Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
7a5cf791 | 17 | #include <linux/ratelimit.h> |
1da177e4 LT |
18 | #include <linux/string.h> |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/security.h> |
28 | #include <linux/seqlock.h> | |
1da177e4 | 29 | #include <linux/bootmem.h> |
ceb5bdc2 NP |
30 | #include <linux/bit_spinlock.h> |
31 | #include <linux/rculist_bl.h> | |
f6041567 | 32 | #include <linux/list_lru.h> |
07f3f05c | 33 | #include "internal.h" |
b2dba1af | 34 | #include "mount.h" |
1da177e4 | 35 | |
789680d1 NP |
36 | /* |
37 | * Usage: | |
873feea0 | 38 | * dcache->d_inode->i_lock protects: |
946e51f2 | 39 | * - i_dentry, d_u.d_alias, d_inode of aliases |
ceb5bdc2 NP |
40 | * dcache_hash_bucket lock protects: |
41 | * - the dcache hash table | |
f1ee6162 N |
42 | * s_roots bl list spinlock protects: |
43 | * - the s_roots list (see __d_drop) | |
19156840 | 44 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
45 | * - the dcache lru lists and counters |
46 | * d_lock protects: | |
47 | * - d_flags | |
48 | * - d_name | |
49 | * - d_lru | |
b7ab39f6 | 50 | * - d_count |
da502956 | 51 | * - d_unhashed() |
2fd6b7f5 NP |
52 | * - d_parent and d_subdirs |
53 | * - childrens' d_child and d_parent | |
946e51f2 | 54 | * - d_u.d_alias, d_inode |
789680d1 NP |
55 | * |
56 | * Ordering: | |
873feea0 | 57 | * dentry->d_inode->i_lock |
b5c84bf6 | 58 | * dentry->d_lock |
19156840 | 59 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 | 60 | * dcache_hash_bucket lock |
f1ee6162 | 61 | * s_roots lock |
789680d1 | 62 | * |
da502956 NP |
63 | * If there is an ancestor relationship: |
64 | * dentry->d_parent->...->d_parent->d_lock | |
65 | * ... | |
66 | * dentry->d_parent->d_lock | |
67 | * dentry->d_lock | |
68 | * | |
69 | * If no ancestor relationship: | |
789680d1 NP |
70 | * if (dentry1 < dentry2) |
71 | * dentry1->d_lock | |
72 | * dentry2->d_lock | |
73 | */ | |
fa3536cc | 74 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
75 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
76 | ||
74c3cbe3 | 77 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 78 | |
949854d0 | 79 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 80 | |
e18b890b | 81 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 82 | |
cdf01226 DH |
83 | const struct qstr empty_name = QSTR_INIT("", 0); |
84 | EXPORT_SYMBOL(empty_name); | |
85 | const struct qstr slash_name = QSTR_INIT("/", 1); | |
86 | EXPORT_SYMBOL(slash_name); | |
87 | ||
1da177e4 LT |
88 | /* |
89 | * This is the single most critical data structure when it comes | |
90 | * to the dcache: the hashtable for lookups. Somebody should try | |
91 | * to make this good - I've just made it work. | |
92 | * | |
93 | * This hash-function tries to avoid losing too many bits of hash | |
94 | * information, yet avoid using a prime hash-size or similar. | |
95 | */ | |
1da177e4 | 96 | |
fa3536cc | 97 | static unsigned int d_hash_shift __read_mostly; |
ceb5bdc2 | 98 | |
b07ad996 | 99 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 100 | |
8387ff25 | 101 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 102 | { |
854d3e63 | 103 | return dentry_hashtable + (hash >> d_hash_shift); |
ceb5bdc2 NP |
104 | } |
105 | ||
94bdd655 AV |
106 | #define IN_LOOKUP_SHIFT 10 |
107 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
108 | ||
109 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
110 | unsigned int hash) | |
111 | { | |
112 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
113 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
114 | } | |
115 | ||
116 | ||
1da177e4 LT |
117 | /* Statistics gathering. */ |
118 | struct dentry_stat_t dentry_stat = { | |
119 | .age_limit = 45, | |
120 | }; | |
121 | ||
3942c07c | 122 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 123 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
124 | |
125 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
126 | |
127 | /* | |
128 | * Here we resort to our own counters instead of using generic per-cpu counters | |
129 | * for consistency with what the vfs inode code does. We are expected to harvest | |
130 | * better code and performance by having our own specialized counters. | |
131 | * | |
132 | * Please note that the loop is done over all possible CPUs, not over all online | |
133 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
134 | * on and off. If one of them goes off, we will just keep their counters. | |
135 | * | |
136 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
137 | * please update all vfs counters to match. | |
138 | */ | |
3942c07c | 139 | static long get_nr_dentry(void) |
3e880fb5 NP |
140 | { |
141 | int i; | |
3942c07c | 142 | long sum = 0; |
3e880fb5 NP |
143 | for_each_possible_cpu(i) |
144 | sum += per_cpu(nr_dentry, i); | |
145 | return sum < 0 ? 0 : sum; | |
146 | } | |
147 | ||
62d36c77 DC |
148 | static long get_nr_dentry_unused(void) |
149 | { | |
150 | int i; | |
151 | long sum = 0; | |
152 | for_each_possible_cpu(i) | |
153 | sum += per_cpu(nr_dentry_unused, i); | |
154 | return sum < 0 ? 0 : sum; | |
155 | } | |
156 | ||
1f7e0616 | 157 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
158 | size_t *lenp, loff_t *ppos) |
159 | { | |
3e880fb5 | 160 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 161 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 162 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
163 | } |
164 | #endif | |
165 | ||
5483f18e LT |
166 | /* |
167 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
168 | * The strings are both count bytes long, and count is non-zero. | |
169 | */ | |
e419b4cc LT |
170 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
171 | ||
172 | #include <asm/word-at-a-time.h> | |
173 | /* | |
174 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
175 | * aligned allocation for this particular component. We don't | |
176 | * strictly need the load_unaligned_zeropad() safety, but it | |
177 | * doesn't hurt either. | |
178 | * | |
179 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
180 | * need the careful unaligned handling. | |
181 | */ | |
94753db5 | 182 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 183 | { |
bfcfaa77 | 184 | unsigned long a,b,mask; |
bfcfaa77 LT |
185 | |
186 | for (;;) { | |
bfe7aa6c | 187 | a = read_word_at_a_time(cs); |
e419b4cc | 188 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
189 | if (tcount < sizeof(unsigned long)) |
190 | break; | |
191 | if (unlikely(a != b)) | |
192 | return 1; | |
193 | cs += sizeof(unsigned long); | |
194 | ct += sizeof(unsigned long); | |
195 | tcount -= sizeof(unsigned long); | |
196 | if (!tcount) | |
197 | return 0; | |
198 | } | |
a5c21dce | 199 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 200 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
201 | } |
202 | ||
bfcfaa77 | 203 | #else |
e419b4cc | 204 | |
94753db5 | 205 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 206 | { |
5483f18e LT |
207 | do { |
208 | if (*cs != *ct) | |
209 | return 1; | |
210 | cs++; | |
211 | ct++; | |
212 | tcount--; | |
213 | } while (tcount); | |
214 | return 0; | |
215 | } | |
216 | ||
e419b4cc LT |
217 | #endif |
218 | ||
94753db5 LT |
219 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
220 | { | |
94753db5 LT |
221 | /* |
222 | * Be careful about RCU walk racing with rename: | |
506458ef | 223 | * use 'READ_ONCE' to fetch the name pointer. |
94753db5 LT |
224 | * |
225 | * NOTE! Even if a rename will mean that the length | |
226 | * was not loaded atomically, we don't care. The | |
227 | * RCU walk will check the sequence count eventually, | |
228 | * and catch it. And we won't overrun the buffer, | |
229 | * because we're reading the name pointer atomically, | |
230 | * and a dentry name is guaranteed to be properly | |
231 | * terminated with a NUL byte. | |
232 | * | |
233 | * End result: even if 'len' is wrong, we'll exit | |
234 | * early because the data cannot match (there can | |
235 | * be no NUL in the ct/tcount data) | |
236 | */ | |
506458ef | 237 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
ae0a843c | 238 | |
6326c71f | 239 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
240 | } |
241 | ||
8d85b484 AV |
242 | struct external_name { |
243 | union { | |
244 | atomic_t count; | |
245 | struct rcu_head head; | |
246 | } u; | |
247 | unsigned char name[]; | |
248 | }; | |
249 | ||
250 | static inline struct external_name *external_name(struct dentry *dentry) | |
251 | { | |
252 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
253 | } | |
254 | ||
9c82ab9c | 255 | static void __d_free(struct rcu_head *head) |
1da177e4 | 256 | { |
9c82ab9c CH |
257 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
258 | ||
8d85b484 AV |
259 | kmem_cache_free(dentry_cache, dentry); |
260 | } | |
261 | ||
262 | static void __d_free_external(struct rcu_head *head) | |
263 | { | |
264 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
8d85b484 | 265 | kfree(external_name(dentry)); |
1da177e4 LT |
266 | kmem_cache_free(dentry_cache, dentry); |
267 | } | |
268 | ||
810bb172 AV |
269 | static inline int dname_external(const struct dentry *dentry) |
270 | { | |
271 | return dentry->d_name.name != dentry->d_iname; | |
272 | } | |
273 | ||
49d31c2f AV |
274 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
275 | { | |
276 | spin_lock(&dentry->d_lock); | |
277 | if (unlikely(dname_external(dentry))) { | |
278 | struct external_name *p = external_name(dentry); | |
279 | atomic_inc(&p->u.count); | |
280 | spin_unlock(&dentry->d_lock); | |
281 | name->name = p->name; | |
282 | } else { | |
283 | memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN); | |
284 | spin_unlock(&dentry->d_lock); | |
285 | name->name = name->inline_name; | |
286 | } | |
287 | } | |
288 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
289 | ||
290 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
291 | { | |
292 | if (unlikely(name->name != name->inline_name)) { | |
293 | struct external_name *p; | |
294 | p = container_of(name->name, struct external_name, name[0]); | |
295 | if (unlikely(atomic_dec_and_test(&p->u.count))) | |
296 | kfree_rcu(p, u.head); | |
297 | } | |
298 | } | |
299 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
300 | ||
4bf46a27 DH |
301 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
302 | struct inode *inode, | |
303 | unsigned type_flags) | |
304 | { | |
305 | unsigned flags; | |
306 | ||
307 | dentry->d_inode = inode; | |
4bf46a27 DH |
308 | flags = READ_ONCE(dentry->d_flags); |
309 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
310 | flags |= type_flags; | |
311 | WRITE_ONCE(dentry->d_flags, flags); | |
312 | } | |
313 | ||
4bf46a27 DH |
314 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
315 | { | |
316 | unsigned flags = READ_ONCE(dentry->d_flags); | |
317 | ||
318 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
319 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 DH |
320 | dentry->d_inode = NULL; |
321 | } | |
322 | ||
b4f0354e AV |
323 | static void dentry_free(struct dentry *dentry) |
324 | { | |
946e51f2 | 325 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
326 | if (unlikely(dname_external(dentry))) { |
327 | struct external_name *p = external_name(dentry); | |
328 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
329 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
330 | return; | |
331 | } | |
332 | } | |
b4f0354e AV |
333 | /* if dentry was never visible to RCU, immediate free is OK */ |
334 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
335 | __d_free(&dentry->d_u.d_rcu); | |
336 | else | |
337 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
338 | } | |
339 | ||
1da177e4 LT |
340 | /* |
341 | * Release the dentry's inode, using the filesystem | |
550dce01 | 342 | * d_iput() operation if defined. |
31e6b01f NP |
343 | */ |
344 | static void dentry_unlink_inode(struct dentry * dentry) | |
345 | __releases(dentry->d_lock) | |
873feea0 | 346 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
347 | { |
348 | struct inode *inode = dentry->d_inode; | |
550dce01 | 349 | bool hashed = !d_unhashed(dentry); |
a528aca7 | 350 | |
550dce01 AV |
351 | if (hashed) |
352 | raw_write_seqcount_begin(&dentry->d_seq); | |
4bf46a27 | 353 | __d_clear_type_and_inode(dentry); |
946e51f2 | 354 | hlist_del_init(&dentry->d_u.d_alias); |
550dce01 AV |
355 | if (hashed) |
356 | raw_write_seqcount_end(&dentry->d_seq); | |
31e6b01f | 357 | spin_unlock(&dentry->d_lock); |
873feea0 | 358 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
359 | if (!inode->i_nlink) |
360 | fsnotify_inoderemove(inode); | |
361 | if (dentry->d_op && dentry->d_op->d_iput) | |
362 | dentry->d_op->d_iput(dentry, inode); | |
363 | else | |
364 | iput(inode); | |
365 | } | |
366 | ||
89dc77bc LT |
367 | /* |
368 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
369 | * is in use - which includes both the "real" per-superblock | |
370 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
371 | * | |
372 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
373 | * on the shrink list (ie not on the superblock LRU list). | |
374 | * | |
375 | * The per-cpu "nr_dentry_unused" counters are updated with | |
376 | * the DCACHE_LRU_LIST bit. | |
377 | * | |
378 | * These helper functions make sure we always follow the | |
379 | * rules. d_lock must be held by the caller. | |
380 | */ | |
381 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
382 | static void d_lru_add(struct dentry *dentry) | |
383 | { | |
384 | D_FLAG_VERIFY(dentry, 0); | |
385 | dentry->d_flags |= DCACHE_LRU_LIST; | |
386 | this_cpu_inc(nr_dentry_unused); | |
387 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
388 | } | |
389 | ||
390 | static void d_lru_del(struct dentry *dentry) | |
391 | { | |
392 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
393 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
394 | this_cpu_dec(nr_dentry_unused); | |
395 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
396 | } | |
397 | ||
398 | static void d_shrink_del(struct dentry *dentry) | |
399 | { | |
400 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
401 | list_del_init(&dentry->d_lru); | |
402 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
403 | this_cpu_dec(nr_dentry_unused); | |
404 | } | |
405 | ||
406 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
407 | { | |
408 | D_FLAG_VERIFY(dentry, 0); | |
409 | list_add(&dentry->d_lru, list); | |
410 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
411 | this_cpu_inc(nr_dentry_unused); | |
412 | } | |
413 | ||
414 | /* | |
415 | * These can only be called under the global LRU lock, ie during the | |
416 | * callback for freeing the LRU list. "isolate" removes it from the | |
417 | * LRU lists entirely, while shrink_move moves it to the indicated | |
418 | * private list. | |
419 | */ | |
3f97b163 | 420 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
421 | { |
422 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
423 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
424 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 425 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
426 | } |
427 | ||
3f97b163 VD |
428 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
429 | struct list_head *list) | |
89dc77bc LT |
430 | { |
431 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
432 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 433 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
434 | } |
435 | ||
789680d1 NP |
436 | /** |
437 | * d_drop - drop a dentry | |
438 | * @dentry: dentry to drop | |
439 | * | |
440 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
441 | * be found through a VFS lookup any more. Note that this is different from | |
442 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
443 | * possible, giving a successful _negative_ lookup, while d_drop will | |
444 | * just make the cache lookup fail. | |
445 | * | |
446 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
447 | * reason (NFS timeouts or autofs deletes). | |
448 | * | |
61647823 N |
449 | * __d_drop requires dentry->d_lock |
450 | * ___d_drop doesn't mark dentry as "unhashed" | |
451 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
789680d1 | 452 | */ |
61647823 | 453 | static void ___d_drop(struct dentry *dentry) |
789680d1 | 454 | { |
0632a9ac AV |
455 | struct hlist_bl_head *b; |
456 | /* | |
457 | * Hashed dentries are normally on the dentry hashtable, | |
458 | * with the exception of those newly allocated by | |
459 | * d_obtain_root, which are always IS_ROOT: | |
460 | */ | |
461 | if (unlikely(IS_ROOT(dentry))) | |
462 | b = &dentry->d_sb->s_roots; | |
463 | else | |
464 | b = d_hash(dentry->d_name.hash); | |
b61625d2 | 465 | |
0632a9ac AV |
466 | hlist_bl_lock(b); |
467 | __hlist_bl_del(&dentry->d_hash); | |
468 | hlist_bl_unlock(b); | |
789680d1 | 469 | } |
61647823 N |
470 | |
471 | void __d_drop(struct dentry *dentry) | |
472 | { | |
0632a9ac AV |
473 | if (!d_unhashed(dentry)) { |
474 | ___d_drop(dentry); | |
475 | dentry->d_hash.pprev = NULL; | |
476 | write_seqcount_invalidate(&dentry->d_seq); | |
477 | } | |
61647823 | 478 | } |
789680d1 NP |
479 | EXPORT_SYMBOL(__d_drop); |
480 | ||
481 | void d_drop(struct dentry *dentry) | |
482 | { | |
789680d1 NP |
483 | spin_lock(&dentry->d_lock); |
484 | __d_drop(dentry); | |
485 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
486 | } |
487 | EXPORT_SYMBOL(d_drop); | |
488 | ||
ba65dc5e AV |
489 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
490 | { | |
491 | struct dentry *next; | |
492 | /* | |
493 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
494 | * attached to the dentry tree | |
495 | */ | |
496 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
497 | if (unlikely(list_empty(&dentry->d_child))) | |
498 | return; | |
499 | __list_del_entry(&dentry->d_child); | |
500 | /* | |
501 | * Cursors can move around the list of children. While we'd been | |
502 | * a normal list member, it didn't matter - ->d_child.next would've | |
503 | * been updated. However, from now on it won't be and for the | |
504 | * things like d_walk() it might end up with a nasty surprise. | |
505 | * Normally d_walk() doesn't care about cursors moving around - | |
506 | * ->d_lock on parent prevents that and since a cursor has no children | |
507 | * of its own, we get through it without ever unlocking the parent. | |
508 | * There is one exception, though - if we ascend from a child that | |
509 | * gets killed as soon as we unlock it, the next sibling is found | |
510 | * using the value left in its ->d_child.next. And if _that_ | |
511 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
512 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
513 | * everything the cursor had been moved past. | |
514 | * | |
515 | * Solution: make sure that the pointer left behind in ->d_child.next | |
516 | * points to something that won't be moving around. I.e. skip the | |
517 | * cursors. | |
518 | */ | |
519 | while (dentry->d_child.next != &parent->d_subdirs) { | |
520 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
521 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
522 | break; | |
523 | dentry->d_child.next = next->d_child.next; | |
524 | } | |
525 | } | |
526 | ||
e55fd011 | 527 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 528 | { |
41edf278 AV |
529 | struct dentry *parent = NULL; |
530 | bool can_free = true; | |
41edf278 | 531 | if (!IS_ROOT(dentry)) |
77812a1e | 532 | parent = dentry->d_parent; |
31e6b01f | 533 | |
0d98439e LT |
534 | /* |
535 | * The dentry is now unrecoverably dead to the world. | |
536 | */ | |
537 | lockref_mark_dead(&dentry->d_lockref); | |
538 | ||
f0023bc6 | 539 | /* |
f0023bc6 SW |
540 | * inform the fs via d_prune that this dentry is about to be |
541 | * unhashed and destroyed. | |
542 | */ | |
29266201 | 543 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
544 | dentry->d_op->d_prune(dentry); |
545 | ||
01b60351 AV |
546 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
547 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
548 | d_lru_del(dentry); | |
01b60351 | 549 | } |
77812a1e NP |
550 | /* if it was on the hash then remove it */ |
551 | __d_drop(dentry); | |
ba65dc5e | 552 | dentry_unlist(dentry, parent); |
03b3b889 AV |
553 | if (parent) |
554 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
555 | if (dentry->d_inode) |
556 | dentry_unlink_inode(dentry); | |
557 | else | |
558 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
559 | this_cpu_dec(nr_dentry); |
560 | if (dentry->d_op && dentry->d_op->d_release) | |
561 | dentry->d_op->d_release(dentry); | |
562 | ||
41edf278 AV |
563 | spin_lock(&dentry->d_lock); |
564 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
565 | dentry->d_flags |= DCACHE_MAY_FREE; | |
566 | can_free = false; | |
567 | } | |
568 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
569 | if (likely(can_free)) |
570 | dentry_free(dentry); | |
e55fd011 AV |
571 | } |
572 | ||
8b987a46 | 573 | static struct dentry *__lock_parent(struct dentry *dentry) |
046b961b | 574 | { |
8b987a46 | 575 | struct dentry *parent; |
046b961b | 576 | rcu_read_lock(); |
c2338f2d | 577 | spin_unlock(&dentry->d_lock); |
046b961b | 578 | again: |
66702eb5 | 579 | parent = READ_ONCE(dentry->d_parent); |
046b961b AV |
580 | spin_lock(&parent->d_lock); |
581 | /* | |
582 | * We can't blindly lock dentry until we are sure | |
583 | * that we won't violate the locking order. | |
584 | * Any changes of dentry->d_parent must have | |
585 | * been done with parent->d_lock held, so | |
586 | * spin_lock() above is enough of a barrier | |
587 | * for checking if it's still our child. | |
588 | */ | |
589 | if (unlikely(parent != dentry->d_parent)) { | |
590 | spin_unlock(&parent->d_lock); | |
591 | goto again; | |
592 | } | |
65d8eb5a AV |
593 | rcu_read_unlock(); |
594 | if (parent != dentry) | |
9f12600f | 595 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
65d8eb5a | 596 | else |
046b961b AV |
597 | parent = NULL; |
598 | return parent; | |
599 | } | |
600 | ||
8b987a46 AV |
601 | static inline struct dentry *lock_parent(struct dentry *dentry) |
602 | { | |
603 | struct dentry *parent = dentry->d_parent; | |
604 | if (IS_ROOT(dentry)) | |
605 | return NULL; | |
606 | if (likely(spin_trylock(&parent->d_lock))) | |
607 | return parent; | |
608 | return __lock_parent(dentry); | |
609 | } | |
610 | ||
a338579f AV |
611 | static inline bool retain_dentry(struct dentry *dentry) |
612 | { | |
613 | WARN_ON(d_in_lookup(dentry)); | |
614 | ||
615 | /* Unreachable? Get rid of it */ | |
616 | if (unlikely(d_unhashed(dentry))) | |
617 | return false; | |
618 | ||
619 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
620 | return false; | |
621 | ||
622 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
623 | if (dentry->d_op->d_delete(dentry)) | |
624 | return false; | |
625 | } | |
62d9956c AV |
626 | /* retain; LRU fodder */ |
627 | dentry->d_lockref.count--; | |
628 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | |
629 | d_lru_add(dentry); | |
630 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) | |
631 | dentry->d_flags |= DCACHE_REFERENCED; | |
a338579f AV |
632 | return true; |
633 | } | |
634 | ||
c1d0c1a2 JO |
635 | /* |
636 | * Finish off a dentry we've decided to kill. | |
637 | * dentry->d_lock must be held, returns with it unlocked. | |
638 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
639 | */ | |
640 | static struct dentry *dentry_kill(struct dentry *dentry) | |
641 | __releases(dentry->d_lock) | |
642 | { | |
643 | struct inode *inode = dentry->d_inode; | |
644 | struct dentry *parent = NULL; | |
645 | ||
646 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
f657a666 | 647 | goto slow_positive; |
c1d0c1a2 JO |
648 | |
649 | if (!IS_ROOT(dentry)) { | |
650 | parent = dentry->d_parent; | |
651 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
f657a666 AV |
652 | parent = __lock_parent(dentry); |
653 | if (likely(inode || !dentry->d_inode)) | |
654 | goto got_locks; | |
655 | /* negative that became positive */ | |
656 | if (parent) | |
657 | spin_unlock(&parent->d_lock); | |
658 | inode = dentry->d_inode; | |
659 | goto slow_positive; | |
c1d0c1a2 JO |
660 | } |
661 | } | |
c1d0c1a2 JO |
662 | __dentry_kill(dentry); |
663 | return parent; | |
664 | ||
f657a666 AV |
665 | slow_positive: |
666 | spin_unlock(&dentry->d_lock); | |
667 | spin_lock(&inode->i_lock); | |
668 | spin_lock(&dentry->d_lock); | |
669 | parent = lock_parent(dentry); | |
670 | got_locks: | |
671 | if (unlikely(dentry->d_lockref.count != 1)) { | |
672 | dentry->d_lockref.count--; | |
673 | } else if (likely(!retain_dentry(dentry))) { | |
674 | __dentry_kill(dentry); | |
675 | return parent; | |
676 | } | |
677 | /* we are keeping it, after all */ | |
678 | if (inode) | |
679 | spin_unlock(&inode->i_lock); | |
680 | if (parent) | |
681 | spin_unlock(&parent->d_lock); | |
c1d0c1a2 | 682 | spin_unlock(&dentry->d_lock); |
f657a666 | 683 | return NULL; |
c1d0c1a2 JO |
684 | } |
685 | ||
360f5479 LT |
686 | /* |
687 | * Try to do a lockless dput(), and return whether that was successful. | |
688 | * | |
689 | * If unsuccessful, we return false, having already taken the dentry lock. | |
690 | * | |
691 | * The caller needs to hold the RCU read lock, so that the dentry is | |
692 | * guaranteed to stay around even if the refcount goes down to zero! | |
693 | */ | |
694 | static inline bool fast_dput(struct dentry *dentry) | |
695 | { | |
696 | int ret; | |
697 | unsigned int d_flags; | |
698 | ||
699 | /* | |
700 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 701 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
702 | */ |
703 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
704 | return lockref_put_or_lock(&dentry->d_lockref); | |
705 | ||
706 | /* | |
707 | * .. otherwise, we can try to just decrement the | |
708 | * lockref optimistically. | |
709 | */ | |
710 | ret = lockref_put_return(&dentry->d_lockref); | |
711 | ||
712 | /* | |
713 | * If the lockref_put_return() failed due to the lock being held | |
714 | * by somebody else, the fast path has failed. We will need to | |
715 | * get the lock, and then check the count again. | |
716 | */ | |
717 | if (unlikely(ret < 0)) { | |
718 | spin_lock(&dentry->d_lock); | |
719 | if (dentry->d_lockref.count > 1) { | |
720 | dentry->d_lockref.count--; | |
721 | spin_unlock(&dentry->d_lock); | |
722 | return 1; | |
723 | } | |
724 | return 0; | |
725 | } | |
726 | ||
727 | /* | |
728 | * If we weren't the last ref, we're done. | |
729 | */ | |
730 | if (ret) | |
731 | return 1; | |
732 | ||
733 | /* | |
734 | * Careful, careful. The reference count went down | |
735 | * to zero, but we don't hold the dentry lock, so | |
736 | * somebody else could get it again, and do another | |
737 | * dput(), and we need to not race with that. | |
738 | * | |
739 | * However, there is a very special and common case | |
740 | * where we don't care, because there is nothing to | |
741 | * do: the dentry is still hashed, it does not have | |
742 | * a 'delete' op, and it's referenced and already on | |
743 | * the LRU list. | |
744 | * | |
745 | * NOTE! Since we aren't locked, these values are | |
746 | * not "stable". However, it is sufficient that at | |
747 | * some point after we dropped the reference the | |
748 | * dentry was hashed and the flags had the proper | |
749 | * value. Other dentry users may have re-gotten | |
750 | * a reference to the dentry and change that, but | |
751 | * our work is done - we can leave the dentry | |
752 | * around with a zero refcount. | |
753 | */ | |
754 | smp_rmb(); | |
66702eb5 | 755 | d_flags = READ_ONCE(dentry->d_flags); |
75a6f82a | 756 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
757 | |
758 | /* Nothing to do? Dropping the reference was all we needed? */ | |
759 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
760 | return 1; | |
761 | ||
762 | /* | |
763 | * Not the fast normal case? Get the lock. We've already decremented | |
764 | * the refcount, but we'll need to re-check the situation after | |
765 | * getting the lock. | |
766 | */ | |
767 | spin_lock(&dentry->d_lock); | |
768 | ||
769 | /* | |
770 | * Did somebody else grab a reference to it in the meantime, and | |
771 | * we're no longer the last user after all? Alternatively, somebody | |
772 | * else could have killed it and marked it dead. Either way, we | |
773 | * don't need to do anything else. | |
774 | */ | |
775 | if (dentry->d_lockref.count) { | |
776 | spin_unlock(&dentry->d_lock); | |
777 | return 1; | |
778 | } | |
779 | ||
780 | /* | |
781 | * Re-get the reference we optimistically dropped. We hold the | |
782 | * lock, and we just tested that it was zero, so we can just | |
783 | * set it to 1. | |
784 | */ | |
785 | dentry->d_lockref.count = 1; | |
786 | return 0; | |
787 | } | |
788 | ||
789 | ||
1da177e4 LT |
790 | /* |
791 | * This is dput | |
792 | * | |
793 | * This is complicated by the fact that we do not want to put | |
794 | * dentries that are no longer on any hash chain on the unused | |
795 | * list: we'd much rather just get rid of them immediately. | |
796 | * | |
797 | * However, that implies that we have to traverse the dentry | |
798 | * tree upwards to the parents which might _also_ now be | |
799 | * scheduled for deletion (it may have been only waiting for | |
800 | * its last child to go away). | |
801 | * | |
802 | * This tail recursion is done by hand as we don't want to depend | |
803 | * on the compiler to always get this right (gcc generally doesn't). | |
804 | * Real recursion would eat up our stack space. | |
805 | */ | |
806 | ||
807 | /* | |
808 | * dput - release a dentry | |
809 | * @dentry: dentry to release | |
810 | * | |
811 | * Release a dentry. This will drop the usage count and if appropriate | |
812 | * call the dentry unlink method as well as removing it from the queues and | |
813 | * releasing its resources. If the parent dentries were scheduled for release | |
814 | * they too may now get deleted. | |
1da177e4 | 815 | */ |
1da177e4 LT |
816 | void dput(struct dentry *dentry) |
817 | { | |
8aab6a27 | 818 | if (unlikely(!dentry)) |
1da177e4 LT |
819 | return; |
820 | ||
821 | repeat: | |
47be6184 WF |
822 | might_sleep(); |
823 | ||
360f5479 LT |
824 | rcu_read_lock(); |
825 | if (likely(fast_dput(dentry))) { | |
826 | rcu_read_unlock(); | |
1da177e4 | 827 | return; |
360f5479 LT |
828 | } |
829 | ||
830 | /* Slow case: now with the dentry lock held */ | |
831 | rcu_read_unlock(); | |
1da177e4 | 832 | |
a338579f | 833 | if (likely(retain_dentry(dentry))) { |
a338579f AV |
834 | spin_unlock(&dentry->d_lock); |
835 | return; | |
1da177e4 | 836 | } |
265ac902 | 837 | |
8cbf74da | 838 | dentry = dentry_kill(dentry); |
47be6184 WF |
839 | if (dentry) { |
840 | cond_resched(); | |
d52b9086 | 841 | goto repeat; |
47be6184 | 842 | } |
1da177e4 | 843 | } |
ec4f8605 | 844 | EXPORT_SYMBOL(dput); |
1da177e4 | 845 | |
1da177e4 | 846 | |
b5c84bf6 | 847 | /* This must be called with d_lock held */ |
dc0474be | 848 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 849 | { |
98474236 | 850 | dentry->d_lockref.count++; |
23044507 NP |
851 | } |
852 | ||
dc0474be | 853 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 854 | { |
98474236 | 855 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
856 | } |
857 | ||
b7ab39f6 NP |
858 | struct dentry *dget_parent(struct dentry *dentry) |
859 | { | |
df3d0bbc | 860 | int gotref; |
b7ab39f6 NP |
861 | struct dentry *ret; |
862 | ||
df3d0bbc WL |
863 | /* |
864 | * Do optimistic parent lookup without any | |
865 | * locking. | |
866 | */ | |
867 | rcu_read_lock(); | |
66702eb5 | 868 | ret = READ_ONCE(dentry->d_parent); |
df3d0bbc WL |
869 | gotref = lockref_get_not_zero(&ret->d_lockref); |
870 | rcu_read_unlock(); | |
871 | if (likely(gotref)) { | |
66702eb5 | 872 | if (likely(ret == READ_ONCE(dentry->d_parent))) |
df3d0bbc WL |
873 | return ret; |
874 | dput(ret); | |
875 | } | |
876 | ||
b7ab39f6 | 877 | repeat: |
a734eb45 NP |
878 | /* |
879 | * Don't need rcu_dereference because we re-check it was correct under | |
880 | * the lock. | |
881 | */ | |
882 | rcu_read_lock(); | |
b7ab39f6 | 883 | ret = dentry->d_parent; |
a734eb45 NP |
884 | spin_lock(&ret->d_lock); |
885 | if (unlikely(ret != dentry->d_parent)) { | |
886 | spin_unlock(&ret->d_lock); | |
887 | rcu_read_unlock(); | |
b7ab39f6 NP |
888 | goto repeat; |
889 | } | |
a734eb45 | 890 | rcu_read_unlock(); |
98474236 WL |
891 | BUG_ON(!ret->d_lockref.count); |
892 | ret->d_lockref.count++; | |
b7ab39f6 | 893 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
894 | return ret; |
895 | } | |
896 | EXPORT_SYMBOL(dget_parent); | |
897 | ||
1da177e4 LT |
898 | /** |
899 | * d_find_alias - grab a hashed alias of inode | |
900 | * @inode: inode in question | |
1da177e4 LT |
901 | * |
902 | * If inode has a hashed alias, or is a directory and has any alias, | |
903 | * acquire the reference to alias and return it. Otherwise return NULL. | |
904 | * Notice that if inode is a directory there can be only one alias and | |
905 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
906 | * of a filesystem, or if the directory was renamed and d_revalidate |
907 | * was the first vfs operation to notice. | |
1da177e4 | 908 | * |
21c0d8fd | 909 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 910 | * any other hashed alias over that one. |
1da177e4 | 911 | */ |
52ed46f0 | 912 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 913 | { |
da502956 | 914 | struct dentry *alias, *discon_alias; |
1da177e4 | 915 | |
da502956 NP |
916 | again: |
917 | discon_alias = NULL; | |
946e51f2 | 918 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 919 | spin_lock(&alias->d_lock); |
1da177e4 | 920 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 921 | if (IS_ROOT(alias) && |
da502956 | 922 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 923 | discon_alias = alias; |
52ed46f0 | 924 | } else { |
dc0474be | 925 | __dget_dlock(alias); |
da502956 NP |
926 | spin_unlock(&alias->d_lock); |
927 | return alias; | |
928 | } | |
929 | } | |
930 | spin_unlock(&alias->d_lock); | |
931 | } | |
932 | if (discon_alias) { | |
933 | alias = discon_alias; | |
934 | spin_lock(&alias->d_lock); | |
935 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
936 | __dget_dlock(alias); |
937 | spin_unlock(&alias->d_lock); | |
938 | return alias; | |
1da177e4 | 939 | } |
da502956 NP |
940 | spin_unlock(&alias->d_lock); |
941 | goto again; | |
1da177e4 | 942 | } |
da502956 | 943 | return NULL; |
1da177e4 LT |
944 | } |
945 | ||
da502956 | 946 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 947 | { |
214fda1f DH |
948 | struct dentry *de = NULL; |
949 | ||
b3d9b7a3 | 950 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 951 | spin_lock(&inode->i_lock); |
52ed46f0 | 952 | de = __d_find_alias(inode); |
873feea0 | 953 | spin_unlock(&inode->i_lock); |
214fda1f | 954 | } |
1da177e4 LT |
955 | return de; |
956 | } | |
ec4f8605 | 957 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
958 | |
959 | /* | |
960 | * Try to kill dentries associated with this inode. | |
961 | * WARNING: you must own a reference to inode. | |
962 | */ | |
963 | void d_prune_aliases(struct inode *inode) | |
964 | { | |
0cdca3f9 | 965 | struct dentry *dentry; |
1da177e4 | 966 | restart: |
873feea0 | 967 | spin_lock(&inode->i_lock); |
946e51f2 | 968 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 969 | spin_lock(&dentry->d_lock); |
98474236 | 970 | if (!dentry->d_lockref.count) { |
29355c39 AV |
971 | struct dentry *parent = lock_parent(dentry); |
972 | if (likely(!dentry->d_lockref.count)) { | |
973 | __dentry_kill(dentry); | |
4a7795d3 | 974 | dput(parent); |
29355c39 AV |
975 | goto restart; |
976 | } | |
977 | if (parent) | |
978 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
979 | } |
980 | spin_unlock(&dentry->d_lock); | |
981 | } | |
873feea0 | 982 | spin_unlock(&inode->i_lock); |
1da177e4 | 983 | } |
ec4f8605 | 984 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 985 | |
3b3f09f4 AV |
986 | /* |
987 | * Lock a dentry from shrink list. | |
8f04da2a JO |
988 | * Called under rcu_read_lock() and dentry->d_lock; the former |
989 | * guarantees that nothing we access will be freed under us. | |
3b3f09f4 | 990 | * Note that dentry is *not* protected from concurrent dentry_kill(), |
8f04da2a JO |
991 | * d_delete(), etc. |
992 | * | |
3b3f09f4 AV |
993 | * Return false if dentry has been disrupted or grabbed, leaving |
994 | * the caller to kick it off-list. Otherwise, return true and have | |
995 | * that dentry's inode and parent both locked. | |
996 | */ | |
997 | static bool shrink_lock_dentry(struct dentry *dentry) | |
1da177e4 | 998 | { |
3b3f09f4 AV |
999 | struct inode *inode; |
1000 | struct dentry *parent; | |
da3bbdd4 | 1001 | |
3b3f09f4 AV |
1002 | if (dentry->d_lockref.count) |
1003 | return false; | |
1004 | ||
1005 | inode = dentry->d_inode; | |
1006 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
3b3f09f4 AV |
1007 | spin_unlock(&dentry->d_lock); |
1008 | spin_lock(&inode->i_lock); | |
ec33679d | 1009 | spin_lock(&dentry->d_lock); |
3b3f09f4 AV |
1010 | if (unlikely(dentry->d_lockref.count)) |
1011 | goto out; | |
1012 | /* changed inode means that somebody had grabbed it */ | |
1013 | if (unlikely(inode != dentry->d_inode)) | |
1014 | goto out; | |
3b3f09f4 | 1015 | } |
046b961b | 1016 | |
3b3f09f4 AV |
1017 | parent = dentry->d_parent; |
1018 | if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) | |
1019 | return true; | |
dd1f6b2e | 1020 | |
3b3f09f4 | 1021 | spin_unlock(&dentry->d_lock); |
3b3f09f4 AV |
1022 | spin_lock(&parent->d_lock); |
1023 | if (unlikely(parent != dentry->d_parent)) { | |
1024 | spin_unlock(&parent->d_lock); | |
1025 | spin_lock(&dentry->d_lock); | |
1026 | goto out; | |
1027 | } | |
1028 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
8f04da2a | 1029 | if (likely(!dentry->d_lockref.count)) |
3b3f09f4 | 1030 | return true; |
3b3f09f4 AV |
1031 | spin_unlock(&parent->d_lock); |
1032 | out: | |
1033 | if (inode) | |
1034 | spin_unlock(&inode->i_lock); | |
3b3f09f4 AV |
1035 | return false; |
1036 | } | |
77812a1e | 1037 | |
3b3f09f4 AV |
1038 | static void shrink_dentry_list(struct list_head *list) |
1039 | { | |
1040 | while (!list_empty(list)) { | |
1041 | struct dentry *dentry, *parent; | |
64fd72e0 | 1042 | |
3b3f09f4 AV |
1043 | dentry = list_entry(list->prev, struct dentry, d_lru); |
1044 | spin_lock(&dentry->d_lock); | |
8f04da2a | 1045 | rcu_read_lock(); |
3b3f09f4 AV |
1046 | if (!shrink_lock_dentry(dentry)) { |
1047 | bool can_free = false; | |
8f04da2a | 1048 | rcu_read_unlock(); |
3b3f09f4 AV |
1049 | d_shrink_del(dentry); |
1050 | if (dentry->d_lockref.count < 0) | |
1051 | can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
64fd72e0 AV |
1052 | spin_unlock(&dentry->d_lock); |
1053 | if (can_free) | |
1054 | dentry_free(dentry); | |
1055 | continue; | |
1056 | } | |
8f04da2a | 1057 | rcu_read_unlock(); |
3b3f09f4 AV |
1058 | d_shrink_del(dentry); |
1059 | parent = dentry->d_parent; | |
ff2fde99 | 1060 | __dentry_kill(dentry); |
3b3f09f4 AV |
1061 | if (parent == dentry) |
1062 | continue; | |
5c47e6d0 AV |
1063 | /* |
1064 | * We need to prune ancestors too. This is necessary to prevent | |
1065 | * quadratic behavior of shrink_dcache_parent(), but is also | |
1066 | * expected to be beneficial in reducing dentry cache | |
1067 | * fragmentation. | |
1068 | */ | |
1069 | dentry = parent; | |
8f04da2a JO |
1070 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) |
1071 | dentry = dentry_kill(dentry); | |
da3bbdd4 | 1072 | } |
3049cfe2 CH |
1073 | } |
1074 | ||
3f97b163 VD |
1075 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1076 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1077 | { |
1078 | struct list_head *freeable = arg; | |
1079 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1080 | ||
1081 | ||
1082 | /* | |
1083 | * we are inverting the lru lock/dentry->d_lock here, | |
1084 | * so use a trylock. If we fail to get the lock, just skip | |
1085 | * it | |
1086 | */ | |
1087 | if (!spin_trylock(&dentry->d_lock)) | |
1088 | return LRU_SKIP; | |
1089 | ||
1090 | /* | |
1091 | * Referenced dentries are still in use. If they have active | |
1092 | * counts, just remove them from the LRU. Otherwise give them | |
1093 | * another pass through the LRU. | |
1094 | */ | |
1095 | if (dentry->d_lockref.count) { | |
3f97b163 | 1096 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1097 | spin_unlock(&dentry->d_lock); |
1098 | return LRU_REMOVED; | |
1099 | } | |
1100 | ||
1101 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1102 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1103 | spin_unlock(&dentry->d_lock); | |
1104 | ||
1105 | /* | |
1106 | * The list move itself will be made by the common LRU code. At | |
1107 | * this point, we've dropped the dentry->d_lock but keep the | |
1108 | * lru lock. This is safe to do, since every list movement is | |
1109 | * protected by the lru lock even if both locks are held. | |
1110 | * | |
1111 | * This is guaranteed by the fact that all LRU management | |
1112 | * functions are intermediated by the LRU API calls like | |
1113 | * list_lru_add and list_lru_del. List movement in this file | |
1114 | * only ever occur through this functions or through callbacks | |
1115 | * like this one, that are called from the LRU API. | |
1116 | * | |
1117 | * The only exceptions to this are functions like | |
1118 | * shrink_dentry_list, and code that first checks for the | |
1119 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1120 | * operating only with stack provided lists after they are | |
1121 | * properly isolated from the main list. It is thus, always a | |
1122 | * local access. | |
1123 | */ | |
1124 | return LRU_ROTATE; | |
1125 | } | |
1126 | ||
3f97b163 | 1127 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1128 | spin_unlock(&dentry->d_lock); |
1129 | ||
1130 | return LRU_REMOVED; | |
1131 | } | |
1132 | ||
3049cfe2 | 1133 | /** |
b48f03b3 DC |
1134 | * prune_dcache_sb - shrink the dcache |
1135 | * @sb: superblock | |
503c358c | 1136 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1137 | * |
503c358c VD |
1138 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1139 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1140 | * function. |
3049cfe2 | 1141 | * |
b48f03b3 DC |
1142 | * This function may fail to free any resources if all the dentries are in |
1143 | * use. | |
3049cfe2 | 1144 | */ |
503c358c | 1145 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1146 | { |
f6041567 DC |
1147 | LIST_HEAD(dispose); |
1148 | long freed; | |
3049cfe2 | 1149 | |
503c358c VD |
1150 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1151 | dentry_lru_isolate, &dispose); | |
f6041567 | 1152 | shrink_dentry_list(&dispose); |
0a234c6d | 1153 | return freed; |
da3bbdd4 | 1154 | } |
23044507 | 1155 | |
4e717f5c | 1156 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1157 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1158 | { |
4e717f5c GC |
1159 | struct list_head *freeable = arg; |
1160 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1161 | |
4e717f5c GC |
1162 | /* |
1163 | * we are inverting the lru lock/dentry->d_lock here, | |
1164 | * so use a trylock. If we fail to get the lock, just skip | |
1165 | * it | |
1166 | */ | |
1167 | if (!spin_trylock(&dentry->d_lock)) | |
1168 | return LRU_SKIP; | |
1169 | ||
3f97b163 | 1170 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1171 | spin_unlock(&dentry->d_lock); |
ec33679d | 1172 | |
4e717f5c | 1173 | return LRU_REMOVED; |
da3bbdd4 KM |
1174 | } |
1175 | ||
4e717f5c | 1176 | |
1da177e4 LT |
1177 | /** |
1178 | * shrink_dcache_sb - shrink dcache for a superblock | |
1179 | * @sb: superblock | |
1180 | * | |
3049cfe2 CH |
1181 | * Shrink the dcache for the specified super block. This is used to free |
1182 | * the dcache before unmounting a file system. | |
1da177e4 | 1183 | */ |
3049cfe2 | 1184 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1185 | { |
4e717f5c GC |
1186 | long freed; |
1187 | ||
1188 | do { | |
1189 | LIST_HEAD(dispose); | |
1190 | ||
1191 | freed = list_lru_walk(&sb->s_dentry_lru, | |
b17c070f | 1192 | dentry_lru_isolate_shrink, &dispose, 1024); |
3049cfe2 | 1193 | |
4e717f5c GC |
1194 | this_cpu_sub(nr_dentry_unused, freed); |
1195 | shrink_dentry_list(&dispose); | |
b17c070f ST |
1196 | cond_resched(); |
1197 | } while (list_lru_count(&sb->s_dentry_lru) > 0); | |
1da177e4 | 1198 | } |
ec4f8605 | 1199 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1200 | |
db14fc3a MS |
1201 | /** |
1202 | * enum d_walk_ret - action to talke during tree walk | |
1203 | * @D_WALK_CONTINUE: contrinue walk | |
1204 | * @D_WALK_QUIT: quit walk | |
1205 | * @D_WALK_NORETRY: quit when retry is needed | |
1206 | * @D_WALK_SKIP: skip this dentry and its children | |
1207 | */ | |
1208 | enum d_walk_ret { | |
1209 | D_WALK_CONTINUE, | |
1210 | D_WALK_QUIT, | |
1211 | D_WALK_NORETRY, | |
1212 | D_WALK_SKIP, | |
1213 | }; | |
c826cb7d | 1214 | |
1da177e4 | 1215 | /** |
db14fc3a MS |
1216 | * d_walk - walk the dentry tree |
1217 | * @parent: start of walk | |
1218 | * @data: data passed to @enter() and @finish() | |
1219 | * @enter: callback when first entering the dentry | |
1220 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1221 | * |
db14fc3a | 1222 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1223 | */ |
db14fc3a MS |
1224 | static void d_walk(struct dentry *parent, void *data, |
1225 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1226 | void (*finish)(void *)) | |
1da177e4 | 1227 | { |
949854d0 | 1228 | struct dentry *this_parent; |
1da177e4 | 1229 | struct list_head *next; |
48f5ec21 | 1230 | unsigned seq = 0; |
db14fc3a MS |
1231 | enum d_walk_ret ret; |
1232 | bool retry = true; | |
949854d0 | 1233 | |
58db63d0 | 1234 | again: |
48f5ec21 | 1235 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1236 | this_parent = parent; |
2fd6b7f5 | 1237 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1238 | |
1239 | ret = enter(data, this_parent); | |
1240 | switch (ret) { | |
1241 | case D_WALK_CONTINUE: | |
1242 | break; | |
1243 | case D_WALK_QUIT: | |
1244 | case D_WALK_SKIP: | |
1245 | goto out_unlock; | |
1246 | case D_WALK_NORETRY: | |
1247 | retry = false; | |
1248 | break; | |
1249 | } | |
1da177e4 LT |
1250 | repeat: |
1251 | next = this_parent->d_subdirs.next; | |
1252 | resume: | |
1253 | while (next != &this_parent->d_subdirs) { | |
1254 | struct list_head *tmp = next; | |
946e51f2 | 1255 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1256 | next = tmp->next; |
2fd6b7f5 | 1257 | |
ba65dc5e AV |
1258 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1259 | continue; | |
1260 | ||
2fd6b7f5 | 1261 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1262 | |
1263 | ret = enter(data, dentry); | |
1264 | switch (ret) { | |
1265 | case D_WALK_CONTINUE: | |
1266 | break; | |
1267 | case D_WALK_QUIT: | |
2fd6b7f5 | 1268 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1269 | goto out_unlock; |
1270 | case D_WALK_NORETRY: | |
1271 | retry = false; | |
1272 | break; | |
1273 | case D_WALK_SKIP: | |
1274 | spin_unlock(&dentry->d_lock); | |
1275 | continue; | |
2fd6b7f5 | 1276 | } |
db14fc3a | 1277 | |
1da177e4 | 1278 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1279 | spin_unlock(&this_parent->d_lock); |
1280 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1281 | this_parent = dentry; |
2fd6b7f5 | 1282 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1283 | goto repeat; |
1284 | } | |
2fd6b7f5 | 1285 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1286 | } |
1287 | /* | |
1288 | * All done at this level ... ascend and resume the search. | |
1289 | */ | |
ca5358ef AV |
1290 | rcu_read_lock(); |
1291 | ascend: | |
1da177e4 | 1292 | if (this_parent != parent) { |
c826cb7d | 1293 | struct dentry *child = this_parent; |
31dec132 AV |
1294 | this_parent = child->d_parent; |
1295 | ||
31dec132 AV |
1296 | spin_unlock(&child->d_lock); |
1297 | spin_lock(&this_parent->d_lock); | |
1298 | ||
ca5358ef AV |
1299 | /* might go back up the wrong parent if we have had a rename. */ |
1300 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1301 | goto rename_retry; |
2159184e AV |
1302 | /* go into the first sibling still alive */ |
1303 | do { | |
1304 | next = child->d_child.next; | |
ca5358ef AV |
1305 | if (next == &this_parent->d_subdirs) |
1306 | goto ascend; | |
1307 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1308 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1309 | rcu_read_unlock(); |
1da177e4 LT |
1310 | goto resume; |
1311 | } | |
ca5358ef | 1312 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1313 | goto rename_retry; |
ca5358ef | 1314 | rcu_read_unlock(); |
db14fc3a MS |
1315 | if (finish) |
1316 | finish(data); | |
1317 | ||
1318 | out_unlock: | |
1319 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1320 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1321 | return; |
58db63d0 NP |
1322 | |
1323 | rename_retry: | |
ca5358ef AV |
1324 | spin_unlock(&this_parent->d_lock); |
1325 | rcu_read_unlock(); | |
1326 | BUG_ON(seq & 1); | |
db14fc3a MS |
1327 | if (!retry) |
1328 | return; | |
48f5ec21 | 1329 | seq = 1; |
58db63d0 | 1330 | goto again; |
1da177e4 | 1331 | } |
db14fc3a | 1332 | |
01619491 IK |
1333 | struct check_mount { |
1334 | struct vfsmount *mnt; | |
1335 | unsigned int mounted; | |
1336 | }; | |
1337 | ||
1338 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1339 | { | |
1340 | struct check_mount *info = data; | |
1341 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1342 | ||
1343 | if (likely(!d_mountpoint(dentry))) | |
1344 | return D_WALK_CONTINUE; | |
1345 | if (__path_is_mountpoint(&path)) { | |
1346 | info->mounted = 1; | |
1347 | return D_WALK_QUIT; | |
1348 | } | |
1349 | return D_WALK_CONTINUE; | |
1350 | } | |
1351 | ||
1352 | /** | |
1353 | * path_has_submounts - check for mounts over a dentry in the | |
1354 | * current namespace. | |
1355 | * @parent: path to check. | |
1356 | * | |
1357 | * Return true if the parent or its subdirectories contain | |
1358 | * a mount point in the current namespace. | |
1359 | */ | |
1360 | int path_has_submounts(const struct path *parent) | |
1361 | { | |
1362 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1363 | ||
1364 | read_seqlock_excl(&mount_lock); | |
1365 | d_walk(parent->dentry, &data, path_check_mount, NULL); | |
1366 | read_sequnlock_excl(&mount_lock); | |
1367 | ||
1368 | return data.mounted; | |
1369 | } | |
1370 | EXPORT_SYMBOL(path_has_submounts); | |
1371 | ||
eed81007 MS |
1372 | /* |
1373 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1374 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1375 | * subtree can become unreachable). | |
1376 | * | |
1ffe46d1 | 1377 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1378 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1379 | */ | |
1380 | int d_set_mounted(struct dentry *dentry) | |
1381 | { | |
1382 | struct dentry *p; | |
1383 | int ret = -ENOENT; | |
1384 | write_seqlock(&rename_lock); | |
1385 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1386 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1387 | spin_lock(&p->d_lock); |
1388 | if (unlikely(d_unhashed(p))) { | |
1389 | spin_unlock(&p->d_lock); | |
1390 | goto out; | |
1391 | } | |
1392 | spin_unlock(&p->d_lock); | |
1393 | } | |
1394 | spin_lock(&dentry->d_lock); | |
1395 | if (!d_unlinked(dentry)) { | |
3895dbf8 EB |
1396 | ret = -EBUSY; |
1397 | if (!d_mountpoint(dentry)) { | |
1398 | dentry->d_flags |= DCACHE_MOUNTED; | |
1399 | ret = 0; | |
1400 | } | |
eed81007 MS |
1401 | } |
1402 | spin_unlock(&dentry->d_lock); | |
1403 | out: | |
1404 | write_sequnlock(&rename_lock); | |
1405 | return ret; | |
1406 | } | |
1407 | ||
1da177e4 | 1408 | /* |
fd517909 | 1409 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1410 | * and move any unused dentries to the end of the unused |
1411 | * list for prune_dcache(). We descend to the next level | |
1412 | * whenever the d_subdirs list is non-empty and continue | |
1413 | * searching. | |
1414 | * | |
1415 | * It returns zero iff there are no unused children, | |
1416 | * otherwise it returns the number of children moved to | |
1417 | * the end of the unused list. This may not be the total | |
1418 | * number of unused children, because select_parent can | |
1419 | * drop the lock and return early due to latency | |
1420 | * constraints. | |
1421 | */ | |
1da177e4 | 1422 | |
db14fc3a MS |
1423 | struct select_data { |
1424 | struct dentry *start; | |
1425 | struct list_head dispose; | |
1426 | int found; | |
1427 | }; | |
23044507 | 1428 | |
db14fc3a MS |
1429 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1430 | { | |
1431 | struct select_data *data = _data; | |
1432 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1433 | |
db14fc3a MS |
1434 | if (data->start == dentry) |
1435 | goto out; | |
2fd6b7f5 | 1436 | |
fe91522a | 1437 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1438 | data->found++; |
fe91522a AV |
1439 | } else { |
1440 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1441 | d_lru_del(dentry); | |
1442 | if (!dentry->d_lockref.count) { | |
1443 | d_shrink_add(dentry, &data->dispose); | |
1444 | data->found++; | |
1445 | } | |
1da177e4 | 1446 | } |
db14fc3a MS |
1447 | /* |
1448 | * We can return to the caller if we have found some (this | |
1449 | * ensures forward progress). We'll be coming back to find | |
1450 | * the rest. | |
1451 | */ | |
fe91522a AV |
1452 | if (!list_empty(&data->dispose)) |
1453 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1454 | out: |
db14fc3a | 1455 | return ret; |
1da177e4 LT |
1456 | } |
1457 | ||
1458 | /** | |
1459 | * shrink_dcache_parent - prune dcache | |
1460 | * @parent: parent of entries to prune | |
1461 | * | |
1462 | * Prune the dcache to remove unused children of the parent dentry. | |
1463 | */ | |
db14fc3a | 1464 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1465 | { |
db14fc3a MS |
1466 | for (;;) { |
1467 | struct select_data data; | |
1da177e4 | 1468 | |
db14fc3a MS |
1469 | INIT_LIST_HEAD(&data.dispose); |
1470 | data.start = parent; | |
1471 | data.found = 0; | |
1472 | ||
1473 | d_walk(parent, &data, select_collect, NULL); | |
1474 | if (!data.found) | |
1475 | break; | |
1476 | ||
1477 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1478 | cond_resched(); |
1479 | } | |
1da177e4 | 1480 | } |
ec4f8605 | 1481 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1482 | |
9c8c10e2 | 1483 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1484 | { |
9c8c10e2 AV |
1485 | /* it has busy descendents; complain about those instead */ |
1486 | if (!list_empty(&dentry->d_subdirs)) | |
1487 | return D_WALK_CONTINUE; | |
42c32608 | 1488 | |
9c8c10e2 AV |
1489 | /* root with refcount 1 is fine */ |
1490 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1491 | return D_WALK_CONTINUE; | |
1492 | ||
1493 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1494 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1495 | dentry, |
1496 | dentry->d_inode ? | |
1497 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1498 | dentry, |
42c32608 AV |
1499 | dentry->d_lockref.count, |
1500 | dentry->d_sb->s_type->name, | |
1501 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1502 | WARN_ON(1); |
1503 | return D_WALK_CONTINUE; | |
1504 | } | |
1505 | ||
1506 | static void do_one_tree(struct dentry *dentry) | |
1507 | { | |
1508 | shrink_dcache_parent(dentry); | |
1509 | d_walk(dentry, dentry, umount_check, NULL); | |
1510 | d_drop(dentry); | |
1511 | dput(dentry); | |
42c32608 AV |
1512 | } |
1513 | ||
1514 | /* | |
1515 | * destroy the dentries attached to a superblock on unmounting | |
1516 | */ | |
1517 | void shrink_dcache_for_umount(struct super_block *sb) | |
1518 | { | |
1519 | struct dentry *dentry; | |
1520 | ||
9c8c10e2 | 1521 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1522 | |
1523 | dentry = sb->s_root; | |
1524 | sb->s_root = NULL; | |
9c8c10e2 | 1525 | do_one_tree(dentry); |
42c32608 | 1526 | |
f1ee6162 N |
1527 | while (!hlist_bl_empty(&sb->s_roots)) { |
1528 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); | |
9c8c10e2 | 1529 | do_one_tree(dentry); |
42c32608 AV |
1530 | } |
1531 | } | |
1532 | ||
8ed936b5 EB |
1533 | struct detach_data { |
1534 | struct select_data select; | |
1535 | struct dentry *mountpoint; | |
1536 | }; | |
1537 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1538 | { |
8ed936b5 | 1539 | struct detach_data *data = _data; |
848ac114 MS |
1540 | |
1541 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1542 | __dget_dlock(dentry); |
1543 | data->mountpoint = dentry; | |
848ac114 MS |
1544 | return D_WALK_QUIT; |
1545 | } | |
1546 | ||
8ed936b5 | 1547 | return select_collect(&data->select, dentry); |
848ac114 MS |
1548 | } |
1549 | ||
1550 | static void check_and_drop(void *_data) | |
1551 | { | |
8ed936b5 | 1552 | struct detach_data *data = _data; |
848ac114 | 1553 | |
81be24d2 | 1554 | if (!data->mountpoint && list_empty(&data->select.dispose)) |
8ed936b5 | 1555 | __d_drop(data->select.start); |
848ac114 MS |
1556 | } |
1557 | ||
1558 | /** | |
1ffe46d1 EB |
1559 | * d_invalidate - detach submounts, prune dcache, and drop |
1560 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1561 | * | |
1ffe46d1 | 1562 | * no dcache lock. |
848ac114 | 1563 | * |
8ed936b5 EB |
1564 | * The final d_drop is done as an atomic operation relative to |
1565 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1566 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1567 | */ |
5542aa2f | 1568 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1569 | { |
1ffe46d1 EB |
1570 | /* |
1571 | * If it's already been dropped, return OK. | |
1572 | */ | |
1573 | spin_lock(&dentry->d_lock); | |
1574 | if (d_unhashed(dentry)) { | |
1575 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1576 | return; |
1ffe46d1 EB |
1577 | } |
1578 | spin_unlock(&dentry->d_lock); | |
1579 | ||
848ac114 MS |
1580 | /* Negative dentries can be dropped without further checks */ |
1581 | if (!dentry->d_inode) { | |
1582 | d_drop(dentry); | |
5542aa2f | 1583 | return; |
848ac114 MS |
1584 | } |
1585 | ||
1586 | for (;;) { | |
8ed936b5 | 1587 | struct detach_data data; |
848ac114 | 1588 | |
8ed936b5 EB |
1589 | data.mountpoint = NULL; |
1590 | INIT_LIST_HEAD(&data.select.dispose); | |
1591 | data.select.start = dentry; | |
1592 | data.select.found = 0; | |
1593 | ||
1594 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1595 | |
81be24d2 | 1596 | if (!list_empty(&data.select.dispose)) |
8ed936b5 | 1597 | shrink_dentry_list(&data.select.dispose); |
81be24d2 AV |
1598 | else if (!data.mountpoint) |
1599 | return; | |
848ac114 | 1600 | |
8ed936b5 EB |
1601 | if (data.mountpoint) { |
1602 | detach_mounts(data.mountpoint); | |
1603 | dput(data.mountpoint); | |
1604 | } | |
848ac114 MS |
1605 | cond_resched(); |
1606 | } | |
848ac114 | 1607 | } |
1ffe46d1 | 1608 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1609 | |
1da177e4 | 1610 | /** |
a4464dbc AV |
1611 | * __d_alloc - allocate a dcache entry |
1612 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1613 | * @name: qstr of the name |
1614 | * | |
1615 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1616 | * available. On a success the dentry is returned. The name passed in is | |
1617 | * copied and the copy passed in may be reused after this call. | |
1618 | */ | |
1619 | ||
a4464dbc | 1620 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1621 | { |
1622 | struct dentry *dentry; | |
1623 | char *dname; | |
285b102d | 1624 | int err; |
1da177e4 | 1625 | |
e12ba74d | 1626 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1627 | if (!dentry) |
1628 | return NULL; | |
1629 | ||
6326c71f LT |
1630 | /* |
1631 | * We guarantee that the inline name is always NUL-terminated. | |
1632 | * This way the memcpy() done by the name switching in rename | |
1633 | * will still always have a NUL at the end, even if we might | |
1634 | * be overwriting an internal NUL character | |
1635 | */ | |
1636 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd | 1637 | if (unlikely(!name)) { |
cdf01226 | 1638 | name = &slash_name; |
798434bd AV |
1639 | dname = dentry->d_iname; |
1640 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1641 | size_t size = offsetof(struct external_name, name[1]); |
5d097056 VD |
1642 | struct external_name *p = kmalloc(size + name->len, |
1643 | GFP_KERNEL_ACCOUNT); | |
8d85b484 | 1644 | if (!p) { |
1da177e4 LT |
1645 | kmem_cache_free(dentry_cache, dentry); |
1646 | return NULL; | |
1647 | } | |
8d85b484 AV |
1648 | atomic_set(&p->u.count, 1); |
1649 | dname = p->name; | |
1da177e4 LT |
1650 | } else { |
1651 | dname = dentry->d_iname; | |
1652 | } | |
1da177e4 LT |
1653 | |
1654 | dentry->d_name.len = name->len; | |
1655 | dentry->d_name.hash = name->hash; | |
1656 | memcpy(dname, name->name, name->len); | |
1657 | dname[name->len] = 0; | |
1658 | ||
6326c71f | 1659 | /* Make sure we always see the terminating NUL character */ |
7088efa9 | 1660 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ |
6326c71f | 1661 | |
98474236 | 1662 | dentry->d_lockref.count = 1; |
dea3667b | 1663 | dentry->d_flags = 0; |
1da177e4 | 1664 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1665 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1666 | dentry->d_inode = NULL; |
a4464dbc AV |
1667 | dentry->d_parent = dentry; |
1668 | dentry->d_sb = sb; | |
1da177e4 LT |
1669 | dentry->d_op = NULL; |
1670 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1671 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1672 | INIT_LIST_HEAD(&dentry->d_lru); |
1673 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1674 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1675 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1676 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1677 | |
285b102d MS |
1678 | if (dentry->d_op && dentry->d_op->d_init) { |
1679 | err = dentry->d_op->d_init(dentry); | |
1680 | if (err) { | |
1681 | if (dname_external(dentry)) | |
1682 | kfree(external_name(dentry)); | |
1683 | kmem_cache_free(dentry_cache, dentry); | |
1684 | return NULL; | |
1685 | } | |
1686 | } | |
1687 | ||
3e880fb5 | 1688 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1689 | |
1da177e4 LT |
1690 | return dentry; |
1691 | } | |
a4464dbc AV |
1692 | |
1693 | /** | |
1694 | * d_alloc - allocate a dcache entry | |
1695 | * @parent: parent of entry to allocate | |
1696 | * @name: qstr of the name | |
1697 | * | |
1698 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1699 | * available. On a success the dentry is returned. The name passed in is | |
1700 | * copied and the copy passed in may be reused after this call. | |
1701 | */ | |
1702 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1703 | { | |
1704 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1705 | if (!dentry) | |
1706 | return NULL; | |
3d56c25e | 1707 | dentry->d_flags |= DCACHE_RCUACCESS; |
a4464dbc AV |
1708 | spin_lock(&parent->d_lock); |
1709 | /* | |
1710 | * don't need child lock because it is not subject | |
1711 | * to concurrency here | |
1712 | */ | |
1713 | __dget_dlock(parent); | |
1714 | dentry->d_parent = parent; | |
946e51f2 | 1715 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1716 | spin_unlock(&parent->d_lock); |
1717 | ||
1718 | return dentry; | |
1719 | } | |
ec4f8605 | 1720 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1721 | |
f9c34674 MS |
1722 | struct dentry *d_alloc_anon(struct super_block *sb) |
1723 | { | |
1724 | return __d_alloc(sb, NULL); | |
1725 | } | |
1726 | EXPORT_SYMBOL(d_alloc_anon); | |
1727 | ||
ba65dc5e AV |
1728 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1729 | { | |
f9c34674 | 1730 | struct dentry *dentry = d_alloc_anon(parent->d_sb); |
ba65dc5e AV |
1731 | if (dentry) { |
1732 | dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR; | |
1733 | dentry->d_parent = dget(parent); | |
1734 | } | |
1735 | return dentry; | |
1736 | } | |
1737 | ||
e1a24bb0 BF |
1738 | /** |
1739 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1740 | * @sb: the superblock | |
1741 | * @name: qstr of the name | |
1742 | * | |
1743 | * For a filesystem that just pins its dentries in memory and never | |
1744 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1745 | */ | |
4b936885 NP |
1746 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1747 | { | |
e1a24bb0 | 1748 | return __d_alloc(sb, name); |
4b936885 NP |
1749 | } |
1750 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1751 | ||
1da177e4 LT |
1752 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1753 | { | |
1754 | struct qstr q; | |
1755 | ||
1756 | q.name = name; | |
8387ff25 | 1757 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1758 | return d_alloc(parent, &q); |
1759 | } | |
ef26ca97 | 1760 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1761 | |
fb045adb NP |
1762 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1763 | { | |
6f7f7caa LT |
1764 | WARN_ON_ONCE(dentry->d_op); |
1765 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1766 | DCACHE_OP_COMPARE | |
1767 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1768 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1769 | DCACHE_OP_DELETE | |
d101a125 | 1770 | DCACHE_OP_REAL)); |
fb045adb NP |
1771 | dentry->d_op = op; |
1772 | if (!op) | |
1773 | return; | |
1774 | if (op->d_hash) | |
1775 | dentry->d_flags |= DCACHE_OP_HASH; | |
1776 | if (op->d_compare) | |
1777 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1778 | if (op->d_revalidate) | |
1779 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1780 | if (op->d_weak_revalidate) |
1781 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1782 | if (op->d_delete) |
1783 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1784 | if (op->d_prune) |
1785 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1786 | if (op->d_real) |
1787 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1788 | |
1789 | } | |
1790 | EXPORT_SYMBOL(d_set_d_op); | |
1791 | ||
df1a085a DH |
1792 | |
1793 | /* | |
1794 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1795 | * @dentry - The dentry to mark | |
1796 | * | |
1797 | * Mark a dentry as falling through to the lower layer (as set with | |
1798 | * d_pin_lower()). This flag may be recorded on the medium. | |
1799 | */ | |
1800 | void d_set_fallthru(struct dentry *dentry) | |
1801 | { | |
1802 | spin_lock(&dentry->d_lock); | |
1803 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1804 | spin_unlock(&dentry->d_lock); | |
1805 | } | |
1806 | EXPORT_SYMBOL(d_set_fallthru); | |
1807 | ||
b18825a7 DH |
1808 | static unsigned d_flags_for_inode(struct inode *inode) |
1809 | { | |
44bdb5e5 | 1810 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1811 | |
1812 | if (!inode) | |
1813 | return DCACHE_MISS_TYPE; | |
1814 | ||
1815 | if (S_ISDIR(inode->i_mode)) { | |
1816 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1817 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1818 | if (unlikely(!inode->i_op->lookup)) | |
1819 | add_flags = DCACHE_AUTODIR_TYPE; | |
1820 | else | |
1821 | inode->i_opflags |= IOP_LOOKUP; | |
1822 | } | |
44bdb5e5 DH |
1823 | goto type_determined; |
1824 | } | |
1825 | ||
1826 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1827 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1828 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1829 | goto type_determined; |
1830 | } | |
1831 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1832 | } |
1833 | ||
44bdb5e5 DH |
1834 | if (unlikely(!S_ISREG(inode->i_mode))) |
1835 | add_flags = DCACHE_SPECIAL_TYPE; | |
1836 | ||
1837 | type_determined: | |
b18825a7 DH |
1838 | if (unlikely(IS_AUTOMOUNT(inode))) |
1839 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1840 | return add_flags; | |
1841 | } | |
1842 | ||
360da900 OH |
1843 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1844 | { | |
b18825a7 | 1845 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1846 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1847 | |
b23fb0a6 | 1848 | spin_lock(&dentry->d_lock); |
de689f5e | 1849 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1850 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1851 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1852 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1853 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1854 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1855 | } |
1856 | ||
1da177e4 LT |
1857 | /** |
1858 | * d_instantiate - fill in inode information for a dentry | |
1859 | * @entry: dentry to complete | |
1860 | * @inode: inode to attach to this dentry | |
1861 | * | |
1862 | * Fill in inode information in the entry. | |
1863 | * | |
1864 | * This turns negative dentries into productive full members | |
1865 | * of society. | |
1866 | * | |
1867 | * NOTE! This assumes that the inode count has been incremented | |
1868 | * (or otherwise set) by the caller to indicate that it is now | |
1869 | * in use by the dcache. | |
1870 | */ | |
1871 | ||
1872 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1873 | { | |
946e51f2 | 1874 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1875 | if (inode) { |
b9680917 | 1876 | security_d_instantiate(entry, inode); |
873feea0 | 1877 | spin_lock(&inode->i_lock); |
de689f5e | 1878 | __d_instantiate(entry, inode); |
873feea0 | 1879 | spin_unlock(&inode->i_lock); |
de689f5e | 1880 | } |
1da177e4 | 1881 | } |
ec4f8605 | 1882 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1883 | |
b70a80e7 MS |
1884 | /** |
1885 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1886 | * @entry: dentry to complete | |
1887 | * @inode: inode to attach to this dentry | |
1888 | * | |
1889 | * Fill in inode information in the entry. If a directory alias is found, then | |
1890 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1891 | * guarantees that a directory inode may never have more than one alias. | |
1892 | */ | |
1893 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1894 | { | |
946e51f2 | 1895 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 | 1896 | |
b9680917 | 1897 | security_d_instantiate(entry, inode); |
b70a80e7 MS |
1898 | spin_lock(&inode->i_lock); |
1899 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1900 | spin_unlock(&inode->i_lock); | |
1901 | iput(inode); | |
1902 | return -EBUSY; | |
1903 | } | |
1904 | __d_instantiate(entry, inode); | |
1905 | spin_unlock(&inode->i_lock); | |
b70a80e7 MS |
1906 | |
1907 | return 0; | |
1908 | } | |
1909 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1910 | ||
adc0e91a AV |
1911 | struct dentry *d_make_root(struct inode *root_inode) |
1912 | { | |
1913 | struct dentry *res = NULL; | |
1914 | ||
1915 | if (root_inode) { | |
f9c34674 | 1916 | res = d_alloc_anon(root_inode->i_sb); |
adc0e91a AV |
1917 | if (res) |
1918 | d_instantiate(res, root_inode); | |
1919 | else | |
1920 | iput(root_inode); | |
1921 | } | |
1922 | return res; | |
1923 | } | |
1924 | EXPORT_SYMBOL(d_make_root); | |
1925 | ||
d891eedb BF |
1926 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1927 | { | |
1928 | struct dentry *alias; | |
1929 | ||
b3d9b7a3 | 1930 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1931 | return NULL; |
946e51f2 | 1932 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1933 | __dget(alias); |
1934 | return alias; | |
1935 | } | |
1936 | ||
46f72b34 SW |
1937 | /** |
1938 | * d_find_any_alias - find any alias for a given inode | |
1939 | * @inode: inode to find an alias for | |
1940 | * | |
1941 | * If any aliases exist for the given inode, take and return a | |
1942 | * reference for one of them. If no aliases exist, return %NULL. | |
1943 | */ | |
1944 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1945 | { |
1946 | struct dentry *de; | |
1947 | ||
1948 | spin_lock(&inode->i_lock); | |
1949 | de = __d_find_any_alias(inode); | |
1950 | spin_unlock(&inode->i_lock); | |
1951 | return de; | |
1952 | } | |
46f72b34 | 1953 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1954 | |
f9c34674 MS |
1955 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, |
1956 | struct inode *inode, | |
1957 | bool disconnected) | |
4ea3ada2 | 1958 | { |
9308a612 | 1959 | struct dentry *res; |
b18825a7 | 1960 | unsigned add_flags; |
4ea3ada2 | 1961 | |
f9c34674 | 1962 | security_d_instantiate(dentry, inode); |
873feea0 | 1963 | spin_lock(&inode->i_lock); |
d891eedb | 1964 | res = __d_find_any_alias(inode); |
9308a612 | 1965 | if (res) { |
873feea0 | 1966 | spin_unlock(&inode->i_lock); |
f9c34674 | 1967 | dput(dentry); |
9308a612 CH |
1968 | goto out_iput; |
1969 | } | |
1970 | ||
1971 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1972 | add_flags = d_flags_for_inode(inode); |
1973 | ||
1974 | if (disconnected) | |
1975 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1976 | |
f9c34674 MS |
1977 | spin_lock(&dentry->d_lock); |
1978 | __d_set_inode_and_type(dentry, inode, add_flags); | |
1979 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
f1ee6162 | 1980 | if (!disconnected) { |
139351f1 LT |
1981 | hlist_bl_lock(&dentry->d_sb->s_roots); |
1982 | hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots); | |
1983 | hlist_bl_unlock(&dentry->d_sb->s_roots); | |
f1ee6162 | 1984 | } |
f9c34674 | 1985 | spin_unlock(&dentry->d_lock); |
873feea0 | 1986 | spin_unlock(&inode->i_lock); |
9308a612 | 1987 | |
f9c34674 | 1988 | return dentry; |
9308a612 CH |
1989 | |
1990 | out_iput: | |
1991 | iput(inode); | |
1992 | return res; | |
4ea3ada2 | 1993 | } |
1a0a397e | 1994 | |
f9c34674 MS |
1995 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) |
1996 | { | |
1997 | return __d_instantiate_anon(dentry, inode, true); | |
1998 | } | |
1999 | EXPORT_SYMBOL(d_instantiate_anon); | |
2000 | ||
2001 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) | |
2002 | { | |
2003 | struct dentry *tmp; | |
2004 | struct dentry *res; | |
2005 | ||
2006 | if (!inode) | |
2007 | return ERR_PTR(-ESTALE); | |
2008 | if (IS_ERR(inode)) | |
2009 | return ERR_CAST(inode); | |
2010 | ||
2011 | res = d_find_any_alias(inode); | |
2012 | if (res) | |
2013 | goto out_iput; | |
2014 | ||
2015 | tmp = d_alloc_anon(inode->i_sb); | |
2016 | if (!tmp) { | |
2017 | res = ERR_PTR(-ENOMEM); | |
2018 | goto out_iput; | |
2019 | } | |
2020 | ||
2021 | return __d_instantiate_anon(tmp, inode, disconnected); | |
2022 | ||
2023 | out_iput: | |
2024 | iput(inode); | |
2025 | return res; | |
2026 | } | |
2027 | ||
1a0a397e BF |
2028 | /** |
2029 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2030 | * @inode: inode to allocate the dentry for | |
2031 | * | |
2032 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2033 | * similar open by handle operations. The returned dentry may be anonymous, | |
2034 | * or may have a full name (if the inode was already in the cache). | |
2035 | * | |
2036 | * When called on a directory inode, we must ensure that the inode only ever | |
2037 | * has one dentry. If a dentry is found, that is returned instead of | |
2038 | * allocating a new one. | |
2039 | * | |
2040 | * On successful return, the reference to the inode has been transferred | |
2041 | * to the dentry. In case of an error the reference on the inode is released. | |
2042 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2043 | * be passed in and the error will be propagated to the return value, | |
2044 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2045 | */ | |
2046 | struct dentry *d_obtain_alias(struct inode *inode) | |
2047 | { | |
f9c34674 | 2048 | return __d_obtain_alias(inode, true); |
1a0a397e | 2049 | } |
adc48720 | 2050 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 2051 | |
1a0a397e BF |
2052 | /** |
2053 | * d_obtain_root - find or allocate a dentry for a given inode | |
2054 | * @inode: inode to allocate the dentry for | |
2055 | * | |
2056 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2057 | * | |
2058 | * We must ensure that directory inodes only ever have one dentry. If a | |
2059 | * dentry is found, that is returned instead of allocating a new one. | |
2060 | * | |
2061 | * On successful return, the reference to the inode has been transferred | |
2062 | * to the dentry. In case of an error the reference on the inode is | |
2063 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2064 | * error will be propagate to the return value, with a %NULL @inode | |
2065 | * replaced by ERR_PTR(-ESTALE). | |
2066 | */ | |
2067 | struct dentry *d_obtain_root(struct inode *inode) | |
2068 | { | |
f9c34674 | 2069 | return __d_obtain_alias(inode, false); |
1a0a397e BF |
2070 | } |
2071 | EXPORT_SYMBOL(d_obtain_root); | |
2072 | ||
9403540c BN |
2073 | /** |
2074 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2075 | * @inode: the inode case-insensitive lookup has found | |
2076 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2077 | * @name: the case-exact name to be associated with the returned dentry | |
2078 | * | |
2079 | * This is to avoid filling the dcache with case-insensitive names to the | |
2080 | * same inode, only the actual correct case is stored in the dcache for | |
2081 | * case-insensitive filesystems. | |
2082 | * | |
2083 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2084 | * already exists in in the dcache, use it and return it. | |
2085 | * | |
2086 | * If no entry exists with the exact case name, allocate new dentry with | |
2087 | * the exact case, and return the spliced entry. | |
2088 | */ | |
e45b590b | 2089 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2090 | struct qstr *name) |
2091 | { | |
d9171b93 | 2092 | struct dentry *found, *res; |
9403540c | 2093 | |
b6520c81 CH |
2094 | /* |
2095 | * First check if a dentry matching the name already exists, | |
2096 | * if not go ahead and create it now. | |
2097 | */ | |
9403540c | 2098 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2099 | if (found) { |
2100 | iput(inode); | |
2101 | return found; | |
2102 | } | |
2103 | if (d_in_lookup(dentry)) { | |
2104 | found = d_alloc_parallel(dentry->d_parent, name, | |
2105 | dentry->d_wait); | |
2106 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2107 | iput(inode); | |
2108 | return found; | |
9403540c | 2109 | } |
d9171b93 AV |
2110 | } else { |
2111 | found = d_alloc(dentry->d_parent, name); | |
2112 | if (!found) { | |
2113 | iput(inode); | |
2114 | return ERR_PTR(-ENOMEM); | |
2115 | } | |
2116 | } | |
2117 | res = d_splice_alias(inode, found); | |
2118 | if (res) { | |
2119 | dput(found); | |
2120 | return res; | |
9403540c | 2121 | } |
4f522a24 | 2122 | return found; |
9403540c | 2123 | } |
ec4f8605 | 2124 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2125 | |
12f8ad4b | 2126 | |
d4c91a8f AV |
2127 | static inline bool d_same_name(const struct dentry *dentry, |
2128 | const struct dentry *parent, | |
2129 | const struct qstr *name) | |
12f8ad4b | 2130 | { |
d4c91a8f AV |
2131 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2132 | if (dentry->d_name.len != name->len) | |
2133 | return false; | |
2134 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2135 | } |
6fa67e70 | 2136 | return parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2137 | dentry->d_name.len, dentry->d_name.name, |
2138 | name) == 0; | |
12f8ad4b LT |
2139 | } |
2140 | ||
31e6b01f NP |
2141 | /** |
2142 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2143 | * @parent: parent dentry | |
2144 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2145 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2146 | * Returns: dentry, or NULL |
2147 | * | |
2148 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2149 | * resolution (store-free path walking) design described in | |
2150 | * Documentation/filesystems/path-lookup.txt. | |
2151 | * | |
2152 | * This is not to be used outside core vfs. | |
2153 | * | |
2154 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2155 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2156 | * without taking d_lock and checking d_seq sequence count against @seq | |
2157 | * returned here. | |
2158 | * | |
15570086 | 2159 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2160 | * function. |
2161 | * | |
2162 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2163 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2164 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2165 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2166 | * |
2167 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2168 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2169 | */ |
8966be90 LT |
2170 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2171 | const struct qstr *name, | |
da53be12 | 2172 | unsigned *seqp) |
31e6b01f | 2173 | { |
26fe5750 | 2174 | u64 hashlen = name->hash_len; |
31e6b01f | 2175 | const unsigned char *str = name->name; |
8387ff25 | 2176 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2177 | struct hlist_bl_node *node; |
31e6b01f NP |
2178 | struct dentry *dentry; |
2179 | ||
2180 | /* | |
2181 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2182 | * required to prevent single threaded performance regressions | |
2183 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2184 | * Keep the two functions in sync. | |
2185 | */ | |
2186 | ||
2187 | /* | |
2188 | * The hash list is protected using RCU. | |
2189 | * | |
2190 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2191 | * races with d_move(). | |
2192 | * | |
2193 | * It is possible that concurrent renames can mess up our list | |
2194 | * walk here and result in missing our dentry, resulting in the | |
2195 | * false-negative result. d_lookup() protects against concurrent | |
2196 | * renames using rename_lock seqlock. | |
2197 | * | |
b0a4bb83 | 2198 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2199 | */ |
b07ad996 | 2200 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2201 | unsigned seq; |
31e6b01f | 2202 | |
31e6b01f | 2203 | seqretry: |
12f8ad4b LT |
2204 | /* |
2205 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2206 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2207 | * |
2208 | * The caller must perform a seqcount check in order | |
da53be12 | 2209 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2210 | * |
2211 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2212 | * we don't wait for the sequence count to stabilize if it | |
2213 | * is in the middle of a sequence change. If we do the slow | |
2214 | * dentry compare, we will do seqretries until it is stable, | |
2215 | * and if we end up with a successful lookup, we actually | |
2216 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2217 | * |
2218 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2219 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2220 | */ |
2221 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2222 | if (dentry->d_parent != parent) |
2223 | continue; | |
2e321806 LT |
2224 | if (d_unhashed(dentry)) |
2225 | continue; | |
12f8ad4b | 2226 | |
830c0f0e | 2227 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
d4c91a8f AV |
2228 | int tlen; |
2229 | const char *tname; | |
26fe5750 LT |
2230 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2231 | continue; | |
d4c91a8f AV |
2232 | tlen = dentry->d_name.len; |
2233 | tname = dentry->d_name.name; | |
2234 | /* we want a consistent (name,len) pair */ | |
2235 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2236 | cpu_relax(); | |
12f8ad4b LT |
2237 | goto seqretry; |
2238 | } | |
6fa67e70 | 2239 | if (parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2240 | tlen, tname, name) != 0) |
2241 | continue; | |
2242 | } else { | |
2243 | if (dentry->d_name.hash_len != hashlen) | |
2244 | continue; | |
2245 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2246 | continue; | |
31e6b01f | 2247 | } |
da53be12 | 2248 | *seqp = seq; |
d4c91a8f | 2249 | return dentry; |
31e6b01f NP |
2250 | } |
2251 | return NULL; | |
2252 | } | |
2253 | ||
1da177e4 LT |
2254 | /** |
2255 | * d_lookup - search for a dentry | |
2256 | * @parent: parent dentry | |
2257 | * @name: qstr of name we wish to find | |
b04f784e | 2258 | * Returns: dentry, or NULL |
1da177e4 | 2259 | * |
b04f784e NP |
2260 | * d_lookup searches the children of the parent dentry for the name in |
2261 | * question. If the dentry is found its reference count is incremented and the | |
2262 | * dentry is returned. The caller must use dput to free the entry when it has | |
2263 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2264 | */ |
da2d8455 | 2265 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2266 | { |
31e6b01f | 2267 | struct dentry *dentry; |
949854d0 | 2268 | unsigned seq; |
1da177e4 | 2269 | |
b8314f93 DY |
2270 | do { |
2271 | seq = read_seqbegin(&rename_lock); | |
2272 | dentry = __d_lookup(parent, name); | |
2273 | if (dentry) | |
1da177e4 LT |
2274 | break; |
2275 | } while (read_seqretry(&rename_lock, seq)); | |
2276 | return dentry; | |
2277 | } | |
ec4f8605 | 2278 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2279 | |
31e6b01f | 2280 | /** |
b04f784e NP |
2281 | * __d_lookup - search for a dentry (racy) |
2282 | * @parent: parent dentry | |
2283 | * @name: qstr of name we wish to find | |
2284 | * Returns: dentry, or NULL | |
2285 | * | |
2286 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2287 | * false-negative result due to unrelated rename activity. | |
2288 | * | |
2289 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2290 | * however it must be used carefully, eg. with a following d_lookup in | |
2291 | * the case of failure. | |
2292 | * | |
2293 | * __d_lookup callers must be commented. | |
2294 | */ | |
a713ca2a | 2295 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2296 | { |
1da177e4 | 2297 | unsigned int hash = name->hash; |
8387ff25 | 2298 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2299 | struct hlist_bl_node *node; |
31e6b01f | 2300 | struct dentry *found = NULL; |
665a7583 | 2301 | struct dentry *dentry; |
1da177e4 | 2302 | |
31e6b01f NP |
2303 | /* |
2304 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2305 | * required to prevent single threaded performance regressions | |
2306 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2307 | * Keep the two functions in sync. | |
2308 | */ | |
2309 | ||
b04f784e NP |
2310 | /* |
2311 | * The hash list is protected using RCU. | |
2312 | * | |
2313 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2314 | * 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. |
b04f784e | 2322 | */ |
1da177e4 LT |
2323 | rcu_read_lock(); |
2324 | ||
b07ad996 | 2325 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2326 | |
1da177e4 LT |
2327 | if (dentry->d_name.hash != hash) |
2328 | continue; | |
1da177e4 LT |
2329 | |
2330 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2331 | if (dentry->d_parent != parent) |
2332 | goto next; | |
d0185c08 LT |
2333 | if (d_unhashed(dentry)) |
2334 | goto next; | |
2335 | ||
d4c91a8f AV |
2336 | if (!d_same_name(dentry, parent, name)) |
2337 | goto next; | |
1da177e4 | 2338 | |
98474236 | 2339 | dentry->d_lockref.count++; |
d0185c08 | 2340 | found = dentry; |
1da177e4 LT |
2341 | spin_unlock(&dentry->d_lock); |
2342 | break; | |
2343 | next: | |
2344 | spin_unlock(&dentry->d_lock); | |
2345 | } | |
2346 | rcu_read_unlock(); | |
2347 | ||
2348 | return found; | |
2349 | } | |
2350 | ||
3e7e241f EB |
2351 | /** |
2352 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2353 | * @dir: Directory to search in | |
2354 | * @name: qstr of name we wish to find | |
2355 | * | |
4f522a24 | 2356 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2357 | */ |
2358 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2359 | { | |
3e7e241f EB |
2360 | /* |
2361 | * Check for a fs-specific hash function. Note that we must | |
2362 | * calculate the standard hash first, as the d_op->d_hash() | |
2363 | * routine may choose to leave the hash value unchanged. | |
2364 | */ | |
8387ff25 | 2365 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2366 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2367 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2368 | if (unlikely(err < 0)) |
2369 | return ERR_PTR(err); | |
3e7e241f | 2370 | } |
4f522a24 | 2371 | return d_lookup(dir, name); |
3e7e241f | 2372 | } |
4f522a24 | 2373 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2374 | |
1da177e4 LT |
2375 | /* |
2376 | * When a file is deleted, we have two options: | |
2377 | * - turn this dentry into a negative dentry | |
2378 | * - unhash this dentry and free it. | |
2379 | * | |
2380 | * Usually, we want to just turn this into | |
2381 | * a negative dentry, but if anybody else is | |
2382 | * currently using the dentry or the inode | |
2383 | * we can't do that and we fall back on removing | |
2384 | * it from the hash queues and waiting for | |
2385 | * it to be deleted later when it has no users | |
2386 | */ | |
2387 | ||
2388 | /** | |
2389 | * d_delete - delete a dentry | |
2390 | * @dentry: The dentry to delete | |
2391 | * | |
2392 | * Turn the dentry into a negative dentry if possible, otherwise | |
2393 | * remove it from the hash queues so it can be deleted later | |
2394 | */ | |
2395 | ||
2396 | void d_delete(struct dentry * dentry) | |
2397 | { | |
c19457f0 AV |
2398 | struct inode *inode = dentry->d_inode; |
2399 | int isdir = d_is_dir(dentry); | |
2400 | ||
2401 | spin_lock(&inode->i_lock); | |
2402 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2403 | /* |
2404 | * Are we the only user? | |
2405 | */ | |
98474236 | 2406 | if (dentry->d_lockref.count == 1) { |
13e3c5e5 | 2407 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2408 | dentry_unlink_inode(dentry); |
c19457f0 | 2409 | } else { |
1da177e4 | 2410 | __d_drop(dentry); |
c19457f0 AV |
2411 | spin_unlock(&dentry->d_lock); |
2412 | spin_unlock(&inode->i_lock); | |
2413 | } | |
7a91bf7f | 2414 | fsnotify_nameremove(dentry, isdir); |
1da177e4 | 2415 | } |
ec4f8605 | 2416 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2417 | |
15d3c589 | 2418 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2419 | { |
15d3c589 | 2420 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
61647823 | 2421 | |
1879fd6a | 2422 | hlist_bl_lock(b); |
b07ad996 | 2423 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2424 | hlist_bl_unlock(b); |
1da177e4 LT |
2425 | } |
2426 | ||
2427 | /** | |
2428 | * d_rehash - add an entry back to the hash | |
2429 | * @entry: dentry to add to the hash | |
2430 | * | |
2431 | * Adds a dentry to the hash according to its name. | |
2432 | */ | |
2433 | ||
2434 | void d_rehash(struct dentry * entry) | |
2435 | { | |
1da177e4 | 2436 | spin_lock(&entry->d_lock); |
15d3c589 | 2437 | __d_rehash(entry); |
1da177e4 | 2438 | spin_unlock(&entry->d_lock); |
1da177e4 | 2439 | } |
ec4f8605 | 2440 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2441 | |
84e710da AV |
2442 | static inline unsigned start_dir_add(struct inode *dir) |
2443 | { | |
2444 | ||
2445 | for (;;) { | |
2446 | unsigned n = dir->i_dir_seq; | |
2447 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2448 | return n; | |
2449 | cpu_relax(); | |
2450 | } | |
2451 | } | |
2452 | ||
2453 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2454 | { | |
2455 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2456 | } | |
2457 | ||
d9171b93 AV |
2458 | static void d_wait_lookup(struct dentry *dentry) |
2459 | { | |
2460 | if (d_in_lookup(dentry)) { | |
2461 | DECLARE_WAITQUEUE(wait, current); | |
2462 | add_wait_queue(dentry->d_wait, &wait); | |
2463 | do { | |
2464 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2465 | spin_unlock(&dentry->d_lock); | |
2466 | schedule(); | |
2467 | spin_lock(&dentry->d_lock); | |
2468 | } while (d_in_lookup(dentry)); | |
2469 | } | |
2470 | } | |
2471 | ||
94bdd655 | 2472 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2473 | const struct qstr *name, |
2474 | wait_queue_head_t *wq) | |
94bdd655 | 2475 | { |
94bdd655 | 2476 | unsigned int hash = name->hash; |
94bdd655 AV |
2477 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2478 | struct hlist_bl_node *node; | |
2479 | struct dentry *new = d_alloc(parent, name); | |
2480 | struct dentry *dentry; | |
2481 | unsigned seq, r_seq, d_seq; | |
2482 | ||
2483 | if (unlikely(!new)) | |
2484 | return ERR_PTR(-ENOMEM); | |
2485 | ||
2486 | retry: | |
2487 | rcu_read_lock(); | |
015555fd | 2488 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
94bdd655 AV |
2489 | r_seq = read_seqbegin(&rename_lock); |
2490 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2491 | if (unlikely(dentry)) { | |
2492 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2493 | rcu_read_unlock(); | |
2494 | goto retry; | |
2495 | } | |
2496 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2497 | rcu_read_unlock(); | |
2498 | dput(dentry); | |
2499 | goto retry; | |
2500 | } | |
2501 | rcu_read_unlock(); | |
2502 | dput(new); | |
2503 | return dentry; | |
2504 | } | |
2505 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2506 | rcu_read_unlock(); | |
2507 | goto retry; | |
2508 | } | |
015555fd WD |
2509 | |
2510 | if (unlikely(seq & 1)) { | |
2511 | rcu_read_unlock(); | |
2512 | goto retry; | |
2513 | } | |
2514 | ||
94bdd655 | 2515 | hlist_bl_lock(b); |
8cc07c80 | 2516 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
94bdd655 AV |
2517 | hlist_bl_unlock(b); |
2518 | rcu_read_unlock(); | |
2519 | goto retry; | |
2520 | } | |
94bdd655 AV |
2521 | /* |
2522 | * No changes for the parent since the beginning of d_lookup(). | |
2523 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2524 | * any potential in-lookup matches are going to stay here until | |
2525 | * we unlock the chain. All fields are stable in everything | |
2526 | * we encounter. | |
2527 | */ | |
2528 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2529 | if (dentry->d_name.hash != hash) | |
2530 | continue; | |
2531 | if (dentry->d_parent != parent) | |
2532 | continue; | |
d4c91a8f AV |
2533 | if (!d_same_name(dentry, parent, name)) |
2534 | continue; | |
94bdd655 | 2535 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2536 | /* now we can try to grab a reference */ |
2537 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2538 | rcu_read_unlock(); | |
2539 | goto retry; | |
2540 | } | |
2541 | ||
2542 | rcu_read_unlock(); | |
2543 | /* | |
2544 | * somebody is likely to be still doing lookup for it; | |
2545 | * wait for them to finish | |
2546 | */ | |
d9171b93 AV |
2547 | spin_lock(&dentry->d_lock); |
2548 | d_wait_lookup(dentry); | |
2549 | /* | |
2550 | * it's not in-lookup anymore; in principle we should repeat | |
2551 | * everything from dcache lookup, but it's likely to be what | |
2552 | * d_lookup() would've found anyway. If it is, just return it; | |
2553 | * otherwise we really have to repeat the whole thing. | |
2554 | */ | |
2555 | if (unlikely(dentry->d_name.hash != hash)) | |
2556 | goto mismatch; | |
2557 | if (unlikely(dentry->d_parent != parent)) | |
2558 | goto mismatch; | |
2559 | if (unlikely(d_unhashed(dentry))) | |
2560 | goto mismatch; | |
d4c91a8f AV |
2561 | if (unlikely(!d_same_name(dentry, parent, name))) |
2562 | goto mismatch; | |
d9171b93 AV |
2563 | /* OK, it *is* a hashed match; return it */ |
2564 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2565 | dput(new); |
2566 | return dentry; | |
2567 | } | |
e7d6ef97 | 2568 | rcu_read_unlock(); |
94bdd655 AV |
2569 | /* we can't take ->d_lock here; it's OK, though. */ |
2570 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2571 | new->d_wait = wq; |
94bdd655 AV |
2572 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2573 | hlist_bl_unlock(b); | |
2574 | return new; | |
d9171b93 AV |
2575 | mismatch: |
2576 | spin_unlock(&dentry->d_lock); | |
2577 | dput(dentry); | |
2578 | goto retry; | |
94bdd655 AV |
2579 | } |
2580 | EXPORT_SYMBOL(d_alloc_parallel); | |
2581 | ||
85c7f810 AV |
2582 | void __d_lookup_done(struct dentry *dentry) |
2583 | { | |
94bdd655 AV |
2584 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2585 | dentry->d_name.hash); | |
2586 | hlist_bl_lock(b); | |
85c7f810 | 2587 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2588 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2589 | wake_up_all(dentry->d_wait); |
2590 | dentry->d_wait = NULL; | |
94bdd655 AV |
2591 | hlist_bl_unlock(b); |
2592 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2593 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2594 | } |
2595 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2596 | |
2597 | /* inode->i_lock held if inode is non-NULL */ | |
2598 | ||
2599 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2600 | { | |
84e710da AV |
2601 | struct inode *dir = NULL; |
2602 | unsigned n; | |
0568d705 | 2603 | spin_lock(&dentry->d_lock); |
84e710da AV |
2604 | if (unlikely(d_in_lookup(dentry))) { |
2605 | dir = dentry->d_parent->d_inode; | |
2606 | n = start_dir_add(dir); | |
85c7f810 | 2607 | __d_lookup_done(dentry); |
84e710da | 2608 | } |
ed782b5a | 2609 | if (inode) { |
0568d705 AV |
2610 | unsigned add_flags = d_flags_for_inode(inode); |
2611 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2612 | raw_write_seqcount_begin(&dentry->d_seq); | |
2613 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2614 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2615 | fsnotify_update_flags(dentry); |
ed782b5a | 2616 | } |
15d3c589 | 2617 | __d_rehash(dentry); |
84e710da AV |
2618 | if (dir) |
2619 | end_dir_add(dir, n); | |
0568d705 AV |
2620 | spin_unlock(&dentry->d_lock); |
2621 | if (inode) | |
2622 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2623 | } |
2624 | ||
34d0d19d AV |
2625 | /** |
2626 | * d_add - add dentry to hash queues | |
2627 | * @entry: dentry to add | |
2628 | * @inode: The inode to attach to this dentry | |
2629 | * | |
2630 | * This adds the entry to the hash queues and initializes @inode. | |
2631 | * The entry was actually filled in earlier during d_alloc(). | |
2632 | */ | |
2633 | ||
2634 | void d_add(struct dentry *entry, struct inode *inode) | |
2635 | { | |
b9680917 AV |
2636 | if (inode) { |
2637 | security_d_instantiate(entry, inode); | |
ed782b5a | 2638 | spin_lock(&inode->i_lock); |
b9680917 | 2639 | } |
ed782b5a | 2640 | __d_add(entry, inode); |
34d0d19d AV |
2641 | } |
2642 | EXPORT_SYMBOL(d_add); | |
2643 | ||
668d0cd5 AV |
2644 | /** |
2645 | * d_exact_alias - find and hash an exact unhashed alias | |
2646 | * @entry: dentry to add | |
2647 | * @inode: The inode to go with this dentry | |
2648 | * | |
2649 | * If an unhashed dentry with the same name/parent and desired | |
2650 | * inode already exists, hash and return it. Otherwise, return | |
2651 | * NULL. | |
2652 | * | |
2653 | * Parent directory should be locked. | |
2654 | */ | |
2655 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2656 | { | |
2657 | struct dentry *alias; | |
668d0cd5 AV |
2658 | unsigned int hash = entry->d_name.hash; |
2659 | ||
2660 | spin_lock(&inode->i_lock); | |
2661 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2662 | /* | |
2663 | * Don't need alias->d_lock here, because aliases with | |
2664 | * d_parent == entry->d_parent are not subject to name or | |
2665 | * parent changes, because the parent inode i_mutex is held. | |
2666 | */ | |
2667 | if (alias->d_name.hash != hash) | |
2668 | continue; | |
2669 | if (alias->d_parent != entry->d_parent) | |
2670 | continue; | |
d4c91a8f | 2671 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2672 | continue; |
2673 | spin_lock(&alias->d_lock); | |
2674 | if (!d_unhashed(alias)) { | |
2675 | spin_unlock(&alias->d_lock); | |
2676 | alias = NULL; | |
2677 | } else { | |
2678 | __dget_dlock(alias); | |
15d3c589 | 2679 | __d_rehash(alias); |
668d0cd5 AV |
2680 | spin_unlock(&alias->d_lock); |
2681 | } | |
2682 | spin_unlock(&inode->i_lock); | |
2683 | return alias; | |
2684 | } | |
2685 | spin_unlock(&inode->i_lock); | |
2686 | return NULL; | |
2687 | } | |
2688 | EXPORT_SYMBOL(d_exact_alias); | |
2689 | ||
fb2d5b86 NP |
2690 | /** |
2691 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2692 | * @dentry: dentry to be updated | |
2693 | * @name: new name | |
2694 | * | |
2695 | * Update a case insensitive dentry with new case of name. | |
2696 | * | |
2697 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2698 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2699 | * lengths). | |
2700 | * | |
2701 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2702 | * keep renames and concurrent inserts, and readdir(2) away). | |
2703 | */ | |
9aba36de | 2704 | void dentry_update_name_case(struct dentry *dentry, const struct qstr *name) |
fb2d5b86 | 2705 | { |
5955102c | 2706 | BUG_ON(!inode_is_locked(dentry->d_parent->d_inode)); |
fb2d5b86 NP |
2707 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2708 | ||
fb2d5b86 | 2709 | spin_lock(&dentry->d_lock); |
31e6b01f | 2710 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2711 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2712 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2713 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2714 | } |
2715 | EXPORT_SYMBOL(dentry_update_name_case); | |
2716 | ||
8d85b484 | 2717 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2718 | { |
8d85b484 AV |
2719 | if (unlikely(dname_external(target))) { |
2720 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2721 | /* |
2722 | * Both external: swap the pointers | |
2723 | */ | |
9a8d5bb4 | 2724 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2725 | } else { |
2726 | /* | |
2727 | * dentry:internal, target:external. Steal target's | |
2728 | * storage and make target internal. | |
2729 | */ | |
321bcf92 BF |
2730 | memcpy(target->d_iname, dentry->d_name.name, |
2731 | dentry->d_name.len + 1); | |
1da177e4 LT |
2732 | dentry->d_name.name = target->d_name.name; |
2733 | target->d_name.name = target->d_iname; | |
2734 | } | |
2735 | } else { | |
8d85b484 | 2736 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2737 | /* |
2738 | * dentry:external, target:internal. Give dentry's | |
2739 | * storage to target and make dentry internal | |
2740 | */ | |
2741 | memcpy(dentry->d_iname, target->d_name.name, | |
2742 | target->d_name.len + 1); | |
2743 | target->d_name.name = dentry->d_name.name; | |
2744 | dentry->d_name.name = dentry->d_iname; | |
2745 | } else { | |
2746 | /* | |
da1ce067 | 2747 | * Both are internal. |
1da177e4 | 2748 | */ |
da1ce067 MS |
2749 | unsigned int i; |
2750 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2751 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2752 | swap(((long *) &dentry->d_iname)[i], | |
2753 | ((long *) &target->d_iname)[i]); | |
2754 | } | |
1da177e4 LT |
2755 | } |
2756 | } | |
a28ddb87 | 2757 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2758 | } |
2759 | ||
8d85b484 AV |
2760 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2761 | { | |
2762 | struct external_name *old_name = NULL; | |
2763 | if (unlikely(dname_external(dentry))) | |
2764 | old_name = external_name(dentry); | |
2765 | if (unlikely(dname_external(target))) { | |
2766 | atomic_inc(&external_name(target)->u.count); | |
2767 | dentry->d_name = target->d_name; | |
2768 | } else { | |
2769 | memcpy(dentry->d_iname, target->d_name.name, | |
2770 | target->d_name.len + 1); | |
2771 | dentry->d_name.name = dentry->d_iname; | |
2772 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2773 | } | |
2774 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2775 | kfree_rcu(old_name, u.head); | |
2776 | } | |
2777 | ||
2fd6b7f5 NP |
2778 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2779 | { | |
2780 | /* | |
2781 | * XXXX: do we really need to take target->d_lock? | |
2782 | */ | |
2783 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2784 | spin_lock(&target->d_parent->d_lock); | |
2785 | else { | |
2786 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2787 | spin_lock(&dentry->d_parent->d_lock); | |
2788 | spin_lock_nested(&target->d_parent->d_lock, | |
2789 | DENTRY_D_LOCK_NESTED); | |
2790 | } else { | |
2791 | spin_lock(&target->d_parent->d_lock); | |
2792 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2793 | DENTRY_D_LOCK_NESTED); | |
2794 | } | |
2795 | } | |
2796 | if (target < dentry) { | |
2797 | spin_lock_nested(&target->d_lock, 2); | |
2798 | spin_lock_nested(&dentry->d_lock, 3); | |
2799 | } else { | |
2800 | spin_lock_nested(&dentry->d_lock, 2); | |
2801 | spin_lock_nested(&target->d_lock, 3); | |
2802 | } | |
2803 | } | |
2804 | ||
986c0194 | 2805 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2806 | { |
2807 | if (target->d_parent != dentry->d_parent) | |
2808 | spin_unlock(&dentry->d_parent->d_lock); | |
2809 | if (target->d_parent != target) | |
2810 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2811 | spin_unlock(&target->d_lock); |
2812 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2813 | } |
2814 | ||
1da177e4 | 2815 | /* |
2fd6b7f5 NP |
2816 | * When switching names, the actual string doesn't strictly have to |
2817 | * be preserved in the target - because we're dropping the target | |
2818 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2819 | * the new name before we switch, unless we are going to rehash |
2820 | * it. Note that if we *do* unhash the target, we are not allowed | |
2821 | * to rehash it without giving it a new name/hash key - whether | |
2822 | * we swap or overwrite the names here, resulting name won't match | |
2823 | * the reality in filesystem; it's only there for d_path() purposes. | |
2824 | * Note that all of this is happening under rename_lock, so the | |
2825 | * any hash lookup seeing it in the middle of manipulations will | |
2826 | * be discarded anyway. So we do not care what happens to the hash | |
2827 | * key in that case. | |
1da177e4 | 2828 | */ |
9eaef27b | 2829 | /* |
18367501 | 2830 | * __d_move - move a dentry |
1da177e4 LT |
2831 | * @dentry: entry to move |
2832 | * @target: new dentry | |
da1ce067 | 2833 | * @exchange: exchange the two dentries |
1da177e4 LT |
2834 | * |
2835 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2836 | * dcache entries should not be moved in this way. Caller must hold |
2837 | * rename_lock, the i_mutex of the source and target directories, | |
2838 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2839 | */ |
da1ce067 MS |
2840 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2841 | bool exchange) | |
1da177e4 | 2842 | { |
84e710da AV |
2843 | struct inode *dir = NULL; |
2844 | unsigned n; | |
1da177e4 LT |
2845 | if (!dentry->d_inode) |
2846 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2847 | ||
2fd6b7f5 NP |
2848 | BUG_ON(d_ancestor(dentry, target)); |
2849 | BUG_ON(d_ancestor(target, dentry)); | |
2850 | ||
2fd6b7f5 | 2851 | dentry_lock_for_move(dentry, target); |
84e710da AV |
2852 | if (unlikely(d_in_lookup(target))) { |
2853 | dir = target->d_parent->d_inode; | |
2854 | n = start_dir_add(dir); | |
85c7f810 | 2855 | __d_lookup_done(target); |
84e710da | 2856 | } |
1da177e4 | 2857 | |
31e6b01f | 2858 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2859 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2860 | |
15d3c589 | 2861 | /* unhash both */ |
0632a9ac AV |
2862 | if (!d_unhashed(dentry)) |
2863 | ___d_drop(dentry); | |
2864 | if (!d_unhashed(target)) | |
2865 | ___d_drop(target); | |
1da177e4 | 2866 | |
1da177e4 | 2867 | /* Switch the names.. */ |
8d85b484 AV |
2868 | if (exchange) |
2869 | swap_names(dentry, target); | |
2870 | else | |
2871 | copy_name(dentry, target); | |
1da177e4 | 2872 | |
15d3c589 AV |
2873 | /* rehash in new place(s) */ |
2874 | __d_rehash(dentry); | |
2875 | if (exchange) | |
2876 | __d_rehash(target); | |
61647823 N |
2877 | else |
2878 | target->d_hash.pprev = NULL; | |
15d3c589 | 2879 | |
63cf427a | 2880 | /* ... and switch them in the tree */ |
1da177e4 | 2881 | if (IS_ROOT(dentry)) { |
63cf427a | 2882 | /* splicing a tree */ |
3d56c25e | 2883 | dentry->d_flags |= DCACHE_RCUACCESS; |
1da177e4 LT |
2884 | dentry->d_parent = target->d_parent; |
2885 | target->d_parent = target; | |
946e51f2 AV |
2886 | list_del_init(&target->d_child); |
2887 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
1da177e4 | 2888 | } else { |
63cf427a | 2889 | /* swapping two dentries */ |
9a8d5bb4 | 2890 | swap(dentry->d_parent, target->d_parent); |
946e51f2 AV |
2891 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
2892 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
63cf427a | 2893 | if (exchange) |
affda484 AV |
2894 | fsnotify_update_flags(target); |
2895 | fsnotify_update_flags(dentry); | |
1da177e4 LT |
2896 | } |
2897 | ||
31e6b01f NP |
2898 | write_seqcount_end(&target->d_seq); |
2899 | write_seqcount_end(&dentry->d_seq); | |
2900 | ||
84e710da AV |
2901 | if (dir) |
2902 | end_dir_add(dir, n); | |
986c0194 | 2903 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2904 | } |
2905 | ||
2906 | /* | |
2907 | * d_move - move a dentry | |
2908 | * @dentry: entry to move | |
2909 | * @target: new dentry | |
2910 | * | |
2911 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2912 | * dcache entries should not be moved in this way. See the locking |
2913 | * requirements for __d_move. | |
18367501 AV |
2914 | */ |
2915 | void d_move(struct dentry *dentry, struct dentry *target) | |
2916 | { | |
2917 | write_seqlock(&rename_lock); | |
da1ce067 | 2918 | __d_move(dentry, target, false); |
1da177e4 | 2919 | write_sequnlock(&rename_lock); |
9eaef27b | 2920 | } |
ec4f8605 | 2921 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2922 | |
da1ce067 MS |
2923 | /* |
2924 | * d_exchange - exchange two dentries | |
2925 | * @dentry1: first dentry | |
2926 | * @dentry2: second dentry | |
2927 | */ | |
2928 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2929 | { | |
2930 | write_seqlock(&rename_lock); | |
2931 | ||
2932 | WARN_ON(!dentry1->d_inode); | |
2933 | WARN_ON(!dentry2->d_inode); | |
2934 | WARN_ON(IS_ROOT(dentry1)); | |
2935 | WARN_ON(IS_ROOT(dentry2)); | |
2936 | ||
2937 | __d_move(dentry1, dentry2, true); | |
2938 | ||
2939 | write_sequnlock(&rename_lock); | |
2940 | } | |
2941 | ||
e2761a11 OH |
2942 | /** |
2943 | * d_ancestor - search for an ancestor | |
2944 | * @p1: ancestor dentry | |
2945 | * @p2: child dentry | |
2946 | * | |
2947 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2948 | * an ancestor of p2, else NULL. | |
9eaef27b | 2949 | */ |
e2761a11 | 2950 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2951 | { |
2952 | struct dentry *p; | |
2953 | ||
871c0067 | 2954 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2955 | if (p->d_parent == p1) |
e2761a11 | 2956 | return p; |
9eaef27b | 2957 | } |
e2761a11 | 2958 | return NULL; |
9eaef27b TM |
2959 | } |
2960 | ||
2961 | /* | |
2962 | * This helper attempts to cope with remotely renamed directories | |
2963 | * | |
2964 | * It assumes that the caller is already holding | |
a03e283b | 2965 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2966 | * |
2967 | * Note: If ever the locking in lock_rename() changes, then please | |
2968 | * remember to update this too... | |
9eaef27b | 2969 | */ |
b5ae6b15 | 2970 | static int __d_unalias(struct inode *inode, |
873feea0 | 2971 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2972 | { |
9902af79 AV |
2973 | struct mutex *m1 = NULL; |
2974 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2975 | int ret = -ESTALE; |
9eaef27b TM |
2976 | |
2977 | /* If alias and dentry share a parent, then no extra locks required */ | |
2978 | if (alias->d_parent == dentry->d_parent) | |
2979 | goto out_unalias; | |
2980 | ||
9eaef27b | 2981 | /* See lock_rename() */ |
9eaef27b TM |
2982 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2983 | goto out_err; | |
2984 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2985 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2986 | goto out_err; |
9902af79 | 2987 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 2988 | out_unalias: |
8ed936b5 | 2989 | __d_move(alias, dentry, false); |
b5ae6b15 | 2990 | ret = 0; |
9eaef27b | 2991 | out_err: |
9eaef27b | 2992 | if (m2) |
9902af79 | 2993 | up_read(m2); |
9eaef27b TM |
2994 | if (m1) |
2995 | mutex_unlock(m1); | |
2996 | return ret; | |
2997 | } | |
2998 | ||
3f70bd51 BF |
2999 | /** |
3000 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
3001 | * @inode: the inode which may have a disconnected dentry | |
3002 | * @dentry: a negative dentry which we want to point to the inode. | |
3003 | * | |
da093a9b BF |
3004 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
3005 | * place of the given dentry and return it, else simply d_add the inode | |
3006 | * to the dentry and return NULL. | |
3f70bd51 | 3007 | * |
908790fa BF |
3008 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
3009 | * we should error out: directories can't have multiple aliases. | |
3010 | * | |
3f70bd51 BF |
3011 | * This is needed in the lookup routine of any filesystem that is exportable |
3012 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
3013 | * | |
3014 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
3015 | * is returned. This matches the expected return value of ->lookup. | |
3016 | * | |
3017 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
3018 | * In that case, we know that the inode will be a regular file, and also this | |
3019 | * will only occur during atomic_open. So we need to check for the dentry | |
3020 | * being already hashed only in the final case. | |
3021 | */ | |
3022 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3023 | { | |
3f70bd51 BF |
3024 | if (IS_ERR(inode)) |
3025 | return ERR_CAST(inode); | |
3026 | ||
770bfad8 DH |
3027 | BUG_ON(!d_unhashed(dentry)); |
3028 | ||
de689f5e | 3029 | if (!inode) |
b5ae6b15 | 3030 | goto out; |
de689f5e | 3031 | |
b9680917 | 3032 | security_d_instantiate(dentry, inode); |
873feea0 | 3033 | spin_lock(&inode->i_lock); |
9eaef27b | 3034 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
3035 | struct dentry *new = __d_find_any_alias(inode); |
3036 | if (unlikely(new)) { | |
a03e283b EB |
3037 | /* The reference to new ensures it remains an alias */ |
3038 | spin_unlock(&inode->i_lock); | |
18367501 | 3039 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3040 | if (unlikely(d_ancestor(new, dentry))) { |
3041 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3042 | dput(new); |
3043 | new = ERR_PTR(-ELOOP); | |
3044 | pr_warn_ratelimited( | |
3045 | "VFS: Lookup of '%s' in %s %s" | |
3046 | " would have caused loop\n", | |
3047 | dentry->d_name.name, | |
3048 | inode->i_sb->s_type->name, | |
3049 | inode->i_sb->s_id); | |
3050 | } else if (!IS_ROOT(new)) { | |
3051 | int err = __d_unalias(inode, dentry, new); | |
18367501 | 3052 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3053 | if (err) { |
3054 | dput(new); | |
3055 | new = ERR_PTR(err); | |
3056 | } | |
18367501 | 3057 | } else { |
b5ae6b15 AV |
3058 | __d_move(new, dentry, false); |
3059 | write_sequnlock(&rename_lock); | |
dd179946 | 3060 | } |
b5ae6b15 AV |
3061 | iput(inode); |
3062 | return new; | |
9eaef27b | 3063 | } |
770bfad8 | 3064 | } |
b5ae6b15 | 3065 | out: |
ed782b5a | 3066 | __d_add(dentry, inode); |
b5ae6b15 | 3067 | return NULL; |
770bfad8 | 3068 | } |
b5ae6b15 | 3069 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3070 | |
1da177e4 LT |
3071 | /* |
3072 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3073 | * | |
3074 | * Trivially implemented using the dcache structure | |
3075 | */ | |
3076 | ||
3077 | /** | |
3078 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3079 | * @new_dentry: new dentry | |
3080 | * @old_dentry: old dentry | |
3081 | * | |
a6e5787f YB |
3082 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3083 | * Returns false otherwise. | |
1da177e4 LT |
3084 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3085 | */ | |
3086 | ||
a6e5787f | 3087 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3088 | { |
a6e5787f | 3089 | bool result; |
949854d0 | 3090 | unsigned seq; |
1da177e4 | 3091 | |
e2761a11 | 3092 | if (new_dentry == old_dentry) |
a6e5787f | 3093 | return true; |
e2761a11 | 3094 | |
e2761a11 | 3095 | do { |
1da177e4 | 3096 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3097 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3098 | /* |
3099 | * Need rcu_readlock to protect against the d_parent trashing | |
3100 | * due to d_move | |
3101 | */ | |
3102 | rcu_read_lock(); | |
e2761a11 | 3103 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3104 | result = true; |
e2761a11 | 3105 | else |
a6e5787f | 3106 | result = false; |
949854d0 | 3107 | rcu_read_unlock(); |
1da177e4 | 3108 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3109 | |
3110 | return result; | |
3111 | } | |
e8f9e5b7 | 3112 | EXPORT_SYMBOL(is_subdir); |
1da177e4 | 3113 | |
db14fc3a | 3114 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3115 | { |
db14fc3a MS |
3116 | struct dentry *root = data; |
3117 | if (dentry != root) { | |
3118 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3119 | return D_WALK_SKIP; | |
1da177e4 | 3120 | |
01ddc4ed MS |
3121 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3122 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3123 | dentry->d_lockref.count--; | |
3124 | } | |
1da177e4 | 3125 | } |
db14fc3a MS |
3126 | return D_WALK_CONTINUE; |
3127 | } | |
58db63d0 | 3128 | |
db14fc3a MS |
3129 | void d_genocide(struct dentry *parent) |
3130 | { | |
3131 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3132 | } |
3133 | ||
60545d0d | 3134 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3135 | { |
60545d0d AV |
3136 | inode_dec_link_count(inode); |
3137 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3138 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3139 | !d_unlinked(dentry)); |
3140 | spin_lock(&dentry->d_parent->d_lock); | |
3141 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3142 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3143 | (unsigned long long)inode->i_ino); | |
3144 | spin_unlock(&dentry->d_lock); | |
3145 | spin_unlock(&dentry->d_parent->d_lock); | |
3146 | d_instantiate(dentry, inode); | |
1da177e4 | 3147 | } |
60545d0d | 3148 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3149 | |
3150 | static __initdata unsigned long dhash_entries; | |
3151 | static int __init set_dhash_entries(char *str) | |
3152 | { | |
3153 | if (!str) | |
3154 | return 0; | |
3155 | dhash_entries = simple_strtoul(str, &str, 0); | |
3156 | return 1; | |
3157 | } | |
3158 | __setup("dhash_entries=", set_dhash_entries); | |
3159 | ||
3160 | static void __init dcache_init_early(void) | |
3161 | { | |
1da177e4 LT |
3162 | /* If hashes are distributed across NUMA nodes, defer |
3163 | * hash allocation until vmalloc space is available. | |
3164 | */ | |
3165 | if (hashdist) | |
3166 | return; | |
3167 | ||
3168 | dentry_hashtable = | |
3169 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3170 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3171 | dhash_entries, |
3172 | 13, | |
3d375d78 | 3173 | HASH_EARLY | HASH_ZERO, |
1da177e4 | 3174 | &d_hash_shift, |
b35d786b | 3175 | NULL, |
31fe62b9 | 3176 | 0, |
1da177e4 | 3177 | 0); |
854d3e63 | 3178 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3179 | } |
3180 | ||
74bf17cf | 3181 | static void __init dcache_init(void) |
1da177e4 | 3182 | { |
3d375d78 | 3183 | /* |
1da177e4 LT |
3184 | * A constructor could be added for stable state like the lists, |
3185 | * but it is probably not worth it because of the cache nature | |
3d375d78 | 3186 | * of the dcache. |
1da177e4 | 3187 | */ |
80344266 DW |
3188 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, |
3189 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, | |
3190 | d_iname); | |
1da177e4 LT |
3191 | |
3192 | /* Hash may have been set up in dcache_init_early */ | |
3193 | if (!hashdist) | |
3194 | return; | |
3195 | ||
3196 | dentry_hashtable = | |
3197 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3198 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3199 | dhash_entries, |
3200 | 13, | |
3d375d78 | 3201 | HASH_ZERO, |
1da177e4 | 3202 | &d_hash_shift, |
b35d786b | 3203 | NULL, |
31fe62b9 | 3204 | 0, |
1da177e4 | 3205 | 0); |
854d3e63 | 3206 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3207 | } |
3208 | ||
3209 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3210 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3211 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3212 | |
1da177e4 LT |
3213 | EXPORT_SYMBOL(d_genocide); |
3214 | ||
1da177e4 LT |
3215 | void __init vfs_caches_init_early(void) |
3216 | { | |
6916363f SAS |
3217 | int i; |
3218 | ||
3219 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3220 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3221 | ||
1da177e4 LT |
3222 | dcache_init_early(); |
3223 | inode_init_early(); | |
3224 | } | |
3225 | ||
4248b0da | 3226 | void __init vfs_caches_init(void) |
1da177e4 | 3227 | { |
6a9b8820 DW |
3228 | names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0, |
3229 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL); | |
1da177e4 | 3230 | |
74bf17cf DC |
3231 | dcache_init(); |
3232 | inode_init(); | |
4248b0da MG |
3233 | files_init(); |
3234 | files_maxfiles_init(); | |
74bf17cf | 3235 | mnt_init(); |
1da177e4 LT |
3236 | bdev_cache_init(); |
3237 | chrdev_init(); | |
3238 | } |