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() |
da549bdd AV |
54 | * - d_parent and d_chilren |
55 | * - childrens' d_sib 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 | |
68279f9c | 81 | static struct kmem_cache *dentry_cache __ro_after_init; |
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); | |
80e5d1ff AV |
87 | const struct qstr dotdot_name = QSTR_INIT("..", 2); |
88 | EXPORT_SYMBOL(dotdot_name); | |
cdf01226 | 89 | |
1da177e4 LT |
90 | /* |
91 | * This is the single most critical data structure when it comes | |
92 | * to the dcache: the hashtable for lookups. Somebody should try | |
93 | * to make this good - I've just made it work. | |
94 | * | |
95 | * This hash-function tries to avoid losing too many bits of hash | |
96 | * information, yet avoid using a prime hash-size or similar. | |
97 | */ | |
1da177e4 | 98 | |
68279f9c | 99 | static unsigned int d_hash_shift __ro_after_init; |
ceb5bdc2 | 100 | |
68279f9c | 101 | static struct hlist_bl_head *dentry_hashtable __ro_after_init; |
ceb5bdc2 | 102 | |
8387ff25 | 103 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 104 | { |
854d3e63 | 105 | return dentry_hashtable + (hash >> d_hash_shift); |
ceb5bdc2 NP |
106 | } |
107 | ||
94bdd655 AV |
108 | #define IN_LOOKUP_SHIFT 10 |
109 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
110 | ||
111 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
112 | unsigned int hash) | |
113 | { | |
114 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
115 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
116 | } | |
117 | ||
c8c0c239 LC |
118 | struct dentry_stat_t { |
119 | long nr_dentry; | |
120 | long nr_unused; | |
121 | long age_limit; /* age in seconds */ | |
122 | long want_pages; /* pages requested by system */ | |
123 | long nr_negative; /* # of unused negative dentries */ | |
124 | long dummy; /* Reserved for future use */ | |
1da177e4 LT |
125 | }; |
126 | ||
3942c07c | 127 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 128 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
af0c9af1 | 129 | static DEFINE_PER_CPU(long, nr_dentry_negative); |
312d3ca8 CH |
130 | |
131 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
c8c0c239 LC |
132 | /* Statistics gathering. */ |
133 | static struct dentry_stat_t dentry_stat = { | |
134 | .age_limit = 45, | |
135 | }; | |
62d36c77 DC |
136 | |
137 | /* | |
138 | * Here we resort to our own counters instead of using generic per-cpu counters | |
139 | * for consistency with what the vfs inode code does. We are expected to harvest | |
140 | * better code and performance by having our own specialized counters. | |
141 | * | |
142 | * Please note that the loop is done over all possible CPUs, not over all online | |
143 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
144 | * on and off. If one of them goes off, we will just keep their counters. | |
145 | * | |
146 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
147 | * please update all vfs counters to match. | |
148 | */ | |
3942c07c | 149 | static long get_nr_dentry(void) |
3e880fb5 NP |
150 | { |
151 | int i; | |
3942c07c | 152 | long sum = 0; |
3e880fb5 NP |
153 | for_each_possible_cpu(i) |
154 | sum += per_cpu(nr_dentry, i); | |
155 | return sum < 0 ? 0 : sum; | |
156 | } | |
157 | ||
62d36c77 DC |
158 | static long get_nr_dentry_unused(void) |
159 | { | |
160 | int i; | |
161 | long sum = 0; | |
162 | for_each_possible_cpu(i) | |
163 | sum += per_cpu(nr_dentry_unused, i); | |
164 | return sum < 0 ? 0 : sum; | |
165 | } | |
166 | ||
af0c9af1 WL |
167 | static long get_nr_dentry_negative(void) |
168 | { | |
169 | int i; | |
170 | long sum = 0; | |
171 | ||
172 | for_each_possible_cpu(i) | |
173 | sum += per_cpu(nr_dentry_negative, i); | |
174 | return sum < 0 ? 0 : sum; | |
175 | } | |
176 | ||
c8c0c239 LC |
177 | static int proc_nr_dentry(struct ctl_table *table, int write, void *buffer, |
178 | size_t *lenp, loff_t *ppos) | |
312d3ca8 | 179 | { |
3e880fb5 | 180 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 181 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
af0c9af1 | 182 | dentry_stat.nr_negative = get_nr_dentry_negative(); |
3942c07c | 183 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 | 184 | } |
c8c0c239 LC |
185 | |
186 | static struct ctl_table fs_dcache_sysctls[] = { | |
187 | { | |
188 | .procname = "dentry-state", | |
189 | .data = &dentry_stat, | |
190 | .maxlen = 6*sizeof(long), | |
191 | .mode = 0444, | |
192 | .proc_handler = proc_nr_dentry, | |
193 | }, | |
194 | { } | |
195 | }; | |
196 | ||
197 | static int __init init_fs_dcache_sysctls(void) | |
198 | { | |
199 | register_sysctl_init("fs", fs_dcache_sysctls); | |
200 | return 0; | |
201 | } | |
202 | fs_initcall(init_fs_dcache_sysctls); | |
312d3ca8 CH |
203 | #endif |
204 | ||
5483f18e LT |
205 | /* |
206 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
207 | * The strings are both count bytes long, and count is non-zero. | |
208 | */ | |
e419b4cc LT |
209 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
210 | ||
211 | #include <asm/word-at-a-time.h> | |
212 | /* | |
213 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
214 | * aligned allocation for this particular component. We don't | |
215 | * strictly need the load_unaligned_zeropad() safety, but it | |
216 | * doesn't hurt either. | |
217 | * | |
218 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
219 | * need the careful unaligned handling. | |
220 | */ | |
94753db5 | 221 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 222 | { |
bfcfaa77 | 223 | unsigned long a,b,mask; |
bfcfaa77 LT |
224 | |
225 | for (;;) { | |
bfe7aa6c | 226 | a = read_word_at_a_time(cs); |
e419b4cc | 227 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
228 | if (tcount < sizeof(unsigned long)) |
229 | break; | |
230 | if (unlikely(a != b)) | |
231 | return 1; | |
232 | cs += sizeof(unsigned long); | |
233 | ct += sizeof(unsigned long); | |
234 | tcount -= sizeof(unsigned long); | |
235 | if (!tcount) | |
236 | return 0; | |
237 | } | |
a5c21dce | 238 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 239 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
240 | } |
241 | ||
bfcfaa77 | 242 | #else |
e419b4cc | 243 | |
94753db5 | 244 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 245 | { |
5483f18e LT |
246 | do { |
247 | if (*cs != *ct) | |
248 | return 1; | |
249 | cs++; | |
250 | ct++; | |
251 | tcount--; | |
252 | } while (tcount); | |
253 | return 0; | |
254 | } | |
255 | ||
e419b4cc LT |
256 | #endif |
257 | ||
94753db5 LT |
258 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
259 | { | |
94753db5 LT |
260 | /* |
261 | * Be careful about RCU walk racing with rename: | |
506458ef | 262 | * use 'READ_ONCE' to fetch the name pointer. |
94753db5 LT |
263 | * |
264 | * NOTE! Even if a rename will mean that the length | |
265 | * was not loaded atomically, we don't care. The | |
266 | * RCU walk will check the sequence count eventually, | |
267 | * and catch it. And we won't overrun the buffer, | |
268 | * because we're reading the name pointer atomically, | |
269 | * and a dentry name is guaranteed to be properly | |
270 | * terminated with a NUL byte. | |
271 | * | |
272 | * End result: even if 'len' is wrong, we'll exit | |
273 | * early because the data cannot match (there can | |
274 | * be no NUL in the ct/tcount data) | |
275 | */ | |
506458ef | 276 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
ae0a843c | 277 | |
6326c71f | 278 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
279 | } |
280 | ||
8d85b484 AV |
281 | struct external_name { |
282 | union { | |
283 | atomic_t count; | |
284 | struct rcu_head head; | |
285 | } u; | |
286 | unsigned char name[]; | |
287 | }; | |
288 | ||
289 | static inline struct external_name *external_name(struct dentry *dentry) | |
290 | { | |
291 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
292 | } | |
293 | ||
9c82ab9c | 294 | static void __d_free(struct rcu_head *head) |
1da177e4 | 295 | { |
9c82ab9c CH |
296 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
297 | ||
8d85b484 AV |
298 | kmem_cache_free(dentry_cache, dentry); |
299 | } | |
300 | ||
301 | static void __d_free_external(struct rcu_head *head) | |
302 | { | |
303 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
2e03b4bc | 304 | kfree(external_name(dentry)); |
f1782c9b | 305 | kmem_cache_free(dentry_cache, dentry); |
1da177e4 LT |
306 | } |
307 | ||
810bb172 AV |
308 | static inline int dname_external(const struct dentry *dentry) |
309 | { | |
310 | return dentry->d_name.name != dentry->d_iname; | |
311 | } | |
312 | ||
49d31c2f AV |
313 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
314 | { | |
315 | spin_lock(&dentry->d_lock); | |
230c6402 | 316 | name->name = dentry->d_name; |
49d31c2f | 317 | if (unlikely(dname_external(dentry))) { |
230c6402 | 318 | atomic_inc(&external_name(dentry)->u.count); |
49d31c2f | 319 | } else { |
6cd00a01 TH |
320 | memcpy(name->inline_name, dentry->d_iname, |
321 | dentry->d_name.len + 1); | |
230c6402 | 322 | name->name.name = name->inline_name; |
49d31c2f | 323 | } |
230c6402 | 324 | spin_unlock(&dentry->d_lock); |
49d31c2f AV |
325 | } |
326 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
327 | ||
328 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
329 | { | |
230c6402 | 330 | if (unlikely(name->name.name != name->inline_name)) { |
49d31c2f | 331 | struct external_name *p; |
230c6402 | 332 | p = container_of(name->name.name, struct external_name, name[0]); |
49d31c2f | 333 | if (unlikely(atomic_dec_and_test(&p->u.count))) |
2e03b4bc | 334 | kfree_rcu(p, u.head); |
49d31c2f AV |
335 | } |
336 | } | |
337 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
338 | ||
4bf46a27 DH |
339 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
340 | struct inode *inode, | |
341 | unsigned type_flags) | |
342 | { | |
343 | unsigned flags; | |
344 | ||
345 | dentry->d_inode = inode; | |
4bf46a27 DH |
346 | flags = READ_ONCE(dentry->d_flags); |
347 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
348 | flags |= type_flags; | |
2fa6b1e0 | 349 | smp_store_release(&dentry->d_flags, flags); |
4bf46a27 DH |
350 | } |
351 | ||
4bf46a27 DH |
352 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
353 | { | |
354 | unsigned flags = READ_ONCE(dentry->d_flags); | |
355 | ||
356 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
357 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 | 358 | dentry->d_inode = NULL; |
af0c9af1 WL |
359 | if (dentry->d_flags & DCACHE_LRU_LIST) |
360 | this_cpu_inc(nr_dentry_negative); | |
4bf46a27 DH |
361 | } |
362 | ||
b4f0354e AV |
363 | static void dentry_free(struct dentry *dentry) |
364 | { | |
946e51f2 | 365 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
366 | if (unlikely(dname_external(dentry))) { |
367 | struct external_name *p = external_name(dentry); | |
368 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
369 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
370 | return; | |
371 | } | |
372 | } | |
b4f0354e | 373 | /* if dentry was never visible to RCU, immediate free is OK */ |
5467a68c | 374 | if (dentry->d_flags & DCACHE_NORCU) |
b4f0354e AV |
375 | __d_free(&dentry->d_u.d_rcu); |
376 | else | |
377 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
378 | } | |
379 | ||
1da177e4 LT |
380 | /* |
381 | * Release the dentry's inode, using the filesystem | |
550dce01 | 382 | * d_iput() operation if defined. |
31e6b01f NP |
383 | */ |
384 | static void dentry_unlink_inode(struct dentry * dentry) | |
385 | __releases(dentry->d_lock) | |
873feea0 | 386 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
387 | { |
388 | struct inode *inode = dentry->d_inode; | |
a528aca7 | 389 | |
4c0d7cd5 | 390 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 391 | __d_clear_type_and_inode(dentry); |
946e51f2 | 392 | hlist_del_init(&dentry->d_u.d_alias); |
4c0d7cd5 | 393 | raw_write_seqcount_end(&dentry->d_seq); |
31e6b01f | 394 | spin_unlock(&dentry->d_lock); |
873feea0 | 395 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
396 | if (!inode->i_nlink) |
397 | fsnotify_inoderemove(inode); | |
398 | if (dentry->d_op && dentry->d_op->d_iput) | |
399 | dentry->d_op->d_iput(dentry, inode); | |
400 | else | |
401 | iput(inode); | |
402 | } | |
403 | ||
89dc77bc LT |
404 | /* |
405 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
406 | * is in use - which includes both the "real" per-superblock | |
407 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
408 | * | |
409 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
410 | * on the shrink list (ie not on the superblock LRU list). | |
411 | * | |
412 | * The per-cpu "nr_dentry_unused" counters are updated with | |
413 | * the DCACHE_LRU_LIST bit. | |
414 | * | |
af0c9af1 WL |
415 | * The per-cpu "nr_dentry_negative" counters are only updated |
416 | * when deleted from or added to the per-superblock LRU list, not | |
417 | * from/to the shrink list. That is to avoid an unneeded dec/inc | |
418 | * pair when moving from LRU to shrink list in select_collect(). | |
419 | * | |
89dc77bc LT |
420 | * These helper functions make sure we always follow the |
421 | * rules. d_lock must be held by the caller. | |
422 | */ | |
423 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
424 | static void d_lru_add(struct dentry *dentry) | |
425 | { | |
426 | D_FLAG_VERIFY(dentry, 0); | |
427 | dentry->d_flags |= DCACHE_LRU_LIST; | |
428 | this_cpu_inc(nr_dentry_unused); | |
af0c9af1 WL |
429 | if (d_is_negative(dentry)) |
430 | this_cpu_inc(nr_dentry_negative); | |
89dc77bc LT |
431 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
432 | } | |
433 | ||
434 | static void d_lru_del(struct dentry *dentry) | |
435 | { | |
436 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
437 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
438 | this_cpu_dec(nr_dentry_unused); | |
af0c9af1 WL |
439 | if (d_is_negative(dentry)) |
440 | this_cpu_dec(nr_dentry_negative); | |
89dc77bc LT |
441 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
442 | } | |
443 | ||
444 | static void d_shrink_del(struct dentry *dentry) | |
445 | { | |
446 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
447 | list_del_init(&dentry->d_lru); | |
448 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
449 | this_cpu_dec(nr_dentry_unused); | |
450 | } | |
451 | ||
452 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
453 | { | |
454 | D_FLAG_VERIFY(dentry, 0); | |
455 | list_add(&dentry->d_lru, list); | |
456 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
457 | this_cpu_inc(nr_dentry_unused); | |
458 | } | |
459 | ||
460 | /* | |
461 | * These can only be called under the global LRU lock, ie during the | |
462 | * callback for freeing the LRU list. "isolate" removes it from the | |
463 | * LRU lists entirely, while shrink_move moves it to the indicated | |
464 | * private list. | |
465 | */ | |
3f97b163 | 466 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
467 | { |
468 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
469 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
470 | this_cpu_dec(nr_dentry_unused); | |
af0c9af1 WL |
471 | if (d_is_negative(dentry)) |
472 | this_cpu_dec(nr_dentry_negative); | |
3f97b163 | 473 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
474 | } |
475 | ||
3f97b163 VD |
476 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
477 | struct list_head *list) | |
89dc77bc LT |
478 | { |
479 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
480 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
af0c9af1 WL |
481 | if (d_is_negative(dentry)) |
482 | this_cpu_dec(nr_dentry_negative); | |
3f97b163 | 483 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
484 | } |
485 | ||
61647823 | 486 | static void ___d_drop(struct dentry *dentry) |
789680d1 | 487 | { |
0632a9ac AV |
488 | struct hlist_bl_head *b; |
489 | /* | |
490 | * Hashed dentries are normally on the dentry hashtable, | |
491 | * with the exception of those newly allocated by | |
492 | * d_obtain_root, which are always IS_ROOT: | |
493 | */ | |
494 | if (unlikely(IS_ROOT(dentry))) | |
495 | b = &dentry->d_sb->s_roots; | |
496 | else | |
497 | b = d_hash(dentry->d_name.hash); | |
b61625d2 | 498 | |
0632a9ac AV |
499 | hlist_bl_lock(b); |
500 | __hlist_bl_del(&dentry->d_hash); | |
501 | hlist_bl_unlock(b); | |
789680d1 | 502 | } |
61647823 N |
503 | |
504 | void __d_drop(struct dentry *dentry) | |
505 | { | |
0632a9ac AV |
506 | if (!d_unhashed(dentry)) { |
507 | ___d_drop(dentry); | |
508 | dentry->d_hash.pprev = NULL; | |
509 | write_seqcount_invalidate(&dentry->d_seq); | |
510 | } | |
61647823 | 511 | } |
789680d1 NP |
512 | EXPORT_SYMBOL(__d_drop); |
513 | ||
961f3c89 MCC |
514 | /** |
515 | * d_drop - drop a dentry | |
516 | * @dentry: dentry to drop | |
517 | * | |
518 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
519 | * be found through a VFS lookup any more. Note that this is different from | |
520 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
521 | * possible, giving a successful _negative_ lookup, while d_drop will | |
522 | * just make the cache lookup fail. | |
523 | * | |
524 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
525 | * reason (NFS timeouts or autofs deletes). | |
526 | * | |
527 | * __d_drop requires dentry->d_lock | |
528 | * | |
529 | * ___d_drop doesn't mark dentry as "unhashed" | |
530 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
531 | */ | |
789680d1 NP |
532 | void d_drop(struct dentry *dentry) |
533 | { | |
789680d1 NP |
534 | spin_lock(&dentry->d_lock); |
535 | __d_drop(dentry); | |
536 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
537 | } |
538 | EXPORT_SYMBOL(d_drop); | |
539 | ||
da549bdd | 540 | static inline void dentry_unlist(struct dentry *dentry) |
ba65dc5e AV |
541 | { |
542 | struct dentry *next; | |
543 | /* | |
544 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
545 | * attached to the dentry tree | |
546 | */ | |
547 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
da549bdd | 548 | if (unlikely(hlist_unhashed(&dentry->d_sib))) |
ba65dc5e | 549 | return; |
da549bdd | 550 | __hlist_del(&dentry->d_sib); |
ba65dc5e AV |
551 | /* |
552 | * Cursors can move around the list of children. While we'd been | |
da549bdd | 553 | * a normal list member, it didn't matter - ->d_sib.next would've |
ba65dc5e AV |
554 | * been updated. However, from now on it won't be and for the |
555 | * things like d_walk() it might end up with a nasty surprise. | |
556 | * Normally d_walk() doesn't care about cursors moving around - | |
557 | * ->d_lock on parent prevents that and since a cursor has no children | |
558 | * of its own, we get through it without ever unlocking the parent. | |
559 | * There is one exception, though - if we ascend from a child that | |
560 | * gets killed as soon as we unlock it, the next sibling is found | |
da549bdd | 561 | * using the value left in its ->d_sib.next. And if _that_ |
ba65dc5e AV |
562 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) |
563 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
564 | * everything the cursor had been moved past. | |
565 | * | |
da549bdd | 566 | * Solution: make sure that the pointer left behind in ->d_sib.next |
ba65dc5e AV |
567 | * points to something that won't be moving around. I.e. skip the |
568 | * cursors. | |
569 | */ | |
da549bdd AV |
570 | while (dentry->d_sib.next) { |
571 | next = hlist_entry(dentry->d_sib.next, struct dentry, d_sib); | |
ba65dc5e AV |
572 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) |
573 | break; | |
da549bdd | 574 | dentry->d_sib.next = next->d_sib.next; |
ba65dc5e AV |
575 | } |
576 | } | |
577 | ||
e55fd011 | 578 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 579 | { |
41edf278 AV |
580 | struct dentry *parent = NULL; |
581 | bool can_free = true; | |
41edf278 | 582 | if (!IS_ROOT(dentry)) |
77812a1e | 583 | parent = dentry->d_parent; |
31e6b01f | 584 | |
0d98439e LT |
585 | /* |
586 | * The dentry is now unrecoverably dead to the world. | |
587 | */ | |
588 | lockref_mark_dead(&dentry->d_lockref); | |
589 | ||
f0023bc6 | 590 | /* |
f0023bc6 SW |
591 | * inform the fs via d_prune that this dentry is about to be |
592 | * unhashed and destroyed. | |
593 | */ | |
29266201 | 594 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
595 | dentry->d_op->d_prune(dentry); |
596 | ||
01b60351 AV |
597 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
598 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
599 | d_lru_del(dentry); | |
01b60351 | 600 | } |
77812a1e NP |
601 | /* if it was on the hash then remove it */ |
602 | __d_drop(dentry); | |
da549bdd | 603 | dentry_unlist(dentry); |
03b3b889 AV |
604 | if (parent) |
605 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
606 | if (dentry->d_inode) |
607 | dentry_unlink_inode(dentry); | |
608 | else | |
609 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
610 | this_cpu_dec(nr_dentry); |
611 | if (dentry->d_op && dentry->d_op->d_release) | |
612 | dentry->d_op->d_release(dentry); | |
613 | ||
41edf278 AV |
614 | spin_lock(&dentry->d_lock); |
615 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
616 | dentry->d_flags |= DCACHE_MAY_FREE; | |
617 | can_free = false; | |
618 | } | |
619 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
620 | if (likely(can_free)) |
621 | dentry_free(dentry); | |
9c5f1d30 | 622 | cond_resched(); |
e55fd011 AV |
623 | } |
624 | ||
8b987a46 | 625 | static struct dentry *__lock_parent(struct dentry *dentry) |
046b961b | 626 | { |
8b987a46 | 627 | struct dentry *parent; |
046b961b | 628 | again: |
66702eb5 | 629 | parent = READ_ONCE(dentry->d_parent); |
046b961b AV |
630 | spin_lock(&parent->d_lock); |
631 | /* | |
632 | * We can't blindly lock dentry until we are sure | |
633 | * that we won't violate the locking order. | |
634 | * Any changes of dentry->d_parent must have | |
635 | * been done with parent->d_lock held, so | |
636 | * spin_lock() above is enough of a barrier | |
637 | * for checking if it's still our child. | |
638 | */ | |
639 | if (unlikely(parent != dentry->d_parent)) { | |
640 | spin_unlock(&parent->d_lock); | |
641 | goto again; | |
642 | } | |
65d8eb5a | 643 | if (parent != dentry) |
9f12600f | 644 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
65d8eb5a | 645 | else |
046b961b AV |
646 | parent = NULL; |
647 | return parent; | |
648 | } | |
649 | ||
8b987a46 AV |
650 | static inline struct dentry *lock_parent(struct dentry *dentry) |
651 | { | |
652 | struct dentry *parent = dentry->d_parent; | |
653 | if (IS_ROOT(dentry)) | |
654 | return NULL; | |
655 | if (likely(spin_trylock(&parent->d_lock))) | |
656 | return parent; | |
339e9e13 AV |
657 | rcu_read_lock(); |
658 | spin_unlock(&dentry->d_lock); | |
659 | parent = __lock_parent(dentry); | |
660 | rcu_read_unlock(); | |
661 | return parent; | |
662 | } | |
663 | ||
664 | /* | |
665 | * Lock a dentry for feeding it to __dentry_kill(). | |
666 | * Called under rcu_read_lock() and dentry->d_lock; the former | |
667 | * guarantees that nothing we access will be freed under us. | |
668 | * Note that dentry is *not* protected from concurrent dentry_kill(), | |
669 | * d_delete(), etc. | |
670 | * | |
671 | * Return false if dentry is busy. Otherwise, return true and have | |
672 | * that dentry's inode and parent both locked. | |
673 | */ | |
674 | ||
675 | static bool lock_for_kill(struct dentry *dentry) | |
676 | { | |
677 | struct inode *inode = dentry->d_inode; | |
678 | struct dentry *parent = dentry->d_parent; | |
679 | ||
680 | if (unlikely(dentry->d_lockref.count)) | |
681 | return false; | |
682 | ||
683 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
684 | goto slow; | |
685 | if (dentry == parent) | |
686 | return true; | |
687 | if (likely(spin_trylock(&parent->d_lock))) | |
688 | return true; | |
689 | ||
690 | if (inode) | |
691 | spin_unlock(&inode->i_lock); | |
692 | slow: | |
693 | spin_unlock(&dentry->d_lock); | |
694 | ||
695 | for (;;) { | |
696 | if (inode) | |
697 | spin_lock(&inode->i_lock); | |
698 | parent = __lock_parent(dentry); | |
699 | if (likely(inode == dentry->d_inode)) | |
700 | break; | |
701 | if (inode) | |
702 | spin_unlock(&inode->i_lock); | |
703 | inode = dentry->d_inode; | |
704 | spin_unlock(&dentry->d_lock); | |
705 | if (parent) | |
706 | spin_unlock(&parent->d_lock); | |
707 | } | |
708 | if (likely(!dentry->d_lockref.count)) | |
709 | return true; | |
710 | if (inode) | |
711 | spin_unlock(&inode->i_lock); | |
712 | if (parent) | |
713 | spin_unlock(&parent->d_lock); | |
714 | return false; | |
8b987a46 AV |
715 | } |
716 | ||
a338579f AV |
717 | static inline bool retain_dentry(struct dentry *dentry) |
718 | { | |
719 | WARN_ON(d_in_lookup(dentry)); | |
720 | ||
721 | /* Unreachable? Get rid of it */ | |
722 | if (unlikely(d_unhashed(dentry))) | |
723 | return false; | |
724 | ||
725 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
726 | return false; | |
727 | ||
728 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
729 | if (dentry->d_op->d_delete(dentry)) | |
730 | return false; | |
731 | } | |
2c567af4 IW |
732 | |
733 | if (unlikely(dentry->d_flags & DCACHE_DONTCACHE)) | |
734 | return false; | |
735 | ||
62d9956c | 736 | /* retain; LRU fodder */ |
62d9956c AV |
737 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
738 | d_lru_add(dentry); | |
739 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) | |
740 | dentry->d_flags |= DCACHE_REFERENCED; | |
a338579f AV |
741 | return true; |
742 | } | |
743 | ||
2c567af4 IW |
744 | void d_mark_dontcache(struct inode *inode) |
745 | { | |
746 | struct dentry *de; | |
747 | ||
748 | spin_lock(&inode->i_lock); | |
749 | hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) { | |
750 | spin_lock(&de->d_lock); | |
751 | de->d_flags |= DCACHE_DONTCACHE; | |
752 | spin_unlock(&de->d_lock); | |
753 | } | |
754 | inode->i_state |= I_DONTCACHE; | |
755 | spin_unlock(&inode->i_lock); | |
756 | } | |
757 | EXPORT_SYMBOL(d_mark_dontcache); | |
758 | ||
c1d0c1a2 JO |
759 | /* |
760 | * Finish off a dentry we've decided to kill. | |
761 | * dentry->d_lock must be held, returns with it unlocked. | |
762 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
763 | */ | |
764 | static struct dentry *dentry_kill(struct dentry *dentry) | |
765 | __releases(dentry->d_lock) | |
766 | { | |
c1d0c1a2 | 767 | |
b06c684d | 768 | dentry->d_lockref.count--; |
339e9e13 AV |
769 | rcu_read_lock(); |
770 | if (likely(lock_for_kill(dentry))) { | |
771 | struct dentry *parent = dentry->d_parent; | |
772 | rcu_read_unlock(); | |
f657a666 | 773 | __dentry_kill(dentry); |
339e9e13 | 774 | return parent != dentry ? parent : NULL; |
f657a666 | 775 | } |
339e9e13 | 776 | rcu_read_unlock(); |
c1d0c1a2 | 777 | spin_unlock(&dentry->d_lock); |
f657a666 | 778 | return NULL; |
c1d0c1a2 JO |
779 | } |
780 | ||
360f5479 LT |
781 | /* |
782 | * Try to do a lockless dput(), and return whether that was successful. | |
783 | * | |
784 | * If unsuccessful, we return false, having already taken the dentry lock. | |
f05441c7 AV |
785 | * In that case refcount is guaranteed to be zero and we have already |
786 | * decided that it's not worth keeping around. | |
360f5479 LT |
787 | * |
788 | * The caller needs to hold the RCU read lock, so that the dentry is | |
789 | * guaranteed to stay around even if the refcount goes down to zero! | |
790 | */ | |
791 | static inline bool fast_dput(struct dentry *dentry) | |
792 | { | |
793 | int ret; | |
794 | unsigned int d_flags; | |
795 | ||
796 | /* | |
15220fbf | 797 | * try to decrement the lockref optimistically. |
360f5479 LT |
798 | */ |
799 | ret = lockref_put_return(&dentry->d_lockref); | |
800 | ||
801 | /* | |
802 | * If the lockref_put_return() failed due to the lock being held | |
803 | * by somebody else, the fast path has failed. We will need to | |
804 | * get the lock, and then check the count again. | |
805 | */ | |
806 | if (unlikely(ret < 0)) { | |
807 | spin_lock(&dentry->d_lock); | |
504e08ce | 808 | if (WARN_ON_ONCE(dentry->d_lockref.count <= 0)) { |
360f5479 | 809 | spin_unlock(&dentry->d_lock); |
7964410f | 810 | return true; |
360f5479 | 811 | } |
504e08ce AV |
812 | dentry->d_lockref.count--; |
813 | goto locked; | |
360f5479 LT |
814 | } |
815 | ||
816 | /* | |
817 | * If we weren't the last ref, we're done. | |
818 | */ | |
819 | if (ret) | |
7964410f | 820 | return true; |
360f5479 LT |
821 | |
822 | /* | |
823 | * Careful, careful. The reference count went down | |
824 | * to zero, but we don't hold the dentry lock, so | |
825 | * somebody else could get it again, and do another | |
826 | * dput(), and we need to not race with that. | |
827 | * | |
828 | * However, there is a very special and common case | |
829 | * where we don't care, because there is nothing to | |
830 | * do: the dentry is still hashed, it does not have | |
831 | * a 'delete' op, and it's referenced and already on | |
832 | * the LRU list. | |
833 | * | |
834 | * NOTE! Since we aren't locked, these values are | |
835 | * not "stable". However, it is sufficient that at | |
836 | * some point after we dropped the reference the | |
837 | * dentry was hashed and the flags had the proper | |
838 | * value. Other dentry users may have re-gotten | |
839 | * a reference to the dentry and change that, but | |
840 | * our work is done - we can leave the dentry | |
841 | * around with a zero refcount. | |
77573fa3 HL |
842 | * |
843 | * Nevertheless, there are two cases that we should kill | |
844 | * the dentry anyway. | |
845 | * 1. free disconnected dentries as soon as their refcount | |
846 | * reached zero. | |
847 | * 2. free dentries if they should not be cached. | |
360f5479 LT |
848 | */ |
849 | smp_rmb(); | |
66702eb5 | 850 | d_flags = READ_ONCE(dentry->d_flags); |
15220fbf | 851 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_OP_DELETE | |
77573fa3 | 852 | DCACHE_DISCONNECTED | DCACHE_DONTCACHE; |
360f5479 LT |
853 | |
854 | /* Nothing to do? Dropping the reference was all we needed? */ | |
855 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
7964410f | 856 | return true; |
360f5479 LT |
857 | |
858 | /* | |
859 | * Not the fast normal case? Get the lock. We've already decremented | |
860 | * the refcount, but we'll need to re-check the situation after | |
861 | * getting the lock. | |
862 | */ | |
863 | spin_lock(&dentry->d_lock); | |
864 | ||
865 | /* | |
866 | * Did somebody else grab a reference to it in the meantime, and | |
867 | * we're no longer the last user after all? Alternatively, somebody | |
868 | * else could have killed it and marked it dead. Either way, we | |
869 | * don't need to do anything else. | |
870 | */ | |
504e08ce | 871 | locked: |
f05441c7 | 872 | if (dentry->d_lockref.count || retain_dentry(dentry)) { |
360f5479 | 873 | spin_unlock(&dentry->d_lock); |
7964410f | 874 | return true; |
360f5479 | 875 | } |
7964410f | 876 | return false; |
360f5479 LT |
877 | } |
878 | ||
879 | ||
1da177e4 LT |
880 | /* |
881 | * This is dput | |
882 | * | |
883 | * This is complicated by the fact that we do not want to put | |
884 | * dentries that are no longer on any hash chain on the unused | |
885 | * list: we'd much rather just get rid of them immediately. | |
886 | * | |
887 | * However, that implies that we have to traverse the dentry | |
888 | * tree upwards to the parents which might _also_ now be | |
889 | * scheduled for deletion (it may have been only waiting for | |
890 | * its last child to go away). | |
891 | * | |
892 | * This tail recursion is done by hand as we don't want to depend | |
893 | * on the compiler to always get this right (gcc generally doesn't). | |
894 | * Real recursion would eat up our stack space. | |
895 | */ | |
896 | ||
897 | /* | |
898 | * dput - release a dentry | |
899 | * @dentry: dentry to release | |
900 | * | |
901 | * Release a dentry. This will drop the usage count and if appropriate | |
902 | * call the dentry unlink method as well as removing it from the queues and | |
903 | * releasing its resources. If the parent dentries were scheduled for release | |
904 | * they too may now get deleted. | |
1da177e4 | 905 | */ |
1da177e4 LT |
906 | void dput(struct dentry *dentry) |
907 | { | |
1088a640 AV |
908 | while (dentry) { |
909 | might_sleep(); | |
1da177e4 | 910 | |
1088a640 AV |
911 | rcu_read_lock(); |
912 | if (likely(fast_dput(dentry))) { | |
913 | rcu_read_unlock(); | |
914 | return; | |
915 | } | |
47be6184 | 916 | |
1088a640 | 917 | /* Slow case: now with the dentry lock held */ |
360f5479 | 918 | rcu_read_unlock(); |
2f42f1eb | 919 | dentry->d_lockref.count = 1; |
1088a640 | 920 | dentry = dentry_kill(dentry); |
47be6184 | 921 | } |
1da177e4 | 922 | } |
ec4f8605 | 923 | EXPORT_SYMBOL(dput); |
1da177e4 | 924 | |
6511f6be | 925 | static void to_shrink_list(struct dentry *dentry, struct list_head *list) |
9bdebc2b AV |
926 | __must_hold(&dentry->d_lock) |
927 | { | |
6511f6be | 928 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { |
9bdebc2b AV |
929 | if (dentry->d_flags & DCACHE_LRU_LIST) |
930 | d_lru_del(dentry); | |
6511f6be | 931 | if (!dentry->d_lockref.count) |
9bdebc2b AV |
932 | d_shrink_add(dentry, list); |
933 | } | |
934 | } | |
935 | ||
936 | void dput_to_list(struct dentry *dentry, struct list_head *list) | |
937 | { | |
938 | rcu_read_lock(); | |
939 | if (likely(fast_dput(dentry))) { | |
940 | rcu_read_unlock(); | |
941 | return; | |
942 | } | |
943 | rcu_read_unlock(); | |
f05441c7 | 944 | to_shrink_list(dentry, list); |
9bdebc2b AV |
945 | spin_unlock(&dentry->d_lock); |
946 | } | |
1da177e4 | 947 | |
b5c84bf6 | 948 | /* This must be called with d_lock held */ |
dc0474be | 949 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 950 | { |
98474236 | 951 | dentry->d_lockref.count++; |
23044507 NP |
952 | } |
953 | ||
dc0474be | 954 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 955 | { |
98474236 | 956 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
957 | } |
958 | ||
b7ab39f6 NP |
959 | struct dentry *dget_parent(struct dentry *dentry) |
960 | { | |
df3d0bbc | 961 | int gotref; |
b7ab39f6 | 962 | struct dentry *ret; |
e8400933 | 963 | unsigned seq; |
b7ab39f6 | 964 | |
df3d0bbc WL |
965 | /* |
966 | * Do optimistic parent lookup without any | |
967 | * locking. | |
968 | */ | |
969 | rcu_read_lock(); | |
e8400933 | 970 | seq = raw_seqcount_begin(&dentry->d_seq); |
66702eb5 | 971 | ret = READ_ONCE(dentry->d_parent); |
df3d0bbc WL |
972 | gotref = lockref_get_not_zero(&ret->d_lockref); |
973 | rcu_read_unlock(); | |
974 | if (likely(gotref)) { | |
e8400933 | 975 | if (!read_seqcount_retry(&dentry->d_seq, seq)) |
df3d0bbc WL |
976 | return ret; |
977 | dput(ret); | |
978 | } | |
979 | ||
b7ab39f6 | 980 | repeat: |
a734eb45 NP |
981 | /* |
982 | * Don't need rcu_dereference because we re-check it was correct under | |
983 | * the lock. | |
984 | */ | |
985 | rcu_read_lock(); | |
b7ab39f6 | 986 | ret = dentry->d_parent; |
a734eb45 NP |
987 | spin_lock(&ret->d_lock); |
988 | if (unlikely(ret != dentry->d_parent)) { | |
989 | spin_unlock(&ret->d_lock); | |
990 | rcu_read_unlock(); | |
b7ab39f6 NP |
991 | goto repeat; |
992 | } | |
a734eb45 | 993 | rcu_read_unlock(); |
98474236 WL |
994 | BUG_ON(!ret->d_lockref.count); |
995 | ret->d_lockref.count++; | |
b7ab39f6 | 996 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
997 | return ret; |
998 | } | |
999 | EXPORT_SYMBOL(dget_parent); | |
1000 | ||
61fec493 AV |
1001 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1002 | { | |
1003 | struct dentry *alias; | |
1004 | ||
1005 | if (hlist_empty(&inode->i_dentry)) | |
1006 | return NULL; | |
1007 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); | |
1008 | __dget(alias); | |
1009 | return alias; | |
1010 | } | |
1011 | ||
1012 | /** | |
1013 | * d_find_any_alias - find any alias for a given inode | |
1014 | * @inode: inode to find an alias for | |
1015 | * | |
1016 | * If any aliases exist for the given inode, take and return a | |
1017 | * reference for one of them. If no aliases exist, return %NULL. | |
1018 | */ | |
1019 | struct dentry *d_find_any_alias(struct inode *inode) | |
1020 | { | |
1021 | struct dentry *de; | |
1022 | ||
1023 | spin_lock(&inode->i_lock); | |
1024 | de = __d_find_any_alias(inode); | |
1025 | spin_unlock(&inode->i_lock); | |
1026 | return de; | |
1027 | } | |
1028 | EXPORT_SYMBOL(d_find_any_alias); | |
1029 | ||
52ed46f0 | 1030 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 1031 | { |
61fec493 AV |
1032 | struct dentry *alias; |
1033 | ||
1034 | if (S_ISDIR(inode->i_mode)) | |
1035 | return __d_find_any_alias(inode); | |
1da177e4 | 1036 | |
946e51f2 | 1037 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 1038 | spin_lock(&alias->d_lock); |
61fec493 | 1039 | if (!d_unhashed(alias)) { |
8d80d7da BF |
1040 | __dget_dlock(alias); |
1041 | spin_unlock(&alias->d_lock); | |
1042 | return alias; | |
1da177e4 | 1043 | } |
da502956 | 1044 | spin_unlock(&alias->d_lock); |
1da177e4 | 1045 | } |
da502956 | 1046 | return NULL; |
1da177e4 LT |
1047 | } |
1048 | ||
961f3c89 MCC |
1049 | /** |
1050 | * d_find_alias - grab a hashed alias of inode | |
1051 | * @inode: inode in question | |
1052 | * | |
1053 | * If inode has a hashed alias, or is a directory and has any alias, | |
1054 | * acquire the reference to alias and return it. Otherwise return NULL. | |
1055 | * Notice that if inode is a directory there can be only one alias and | |
1056 | * it can be unhashed only if it has no children, or if it is the root | |
1057 | * of a filesystem, or if the directory was renamed and d_revalidate | |
1058 | * was the first vfs operation to notice. | |
1059 | * | |
1060 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer | |
1061 | * any other hashed alias over that one. | |
1062 | */ | |
da502956 | 1063 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 1064 | { |
214fda1f DH |
1065 | struct dentry *de = NULL; |
1066 | ||
b3d9b7a3 | 1067 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 1068 | spin_lock(&inode->i_lock); |
52ed46f0 | 1069 | de = __d_find_alias(inode); |
873feea0 | 1070 | spin_unlock(&inode->i_lock); |
214fda1f | 1071 | } |
1da177e4 LT |
1072 | return de; |
1073 | } | |
ec4f8605 | 1074 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 | 1075 | |
bca585d2 AV |
1076 | /* |
1077 | * Caller MUST be holding rcu_read_lock() and be guaranteed | |
1078 | * that inode won't get freed until rcu_read_unlock(). | |
1079 | */ | |
1080 | struct dentry *d_find_alias_rcu(struct inode *inode) | |
1081 | { | |
1082 | struct hlist_head *l = &inode->i_dentry; | |
1083 | struct dentry *de = NULL; | |
1084 | ||
1085 | spin_lock(&inode->i_lock); | |
1086 | // ->i_dentry and ->i_rcu are colocated, but the latter won't be | |
1087 | // used without having I_FREEING set, which means no aliases left | |
1088 | if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) { | |
1089 | if (S_ISDIR(inode->i_mode)) { | |
1090 | de = hlist_entry(l->first, struct dentry, d_u.d_alias); | |
1091 | } else { | |
1092 | hlist_for_each_entry(de, l, d_u.d_alias) | |
1093 | if (!d_unhashed(de)) | |
1094 | break; | |
1095 | } | |
1096 | } | |
1097 | spin_unlock(&inode->i_lock); | |
1098 | return de; | |
1099 | } | |
1100 | ||
1da177e4 LT |
1101 | /* |
1102 | * Try to kill dentries associated with this inode. | |
1103 | * WARNING: you must own a reference to inode. | |
1104 | */ | |
1105 | void d_prune_aliases(struct inode *inode) | |
1106 | { | |
0cdca3f9 | 1107 | struct dentry *dentry; |
1da177e4 | 1108 | restart: |
873feea0 | 1109 | spin_lock(&inode->i_lock); |
946e51f2 | 1110 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 1111 | spin_lock(&dentry->d_lock); |
98474236 | 1112 | if (!dentry->d_lockref.count) { |
29355c39 AV |
1113 | struct dentry *parent = lock_parent(dentry); |
1114 | if (likely(!dentry->d_lockref.count)) { | |
1115 | __dentry_kill(dentry); | |
4a7795d3 | 1116 | dput(parent); |
29355c39 AV |
1117 | goto restart; |
1118 | } | |
1119 | if (parent) | |
1120 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
1121 | } |
1122 | spin_unlock(&dentry->d_lock); | |
1123 | } | |
873feea0 | 1124 | spin_unlock(&inode->i_lock); |
1da177e4 | 1125 | } |
ec4f8605 | 1126 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 1127 | |
3fcf5356 AV |
1128 | static inline void shrink_kill(struct dentry *victim, struct list_head *list) |
1129 | { | |
1130 | struct dentry *parent = victim->d_parent; | |
6511f6be AV |
1131 | if (parent != victim) { |
1132 | --parent->d_lockref.count; | |
1133 | to_shrink_list(parent, list); | |
1134 | } | |
3fcf5356 AV |
1135 | __dentry_kill(victim); |
1136 | } | |
1137 | ||
9bdebc2b | 1138 | void shrink_dentry_list(struct list_head *list) |
3b3f09f4 AV |
1139 | { |
1140 | while (!list_empty(list)) { | |
3fcf5356 | 1141 | struct dentry *dentry; |
64fd72e0 | 1142 | |
3b3f09f4 AV |
1143 | dentry = list_entry(list->prev, struct dentry, d_lru); |
1144 | spin_lock(&dentry->d_lock); | |
8f04da2a | 1145 | rcu_read_lock(); |
339e9e13 | 1146 | if (!lock_for_kill(dentry)) { |
cd9f84f3 | 1147 | bool can_free; |
8f04da2a | 1148 | rcu_read_unlock(); |
3b3f09f4 | 1149 | d_shrink_del(dentry); |
cd9f84f3 | 1150 | can_free = dentry->d_flags & DCACHE_MAY_FREE; |
64fd72e0 AV |
1151 | spin_unlock(&dentry->d_lock); |
1152 | if (can_free) | |
1153 | dentry_free(dentry); | |
1154 | continue; | |
1155 | } | |
8f04da2a | 1156 | rcu_read_unlock(); |
3b3f09f4 | 1157 | d_shrink_del(dentry); |
3fcf5356 | 1158 | shrink_kill(dentry, list); |
da3bbdd4 | 1159 | } |
3049cfe2 CH |
1160 | } |
1161 | ||
3f97b163 VD |
1162 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1163 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1164 | { |
1165 | struct list_head *freeable = arg; | |
1166 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1167 | ||
1168 | ||
1169 | /* | |
1170 | * we are inverting the lru lock/dentry->d_lock here, | |
1171 | * so use a trylock. If we fail to get the lock, just skip | |
1172 | * it | |
1173 | */ | |
1174 | if (!spin_trylock(&dentry->d_lock)) | |
1175 | return LRU_SKIP; | |
1176 | ||
1177 | /* | |
1178 | * Referenced dentries are still in use. If they have active | |
1179 | * counts, just remove them from the LRU. Otherwise give them | |
1180 | * another pass through the LRU. | |
1181 | */ | |
1182 | if (dentry->d_lockref.count) { | |
3f97b163 | 1183 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1184 | spin_unlock(&dentry->d_lock); |
1185 | return LRU_REMOVED; | |
1186 | } | |
1187 | ||
1188 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1189 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1190 | spin_unlock(&dentry->d_lock); | |
1191 | ||
1192 | /* | |
1193 | * The list move itself will be made by the common LRU code. At | |
1194 | * this point, we've dropped the dentry->d_lock but keep the | |
1195 | * lru lock. This is safe to do, since every list movement is | |
1196 | * protected by the lru lock even if both locks are held. | |
1197 | * | |
1198 | * This is guaranteed by the fact that all LRU management | |
1199 | * functions are intermediated by the LRU API calls like | |
1200 | * list_lru_add and list_lru_del. List movement in this file | |
1201 | * only ever occur through this functions or through callbacks | |
1202 | * like this one, that are called from the LRU API. | |
1203 | * | |
1204 | * The only exceptions to this are functions like | |
1205 | * shrink_dentry_list, and code that first checks for the | |
1206 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1207 | * operating only with stack provided lists after they are | |
1208 | * properly isolated from the main list. It is thus, always a | |
1209 | * local access. | |
1210 | */ | |
1211 | return LRU_ROTATE; | |
1212 | } | |
1213 | ||
3f97b163 | 1214 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1215 | spin_unlock(&dentry->d_lock); |
1216 | ||
1217 | return LRU_REMOVED; | |
1218 | } | |
1219 | ||
3049cfe2 | 1220 | /** |
b48f03b3 DC |
1221 | * prune_dcache_sb - shrink the dcache |
1222 | * @sb: superblock | |
503c358c | 1223 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1224 | * |
503c358c VD |
1225 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1226 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1227 | * function. |
3049cfe2 | 1228 | * |
b48f03b3 DC |
1229 | * This function may fail to free any resources if all the dentries are in |
1230 | * use. | |
3049cfe2 | 1231 | */ |
503c358c | 1232 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1233 | { |
f6041567 DC |
1234 | LIST_HEAD(dispose); |
1235 | long freed; | |
3049cfe2 | 1236 | |
503c358c VD |
1237 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1238 | dentry_lru_isolate, &dispose); | |
f6041567 | 1239 | shrink_dentry_list(&dispose); |
0a234c6d | 1240 | return freed; |
da3bbdd4 | 1241 | } |
23044507 | 1242 | |
4e717f5c | 1243 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1244 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1245 | { |
4e717f5c GC |
1246 | struct list_head *freeable = arg; |
1247 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1248 | |
4e717f5c GC |
1249 | /* |
1250 | * we are inverting the lru lock/dentry->d_lock here, | |
1251 | * so use a trylock. If we fail to get the lock, just skip | |
1252 | * it | |
1253 | */ | |
1254 | if (!spin_trylock(&dentry->d_lock)) | |
1255 | return LRU_SKIP; | |
1256 | ||
3f97b163 | 1257 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1258 | spin_unlock(&dentry->d_lock); |
ec33679d | 1259 | |
4e717f5c | 1260 | return LRU_REMOVED; |
da3bbdd4 KM |
1261 | } |
1262 | ||
4e717f5c | 1263 | |
1da177e4 LT |
1264 | /** |
1265 | * shrink_dcache_sb - shrink dcache for a superblock | |
1266 | * @sb: superblock | |
1267 | * | |
3049cfe2 CH |
1268 | * Shrink the dcache for the specified super block. This is used to free |
1269 | * the dcache before unmounting a file system. | |
1da177e4 | 1270 | */ |
3049cfe2 | 1271 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1272 | { |
4e717f5c GC |
1273 | do { |
1274 | LIST_HEAD(dispose); | |
1275 | ||
1dbd449c | 1276 | list_lru_walk(&sb->s_dentry_lru, |
b17c070f | 1277 | dentry_lru_isolate_shrink, &dispose, 1024); |
4e717f5c | 1278 | shrink_dentry_list(&dispose); |
b17c070f | 1279 | } while (list_lru_count(&sb->s_dentry_lru) > 0); |
1da177e4 | 1280 | } |
ec4f8605 | 1281 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1282 | |
db14fc3a MS |
1283 | /** |
1284 | * enum d_walk_ret - action to talke during tree walk | |
1285 | * @D_WALK_CONTINUE: contrinue walk | |
1286 | * @D_WALK_QUIT: quit walk | |
1287 | * @D_WALK_NORETRY: quit when retry is needed | |
1288 | * @D_WALK_SKIP: skip this dentry and its children | |
1289 | */ | |
1290 | enum d_walk_ret { | |
1291 | D_WALK_CONTINUE, | |
1292 | D_WALK_QUIT, | |
1293 | D_WALK_NORETRY, | |
1294 | D_WALK_SKIP, | |
1295 | }; | |
c826cb7d | 1296 | |
1da177e4 | 1297 | /** |
db14fc3a MS |
1298 | * d_walk - walk the dentry tree |
1299 | * @parent: start of walk | |
1300 | * @data: data passed to @enter() and @finish() | |
1301 | * @enter: callback when first entering the dentry | |
1da177e4 | 1302 | * |
3a8e3611 | 1303 | * The @enter() callbacks are called with d_lock held. |
1da177e4 | 1304 | */ |
db14fc3a | 1305 | static void d_walk(struct dentry *parent, void *data, |
3a8e3611 | 1306 | enum d_walk_ret (*enter)(void *, struct dentry *)) |
1da177e4 | 1307 | { |
da549bdd | 1308 | struct dentry *this_parent, *dentry; |
48f5ec21 | 1309 | unsigned seq = 0; |
db14fc3a MS |
1310 | enum d_walk_ret ret; |
1311 | bool retry = true; | |
949854d0 | 1312 | |
58db63d0 | 1313 | again: |
48f5ec21 | 1314 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1315 | this_parent = parent; |
2fd6b7f5 | 1316 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1317 | |
1318 | ret = enter(data, this_parent); | |
1319 | switch (ret) { | |
1320 | case D_WALK_CONTINUE: | |
1321 | break; | |
1322 | case D_WALK_QUIT: | |
1323 | case D_WALK_SKIP: | |
1324 | goto out_unlock; | |
1325 | case D_WALK_NORETRY: | |
1326 | retry = false; | |
1327 | break; | |
1328 | } | |
1da177e4 | 1329 | repeat: |
da549bdd | 1330 | dentry = d_first_child(this_parent); |
1da177e4 | 1331 | resume: |
da549bdd | 1332 | hlist_for_each_entry_from(dentry, d_sib) { |
ba65dc5e AV |
1333 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1334 | continue; | |
1335 | ||
2fd6b7f5 | 1336 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1337 | |
1338 | ret = enter(data, dentry); | |
1339 | switch (ret) { | |
1340 | case D_WALK_CONTINUE: | |
1341 | break; | |
1342 | case D_WALK_QUIT: | |
2fd6b7f5 | 1343 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1344 | goto out_unlock; |
1345 | case D_WALK_NORETRY: | |
1346 | retry = false; | |
1347 | break; | |
1348 | case D_WALK_SKIP: | |
1349 | spin_unlock(&dentry->d_lock); | |
1350 | continue; | |
2fd6b7f5 | 1351 | } |
db14fc3a | 1352 | |
da549bdd | 1353 | if (!hlist_empty(&dentry->d_children)) { |
2fd6b7f5 | 1354 | spin_unlock(&this_parent->d_lock); |
5facae4f | 1355 | spin_release(&dentry->d_lock.dep_map, _RET_IP_); |
1da177e4 | 1356 | this_parent = dentry; |
2fd6b7f5 | 1357 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1358 | goto repeat; |
1359 | } | |
2fd6b7f5 | 1360 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1361 | } |
1362 | /* | |
1363 | * All done at this level ... ascend and resume the search. | |
1364 | */ | |
ca5358ef AV |
1365 | rcu_read_lock(); |
1366 | ascend: | |
1da177e4 | 1367 | if (this_parent != parent) { |
da549bdd AV |
1368 | dentry = this_parent; |
1369 | this_parent = dentry->d_parent; | |
31dec132 | 1370 | |
da549bdd | 1371 | spin_unlock(&dentry->d_lock); |
31dec132 AV |
1372 | spin_lock(&this_parent->d_lock); |
1373 | ||
ca5358ef AV |
1374 | /* might go back up the wrong parent if we have had a rename. */ |
1375 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1376 | goto rename_retry; |
2159184e | 1377 | /* go into the first sibling still alive */ |
da549bdd AV |
1378 | hlist_for_each_entry_continue(dentry, d_sib) { |
1379 | if (likely(!(dentry->d_flags & DCACHE_DENTRY_KILLED))) { | |
1380 | rcu_read_unlock(); | |
1381 | goto resume; | |
1382 | } | |
1383 | } | |
1384 | goto ascend; | |
1da177e4 | 1385 | } |
ca5358ef | 1386 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1387 | goto rename_retry; |
ca5358ef | 1388 | rcu_read_unlock(); |
db14fc3a MS |
1389 | |
1390 | out_unlock: | |
1391 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1392 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1393 | return; |
58db63d0 NP |
1394 | |
1395 | rename_retry: | |
ca5358ef AV |
1396 | spin_unlock(&this_parent->d_lock); |
1397 | rcu_read_unlock(); | |
1398 | BUG_ON(seq & 1); | |
db14fc3a MS |
1399 | if (!retry) |
1400 | return; | |
48f5ec21 | 1401 | seq = 1; |
58db63d0 | 1402 | goto again; |
1da177e4 | 1403 | } |
db14fc3a | 1404 | |
01619491 IK |
1405 | struct check_mount { |
1406 | struct vfsmount *mnt; | |
1407 | unsigned int mounted; | |
1408 | }; | |
1409 | ||
1410 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1411 | { | |
1412 | struct check_mount *info = data; | |
1413 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1414 | ||
1415 | if (likely(!d_mountpoint(dentry))) | |
1416 | return D_WALK_CONTINUE; | |
1417 | if (__path_is_mountpoint(&path)) { | |
1418 | info->mounted = 1; | |
1419 | return D_WALK_QUIT; | |
1420 | } | |
1421 | return D_WALK_CONTINUE; | |
1422 | } | |
1423 | ||
1424 | /** | |
1425 | * path_has_submounts - check for mounts over a dentry in the | |
1426 | * current namespace. | |
1427 | * @parent: path to check. | |
1428 | * | |
1429 | * Return true if the parent or its subdirectories contain | |
1430 | * a mount point in the current namespace. | |
1431 | */ | |
1432 | int path_has_submounts(const struct path *parent) | |
1433 | { | |
1434 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1435 | ||
1436 | read_seqlock_excl(&mount_lock); | |
3a8e3611 | 1437 | d_walk(parent->dentry, &data, path_check_mount); |
01619491 IK |
1438 | read_sequnlock_excl(&mount_lock); |
1439 | ||
1440 | return data.mounted; | |
1441 | } | |
1442 | EXPORT_SYMBOL(path_has_submounts); | |
1443 | ||
eed81007 MS |
1444 | /* |
1445 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1446 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1447 | * subtree can become unreachable). | |
1448 | * | |
1ffe46d1 | 1449 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1450 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1451 | */ | |
1452 | int d_set_mounted(struct dentry *dentry) | |
1453 | { | |
1454 | struct dentry *p; | |
1455 | int ret = -ENOENT; | |
1456 | write_seqlock(&rename_lock); | |
1457 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1458 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1459 | spin_lock(&p->d_lock); |
1460 | if (unlikely(d_unhashed(p))) { | |
1461 | spin_unlock(&p->d_lock); | |
1462 | goto out; | |
1463 | } | |
1464 | spin_unlock(&p->d_lock); | |
1465 | } | |
1466 | spin_lock(&dentry->d_lock); | |
1467 | if (!d_unlinked(dentry)) { | |
3895dbf8 EB |
1468 | ret = -EBUSY; |
1469 | if (!d_mountpoint(dentry)) { | |
1470 | dentry->d_flags |= DCACHE_MOUNTED; | |
1471 | ret = 0; | |
1472 | } | |
eed81007 MS |
1473 | } |
1474 | spin_unlock(&dentry->d_lock); | |
1475 | out: | |
1476 | write_sequnlock(&rename_lock); | |
1477 | return ret; | |
1478 | } | |
1479 | ||
1da177e4 | 1480 | /* |
fd517909 | 1481 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1482 | * and move any unused dentries to the end of the unused |
1483 | * list for prune_dcache(). We descend to the next level | |
da549bdd | 1484 | * whenever the d_children list is non-empty and continue |
1da177e4 LT |
1485 | * searching. |
1486 | * | |
1487 | * It returns zero iff there are no unused children, | |
1488 | * otherwise it returns the number of children moved to | |
1489 | * the end of the unused list. This may not be the total | |
1490 | * number of unused children, because select_parent can | |
1491 | * drop the lock and return early due to latency | |
1492 | * constraints. | |
1493 | */ | |
1da177e4 | 1494 | |
db14fc3a MS |
1495 | struct select_data { |
1496 | struct dentry *start; | |
9bdebc2b AV |
1497 | union { |
1498 | long found; | |
1499 | struct dentry *victim; | |
1500 | }; | |
db14fc3a | 1501 | struct list_head dispose; |
db14fc3a | 1502 | }; |
23044507 | 1503 | |
db14fc3a MS |
1504 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1505 | { | |
1506 | struct select_data *data = _data; | |
1507 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1508 | |
db14fc3a MS |
1509 | if (data->start == dentry) |
1510 | goto out; | |
2fd6b7f5 | 1511 | |
fe91522a | 1512 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1513 | data->found++; |
fe91522a AV |
1514 | } else { |
1515 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1516 | d_lru_del(dentry); | |
1517 | if (!dentry->d_lockref.count) { | |
1518 | d_shrink_add(dentry, &data->dispose); | |
1519 | data->found++; | |
1520 | } | |
1da177e4 | 1521 | } |
db14fc3a MS |
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 | */ | |
fe91522a AV |
1527 | if (!list_empty(&data->dispose)) |
1528 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1529 | out: |
db14fc3a | 1530 | return ret; |
1da177e4 LT |
1531 | } |
1532 | ||
9bdebc2b AV |
1533 | static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry) |
1534 | { | |
1535 | struct select_data *data = _data; | |
1536 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1537 | ||
1538 | if (data->start == dentry) | |
1539 | goto out; | |
1540 | ||
1541 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
1542 | if (!dentry->d_lockref.count) { | |
1543 | rcu_read_lock(); | |
1544 | data->victim = dentry; | |
1545 | return D_WALK_QUIT; | |
1546 | } | |
1547 | } else { | |
1548 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1549 | d_lru_del(dentry); | |
1550 | if (!dentry->d_lockref.count) | |
1551 | d_shrink_add(dentry, &data->dispose); | |
1552 | } | |
1553 | /* | |
1554 | * We can return to the caller if we have found some (this | |
1555 | * ensures forward progress). We'll be coming back to find | |
1556 | * the rest. | |
1557 | */ | |
1558 | if (!list_empty(&data->dispose)) | |
1559 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1560 | out: | |
1561 | return ret; | |
1562 | } | |
1563 | ||
1da177e4 LT |
1564 | /** |
1565 | * shrink_dcache_parent - prune dcache | |
1566 | * @parent: parent of entries to prune | |
1567 | * | |
1568 | * Prune the dcache to remove unused children of the parent dentry. | |
1569 | */ | |
db14fc3a | 1570 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1571 | { |
db14fc3a | 1572 | for (;;) { |
9bdebc2b | 1573 | struct select_data data = {.start = parent}; |
1da177e4 | 1574 | |
db14fc3a | 1575 | INIT_LIST_HEAD(&data.dispose); |
3a8e3611 | 1576 | d_walk(parent, &data, select_collect); |
4fb48871 AV |
1577 | |
1578 | if (!list_empty(&data.dispose)) { | |
1579 | shrink_dentry_list(&data.dispose); | |
1580 | continue; | |
1581 | } | |
1582 | ||
1583 | cond_resched(); | |
db14fc3a MS |
1584 | if (!data.found) |
1585 | break; | |
9bdebc2b AV |
1586 | data.victim = NULL; |
1587 | d_walk(parent, &data, select_collect2); | |
1588 | if (data.victim) { | |
9bdebc2b | 1589 | spin_lock(&data.victim->d_lock); |
339e9e13 | 1590 | if (!lock_for_kill(data.victim)) { |
9bdebc2b AV |
1591 | spin_unlock(&data.victim->d_lock); |
1592 | rcu_read_unlock(); | |
1593 | } else { | |
1594 | rcu_read_unlock(); | |
3fcf5356 | 1595 | shrink_kill(data.victim, &data.dispose); |
9bdebc2b AV |
1596 | } |
1597 | } | |
1598 | if (!list_empty(&data.dispose)) | |
1599 | shrink_dentry_list(&data.dispose); | |
421348f1 | 1600 | } |
1da177e4 | 1601 | } |
ec4f8605 | 1602 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1603 | |
9c8c10e2 | 1604 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1605 | { |
9c8c10e2 | 1606 | /* it has busy descendents; complain about those instead */ |
da549bdd | 1607 | if (!hlist_empty(&dentry->d_children)) |
9c8c10e2 | 1608 | return D_WALK_CONTINUE; |
42c32608 | 1609 | |
9c8c10e2 AV |
1610 | /* root with refcount 1 is fine */ |
1611 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1612 | return D_WALK_CONTINUE; | |
1613 | ||
8c8e7dba | 1614 | WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} " |
9c8c10e2 | 1615 | " still in use (%d) [unmount of %s %s]\n", |
42c32608 AV |
1616 | dentry, |
1617 | dentry->d_inode ? | |
1618 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1619 | dentry, |
42c32608 AV |
1620 | dentry->d_lockref.count, |
1621 | dentry->d_sb->s_type->name, | |
1622 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1623 | return D_WALK_CONTINUE; |
1624 | } | |
1625 | ||
1626 | static void do_one_tree(struct dentry *dentry) | |
1627 | { | |
1628 | shrink_dcache_parent(dentry); | |
3a8e3611 | 1629 | d_walk(dentry, dentry, umount_check); |
9c8c10e2 AV |
1630 | d_drop(dentry); |
1631 | dput(dentry); | |
42c32608 AV |
1632 | } |
1633 | ||
1634 | /* | |
1635 | * destroy the dentries attached to a superblock on unmounting | |
1636 | */ | |
1637 | void shrink_dcache_for_umount(struct super_block *sb) | |
1638 | { | |
1639 | struct dentry *dentry; | |
1640 | ||
9c8c10e2 | 1641 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1642 | |
1643 | dentry = sb->s_root; | |
1644 | sb->s_root = NULL; | |
9c8c10e2 | 1645 | do_one_tree(dentry); |
42c32608 | 1646 | |
f1ee6162 N |
1647 | while (!hlist_bl_empty(&sb->s_roots)) { |
1648 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); | |
9c8c10e2 | 1649 | do_one_tree(dentry); |
42c32608 AV |
1650 | } |
1651 | } | |
1652 | ||
ff17fa56 | 1653 | static enum d_walk_ret find_submount(void *_data, struct dentry *dentry) |
848ac114 | 1654 | { |
ff17fa56 | 1655 | struct dentry **victim = _data; |
848ac114 | 1656 | if (d_mountpoint(dentry)) { |
8ed936b5 | 1657 | __dget_dlock(dentry); |
ff17fa56 | 1658 | *victim = dentry; |
848ac114 MS |
1659 | return D_WALK_QUIT; |
1660 | } | |
ff17fa56 | 1661 | return D_WALK_CONTINUE; |
848ac114 MS |
1662 | } |
1663 | ||
1664 | /** | |
1ffe46d1 EB |
1665 | * d_invalidate - detach submounts, prune dcache, and drop |
1666 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
848ac114 | 1667 | */ |
5542aa2f | 1668 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1669 | { |
ff17fa56 | 1670 | bool had_submounts = false; |
1ffe46d1 EB |
1671 | spin_lock(&dentry->d_lock); |
1672 | if (d_unhashed(dentry)) { | |
1673 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1674 | return; |
1ffe46d1 | 1675 | } |
ff17fa56 | 1676 | __d_drop(dentry); |
1ffe46d1 EB |
1677 | spin_unlock(&dentry->d_lock); |
1678 | ||
848ac114 | 1679 | /* Negative dentries can be dropped without further checks */ |
ff17fa56 | 1680 | if (!dentry->d_inode) |
5542aa2f | 1681 | return; |
848ac114 | 1682 | |
ff17fa56 | 1683 | shrink_dcache_parent(dentry); |
848ac114 | 1684 | for (;;) { |
ff17fa56 | 1685 | struct dentry *victim = NULL; |
3a8e3611 | 1686 | d_walk(dentry, &victim, find_submount); |
ff17fa56 AV |
1687 | if (!victim) { |
1688 | if (had_submounts) | |
1689 | shrink_dcache_parent(dentry); | |
81be24d2 | 1690 | return; |
8ed936b5 | 1691 | } |
ff17fa56 AV |
1692 | had_submounts = true; |
1693 | detach_mounts(victim); | |
1694 | dput(victim); | |
848ac114 | 1695 | } |
848ac114 | 1696 | } |
1ffe46d1 | 1697 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1698 | |
1da177e4 | 1699 | /** |
a4464dbc AV |
1700 | * __d_alloc - allocate a dcache entry |
1701 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1702 | * @name: qstr of the name |
1703 | * | |
1704 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1705 | * available. On a success the dentry is returned. The name passed in is | |
1706 | * copied and the copy passed in may be reused after this call. | |
1707 | */ | |
1708 | ||
5c8b0dfc | 1709 | static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1710 | { |
1711 | struct dentry *dentry; | |
1712 | char *dname; | |
285b102d | 1713 | int err; |
1da177e4 | 1714 | |
f53bf711 MS |
1715 | dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru, |
1716 | GFP_KERNEL); | |
1da177e4 LT |
1717 | if (!dentry) |
1718 | return NULL; | |
1719 | ||
6326c71f LT |
1720 | /* |
1721 | * We guarantee that the inline name is always NUL-terminated. | |
1722 | * This way the memcpy() done by the name switching in rename | |
1723 | * will still always have a NUL at the end, even if we might | |
1724 | * be overwriting an internal NUL character | |
1725 | */ | |
1726 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd | 1727 | if (unlikely(!name)) { |
cdf01226 | 1728 | name = &slash_name; |
798434bd AV |
1729 | dname = dentry->d_iname; |
1730 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1731 | size_t size = offsetof(struct external_name, name[1]); |
2e03b4bc VB |
1732 | struct external_name *p = kmalloc(size + name->len, |
1733 | GFP_KERNEL_ACCOUNT | | |
1734 | __GFP_RECLAIMABLE); | |
1735 | if (!p) { | |
1da177e4 LT |
1736 | kmem_cache_free(dentry_cache, dentry); |
1737 | return NULL; | |
1738 | } | |
2e03b4bc VB |
1739 | atomic_set(&p->u.count, 1); |
1740 | dname = p->name; | |
1da177e4 LT |
1741 | } else { |
1742 | dname = dentry->d_iname; | |
1743 | } | |
1da177e4 LT |
1744 | |
1745 | dentry->d_name.len = name->len; | |
1746 | dentry->d_name.hash = name->hash; | |
1747 | memcpy(dname, name->name, name->len); | |
1748 | dname[name->len] = 0; | |
1749 | ||
6326c71f | 1750 | /* Make sure we always see the terminating NUL character */ |
7088efa9 | 1751 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ |
6326c71f | 1752 | |
98474236 | 1753 | dentry->d_lockref.count = 1; |
dea3667b | 1754 | dentry->d_flags = 0; |
1da177e4 | 1755 | spin_lock_init(&dentry->d_lock); |
26475371 | 1756 | seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock); |
1da177e4 | 1757 | dentry->d_inode = NULL; |
a4464dbc AV |
1758 | dentry->d_parent = dentry; |
1759 | dentry->d_sb = sb; | |
1da177e4 LT |
1760 | dentry->d_op = NULL; |
1761 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1762 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 | 1763 | INIT_LIST_HEAD(&dentry->d_lru); |
da549bdd | 1764 | INIT_HLIST_HEAD(&dentry->d_children); |
946e51f2 | 1765 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
da549bdd | 1766 | INIT_HLIST_NODE(&dentry->d_sib); |
a4464dbc | 1767 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1768 | |
285b102d MS |
1769 | if (dentry->d_op && dentry->d_op->d_init) { |
1770 | err = dentry->d_op->d_init(dentry); | |
1771 | if (err) { | |
1772 | if (dname_external(dentry)) | |
1773 | kfree(external_name(dentry)); | |
1774 | kmem_cache_free(dentry_cache, dentry); | |
1775 | return NULL; | |
1776 | } | |
1777 | } | |
1778 | ||
3e880fb5 | 1779 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1780 | |
1da177e4 LT |
1781 | return dentry; |
1782 | } | |
a4464dbc AV |
1783 | |
1784 | /** | |
1785 | * d_alloc - allocate a dcache entry | |
1786 | * @parent: parent of entry to allocate | |
1787 | * @name: qstr of the name | |
1788 | * | |
1789 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1790 | * available. On a success the dentry is returned. The name passed in is | |
1791 | * copied and the copy passed in may be reused after this call. | |
1792 | */ | |
1793 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1794 | { | |
1795 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1796 | if (!dentry) | |
1797 | return NULL; | |
a4464dbc AV |
1798 | spin_lock(&parent->d_lock); |
1799 | /* | |
1800 | * don't need child lock because it is not subject | |
1801 | * to concurrency here | |
1802 | */ | |
1803 | __dget_dlock(parent); | |
1804 | dentry->d_parent = parent; | |
da549bdd | 1805 | hlist_add_head(&dentry->d_sib, &parent->d_children); |
a4464dbc AV |
1806 | spin_unlock(&parent->d_lock); |
1807 | ||
1808 | return dentry; | |
1809 | } | |
ec4f8605 | 1810 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1811 | |
f9c34674 MS |
1812 | struct dentry *d_alloc_anon(struct super_block *sb) |
1813 | { | |
1814 | return __d_alloc(sb, NULL); | |
1815 | } | |
1816 | EXPORT_SYMBOL(d_alloc_anon); | |
1817 | ||
ba65dc5e AV |
1818 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1819 | { | |
f9c34674 | 1820 | struct dentry *dentry = d_alloc_anon(parent->d_sb); |
ba65dc5e | 1821 | if (dentry) { |
5467a68c | 1822 | dentry->d_flags |= DCACHE_DENTRY_CURSOR; |
ba65dc5e AV |
1823 | dentry->d_parent = dget(parent); |
1824 | } | |
1825 | return dentry; | |
1826 | } | |
1827 | ||
e1a24bb0 BF |
1828 | /** |
1829 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1830 | * @sb: the superblock | |
1831 | * @name: qstr of the name | |
1832 | * | |
1833 | * For a filesystem that just pins its dentries in memory and never | |
1834 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
5467a68c AV |
1835 | * This is used for pipes, sockets et.al. - the stuff that should |
1836 | * never be anyone's children or parents. Unlike all other | |
1837 | * dentries, these will not have RCU delay between dropping the | |
1838 | * last reference and freeing them. | |
ab1152dd AV |
1839 | * |
1840 | * The only user is alloc_file_pseudo() and that's what should | |
1841 | * be considered a public interface. Don't use directly. | |
e1a24bb0 | 1842 | */ |
4b936885 NP |
1843 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1844 | { | |
5467a68c AV |
1845 | struct dentry *dentry = __d_alloc(sb, name); |
1846 | if (likely(dentry)) | |
1847 | dentry->d_flags |= DCACHE_NORCU; | |
1848 | return dentry; | |
4b936885 | 1849 | } |
4b936885 | 1850 | |
1da177e4 LT |
1851 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1852 | { | |
1853 | struct qstr q; | |
1854 | ||
1855 | q.name = name; | |
8387ff25 | 1856 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1857 | return d_alloc(parent, &q); |
1858 | } | |
ef26ca97 | 1859 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1860 | |
fb045adb NP |
1861 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1862 | { | |
6f7f7caa LT |
1863 | WARN_ON_ONCE(dentry->d_op); |
1864 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1865 | DCACHE_OP_COMPARE | |
1866 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1867 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1868 | DCACHE_OP_DELETE | |
d101a125 | 1869 | DCACHE_OP_REAL)); |
fb045adb NP |
1870 | dentry->d_op = op; |
1871 | if (!op) | |
1872 | return; | |
1873 | if (op->d_hash) | |
1874 | dentry->d_flags |= DCACHE_OP_HASH; | |
1875 | if (op->d_compare) | |
1876 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1877 | if (op->d_revalidate) | |
1878 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1879 | if (op->d_weak_revalidate) |
1880 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1881 | if (op->d_delete) |
1882 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1883 | if (op->d_prune) |
1884 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1885 | if (op->d_real) |
1886 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1887 | |
1888 | } | |
1889 | EXPORT_SYMBOL(d_set_d_op); | |
1890 | ||
df1a085a DH |
1891 | |
1892 | /* | |
1893 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1894 | * @dentry - The dentry to mark | |
1895 | * | |
1896 | * Mark a dentry as falling through to the lower layer (as set with | |
1897 | * d_pin_lower()). This flag may be recorded on the medium. | |
1898 | */ | |
1899 | void d_set_fallthru(struct dentry *dentry) | |
1900 | { | |
1901 | spin_lock(&dentry->d_lock); | |
1902 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1903 | spin_unlock(&dentry->d_lock); | |
1904 | } | |
1905 | EXPORT_SYMBOL(d_set_fallthru); | |
1906 | ||
b18825a7 DH |
1907 | static unsigned d_flags_for_inode(struct inode *inode) |
1908 | { | |
44bdb5e5 | 1909 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1910 | |
1911 | if (!inode) | |
1912 | return DCACHE_MISS_TYPE; | |
1913 | ||
1914 | if (S_ISDIR(inode->i_mode)) { | |
1915 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1916 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1917 | if (unlikely(!inode->i_op->lookup)) | |
1918 | add_flags = DCACHE_AUTODIR_TYPE; | |
1919 | else | |
1920 | inode->i_opflags |= IOP_LOOKUP; | |
1921 | } | |
44bdb5e5 DH |
1922 | goto type_determined; |
1923 | } | |
1924 | ||
1925 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1926 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1927 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1928 | goto type_determined; |
1929 | } | |
1930 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1931 | } |
1932 | ||
44bdb5e5 DH |
1933 | if (unlikely(!S_ISREG(inode->i_mode))) |
1934 | add_flags = DCACHE_SPECIAL_TYPE; | |
1935 | ||
1936 | type_determined: | |
b18825a7 DH |
1937 | if (unlikely(IS_AUTOMOUNT(inode))) |
1938 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1939 | return add_flags; | |
1940 | } | |
1941 | ||
360da900 OH |
1942 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1943 | { | |
b18825a7 | 1944 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1945 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1946 | |
b23fb0a6 | 1947 | spin_lock(&dentry->d_lock); |
af0c9af1 WL |
1948 | /* |
1949 | * Decrement negative dentry count if it was in the LRU list. | |
1950 | */ | |
1951 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1952 | this_cpu_dec(nr_dentry_negative); | |
de689f5e | 1953 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1954 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1955 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1956 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1957 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1958 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1959 | } |
1960 | ||
1da177e4 LT |
1961 | /** |
1962 | * d_instantiate - fill in inode information for a dentry | |
1963 | * @entry: dentry to complete | |
1964 | * @inode: inode to attach to this dentry | |
1965 | * | |
1966 | * Fill in inode information in the entry. | |
1967 | * | |
1968 | * This turns negative dentries into productive full members | |
1969 | * of society. | |
1970 | * | |
1971 | * NOTE! This assumes that the inode count has been incremented | |
1972 | * (or otherwise set) by the caller to indicate that it is now | |
1973 | * in use by the dcache. | |
1974 | */ | |
1975 | ||
1976 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1977 | { | |
946e51f2 | 1978 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1979 | if (inode) { |
b9680917 | 1980 | security_d_instantiate(entry, inode); |
873feea0 | 1981 | spin_lock(&inode->i_lock); |
de689f5e | 1982 | __d_instantiate(entry, inode); |
873feea0 | 1983 | spin_unlock(&inode->i_lock); |
de689f5e | 1984 | } |
1da177e4 | 1985 | } |
ec4f8605 | 1986 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1987 | |
1e2e547a AV |
1988 | /* |
1989 | * This should be equivalent to d_instantiate() + unlock_new_inode(), | |
1990 | * with lockdep-related part of unlock_new_inode() done before | |
1991 | * anything else. Use that instead of open-coding d_instantiate()/ | |
1992 | * unlock_new_inode() combinations. | |
1993 | */ | |
1994 | void d_instantiate_new(struct dentry *entry, struct inode *inode) | |
1995 | { | |
1996 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); | |
1997 | BUG_ON(!inode); | |
1998 | lockdep_annotate_inode_mutex_key(inode); | |
1999 | security_d_instantiate(entry, inode); | |
2000 | spin_lock(&inode->i_lock); | |
2001 | __d_instantiate(entry, inode); | |
2002 | WARN_ON(!(inode->i_state & I_NEW)); | |
c2b6d621 | 2003 | inode->i_state &= ~I_NEW & ~I_CREATING; |
1e2e547a AV |
2004 | smp_mb(); |
2005 | wake_up_bit(&inode->i_state, __I_NEW); | |
2006 | spin_unlock(&inode->i_lock); | |
2007 | } | |
2008 | EXPORT_SYMBOL(d_instantiate_new); | |
2009 | ||
adc0e91a AV |
2010 | struct dentry *d_make_root(struct inode *root_inode) |
2011 | { | |
2012 | struct dentry *res = NULL; | |
2013 | ||
2014 | if (root_inode) { | |
f9c34674 | 2015 | res = d_alloc_anon(root_inode->i_sb); |
5467a68c | 2016 | if (res) |
adc0e91a | 2017 | d_instantiate(res, root_inode); |
5467a68c | 2018 | else |
adc0e91a AV |
2019 | iput(root_inode); |
2020 | } | |
2021 | return res; | |
2022 | } | |
2023 | EXPORT_SYMBOL(d_make_root); | |
2024 | ||
f9c34674 MS |
2025 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, |
2026 | struct inode *inode, | |
2027 | bool disconnected) | |
4ea3ada2 | 2028 | { |
9308a612 | 2029 | struct dentry *res; |
b18825a7 | 2030 | unsigned add_flags; |
4ea3ada2 | 2031 | |
f9c34674 | 2032 | security_d_instantiate(dentry, inode); |
873feea0 | 2033 | spin_lock(&inode->i_lock); |
d891eedb | 2034 | res = __d_find_any_alias(inode); |
9308a612 | 2035 | if (res) { |
873feea0 | 2036 | spin_unlock(&inode->i_lock); |
f9c34674 | 2037 | dput(dentry); |
9308a612 CH |
2038 | goto out_iput; |
2039 | } | |
2040 | ||
2041 | /* attach a disconnected dentry */ | |
1a0a397e BF |
2042 | add_flags = d_flags_for_inode(inode); |
2043 | ||
2044 | if (disconnected) | |
2045 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 2046 | |
f9c34674 MS |
2047 | spin_lock(&dentry->d_lock); |
2048 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2049 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
f1ee6162 | 2050 | if (!disconnected) { |
139351f1 LT |
2051 | hlist_bl_lock(&dentry->d_sb->s_roots); |
2052 | hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots); | |
2053 | hlist_bl_unlock(&dentry->d_sb->s_roots); | |
f1ee6162 | 2054 | } |
f9c34674 | 2055 | spin_unlock(&dentry->d_lock); |
873feea0 | 2056 | spin_unlock(&inode->i_lock); |
9308a612 | 2057 | |
f9c34674 | 2058 | return dentry; |
9308a612 CH |
2059 | |
2060 | out_iput: | |
2061 | iput(inode); | |
2062 | return res; | |
4ea3ada2 | 2063 | } |
1a0a397e | 2064 | |
f9c34674 MS |
2065 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) |
2066 | { | |
2067 | return __d_instantiate_anon(dentry, inode, true); | |
2068 | } | |
2069 | EXPORT_SYMBOL(d_instantiate_anon); | |
2070 | ||
2071 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) | |
2072 | { | |
2073 | struct dentry *tmp; | |
2074 | struct dentry *res; | |
2075 | ||
2076 | if (!inode) | |
2077 | return ERR_PTR(-ESTALE); | |
2078 | if (IS_ERR(inode)) | |
2079 | return ERR_CAST(inode); | |
2080 | ||
2081 | res = d_find_any_alias(inode); | |
2082 | if (res) | |
2083 | goto out_iput; | |
2084 | ||
2085 | tmp = d_alloc_anon(inode->i_sb); | |
2086 | if (!tmp) { | |
2087 | res = ERR_PTR(-ENOMEM); | |
2088 | goto out_iput; | |
2089 | } | |
2090 | ||
2091 | return __d_instantiate_anon(tmp, inode, disconnected); | |
2092 | ||
2093 | out_iput: | |
2094 | iput(inode); | |
2095 | return res; | |
2096 | } | |
2097 | ||
1a0a397e BF |
2098 | /** |
2099 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2100 | * @inode: inode to allocate the dentry for | |
2101 | * | |
2102 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2103 | * similar open by handle operations. The returned dentry may be anonymous, | |
2104 | * or may have a full name (if the inode was already in the cache). | |
2105 | * | |
2106 | * When called on a directory inode, we must ensure that the inode only ever | |
2107 | * has one dentry. If a dentry is found, that is returned instead of | |
2108 | * allocating a new one. | |
2109 | * | |
2110 | * On successful return, the reference to the inode has been transferred | |
2111 | * to the dentry. In case of an error the reference on the inode is released. | |
2112 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2113 | * be passed in and the error will be propagated to the return value, | |
2114 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2115 | */ | |
2116 | struct dentry *d_obtain_alias(struct inode *inode) | |
2117 | { | |
f9c34674 | 2118 | return __d_obtain_alias(inode, true); |
1a0a397e | 2119 | } |
adc48720 | 2120 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 2121 | |
1a0a397e BF |
2122 | /** |
2123 | * d_obtain_root - find or allocate a dentry for a given inode | |
2124 | * @inode: inode to allocate the dentry for | |
2125 | * | |
2126 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2127 | * | |
2128 | * We must ensure that directory inodes only ever have one dentry. If a | |
2129 | * dentry is found, that is returned instead of allocating a new one. | |
2130 | * | |
2131 | * On successful return, the reference to the inode has been transferred | |
2132 | * to the dentry. In case of an error the reference on the inode is | |
2133 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2134 | * error will be propagate to the return value, with a %NULL @inode | |
2135 | * replaced by ERR_PTR(-ESTALE). | |
2136 | */ | |
2137 | struct dentry *d_obtain_root(struct inode *inode) | |
2138 | { | |
f9c34674 | 2139 | return __d_obtain_alias(inode, false); |
1a0a397e BF |
2140 | } |
2141 | EXPORT_SYMBOL(d_obtain_root); | |
2142 | ||
9403540c BN |
2143 | /** |
2144 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2145 | * @inode: the inode case-insensitive lookup has found | |
2146 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2147 | * @name: the case-exact name to be associated with the returned dentry | |
2148 | * | |
2149 | * This is to avoid filling the dcache with case-insensitive names to the | |
2150 | * same inode, only the actual correct case is stored in the dcache for | |
2151 | * case-insensitive filesystems. | |
2152 | * | |
3d742d4b RD |
2153 | * For a case-insensitive lookup match and if the case-exact dentry |
2154 | * already exists in the dcache, use it and return it. | |
9403540c BN |
2155 | * |
2156 | * If no entry exists with the exact case name, allocate new dentry with | |
2157 | * the exact case, and return the spliced entry. | |
2158 | */ | |
e45b590b | 2159 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2160 | struct qstr *name) |
2161 | { | |
d9171b93 | 2162 | struct dentry *found, *res; |
9403540c | 2163 | |
b6520c81 CH |
2164 | /* |
2165 | * First check if a dentry matching the name already exists, | |
2166 | * if not go ahead and create it now. | |
2167 | */ | |
9403540c | 2168 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2169 | if (found) { |
2170 | iput(inode); | |
2171 | return found; | |
2172 | } | |
2173 | if (d_in_lookup(dentry)) { | |
2174 | found = d_alloc_parallel(dentry->d_parent, name, | |
2175 | dentry->d_wait); | |
2176 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2177 | iput(inode); | |
2178 | return found; | |
9403540c | 2179 | } |
d9171b93 AV |
2180 | } else { |
2181 | found = d_alloc(dentry->d_parent, name); | |
2182 | if (!found) { | |
2183 | iput(inode); | |
2184 | return ERR_PTR(-ENOMEM); | |
2185 | } | |
2186 | } | |
2187 | res = d_splice_alias(inode, found); | |
2188 | if (res) { | |
40a3cb0d | 2189 | d_lookup_done(found); |
d9171b93 AV |
2190 | dput(found); |
2191 | return res; | |
9403540c | 2192 | } |
4f522a24 | 2193 | return found; |
9403540c | 2194 | } |
ec4f8605 | 2195 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2196 | |
4f48d5da XL |
2197 | /** |
2198 | * d_same_name - compare dentry name with case-exact name | |
2199 | * @parent: parent dentry | |
2200 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2201 | * @name: the case-exact name to be associated with the returned dentry | |
2202 | * | |
2203 | * Return: true if names are same, or false | |
2204 | */ | |
2205 | bool d_same_name(const struct dentry *dentry, const struct dentry *parent, | |
2206 | const struct qstr *name) | |
12f8ad4b | 2207 | { |
d4c91a8f AV |
2208 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2209 | if (dentry->d_name.len != name->len) | |
2210 | return false; | |
2211 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2212 | } |
6fa67e70 | 2213 | return parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2214 | dentry->d_name.len, dentry->d_name.name, |
2215 | name) == 0; | |
12f8ad4b | 2216 | } |
4f48d5da | 2217 | EXPORT_SYMBOL_GPL(d_same_name); |
12f8ad4b | 2218 | |
ae2a8236 LT |
2219 | /* |
2220 | * This is __d_lookup_rcu() when the parent dentry has | |
2221 | * DCACHE_OP_COMPARE, which makes things much nastier. | |
2222 | */ | |
2223 | static noinline struct dentry *__d_lookup_rcu_op_compare( | |
2224 | const struct dentry *parent, | |
2225 | const struct qstr *name, | |
2226 | unsigned *seqp) | |
2227 | { | |
2228 | u64 hashlen = name->hash_len; | |
2229 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); | |
2230 | struct hlist_bl_node *node; | |
2231 | struct dentry *dentry; | |
2232 | ||
2233 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | |
2234 | int tlen; | |
2235 | const char *tname; | |
2236 | unsigned seq; | |
2237 | ||
2238 | seqretry: | |
2239 | seq = raw_seqcount_begin(&dentry->d_seq); | |
2240 | if (dentry->d_parent != parent) | |
2241 | continue; | |
2242 | if (d_unhashed(dentry)) | |
2243 | continue; | |
2244 | if (dentry->d_name.hash != hashlen_hash(hashlen)) | |
2245 | continue; | |
2246 | tlen = dentry->d_name.len; | |
2247 | tname = dentry->d_name.name; | |
2248 | /* we want a consistent (name,len) pair */ | |
2249 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2250 | cpu_relax(); | |
2251 | goto seqretry; | |
2252 | } | |
2253 | if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0) | |
2254 | continue; | |
2255 | *seqp = seq; | |
2256 | return dentry; | |
2257 | } | |
2258 | return NULL; | |
2259 | } | |
2260 | ||
31e6b01f NP |
2261 | /** |
2262 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2263 | * @parent: parent dentry | |
2264 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2265 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2266 | * Returns: dentry, or NULL |
2267 | * | |
2268 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2269 | * resolution (store-free path walking) design described in | |
2270 | * Documentation/filesystems/path-lookup.txt. | |
2271 | * | |
2272 | * This is not to be used outside core vfs. | |
2273 | * | |
2274 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2275 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2276 | * without taking d_lock and checking d_seq sequence count against @seq | |
2277 | * returned here. | |
2278 | * | |
15570086 | 2279 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2280 | * function. |
2281 | * | |
2282 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2283 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2284 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2285 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2286 | * |
2287 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2288 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2289 | */ |
8966be90 LT |
2290 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2291 | const struct qstr *name, | |
da53be12 | 2292 | unsigned *seqp) |
31e6b01f | 2293 | { |
26fe5750 | 2294 | u64 hashlen = name->hash_len; |
31e6b01f | 2295 | const unsigned char *str = name->name; |
8387ff25 | 2296 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2297 | struct hlist_bl_node *node; |
31e6b01f NP |
2298 | struct dentry *dentry; |
2299 | ||
2300 | /* | |
2301 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2302 | * required to prevent single threaded performance regressions | |
2303 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2304 | * Keep the two functions in sync. | |
2305 | */ | |
2306 | ||
ae2a8236 LT |
2307 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) |
2308 | return __d_lookup_rcu_op_compare(parent, name, seqp); | |
2309 | ||
31e6b01f NP |
2310 | /* |
2311 | * The hash list is protected using RCU. | |
2312 | * | |
2313 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2314 | * races with d_move(). | |
2315 | * | |
2316 | * It is possible that concurrent renames can mess up our list | |
2317 | * walk here and result in missing our dentry, resulting in the | |
2318 | * false-negative result. d_lookup() protects against concurrent | |
2319 | * renames using rename_lock seqlock. | |
2320 | * | |
b0a4bb83 | 2321 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2322 | */ |
b07ad996 | 2323 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2324 | unsigned seq; |
31e6b01f | 2325 | |
12f8ad4b LT |
2326 | /* |
2327 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2328 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2329 | * |
2330 | * The caller must perform a seqcount check in order | |
da53be12 | 2331 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2332 | * |
2333 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2334 | * we don't wait for the sequence count to stabilize if it | |
2335 | * is in the middle of a sequence change. If we do the slow | |
2336 | * dentry compare, we will do seqretries until it is stable, | |
2337 | * and if we end up with a successful lookup, we actually | |
2338 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2339 | * |
2340 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2341 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2342 | */ |
2343 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2344 | if (dentry->d_parent != parent) |
2345 | continue; | |
2e321806 LT |
2346 | if (d_unhashed(dentry)) |
2347 | continue; | |
ae2a8236 LT |
2348 | if (dentry->d_name.hash_len != hashlen) |
2349 | continue; | |
2350 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2351 | continue; | |
da53be12 | 2352 | *seqp = seq; |
d4c91a8f | 2353 | return dentry; |
31e6b01f NP |
2354 | } |
2355 | return NULL; | |
2356 | } | |
2357 | ||
1da177e4 LT |
2358 | /** |
2359 | * d_lookup - search for a dentry | |
2360 | * @parent: parent dentry | |
2361 | * @name: qstr of name we wish to find | |
b04f784e | 2362 | * Returns: dentry, or NULL |
1da177e4 | 2363 | * |
b04f784e NP |
2364 | * d_lookup searches the children of the parent dentry for the name in |
2365 | * question. If the dentry is found its reference count is incremented and the | |
2366 | * dentry is returned. The caller must use dput to free the entry when it has | |
2367 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2368 | */ |
da2d8455 | 2369 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2370 | { |
31e6b01f | 2371 | struct dentry *dentry; |
949854d0 | 2372 | unsigned seq; |
1da177e4 | 2373 | |
b8314f93 DY |
2374 | do { |
2375 | seq = read_seqbegin(&rename_lock); | |
2376 | dentry = __d_lookup(parent, name); | |
2377 | if (dentry) | |
1da177e4 LT |
2378 | break; |
2379 | } while (read_seqretry(&rename_lock, seq)); | |
2380 | return dentry; | |
2381 | } | |
ec4f8605 | 2382 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2383 | |
31e6b01f | 2384 | /** |
b04f784e NP |
2385 | * __d_lookup - search for a dentry (racy) |
2386 | * @parent: parent dentry | |
2387 | * @name: qstr of name we wish to find | |
2388 | * Returns: dentry, or NULL | |
2389 | * | |
2390 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2391 | * false-negative result due to unrelated rename activity. | |
2392 | * | |
2393 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2394 | * however it must be used carefully, eg. with a following d_lookup in | |
2395 | * the case of failure. | |
2396 | * | |
2397 | * __d_lookup callers must be commented. | |
2398 | */ | |
a713ca2a | 2399 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2400 | { |
1da177e4 | 2401 | unsigned int hash = name->hash; |
8387ff25 | 2402 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2403 | struct hlist_bl_node *node; |
31e6b01f | 2404 | struct dentry *found = NULL; |
665a7583 | 2405 | struct dentry *dentry; |
1da177e4 | 2406 | |
31e6b01f NP |
2407 | /* |
2408 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2409 | * required to prevent single threaded performance regressions | |
2410 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2411 | * Keep the two functions in sync. | |
2412 | */ | |
2413 | ||
b04f784e NP |
2414 | /* |
2415 | * The hash list is protected using RCU. | |
2416 | * | |
2417 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2418 | * with d_move(). | |
2419 | * | |
2420 | * It is possible that concurrent renames can mess up our list | |
2421 | * walk here and result in missing our dentry, resulting in the | |
2422 | * false-negative result. d_lookup() protects against concurrent | |
2423 | * renames using rename_lock seqlock. | |
2424 | * | |
b0a4bb83 | 2425 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2426 | */ |
1da177e4 LT |
2427 | rcu_read_lock(); |
2428 | ||
b07ad996 | 2429 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2430 | |
1da177e4 LT |
2431 | if (dentry->d_name.hash != hash) |
2432 | continue; | |
1da177e4 LT |
2433 | |
2434 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2435 | if (dentry->d_parent != parent) |
2436 | goto next; | |
d0185c08 LT |
2437 | if (d_unhashed(dentry)) |
2438 | goto next; | |
2439 | ||
d4c91a8f AV |
2440 | if (!d_same_name(dentry, parent, name)) |
2441 | goto next; | |
1da177e4 | 2442 | |
98474236 | 2443 | dentry->d_lockref.count++; |
d0185c08 | 2444 | found = dentry; |
1da177e4 LT |
2445 | spin_unlock(&dentry->d_lock); |
2446 | break; | |
2447 | next: | |
2448 | spin_unlock(&dentry->d_lock); | |
2449 | } | |
2450 | rcu_read_unlock(); | |
2451 | ||
2452 | return found; | |
2453 | } | |
2454 | ||
3e7e241f EB |
2455 | /** |
2456 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2457 | * @dir: Directory to search in | |
2458 | * @name: qstr of name we wish to find | |
2459 | * | |
4f522a24 | 2460 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2461 | */ |
2462 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2463 | { | |
3e7e241f EB |
2464 | /* |
2465 | * Check for a fs-specific hash function. Note that we must | |
2466 | * calculate the standard hash first, as the d_op->d_hash() | |
2467 | * routine may choose to leave the hash value unchanged. | |
2468 | */ | |
8387ff25 | 2469 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2470 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2471 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2472 | if (unlikely(err < 0)) |
2473 | return ERR_PTR(err); | |
3e7e241f | 2474 | } |
4f522a24 | 2475 | return d_lookup(dir, name); |
3e7e241f | 2476 | } |
4f522a24 | 2477 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2478 | |
1da177e4 LT |
2479 | /* |
2480 | * When a file is deleted, we have two options: | |
2481 | * - turn this dentry into a negative dentry | |
2482 | * - unhash this dentry and free it. | |
2483 | * | |
2484 | * Usually, we want to just turn this into | |
2485 | * a negative dentry, but if anybody else is | |
2486 | * currently using the dentry or the inode | |
2487 | * we can't do that and we fall back on removing | |
2488 | * it from the hash queues and waiting for | |
2489 | * it to be deleted later when it has no users | |
2490 | */ | |
2491 | ||
2492 | /** | |
2493 | * d_delete - delete a dentry | |
2494 | * @dentry: The dentry to delete | |
2495 | * | |
2496 | * Turn the dentry into a negative dentry if possible, otherwise | |
2497 | * remove it from the hash queues so it can be deleted later | |
2498 | */ | |
2499 | ||
2500 | void d_delete(struct dentry * dentry) | |
2501 | { | |
c19457f0 | 2502 | struct inode *inode = dentry->d_inode; |
c19457f0 AV |
2503 | |
2504 | spin_lock(&inode->i_lock); | |
2505 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2506 | /* |
2507 | * Are we the only user? | |
2508 | */ | |
98474236 | 2509 | if (dentry->d_lockref.count == 1) { |
13e3c5e5 | 2510 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2511 | dentry_unlink_inode(dentry); |
c19457f0 | 2512 | } else { |
1da177e4 | 2513 | __d_drop(dentry); |
c19457f0 AV |
2514 | spin_unlock(&dentry->d_lock); |
2515 | spin_unlock(&inode->i_lock); | |
2516 | } | |
1da177e4 | 2517 | } |
ec4f8605 | 2518 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2519 | |
15d3c589 | 2520 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2521 | { |
15d3c589 | 2522 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
61647823 | 2523 | |
1879fd6a | 2524 | hlist_bl_lock(b); |
b07ad996 | 2525 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2526 | hlist_bl_unlock(b); |
1da177e4 LT |
2527 | } |
2528 | ||
2529 | /** | |
2530 | * d_rehash - add an entry back to the hash | |
2531 | * @entry: dentry to add to the hash | |
2532 | * | |
2533 | * Adds a dentry to the hash according to its name. | |
2534 | */ | |
2535 | ||
2536 | void d_rehash(struct dentry * entry) | |
2537 | { | |
1da177e4 | 2538 | spin_lock(&entry->d_lock); |
15d3c589 | 2539 | __d_rehash(entry); |
1da177e4 | 2540 | spin_unlock(&entry->d_lock); |
1da177e4 | 2541 | } |
ec4f8605 | 2542 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2543 | |
84e710da AV |
2544 | static inline unsigned start_dir_add(struct inode *dir) |
2545 | { | |
93f6d4e1 | 2546 | preempt_disable_nested(); |
84e710da AV |
2547 | for (;;) { |
2548 | unsigned n = dir->i_dir_seq; | |
2549 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2550 | return n; | |
2551 | cpu_relax(); | |
2552 | } | |
2553 | } | |
2554 | ||
50417d22 SAS |
2555 | static inline void end_dir_add(struct inode *dir, unsigned int n, |
2556 | wait_queue_head_t *d_wait) | |
84e710da AV |
2557 | { |
2558 | smp_store_release(&dir->i_dir_seq, n + 2); | |
93f6d4e1 | 2559 | preempt_enable_nested(); |
50417d22 | 2560 | wake_up_all(d_wait); |
84e710da AV |
2561 | } |
2562 | ||
d9171b93 AV |
2563 | static void d_wait_lookup(struct dentry *dentry) |
2564 | { | |
2565 | if (d_in_lookup(dentry)) { | |
2566 | DECLARE_WAITQUEUE(wait, current); | |
2567 | add_wait_queue(dentry->d_wait, &wait); | |
2568 | do { | |
2569 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2570 | spin_unlock(&dentry->d_lock); | |
2571 | schedule(); | |
2572 | spin_lock(&dentry->d_lock); | |
2573 | } while (d_in_lookup(dentry)); | |
2574 | } | |
2575 | } | |
2576 | ||
94bdd655 | 2577 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2578 | const struct qstr *name, |
2579 | wait_queue_head_t *wq) | |
94bdd655 | 2580 | { |
94bdd655 | 2581 | unsigned int hash = name->hash; |
94bdd655 AV |
2582 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2583 | struct hlist_bl_node *node; | |
2584 | struct dentry *new = d_alloc(parent, name); | |
2585 | struct dentry *dentry; | |
2586 | unsigned seq, r_seq, d_seq; | |
2587 | ||
2588 | if (unlikely(!new)) | |
2589 | return ERR_PTR(-ENOMEM); | |
2590 | ||
2591 | retry: | |
2592 | rcu_read_lock(); | |
015555fd | 2593 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
94bdd655 AV |
2594 | r_seq = read_seqbegin(&rename_lock); |
2595 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2596 | if (unlikely(dentry)) { | |
2597 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2598 | rcu_read_unlock(); | |
2599 | goto retry; | |
2600 | } | |
2601 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2602 | rcu_read_unlock(); | |
2603 | dput(dentry); | |
2604 | goto retry; | |
2605 | } | |
2606 | rcu_read_unlock(); | |
2607 | dput(new); | |
2608 | return dentry; | |
2609 | } | |
2610 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2611 | rcu_read_unlock(); | |
2612 | goto retry; | |
2613 | } | |
015555fd WD |
2614 | |
2615 | if (unlikely(seq & 1)) { | |
2616 | rcu_read_unlock(); | |
2617 | goto retry; | |
2618 | } | |
2619 | ||
94bdd655 | 2620 | hlist_bl_lock(b); |
8cc07c80 | 2621 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
94bdd655 AV |
2622 | hlist_bl_unlock(b); |
2623 | rcu_read_unlock(); | |
2624 | goto retry; | |
2625 | } | |
94bdd655 AV |
2626 | /* |
2627 | * No changes for the parent since the beginning of d_lookup(). | |
2628 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2629 | * any potential in-lookup matches are going to stay here until | |
2630 | * we unlock the chain. All fields are stable in everything | |
2631 | * we encounter. | |
2632 | */ | |
2633 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2634 | if (dentry->d_name.hash != hash) | |
2635 | continue; | |
2636 | if (dentry->d_parent != parent) | |
2637 | continue; | |
d4c91a8f AV |
2638 | if (!d_same_name(dentry, parent, name)) |
2639 | continue; | |
94bdd655 | 2640 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2641 | /* now we can try to grab a reference */ |
2642 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2643 | rcu_read_unlock(); | |
2644 | goto retry; | |
2645 | } | |
2646 | ||
2647 | rcu_read_unlock(); | |
2648 | /* | |
2649 | * somebody is likely to be still doing lookup for it; | |
2650 | * wait for them to finish | |
2651 | */ | |
d9171b93 AV |
2652 | spin_lock(&dentry->d_lock); |
2653 | d_wait_lookup(dentry); | |
2654 | /* | |
2655 | * it's not in-lookup anymore; in principle we should repeat | |
2656 | * everything from dcache lookup, but it's likely to be what | |
2657 | * d_lookup() would've found anyway. If it is, just return it; | |
2658 | * otherwise we really have to repeat the whole thing. | |
2659 | */ | |
2660 | if (unlikely(dentry->d_name.hash != hash)) | |
2661 | goto mismatch; | |
2662 | if (unlikely(dentry->d_parent != parent)) | |
2663 | goto mismatch; | |
2664 | if (unlikely(d_unhashed(dentry))) | |
2665 | goto mismatch; | |
d4c91a8f AV |
2666 | if (unlikely(!d_same_name(dentry, parent, name))) |
2667 | goto mismatch; | |
d9171b93 AV |
2668 | /* OK, it *is* a hashed match; return it */ |
2669 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2670 | dput(new); |
2671 | return dentry; | |
2672 | } | |
e7d6ef97 | 2673 | rcu_read_unlock(); |
94bdd655 AV |
2674 | /* we can't take ->d_lock here; it's OK, though. */ |
2675 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2676 | new->d_wait = wq; |
94bdd655 AV |
2677 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2678 | hlist_bl_unlock(b); | |
2679 | return new; | |
d9171b93 AV |
2680 | mismatch: |
2681 | spin_unlock(&dentry->d_lock); | |
2682 | dput(dentry); | |
2683 | goto retry; | |
94bdd655 AV |
2684 | } |
2685 | EXPORT_SYMBOL(d_alloc_parallel); | |
2686 | ||
45f78b0a SAS |
2687 | /* |
2688 | * - Unhash the dentry | |
2689 | * - Retrieve and clear the waitqueue head in dentry | |
2690 | * - Return the waitqueue head | |
2691 | */ | |
2692 | static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry) | |
85c7f810 | 2693 | { |
45f78b0a SAS |
2694 | wait_queue_head_t *d_wait; |
2695 | struct hlist_bl_head *b; | |
2696 | ||
2697 | lockdep_assert_held(&dentry->d_lock); | |
2698 | ||
2699 | b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash); | |
94bdd655 | 2700 | hlist_bl_lock(b); |
85c7f810 | 2701 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2702 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
45f78b0a | 2703 | d_wait = dentry->d_wait; |
d9171b93 | 2704 | dentry->d_wait = NULL; |
94bdd655 AV |
2705 | hlist_bl_unlock(b); |
2706 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2707 | INIT_LIST_HEAD(&dentry->d_lru); |
45f78b0a SAS |
2708 | return d_wait; |
2709 | } | |
2710 | ||
2711 | void __d_lookup_unhash_wake(struct dentry *dentry) | |
2712 | { | |
2713 | spin_lock(&dentry->d_lock); | |
2714 | wake_up_all(__d_lookup_unhash(dentry)); | |
2715 | spin_unlock(&dentry->d_lock); | |
85c7f810 | 2716 | } |
45f78b0a | 2717 | EXPORT_SYMBOL(__d_lookup_unhash_wake); |
ed782b5a AV |
2718 | |
2719 | /* inode->i_lock held if inode is non-NULL */ | |
2720 | ||
2721 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2722 | { | |
45f78b0a | 2723 | wait_queue_head_t *d_wait; |
84e710da AV |
2724 | struct inode *dir = NULL; |
2725 | unsigned n; | |
0568d705 | 2726 | spin_lock(&dentry->d_lock); |
84e710da AV |
2727 | if (unlikely(d_in_lookup(dentry))) { |
2728 | dir = dentry->d_parent->d_inode; | |
2729 | n = start_dir_add(dir); | |
45f78b0a | 2730 | d_wait = __d_lookup_unhash(dentry); |
84e710da | 2731 | } |
ed782b5a | 2732 | if (inode) { |
0568d705 AV |
2733 | unsigned add_flags = d_flags_for_inode(inode); |
2734 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2735 | raw_write_seqcount_begin(&dentry->d_seq); | |
2736 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2737 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2738 | fsnotify_update_flags(dentry); |
ed782b5a | 2739 | } |
15d3c589 | 2740 | __d_rehash(dentry); |
84e710da | 2741 | if (dir) |
50417d22 | 2742 | end_dir_add(dir, n, d_wait); |
0568d705 AV |
2743 | spin_unlock(&dentry->d_lock); |
2744 | if (inode) | |
2745 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2746 | } |
2747 | ||
34d0d19d AV |
2748 | /** |
2749 | * d_add - add dentry to hash queues | |
2750 | * @entry: dentry to add | |
2751 | * @inode: The inode to attach to this dentry | |
2752 | * | |
2753 | * This adds the entry to the hash queues and initializes @inode. | |
2754 | * The entry was actually filled in earlier during d_alloc(). | |
2755 | */ | |
2756 | ||
2757 | void d_add(struct dentry *entry, struct inode *inode) | |
2758 | { | |
b9680917 AV |
2759 | if (inode) { |
2760 | security_d_instantiate(entry, inode); | |
ed782b5a | 2761 | spin_lock(&inode->i_lock); |
b9680917 | 2762 | } |
ed782b5a | 2763 | __d_add(entry, inode); |
34d0d19d AV |
2764 | } |
2765 | EXPORT_SYMBOL(d_add); | |
2766 | ||
668d0cd5 AV |
2767 | /** |
2768 | * d_exact_alias - find and hash an exact unhashed alias | |
2769 | * @entry: dentry to add | |
2770 | * @inode: The inode to go with this dentry | |
2771 | * | |
2772 | * If an unhashed dentry with the same name/parent and desired | |
2773 | * inode already exists, hash and return it. Otherwise, return | |
2774 | * NULL. | |
2775 | * | |
2776 | * Parent directory should be locked. | |
2777 | */ | |
2778 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2779 | { | |
2780 | struct dentry *alias; | |
668d0cd5 AV |
2781 | unsigned int hash = entry->d_name.hash; |
2782 | ||
2783 | spin_lock(&inode->i_lock); | |
2784 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2785 | /* | |
2786 | * Don't need alias->d_lock here, because aliases with | |
2787 | * d_parent == entry->d_parent are not subject to name or | |
2788 | * parent changes, because the parent inode i_mutex is held. | |
2789 | */ | |
2790 | if (alias->d_name.hash != hash) | |
2791 | continue; | |
2792 | if (alias->d_parent != entry->d_parent) | |
2793 | continue; | |
d4c91a8f | 2794 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2795 | continue; |
2796 | spin_lock(&alias->d_lock); | |
2797 | if (!d_unhashed(alias)) { | |
2798 | spin_unlock(&alias->d_lock); | |
2799 | alias = NULL; | |
2800 | } else { | |
2801 | __dget_dlock(alias); | |
15d3c589 | 2802 | __d_rehash(alias); |
668d0cd5 AV |
2803 | spin_unlock(&alias->d_lock); |
2804 | } | |
2805 | spin_unlock(&inode->i_lock); | |
2806 | return alias; | |
2807 | } | |
2808 | spin_unlock(&inode->i_lock); | |
2809 | return NULL; | |
2810 | } | |
2811 | EXPORT_SYMBOL(d_exact_alias); | |
2812 | ||
8d85b484 | 2813 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2814 | { |
8d85b484 AV |
2815 | if (unlikely(dname_external(target))) { |
2816 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2817 | /* |
2818 | * Both external: swap the pointers | |
2819 | */ | |
9a8d5bb4 | 2820 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2821 | } else { |
2822 | /* | |
2823 | * dentry:internal, target:external. Steal target's | |
2824 | * storage and make target internal. | |
2825 | */ | |
321bcf92 BF |
2826 | memcpy(target->d_iname, dentry->d_name.name, |
2827 | dentry->d_name.len + 1); | |
1da177e4 LT |
2828 | dentry->d_name.name = target->d_name.name; |
2829 | target->d_name.name = target->d_iname; | |
2830 | } | |
2831 | } else { | |
8d85b484 | 2832 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2833 | /* |
2834 | * dentry:external, target:internal. Give dentry's | |
2835 | * storage to target and make dentry internal | |
2836 | */ | |
2837 | memcpy(dentry->d_iname, target->d_name.name, | |
2838 | target->d_name.len + 1); | |
2839 | target->d_name.name = dentry->d_name.name; | |
2840 | dentry->d_name.name = dentry->d_iname; | |
2841 | } else { | |
2842 | /* | |
da1ce067 | 2843 | * Both are internal. |
1da177e4 | 2844 | */ |
da1ce067 MS |
2845 | unsigned int i; |
2846 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2847 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2848 | swap(((long *) &dentry->d_iname)[i], | |
2849 | ((long *) &target->d_iname)[i]); | |
2850 | } | |
1da177e4 LT |
2851 | } |
2852 | } | |
a28ddb87 | 2853 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2854 | } |
2855 | ||
8d85b484 AV |
2856 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2857 | { | |
2858 | struct external_name *old_name = NULL; | |
2859 | if (unlikely(dname_external(dentry))) | |
2860 | old_name = external_name(dentry); | |
2861 | if (unlikely(dname_external(target))) { | |
2862 | atomic_inc(&external_name(target)->u.count); | |
2863 | dentry->d_name = target->d_name; | |
2864 | } else { | |
2865 | memcpy(dentry->d_iname, target->d_name.name, | |
2866 | target->d_name.len + 1); | |
2867 | dentry->d_name.name = dentry->d_iname; | |
2868 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2869 | } | |
2870 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2e03b4bc | 2871 | kfree_rcu(old_name, u.head); |
8d85b484 AV |
2872 | } |
2873 | ||
9eaef27b | 2874 | /* |
18367501 | 2875 | * __d_move - move a dentry |
1da177e4 LT |
2876 | * @dentry: entry to move |
2877 | * @target: new dentry | |
da1ce067 | 2878 | * @exchange: exchange the two dentries |
1da177e4 LT |
2879 | * |
2880 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2881 | * dcache entries should not be moved in this way. Caller must hold |
2882 | * rename_lock, the i_mutex of the source and target directories, | |
2883 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2884 | */ |
da1ce067 MS |
2885 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2886 | bool exchange) | |
1da177e4 | 2887 | { |
42177007 | 2888 | struct dentry *old_parent, *p; |
45f78b0a | 2889 | wait_queue_head_t *d_wait; |
84e710da AV |
2890 | struct inode *dir = NULL; |
2891 | unsigned n; | |
1da177e4 | 2892 | |
42177007 AV |
2893 | WARN_ON(!dentry->d_inode); |
2894 | if (WARN_ON(dentry == target)) | |
2895 | return; | |
2896 | ||
2fd6b7f5 | 2897 | BUG_ON(d_ancestor(target, dentry)); |
42177007 AV |
2898 | old_parent = dentry->d_parent; |
2899 | p = d_ancestor(old_parent, target); | |
2900 | if (IS_ROOT(dentry)) { | |
2901 | BUG_ON(p); | |
2902 | spin_lock(&target->d_parent->d_lock); | |
2903 | } else if (!p) { | |
2904 | /* target is not a descendent of dentry->d_parent */ | |
2905 | spin_lock(&target->d_parent->d_lock); | |
2906 | spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED); | |
2907 | } else { | |
2908 | BUG_ON(p == dentry); | |
2909 | spin_lock(&old_parent->d_lock); | |
2910 | if (p != target) | |
2911 | spin_lock_nested(&target->d_parent->d_lock, | |
2912 | DENTRY_D_LOCK_NESTED); | |
2913 | } | |
2914 | spin_lock_nested(&dentry->d_lock, 2); | |
2915 | spin_lock_nested(&target->d_lock, 3); | |
2fd6b7f5 | 2916 | |
84e710da AV |
2917 | if (unlikely(d_in_lookup(target))) { |
2918 | dir = target->d_parent->d_inode; | |
2919 | n = start_dir_add(dir); | |
45f78b0a | 2920 | d_wait = __d_lookup_unhash(target); |
84e710da | 2921 | } |
1da177e4 | 2922 | |
31e6b01f | 2923 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2924 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2925 | |
15d3c589 | 2926 | /* unhash both */ |
0632a9ac AV |
2927 | if (!d_unhashed(dentry)) |
2928 | ___d_drop(dentry); | |
2929 | if (!d_unhashed(target)) | |
2930 | ___d_drop(target); | |
1da177e4 | 2931 | |
076515fc AV |
2932 | /* ... and switch them in the tree */ |
2933 | dentry->d_parent = target->d_parent; | |
2934 | if (!exchange) { | |
8d85b484 | 2935 | copy_name(dentry, target); |
61647823 | 2936 | target->d_hash.pprev = NULL; |
076515fc | 2937 | dentry->d_parent->d_lockref.count++; |
5467a68c | 2938 | if (dentry != old_parent) /* wasn't IS_ROOT */ |
076515fc | 2939 | WARN_ON(!--old_parent->d_lockref.count); |
1da177e4 | 2940 | } else { |
076515fc AV |
2941 | target->d_parent = old_parent; |
2942 | swap_names(dentry, target); | |
da549bdd AV |
2943 | if (!hlist_unhashed(&target->d_sib)) |
2944 | __hlist_del(&target->d_sib); | |
2945 | hlist_add_head(&target->d_sib, &target->d_parent->d_children); | |
076515fc AV |
2946 | __d_rehash(target); |
2947 | fsnotify_update_flags(target); | |
1da177e4 | 2948 | } |
da549bdd AV |
2949 | if (!hlist_unhashed(&dentry->d_sib)) |
2950 | __hlist_del(&dentry->d_sib); | |
2951 | hlist_add_head(&dentry->d_sib, &dentry->d_parent->d_children); | |
076515fc AV |
2952 | __d_rehash(dentry); |
2953 | fsnotify_update_flags(dentry); | |
0bf3d5c1 | 2954 | fscrypt_handle_d_move(dentry); |
1da177e4 | 2955 | |
31e6b01f NP |
2956 | write_seqcount_end(&target->d_seq); |
2957 | write_seqcount_end(&dentry->d_seq); | |
2958 | ||
84e710da | 2959 | if (dir) |
50417d22 | 2960 | end_dir_add(dir, n, d_wait); |
076515fc AV |
2961 | |
2962 | if (dentry->d_parent != old_parent) | |
2963 | spin_unlock(&dentry->d_parent->d_lock); | |
2964 | if (dentry != old_parent) | |
2965 | spin_unlock(&old_parent->d_lock); | |
2966 | spin_unlock(&target->d_lock); | |
2967 | spin_unlock(&dentry->d_lock); | |
18367501 AV |
2968 | } |
2969 | ||
2970 | /* | |
2971 | * d_move - move a dentry | |
2972 | * @dentry: entry to move | |
2973 | * @target: new dentry | |
2974 | * | |
2975 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2976 | * dcache entries should not be moved in this way. See the locking |
2977 | * requirements for __d_move. | |
18367501 AV |
2978 | */ |
2979 | void d_move(struct dentry *dentry, struct dentry *target) | |
2980 | { | |
2981 | write_seqlock(&rename_lock); | |
da1ce067 | 2982 | __d_move(dentry, target, false); |
1da177e4 | 2983 | write_sequnlock(&rename_lock); |
9eaef27b | 2984 | } |
ec4f8605 | 2985 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2986 | |
da1ce067 MS |
2987 | /* |
2988 | * d_exchange - exchange two dentries | |
2989 | * @dentry1: first dentry | |
2990 | * @dentry2: second dentry | |
2991 | */ | |
2992 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2993 | { | |
2994 | write_seqlock(&rename_lock); | |
2995 | ||
2996 | WARN_ON(!dentry1->d_inode); | |
2997 | WARN_ON(!dentry2->d_inode); | |
2998 | WARN_ON(IS_ROOT(dentry1)); | |
2999 | WARN_ON(IS_ROOT(dentry2)); | |
3000 | ||
3001 | __d_move(dentry1, dentry2, true); | |
3002 | ||
3003 | write_sequnlock(&rename_lock); | |
3004 | } | |
3005 | ||
e2761a11 OH |
3006 | /** |
3007 | * d_ancestor - search for an ancestor | |
3008 | * @p1: ancestor dentry | |
3009 | * @p2: child dentry | |
3010 | * | |
3011 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
3012 | * an ancestor of p2, else NULL. | |
9eaef27b | 3013 | */ |
e2761a11 | 3014 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
3015 | { |
3016 | struct dentry *p; | |
3017 | ||
871c0067 | 3018 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 3019 | if (p->d_parent == p1) |
e2761a11 | 3020 | return p; |
9eaef27b | 3021 | } |
e2761a11 | 3022 | return NULL; |
9eaef27b TM |
3023 | } |
3024 | ||
3025 | /* | |
3026 | * This helper attempts to cope with remotely renamed directories | |
3027 | * | |
3028 | * It assumes that the caller is already holding | |
a03e283b | 3029 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
3030 | * |
3031 | * Note: If ever the locking in lock_rename() changes, then please | |
3032 | * remember to update this too... | |
9eaef27b | 3033 | */ |
b5ae6b15 | 3034 | static int __d_unalias(struct inode *inode, |
873feea0 | 3035 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 3036 | { |
9902af79 AV |
3037 | struct mutex *m1 = NULL; |
3038 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 3039 | int ret = -ESTALE; |
9eaef27b TM |
3040 | |
3041 | /* If alias and dentry share a parent, then no extra locks required */ | |
3042 | if (alias->d_parent == dentry->d_parent) | |
3043 | goto out_unalias; | |
3044 | ||
9eaef27b | 3045 | /* See lock_rename() */ |
9eaef27b TM |
3046 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
3047 | goto out_err; | |
3048 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 3049 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 3050 | goto out_err; |
9902af79 | 3051 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 3052 | out_unalias: |
8ed936b5 | 3053 | __d_move(alias, dentry, false); |
b5ae6b15 | 3054 | ret = 0; |
9eaef27b | 3055 | out_err: |
9eaef27b | 3056 | if (m2) |
9902af79 | 3057 | up_read(m2); |
9eaef27b TM |
3058 | if (m1) |
3059 | mutex_unlock(m1); | |
3060 | return ret; | |
3061 | } | |
3062 | ||
3f70bd51 BF |
3063 | /** |
3064 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
3065 | * @inode: the inode which may have a disconnected dentry | |
3066 | * @dentry: a negative dentry which we want to point to the inode. | |
3067 | * | |
da093a9b BF |
3068 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
3069 | * place of the given dentry and return it, else simply d_add the inode | |
3070 | * to the dentry and return NULL. | |
3f70bd51 | 3071 | * |
908790fa BF |
3072 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
3073 | * we should error out: directories can't have multiple aliases. | |
3074 | * | |
3f70bd51 BF |
3075 | * This is needed in the lookup routine of any filesystem that is exportable |
3076 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
3077 | * | |
3078 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
3079 | * is returned. This matches the expected return value of ->lookup. | |
3080 | * | |
3081 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
3082 | * In that case, we know that the inode will be a regular file, and also this | |
3083 | * will only occur during atomic_open. So we need to check for the dentry | |
3084 | * being already hashed only in the final case. | |
3085 | */ | |
3086 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3087 | { | |
3f70bd51 BF |
3088 | if (IS_ERR(inode)) |
3089 | return ERR_CAST(inode); | |
3090 | ||
770bfad8 DH |
3091 | BUG_ON(!d_unhashed(dentry)); |
3092 | ||
de689f5e | 3093 | if (!inode) |
b5ae6b15 | 3094 | goto out; |
de689f5e | 3095 | |
b9680917 | 3096 | security_d_instantiate(dentry, inode); |
873feea0 | 3097 | spin_lock(&inode->i_lock); |
9eaef27b | 3098 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
3099 | struct dentry *new = __d_find_any_alias(inode); |
3100 | if (unlikely(new)) { | |
a03e283b EB |
3101 | /* The reference to new ensures it remains an alias */ |
3102 | spin_unlock(&inode->i_lock); | |
18367501 | 3103 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3104 | if (unlikely(d_ancestor(new, dentry))) { |
3105 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3106 | dput(new); |
3107 | new = ERR_PTR(-ELOOP); | |
3108 | pr_warn_ratelimited( | |
3109 | "VFS: Lookup of '%s' in %s %s" | |
3110 | " would have caused loop\n", | |
3111 | dentry->d_name.name, | |
3112 | inode->i_sb->s_type->name, | |
3113 | inode->i_sb->s_id); | |
3114 | } else if (!IS_ROOT(new)) { | |
076515fc | 3115 | struct dentry *old_parent = dget(new->d_parent); |
b5ae6b15 | 3116 | int err = __d_unalias(inode, dentry, new); |
18367501 | 3117 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3118 | if (err) { |
3119 | dput(new); | |
3120 | new = ERR_PTR(err); | |
3121 | } | |
076515fc | 3122 | dput(old_parent); |
18367501 | 3123 | } else { |
b5ae6b15 AV |
3124 | __d_move(new, dentry, false); |
3125 | write_sequnlock(&rename_lock); | |
dd179946 | 3126 | } |
b5ae6b15 AV |
3127 | iput(inode); |
3128 | return new; | |
9eaef27b | 3129 | } |
770bfad8 | 3130 | } |
b5ae6b15 | 3131 | out: |
ed782b5a | 3132 | __d_add(dentry, inode); |
b5ae6b15 | 3133 | return NULL; |
770bfad8 | 3134 | } |
b5ae6b15 | 3135 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3136 | |
1da177e4 LT |
3137 | /* |
3138 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3139 | * | |
3140 | * Trivially implemented using the dcache structure | |
3141 | */ | |
3142 | ||
3143 | /** | |
3144 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3145 | * @new_dentry: new dentry | |
3146 | * @old_dentry: old dentry | |
3147 | * | |
a6e5787f YB |
3148 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3149 | * Returns false otherwise. | |
1da177e4 LT |
3150 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3151 | */ | |
3152 | ||
a6e5787f | 3153 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3154 | { |
a6e5787f | 3155 | bool result; |
949854d0 | 3156 | unsigned seq; |
1da177e4 | 3157 | |
e2761a11 | 3158 | if (new_dentry == old_dentry) |
a6e5787f | 3159 | return true; |
e2761a11 | 3160 | |
e2761a11 | 3161 | do { |
1da177e4 | 3162 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3163 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3164 | /* |
3165 | * Need rcu_readlock to protect against the d_parent trashing | |
3166 | * due to d_move | |
3167 | */ | |
3168 | rcu_read_lock(); | |
e2761a11 | 3169 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3170 | result = true; |
e2761a11 | 3171 | else |
a6e5787f | 3172 | result = false; |
949854d0 | 3173 | rcu_read_unlock(); |
1da177e4 | 3174 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3175 | |
3176 | return result; | |
3177 | } | |
e8f9e5b7 | 3178 | EXPORT_SYMBOL(is_subdir); |
1da177e4 | 3179 | |
db14fc3a | 3180 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3181 | { |
db14fc3a MS |
3182 | struct dentry *root = data; |
3183 | if (dentry != root) { | |
3184 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3185 | return D_WALK_SKIP; | |
1da177e4 | 3186 | |
01ddc4ed MS |
3187 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3188 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3189 | dentry->d_lockref.count--; | |
3190 | } | |
1da177e4 | 3191 | } |
db14fc3a MS |
3192 | return D_WALK_CONTINUE; |
3193 | } | |
58db63d0 | 3194 | |
db14fc3a MS |
3195 | void d_genocide(struct dentry *parent) |
3196 | { | |
3a8e3611 | 3197 | d_walk(parent, parent, d_genocide_kill); |
1da177e4 LT |
3198 | } |
3199 | ||
771eb4fe | 3200 | void d_mark_tmpfile(struct file *file, struct inode *inode) |
1da177e4 | 3201 | { |
863f144f MS |
3202 | struct dentry *dentry = file->f_path.dentry; |
3203 | ||
60545d0d | 3204 | BUG_ON(dentry->d_name.name != dentry->d_iname || |
946e51f2 | 3205 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3206 | !d_unlinked(dentry)); |
3207 | spin_lock(&dentry->d_parent->d_lock); | |
3208 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3209 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3210 | (unsigned long long)inode->i_ino); | |
3211 | spin_unlock(&dentry->d_lock); | |
3212 | spin_unlock(&dentry->d_parent->d_lock); | |
771eb4fe KO |
3213 | } |
3214 | EXPORT_SYMBOL(d_mark_tmpfile); | |
3215 | ||
3216 | void d_tmpfile(struct file *file, struct inode *inode) | |
3217 | { | |
3218 | struct dentry *dentry = file->f_path.dentry; | |
3219 | ||
3220 | inode_dec_link_count(inode); | |
3221 | d_mark_tmpfile(file, inode); | |
60545d0d | 3222 | d_instantiate(dentry, inode); |
1da177e4 | 3223 | } |
60545d0d | 3224 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3225 | |
3226 | static __initdata unsigned long dhash_entries; | |
3227 | static int __init set_dhash_entries(char *str) | |
3228 | { | |
3229 | if (!str) | |
3230 | return 0; | |
3231 | dhash_entries = simple_strtoul(str, &str, 0); | |
3232 | return 1; | |
3233 | } | |
3234 | __setup("dhash_entries=", set_dhash_entries); | |
3235 | ||
3236 | static void __init dcache_init_early(void) | |
3237 | { | |
1da177e4 LT |
3238 | /* If hashes are distributed across NUMA nodes, defer |
3239 | * hash allocation until vmalloc space is available. | |
3240 | */ | |
3241 | if (hashdist) | |
3242 | return; | |
3243 | ||
3244 | dentry_hashtable = | |
3245 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3246 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3247 | dhash_entries, |
3248 | 13, | |
3d375d78 | 3249 | HASH_EARLY | HASH_ZERO, |
1da177e4 | 3250 | &d_hash_shift, |
b35d786b | 3251 | NULL, |
31fe62b9 | 3252 | 0, |
1da177e4 | 3253 | 0); |
854d3e63 | 3254 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3255 | } |
3256 | ||
74bf17cf | 3257 | static void __init dcache_init(void) |
1da177e4 | 3258 | { |
3d375d78 | 3259 | /* |
1da177e4 LT |
3260 | * A constructor could be added for stable state like the lists, |
3261 | * but it is probably not worth it because of the cache nature | |
3d375d78 | 3262 | * of the dcache. |
1da177e4 | 3263 | */ |
80344266 DW |
3264 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, |
3265 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, | |
3266 | d_iname); | |
1da177e4 LT |
3267 | |
3268 | /* Hash may have been set up in dcache_init_early */ | |
3269 | if (!hashdist) | |
3270 | return; | |
3271 | ||
3272 | dentry_hashtable = | |
3273 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3274 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3275 | dhash_entries, |
3276 | 13, | |
3d375d78 | 3277 | HASH_ZERO, |
1da177e4 | 3278 | &d_hash_shift, |
b35d786b | 3279 | NULL, |
31fe62b9 | 3280 | 0, |
1da177e4 | 3281 | 0); |
854d3e63 | 3282 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3283 | } |
3284 | ||
3285 | /* SLAB cache for __getname() consumers */ | |
68279f9c | 3286 | struct kmem_cache *names_cachep __ro_after_init; |
ec4f8605 | 3287 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3288 | |
1da177e4 LT |
3289 | void __init vfs_caches_init_early(void) |
3290 | { | |
6916363f SAS |
3291 | int i; |
3292 | ||
3293 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3294 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3295 | ||
1da177e4 LT |
3296 | dcache_init_early(); |
3297 | inode_init_early(); | |
3298 | } | |
3299 | ||
4248b0da | 3300 | void __init vfs_caches_init(void) |
1da177e4 | 3301 | { |
6a9b8820 DW |
3302 | names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0, |
3303 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL); | |
1da177e4 | 3304 | |
74bf17cf DC |
3305 | dcache_init(); |
3306 | inode_init(); | |
4248b0da MG |
3307 | files_init(); |
3308 | files_maxfiles_init(); | |
74bf17cf | 3309 | mnt_init(); |
1da177e4 LT |
3310 | bdev_cache_init(); |
3311 | chrdev_init(); | |
3312 | } |