1f | 1012 | } |
1da177e4 LT |
1013 | return de; |
1014 | } | |
ec4f8605 | 1015 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
1016 | |
1017 | /* | |
1018 | * Try to kill dentries associated with this inode. | |
1019 | * WARNING: you must own a reference to inode. | |
1020 | */ | |
1021 | void d_prune_aliases(struct inode *inode) | |
1022 | { | |
0cdca3f9 | 1023 | struct dentry *dentry; |
1da177e4 | 1024 | restart: |
873feea0 | 1025 | spin_lock(&inode->i_lock); |
946e51f2 | 1026 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 1027 | spin_lock(&dentry->d_lock); |
98474236 | 1028 | if (!dentry->d_lockref.count) { |
29355c39 AV |
1029 | struct dentry *parent = lock_parent(dentry); |
1030 | if (likely(!dentry->d_lockref.count)) { | |
1031 | __dentry_kill(dentry); | |
4a7795d3 | 1032 | dput(parent); |
29355c39 AV |
1033 | goto restart; |
1034 | } | |
1035 | if (parent) | |
1036 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
1037 | } |
1038 | spin_unlock(&dentry->d_lock); | |
1039 | } | |
873feea0 | 1040 | spin_unlock(&inode->i_lock); |
1da177e4 | 1041 | } |
ec4f8605 | 1042 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 1043 | |
3b3f09f4 AV |
1044 | /* |
1045 | * Lock a dentry from shrink list. | |
8f04da2a JO |
1046 | * Called under rcu_read_lock() and dentry->d_lock; the former |
1047 | * guarantees that nothing we access will be freed under us. | |
3b3f09f4 | 1048 | * Note that dentry is *not* protected from concurrent dentry_kill(), |
8f04da2a JO |
1049 | * d_delete(), etc. |
1050 | * | |
3b3f09f4 AV |
1051 | * Return false if dentry has been disrupted or grabbed, leaving |
1052 | * the caller to kick it off-list. Otherwise, return true and have | |
1053 | * that dentry's inode and parent both locked. | |
1054 | */ | |
1055 | static bool shrink_lock_dentry(struct dentry *dentry) | |
1da177e4 | 1056 | { |
3b3f09f4 AV |
1057 | struct inode *inode; |
1058 | struct dentry *parent; | |
da3bbdd4 | 1059 | |
3b3f09f4 AV |
1060 | if (dentry->d_lockref.count) |
1061 | return false; | |
1062 | ||
1063 | inode = dentry->d_inode; | |
1064 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
3b3f09f4 AV |
1065 | spin_unlock(&dentry->d_lock); |
1066 | spin_lock(&inode->i_lock); | |
ec33679d | 1067 | spin_lock(&dentry->d_lock); |
3b3f09f4 AV |
1068 | if (unlikely(dentry->d_lockref.count)) |
1069 | goto out; | |
1070 | /* changed inode means that somebody had grabbed it */ | |
1071 | if (unlikely(inode != dentry->d_inode)) | |
1072 | goto out; | |
3b3f09f4 | 1073 | } |
046b961b | 1074 | |
3b3f09f4 AV |
1075 | parent = dentry->d_parent; |
1076 | if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) | |
1077 | return true; | |
dd1f6b2e | 1078 | |
3b3f09f4 | 1079 | spin_unlock(&dentry->d_lock); |
3b3f09f4 AV |
1080 | spin_lock(&parent->d_lock); |
1081 | if (unlikely(parent != dentry->d_parent)) { | |
1082 | spin_unlock(&parent->d_lock); | |
1083 | spin_lock(&dentry->d_lock); | |
1084 | goto out; | |
1085 | } | |
1086 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
8f04da2a | 1087 | if (likely(!dentry->d_lockref.count)) |
3b3f09f4 | 1088 | return true; |
3b3f09f4 AV |
1089 | spin_unlock(&parent->d_lock); |
1090 | out: | |
1091 | if (inode) | |
1092 | spin_unlock(&inode->i_lock); | |
3b3f09f4 AV |
1093 | return false; |
1094 | } | |
77812a1e | 1095 | |
9bdebc2b | 1096 | void shrink_dentry_list(struct list_head *list) |
3b3f09f4 AV |
1097 | { |
1098 | while (!list_empty(list)) { | |
1099 | struct dentry *dentry, *parent; | |
64fd72e0 | 1100 | |
3b3f09f4 AV |
1101 | dentry = list_entry(list->prev, struct dentry, d_lru); |
1102 | spin_lock(&dentry->d_lock); | |
8f04da2a | 1103 | rcu_read_lock(); |
3b3f09f4 AV |
1104 | if (!shrink_lock_dentry(dentry)) { |
1105 | bool can_free = false; | |
8f04da2a | 1106 | rcu_read_unlock(); |
3b3f09f4 AV |
1107 | d_shrink_del(dentry); |
1108 | if (dentry->d_lockref.count < 0) | |
1109 | can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
64fd72e0 AV |
1110 | spin_unlock(&dentry->d_lock); |
1111 | if (can_free) | |
1112 | dentry_free(dentry); | |
1113 | continue; | |
1114 | } | |
8f04da2a | 1115 | rcu_read_unlock(); |
3b3f09f4 AV |
1116 | d_shrink_del(dentry); |
1117 | parent = dentry->d_parent; | |
9bdebc2b AV |
1118 | if (parent != dentry) |
1119 | __dput_to_list(parent, list); | |
ff2fde99 | 1120 | __dentry_kill(dentry); |
da3bbdd4 | 1121 | } |
3049cfe2 CH |
1122 | } |
1123 | ||
3f97b163 VD |
1124 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1125 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1126 | { |
1127 | struct list_head *freeable = arg; | |
1128 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1129 | ||
1130 | ||
1131 | /* | |
1132 | * we are inverting the lru lock/dentry->d_lock here, | |
1133 | * so use a trylock. If we fail to get the lock, just skip | |
1134 | * it | |
1135 | */ | |
1136 | if (!spin_trylock(&dentry->d_lock)) | |
1137 | return LRU_SKIP; | |
1138 | ||
1139 | /* | |
1140 | * Referenced dentries are still in use. If they have active | |
1141 | * counts, just remove them from the LRU. Otherwise give them | |
1142 | * another pass through the LRU. | |
1143 | */ | |
1144 | if (dentry->d_lockref.count) { | |
3f97b163 | 1145 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1146 | spin_unlock(&dentry->d_lock); |
1147 | return LRU_REMOVED; | |
1148 | } | |
1149 | ||
1150 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1151 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1152 | spin_unlock(&dentry->d_lock); | |
1153 | ||
1154 | /* | |
1155 | * The list move itself will be made by the common LRU code. At | |
1156 | * this point, we've dropped the dentry->d_lock but keep the | |
1157 | * lru lock. This is safe to do, since every list movement is | |
1158 | * protected by the lru lock even if both locks are held. | |
1159 | * | |
1160 | * This is guaranteed by the fact that all LRU management | |
1161 | * functions are intermediated by the LRU API calls like | |
1162 | * list_lru_add and list_lru_del. List movement in this file | |
1163 | * only ever occur through this functions or through callbacks | |
1164 | * like this one, that are called from the LRU API. | |
1165 | * | |
1166 | * The only exceptions to this are functions like | |
1167 | * shrink_dentry_list, and code that first checks for the | |
1168 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1169 | * operating only with stack provided lists after they are | |
1170 | * properly isolated from the main list. It is thus, always a | |
1171 | * local access. | |
1172 | */ | |
1173 | return LRU_ROTATE; | |
1174 | } | |
1175 | ||
3f97b163 | 1176 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1177 | spin_unlock(&dentry->d_lock); |
1178 | ||
1179 | return LRU_REMOVED; | |
1180 | } | |
1181 | ||
3049cfe2 | 1182 | /** |
b48f03b3 DC |
1183 | * prune_dcache_sb - shrink the dcache |
1184 | * @sb: superblock | |
503c358c | 1185 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1186 | * |
503c358c VD |
1187 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1188 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1189 | * function. |
3049cfe2 | 1190 | * |
b48f03b3 DC |
1191 | * This function may fail to free any resources if all the dentries are in |
1192 | * use. | |
3049cfe2 | 1193 | */ |
503c358c | 1194 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1195 | { |
f6041567 DC |
1196 | LIST_HEAD(dispose); |
1197 | long freed; | |
3049cfe2 | 1198 | |
503c358c VD |
1199 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1200 | dentry_lru_isolate, &dispose); | |
f6041567 | 1201 | shrink_dentry_list(&dispose); |
0a234c6d | 1202 | return freed; |
da3bbdd4 | 1203 | } |
23044507 | 1204 | |
4e717f5c | 1205 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1206 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1207 | { |
4e717f5c GC |
1208 | struct list_head *freeable = arg; |
1209 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1210 | |
4e717f5c GC |
1211 | /* |
1212 | * we are inverting the lru lock/dentry->d_lock here, | |
1213 | * so use a trylock. If we fail to get the lock, just skip | |
1214 | * it | |
1215 | */ | |
1216 | if (!spin_trylock(&dentry->d_lock)) | |
1217 | return LRU_SKIP; | |
1218 | ||
3f97b163 | 1219 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1220 | spin_unlock(&dentry->d_lock); |
ec33679d | 1221 | |
4e717f5c | 1222 | return LRU_REMOVED; |
da3bbdd4 KM |
1223 | } |
1224 | ||
4e717f5c | 1225 | |
1da177e4 LT |
1226 | /** |
1227 | * shrink_dcache_sb - shrink dcache for a superblock | |
1228 | * @sb: superblock | |
1229 | * | |
3049cfe2 CH |
1230 | * Shrink the dcache for the specified super block. This is used to free |
1231 | * the dcache before unmounting a file system. | |
1da177e4 | 1232 | */ |
3049cfe2 | 1233 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1234 | { |
4e717f5c GC |
1235 | do { |
1236 | LIST_HEAD(dispose); | |
1237 | ||
1dbd449c | 1238 | list_lru_walk(&sb->s_dentry_lru, |
b17c070f | 1239 | dentry_lru_isolate_shrink, &dispose, 1024); |
4e717f5c | 1240 | shrink_dentry_list(&dispose); |
b17c070f | 1241 | } while (list_lru_count(&sb->s_dentry_lru) > 0); |
1da177e4 | 1242 | } |
ec4f8605 | 1243 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1244 | |
db14fc3a MS |
1245 | /** |
1246 | * enum d_walk_ret - action to talke during tree walk | |
1247 | * @D_WALK_CONTINUE: contrinue walk | |
1248 | * @D_WALK_QUIT: quit walk | |
1249 | * @D_WALK_NORETRY: quit when retry is needed | |
1250 | * @D_WALK_SKIP: skip this dentry and its children | |
1251 | */ | |
1252 | enum d_walk_ret { | |
1253 | D_WALK_CONTINUE, | |
1254 | D_WALK_QUIT, | |
1255 | D_WALK_NORETRY, | |
1256 | D_WALK_SKIP, | |
1257 | }; | |
c826cb7d | 1258 | |
1da177e4 | 1259 | /** |
db14fc3a MS |
1260 | * d_walk - walk the dentry tree |
1261 | * @parent: start of walk | |
1262 | * @data: data passed to @enter() and @finish() | |
1263 | * @enter: callback when first entering the dentry | |
1da177e4 | 1264 | * |
3a8e3611 | 1265 | * The @enter() callbacks are called with d_lock held. |
1da177e4 | 1266 | */ |
8994cce9 | 1267 | void d_walk(struct dentry *parent, void *data, |
3a8e3611 | 1268 | enum d_walk_ret (*enter)(void *, struct dentry *)) |
1da177e4 | 1269 | { |
949854d0 | 1270 | struct dentry *this_parent; |
1da177e4 | 1271 | struct list_head *next; |
48f5ec21 | 1272 | unsigned seq = 0; |
db14fc3a MS |
1273 | enum d_walk_ret ret; |
1274 | bool retry = true; | |
949854d0 | 1275 | |
58db63d0 | 1276 | again: |
48f5ec21 | 1277 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1278 | this_parent = parent; |
2fd6b7f5 | 1279 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1280 | |
1281 | ret = enter(data, this_parent); | |
1282 | switch (ret) { | |
1283 | case D_WALK_CONTINUE: | |
1284 | break; | |
1285 | case D_WALK_QUIT: | |
1286 | case D_WALK_SKIP: | |
1287 | goto out_unlock; | |
1288 | case D_WALK_NORETRY: | |
1289 | retry = false; | |
1290 | break; | |
1291 | } | |
1da177e4 LT |
1292 | repeat: |
1293 | next = this_parent->d_subdirs.next; | |
1294 | resume: | |
1295 | while (next != &this_parent->d_subdirs) { | |
1296 | struct list_head *tmp = next; | |
946e51f2 | 1297 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1298 | next = tmp->next; |
2fd6b7f5 | 1299 | |
ba65dc5e AV |
1300 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1301 | continue; | |
1302 | ||
2fd6b7f5 | 1303 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1304 | |
1305 | ret = enter(data, dentry); | |
1306 | switch (ret) { | |
1307 | case D_WALK_CONTINUE: | |
1308 | break; | |
1309 | case D_WALK_QUIT: | |
2fd6b7f5 | 1310 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1311 | goto out_unlock; |
1312 | case D_WALK_NORETRY: | |
1313 | retry = false; | |
1314 | break; | |
1315 | case D_WALK_SKIP: | |
1316 | spin_unlock(&dentry->d_lock); | |
1317 | continue; | |
2fd6b7f5 | 1318 | } |
db14fc3a | 1319 | |
1da177e4 | 1320 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1321 | spin_unlock(&this_parent->d_lock); |
1322 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1323 | this_parent = dentry; |
2fd6b7f5 | 1324 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1325 | goto repeat; |
1326 | } | |
2fd6b7f5 | 1327 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1328 | } |
1329 | /* | |
1330 | * All done at this level ... ascend and resume the search. | |
1331 | */ | |
ca5358ef AV |
1332 | rcu_read_lock(); |
1333 | ascend: | |
1da177e4 | 1334 | if (this_parent != parent) { |
c826cb7d | 1335 | struct dentry *child = this_parent; |
31dec132 AV |
1336 | this_parent = child->d_parent; |
1337 | ||
31dec132 AV |
1338 | spin_unlock(&child->d_lock); |
1339 | spin_lock(&this_parent->d_lock); | |
1340 | ||
ca5358ef AV |
1341 | /* might go back up the wrong parent if we have had a rename. */ |
1342 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1343 | goto rename_retry; |
2159184e AV |
1344 | /* go into the first sibling still alive */ |
1345 | do { | |
1346 | next = child->d_child.next; | |
ca5358ef AV |
1347 | if (next == &this_parent->d_subdirs) |
1348 | goto ascend; | |
1349 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1350 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1351 | rcu_read_unlock(); |
1da177e4 LT |
1352 | goto resume; |
1353 | } | |
ca5358ef | 1354 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1355 | goto rename_retry; |
ca5358ef | 1356 | rcu_read_unlock(); |
db14fc3a MS |
1357 | |
1358 | out_unlock: | |
1359 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1360 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1361 | return; |
58db63d0 NP |
1362 | |
1363 | rename_retry: | |
ca5358ef AV |
1364 | spin_unlock(&this_parent->d_lock); |
1365 | rcu_read_unlock(); | |
1366 | BUG_ON(seq & 1); | |
db14fc3a MS |
1367 | if (!retry) |
1368 | return; | |
48f5ec21 | 1369 | seq = 1; |
58db63d0 | 1370 | goto again; |
1da177e4 | 1371 | } |
8994cce9 | 1372 | EXPORT_SYMBOL_GPL(d_walk); |
db14fc3a | 1373 | |
01619491 IK |
1374 | struct check_mount { |
1375 | struct vfsmount *mnt; | |
1376 | unsigned int mounted; | |
1377 | }; | |
1378 | ||
1379 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1380 | { | |
1381 | struct check_mount *info = data; | |
1382 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1383 | ||
1384 | if (likely(!d_mountpoint(dentry))) | |
1385 | return D_WALK_CONTINUE; | |
1386 | if (__path_is_mountpoint(&path)) { | |
1387 | info->mounted = 1; | |
1388 | return D_WALK_QUIT; | |
1389 | } | |
1390 | return D_WALK_CONTINUE; | |
1391 | } | |
1392 | ||
1393 | /** | |
1394 | * path_has_submounts - check for mounts over a dentry in the | |
1395 | * current namespace. | |
1396 | * @parent: path to check. | |
1397 | * | |
1398 | * Return true if the parent or its subdirectories contain | |
1399 | * a mount point in the current namespace. | |
1400 | */ | |
1401 | int path_has_submounts(const struct path *parent) | |
1402 | { | |
1403 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1404 | ||
1405 | read_seqlock_excl(&mount_lock); | |
3a8e3611 | 1406 | d_walk(parent->dentry, &data, path_check_mount); |
01619491 IK |
1407 | read_sequnlock_excl(&mount_lock); |
1408 | ||
1409 | return data.mounted; | |
1410 | } | |
1411 | EXPORT_SYMBOL(path_has_submounts); | |
1412 | ||
eed81007 MS |
1413 | /* |
1414 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1415 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1416 | * subtree can become unreachable). | |
1417 | * | |
1ffe46d1 | 1418 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1419 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1420 | */ | |
1421 | int d_set_mounted(struct dentry *dentry) | |
1422 | { | |
1423 | struct dentry *p; | |
1424 | int ret = -ENOENT; | |
1425 | write_seqlock(&rename_lock); | |
1426 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1427 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1428 | spin_lock(&p->d_lock); |
1429 | if (unlikely(d_unhashed(p))) { | |
1430 | spin_unlock(&p->d_lock); | |
1431 | goto out; | |
1432 | } | |
1433 | spin_unlock(&p->d_lock); | |
1434 | } | |
1435 | spin_lock(&dentry->d_lock); | |
1436 | if (!d_unlinked(dentry)) { | |
3895dbf8 EB |
1437 | ret = -EBUSY; |
1438 | if (!d_mountpoint(dentry)) { | |
1439 | dentry->d_flags |= DCACHE_MOUNTED; | |
1440 | ret = 0; | |
1441 | } | |
eed81007 MS |
1442 | } |
1443 | spin_unlock(&dentry->d_lock); | |
1444 | out: | |
1445 | write_sequnlock(&rename_lock); | |
1446 | return ret; | |
1447 | } | |
1448 | ||
1da177e4 | 1449 | /* |
fd517909 | 1450 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1451 | * and move any unused dentries to the end of the unused |
1452 | * list for prune_dcache(). We descend to the next level | |
1453 | * whenever the d_subdirs list is non-empty and continue | |
1454 | * searching. | |
1455 | * | |
1456 | * It returns zero iff there are no unused children, | |
1457 | * otherwise it returns the number of children moved to | |
1458 | * the end of the unused list. This may not be the total | |
1459 | * number of unused children, because select_parent can | |
1460 | * drop the lock and return early due to latency | |
1461 | * constraints. | |
1462 | */ | |
1da177e4 | 1463 | |
db14fc3a MS |
1464 | struct select_data { |
1465 | struct dentry *start; | |
9bdebc2b AV |
1466 | union { |
1467 | long found; | |
1468 | struct dentry *victim; | |
1469 | }; | |
db14fc3a | 1470 | struct list_head dispose; |
db14fc3a | 1471 | }; |
23044507 | 1472 | |
db14fc3a MS |
1473 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1474 | { | |
1475 | struct select_data *data = _data; | |
1476 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1477 | |
db14fc3a MS |
1478 | if (data->start == dentry) |
1479 | goto out; | |
2fd6b7f5 | 1480 | |
fe91522a | 1481 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1482 | data->found++; |
fe91522a AV |
1483 | } else { |
1484 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1485 | d_lru_del(dentry); | |
1486 | if (!dentry->d_lockref.count) { | |
1487 | d_shrink_add(dentry, &data->dispose); | |
1488 | data->found++; | |
1489 | } | |
1da177e4 | 1490 | } |
db14fc3a MS |
1491 | /* |
1492 | * We can return to the caller if we have found some (this | |
1493 | * ensures forward progress). We'll be coming back to find | |
1494 | * the rest. | |
1495 | */ | |
fe91522a AV |
1496 | if (!list_empty(&data->dispose)) |
1497 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1498 | out: |
db14fc3a | 1499 | return ret; |
1da177e4 LT |
1500 | } |
1501 | ||
9bdebc2b AV |
1502 | static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry) |
1503 | { | |
1504 | struct select_data *data = _data; | |
1505 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1506 | ||
1507 | if (data->start == dentry) | |
1508 | goto out; | |
1509 | ||
1510 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
1511 | if (!dentry->d_lockref.count) { | |
1512 | rcu_read_lock(); | |
1513 | data->victim = dentry; | |
1514 | return D_WALK_QUIT; | |
1515 | } | |
1516 | } else { | |
1517 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1518 | d_lru_del(dentry); | |
1519 | if (!dentry->d_lockref.count) | |
1520 | d_shrink_add(dentry, &data->dispose); | |
1521 | } | |
1522 | /* | |
1523 | * We can return to the caller if we have found some (this | |
1524 | * ensures forward progress). We'll be coming back to find | |
1525 | * the rest. | |
1526 | */ | |
1527 | if (!list_empty(&data->dispose)) | |
1528 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1529 | out: | |
1530 | return ret; | |
1531 | } | |
1532 | ||
1da177e4 LT |
1533 | /** |
1534 | * shrink_dcache_parent - prune dcache | |
1535 | * @parent: parent of entries to prune | |
1536 | * | |
1537 | * Prune the dcache to remove unused children of the parent dentry. | |
1538 | */ | |
db14fc3a | 1539 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1540 | { |
db14fc3a | 1541 | for (;;) { |
9bdebc2b | 1542 | struct select_data data = {.start = parent}; |
1da177e4 | 1543 | |
db14fc3a | 1544 | INIT_LIST_HEAD(&data.dispose); |
3a8e3611 | 1545 | d_walk(parent, &data, select_collect); |
4fb48871 AV |
1546 | |
1547 | if (!list_empty(&data.dispose)) { | |
1548 | shrink_dentry_list(&data.dispose); | |
1549 | continue; | |
1550 | } | |
1551 | ||
1552 | cond_resched(); | |
db14fc3a MS |
1553 | if (!data.found) |
1554 | break; | |
9bdebc2b AV |
1555 | data.victim = NULL; |
1556 | d_walk(parent, &data, select_collect2); | |
1557 | if (data.victim) { | |
1558 | struct dentry *parent; | |
1559 | spin_lock(&data.victim->d_lock); | |
1560 | if (!shrink_lock_dentry(data.victim)) { | |
1561 | spin_unlock(&data.victim->d_lock); | |
1562 | rcu_read_unlock(); | |
1563 | } else { | |
1564 | rcu_read_unlock(); | |
1565 | parent = data.victim->d_parent; | |
1566 | if (parent != data.victim) | |
1567 | __dput_to_list(parent, &data.dispose); | |
1568 | __dentry_kill(data.victim); | |
1569 | } | |
1570 | } | |
1571 | if (!list_empty(&data.dispose)) | |
1572 | shrink_dentry_list(&data.dispose); | |
421348f1 | 1573 | } |
1da177e4 | 1574 | } |
ec4f8605 | 1575 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1576 | |
9c8c10e2 | 1577 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1578 | { |
9c8c10e2 AV |
1579 | /* it has busy descendents; complain about those instead */ |
1580 | if (!list_empty(&dentry->d_subdirs)) | |
1581 | return D_WALK_CONTINUE; | |
42c32608 | 1582 | |
9c8c10e2 AV |
1583 | /* root with refcount 1 is fine */ |
1584 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1585 | return D_WALK_CONTINUE; | |
1586 | ||
1587 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1588 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1589 | dentry, |
1590 | dentry->d_inode ? | |
1591 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1592 | dentry, |
42c32608 AV |
1593 | dentry->d_lockref.count, |
1594 | dentry->d_sb->s_type->name, | |
1595 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1596 | WARN_ON(1); |
1597 | return D_WALK_CONTINUE; | |
1598 | } | |
1599 | ||
1600 | static void do_one_tree(struct dentry *dentry) | |
1601 | { | |
1602 | shrink_dcache_parent(dentry); | |
3a8e3611 | 1603 | d_walk(dentry, dentry, umount_check); |
9c8c10e2 AV |
1604 | d_drop(dentry); |
1605 | dput(dentry); | |
42c32608 AV |
1606 | } |
1607 | ||
1608 | /* | |
1609 | * destroy the dentries attached to a superblock on unmounting | |
1610 | */ | |
1611 | void shrink_dcache_for_umount(struct super_block *sb) | |
1612 | { | |
1613 | struct dentry *dentry; | |
1614 | ||
9c8c10e2 | 1615 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1616 | |
1617 | dentry = sb->s_root; | |
1618 | sb->s_root = NULL; | |
9c8c10e2 | 1619 | do_one_tree(dentry); |
42c32608 | 1620 | |
f1ee6162 N |
1621 | while (!hlist_bl_empty(&sb->s_roots)) { |
1622 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); | |
9c8c10e2 | 1623 | do_one_tree(dentry); |
42c32608 AV |
1624 | } |
1625 | } | |
1626 | ||
ff17fa56 | 1627 | static enum d_walk_ret find_submount(void *_data, struct dentry *dentry) |
848ac114 | 1628 | { |
ff17fa56 | 1629 | struct dentry **victim = _data; |
848ac114 | 1630 | if (d_mountpoint(dentry)) { |
8ed936b5 | 1631 | __dget_dlock(dentry); |
ff17fa56 | 1632 | *victim = dentry; |
848ac114 MS |
1633 | return D_WALK_QUIT; |
1634 | } | |
ff17fa56 | 1635 | return D_WALK_CONTINUE; |
848ac114 MS |
1636 | } |
1637 | ||
1638 | /** | |
1ffe46d1 EB |
1639 | * d_invalidate - detach submounts, prune dcache, and drop |
1640 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
848ac114 | 1641 | */ |
5542aa2f | 1642 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1643 | { |
ff17fa56 | 1644 | bool had_submounts = false; |
1ffe46d1 EB |
1645 | spin_lock(&dentry->d_lock); |
1646 | if (d_unhashed(dentry)) { | |
1647 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1648 | return; |
1ffe46d1 | 1649 | } |
ff17fa56 | 1650 | __d_drop(dentry); |
1ffe46d1 EB |
1651 | spin_unlock(&dentry->d_lock); |
1652 | ||
848ac114 | 1653 | /* Negative dentries can be dropped without further checks */ |
ff17fa56 | 1654 | if (!dentry->d_inode) |
5542aa2f | 1655 | return; |
848ac114 | 1656 | |
ff17fa56 | 1657 | shrink_dcache_parent(dentry); |
848ac114 | 1658 | for (;;) { |
ff17fa56 | 1659 | struct dentry *victim = NULL; |
3a8e3611 | 1660 | d_walk(dentry, &victim, find_submount); |
ff17fa56 AV |
1661 | if (!victim) { |
1662 | if (had_submounts) | |
1663 | shrink_dcache_parent(dentry); | |
81be24d2 | 1664 | return; |
8ed936b5 | 1665 | } |
ff17fa56 AV |
1666 | had_submounts = true; |
1667 | detach_mounts(victim); | |
1668 | dput(victim); | |
848ac114 | 1669 | } |
848ac114 | 1670 | } |
1ffe46d1 | 1671 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1672 | |
1da177e4 | 1673 | /** |
a4464dbc AV |
1674 | * __d_alloc - allocate a dcache entry |
1675 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1676 | * @name: qstr of the name |
1677 | * | |
1678 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1679 | * available. On a success the dentry is returned. The name passed in is | |
1680 | * copied and the copy passed in may be reused after this call. | |
1681 | */ | |
1682 | ||
a4464dbc | 1683 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1684 | { |
1685 | struct dentry *dentry; | |
1686 | char *dname; | |
285b102d | 1687 | int err; |
1da177e4 | 1688 | |
e12ba74d | 1689 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1690 | if (!dentry) |
1691 | return NULL; | |
1692 | ||
6326c71f LT |
1693 | /* |
1694 | * We guarantee that the inline name is always NUL-terminated. | |
1695 | * This way the memcpy() done by the name switching in rename | |
1696 | * will still always have a NUL at the end, even if we might | |
1697 | * be overwriting an internal NUL character | |
1698 | */ | |
1699 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd | 1700 | if (unlikely(!name)) { |
cdf01226 | 1701 | name = &slash_name; |
798434bd AV |
1702 | dname = dentry->d_iname; |
1703 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1704 | size_t size = offsetof(struct external_name, name[1]); |
2e03b4bc VB |
1705 | struct external_name *p = kmalloc(size + name->len, |
1706 | GFP_KERNEL_ACCOUNT | | |
1707 | __GFP_RECLAIMABLE); | |
1708 | if (!p) { | |
1da177e4 LT |
1709 | kmem_cache_free(dentry_cache, dentry); |
1710 | return NULL; | |
1711 | } | |
2e03b4bc VB |
1712 | atomic_set(&p->u.count, 1); |
1713 | dname = p->name; | |
1da177e4 LT |
1714 | } else { |
1715 | dname = dentry->d_iname; | |
1716 | } | |
1da177e4 LT |
1717 | |
1718 | dentry->d_name.len = name->len; | |
1719 | dentry->d_name.hash = name->hash; | |
1720 | memcpy(dname, name->name, name->len); | |
1721 | dname[name->len] = 0; | |
1722 | ||
6326c71f | 1723 | /* Make sure we always see the terminating NUL character */ |
7088efa9 | 1724 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ |
6326c71f | 1725 | |
98474236 | 1726 | dentry->d_lockref.count = 1; |
dea3667b | 1727 | dentry->d_flags = 0; |
1da177e4 | 1728 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1729 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1730 | dentry->d_inode = NULL; |
a4464dbc AV |
1731 | dentry->d_parent = dentry; |
1732 | dentry->d_sb = sb; | |
1da177e4 LT |
1733 | dentry->d_op = NULL; |
1734 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1735 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1736 | INIT_LIST_HEAD(&dentry->d_lru); |
1737 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1738 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1739 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1740 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1741 | |
285b102d MS |
1742 | if (dentry->d_op && dentry->d_op->d_init) { |
1743 | err = dentry->d_op->d_init(dentry); | |
1744 | if (err) { | |
1745 | if (dname_external(dentry)) | |
1746 | kfree(external_name(dentry)); | |
1747 | kmem_cache_free(dentry_cache, dentry); | |
1748 | return NULL; | |
1749 | } | |
1750 | } | |
1751 | ||
3e880fb5 | 1752 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1753 | |
1da177e4 LT |
1754 | return dentry; |
1755 | } | |
a4464dbc AV |
1756 | |
1757 | /** | |
1758 | * d_alloc - allocate a dcache entry | |
1759 | * @parent: parent of entry to allocate | |
1760 | * @name: qstr of the name | |
1761 | * | |
1762 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1763 | * available. On a success the dentry is returned. The name passed in is | |
1764 | * copied and the copy passed in may be reused after this call. | |
1765 | */ | |
1766 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1767 | { | |
1768 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1769 | if (!dentry) | |
1770 | return NULL; | |
a4464dbc AV |
1771 | spin_lock(&parent->d_lock); |
1772 | /* | |
1773 | * don't need child lock because it is not subject | |
1774 | * to concurrency here | |
1775 | */ | |
1776 | __dget_dlock(parent); | |
1777 | dentry->d_parent = parent; | |
946e51f2 | 1778 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1779 | spin_unlock(&parent->d_lock); |
1780 | ||
1781 | return dentry; | |
1782 | } | |
ec4f8605 | 1783 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1784 | |
f9c34674 MS |
1785 | struct dentry *d_alloc_anon(struct super_block *sb) |
1786 | { | |
1787 | return __d_alloc(sb, NULL); | |
1788 | } | |
1789 | EXPORT_SYMBOL(d_alloc_anon); | |
1790 | ||
ba65dc5e AV |
1791 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1792 | { | |
f9c34674 | 1793 | struct dentry *dentry = d_alloc_anon(parent->d_sb); |
ba65dc5e | 1794 | if (dentry) { |
5467a68c | 1795 | dentry->d_flags |= DCACHE_DENTRY_CURSOR; |
ba65dc5e AV |
1796 | dentry->d_parent = dget(parent); |
1797 | } | |
1798 | return dentry; | |
1799 | } | |
1800 | ||
e1a24bb0 BF |
1801 | /** |
1802 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1803 | * @sb: the superblock | |
1804 | * @name: qstr of the name | |
1805 | * | |
1806 | * For a filesystem that just pins its dentries in memory and never | |
1807 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
5467a68c AV |
1808 | * This is used for pipes, sockets et.al. - the stuff that should |
1809 | * never be anyone's children or parents. Unlike all other | |
1810 | * dentries, these will not have RCU delay between dropping the | |
1811 | * last reference and freeing them. | |
ab1152dd AV |
1812 | * |
1813 | * The only user is alloc_file_pseudo() and that's what should | |
1814 | * be considered a public interface. Don't use directly. | |
e1a24bb0 | 1815 | */ |
4b936885 NP |
1816 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1817 | { | |
5467a68c AV |
1818 | struct dentry *dentry = __d_alloc(sb, name); |
1819 | if (likely(dentry)) | |
1820 | dentry->d_flags |= DCACHE_NORCU; | |
1821 | return dentry; | |
4b936885 | 1822 | } |
4b936885 | 1823 | |
1da177e4 LT |
1824 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1825 | { | |
1826 | struct qstr q; | |
1827 | ||
1828 | q.name = name; | |
8387ff25 | 1829 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1830 | return d_alloc(parent, &q); |
1831 | } | |
ef26ca97 | 1832 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1833 | |
fb045adb NP |
1834 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1835 | { | |
6f7f7caa LT |
1836 | WARN_ON_ONCE(dentry->d_op); |
1837 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1838 | DCACHE_OP_COMPARE | |
1839 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1840 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1841 | DCACHE_OP_DELETE | |
d101a125 | 1842 | DCACHE_OP_REAL)); |
fb045adb NP |
1843 | dentry->d_op = op; |
1844 | if (!op) | |
1845 | return; | |
1846 | if (op->d_hash) | |
1847 | dentry->d_flags |= DCACHE_OP_HASH; | |
1848 | if (op->d_compare) | |
1849 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1850 | if (op->d_revalidate) | |
1851 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1852 | if (op->d_weak_revalidate) |
1853 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1854 | if (op->d_delete) |
1855 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1856 | if (op->d_prune) |
1857 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1858 | if (op->d_real) |
1859 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1860 | |
1861 | } | |
1862 | EXPORT_SYMBOL(d_set_d_op); | |
1863 | ||
df1a085a DH |
1864 | |
1865 | /* | |
1866 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1867 | * @dentry - The dentry to mark | |
1868 | * | |
1869 | * Mark a dentry as falling through to the lower layer (as set with | |
1870 | * d_pin_lower()). This flag may be recorded on the medium. | |
1871 | */ | |
1872 | void d_set_fallthru(struct dentry *dentry) | |
1873 | { | |
1874 | spin_lock(&dentry->d_lock); | |
1875 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1876 | spin_unlock(&dentry->d_lock); | |
1877 | } | |
1878 | EXPORT_SYMBOL(d_set_fallthru); | |
1879 | ||
b18825a7 DH |
1880 | static unsigned d_flags_for_inode(struct inode *inode) |
1881 | { | |
44bdb5e5 | 1882 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1883 | |
1884 | if (!inode) | |
1885 | return DCACHE_MISS_TYPE; | |
1886 | ||
1887 | if (S_ISDIR(inode->i_mode)) { | |
1888 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1889 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1890 | if (unlikely(!inode->i_op->lookup)) | |
1891 | add_flags = DCACHE_AUTODIR_TYPE; | |
1892 | else | |
1893 | inode->i_opflags |= IOP_LOOKUP; | |
1894 | } | |
44bdb5e5 DH |
1895 | goto type_determined; |
1896 | } | |
1897 | ||
1898 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1899 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1900 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1901 | goto type_determined; |
1902 | } | |
1903 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1904 | } |
1905 | ||
44bdb5e5 DH |
1906 | if (unlikely(!S_ISREG(inode->i_mode))) |
1907 | add_flags = DCACHE_SPECIAL_TYPE; | |
1908 | ||
1909 | type_determined: | |
b18825a7 DH |
1910 | if (unlikely(IS_AUTOMOUNT(inode))) |
1911 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1912 | return add_flags; | |
1913 | } | |
1914 | ||
360da900 OH |
1915 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1916 | { | |
b18825a7 | 1917 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1918 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1919 | |
b23fb0a6 | 1920 | spin_lock(&dentry->d_lock); |
af0c9af1 WL |
1921 | /* |
1922 | * Decrement negative dentry count if it was in the LRU list. | |
1923 | */ | |
1924 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1925 | this_cpu_dec(nr_dentry_negative); | |
de689f5e | 1926 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1927 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1928 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1929 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1930 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1931 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1932 | } |
1933 | ||
1da177e4 LT |
1934 | /** |
1935 | * d_instantiate - fill in inode information for a dentry | |
1936 | * @entry: dentry to complete | |
1937 | * @inode: inode to attach to this dentry | |
1938 | * | |
1939 | * Fill in inode information in the entry. | |
1940 | * | |
1941 | * This turns negative dentries into productive full members | |
1942 | * of society. | |
1943 | * | |
1944 | * NOTE! This assumes that the inode count has been incremented | |
1945 | * (or otherwise set) by the caller to indicate that it is now | |
1946 | * in use by the dcache. | |
1947 | */ | |
1948 | ||
1949 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1950 | { | |
946e51f2 | 1951 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1952 | if (inode) { |
b9680917 | 1953 | security_d_instantiate(entry, inode); |
873feea0 | 1954 | spin_lock(&inode->i_lock); |
de689f5e | 1955 | __d_instantiate(entry, inode); |
873feea0 | 1956 | spin_unlock(&inode->i_lock); |
de689f5e | 1957 | } |
1da177e4 | 1958 | } |
ec4f8605 | 1959 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1960 | |
1e2e547a AV |
1961 | /* |
1962 | * This should be equivalent to d_instantiate() + unlock_new_inode(), | |
1963 | * with lockdep-related part of unlock_new_inode() done before | |
1964 | * anything else. Use that instead of open-coding d_instantiate()/ | |
1965 | * unlock_new_inode() combinations. | |
1966 | */ | |
1967 | void d_instantiate_new(struct dentry *entry, struct inode *inode) | |
1968 | { | |
1969 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); | |
1970 | BUG_ON(!inode); | |
1971 | lockdep_annotate_inode_mutex_key(inode); | |
1972 | security_d_instantiate(entry, inode); | |
1973 | spin_lock(&inode->i_lock); | |
1974 | __d_instantiate(entry, inode); | |
1975 | WARN_ON(!(inode->i_state & I_NEW)); | |
c2b6d621 | 1976 | inode->i_state &= ~I_NEW & ~I_CREATING; |
1e2e547a AV |
1977 | smp_mb(); |
1978 | wake_up_bit(&inode->i_state, __I_NEW); | |
1979 | spin_unlock(&inode->i_lock); | |
1980 | } | |
1981 | EXPORT_SYMBOL(d_instantiate_new); | |
1982 | ||
adc0e91a AV |
1983 | struct dentry *d_make_root(struct inode *root_inode) |
1984 | { | |
1985 | struct dentry *res = NULL; | |
1986 | ||
1987 | if (root_inode) { | |
f9c34674 | 1988 | res = d_alloc_anon(root_inode->i_sb); |
5467a68c | 1989 | if (res) |
adc0e91a | 1990 | d_instantiate(res, root_inode); |
5467a68c | 1991 | else |
adc0e91a AV |
1992 | iput(root_inode); |
1993 | } | |
1994 | return res; | |
1995 | } | |
1996 | EXPORT_SYMBOL(d_make_root); | |
1997 | ||
f9c34674 MS |
1998 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, |
1999 | struct inode *inode, | |
2000 | bool disconnected) | |
4ea3ada2 | 2001 | { |
9308a612 | 2002 | struct dentry *res; |
b18825a7 | 2003 | unsigned add_flags; |
4ea3ada2 | 2004 | |
f9c34674 | 2005 | security_d_instantiate(dentry, inode); |
873feea0 | 2006 | spin_lock(&inode->i_lock); |
d891eedb | 2007 | res = __d_find_any_alias(inode); |
9308a612 | 2008 | if (res) { |
873feea0 | 2009 | spin_unlock(&inode->i_lock); |
f9c34674 | 2010 | dput(dentry); |
9308a612 CH |
2011 | goto out_iput; |
2012 | } | |
2013 | ||
2014 | /* attach a disconnected dentry */ | |
1a0a397e BF |
2015 | add_flags = d_flags_for_inode(inode); |
2016 | ||
2017 | if (disconnected) | |
2018 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 2019 | |
f9c34674 MS |
2020 | spin_lock(&dentry->d_lock); |
2021 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2022 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
f1ee6162 | 2023 | if (!disconnected) { |
139351f1 LT |
2024 | hlist_bl_lock(&dentry->d_sb->s_roots); |
2025 | hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots); | |
2026 | hlist_bl_unlock(&dentry->d_sb->s_roots); | |
f1ee6162 | 2027 | } |
f9c34674 | 2028 | spin_unlock(&dentry->d_lock); |
873feea0 | 2029 | spin_unlock(&inode->i_lock); |
9308a612 | 2030 | |
f9c34674 | 2031 | return dentry; |
9308a612 CH |
2032 | |
2033 | out_iput: | |
2034 | iput(inode); | |
2035 | return res; | |
4ea3ada2 | 2036 | } |
1a0a397e | 2037 | |
f9c34674 MS |
2038 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) |
2039 | { | |
2040 | return __d_instantiate_anon(dentry, inode, true); | |
2041 | } | |
2042 | EXPORT_SYMBOL(d_instantiate_anon); | |
2043 | ||
2044 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) | |
2045 | { | |
2046 | struct dentry *tmp; | |
2047 | struct dentry *res; | |
2048 | ||
2049 | if (!inode) | |
2050 | return ERR_PTR(-ESTALE); | |
2051 | if (IS_ERR(inode)) | |
2052 | return ERR_CAST(inode); | |
2053 | ||
2054 | res = d_find_any_alias(inode); | |
2055 | if (res) | |
2056 | goto out_iput; | |
2057 | ||
2058 | tmp = d_alloc_anon(inode->i_sb); | |
2059 | if (!tmp) { | |
2060 | res = ERR_PTR(-ENOMEM); | |
2061 | goto out_iput; | |
2062 | } | |
2063 | ||
2064 | return __d_instantiate_anon(tmp, inode, disconnected); | |
2065 | ||
2066 | out_iput: | |
2067 | iput(inode); | |
2068 | return res; | |
2069 | } | |
2070 | ||
1a0a397e BF |
2071 | /** |
2072 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2073 | * @inode: inode to allocate the dentry for | |
2074 | * | |
2075 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2076 | * similar open by handle operations. The returned dentry may be anonymous, | |
2077 | * or may have a full name (if the inode was already in the cache). | |
2078 | * | |
2079 | * When called on a directory inode, we must ensure that the inode only ever | |
2080 | * has one dentry. If a dentry is found, that is returned instead of | |
2081 | * allocating a new one. | |
2082 | * | |
2083 | * On successful return, the reference to the inode has been transferred | |
2084 | * to the dentry. In case of an error the reference on the inode is released. | |
2085 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2086 | * be passed in and the error will be propagated to the return value, | |
2087 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2088 | */ | |
2089 | struct dentry *d_obtain_alias(struct inode *inode) | |
2090 | { | |
f9c34674 | 2091 | return __d_obtain_alias(inode, true); |
1a0a397e | 2092 | } |
adc48720 | 2093 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 2094 | |
1a0a397e BF |
2095 | /** |
2096 | * d_obtain_root - find or allocate a dentry for a given inode | |
2097 | * @inode: inode to allocate the dentry for | |
2098 | * | |
2099 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2100 | * | |
2101 | * We must ensure that directory inodes only ever have one dentry. If a | |
2102 | * dentry is found, that is returned instead of allocating a new one. | |
2103 | * | |
2104 | * On successful return, the reference to the inode has been transferred | |
2105 | * to the dentry. In case of an error the reference on the inode is | |
2106 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2107 | * error will be propagate to the return value, with a %NULL @inode | |
2108 | * replaced by ERR_PTR(-ESTALE). | |
2109 | */ | |
2110 | struct dentry *d_obtain_root(struct inode *inode) | |
2111 | { | |
f9c34674 | 2112 | return __d_obtain_alias(inode, false); |
1a0a397e BF |
2113 | } |
2114 | EXPORT_SYMBOL(d_obtain_root); | |
2115 | ||
9403540c BN |
2116 | /** |
2117 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2118 | * @inode: the inode case-insensitive lookup has found | |
2119 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2120 | * @name: the case-exact name to be associated with the returned dentry | |
2121 | * | |
2122 | * This is to avoid filling the dcache with case-insensitive names to the | |
2123 | * same inode, only the actual correct case is stored in the dcache for | |
2124 | * case-insensitive filesystems. | |
2125 | * | |
2126 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2127 | * already exists in in the dcache, use it and return it. | |
2128 | * | |
2129 | * If no entry exists with the exact case name, allocate new dentry with | |
2130 | * the exact case, and return the spliced entry. | |
2131 | */ | |
e45b590b | 2132 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2133 | struct qstr *name) |
2134 | { | |
d9171b93 | 2135 | struct dentry *found, *res; |
9403540c | 2136 | |
b6520c81 CH |
2137 | /* |
2138 | * First check if a dentry matching the name already exists, | |
2139 | * if not go ahead and create it now. | |
2140 | */ | |
9403540c | 2141 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2142 | if (found) { |
2143 | iput(inode); | |
2144 | return found; | |
2145 | } | |
2146 | if (d_in_lookup(dentry)) { | |
2147 | found = d_alloc_parallel(dentry->d_parent, name, | |
2148 | dentry->d_wait); | |
2149 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2150 | iput(inode); | |
2151 | return found; | |
9403540c | 2152 | } |
d9171b93 AV |
2153 | } else { |
2154 | found = d_alloc(dentry->d_parent, name); | |
2155 | if (!found) { | |
2156 | iput(inode); | |
2157 | return ERR_PTR(-ENOMEM); | |
2158 | } | |
2159 | } | |
2160 | res = d_splice_alias(inode, found); | |
2161 | if (res) { | |
2162 | dput(found); | |
2163 | return res; | |
9403540c | 2164 | } |
4f522a24 | 2165 | return found; |
9403540c | 2166 | } |
ec4f8605 | 2167 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2168 | |
12f8ad4b | 2169 | |
d4c91a8f AV |
2170 | static inline bool d_same_name(const struct dentry *dentry, |
2171 | const struct dentry *parent, | |
2172 | const struct qstr *name) | |
12f8ad4b | 2173 | { |
d4c91a8f AV |
2174 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2175 | if (dentry->d_name.len != name->len) | |
2176 | return false; | |
2177 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2178 | } |
6fa67e70 | 2179 | return parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2180 | dentry->d_name.len, dentry->d_name.name, |
2181 | name) == 0; | |
12f8ad4b LT |
2182 | } |
2183 | ||
31e6b01f NP |
2184 | /** |
2185 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2186 | * @parent: parent dentry | |
2187 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2188 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2189 | * Returns: dentry, or NULL |
2190 | * | |
2191 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2192 | * resolution (store-free path walking) design described in | |
2193 | * Documentation/filesystems/path-lookup.txt. | |
2194 | * | |
2195 | * This is not to be used outside core vfs. | |
2196 | * | |
2197 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2198 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2199 | * without taking d_lock and checking d_seq sequence count against @seq | |
2200 | * returned here. | |
2201 | * | |
15570086 | 2202 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2203 | * function. |
2204 | * | |
2205 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2206 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2207 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2208 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2209 | * |
2210 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2211 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2212 | */ |
8966be90 LT |
2213 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2214 | const struct qstr *name, | |
da53be12 | 2215 | unsigned *seqp) |
31e6b01f | 2216 | { |
26fe5750 | 2217 | u64 hashlen = name->hash_len; |
31e6b01f | 2218 | const unsigned char *str = name->name; |
8387ff25 | 2219 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2220 | struct hlist_bl_node *node; |
31e6b01f NP |
2221 | struct dentry *dentry; |
2222 | ||
2223 | /* | |
2224 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2225 | * required to prevent single threaded performance regressions | |
2226 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2227 | * Keep the two functions in sync. | |
2228 | */ | |
2229 | ||
2230 | /* | |
2231 | * The hash list is protected using RCU. | |
2232 | * | |
2233 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2234 | * races with d_move(). | |
2235 | * | |
2236 | * It is possible that concurrent renames can mess up our list | |
2237 | * walk here and result in missing our dentry, resulting in the | |
2238 | * false-negative result. d_lookup() protects against concurrent | |
2239 | * renames using rename_lock seqlock. | |
2240 | * | |
b0a4bb83 | 2241 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2242 | */ |
b07ad996 | 2243 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2244 | unsigned seq; |
31e6b01f | 2245 | |
31e6b01f | 2246 | seqretry: |
12f8ad4b LT |
2247 | /* |
2248 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2249 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2250 | * |
2251 | * The caller must perform a seqcount check in order | |
da53be12 | 2252 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2253 | * |
2254 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2255 | * we don't wait for the sequence count to stabilize if it | |
2256 | * is in the middle of a sequence change. If we do the slow | |
2257 | * dentry compare, we will do seqretries until it is stable, | |
2258 | * and if we end up with a successful lookup, we actually | |
2259 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2260 | * |
2261 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2262 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2263 | */ |
2264 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2265 | if (dentry->d_parent != parent) |
2266 | continue; | |
2e321806 LT |
2267 | if (d_unhashed(dentry)) |
2268 | continue; | |
12f8ad4b | 2269 | |
830c0f0e | 2270 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
d4c91a8f AV |
2271 | int tlen; |
2272 | const char *tname; | |
26fe5750 LT |
2273 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2274 | continue; | |
d4c91a8f AV |
2275 | tlen = dentry->d_name.len; |
2276 | tname = dentry->d_name.name; | |
2277 | /* we want a consistent (name,len) pair */ | |
2278 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2279 | cpu_relax(); | |
12f8ad4b LT |
2280 | goto seqretry; |
2281 | } | |
6fa67e70 | 2282 | if (parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2283 | tlen, tname, name) != 0) |
2284 | continue; | |
2285 | } else { | |
2286 | if (dentry->d_name.hash_len != hashlen) | |
2287 | continue; | |
2288 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2289 | continue; | |
31e6b01f | 2290 | } |
da53be12 | 2291 | *seqp = seq; |
d4c91a8f | 2292 | return dentry; |
31e6b01f NP |
2293 | } |
2294 | return NULL; | |
2295 | } | |
2296 | ||
1da177e4 LT |
2297 | /** |
2298 | * d_lookup - search for a dentry | |
2299 | * @parent: parent dentry | |
2300 | * @name: qstr of name we wish to find | |
b04f784e | 2301 | * Returns: dentry, or NULL |
1da177e4 | 2302 | * |
b04f784e NP |
2303 | * d_lookup searches the children of the parent dentry for the name in |
2304 | * question. If the dentry is found its reference count is incremented and the | |
2305 | * dentry is returned. The caller must use dput to free the entry when it has | |
2306 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2307 | */ |
da2d8455 | 2308 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2309 | { |
31e6b01f | 2310 | struct dentry *dentry; |
949854d0 | 2311 | unsigned seq; |
1da177e4 | 2312 | |
b8314f93 DY |
2313 | do { |
2314 | seq = read_seqbegin(&rename_lock); | |
2315 | dentry = __d_lookup(parent, name); | |
2316 | if (dentry) | |
1da177e4 LT |
2317 | break; |
2318 | } while (read_seqretry(&rename_lock, seq)); | |
2319 | return dentry; | |
2320 | } | |
ec4f8605 | 2321 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2322 | |
31e6b01f | 2323 | /** |
b04f784e NP |
2324 | * __d_lookup - search for a dentry (racy) |
2325 | * @parent: parent dentry | |
2326 | * @name: qstr of name we wish to find | |
2327 | * Returns: dentry, or NULL | |
2328 | * | |
2329 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2330 | * false-negative result due to unrelated rename activity. | |
2331 | * | |
2332 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2333 | * however it must be used carefully, eg. with a following d_lookup in | |
2334 | * the case of failure. | |
2335 | * | |
2336 | * __d_lookup callers must be commented. | |
2337 | */ | |
a713ca2a | 2338 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2339 | { |
1da177e4 | 2340 | unsigned int hash = name->hash; |
8387ff25 | 2341 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2342 | struct hlist_bl_node *node; |
31e6b01f | 2343 | struct dentry *found = NULL; |
665a7583 | 2344 | struct dentry *dentry; |
1da177e4 | 2345 | |
31e6b01f NP |
2346 | /* |
2347 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2348 | * required to prevent single threaded performance regressions | |
2349 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2350 | * Keep the two functions in sync. | |
2351 | */ | |
2352 | ||
b04f784e NP |
2353 | /* |
2354 | * The hash list is protected using RCU. | |
2355 | * | |
2356 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2357 | * with d_move(). | |
2358 | * | |
2359 | * It is possible that concurrent renames can mess up our list | |
2360 | * walk here and result in missing our dentry, resulting in the | |
2361 | * false-negative result. d_lookup() protects against concurrent | |
2362 | * renames using rename_lock seqlock. | |
2363 | * | |
b0a4bb83 | 2364 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2365 | */ |
1da177e4 LT |
2366 | rcu_read_lock(); |
2367 | ||
b07ad996 | 2368 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2369 | |
1da177e4 LT |
2370 | if (dentry->d_name.hash != hash) |
2371 | continue; | |
1da177e4 LT |
2372 | |
2373 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2374 | if (dentry->d_parent != parent) |
2375 | goto next; | |
d0185c08 LT |
2376 | if (d_unhashed(dentry)) |
2377 | goto next; | |
2378 | ||
d4c91a8f AV |
2379 | if (!d_same_name(dentry, parent, name)) |
2380 | goto next; | |
1da177e4 | 2381 | |
98474236 | 2382 | dentry->d_lockref.count++; |
d0185c08 | 2383 | found = dentry; |
1da177e4 LT |
2384 | spin_unlock(&dentry->d_lock); |
2385 | break; | |
2386 | next: | |
2387 | spin_unlock(&dentry->d_lock); | |
2388 | } | |
2389 | rcu_read_unlock(); | |
2390 | ||
2391 | return found; | |
2392 | } | |
2393 | ||
3e7e241f EB |
2394 | /** |
2395 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2396 | * @dir: Directory to search in | |
2397 | * @name: qstr of name we wish to find | |
2398 | * | |
4f522a24 | 2399 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2400 | */ |
2401 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2402 | { | |
3e7e241f EB |
2403 | /* |
2404 | * Check for a fs-specific hash function. Note that we must | |
2405 | * calculate the standard hash first, as the d_op->d_hash() | |
2406 | * routine may choose to leave the hash value unchanged. | |
2407 | */ | |
8387ff25 | 2408 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2409 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2410 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2411 | if (unlikely(err < 0)) |
2412 | return ERR_PTR(err); | |
3e7e241f | 2413 | } |
4f522a24 | 2414 | return d_lookup(dir, name); |
3e7e241f | 2415 | } |
4f522a24 | 2416 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2417 | |
1da177e4 LT |
2418 | /* |
2419 | * When a file is deleted, we have two options: | |
2420 | * - turn this dentry into a negative dentry | |
2421 | * - unhash this dentry and free it. | |
2422 | * | |
2423 | * Usually, we want to just turn this into | |
2424 | * a negative dentry, but if anybody else is | |
2425 | * currently using the dentry or the inode | |
2426 | * we can't do that and we fall back on removing | |
2427 | * it from the hash queues and waiting for | |
2428 | * it to be deleted later when it has no users | |
2429 | */ | |
2430 | ||
2431 | /** | |
2432 | * d_delete - delete a dentry | |
2433 | * @dentry: The dentry to delete | |
2434 | * | |
2435 | * Turn the dentry into a negative dentry if possible, otherwise | |
2436 | * remove it from the hash queues so it can be deleted later | |
2437 | */ | |
2438 | ||
2439 | void d_delete(struct dentry * dentry) | |
2440 | { | |
c19457f0 | 2441 | struct inode *inode = dentry->d_inode; |
c19457f0 AV |
2442 | |
2443 | spin_lock(&inode->i_lock); | |
2444 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2445 | /* |
2446 | * Are we the only user? | |
2447 | */ | |
98474236 | 2448 | if (dentry->d_lockref.count == 1) { |
13e3c5e5 | 2449 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2450 | dentry_unlink_inode(dentry); |
c19457f0 | 2451 | } else { |
1da177e4 | 2452 | __d_drop(dentry); |
c19457f0 AV |
2453 | spin_unlock(&dentry->d_lock); |
2454 | spin_unlock(&inode->i_lock); | |
2455 | } | |
1da177e4 | 2456 | } |
ec4f8605 | 2457 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2458 | |
15d3c589 | 2459 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2460 | { |
15d3c589 | 2461 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
61647823 | 2462 | |
1879fd6a | 2463 | hlist_bl_lock(b); |
b07ad996 | 2464 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2465 | hlist_bl_unlock(b); |
1da177e4 LT |
2466 | } |
2467 | ||
2468 | /** | |
2469 | * d_rehash - add an entry back to the hash | |
2470 | * @entry: dentry to add to the hash | |
2471 | * | |
2472 | * Adds a dentry to the hash according to its name. | |
2473 | */ | |
2474 | ||
2475 | void d_rehash(struct dentry * entry) | |
2476 | { | |
1da177e4 | 2477 | spin_lock(&entry->d_lock); |
15d3c589 | 2478 | __d_rehash(entry); |
1da177e4 | 2479 | spin_unlock(&entry->d_lock); |
1da177e4 | 2480 | } |
ec4f8605 | 2481 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2482 | |
84e710da AV |
2483 | static inline unsigned start_dir_add(struct inode *dir) |
2484 | { | |
2485 | ||
2486 | for (;;) { | |
2487 | unsigned n = dir->i_dir_seq; | |
2488 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2489 | return n; | |
2490 | cpu_relax(); | |
2491 | } | |
2492 | } | |
2493 | ||
2494 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2495 | { | |
2496 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2497 | } | |
2498 | ||
d9171b93 AV |
2499 | static void d_wait_lookup(struct dentry *dentry) |
2500 | { | |
2501 | if (d_in_lookup(dentry)) { | |
2502 | DECLARE_WAITQUEUE(wait, current); | |
2503 | add_wait_queue(dentry->d_wait, &wait); | |
2504 | do { | |
2505 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2506 | spin_unlock(&dentry->d_lock); | |
2507 | schedule(); | |
2508 | spin_lock(&dentry->d_lock); | |
2509 | } while (d_in_lookup(dentry)); | |
2510 | } | |
2511 | } | |
2512 | ||
94bdd655 | 2513 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2514 | const struct qstr *name, |
2515 | wait_queue_head_t *wq) | |
94bdd655 | 2516 | { |
94bdd655 | 2517 | unsigned int hash = name->hash; |
94bdd655 AV |
2518 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2519 | struct hlist_bl_node *node; | |
2520 | struct dentry *new = d_alloc(parent, name); | |
2521 | struct dentry *dentry; | |
2522 | unsigned seq, r_seq, d_seq; | |
2523 | ||
2524 | if (unlikely(!new)) | |
2525 | return ERR_PTR(-ENOMEM); | |
2526 | ||
2527 | retry: | |
2528 | rcu_read_lock(); | |
015555fd | 2529 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
94bdd655 AV |
2530 | r_seq = read_seqbegin(&rename_lock); |
2531 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2532 | if (unlikely(dentry)) { | |
2533 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2534 | rcu_read_unlock(); | |
2535 | goto retry; | |
2536 | } | |
2537 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2538 | rcu_read_unlock(); | |
2539 | dput(dentry); | |
2540 | goto retry; | |
2541 | } | |
2542 | rcu_read_unlock(); | |
2543 | dput(new); | |
2544 | return dentry; | |
2545 | } | |
2546 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2547 | rcu_read_unlock(); | |
2548 | goto retry; | |
2549 | } | |
015555fd WD |
2550 | |
2551 | if (unlikely(seq & 1)) { | |
2552 | rcu_read_unlock(); | |
2553 | goto retry; | |
2554 | } | |
2555 | ||
94bdd655 | 2556 | hlist_bl_lock(b); |
8cc07c80 | 2557 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
94bdd655 AV |
2558 | hlist_bl_unlock(b); |
2559 | rcu_read_unlock(); | |
2560 | goto retry; | |
2561 | } | |
94bdd655 AV |
2562 | /* |
2563 | * No changes for the parent since the beginning of d_lookup(). | |
2564 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2565 | * any potential in-lookup matches are going to stay here until | |
2566 | * we unlock the chain. All fields are stable in everything | |
2567 | * we encounter. | |
2568 | */ | |
2569 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2570 | if (dentry->d_name.hash != hash) | |
2571 | continue; | |
2572 | if (dentry->d_parent != parent) | |
2573 | continue; | |
d4c91a8f AV |
2574 | if (!d_same_name(dentry, parent, name)) |
2575 | continue; | |
94bdd655 | 2576 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2577 | /* now we can try to grab a reference */ |
2578 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2579 | rcu_read_unlock(); | |
2580 | goto retry; | |
2581 | } | |
2582 | ||
2583 | rcu_read_unlock(); | |
2584 | /* | |
2585 | * somebody is likely to be still doing lookup for it; | |
2586 | * wait for them to finish | |
2587 | */ | |
d9171b93 AV |
2588 | spin_lock(&dentry->d_lock); |
2589 | d_wait_lookup(dentry); | |
2590 | /* | |
2591 | * it's not in-lookup anymore; in principle we should repeat | |
2592 | * everything from dcache lookup, but it's likely to be what | |
2593 | * d_lookup() would've found anyway. If it is, just return it; | |
2594 | * otherwise we really have to repeat the whole thing. | |
2595 | */ | |
2596 | if (unlikely(dentry->d_name.hash != hash)) | |
2597 | goto mismatch; | |
2598 | if (unlikely(dentry->d_parent != parent)) | |
2599 | goto mismatch; | |
2600 | if (unlikely(d_unhashed(dentry))) | |
2601 | goto mismatch; | |
d4c91a8f AV |
2602 | if (unlikely(!d_same_name(dentry, parent, name))) |
2603 | goto mismatch; | |
d9171b93 AV |
2604 | /* OK, it *is* a hashed match; return it */ |
2605 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2606 | dput(new); |
2607 | return dentry; | |
2608 | } | |
e7d6ef97 | 2609 | rcu_read_unlock(); |
94bdd655 AV |
2610 | /* we can't take ->d_lock here; it's OK, though. */ |
2611 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2612 | new->d_wait = wq; |
94bdd655 AV |
2613 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2614 | hlist_bl_unlock(b); | |
2615 | return new; | |
d9171b93 AV |
2616 | mismatch: |
2617 | spin_unlock(&dentry->d_lock); | |
2618 | dput(dentry); | |
2619 | goto retry; | |
94bdd655 AV |
2620 | } |
2621 | EXPORT_SYMBOL(d_alloc_parallel); | |
2622 | ||
85c7f810 AV |
2623 | void __d_lookup_done(struct dentry *dentry) |
2624 | { | |
94bdd655 AV |
2625 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2626 | dentry->d_name.hash); | |
2627 | hlist_bl_lock(b); | |
85c7f810 | 2628 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2629 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2630 | wake_up_all(dentry->d_wait); |
2631 | dentry->d_wait = NULL; | |
94bdd655 AV |
2632 | hlist_bl_unlock(b); |
2633 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2634 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2635 | } |
2636 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2637 | |
2638 | /* inode->i_lock held if inode is non-NULL */ | |
2639 | ||
2640 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2641 | { | |
84e710da AV |
2642 | struct inode *dir = NULL; |
2643 | unsigned n; | |
0568d705 | 2644 | spin_lock(&dentry->d_lock); |
84e710da AV |
2645 | if (unlikely(d_in_lookup(dentry))) { |
2646 | dir = dentry->d_parent->d_inode; | |
2647 | n = start_dir_add(dir); | |
85c7f810 | 2648 | __d_lookup_done(dentry); |
84e710da | 2649 | } |
ed782b5a | 2650 | if (inode) { |
0568d705 AV |
2651 | unsigned add_flags = d_flags_for_inode(inode); |
2652 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2653 | raw_write_seqcount_begin(&dentry->d_seq); | |
2654 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2655 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2656 | fsnotify_update_flags(dentry); |
ed782b5a | 2657 | } |
15d3c589 | 2658 | __d_rehash(dentry); |
84e710da AV |
2659 | if (dir) |
2660 | end_dir_add(dir, n); | |
0568d705 AV |
2661 | spin_unlock(&dentry->d_lock); |
2662 | if (inode) | |
2663 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2664 | } |
2665 | ||
34d0d19d AV |
2666 | /** |
2667 | * d_add - add dentry to hash queues | |
2668 | * @entry: dentry to add | |
2669 | * @inode: The inode to attach to this dentry | |
2670 | * | |
2671 | * This adds the entry to the hash queues and initializes @inode. | |
2672 | * The entry was actually filled in earlier during d_alloc(). | |
2673 | */ | |
2674 | ||
2675 | void d_add(struct dentry *entry, struct inode *inode) | |
2676 | { | |
b9680917 AV |
2677 | if (inode) { |
2678 | security_d_instantiate(entry, inode); | |
ed782b5a | 2679 | spin_lock(&inode->i_lock); |
b9680917 | 2680 | } |
ed782b5a | 2681 | __d_add(entry, inode); |
34d0d19d AV |
2682 | } |
2683 | EXPORT_SYMBOL(d_add); | |
2684 | ||
668d0cd5 AV |
2685 | /** |
2686 | * d_exact_alias - find and hash an exact unhashed alias | |
2687 | * @entry: dentry to add | |
2688 | * @inode: The inode to go with this dentry | |
2689 | * | |
2690 | * If an unhashed dentry with the same name/parent and desired | |
2691 | * inode already exists, hash and return it. Otherwise, return | |
2692 | * NULL. | |
2693 | * | |
2694 | * Parent directory should be locked. | |
2695 | */ | |
2696 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2697 | { | |
2698 | struct dentry *alias; | |
668d0cd5 AV |
2699 | unsigned int hash = entry->d_name.hash; |
2700 | ||
2701 | spin_lock(&inode->i_lock); | |
2702 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2703 | /* | |
2704 | * Don't need alias->d_lock here, because aliases with | |
2705 | * d_parent == entry->d_parent are not subject to name or | |
2706 | * parent changes, because the parent inode i_mutex is held. | |
2707 | */ | |
2708 | if (alias->d_name.hash != hash) | |
2709 | continue; | |
2710 | if (alias->d_parent != entry->d_parent) | |
2711 | continue; | |
d4c91a8f | 2712 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2713 | continue; |
2714 | spin_lock(&alias->d_lock); | |
2715 | if (!d_unhashed(alias)) { | |
2716 | spin_unlock(&alias->d_lock); | |
2717 | alias = NULL; | |
2718 | } else { | |
2719 | __dget_dlock(alias); | |
15d3c589 | 2720 | __d_rehash(alias); |
668d0cd5 AV |
2721 | spin_unlock(&alias->d_lock); |
2722 | } | |
2723 | spin_unlock(&inode->i_lock); | |
2724 | return alias; | |
2725 | } | |
2726 | spin_unlock(&inode->i_lock); | |
2727 | return NULL; | |
2728 | } | |
2729 | EXPORT_SYMBOL(d_exact_alias); | |
2730 | ||
8d85b484 | 2731 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2732 | { |
8d85b484 AV |
2733 | if (unlikely(dname_external(target))) { |
2734 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2735 | /* |
2736 | * Both external: swap the pointers | |
2737 | */ | |
9a8d5bb4 | 2738 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2739 | } else { |
2740 | /* | |
2741 | * dentry:internal, target:external. Steal target's | |
2742 | * storage and make target internal. | |
2743 | */ | |
321bcf92 BF |
2744 | memcpy(target->d_iname, dentry->d_name.name, |
2745 | dentry->d_name.len + 1); | |
1da177e4 LT |
2746 | dentry->d_name.name = target->d_name.name; |
2747 | target->d_name.name = target->d_iname; | |
2748 | } | |
2749 | } else { | |
8d85b484 | 2750 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2751 | /* |
2752 | * dentry:external, target:internal. Give dentry's | |
2753 | * storage to target and make dentry internal | |
2754 | */ | |
2755 | memcpy(dentry->d_iname, target->d_name.name, | |
2756 | target->d_name.len + 1); | |
2757 | target->d_name.name = dentry->d_name.name; | |
2758 | dentry->d_name.name = dentry->d_iname; | |
2759 | } else { | |
2760 | /* | |
da1ce067 | 2761 | * Both are internal. |
1da177e4 | 2762 | */ |
da1ce067 MS |
2763 | unsigned int i; |
2764 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2765 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2766 | swap(((long *) &dentry->d_iname)[i], | |
2767 | ((long *) &target->d_iname)[i]); | |
2768 | } | |
1da177e4 LT |
2769 | } |
2770 | } | |
a28ddb87 | 2771 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2772 | } |
2773 | ||
8d85b484 AV |
2774 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2775 | { | |
2776 | struct external_name *old_name = NULL; | |
2777 | if (unlikely(dname_external(dentry))) | |
2778 | old_name = external_name(dentry); | |
2779 | if (unlikely(dname_external(target))) { | |
2780 | atomic_inc(&external_name(target)->u.count); | |
2781 | dentry->d_name = target->d_name; | |
2782 | } else { | |
2783 | memcpy(dentry->d_iname, target->d_name.name, | |
2784 | target->d_name.len + 1); | |
2785 | dentry->d_name.name = dentry->d_iname; | |
2786 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2787 | } | |
2788 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2e03b4bc | 2789 | kfree_rcu(old_name, u.head); |
8d85b484 AV |
2790 | } |
2791 | ||
9eaef27b | 2792 | /* |
18367501 | 2793 | * __d_move - move a dentry |
1da177e4 LT |
2794 | * @dentry: entry to move |
2795 | * @target: new dentry | |
da1ce067 | 2796 | * @exchange: exchange the two dentries |
1da177e4 LT |
2797 | * |
2798 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2799 | * dcache entries should not be moved in this way. Caller must hold |
2800 | * rename_lock, the i_mutex of the source and target directories, | |
2801 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2802 | */ |
da1ce067 MS |
2803 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2804 | bool exchange) | |
1da177e4 | 2805 | { |
42177007 | 2806 | struct dentry *old_parent, *p; |
84e710da AV |
2807 | struct inode *dir = NULL; |
2808 | unsigned n; | |
1da177e4 | 2809 | |
42177007 AV |
2810 | WARN_ON(!dentry->d_inode); |
2811 | if (WARN_ON(dentry == target)) | |
2812 | return; | |
2813 | ||
2fd6b7f5 | 2814 | BUG_ON(d_ancestor(target, dentry)); |
42177007 AV |
2815 | old_parent = dentry->d_parent; |
2816 | p = d_ancestor(old_parent, target); | |
2817 | if (IS_ROOT(dentry)) { | |
2818 | BUG_ON(p); | |
2819 | spin_lock(&target->d_parent->d_lock); | |
2820 | } else if (!p) { | |
2821 | /* target is not a descendent of dentry->d_parent */ | |
2822 | spin_lock(&target->d_parent->d_lock); | |
2823 | spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED); | |
2824 | } else { | |
2825 | BUG_ON(p == dentry); | |
2826 | spin_lock(&old_parent->d_lock); | |
2827 | if (p != target) | |
2828 | spin_lock_nested(&target->d_parent->d_lock, | |
2829 | DENTRY_D_LOCK_NESTED); | |
2830 | } | |
2831 | spin_lock_nested(&dentry->d_lock, 2); | |
2832 | spin_lock_nested(&target->d_lock, 3); | |
2fd6b7f5 | 2833 | |
84e710da AV |
2834 | if (unlikely(d_in_lookup(target))) { |
2835 | dir = target->d_parent->d_inode; | |
2836 | n = start_dir_add(dir); | |
85c7f810 | 2837 | __d_lookup_done(target); |
84e710da | 2838 | } |
1da177e4 | 2839 | |
31e6b01f | 2840 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2841 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2842 | |
15d3c589 | 2843 | /* unhash both */ |
0632a9ac AV |
2844 | if (!d_unhashed(dentry)) |
2845 | ___d_drop(dentry); | |
2846 | if (!d_unhashed(target)) | |
2847 | ___d_drop(target); | |
1da177e4 | 2848 | |
076515fc AV |
2849 | /* ... and switch them in the tree */ |
2850 | dentry->d_parent = target->d_parent; | |
2851 | if (!exchange) { | |
8d85b484 | 2852 | copy_name(dentry, target); |
61647823 | 2853 | target->d_hash.pprev = NULL; |
076515fc | 2854 | dentry->d_parent->d_lockref.count++; |
5467a68c | 2855 | if (dentry != old_parent) /* wasn't IS_ROOT */ |
076515fc | 2856 | WARN_ON(!--old_parent->d_lockref.count); |
1da177e4 | 2857 | } else { |
076515fc AV |
2858 | target->d_parent = old_parent; |
2859 | swap_names(dentry, target); | |
946e51f2 | 2860 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
076515fc AV |
2861 | __d_rehash(target); |
2862 | fsnotify_update_flags(target); | |
1da177e4 | 2863 | } |
076515fc AV |
2864 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); |
2865 | __d_rehash(dentry); | |
2866 | fsnotify_update_flags(dentry); | |
0bf3d5c1 | 2867 | fscrypt_handle_d_move(dentry); |
1da177e4 | 2868 | |
31e6b01f NP |
2869 | write_seqcount_end(&target->d_seq); |
2870 | write_seqcount_end(&dentry->d_seq); | |
2871 | ||
84e710da AV |
2872 | if (dir) |
2873 | end_dir_add(dir, n); | |
076515fc AV |
2874 | |
2875 | if (dentry->d_parent != old_parent) | |
2876 | spin_unlock(&dentry->d_parent->d_lock); | |
2877 | if (dentry != old_parent) | |
2878 | spin_unlock(&old_parent->d_lock); | |
2879 | spin_unlock(&target->d_lock); | |
2880 | spin_unlock(&dentry->d_lock); | |
18367501 AV |
2881 | } |
2882 | ||
2883 | /* | |
2884 | * d_move - move a dentry | |
2885 | * @dentry: entry to move | |
2886 | * @target: new dentry | |
2887 | * | |
2888 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2889 | * dcache entries should not be moved in this way. See the locking |
2890 | * requirements for __d_move. | |
18367501 AV |
2891 | */ |
2892 | void d_move(struct dentry *dentry, struct dentry *target) | |
2893 | { | |
2894 | write_seqlock(&rename_lock); | |
da1ce067 | 2895 | __d_move(dentry, target, false); |
1da177e4 | 2896 | write_sequnlock(&rename_lock); |
9eaef27b | 2897 | } |
ec4f8605 | 2898 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2899 | |
da1ce067 MS |
2900 | /* |
2901 | * d_exchange - exchange two dentries | |
2902 | * @dentry1: first dentry | |
2903 | * @dentry2: second dentry | |
2904 | */ | |
2905 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2906 | { | |
2907 | write_seqlock(&rename_lock); | |
2908 | ||
2909 | WARN_ON(!dentry1->d_inode); | |
2910 | WARN_ON(!dentry2->d_inode); | |
2911 | WARN_ON(IS_ROOT(dentry1)); | |
2912 | WARN_ON(IS_ROOT(dentry2)); | |
2913 | ||
2914 | __d_move(dentry1, dentry2, true); | |
2915 | ||
2916 | write_sequnlock(&rename_lock); | |
2917 | } | |
8994cce9 | 2918 | EXPORT_SYMBOL_GPL(d_exchange); |
da1ce067 | 2919 | |
e2761a11 OH |
2920 | /** |
2921 | * d_ancestor - search for an ancestor | |
2922 | * @p1: ancestor dentry | |
2923 | * @p2: child dentry | |
2924 | * | |
2925 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2926 | * an ancestor of p2, else NULL. | |
9eaef27b | 2927 | */ |
e2761a11 | 2928 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2929 | { |
2930 | struct dentry *p; | |
2931 | ||
871c0067 | 2932 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2933 | if (p->d_parent == p1) |
e2761a11 | 2934 | return p; |
9eaef27b | 2935 | } |
e2761a11 | 2936 | return NULL; |
9eaef27b TM |
2937 | } |
2938 | ||
2939 | /* | |
2940 | * This helper attempts to cope with remotely renamed directories | |
2941 | * | |
2942 | * It assumes that the caller is already holding | |
a03e283b | 2943 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2944 | * |
2945 | * Note: If ever the locking in lock_rename() changes, then please | |
2946 | * remember to update this too... | |
9eaef27b | 2947 | */ |
b5ae6b15 | 2948 | static int __d_unalias(struct inode *inode, |
873feea0 | 2949 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2950 | { |
9902af79 AV |
2951 | struct mutex *m1 = NULL; |
2952 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2953 | int ret = -ESTALE; |
9eaef27b TM |
2954 | |
2955 | /* If alias and dentry share a parent, then no extra locks required */ | |
2956 | if (alias->d_parent == dentry->d_parent) | |
2957 | goto out_unalias; | |
2958 | ||
9eaef27b | 2959 | /* See lock_rename() */ |
9eaef27b TM |
2960 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2961 | goto out_err; | |
2962 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2963 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2964 | goto out_err; |
9902af79 | 2965 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 2966 | out_unalias: |
8ed936b5 | 2967 | __d_move(alias, dentry, false); |
b5ae6b15 | 2968 | ret = 0; |
9eaef27b | 2969 | out_err: |
9eaef27b | 2970 | if (m2) |
9902af79 | 2971 | up_read(m2); |
9eaef27b TM |
2972 | if (m1) |
2973 | mutex_unlock(m1); | |
2974 | return ret; | |
2975 | } | |
2976 | ||
3f70bd51 BF |
2977 | /** |
2978 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2979 | * @inode: the inode which may have a disconnected dentry | |
2980 | * @dentry: a negative dentry which we want to point to the inode. | |
2981 | * | |
da093a9b BF |
2982 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2983 | * place of the given dentry and return it, else simply d_add the inode | |
2984 | * to the dentry and return NULL. | |
3f70bd51 | 2985 | * |
908790fa BF |
2986 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2987 | * we should error out: directories can't have multiple aliases. | |
2988 | * | |
3f70bd51 BF |
2989 | * This is needed in the lookup routine of any filesystem that is exportable |
2990 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2991 | * | |
2992 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2993 | * is returned. This matches the expected return value of ->lookup. | |
2994 | * | |
2995 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2996 | * In that case, we know that the inode will be a regular file, and also this | |
2997 | * will only occur during atomic_open. So we need to check for the dentry | |
2998 | * being already hashed only in the final case. | |
2999 | */ | |
3000 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3001 | { | |
3f70bd51 BF |
3002 | if (IS_ERR(inode)) |
3003 | return ERR_CAST(inode); | |
3004 | ||
770bfad8 DH |
3005 | BUG_ON(!d_unhashed(dentry)); |
3006 | ||
de689f5e | 3007 | if (!inode) |
b5ae6b15 | 3008 | goto out; |
de689f5e | 3009 | |
b9680917 | 3010 | security_d_instantiate(dentry, inode); |
873feea0 | 3011 | spin_lock(&inode->i_lock); |
9eaef27b | 3012 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
3013 | struct dentry *new = __d_find_any_alias(inode); |
3014 | if (unlikely(new)) { | |
a03e283b EB |
3015 | /* The reference to new ensures it remains an alias */ |
3016 | spin_unlock(&inode->i_lock); | |
18367501 | 3017 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3018 | if (unlikely(d_ancestor(new, dentry))) { |
3019 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3020 | dput(new); |
3021 | new = ERR_PTR(-ELOOP); | |
3022 | pr_warn_ratelimited( | |
3023 | "VFS: Lookup of '%s' in %s %s" | |
3024 | " would have caused loop\n", | |
3025 | dentry->d_name.name, | |
3026 | inode->i_sb->s_type->name, | |
3027 | inode->i_sb->s_id); | |
3028 | } else if (!IS_ROOT(new)) { | |
076515fc | 3029 | struct dentry *old_parent = dget(new->d_parent); |
b5ae6b15 | 3030 | int err = __d_unalias(inode, dentry, new); |
18367501 | 3031 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3032 | if (err) { |
3033 | dput(new); | |
3034 | new = ERR_PTR(err); | |
3035 | } | |
076515fc | 3036 | dput(old_parent); |
18367501 | 3037 | } else { |
b5ae6b15 AV |
3038 | __d_move(new, dentry, false); |
3039 | write_sequnlock(&rename_lock); | |
dd179946 | 3040 | } |
b5ae6b15 AV |
3041 | iput(inode); |
3042 | return new; | |
9eaef27b | 3043 | } |
770bfad8 | 3044 | } |
b5ae6b15 | 3045 | out: |
ed782b5a | 3046 | __d_add(dentry, inode); |
b5ae6b15 | 3047 | return NULL; |
770bfad8 | 3048 | } |
b5ae6b15 | 3049 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3050 | |
1da177e4 LT |
3051 | /* |
3052 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3053 | * | |
3054 | * Trivially implemented using the dcache structure | |
3055 | */ | |
3056 | ||
3057 | /** | |
3058 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3059 | * @new_dentry: new dentry | |
3060 | * @old_dentry: old dentry | |
3061 | * | |
a6e5787f YB |
3062 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3063 | * Returns false otherwise. | |
1da177e4 LT |
3064 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3065 | */ | |
3066 | ||
a6e5787f | 3067 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3068 | { |
a6e5787f | 3069 | bool result; |
949854d0 | 3070 | unsigned seq; |
1da177e4 | 3071 | |
e2761a11 | 3072 | if (new_dentry == old_dentry) |
a6e5787f | 3073 | return true; |
e2761a11 | 3074 | |
e2761a11 | 3075 | do { |
1da177e4 | 3076 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3077 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3078 | /* |
3079 | * Need rcu_readlock to protect against the d_parent trashing | |
3080 | * due to d_move | |
3081 | */ | |
3082 | rcu_read_lock(); | |
e2761a11 | 3083 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3084 | result = true; |
e2761a11 | 3085 | else |
a6e5787f | 3086 | result = false; |
949854d0 | 3087 | rcu_read_unlock(); |
1da177e4 | 3088 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3089 | |
3090 | return result; | |
3091 | } | |
e8f9e5b7 | 3092 | EXPORT_SYMBOL(is_subdir); |
1da177e4 | 3093 | |
db14fc3a | 3094 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3095 | { |
db14fc3a MS |
3096 | struct dentry *root = data; |
3097 | if (dentry != root) { | |
3098 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3099 | return D_WALK_SKIP; | |
1da177e4 | 3100 | |
01ddc4ed MS |
3101 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3102 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3103 | dentry->d_lockref.count--; | |
3104 | } | |
1da177e4 | 3105 | } |
db14fc3a MS |
3106 | return D_WALK_CONTINUE; |
3107 | } | |
58db63d0 | 3108 | |
db14fc3a MS |
3109 | void d_genocide(struct dentry *parent) |
3110 | { | |
3a8e3611 | 3111 | d_walk(parent, parent, d_genocide_kill); |
1da177e4 LT |
3112 | } |
3113 | ||
cbd4a5bc AV |
3114 | EXPORT_SYMBOL(d_genocide); |
3115 | ||
60545d0d | 3116 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3117 | { |
60545d0d AV |
3118 | inode_dec_link_count(inode); |
3119 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3120 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3121 | !d_unlinked(dentry)); |
3122 | spin_lock(&dentry->d_parent->d_lock); | |
3123 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3124 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3125 | (unsigned long long)inode->i_ino); | |
3126 | spin_unlock(&dentry->d_lock); | |
3127 | spin_unlock(&dentry->d_parent->d_lock); | |
3128 | d_instantiate(dentry, inode); | |
1da177e4 | 3129 | } |
60545d0d | 3130 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3131 | |
3132 | static __initdata unsigned long dhash_entries; | |
3133 | static int __init set_dhash_entries(char *str) | |
3134 | { | |
3135 | if (!str) | |
3136 | return 0; | |
3137 | dhash_entries = simple_strtoul(str, &str, 0); | |
3138 | return 1; | |
3139 | } | |
3140 | __setup("dhash_entries=", set_dhash_entries); | |
3141 | ||
3142 | static void __init dcache_init_early(void) | |
3143 | { | |
1da177e4 LT |
3144 | /* If hashes are distributed across NUMA nodes, defer |
3145 | * hash allocation until vmalloc space is available. | |
3146 | */ | |
3147 | if (hashdist) | |
3148 | return; | |
3149 | ||
3150 | dentry_hashtable = | |
3151 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3152 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3153 | dhash_entries, |
3154 | 13, | |
3d375d78 | 3155 | HASH_EARLY | HASH_ZERO, |
1da177e4 | 3156 | &d_hash_shift, |
b35d786b | 3157 | NULL, |
31fe62b9 | 3158 | 0, |
1da177e4 | 3159 | 0); |
854d3e63 | 3160 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3161 | } |
3162 | ||
74bf17cf | 3163 | static void __init dcache_init(void) |
1da177e4 | 3164 | { |
3d375d78 | 3165 | /* |
1da177e4 LT |
3166 | * A constructor could be added for stable state like the lists, |
3167 | * but it is probably not worth it because of the cache nature | |
3d375d78 | 3168 | * of the dcache. |
1da177e4 | 3169 | */ |
80344266 DW |
3170 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, |
3171 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, | |
3172 | d_iname); | |
1da177e4 LT |
3173 | |
3174 | /* Hash may have been set up in dcache_init_early */ | |
3175 | if (!hashdist) | |
3176 | return; | |
3177 | ||
3178 | dentry_hashtable = | |
3179 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3180 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3181 | dhash_entries, |
3182 | 13, | |
3d375d78 | 3183 | HASH_ZERO, |
1da177e4 | 3184 | &d_hash_shift, |
b35d786b | 3185 | NULL, |
31fe62b9 | 3186 | 0, |
1da177e4 | 3187 | 0); |
854d3e63 | 3188 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3189 | } |
3190 | ||
3191 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3192 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3193 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3194 | |
1da177e4 LT |
3195 | void __init vfs_caches_init_early(void) |
3196 | { | |
6916363f SAS |
3197 | int i; |
3198 | ||
3199 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3200 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3201 | ||
1da177e4 LT |
3202 | dcache_init_early(); |
3203 | inode_init_early(); | |
3204 | } | |
3205 | ||
4248b0da | 3206 | void __init vfs_caches_init(void) |
1da177e4 | 3207 | { |
6a9b8820 DW |
3208 | names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0, |
3209 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL); | |
1da177e4 | 3210 | |
74bf17cf DC |
3211 | dcache_init(); |
3212 | inode_init(); | |
4248b0da MG |
3213 | files_init(); |
3214 | files_maxfiles_init(); | |
74bf17cf | 3215 | mnt_init(); |
1da177e4 LT |
3216 | bdev_cache_init(); |
3217 | chrdev_init(); | |
3218 | } |