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: | |
076515fc | 70 | * arbitrary, since it's serialized on rename_lock |
789680d1 | 71 | */ |
fa3536cc | 72 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
73 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
74 | ||
74c3cbe3 | 75 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 76 | |
949854d0 | 77 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 78 | |
e18b890b | 79 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 80 | |
cdf01226 DH |
81 | const struct qstr empty_name = QSTR_INIT("", 0); |
82 | EXPORT_SYMBOL(empty_name); | |
83 | const struct qstr slash_name = QSTR_INIT("/", 1); | |
84 | EXPORT_SYMBOL(slash_name); | |
85 | ||
1da177e4 LT |
86 | /* |
87 | * This is the single most critical data structure when it comes | |
88 | * to the dcache: the hashtable for lookups. Somebody should try | |
89 | * to make this good - I've just made it work. | |
90 | * | |
91 | * This hash-function tries to avoid losing too many bits of hash | |
92 | * information, yet avoid using a prime hash-size or similar. | |
93 | */ | |
1da177e4 | 94 | |
fa3536cc | 95 | static unsigned int d_hash_shift __read_mostly; |
ceb5bdc2 | 96 | |
b07ad996 | 97 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 98 | |
8387ff25 | 99 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 100 | { |
854d3e63 | 101 | return dentry_hashtable + (hash >> d_hash_shift); |
ceb5bdc2 NP |
102 | } |
103 | ||
94bdd655 AV |
104 | #define IN_LOOKUP_SHIFT 10 |
105 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
106 | ||
107 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
108 | unsigned int hash) | |
109 | { | |
110 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
111 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
112 | } | |
113 | ||
114 | ||
1da177e4 LT |
115 | /* Statistics gathering. */ |
116 | struct dentry_stat_t dentry_stat = { | |
117 | .age_limit = 45, | |
118 | }; | |
119 | ||
3942c07c | 120 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 121 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
122 | |
123 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
124 | |
125 | /* | |
126 | * Here we resort to our own counters instead of using generic per-cpu counters | |
127 | * for consistency with what the vfs inode code does. We are expected to harvest | |
128 | * better code and performance by having our own specialized counters. | |
129 | * | |
130 | * Please note that the loop is done over all possible CPUs, not over all online | |
131 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
132 | * on and off. If one of them goes off, we will just keep their counters. | |
133 | * | |
134 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
135 | * please update all vfs counters to match. | |
136 | */ | |
3942c07c | 137 | static long get_nr_dentry(void) |
3e880fb5 NP |
138 | { |
139 | int i; | |
3942c07c | 140 | long sum = 0; |
3e880fb5 NP |
141 | for_each_possible_cpu(i) |
142 | sum += per_cpu(nr_dentry, i); | |
143 | return sum < 0 ? 0 : sum; | |
144 | } | |
145 | ||
62d36c77 DC |
146 | static long get_nr_dentry_unused(void) |
147 | { | |
148 | int i; | |
149 | long sum = 0; | |
150 | for_each_possible_cpu(i) | |
151 | sum += per_cpu(nr_dentry_unused, i); | |
152 | return sum < 0 ? 0 : sum; | |
153 | } | |
154 | ||
1f7e0616 | 155 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
156 | size_t *lenp, loff_t *ppos) |
157 | { | |
3e880fb5 | 158 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 159 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 160 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
161 | } |
162 | #endif | |
163 | ||
5483f18e LT |
164 | /* |
165 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
166 | * The strings are both count bytes long, and count is non-zero. | |
167 | */ | |
e419b4cc LT |
168 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
169 | ||
170 | #include <asm/word-at-a-time.h> | |
171 | /* | |
172 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
173 | * aligned allocation for this particular component. We don't | |
174 | * strictly need the load_unaligned_zeropad() safety, but it | |
175 | * doesn't hurt either. | |
176 | * | |
177 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
178 | * need the careful unaligned handling. | |
179 | */ | |
94753db5 | 180 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 181 | { |
bfcfaa77 | 182 | unsigned long a,b,mask; |
bfcfaa77 LT |
183 | |
184 | for (;;) { | |
bfe7aa6c | 185 | a = read_word_at_a_time(cs); |
e419b4cc | 186 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
187 | if (tcount < sizeof(unsigned long)) |
188 | break; | |
189 | if (unlikely(a != b)) | |
190 | return 1; | |
191 | cs += sizeof(unsigned long); | |
192 | ct += sizeof(unsigned long); | |
193 | tcount -= sizeof(unsigned long); | |
194 | if (!tcount) | |
195 | return 0; | |
196 | } | |
a5c21dce | 197 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 198 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
199 | } |
200 | ||
bfcfaa77 | 201 | #else |
e419b4cc | 202 | |
94753db5 | 203 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 204 | { |
5483f18e LT |
205 | do { |
206 | if (*cs != *ct) | |
207 | return 1; | |
208 | cs++; | |
209 | ct++; | |
210 | tcount--; | |
211 | } while (tcount); | |
212 | return 0; | |
213 | } | |
214 | ||
e419b4cc LT |
215 | #endif |
216 | ||
94753db5 LT |
217 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
218 | { | |
94753db5 LT |
219 | /* |
220 | * Be careful about RCU walk racing with rename: | |
506458ef | 221 | * use 'READ_ONCE' to fetch the name pointer. |
94753db5 LT |
222 | * |
223 | * NOTE! Even if a rename will mean that the length | |
224 | * was not loaded atomically, we don't care. The | |
225 | * RCU walk will check the sequence count eventually, | |
226 | * and catch it. And we won't overrun the buffer, | |
227 | * because we're reading the name pointer atomically, | |
228 | * and a dentry name is guaranteed to be properly | |
229 | * terminated with a NUL byte. | |
230 | * | |
231 | * End result: even if 'len' is wrong, we'll exit | |
232 | * early because the data cannot match (there can | |
233 | * be no NUL in the ct/tcount data) | |
234 | */ | |
506458ef | 235 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
ae0a843c | 236 | |
6326c71f | 237 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
238 | } |
239 | ||
8d85b484 AV |
240 | struct external_name { |
241 | union { | |
242 | atomic_t count; | |
243 | struct rcu_head head; | |
244 | } u; | |
245 | unsigned char name[]; | |
246 | }; | |
247 | ||
248 | static inline struct external_name *external_name(struct dentry *dentry) | |
249 | { | |
250 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
251 | } | |
252 | ||
9c82ab9c | 253 | static void __d_free(struct rcu_head *head) |
1da177e4 | 254 | { |
9c82ab9c CH |
255 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
256 | ||
8d85b484 AV |
257 | kmem_cache_free(dentry_cache, dentry); |
258 | } | |
259 | ||
f1782c9b RG |
260 | static void __d_free_external_name(struct rcu_head *head) |
261 | { | |
262 | struct external_name *name = container_of(head, struct external_name, | |
263 | u.head); | |
264 | ||
265 | mod_node_page_state(page_pgdat(virt_to_page(name)), | |
266 | NR_INDIRECTLY_RECLAIMABLE_BYTES, | |
267 | -ksize(name)); | |
268 | ||
269 | kfree(name); | |
270 | } | |
271 | ||
8d85b484 AV |
272 | static void __d_free_external(struct rcu_head *head) |
273 | { | |
274 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
f1782c9b RG |
275 | |
276 | __d_free_external_name(&external_name(dentry)->u.head); | |
277 | ||
278 | kmem_cache_free(dentry_cache, dentry); | |
1da177e4 LT |
279 | } |
280 | ||
810bb172 AV |
281 | static inline int dname_external(const struct dentry *dentry) |
282 | { | |
283 | return dentry->d_name.name != dentry->d_iname; | |
284 | } | |
285 | ||
49d31c2f AV |
286 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
287 | { | |
288 | spin_lock(&dentry->d_lock); | |
289 | if (unlikely(dname_external(dentry))) { | |
290 | struct external_name *p = external_name(dentry); | |
291 | atomic_inc(&p->u.count); | |
292 | spin_unlock(&dentry->d_lock); | |
293 | name->name = p->name; | |
294 | } else { | |
295 | memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN); | |
296 | spin_unlock(&dentry->d_lock); | |
297 | name->name = name->inline_name; | |
298 | } | |
299 | } | |
300 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
301 | ||
302 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
303 | { | |
304 | if (unlikely(name->name != name->inline_name)) { | |
305 | struct external_name *p; | |
306 | p = container_of(name->name, struct external_name, name[0]); | |
307 | if (unlikely(atomic_dec_and_test(&p->u.count))) | |
f1782c9b | 308 | call_rcu(&p->u.head, __d_free_external_name); |
49d31c2f AV |
309 | } |
310 | } | |
311 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
312 | ||
4bf46a27 DH |
313 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
314 | struct inode *inode, | |
315 | unsigned type_flags) | |
316 | { | |
317 | unsigned flags; | |
318 | ||
319 | dentry->d_inode = inode; | |
4bf46a27 DH |
320 | flags = READ_ONCE(dentry->d_flags); |
321 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
322 | flags |= type_flags; | |
323 | WRITE_ONCE(dentry->d_flags, flags); | |
324 | } | |
325 | ||
4bf46a27 DH |
326 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
327 | { | |
328 | unsigned flags = READ_ONCE(dentry->d_flags); | |
329 | ||
330 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
331 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 DH |
332 | dentry->d_inode = NULL; |
333 | } | |
334 | ||
b4f0354e AV |
335 | static void dentry_free(struct dentry *dentry) |
336 | { | |
946e51f2 | 337 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
338 | if (unlikely(dname_external(dentry))) { |
339 | struct external_name *p = external_name(dentry); | |
340 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
341 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
342 | return; | |
343 | } | |
344 | } | |
b4f0354e AV |
345 | /* if dentry was never visible to RCU, immediate free is OK */ |
346 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
347 | __d_free(&dentry->d_u.d_rcu); | |
348 | else | |
349 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
350 | } | |
351 | ||
1da177e4 LT |
352 | /* |
353 | * Release the dentry's inode, using the filesystem | |
550dce01 | 354 | * d_iput() operation if defined. |
31e6b01f NP |
355 | */ |
356 | static void dentry_unlink_inode(struct dentry * dentry) | |
357 | __releases(dentry->d_lock) | |
873feea0 | 358 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
359 | { |
360 | struct inode *inode = dentry->d_inode; | |
550dce01 | 361 | bool hashed = !d_unhashed(dentry); |
a528aca7 | 362 | |
550dce01 AV |
363 | if (hashed) |
364 | raw_write_seqcount_begin(&dentry->d_seq); | |
4bf46a27 | 365 | __d_clear_type_and_inode(dentry); |
946e51f2 | 366 | hlist_del_init(&dentry->d_u.d_alias); |
550dce01 AV |
367 | if (hashed) |
368 | raw_write_seqcount_end(&dentry->d_seq); | |
31e6b01f | 369 | spin_unlock(&dentry->d_lock); |
873feea0 | 370 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
371 | if (!inode->i_nlink) |
372 | fsnotify_inoderemove(inode); | |
373 | if (dentry->d_op && dentry->d_op->d_iput) | |
374 | dentry->d_op->d_iput(dentry, inode); | |
375 | else | |
376 | iput(inode); | |
377 | } | |
378 | ||
89dc77bc LT |
379 | /* |
380 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
381 | * is in use - which includes both the "real" per-superblock | |
382 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
383 | * | |
384 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
385 | * on the shrink list (ie not on the superblock LRU list). | |
386 | * | |
387 | * The per-cpu "nr_dentry_unused" counters are updated with | |
388 | * the DCACHE_LRU_LIST bit. | |
389 | * | |
390 | * These helper functions make sure we always follow the | |
391 | * rules. d_lock must be held by the caller. | |
392 | */ | |
393 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
394 | static void d_lru_add(struct dentry *dentry) | |
395 | { | |
396 | D_FLAG_VERIFY(dentry, 0); | |
397 | dentry->d_flags |= DCACHE_LRU_LIST; | |
398 | this_cpu_inc(nr_dentry_unused); | |
399 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
400 | } | |
401 | ||
402 | static void d_lru_del(struct dentry *dentry) | |
403 | { | |
404 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
405 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
406 | this_cpu_dec(nr_dentry_unused); | |
407 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
408 | } | |
409 | ||
410 | static void d_shrink_del(struct dentry *dentry) | |
411 | { | |
412 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
413 | list_del_init(&dentry->d_lru); | |
414 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
415 | this_cpu_dec(nr_dentry_unused); | |
416 | } | |
417 | ||
418 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
419 | { | |
420 | D_FLAG_VERIFY(dentry, 0); | |
421 | list_add(&dentry->d_lru, list); | |
422 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
423 | this_cpu_inc(nr_dentry_unused); | |
424 | } | |
425 | ||
426 | /* | |
427 | * These can only be called under the global LRU lock, ie during the | |
428 | * callback for freeing the LRU list. "isolate" removes it from the | |
429 | * LRU lists entirely, while shrink_move moves it to the indicated | |
430 | * private list. | |
431 | */ | |
3f97b163 | 432 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
433 | { |
434 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
435 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
436 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 437 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
438 | } |
439 | ||
3f97b163 VD |
440 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
441 | struct list_head *list) | |
89dc77bc LT |
442 | { |
443 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
444 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 445 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
446 | } |
447 | ||
789680d1 NP |
448 | /** |
449 | * d_drop - drop a dentry | |
450 | * @dentry: dentry to drop | |
451 | * | |
452 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
453 | * be found through a VFS lookup any more. Note that this is different from | |
454 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
455 | * possible, giving a successful _negative_ lookup, while d_drop will | |
456 | * just make the cache lookup fail. | |
457 | * | |
458 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
459 | * reason (NFS timeouts or autofs deletes). | |
460 | * | |
61647823 N |
461 | * __d_drop requires dentry->d_lock |
462 | * ___d_drop doesn't mark dentry as "unhashed" | |
463 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
789680d1 | 464 | */ |
61647823 | 465 | static void ___d_drop(struct dentry *dentry) |
789680d1 | 466 | { |
0632a9ac AV |
467 | struct hlist_bl_head *b; |
468 | /* | |
469 | * Hashed dentries are normally on the dentry hashtable, | |
470 | * with the exception of those newly allocated by | |
471 | * d_obtain_root, which are always IS_ROOT: | |
472 | */ | |
473 | if (unlikely(IS_ROOT(dentry))) | |
474 | b = &dentry->d_sb->s_roots; | |
475 | else | |
476 | b = d_hash(dentry->d_name.hash); | |
b61625d2 | 477 | |
0632a9ac AV |
478 | hlist_bl_lock(b); |
479 | __hlist_bl_del(&dentry->d_hash); | |
480 | hlist_bl_unlock(b); | |
789680d1 | 481 | } |
61647823 N |
482 | |
483 | void __d_drop(struct dentry *dentry) | |
484 | { | |
0632a9ac AV |
485 | if (!d_unhashed(dentry)) { |
486 | ___d_drop(dentry); | |
487 | dentry->d_hash.pprev = NULL; | |
488 | write_seqcount_invalidate(&dentry->d_seq); | |
489 | } | |
61647823 | 490 | } |
789680d1 NP |
491 | EXPORT_SYMBOL(__d_drop); |
492 | ||
493 | void d_drop(struct dentry *dentry) | |
494 | { | |
789680d1 NP |
495 | spin_lock(&dentry->d_lock); |
496 | __d_drop(dentry); | |
497 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
498 | } |
499 | EXPORT_SYMBOL(d_drop); | |
500 | ||
ba65dc5e AV |
501 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
502 | { | |
503 | struct dentry *next; | |
504 | /* | |
505 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
506 | * attached to the dentry tree | |
507 | */ | |
508 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
509 | if (unlikely(list_empty(&dentry->d_child))) | |
510 | return; | |
511 | __list_del_entry(&dentry->d_child); | |
512 | /* | |
513 | * Cursors can move around the list of children. While we'd been | |
514 | * a normal list member, it didn't matter - ->d_child.next would've | |
515 | * been updated. However, from now on it won't be and for the | |
516 | * things like d_walk() it might end up with a nasty surprise. | |
517 | * Normally d_walk() doesn't care about cursors moving around - | |
518 | * ->d_lock on parent prevents that and since a cursor has no children | |
519 | * of its own, we get through it without ever unlocking the parent. | |
520 | * There is one exception, though - if we ascend from a child that | |
521 | * gets killed as soon as we unlock it, the next sibling is found | |
522 | * using the value left in its ->d_child.next. And if _that_ | |
523 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
524 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
525 | * everything the cursor had been moved past. | |
526 | * | |
527 | * Solution: make sure that the pointer left behind in ->d_child.next | |
528 | * points to something that won't be moving around. I.e. skip the | |
529 | * cursors. | |
530 | */ | |
531 | while (dentry->d_child.next != &parent->d_subdirs) { | |
532 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
533 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
534 | break; | |
535 | dentry->d_child.next = next->d_child.next; | |
536 | } | |
537 | } | |
538 | ||
e55fd011 | 539 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 540 | { |
41edf278 AV |
541 | struct dentry *parent = NULL; |
542 | bool can_free = true; | |
41edf278 | 543 | if (!IS_ROOT(dentry)) |
77812a1e | 544 | parent = dentry->d_parent; |
31e6b01f | 545 | |
0d98439e LT |
546 | /* |
547 | * The dentry is now unrecoverably dead to the world. | |
548 | */ | |
549 | lockref_mark_dead(&dentry->d_lockref); | |
550 | ||
f0023bc6 | 551 | /* |
f0023bc6 SW |
552 | * inform the fs via d_prune that this dentry is about to be |
553 | * unhashed and destroyed. | |
554 | */ | |
29266201 | 555 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
556 | dentry->d_op->d_prune(dentry); |
557 | ||
01b60351 AV |
558 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
559 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
560 | d_lru_del(dentry); | |
01b60351 | 561 | } |
77812a1e NP |
562 | /* if it was on the hash then remove it */ |
563 | __d_drop(dentry); | |
ba65dc5e | 564 | dentry_unlist(dentry, parent); |
03b3b889 AV |
565 | if (parent) |
566 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
567 | if (dentry->d_inode) |
568 | dentry_unlink_inode(dentry); | |
569 | else | |
570 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
571 | this_cpu_dec(nr_dentry); |
572 | if (dentry->d_op && dentry->d_op->d_release) | |
573 | dentry->d_op->d_release(dentry); | |
574 | ||
41edf278 AV |
575 | spin_lock(&dentry->d_lock); |
576 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
577 | dentry->d_flags |= DCACHE_MAY_FREE; | |
578 | can_free = false; | |
579 | } | |
580 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
581 | if (likely(can_free)) |
582 | dentry_free(dentry); | |
e55fd011 AV |
583 | } |
584 | ||
8b987a46 | 585 | static struct dentry *__lock_parent(struct dentry *dentry) |
046b961b | 586 | { |
8b987a46 | 587 | struct dentry *parent; |
046b961b | 588 | rcu_read_lock(); |
c2338f2d | 589 | spin_unlock(&dentry->d_lock); |
046b961b | 590 | again: |
66702eb5 | 591 | parent = READ_ONCE(dentry->d_parent); |
046b961b AV |
592 | spin_lock(&parent->d_lock); |
593 | /* | |
594 | * We can't blindly lock dentry until we are sure | |
595 | * that we won't violate the locking order. | |
596 | * Any changes of dentry->d_parent must have | |
597 | * been done with parent->d_lock held, so | |
598 | * spin_lock() above is enough of a barrier | |
599 | * for checking if it's still our child. | |
600 | */ | |
601 | if (unlikely(parent != dentry->d_parent)) { | |
602 | spin_unlock(&parent->d_lock); | |
603 | goto again; | |
604 | } | |
65d8eb5a AV |
605 | rcu_read_unlock(); |
606 | if (parent != dentry) | |
9f12600f | 607 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
65d8eb5a | 608 | else |
046b961b AV |
609 | parent = NULL; |
610 | return parent; | |
611 | } | |
612 | ||
8b987a46 AV |
613 | static inline struct dentry *lock_parent(struct dentry *dentry) |
614 | { | |
615 | struct dentry *parent = dentry->d_parent; | |
616 | if (IS_ROOT(dentry)) | |
617 | return NULL; | |
618 | if (likely(spin_trylock(&parent->d_lock))) | |
619 | return parent; | |
620 | return __lock_parent(dentry); | |
621 | } | |
622 | ||
a338579f AV |
623 | static inline bool retain_dentry(struct dentry *dentry) |
624 | { | |
625 | WARN_ON(d_in_lookup(dentry)); | |
626 | ||
627 | /* Unreachable? Get rid of it */ | |
628 | if (unlikely(d_unhashed(dentry))) | |
629 | return false; | |
630 | ||
631 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
632 | return false; | |
633 | ||
634 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
635 | if (dentry->d_op->d_delete(dentry)) | |
636 | return false; | |
637 | } | |
62d9956c AV |
638 | /* retain; LRU fodder */ |
639 | dentry->d_lockref.count--; | |
640 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | |
641 | d_lru_add(dentry); | |
642 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) | |
643 | dentry->d_flags |= DCACHE_REFERENCED; | |
a338579f AV |
644 | return true; |
645 | } | |
646 | ||
c1d0c1a2 JO |
647 | /* |
648 | * Finish off a dentry we've decided to kill. | |
649 | * dentry->d_lock must be held, returns with it unlocked. | |
650 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
651 | */ | |
652 | static struct dentry *dentry_kill(struct dentry *dentry) | |
653 | __releases(dentry->d_lock) | |
654 | { | |
655 | struct inode *inode = dentry->d_inode; | |
656 | struct dentry *parent = NULL; | |
657 | ||
658 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
f657a666 | 659 | goto slow_positive; |
c1d0c1a2 JO |
660 | |
661 | if (!IS_ROOT(dentry)) { | |
662 | parent = dentry->d_parent; | |
663 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
f657a666 AV |
664 | parent = __lock_parent(dentry); |
665 | if (likely(inode || !dentry->d_inode)) | |
666 | goto got_locks; | |
667 | /* negative that became positive */ | |
668 | if (parent) | |
669 | spin_unlock(&parent->d_lock); | |
670 | inode = dentry->d_inode; | |
671 | goto slow_positive; | |
c1d0c1a2 JO |
672 | } |
673 | } | |
c1d0c1a2 JO |
674 | __dentry_kill(dentry); |
675 | return parent; | |
676 | ||
f657a666 AV |
677 | slow_positive: |
678 | spin_unlock(&dentry->d_lock); | |
679 | spin_lock(&inode->i_lock); | |
680 | spin_lock(&dentry->d_lock); | |
681 | parent = lock_parent(dentry); | |
682 | got_locks: | |
683 | if (unlikely(dentry->d_lockref.count != 1)) { | |
684 | dentry->d_lockref.count--; | |
685 | } else if (likely(!retain_dentry(dentry))) { | |
686 | __dentry_kill(dentry); | |
687 | return parent; | |
688 | } | |
689 | /* we are keeping it, after all */ | |
690 | if (inode) | |
691 | spin_unlock(&inode->i_lock); | |
692 | if (parent) | |
693 | spin_unlock(&parent->d_lock); | |
c1d0c1a2 | 694 | spin_unlock(&dentry->d_lock); |
f657a666 | 695 | return NULL; |
c1d0c1a2 JO |
696 | } |
697 | ||
360f5479 LT |
698 | /* |
699 | * Try to do a lockless dput(), and return whether that was successful. | |
700 | * | |
701 | * If unsuccessful, we return false, having already taken the dentry lock. | |
702 | * | |
703 | * The caller needs to hold the RCU read lock, so that the dentry is | |
704 | * guaranteed to stay around even if the refcount goes down to zero! | |
705 | */ | |
706 | static inline bool fast_dput(struct dentry *dentry) | |
707 | { | |
708 | int ret; | |
709 | unsigned int d_flags; | |
710 | ||
711 | /* | |
712 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 713 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
714 | */ |
715 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
716 | return lockref_put_or_lock(&dentry->d_lockref); | |
717 | ||
718 | /* | |
719 | * .. otherwise, we can try to just decrement the | |
720 | * lockref optimistically. | |
721 | */ | |
722 | ret = lockref_put_return(&dentry->d_lockref); | |
723 | ||
724 | /* | |
725 | * If the lockref_put_return() failed due to the lock being held | |
726 | * by somebody else, the fast path has failed. We will need to | |
727 | * get the lock, and then check the count again. | |
728 | */ | |
729 | if (unlikely(ret < 0)) { | |
730 | spin_lock(&dentry->d_lock); | |
731 | if (dentry->d_lockref.count > 1) { | |
732 | dentry->d_lockref.count--; | |
733 | spin_unlock(&dentry->d_lock); | |
734 | return 1; | |
735 | } | |
736 | return 0; | |
737 | } | |
738 | ||
739 | /* | |
740 | * If we weren't the last ref, we're done. | |
741 | */ | |
742 | if (ret) | |
743 | return 1; | |
744 | ||
745 | /* | |
746 | * Careful, careful. The reference count went down | |
747 | * to zero, but we don't hold the dentry lock, so | |
748 | * somebody else could get it again, and do another | |
749 | * dput(), and we need to not race with that. | |
750 | * | |
751 | * However, there is a very special and common case | |
752 | * where we don't care, because there is nothing to | |
753 | * do: the dentry is still hashed, it does not have | |
754 | * a 'delete' op, and it's referenced and already on | |
755 | * the LRU list. | |
756 | * | |
757 | * NOTE! Since we aren't locked, these values are | |
758 | * not "stable". However, it is sufficient that at | |
759 | * some point after we dropped the reference the | |
760 | * dentry was hashed and the flags had the proper | |
761 | * value. Other dentry users may have re-gotten | |
762 | * a reference to the dentry and change that, but | |
763 | * our work is done - we can leave the dentry | |
764 | * around with a zero refcount. | |
765 | */ | |
766 | smp_rmb(); | |
66702eb5 | 767 | d_flags = READ_ONCE(dentry->d_flags); |
75a6f82a | 768 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
769 | |
770 | /* Nothing to do? Dropping the reference was all we needed? */ | |
771 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
772 | return 1; | |
773 | ||
774 | /* | |
775 | * Not the fast normal case? Get the lock. We've already decremented | |
776 | * the refcount, but we'll need to re-check the situation after | |
777 | * getting the lock. | |
778 | */ | |
779 | spin_lock(&dentry->d_lock); | |
780 | ||
781 | /* | |
782 | * Did somebody else grab a reference to it in the meantime, and | |
783 | * we're no longer the last user after all? Alternatively, somebody | |
784 | * else could have killed it and marked it dead. Either way, we | |
785 | * don't need to do anything else. | |
786 | */ | |
787 | if (dentry->d_lockref.count) { | |
788 | spin_unlock(&dentry->d_lock); | |
789 | return 1; | |
790 | } | |
791 | ||
792 | /* | |
793 | * Re-get the reference we optimistically dropped. We hold the | |
794 | * lock, and we just tested that it was zero, so we can just | |
795 | * set it to 1. | |
796 | */ | |
797 | dentry->d_lockref.count = 1; | |
798 | return 0; | |
799 | } | |
800 | ||
801 | ||
1da177e4 LT |
802 | /* |
803 | * This is dput | |
804 | * | |
805 | * This is complicated by the fact that we do not want to put | |
806 | * dentries that are no longer on any hash chain on the unused | |
807 | * list: we'd much rather just get rid of them immediately. | |
808 | * | |
809 | * However, that implies that we have to traverse the dentry | |
810 | * tree upwards to the parents which might _also_ now be | |
811 | * scheduled for deletion (it may have been only waiting for | |
812 | * its last child to go away). | |
813 | * | |
814 | * This tail recursion is done by hand as we don't want to depend | |
815 | * on the compiler to always get this right (gcc generally doesn't). | |
816 | * Real recursion would eat up our stack space. | |
817 | */ | |
818 | ||
819 | /* | |
820 | * dput - release a dentry | |
821 | * @dentry: dentry to release | |
822 | * | |
823 | * Release a dentry. This will drop the usage count and if appropriate | |
824 | * call the dentry unlink method as well as removing it from the queues and | |
825 | * releasing its resources. If the parent dentries were scheduled for release | |
826 | * they too may now get deleted. | |
1da177e4 | 827 | */ |
1da177e4 LT |
828 | void dput(struct dentry *dentry) |
829 | { | |
8aab6a27 | 830 | if (unlikely(!dentry)) |
1da177e4 LT |
831 | return; |
832 | ||
833 | repeat: | |
47be6184 WF |
834 | might_sleep(); |
835 | ||
360f5479 LT |
836 | rcu_read_lock(); |
837 | if (likely(fast_dput(dentry))) { | |
838 | rcu_read_unlock(); | |
1da177e4 | 839 | return; |
360f5479 LT |
840 | } |
841 | ||
842 | /* Slow case: now with the dentry lock held */ | |
843 | rcu_read_unlock(); | |
1da177e4 | 844 | |
a338579f | 845 | if (likely(retain_dentry(dentry))) { |
a338579f AV |
846 | spin_unlock(&dentry->d_lock); |
847 | return; | |
1da177e4 | 848 | } |
265ac902 | 849 | |
8cbf74da | 850 | dentry = dentry_kill(dentry); |
47be6184 WF |
851 | if (dentry) { |
852 | cond_resched(); | |
d52b9086 | 853 | goto repeat; |
47be6184 | 854 | } |
1da177e4 | 855 | } |
ec4f8605 | 856 | EXPORT_SYMBOL(dput); |
1da177e4 | 857 | |
1da177e4 | 858 | |
b5c84bf6 | 859 | /* This must be called with d_lock held */ |
dc0474be | 860 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 861 | { |
98474236 | 862 | dentry->d_lockref.count++; |
23044507 NP |
863 | } |
864 | ||
dc0474be | 865 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 866 | { |
98474236 | 867 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
868 | } |
869 | ||
b7ab39f6 NP |
870 | struct dentry *dget_parent(struct dentry *dentry) |
871 | { | |
df3d0bbc | 872 | int gotref; |
b7ab39f6 NP |
873 | struct dentry *ret; |
874 | ||
df3d0bbc WL |
875 | /* |
876 | * Do optimistic parent lookup without any | |
877 | * locking. | |
878 | */ | |
879 | rcu_read_lock(); | |
66702eb5 | 880 | ret = READ_ONCE(dentry->d_parent); |
df3d0bbc WL |
881 | gotref = lockref_get_not_zero(&ret->d_lockref); |
882 | rcu_read_unlock(); | |
883 | if (likely(gotref)) { | |
66702eb5 | 884 | if (likely(ret == READ_ONCE(dentry->d_parent))) |
df3d0bbc WL |
885 | return ret; |
886 | dput(ret); | |
887 | } | |
888 | ||
b7ab39f6 | 889 | repeat: |
a734eb45 NP |
890 | /* |
891 | * Don't need rcu_dereference because we re-check it was correct under | |
892 | * the lock. | |
893 | */ | |
894 | rcu_read_lock(); | |
b7ab39f6 | 895 | ret = dentry->d_parent; |
a734eb45 NP |
896 | spin_lock(&ret->d_lock); |
897 | if (unlikely(ret != dentry->d_parent)) { | |
898 | spin_unlock(&ret->d_lock); | |
899 | rcu_read_unlock(); | |
b7ab39f6 NP |
900 | goto repeat; |
901 | } | |
a734eb45 | 902 | rcu_read_unlock(); |
98474236 WL |
903 | BUG_ON(!ret->d_lockref.count); |
904 | ret->d_lockref.count++; | |
b7ab39f6 | 905 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
906 | return ret; |
907 | } | |
908 | EXPORT_SYMBOL(dget_parent); | |
909 | ||
1da177e4 LT |
910 | /** |
911 | * d_find_alias - grab a hashed alias of inode | |
912 | * @inode: inode in question | |
1da177e4 LT |
913 | * |
914 | * If inode has a hashed alias, or is a directory and has any alias, | |
915 | * acquire the reference to alias and return it. Otherwise return NULL. | |
916 | * Notice that if inode is a directory there can be only one alias and | |
917 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
918 | * of a filesystem, or if the directory was renamed and d_revalidate |
919 | * was the first vfs operation to notice. | |
1da177e4 | 920 | * |
21c0d8fd | 921 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 922 | * any other hashed alias over that one. |
1da177e4 | 923 | */ |
52ed46f0 | 924 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 925 | { |
da502956 | 926 | struct dentry *alias, *discon_alias; |
1da177e4 | 927 | |
da502956 NP |
928 | again: |
929 | discon_alias = NULL; | |
946e51f2 | 930 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 931 | spin_lock(&alias->d_lock); |
1da177e4 | 932 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 933 | if (IS_ROOT(alias) && |
da502956 | 934 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 935 | discon_alias = alias; |
52ed46f0 | 936 | } else { |
dc0474be | 937 | __dget_dlock(alias); |
da502956 NP |
938 | spin_unlock(&alias->d_lock); |
939 | return alias; | |
940 | } | |
941 | } | |
942 | spin_unlock(&alias->d_lock); | |
943 | } | |
944 | if (discon_alias) { | |
945 | alias = discon_alias; | |
946 | spin_lock(&alias->d_lock); | |
947 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
948 | __dget_dlock(alias); |
949 | spin_unlock(&alias->d_lock); | |
950 | return alias; | |
1da177e4 | 951 | } |
da502956 NP |
952 | spin_unlock(&alias->d_lock); |
953 | goto again; | |
1da177e4 | 954 | } |
da502956 | 955 | return NULL; |
1da177e4 LT |
956 | } |
957 | ||
da502956 | 958 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 959 | { |
214fda1f DH |
960 | struct dentry *de = NULL; |
961 | ||
b3d9b7a3 | 962 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 963 | spin_lock(&inode->i_lock); |
52ed46f0 | 964 | de = __d_find_alias(inode); |
873feea0 | 965 | spin_unlock(&inode->i_lock); |
214fda1f | 966 | } |
1da177e4 LT |
967 | return de; |
968 | } | |
ec4f8605 | 969 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
970 | |
971 | /* | |
972 | * Try to kill dentries associated with this inode. | |
973 | * WARNING: you must own a reference to inode. | |
974 | */ | |
975 | void d_prune_aliases(struct inode *inode) | |
976 | { | |
0cdca3f9 | 977 | struct dentry *dentry; |
1da177e4 | 978 | restart: |
873feea0 | 979 | spin_lock(&inode->i_lock); |
946e51f2 | 980 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 981 | spin_lock(&dentry->d_lock); |
98474236 | 982 | if (!dentry->d_lockref.count) { |
29355c39 AV |
983 | struct dentry *parent = lock_parent(dentry); |
984 | if (likely(!dentry->d_lockref.count)) { | |
985 | __dentry_kill(dentry); | |
4a7795d3 | 986 | dput(parent); |
29355c39 AV |
987 | goto restart; |
988 | } | |
989 | if (parent) | |
990 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
991 | } |
992 | spin_unlock(&dentry->d_lock); | |
993 | } | |
873feea0 | 994 | spin_unlock(&inode->i_lock); |
1da177e4 | 995 | } |
ec4f8605 | 996 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 997 | |
3b3f09f4 AV |
998 | /* |
999 | * Lock a dentry from shrink list. | |
8f04da2a JO |
1000 | * Called under rcu_read_lock() and dentry->d_lock; the former |
1001 | * guarantees that nothing we access will be freed under us. | |
3b3f09f4 | 1002 | * Note that dentry is *not* protected from concurrent dentry_kill(), |
8f04da2a JO |
1003 | * d_delete(), etc. |
1004 | * | |
3b3f09f4 AV |
1005 | * Return false if dentry has been disrupted or grabbed, leaving |
1006 | * the caller to kick it off-list. Otherwise, return true and have | |
1007 | * that dentry's inode and parent both locked. | |
1008 | */ | |
1009 | static bool shrink_lock_dentry(struct dentry *dentry) | |
1da177e4 | 1010 | { |
3b3f09f4 AV |
1011 | struct inode *inode; |
1012 | struct dentry *parent; | |
da3bbdd4 | 1013 | |
3b3f09f4 AV |
1014 | if (dentry->d_lockref.count) |
1015 | return false; | |
1016 | ||
1017 | inode = dentry->d_inode; | |
1018 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
3b3f09f4 AV |
1019 | spin_unlock(&dentry->d_lock); |
1020 | spin_lock(&inode->i_lock); | |
ec33679d | 1021 | spin_lock(&dentry->d_lock); |
3b3f09f4 AV |
1022 | if (unlikely(dentry->d_lockref.count)) |
1023 | goto out; | |
1024 | /* changed inode means that somebody had grabbed it */ | |
1025 | if (unlikely(inode != dentry->d_inode)) | |
1026 | goto out; | |
3b3f09f4 | 1027 | } |
046b961b | 1028 | |
3b3f09f4 AV |
1029 | parent = dentry->d_parent; |
1030 | if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) | |
1031 | return true; | |
dd1f6b2e | 1032 | |
3b3f09f4 | 1033 | spin_unlock(&dentry->d_lock); |
3b3f09f4 AV |
1034 | spin_lock(&parent->d_lock); |
1035 | if (unlikely(parent != dentry->d_parent)) { | |
1036 | spin_unlock(&parent->d_lock); | |
1037 | spin_lock(&dentry->d_lock); | |
1038 | goto out; | |
1039 | } | |
1040 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
8f04da2a | 1041 | if (likely(!dentry->d_lockref.count)) |
3b3f09f4 | 1042 | return true; |
3b3f09f4 AV |
1043 | spin_unlock(&parent->d_lock); |
1044 | out: | |
1045 | if (inode) | |
1046 | spin_unlock(&inode->i_lock); | |
3b3f09f4 AV |
1047 | return false; |
1048 | } | |
77812a1e | 1049 | |
3b3f09f4 AV |
1050 | static void shrink_dentry_list(struct list_head *list) |
1051 | { | |
1052 | while (!list_empty(list)) { | |
1053 | struct dentry *dentry, *parent; | |
64fd72e0 | 1054 | |
3b3f09f4 AV |
1055 | dentry = list_entry(list->prev, struct dentry, d_lru); |
1056 | spin_lock(&dentry->d_lock); | |
8f04da2a | 1057 | rcu_read_lock(); |
3b3f09f4 AV |
1058 | if (!shrink_lock_dentry(dentry)) { |
1059 | bool can_free = false; | |
8f04da2a | 1060 | rcu_read_unlock(); |
3b3f09f4 AV |
1061 | d_shrink_del(dentry); |
1062 | if (dentry->d_lockref.count < 0) | |
1063 | can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
64fd72e0 AV |
1064 | spin_unlock(&dentry->d_lock); |
1065 | if (can_free) | |
1066 | dentry_free(dentry); | |
1067 | continue; | |
1068 | } | |
8f04da2a | 1069 | rcu_read_unlock(); |
3b3f09f4 AV |
1070 | d_shrink_del(dentry); |
1071 | parent = dentry->d_parent; | |
ff2fde99 | 1072 | __dentry_kill(dentry); |
3b3f09f4 AV |
1073 | if (parent == dentry) |
1074 | continue; | |
5c47e6d0 AV |
1075 | /* |
1076 | * We need to prune ancestors too. This is necessary to prevent | |
1077 | * quadratic behavior of shrink_dcache_parent(), but is also | |
1078 | * expected to be beneficial in reducing dentry cache | |
1079 | * fragmentation. | |
1080 | */ | |
1081 | dentry = parent; | |
8f04da2a JO |
1082 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) |
1083 | dentry = dentry_kill(dentry); | |
da3bbdd4 | 1084 | } |
3049cfe2 CH |
1085 | } |
1086 | ||
3f97b163 VD |
1087 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1088 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1089 | { |
1090 | struct list_head *freeable = arg; | |
1091 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1092 | ||
1093 | ||
1094 | /* | |
1095 | * we are inverting the lru lock/dentry->d_lock here, | |
1096 | * so use a trylock. If we fail to get the lock, just skip | |
1097 | * it | |
1098 | */ | |
1099 | if (!spin_trylock(&dentry->d_lock)) | |
1100 | return LRU_SKIP; | |
1101 | ||
1102 | /* | |
1103 | * Referenced dentries are still in use. If they have active | |
1104 | * counts, just remove them from the LRU. Otherwise give them | |
1105 | * another pass through the LRU. | |
1106 | */ | |
1107 | if (dentry->d_lockref.count) { | |
3f97b163 | 1108 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1109 | spin_unlock(&dentry->d_lock); |
1110 | return LRU_REMOVED; | |
1111 | } | |
1112 | ||
1113 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1114 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1115 | spin_unlock(&dentry->d_lock); | |
1116 | ||
1117 | /* | |
1118 | * The list move itself will be made by the common LRU code. At | |
1119 | * this point, we've dropped the dentry->d_lock but keep the | |
1120 | * lru lock. This is safe to do, since every list movement is | |
1121 | * protected by the lru lock even if both locks are held. | |
1122 | * | |
1123 | * This is guaranteed by the fact that all LRU management | |
1124 | * functions are intermediated by the LRU API calls like | |
1125 | * list_lru_add and list_lru_del. List movement in this file | |
1126 | * only ever occur through this functions or through callbacks | |
1127 | * like this one, that are called from the LRU API. | |
1128 | * | |
1129 | * The only exceptions to this are functions like | |
1130 | * shrink_dentry_list, and code that first checks for the | |
1131 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1132 | * operating only with stack provided lists after they are | |
1133 | * properly isolated from the main list. It is thus, always a | |
1134 | * local access. | |
1135 | */ | |
1136 | return LRU_ROTATE; | |
1137 | } | |
1138 | ||
3f97b163 | 1139 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1140 | spin_unlock(&dentry->d_lock); |
1141 | ||
1142 | return LRU_REMOVED; | |
1143 | } | |
1144 | ||
3049cfe2 | 1145 | /** |
b48f03b3 DC |
1146 | * prune_dcache_sb - shrink the dcache |
1147 | * @sb: superblock | |
503c358c | 1148 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1149 | * |
503c358c VD |
1150 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1151 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1152 | * function. |
3049cfe2 | 1153 | * |
b48f03b3 DC |
1154 | * This function may fail to free any resources if all the dentries are in |
1155 | * use. | |
3049cfe2 | 1156 | */ |
503c358c | 1157 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1158 | { |
f6041567 DC |
1159 | LIST_HEAD(dispose); |
1160 | long freed; | |
3049cfe2 | 1161 | |
503c358c VD |
1162 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1163 | dentry_lru_isolate, &dispose); | |
f6041567 | 1164 | shrink_dentry_list(&dispose); |
0a234c6d | 1165 | return freed; |
da3bbdd4 | 1166 | } |
23044507 | 1167 | |
4e717f5c | 1168 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1169 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1170 | { |
4e717f5c GC |
1171 | struct list_head *freeable = arg; |
1172 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1173 | |
4e717f5c GC |
1174 | /* |
1175 | * we are inverting the lru lock/dentry->d_lock here, | |
1176 | * so use a trylock. If we fail to get the lock, just skip | |
1177 | * it | |
1178 | */ | |
1179 | if (!spin_trylock(&dentry->d_lock)) | |
1180 | return LRU_SKIP; | |
1181 | ||
3f97b163 | 1182 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1183 | spin_unlock(&dentry->d_lock); |
ec33679d | 1184 | |
4e717f5c | 1185 | return LRU_REMOVED; |
da3bbdd4 KM |
1186 | } |
1187 | ||
4e717f5c | 1188 | |
1da177e4 LT |
1189 | /** |
1190 | * shrink_dcache_sb - shrink dcache for a superblock | |
1191 | * @sb: superblock | |
1192 | * | |
3049cfe2 CH |
1193 | * Shrink the dcache for the specified super block. This is used to free |
1194 | * the dcache before unmounting a file system. | |
1da177e4 | 1195 | */ |
3049cfe2 | 1196 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1197 | { |
4e717f5c GC |
1198 | long freed; |
1199 | ||
1200 | do { | |
1201 | LIST_HEAD(dispose); | |
1202 | ||
1203 | freed = list_lru_walk(&sb->s_dentry_lru, | |
b17c070f | 1204 | dentry_lru_isolate_shrink, &dispose, 1024); |
3049cfe2 | 1205 | |
4e717f5c GC |
1206 | this_cpu_sub(nr_dentry_unused, freed); |
1207 | shrink_dentry_list(&dispose); | |
b17c070f ST |
1208 | cond_resched(); |
1209 | } while (list_lru_count(&sb->s_dentry_lru) > 0); | |
1da177e4 | 1210 | } |
ec4f8605 | 1211 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1212 | |
db14fc3a MS |
1213 | /** |
1214 | * enum d_walk_ret - action to talke during tree walk | |
1215 | * @D_WALK_CONTINUE: contrinue walk | |
1216 | * @D_WALK_QUIT: quit walk | |
1217 | * @D_WALK_NORETRY: quit when retry is needed | |
1218 | * @D_WALK_SKIP: skip this dentry and its children | |
1219 | */ | |
1220 | enum d_walk_ret { | |
1221 | D_WALK_CONTINUE, | |
1222 | D_WALK_QUIT, | |
1223 | D_WALK_NORETRY, | |
1224 | D_WALK_SKIP, | |
1225 | }; | |
c826cb7d | 1226 | |
1da177e4 | 1227 | /** |
db14fc3a MS |
1228 | * d_walk - walk the dentry tree |
1229 | * @parent: start of walk | |
1230 | * @data: data passed to @enter() and @finish() | |
1231 | * @enter: callback when first entering the dentry | |
1232 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1233 | * |
db14fc3a | 1234 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1235 | */ |
db14fc3a MS |
1236 | static void d_walk(struct dentry *parent, void *data, |
1237 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1238 | void (*finish)(void *)) | |
1da177e4 | 1239 | { |
949854d0 | 1240 | struct dentry *this_parent; |
1da177e4 | 1241 | struct list_head *next; |
48f5ec21 | 1242 | unsigned seq = 0; |
db14fc3a MS |
1243 | enum d_walk_ret ret; |
1244 | bool retry = true; | |
949854d0 | 1245 | |
58db63d0 | 1246 | again: |
48f5ec21 | 1247 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1248 | this_parent = parent; |
2fd6b7f5 | 1249 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1250 | |
1251 | ret = enter(data, this_parent); | |
1252 | switch (ret) { | |
1253 | case D_WALK_CONTINUE: | |
1254 | break; | |
1255 | case D_WALK_QUIT: | |
1256 | case D_WALK_SKIP: | |
1257 | goto out_unlock; | |
1258 | case D_WALK_NORETRY: | |
1259 | retry = false; | |
1260 | break; | |
1261 | } | |
1da177e4 LT |
1262 | repeat: |
1263 | next = this_parent->d_subdirs.next; | |
1264 | resume: | |
1265 | while (next != &this_parent->d_subdirs) { | |
1266 | struct list_head *tmp = next; | |
946e51f2 | 1267 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1268 | next = tmp->next; |
2fd6b7f5 | 1269 | |
ba65dc5e AV |
1270 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1271 | continue; | |
1272 | ||
2fd6b7f5 | 1273 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1274 | |
1275 | ret = enter(data, dentry); | |
1276 | switch (ret) { | |
1277 | case D_WALK_CONTINUE: | |
1278 | break; | |
1279 | case D_WALK_QUIT: | |
2fd6b7f5 | 1280 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1281 | goto out_unlock; |
1282 | case D_WALK_NORETRY: | |
1283 | retry = false; | |
1284 | break; | |
1285 | case D_WALK_SKIP: | |
1286 | spin_unlock(&dentry->d_lock); | |
1287 | continue; | |
2fd6b7f5 | 1288 | } |
db14fc3a | 1289 | |
1da177e4 | 1290 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1291 | spin_unlock(&this_parent->d_lock); |
1292 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1293 | this_parent = dentry; |
2fd6b7f5 | 1294 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1295 | goto repeat; |
1296 | } | |
2fd6b7f5 | 1297 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1298 | } |
1299 | /* | |
1300 | * All done at this level ... ascend and resume the search. | |
1301 | */ | |
ca5358ef AV |
1302 | rcu_read_lock(); |
1303 | ascend: | |
1da177e4 | 1304 | if (this_parent != parent) { |
c826cb7d | 1305 | struct dentry *child = this_parent; |
31dec132 AV |
1306 | this_parent = child->d_parent; |
1307 | ||
31dec132 AV |
1308 | spin_unlock(&child->d_lock); |
1309 | spin_lock(&this_parent->d_lock); | |
1310 | ||
ca5358ef AV |
1311 | /* might go back up the wrong parent if we have had a rename. */ |
1312 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1313 | goto rename_retry; |
2159184e AV |
1314 | /* go into the first sibling still alive */ |
1315 | do { | |
1316 | next = child->d_child.next; | |
ca5358ef AV |
1317 | if (next == &this_parent->d_subdirs) |
1318 | goto ascend; | |
1319 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1320 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1321 | rcu_read_unlock(); |
1da177e4 LT |
1322 | goto resume; |
1323 | } | |
ca5358ef | 1324 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1325 | goto rename_retry; |
ca5358ef | 1326 | rcu_read_unlock(); |
db14fc3a MS |
1327 | if (finish) |
1328 | finish(data); | |
1329 | ||
1330 | out_unlock: | |
1331 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1332 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1333 | return; |
58db63d0 NP |
1334 | |
1335 | rename_retry: | |
ca5358ef AV |
1336 | spin_unlock(&this_parent->d_lock); |
1337 | rcu_read_unlock(); | |
1338 | BUG_ON(seq & 1); | |
db14fc3a MS |
1339 | if (!retry) |
1340 | return; | |
48f5ec21 | 1341 | seq = 1; |
58db63d0 | 1342 | goto again; |
1da177e4 | 1343 | } |
db14fc3a | 1344 | |
01619491 IK |
1345 | struct check_mount { |
1346 | struct vfsmount *mnt; | |
1347 | unsigned int mounted; | |
1348 | }; | |
1349 | ||
1350 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1351 | { | |
1352 | struct check_mount *info = data; | |
1353 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1354 | ||
1355 | if (likely(!d_mountpoint(dentry))) | |
1356 | return D_WALK_CONTINUE; | |
1357 | if (__path_is_mountpoint(&path)) { | |
1358 | info->mounted = 1; | |
1359 | return D_WALK_QUIT; | |
1360 | } | |
1361 | return D_WALK_CONTINUE; | |
1362 | } | |
1363 | ||
1364 | /** | |
1365 | * path_has_submounts - check for mounts over a dentry in the | |
1366 | * current namespace. | |
1367 | * @parent: path to check. | |
1368 | * | |
1369 | * Return true if the parent or its subdirectories contain | |
1370 | * a mount point in the current namespace. | |
1371 | */ | |
1372 | int path_has_submounts(const struct path *parent) | |
1373 | { | |
1374 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1375 | ||
1376 | read_seqlock_excl(&mount_lock); | |
1377 | d_walk(parent->dentry, &data, path_check_mount, NULL); | |
1378 | read_sequnlock_excl(&mount_lock); | |
1379 | ||
1380 | return data.mounted; | |
1381 | } | |
1382 | EXPORT_SYMBOL(path_has_submounts); | |
1383 | ||
eed81007 MS |
1384 | /* |
1385 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1386 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1387 | * subtree can become unreachable). | |
1388 | * | |
1ffe46d1 | 1389 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1390 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1391 | */ | |
1392 | int d_set_mounted(struct dentry *dentry) | |
1393 | { | |
1394 | struct dentry *p; | |
1395 | int ret = -ENOENT; | |
1396 | write_seqlock(&rename_lock); | |
1397 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1398 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1399 | spin_lock(&p->d_lock); |
1400 | if (unlikely(d_unhashed(p))) { | |
1401 | spin_unlock(&p->d_lock); | |
1402 | goto out; | |
1403 | } | |
1404 | spin_unlock(&p->d_lock); | |
1405 | } | |
1406 | spin_lock(&dentry->d_lock); | |
1407 | if (!d_unlinked(dentry)) { | |
3895dbf8 EB |
1408 | ret = -EBUSY; |
1409 | if (!d_mountpoint(dentry)) { | |
1410 | dentry->d_flags |= DCACHE_MOUNTED; | |
1411 | ret = 0; | |
1412 | } | |
eed81007 MS |
1413 | } |
1414 | spin_unlock(&dentry->d_lock); | |
1415 | out: | |
1416 | write_sequnlock(&rename_lock); | |
1417 | return ret; | |
1418 | } | |
1419 | ||
1da177e4 | 1420 | /* |
fd517909 | 1421 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1422 | * and move any unused dentries to the end of the unused |
1423 | * list for prune_dcache(). We descend to the next level | |
1424 | * whenever the d_subdirs list is non-empty and continue | |
1425 | * searching. | |
1426 | * | |
1427 | * It returns zero iff there are no unused children, | |
1428 | * otherwise it returns the number of children moved to | |
1429 | * the end of the unused list. This may not be the total | |
1430 | * number of unused children, because select_parent can | |
1431 | * drop the lock and return early due to latency | |
1432 | * constraints. | |
1433 | */ | |
1da177e4 | 1434 | |
db14fc3a MS |
1435 | struct select_data { |
1436 | struct dentry *start; | |
1437 | struct list_head dispose; | |
1438 | int found; | |
1439 | }; | |
23044507 | 1440 | |
db14fc3a MS |
1441 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1442 | { | |
1443 | struct select_data *data = _data; | |
1444 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1445 | |
db14fc3a MS |
1446 | if (data->start == dentry) |
1447 | goto out; | |
2fd6b7f5 | 1448 | |
fe91522a | 1449 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1450 | data->found++; |
fe91522a AV |
1451 | } else { |
1452 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1453 | d_lru_del(dentry); | |
1454 | if (!dentry->d_lockref.count) { | |
1455 | d_shrink_add(dentry, &data->dispose); | |
1456 | data->found++; | |
1457 | } | |
1da177e4 | 1458 | } |
db14fc3a MS |
1459 | /* |
1460 | * We can return to the caller if we have found some (this | |
1461 | * ensures forward progress). We'll be coming back to find | |
1462 | * the rest. | |
1463 | */ | |
fe91522a AV |
1464 | if (!list_empty(&data->dispose)) |
1465 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1466 | out: |
db14fc3a | 1467 | return ret; |
1da177e4 LT |
1468 | } |
1469 | ||
1470 | /** | |
1471 | * shrink_dcache_parent - prune dcache | |
1472 | * @parent: parent of entries to prune | |
1473 | * | |
1474 | * Prune the dcache to remove unused children of the parent dentry. | |
1475 | */ | |
db14fc3a | 1476 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1477 | { |
db14fc3a MS |
1478 | for (;;) { |
1479 | struct select_data data; | |
1da177e4 | 1480 | |
db14fc3a MS |
1481 | INIT_LIST_HEAD(&data.dispose); |
1482 | data.start = parent; | |
1483 | data.found = 0; | |
1484 | ||
1485 | d_walk(parent, &data, select_collect, NULL); | |
1486 | if (!data.found) | |
1487 | break; | |
1488 | ||
1489 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1490 | cond_resched(); |
1491 | } | |
1da177e4 | 1492 | } |
ec4f8605 | 1493 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1494 | |
9c8c10e2 | 1495 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1496 | { |
9c8c10e2 AV |
1497 | /* it has busy descendents; complain about those instead */ |
1498 | if (!list_empty(&dentry->d_subdirs)) | |
1499 | return D_WALK_CONTINUE; | |
42c32608 | 1500 | |
9c8c10e2 AV |
1501 | /* root with refcount 1 is fine */ |
1502 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1503 | return D_WALK_CONTINUE; | |
1504 | ||
1505 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1506 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1507 | dentry, |
1508 | dentry->d_inode ? | |
1509 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1510 | dentry, |
42c32608 AV |
1511 | dentry->d_lockref.count, |
1512 | dentry->d_sb->s_type->name, | |
1513 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1514 | WARN_ON(1); |
1515 | return D_WALK_CONTINUE; | |
1516 | } | |
1517 | ||
1518 | static void do_one_tree(struct dentry *dentry) | |
1519 | { | |
1520 | shrink_dcache_parent(dentry); | |
1521 | d_walk(dentry, dentry, umount_check, NULL); | |
1522 | d_drop(dentry); | |
1523 | dput(dentry); | |
42c32608 AV |
1524 | } |
1525 | ||
1526 | /* | |
1527 | * destroy the dentries attached to a superblock on unmounting | |
1528 | */ | |
1529 | void shrink_dcache_for_umount(struct super_block *sb) | |
1530 | { | |
1531 | struct dentry *dentry; | |
1532 | ||
9c8c10e2 | 1533 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1534 | |
1535 | dentry = sb->s_root; | |
1536 | sb->s_root = NULL; | |
9c8c10e2 | 1537 | do_one_tree(dentry); |
42c32608 | 1538 | |
f1ee6162 N |
1539 | while (!hlist_bl_empty(&sb->s_roots)) { |
1540 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); | |
9c8c10e2 | 1541 | do_one_tree(dentry); |
42c32608 AV |
1542 | } |
1543 | } | |
1544 | ||
8ed936b5 EB |
1545 | struct detach_data { |
1546 | struct select_data select; | |
1547 | struct dentry *mountpoint; | |
1548 | }; | |
1549 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1550 | { |
8ed936b5 | 1551 | struct detach_data *data = _data; |
848ac114 MS |
1552 | |
1553 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1554 | __dget_dlock(dentry); |
1555 | data->mountpoint = dentry; | |
848ac114 MS |
1556 | return D_WALK_QUIT; |
1557 | } | |
1558 | ||
8ed936b5 | 1559 | return select_collect(&data->select, dentry); |
848ac114 MS |
1560 | } |
1561 | ||
1562 | static void check_and_drop(void *_data) | |
1563 | { | |
8ed936b5 | 1564 | struct detach_data *data = _data; |
848ac114 | 1565 | |
81be24d2 | 1566 | if (!data->mountpoint && list_empty(&data->select.dispose)) |
8ed936b5 | 1567 | __d_drop(data->select.start); |
848ac114 MS |
1568 | } |
1569 | ||
1570 | /** | |
1ffe46d1 EB |
1571 | * d_invalidate - detach submounts, prune dcache, and drop |
1572 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1573 | * | |
1ffe46d1 | 1574 | * no dcache lock. |
848ac114 | 1575 | * |
8ed936b5 EB |
1576 | * The final d_drop is done as an atomic operation relative to |
1577 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1578 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1579 | */ |
5542aa2f | 1580 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1581 | { |
1ffe46d1 EB |
1582 | /* |
1583 | * If it's already been dropped, return OK. | |
1584 | */ | |
1585 | spin_lock(&dentry->d_lock); | |
1586 | if (d_unhashed(dentry)) { | |
1587 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1588 | return; |
1ffe46d1 EB |
1589 | } |
1590 | spin_unlock(&dentry->d_lock); | |
1591 | ||
848ac114 MS |
1592 | /* Negative dentries can be dropped without further checks */ |
1593 | if (!dentry->d_inode) { | |
1594 | d_drop(dentry); | |
5542aa2f | 1595 | return; |
848ac114 MS |
1596 | } |
1597 | ||
1598 | for (;;) { | |
8ed936b5 | 1599 | struct detach_data data; |
848ac114 | 1600 | |
8ed936b5 EB |
1601 | data.mountpoint = NULL; |
1602 | INIT_LIST_HEAD(&data.select.dispose); | |
1603 | data.select.start = dentry; | |
1604 | data.select.found = 0; | |
1605 | ||
1606 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1607 | |
81be24d2 | 1608 | if (!list_empty(&data.select.dispose)) |
8ed936b5 | 1609 | shrink_dentry_list(&data.select.dispose); |
81be24d2 AV |
1610 | else if (!data.mountpoint) |
1611 | return; | |
848ac114 | 1612 | |
8ed936b5 EB |
1613 | if (data.mountpoint) { |
1614 | detach_mounts(data.mountpoint); | |
1615 | dput(data.mountpoint); | |
1616 | } | |
848ac114 MS |
1617 | cond_resched(); |
1618 | } | |
848ac114 | 1619 | } |
1ffe46d1 | 1620 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1621 | |
1da177e4 | 1622 | /** |
a4464dbc AV |
1623 | * __d_alloc - allocate a dcache entry |
1624 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1625 | * @name: qstr of the name |
1626 | * | |
1627 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1628 | * available. On a success the dentry is returned. The name passed in is | |
1629 | * copied and the copy passed in may be reused after this call. | |
1630 | */ | |
1631 | ||
a4464dbc | 1632 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 | 1633 | { |
f1782c9b | 1634 | struct external_name *ext = NULL; |
1da177e4 LT |
1635 | struct dentry *dentry; |
1636 | char *dname; | |
285b102d | 1637 | int err; |
1da177e4 | 1638 | |
e12ba74d | 1639 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1640 | if (!dentry) |
1641 | return NULL; | |
1642 | ||
6326c71f LT |
1643 | /* |
1644 | * We guarantee that the inline name is always NUL-terminated. | |
1645 | * This way the memcpy() done by the name switching in rename | |
1646 | * will still always have a NUL at the end, even if we might | |
1647 | * be overwriting an internal NUL character | |
1648 | */ | |
1649 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd | 1650 | if (unlikely(!name)) { |
cdf01226 | 1651 | name = &slash_name; |
798434bd AV |
1652 | dname = dentry->d_iname; |
1653 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1654 | size_t size = offsetof(struct external_name, name[1]); |
f1782c9b RG |
1655 | |
1656 | ext = kmalloc(size + name->len, GFP_KERNEL_ACCOUNT); | |
1657 | if (!ext) { | |
1da177e4 LT |
1658 | kmem_cache_free(dentry_cache, dentry); |
1659 | return NULL; | |
1660 | } | |
f1782c9b RG |
1661 | atomic_set(&ext->u.count, 1); |
1662 | dname = ext->name; | |
1da177e4 LT |
1663 | } else { |
1664 | dname = dentry->d_iname; | |
1665 | } | |
1da177e4 LT |
1666 | |
1667 | dentry->d_name.len = name->len; | |
1668 | dentry->d_name.hash = name->hash; | |
1669 | memcpy(dname, name->name, name->len); | |
1670 | dname[name->len] = 0; | |
1671 | ||
6326c71f | 1672 | /* Make sure we always see the terminating NUL character */ |
7088efa9 | 1673 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ |
6326c71f | 1674 | |
98474236 | 1675 | dentry->d_lockref.count = 1; |
dea3667b | 1676 | dentry->d_flags = 0; |
1da177e4 | 1677 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1678 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1679 | dentry->d_inode = NULL; |
a4464dbc AV |
1680 | dentry->d_parent = dentry; |
1681 | dentry->d_sb = sb; | |
1da177e4 LT |
1682 | dentry->d_op = NULL; |
1683 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1684 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1685 | INIT_LIST_HEAD(&dentry->d_lru); |
1686 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1687 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1688 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1689 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1690 | |
285b102d MS |
1691 | if (dentry->d_op && dentry->d_op->d_init) { |
1692 | err = dentry->d_op->d_init(dentry); | |
1693 | if (err) { | |
1694 | if (dname_external(dentry)) | |
1695 | kfree(external_name(dentry)); | |
1696 | kmem_cache_free(dentry_cache, dentry); | |
1697 | return NULL; | |
1698 | } | |
1699 | } | |
1700 | ||
f1782c9b RG |
1701 | if (unlikely(ext)) { |
1702 | pg_data_t *pgdat = page_pgdat(virt_to_page(ext)); | |
1703 | mod_node_page_state(pgdat, NR_INDIRECTLY_RECLAIMABLE_BYTES, | |
1704 | ksize(ext)); | |
1705 | } | |
1706 | ||
3e880fb5 | 1707 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1708 | |
1da177e4 LT |
1709 | return dentry; |
1710 | } | |
a4464dbc AV |
1711 | |
1712 | /** | |
1713 | * d_alloc - allocate a dcache entry | |
1714 | * @parent: parent of entry to allocate | |
1715 | * @name: qstr of the name | |
1716 | * | |
1717 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1718 | * available. On a success the dentry is returned. The name passed in is | |
1719 | * copied and the copy passed in may be reused after this call. | |
1720 | */ | |
1721 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1722 | { | |
1723 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1724 | if (!dentry) | |
1725 | return NULL; | |
3d56c25e | 1726 | dentry->d_flags |= DCACHE_RCUACCESS; |
a4464dbc AV |
1727 | spin_lock(&parent->d_lock); |
1728 | /* | |
1729 | * don't need child lock because it is not subject | |
1730 | * to concurrency here | |
1731 | */ | |
1732 | __dget_dlock(parent); | |
1733 | dentry->d_parent = parent; | |
946e51f2 | 1734 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1735 | spin_unlock(&parent->d_lock); |
1736 | ||
1737 | return dentry; | |
1738 | } | |
ec4f8605 | 1739 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1740 | |
f9c34674 MS |
1741 | struct dentry *d_alloc_anon(struct super_block *sb) |
1742 | { | |
1743 | return __d_alloc(sb, NULL); | |
1744 | } | |
1745 | EXPORT_SYMBOL(d_alloc_anon); | |
1746 | ||
ba65dc5e AV |
1747 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1748 | { | |
f9c34674 | 1749 | struct dentry *dentry = d_alloc_anon(parent->d_sb); |
ba65dc5e AV |
1750 | if (dentry) { |
1751 | dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR; | |
1752 | dentry->d_parent = dget(parent); | |
1753 | } | |
1754 | return dentry; | |
1755 | } | |
1756 | ||
e1a24bb0 BF |
1757 | /** |
1758 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1759 | * @sb: the superblock | |
1760 | * @name: qstr of the name | |
1761 | * | |
1762 | * For a filesystem that just pins its dentries in memory and never | |
1763 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1764 | */ | |
4b936885 NP |
1765 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1766 | { | |
e1a24bb0 | 1767 | return __d_alloc(sb, name); |
4b936885 NP |
1768 | } |
1769 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1770 | ||
1da177e4 LT |
1771 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1772 | { | |
1773 | struct qstr q; | |
1774 | ||
1775 | q.name = name; | |
8387ff25 | 1776 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1777 | return d_alloc(parent, &q); |
1778 | } | |
ef26ca97 | 1779 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1780 | |
fb045adb NP |
1781 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1782 | { | |
6f7f7caa LT |
1783 | WARN_ON_ONCE(dentry->d_op); |
1784 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1785 | DCACHE_OP_COMPARE | |
1786 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1787 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1788 | DCACHE_OP_DELETE | |
d101a125 | 1789 | DCACHE_OP_REAL)); |
fb045adb NP |
1790 | dentry->d_op = op; |
1791 | if (!op) | |
1792 | return; | |
1793 | if (op->d_hash) | |
1794 | dentry->d_flags |= DCACHE_OP_HASH; | |
1795 | if (op->d_compare) | |
1796 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1797 | if (op->d_revalidate) | |
1798 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1799 | if (op->d_weak_revalidate) |
1800 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1801 | if (op->d_delete) |
1802 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1803 | if (op->d_prune) |
1804 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1805 | if (op->d_real) |
1806 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1807 | |
1808 | } | |
1809 | EXPORT_SYMBOL(d_set_d_op); | |
1810 | ||
df1a085a DH |
1811 | |
1812 | /* | |
1813 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1814 | * @dentry - The dentry to mark | |
1815 | * | |
1816 | * Mark a dentry as falling through to the lower layer (as set with | |
1817 | * d_pin_lower()). This flag may be recorded on the medium. | |
1818 | */ | |
1819 | void d_set_fallthru(struct dentry *dentry) | |
1820 | { | |
1821 | spin_lock(&dentry->d_lock); | |
1822 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1823 | spin_unlock(&dentry->d_lock); | |
1824 | } | |
1825 | EXPORT_SYMBOL(d_set_fallthru); | |
1826 | ||
b18825a7 DH |
1827 | static unsigned d_flags_for_inode(struct inode *inode) |
1828 | { | |
44bdb5e5 | 1829 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1830 | |
1831 | if (!inode) | |
1832 | return DCACHE_MISS_TYPE; | |
1833 | ||
1834 | if (S_ISDIR(inode->i_mode)) { | |
1835 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1836 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1837 | if (unlikely(!inode->i_op->lookup)) | |
1838 | add_flags = DCACHE_AUTODIR_TYPE; | |
1839 | else | |
1840 | inode->i_opflags |= IOP_LOOKUP; | |
1841 | } | |
44bdb5e5 DH |
1842 | goto type_determined; |
1843 | } | |
1844 | ||
1845 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1846 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1847 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1848 | goto type_determined; |
1849 | } | |
1850 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1851 | } |
1852 | ||
44bdb5e5 DH |
1853 | if (unlikely(!S_ISREG(inode->i_mode))) |
1854 | add_flags = DCACHE_SPECIAL_TYPE; | |
1855 | ||
1856 | type_determined: | |
b18825a7 DH |
1857 | if (unlikely(IS_AUTOMOUNT(inode))) |
1858 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1859 | return add_flags; | |
1860 | } | |
1861 | ||
360da900 OH |
1862 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1863 | { | |
b18825a7 | 1864 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1865 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1866 | |
b23fb0a6 | 1867 | spin_lock(&dentry->d_lock); |
de689f5e | 1868 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1869 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1870 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1871 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1872 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1873 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1874 | } |
1875 | ||
1da177e4 LT |
1876 | /** |
1877 | * d_instantiate - fill in inode information for a dentry | |
1878 | * @entry: dentry to complete | |
1879 | * @inode: inode to attach to this dentry | |
1880 | * | |
1881 | * Fill in inode information in the entry. | |
1882 | * | |
1883 | * This turns negative dentries into productive full members | |
1884 | * of society. | |
1885 | * | |
1886 | * NOTE! This assumes that the inode count has been incremented | |
1887 | * (or otherwise set) by the caller to indicate that it is now | |
1888 | * in use by the dcache. | |
1889 | */ | |
1890 | ||
1891 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1892 | { | |
946e51f2 | 1893 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1894 | if (inode) { |
b9680917 | 1895 | security_d_instantiate(entry, inode); |
873feea0 | 1896 | spin_lock(&inode->i_lock); |
de689f5e | 1897 | __d_instantiate(entry, inode); |
873feea0 | 1898 | spin_unlock(&inode->i_lock); |
de689f5e | 1899 | } |
1da177e4 | 1900 | } |
ec4f8605 | 1901 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1902 | |
b70a80e7 MS |
1903 | /** |
1904 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1905 | * @entry: dentry to complete | |
1906 | * @inode: inode to attach to this dentry | |
1907 | * | |
1908 | * Fill in inode information in the entry. If a directory alias is found, then | |
1909 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1910 | * guarantees that a directory inode may never have more than one alias. | |
1911 | */ | |
1912 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1913 | { | |
946e51f2 | 1914 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 | 1915 | |
b9680917 | 1916 | security_d_instantiate(entry, inode); |
b70a80e7 MS |
1917 | spin_lock(&inode->i_lock); |
1918 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1919 | spin_unlock(&inode->i_lock); | |
1920 | iput(inode); | |
1921 | return -EBUSY; | |
1922 | } | |
1923 | __d_instantiate(entry, inode); | |
1924 | spin_unlock(&inode->i_lock); | |
b70a80e7 MS |
1925 | |
1926 | return 0; | |
1927 | } | |
1928 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1929 | ||
adc0e91a AV |
1930 | struct dentry *d_make_root(struct inode *root_inode) |
1931 | { | |
1932 | struct dentry *res = NULL; | |
1933 | ||
1934 | if (root_inode) { | |
f9c34674 | 1935 | res = d_alloc_anon(root_inode->i_sb); |
adc0e91a AV |
1936 | if (res) |
1937 | d_instantiate(res, root_inode); | |
1938 | else | |
1939 | iput(root_inode); | |
1940 | } | |
1941 | return res; | |
1942 | } | |
1943 | EXPORT_SYMBOL(d_make_root); | |
1944 | ||
d891eedb BF |
1945 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1946 | { | |
1947 | struct dentry *alias; | |
1948 | ||
b3d9b7a3 | 1949 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1950 | return NULL; |
946e51f2 | 1951 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1952 | __dget(alias); |
1953 | return alias; | |
1954 | } | |
1955 | ||
46f72b34 SW |
1956 | /** |
1957 | * d_find_any_alias - find any alias for a given inode | |
1958 | * @inode: inode to find an alias for | |
1959 | * | |
1960 | * If any aliases exist for the given inode, take and return a | |
1961 | * reference for one of them. If no aliases exist, return %NULL. | |
1962 | */ | |
1963 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1964 | { |
1965 | struct dentry *de; | |
1966 | ||
1967 | spin_lock(&inode->i_lock); | |
1968 | de = __d_find_any_alias(inode); | |
1969 | spin_unlock(&inode->i_lock); | |
1970 | return de; | |
1971 | } | |
46f72b34 | 1972 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1973 | |
f9c34674 MS |
1974 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, |
1975 | struct inode *inode, | |
1976 | bool disconnected) | |
4ea3ada2 | 1977 | { |
9308a612 | 1978 | struct dentry *res; |
b18825a7 | 1979 | unsigned add_flags; |
4ea3ada2 | 1980 | |
f9c34674 | 1981 | security_d_instantiate(dentry, inode); |
873feea0 | 1982 | spin_lock(&inode->i_lock); |
d891eedb | 1983 | res = __d_find_any_alias(inode); |
9308a612 | 1984 | if (res) { |
873feea0 | 1985 | spin_unlock(&inode->i_lock); |
f9c34674 | 1986 | dput(dentry); |
9308a612 CH |
1987 | goto out_iput; |
1988 | } | |
1989 | ||
1990 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1991 | add_flags = d_flags_for_inode(inode); |
1992 | ||
1993 | if (disconnected) | |
1994 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1995 | |
f9c34674 MS |
1996 | spin_lock(&dentry->d_lock); |
1997 | __d_set_inode_and_type(dentry, inode, add_flags); | |
1998 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
f1ee6162 | 1999 | if (!disconnected) { |
139351f1 LT |
2000 | hlist_bl_lock(&dentry->d_sb->s_roots); |
2001 | hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots); | |
2002 | hlist_bl_unlock(&dentry->d_sb->s_roots); | |
f1ee6162 | 2003 | } |
f9c34674 | 2004 | spin_unlock(&dentry->d_lock); |
873feea0 | 2005 | spin_unlock(&inode->i_lock); |
9308a612 | 2006 | |
f9c34674 | 2007 | return dentry; |
9308a612 CH |
2008 | |
2009 | out_iput: | |
2010 | iput(inode); | |
2011 | return res; | |
4ea3ada2 | 2012 | } |
1a0a397e | 2013 | |
f9c34674 MS |
2014 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) |
2015 | { | |
2016 | return __d_instantiate_anon(dentry, inode, true); | |
2017 | } | |
2018 | EXPORT_SYMBOL(d_instantiate_anon); | |
2019 | ||
2020 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) | |
2021 | { | |
2022 | struct dentry *tmp; | |
2023 | struct dentry *res; | |
2024 | ||
2025 | if (!inode) | |
2026 | return ERR_PTR(-ESTALE); | |
2027 | if (IS_ERR(inode)) | |
2028 | return ERR_CAST(inode); | |
2029 | ||
2030 | res = d_find_any_alias(inode); | |
2031 | if (res) | |
2032 | goto out_iput; | |
2033 | ||
2034 | tmp = d_alloc_anon(inode->i_sb); | |
2035 | if (!tmp) { | |
2036 | res = ERR_PTR(-ENOMEM); | |
2037 | goto out_iput; | |
2038 | } | |
2039 | ||
2040 | return __d_instantiate_anon(tmp, inode, disconnected); | |
2041 | ||
2042 | out_iput: | |
2043 | iput(inode); | |
2044 | return res; | |
2045 | } | |
2046 | ||
1a0a397e BF |
2047 | /** |
2048 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2049 | * @inode: inode to allocate the dentry for | |
2050 | * | |
2051 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2052 | * similar open by handle operations. The returned dentry may be anonymous, | |
2053 | * or may have a full name (if the inode was already in the cache). | |
2054 | * | |
2055 | * When called on a directory inode, we must ensure that the inode only ever | |
2056 | * has one dentry. If a dentry is found, that is returned instead of | |
2057 | * allocating a new one. | |
2058 | * | |
2059 | * On successful return, the reference to the inode has been transferred | |
2060 | * to the dentry. In case of an error the reference on the inode is released. | |
2061 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2062 | * be passed in and the error will be propagated to the return value, | |
2063 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2064 | */ | |
2065 | struct dentry *d_obtain_alias(struct inode *inode) | |
2066 | { | |
f9c34674 | 2067 | return __d_obtain_alias(inode, true); |
1a0a397e | 2068 | } |
adc48720 | 2069 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 2070 | |
1a0a397e BF |
2071 | /** |
2072 | * d_obtain_root - find or allocate a dentry for a given inode | |
2073 | * @inode: inode to allocate the dentry for | |
2074 | * | |
2075 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2076 | * | |
2077 | * We must ensure that directory inodes only ever have one dentry. If a | |
2078 | * dentry is found, that is returned instead of allocating a new one. | |
2079 | * | |
2080 | * On successful return, the reference to the inode has been transferred | |
2081 | * to the dentry. In case of an error the reference on the inode is | |
2082 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2083 | * error will be propagate to the return value, with a %NULL @inode | |
2084 | * replaced by ERR_PTR(-ESTALE). | |
2085 | */ | |
2086 | struct dentry *d_obtain_root(struct inode *inode) | |
2087 | { | |
f9c34674 | 2088 | return __d_obtain_alias(inode, false); |
1a0a397e BF |
2089 | } |
2090 | EXPORT_SYMBOL(d_obtain_root); | |
2091 | ||
9403540c BN |
2092 | /** |
2093 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2094 | * @inode: the inode case-insensitive lookup has found | |
2095 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2096 | * @name: the case-exact name to be associated with the returned dentry | |
2097 | * | |
2098 | * This is to avoid filling the dcache with case-insensitive names to the | |
2099 | * same inode, only the actual correct case is stored in the dcache for | |
2100 | * case-insensitive filesystems. | |
2101 | * | |
2102 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2103 | * already exists in in the dcache, use it and return it. | |
2104 | * | |
2105 | * If no entry exists with the exact case name, allocate new dentry with | |
2106 | * the exact case, and return the spliced entry. | |
2107 | */ | |
e45b590b | 2108 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2109 | struct qstr *name) |
2110 | { | |
d9171b93 | 2111 | struct dentry *found, *res; |
9403540c | 2112 | |
b6520c81 CH |
2113 | /* |
2114 | * First check if a dentry matching the name already exists, | |
2115 | * if not go ahead and create it now. | |
2116 | */ | |
9403540c | 2117 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2118 | if (found) { |
2119 | iput(inode); | |
2120 | return found; | |
2121 | } | |
2122 | if (d_in_lookup(dentry)) { | |
2123 | found = d_alloc_parallel(dentry->d_parent, name, | |
2124 | dentry->d_wait); | |
2125 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2126 | iput(inode); | |
2127 | return found; | |
9403540c | 2128 | } |
d9171b93 AV |
2129 | } else { |
2130 | found = d_alloc(dentry->d_parent, name); | |
2131 | if (!found) { | |
2132 | iput(inode); | |
2133 | return ERR_PTR(-ENOMEM); | |
2134 | } | |
2135 | } | |
2136 | res = d_splice_alias(inode, found); | |
2137 | if (res) { | |
2138 | dput(found); | |
2139 | return res; | |
9403540c | 2140 | } |
4f522a24 | 2141 | return found; |
9403540c | 2142 | } |
ec4f8605 | 2143 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2144 | |
12f8ad4b | 2145 | |
d4c91a8f AV |
2146 | static inline bool d_same_name(const struct dentry *dentry, |
2147 | const struct dentry *parent, | |
2148 | const struct qstr *name) | |
12f8ad4b | 2149 | { |
d4c91a8f AV |
2150 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2151 | if (dentry->d_name.len != name->len) | |
2152 | return false; | |
2153 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2154 | } |
6fa67e70 | 2155 | return parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2156 | dentry->d_name.len, dentry->d_name.name, |
2157 | name) == 0; | |
12f8ad4b LT |
2158 | } |
2159 | ||
31e6b01f NP |
2160 | /** |
2161 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2162 | * @parent: parent dentry | |
2163 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2164 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2165 | * Returns: dentry, or NULL |
2166 | * | |
2167 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2168 | * resolution (store-free path walking) design described in | |
2169 | * Documentation/filesystems/path-lookup.txt. | |
2170 | * | |
2171 | * This is not to be used outside core vfs. | |
2172 | * | |
2173 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2174 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2175 | * without taking d_lock and checking d_seq sequence count against @seq | |
2176 | * returned here. | |
2177 | * | |
15570086 | 2178 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2179 | * function. |
2180 | * | |
2181 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2182 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2183 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2184 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2185 | * |
2186 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2187 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2188 | */ |
8966be90 LT |
2189 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2190 | const struct qstr *name, | |
da53be12 | 2191 | unsigned *seqp) |
31e6b01f | 2192 | { |
26fe5750 | 2193 | u64 hashlen = name->hash_len; |
31e6b01f | 2194 | const unsigned char *str = name->name; |
8387ff25 | 2195 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2196 | struct hlist_bl_node *node; |
31e6b01f NP |
2197 | struct dentry *dentry; |
2198 | ||
2199 | /* | |
2200 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2201 | * required to prevent single threaded performance regressions | |
2202 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2203 | * Keep the two functions in sync. | |
2204 | */ | |
2205 | ||
2206 | /* | |
2207 | * The hash list is protected using RCU. | |
2208 | * | |
2209 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2210 | * races with d_move(). | |
2211 | * | |
2212 | * It is possible that concurrent renames can mess up our list | |
2213 | * walk here and result in missing our dentry, resulting in the | |
2214 | * false-negative result. d_lookup() protects against concurrent | |
2215 | * renames using rename_lock seqlock. | |
2216 | * | |
b0a4bb83 | 2217 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2218 | */ |
b07ad996 | 2219 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2220 | unsigned seq; |
31e6b01f | 2221 | |
31e6b01f | 2222 | seqretry: |
12f8ad4b LT |
2223 | /* |
2224 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2225 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2226 | * |
2227 | * The caller must perform a seqcount check in order | |
da53be12 | 2228 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2229 | * |
2230 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2231 | * we don't wait for the sequence count to stabilize if it | |
2232 | * is in the middle of a sequence change. If we do the slow | |
2233 | * dentry compare, we will do seqretries until it is stable, | |
2234 | * and if we end up with a successful lookup, we actually | |
2235 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2236 | * |
2237 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2238 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2239 | */ |
2240 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2241 | if (dentry->d_parent != parent) |
2242 | continue; | |
2e321806 LT |
2243 | if (d_unhashed(dentry)) |
2244 | continue; | |
12f8ad4b | 2245 | |
830c0f0e | 2246 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
d4c91a8f AV |
2247 | int tlen; |
2248 | const char *tname; | |
26fe5750 LT |
2249 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2250 | continue; | |
d4c91a8f AV |
2251 | tlen = dentry->d_name.len; |
2252 | tname = dentry->d_name.name; | |
2253 | /* we want a consistent (name,len) pair */ | |
2254 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2255 | cpu_relax(); | |
12f8ad4b LT |
2256 | goto seqretry; |
2257 | } | |
6fa67e70 | 2258 | if (parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2259 | tlen, tname, name) != 0) |
2260 | continue; | |
2261 | } else { | |
2262 | if (dentry->d_name.hash_len != hashlen) | |
2263 | continue; | |
2264 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2265 | continue; | |
31e6b01f | 2266 | } |
da53be12 | 2267 | *seqp = seq; |
d4c91a8f | 2268 | return dentry; |
31e6b01f NP |
2269 | } |
2270 | return NULL; | |
2271 | } | |
2272 | ||
1da177e4 LT |
2273 | /** |
2274 | * d_lookup - search for a dentry | |
2275 | * @parent: parent dentry | |
2276 | * @name: qstr of name we wish to find | |
b04f784e | 2277 | * Returns: dentry, or NULL |
1da177e4 | 2278 | * |
b04f784e NP |
2279 | * d_lookup searches the children of the parent dentry for the name in |
2280 | * question. If the dentry is found its reference count is incremented and the | |
2281 | * dentry is returned. The caller must use dput to free the entry when it has | |
2282 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2283 | */ |
da2d8455 | 2284 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2285 | { |
31e6b01f | 2286 | struct dentry *dentry; |
949854d0 | 2287 | unsigned seq; |
1da177e4 | 2288 | |
b8314f93 DY |
2289 | do { |
2290 | seq = read_seqbegin(&rename_lock); | |
2291 | dentry = __d_lookup(parent, name); | |
2292 | if (dentry) | |
1da177e4 LT |
2293 | break; |
2294 | } while (read_seqretry(&rename_lock, seq)); | |
2295 | return dentry; | |
2296 | } | |
ec4f8605 | 2297 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2298 | |
31e6b01f | 2299 | /** |
b04f784e NP |
2300 | * __d_lookup - search for a dentry (racy) |
2301 | * @parent: parent dentry | |
2302 | * @name: qstr of name we wish to find | |
2303 | * Returns: dentry, or NULL | |
2304 | * | |
2305 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2306 | * false-negative result due to unrelated rename activity. | |
2307 | * | |
2308 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2309 | * however it must be used carefully, eg. with a following d_lookup in | |
2310 | * the case of failure. | |
2311 | * | |
2312 | * __d_lookup callers must be commented. | |
2313 | */ | |
a713ca2a | 2314 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2315 | { |
1da177e4 | 2316 | unsigned int hash = name->hash; |
8387ff25 | 2317 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2318 | struct hlist_bl_node *node; |
31e6b01f | 2319 | struct dentry *found = NULL; |
665a7583 | 2320 | struct dentry *dentry; |
1da177e4 | 2321 | |
31e6b01f NP |
2322 | /* |
2323 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2324 | * required to prevent single threaded performance regressions | |
2325 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2326 | * Keep the two functions in sync. | |
2327 | */ | |
2328 | ||
b04f784e NP |
2329 | /* |
2330 | * The hash list is protected using RCU. | |
2331 | * | |
2332 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2333 | * with d_move(). | |
2334 | * | |
2335 | * It is possible that concurrent renames can mess up our list | |
2336 | * walk here and result in missing our dentry, resulting in the | |
2337 | * false-negative result. d_lookup() protects against concurrent | |
2338 | * renames using rename_lock seqlock. | |
2339 | * | |
b0a4bb83 | 2340 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2341 | */ |
1da177e4 LT |
2342 | rcu_read_lock(); |
2343 | ||
b07ad996 | 2344 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2345 | |
1da177e4 LT |
2346 | if (dentry->d_name.hash != hash) |
2347 | continue; | |
1da177e4 LT |
2348 | |
2349 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2350 | if (dentry->d_parent != parent) |
2351 | goto next; | |
d0185c08 LT |
2352 | if (d_unhashed(dentry)) |
2353 | goto next; | |
2354 | ||
d4c91a8f AV |
2355 | if (!d_same_name(dentry, parent, name)) |
2356 | goto next; | |
1da177e4 | 2357 | |
98474236 | 2358 | dentry->d_lockref.count++; |
d0185c08 | 2359 | found = dentry; |
1da177e4 LT |
2360 | spin_unlock(&dentry->d_lock); |
2361 | break; | |
2362 | next: | |
2363 | spin_unlock(&dentry->d_lock); | |
2364 | } | |
2365 | rcu_read_unlock(); | |
2366 | ||
2367 | return found; | |
2368 | } | |
2369 | ||
3e7e241f EB |
2370 | /** |
2371 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2372 | * @dir: Directory to search in | |
2373 | * @name: qstr of name we wish to find | |
2374 | * | |
4f522a24 | 2375 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2376 | */ |
2377 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2378 | { | |
3e7e241f EB |
2379 | /* |
2380 | * Check for a fs-specific hash function. Note that we must | |
2381 | * calculate the standard hash first, as the d_op->d_hash() | |
2382 | * routine may choose to leave the hash value unchanged. | |
2383 | */ | |
8387ff25 | 2384 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2385 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2386 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2387 | if (unlikely(err < 0)) |
2388 | return ERR_PTR(err); | |
3e7e241f | 2389 | } |
4f522a24 | 2390 | return d_lookup(dir, name); |
3e7e241f | 2391 | } |
4f522a24 | 2392 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2393 | |
1da177e4 LT |
2394 | /* |
2395 | * When a file is deleted, we have two options: | |
2396 | * - turn this dentry into a negative dentry | |
2397 | * - unhash this dentry and free it. | |
2398 | * | |
2399 | * Usually, we want to just turn this into | |
2400 | * a negative dentry, but if anybody else is | |
2401 | * currently using the dentry or the inode | |
2402 | * we can't do that and we fall back on removing | |
2403 | * it from the hash queues and waiting for | |
2404 | * it to be deleted later when it has no users | |
2405 | */ | |
2406 | ||
2407 | /** | |
2408 | * d_delete - delete a dentry | |
2409 | * @dentry: The dentry to delete | |
2410 | * | |
2411 | * Turn the dentry into a negative dentry if possible, otherwise | |
2412 | * remove it from the hash queues so it can be deleted later | |
2413 | */ | |
2414 | ||
2415 | void d_delete(struct dentry * dentry) | |
2416 | { | |
c19457f0 AV |
2417 | struct inode *inode = dentry->d_inode; |
2418 | int isdir = d_is_dir(dentry); | |
2419 | ||
2420 | spin_lock(&inode->i_lock); | |
2421 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2422 | /* |
2423 | * Are we the only user? | |
2424 | */ | |
98474236 | 2425 | if (dentry->d_lockref.count == 1) { |
13e3c5e5 | 2426 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2427 | dentry_unlink_inode(dentry); |
c19457f0 | 2428 | } else { |
1da177e4 | 2429 | __d_drop(dentry); |
c19457f0 AV |
2430 | spin_unlock(&dentry->d_lock); |
2431 | spin_unlock(&inode->i_lock); | |
2432 | } | |
7a91bf7f | 2433 | fsnotify_nameremove(dentry, isdir); |
1da177e4 | 2434 | } |
ec4f8605 | 2435 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2436 | |
15d3c589 | 2437 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2438 | { |
15d3c589 | 2439 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
61647823 | 2440 | |
1879fd6a | 2441 | hlist_bl_lock(b); |
b07ad996 | 2442 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2443 | hlist_bl_unlock(b); |
1da177e4 LT |
2444 | } |
2445 | ||
2446 | /** | |
2447 | * d_rehash - add an entry back to the hash | |
2448 | * @entry: dentry to add to the hash | |
2449 | * | |
2450 | * Adds a dentry to the hash according to its name. | |
2451 | */ | |
2452 | ||
2453 | void d_rehash(struct dentry * entry) | |
2454 | { | |
1da177e4 | 2455 | spin_lock(&entry->d_lock); |
15d3c589 | 2456 | __d_rehash(entry); |
1da177e4 | 2457 | spin_unlock(&entry->d_lock); |
1da177e4 | 2458 | } |
ec4f8605 | 2459 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2460 | |
84e710da AV |
2461 | static inline unsigned start_dir_add(struct inode *dir) |
2462 | { | |
2463 | ||
2464 | for (;;) { | |
2465 | unsigned n = dir->i_dir_seq; | |
2466 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2467 | return n; | |
2468 | cpu_relax(); | |
2469 | } | |
2470 | } | |
2471 | ||
2472 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2473 | { | |
2474 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2475 | } | |
2476 | ||
d9171b93 AV |
2477 | static void d_wait_lookup(struct dentry *dentry) |
2478 | { | |
2479 | if (d_in_lookup(dentry)) { | |
2480 | DECLARE_WAITQUEUE(wait, current); | |
2481 | add_wait_queue(dentry->d_wait, &wait); | |
2482 | do { | |
2483 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2484 | spin_unlock(&dentry->d_lock); | |
2485 | schedule(); | |
2486 | spin_lock(&dentry->d_lock); | |
2487 | } while (d_in_lookup(dentry)); | |
2488 | } | |
2489 | } | |
2490 | ||
94bdd655 | 2491 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2492 | const struct qstr *name, |
2493 | wait_queue_head_t *wq) | |
94bdd655 | 2494 | { |
94bdd655 | 2495 | unsigned int hash = name->hash; |
94bdd655 AV |
2496 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2497 | struct hlist_bl_node *node; | |
2498 | struct dentry *new = d_alloc(parent, name); | |
2499 | struct dentry *dentry; | |
2500 | unsigned seq, r_seq, d_seq; | |
2501 | ||
2502 | if (unlikely(!new)) | |
2503 | return ERR_PTR(-ENOMEM); | |
2504 | ||
2505 | retry: | |
2506 | rcu_read_lock(); | |
015555fd | 2507 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
94bdd655 AV |
2508 | r_seq = read_seqbegin(&rename_lock); |
2509 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2510 | if (unlikely(dentry)) { | |
2511 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2512 | rcu_read_unlock(); | |
2513 | goto retry; | |
2514 | } | |
2515 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2516 | rcu_read_unlock(); | |
2517 | dput(dentry); | |
2518 | goto retry; | |
2519 | } | |
2520 | rcu_read_unlock(); | |
2521 | dput(new); | |
2522 | return dentry; | |
2523 | } | |
2524 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2525 | rcu_read_unlock(); | |
2526 | goto retry; | |
2527 | } | |
015555fd WD |
2528 | |
2529 | if (unlikely(seq & 1)) { | |
2530 | rcu_read_unlock(); | |
2531 | goto retry; | |
2532 | } | |
2533 | ||
94bdd655 | 2534 | hlist_bl_lock(b); |
8cc07c80 | 2535 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
94bdd655 AV |
2536 | hlist_bl_unlock(b); |
2537 | rcu_read_unlock(); | |
2538 | goto retry; | |
2539 | } | |
94bdd655 AV |
2540 | /* |
2541 | * No changes for the parent since the beginning of d_lookup(). | |
2542 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2543 | * any potential in-lookup matches are going to stay here until | |
2544 | * we unlock the chain. All fields are stable in everything | |
2545 | * we encounter. | |
2546 | */ | |
2547 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2548 | if (dentry->d_name.hash != hash) | |
2549 | continue; | |
2550 | if (dentry->d_parent != parent) | |
2551 | continue; | |
d4c91a8f AV |
2552 | if (!d_same_name(dentry, parent, name)) |
2553 | continue; | |
94bdd655 | 2554 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2555 | /* now we can try to grab a reference */ |
2556 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2557 | rcu_read_unlock(); | |
2558 | goto retry; | |
2559 | } | |
2560 | ||
2561 | rcu_read_unlock(); | |
2562 | /* | |
2563 | * somebody is likely to be still doing lookup for it; | |
2564 | * wait for them to finish | |
2565 | */ | |
d9171b93 AV |
2566 | spin_lock(&dentry->d_lock); |
2567 | d_wait_lookup(dentry); | |
2568 | /* | |
2569 | * it's not in-lookup anymore; in principle we should repeat | |
2570 | * everything from dcache lookup, but it's likely to be what | |
2571 | * d_lookup() would've found anyway. If it is, just return it; | |
2572 | * otherwise we really have to repeat the whole thing. | |
2573 | */ | |
2574 | if (unlikely(dentry->d_name.hash != hash)) | |
2575 | goto mismatch; | |
2576 | if (unlikely(dentry->d_parent != parent)) | |
2577 | goto mismatch; | |
2578 | if (unlikely(d_unhashed(dentry))) | |
2579 | goto mismatch; | |
d4c91a8f AV |
2580 | if (unlikely(!d_same_name(dentry, parent, name))) |
2581 | goto mismatch; | |
d9171b93 AV |
2582 | /* OK, it *is* a hashed match; return it */ |
2583 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2584 | dput(new); |
2585 | return dentry; | |
2586 | } | |
e7d6ef97 | 2587 | rcu_read_unlock(); |
94bdd655 AV |
2588 | /* we can't take ->d_lock here; it's OK, though. */ |
2589 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2590 | new->d_wait = wq; |
94bdd655 AV |
2591 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2592 | hlist_bl_unlock(b); | |
2593 | return new; | |
d9171b93 AV |
2594 | mismatch: |
2595 | spin_unlock(&dentry->d_lock); | |
2596 | dput(dentry); | |
2597 | goto retry; | |
94bdd655 AV |
2598 | } |
2599 | EXPORT_SYMBOL(d_alloc_parallel); | |
2600 | ||
85c7f810 AV |
2601 | void __d_lookup_done(struct dentry *dentry) |
2602 | { | |
94bdd655 AV |
2603 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2604 | dentry->d_name.hash); | |
2605 | hlist_bl_lock(b); | |
85c7f810 | 2606 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2607 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2608 | wake_up_all(dentry->d_wait); |
2609 | dentry->d_wait = NULL; | |
94bdd655 AV |
2610 | hlist_bl_unlock(b); |
2611 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2612 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2613 | } |
2614 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2615 | |
2616 | /* inode->i_lock held if inode is non-NULL */ | |
2617 | ||
2618 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2619 | { | |
84e710da AV |
2620 | struct inode *dir = NULL; |
2621 | unsigned n; | |
0568d705 | 2622 | spin_lock(&dentry->d_lock); |
84e710da AV |
2623 | if (unlikely(d_in_lookup(dentry))) { |
2624 | dir = dentry->d_parent->d_inode; | |
2625 | n = start_dir_add(dir); | |
85c7f810 | 2626 | __d_lookup_done(dentry); |
84e710da | 2627 | } |
ed782b5a | 2628 | if (inode) { |
0568d705 AV |
2629 | unsigned add_flags = d_flags_for_inode(inode); |
2630 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2631 | raw_write_seqcount_begin(&dentry->d_seq); | |
2632 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2633 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2634 | fsnotify_update_flags(dentry); |
ed782b5a | 2635 | } |
15d3c589 | 2636 | __d_rehash(dentry); |
84e710da AV |
2637 | if (dir) |
2638 | end_dir_add(dir, n); | |
0568d705 AV |
2639 | spin_unlock(&dentry->d_lock); |
2640 | if (inode) | |
2641 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2642 | } |
2643 | ||
34d0d19d AV |
2644 | /** |
2645 | * d_add - add dentry to hash queues | |
2646 | * @entry: dentry to add | |
2647 | * @inode: The inode to attach to this dentry | |
2648 | * | |
2649 | * This adds the entry to the hash queues and initializes @inode. | |
2650 | * The entry was actually filled in earlier during d_alloc(). | |
2651 | */ | |
2652 | ||
2653 | void d_add(struct dentry *entry, struct inode *inode) | |
2654 | { | |
b9680917 AV |
2655 | if (inode) { |
2656 | security_d_instantiate(entry, inode); | |
ed782b5a | 2657 | spin_lock(&inode->i_lock); |
b9680917 | 2658 | } |
ed782b5a | 2659 | __d_add(entry, inode); |
34d0d19d AV |
2660 | } |
2661 | EXPORT_SYMBOL(d_add); | |
2662 | ||
668d0cd5 AV |
2663 | /** |
2664 | * d_exact_alias - find and hash an exact unhashed alias | |
2665 | * @entry: dentry to add | |
2666 | * @inode: The inode to go with this dentry | |
2667 | * | |
2668 | * If an unhashed dentry with the same name/parent and desired | |
2669 | * inode already exists, hash and return it. Otherwise, return | |
2670 | * NULL. | |
2671 | * | |
2672 | * Parent directory should be locked. | |
2673 | */ | |
2674 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2675 | { | |
2676 | struct dentry *alias; | |
668d0cd5 AV |
2677 | unsigned int hash = entry->d_name.hash; |
2678 | ||
2679 | spin_lock(&inode->i_lock); | |
2680 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2681 | /* | |
2682 | * Don't need alias->d_lock here, because aliases with | |
2683 | * d_parent == entry->d_parent are not subject to name or | |
2684 | * parent changes, because the parent inode i_mutex is held. | |
2685 | */ | |
2686 | if (alias->d_name.hash != hash) | |
2687 | continue; | |
2688 | if (alias->d_parent != entry->d_parent) | |
2689 | continue; | |
d4c91a8f | 2690 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2691 | continue; |
2692 | spin_lock(&alias->d_lock); | |
2693 | if (!d_unhashed(alias)) { | |
2694 | spin_unlock(&alias->d_lock); | |
2695 | alias = NULL; | |
2696 | } else { | |
2697 | __dget_dlock(alias); | |
15d3c589 | 2698 | __d_rehash(alias); |
668d0cd5 AV |
2699 | spin_unlock(&alias->d_lock); |
2700 | } | |
2701 | spin_unlock(&inode->i_lock); | |
2702 | return alias; | |
2703 | } | |
2704 | spin_unlock(&inode->i_lock); | |
2705 | return NULL; | |
2706 | } | |
2707 | EXPORT_SYMBOL(d_exact_alias); | |
2708 | ||
fb2d5b86 NP |
2709 | /** |
2710 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2711 | * @dentry: dentry to be updated | |
2712 | * @name: new name | |
2713 | * | |
2714 | * Update a case insensitive dentry with new case of name. | |
2715 | * | |
2716 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2717 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2718 | * lengths). | |
2719 | * | |
2720 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2721 | * keep renames and concurrent inserts, and readdir(2) away). | |
2722 | */ | |
9aba36de | 2723 | void dentry_update_name_case(struct dentry *dentry, const struct qstr *name) |
fb2d5b86 | 2724 | { |
5955102c | 2725 | BUG_ON(!inode_is_locked(dentry->d_parent->d_inode)); |
fb2d5b86 NP |
2726 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2727 | ||
fb2d5b86 | 2728 | spin_lock(&dentry->d_lock); |
31e6b01f | 2729 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2730 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2731 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2732 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2733 | } |
2734 | EXPORT_SYMBOL(dentry_update_name_case); | |
2735 | ||
8d85b484 | 2736 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2737 | { |
8d85b484 AV |
2738 | if (unlikely(dname_external(target))) { |
2739 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2740 | /* |
2741 | * Both external: swap the pointers | |
2742 | */ | |
9a8d5bb4 | 2743 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2744 | } else { |
2745 | /* | |
2746 | * dentry:internal, target:external. Steal target's | |
2747 | * storage and make target internal. | |
2748 | */ | |
321bcf92 BF |
2749 | memcpy(target->d_iname, dentry->d_name.name, |
2750 | dentry->d_name.len + 1); | |
1da177e4 LT |
2751 | dentry->d_name.name = target->d_name.name; |
2752 | target->d_name.name = target->d_iname; | |
2753 | } | |
2754 | } else { | |
8d85b484 | 2755 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2756 | /* |
2757 | * dentry:external, target:internal. Give dentry's | |
2758 | * storage to target and make dentry internal | |
2759 | */ | |
2760 | memcpy(dentry->d_iname, target->d_name.name, | |
2761 | target->d_name.len + 1); | |
2762 | target->d_name.name = dentry->d_name.name; | |
2763 | dentry->d_name.name = dentry->d_iname; | |
2764 | } else { | |
2765 | /* | |
da1ce067 | 2766 | * Both are internal. |
1da177e4 | 2767 | */ |
da1ce067 MS |
2768 | unsigned int i; |
2769 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2770 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2771 | swap(((long *) &dentry->d_iname)[i], | |
2772 | ((long *) &target->d_iname)[i]); | |
2773 | } | |
1da177e4 LT |
2774 | } |
2775 | } | |
a28ddb87 | 2776 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2777 | } |
2778 | ||
8d85b484 AV |
2779 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2780 | { | |
2781 | struct external_name *old_name = NULL; | |
2782 | if (unlikely(dname_external(dentry))) | |
2783 | old_name = external_name(dentry); | |
2784 | if (unlikely(dname_external(target))) { | |
2785 | atomic_inc(&external_name(target)->u.count); | |
2786 | dentry->d_name = target->d_name; | |
2787 | } else { | |
2788 | memcpy(dentry->d_iname, target->d_name.name, | |
2789 | target->d_name.len + 1); | |
2790 | dentry->d_name.name = dentry->d_iname; | |
2791 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2792 | } | |
2793 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
f1782c9b | 2794 | call_rcu(&old_name->u.head, __d_free_external_name); |
8d85b484 AV |
2795 | } |
2796 | ||
9eaef27b | 2797 | /* |
18367501 | 2798 | * __d_move - move a dentry |
1da177e4 LT |
2799 | * @dentry: entry to move |
2800 | * @target: new dentry | |
da1ce067 | 2801 | * @exchange: exchange the two dentries |
1da177e4 LT |
2802 | * |
2803 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2804 | * dcache entries should not be moved in this way. Caller must hold |
2805 | * rename_lock, the i_mutex of the source and target directories, | |
2806 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2807 | */ |
da1ce067 MS |
2808 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2809 | bool exchange) | |
1da177e4 | 2810 | { |
42177007 | 2811 | struct dentry *old_parent, *p; |
84e710da AV |
2812 | struct inode *dir = NULL; |
2813 | unsigned n; | |
1da177e4 | 2814 | |
42177007 AV |
2815 | WARN_ON(!dentry->d_inode); |
2816 | if (WARN_ON(dentry == target)) | |
2817 | return; | |
2818 | ||
2fd6b7f5 | 2819 | BUG_ON(d_ancestor(target, dentry)); |
42177007 AV |
2820 | old_parent = dentry->d_parent; |
2821 | p = d_ancestor(old_parent, target); | |
2822 | if (IS_ROOT(dentry)) { | |
2823 | BUG_ON(p); | |
2824 | spin_lock(&target->d_parent->d_lock); | |
2825 | } else if (!p) { | |
2826 | /* target is not a descendent of dentry->d_parent */ | |
2827 | spin_lock(&target->d_parent->d_lock); | |
2828 | spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED); | |
2829 | } else { | |
2830 | BUG_ON(p == dentry); | |
2831 | spin_lock(&old_parent->d_lock); | |
2832 | if (p != target) | |
2833 | spin_lock_nested(&target->d_parent->d_lock, | |
2834 | DENTRY_D_LOCK_NESTED); | |
2835 | } | |
2836 | spin_lock_nested(&dentry->d_lock, 2); | |
2837 | spin_lock_nested(&target->d_lock, 3); | |
2fd6b7f5 | 2838 | |
84e710da AV |
2839 | if (unlikely(d_in_lookup(target))) { |
2840 | dir = target->d_parent->d_inode; | |
2841 | n = start_dir_add(dir); | |
85c7f810 | 2842 | __d_lookup_done(target); |
84e710da | 2843 | } |
1da177e4 | 2844 | |
31e6b01f | 2845 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2846 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2847 | |
15d3c589 | 2848 | /* unhash both */ |
0632a9ac AV |
2849 | if (!d_unhashed(dentry)) |
2850 | ___d_drop(dentry); | |
2851 | if (!d_unhashed(target)) | |
2852 | ___d_drop(target); | |
1da177e4 | 2853 | |
076515fc AV |
2854 | /* ... and switch them in the tree */ |
2855 | dentry->d_parent = target->d_parent; | |
2856 | if (!exchange) { | |
8d85b484 | 2857 | copy_name(dentry, target); |
61647823 | 2858 | target->d_hash.pprev = NULL; |
076515fc AV |
2859 | dentry->d_parent->d_lockref.count++; |
2860 | if (dentry == old_parent) | |
2861 | dentry->d_flags |= DCACHE_RCUACCESS; | |
2862 | else | |
2863 | WARN_ON(!--old_parent->d_lockref.count); | |
1da177e4 | 2864 | } else { |
076515fc AV |
2865 | target->d_parent = old_parent; |
2866 | swap_names(dentry, target); | |
946e51f2 | 2867 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
076515fc AV |
2868 | __d_rehash(target); |
2869 | fsnotify_update_flags(target); | |
1da177e4 | 2870 | } |
076515fc AV |
2871 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); |
2872 | __d_rehash(dentry); | |
2873 | fsnotify_update_flags(dentry); | |
1da177e4 | 2874 | |
31e6b01f NP |
2875 | write_seqcount_end(&target->d_seq); |
2876 | write_seqcount_end(&dentry->d_seq); | |
2877 | ||
84e710da AV |
2878 | if (dir) |
2879 | end_dir_add(dir, n); | |
076515fc AV |
2880 | |
2881 | if (dentry->d_parent != old_parent) | |
2882 | spin_unlock(&dentry->d_parent->d_lock); | |
2883 | if (dentry != old_parent) | |
2884 | spin_unlock(&old_parent->d_lock); | |
2885 | spin_unlock(&target->d_lock); | |
2886 | spin_unlock(&dentry->d_lock); | |
18367501 AV |
2887 | } |
2888 | ||
2889 | /* | |
2890 | * d_move - move a dentry | |
2891 | * @dentry: entry to move | |
2892 | * @target: new dentry | |
2893 | * | |
2894 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2895 | * dcache entries should not be moved in this way. See the locking |
2896 | * requirements for __d_move. | |
18367501 AV |
2897 | */ |
2898 | void d_move(struct dentry *dentry, struct dentry *target) | |
2899 | { | |
2900 | write_seqlock(&rename_lock); | |
da1ce067 | 2901 | __d_move(dentry, target, false); |
1da177e4 | 2902 | write_sequnlock(&rename_lock); |
9eaef27b | 2903 | } |
ec4f8605 | 2904 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2905 | |
da1ce067 MS |
2906 | /* |
2907 | * d_exchange - exchange two dentries | |
2908 | * @dentry1: first dentry | |
2909 | * @dentry2: second dentry | |
2910 | */ | |
2911 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2912 | { | |
2913 | write_seqlock(&rename_lock); | |
2914 | ||
2915 | WARN_ON(!dentry1->d_inode); | |
2916 | WARN_ON(!dentry2->d_inode); | |
2917 | WARN_ON(IS_ROOT(dentry1)); | |
2918 | WARN_ON(IS_ROOT(dentry2)); | |
2919 | ||
2920 | __d_move(dentry1, dentry2, true); | |
2921 | ||
2922 | write_sequnlock(&rename_lock); | |
2923 | } | |
2924 | ||
e2761a11 OH |
2925 | /** |
2926 | * d_ancestor - search for an ancestor | |
2927 | * @p1: ancestor dentry | |
2928 | * @p2: child dentry | |
2929 | * | |
2930 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2931 | * an ancestor of p2, else NULL. | |
9eaef27b | 2932 | */ |
e2761a11 | 2933 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2934 | { |
2935 | struct dentry *p; | |
2936 | ||
871c0067 | 2937 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2938 | if (p->d_parent == p1) |
e2761a11 | 2939 | return p; |
9eaef27b | 2940 | } |
e2761a11 | 2941 | return NULL; |
9eaef27b TM |
2942 | } |
2943 | ||
2944 | /* | |
2945 | * This helper attempts to cope with remotely renamed directories | |
2946 | * | |
2947 | * It assumes that the caller is already holding | |
a03e283b | 2948 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2949 | * |
2950 | * Note: If ever the locking in lock_rename() changes, then please | |
2951 | * remember to update this too... | |
9eaef27b | 2952 | */ |
b5ae6b15 | 2953 | static int __d_unalias(struct inode *inode, |
873feea0 | 2954 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2955 | { |
9902af79 AV |
2956 | struct mutex *m1 = NULL; |
2957 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2958 | int ret = -ESTALE; |
9eaef27b TM |
2959 | |
2960 | /* If alias and dentry share a parent, then no extra locks required */ | |
2961 | if (alias->d_parent == dentry->d_parent) | |
2962 | goto out_unalias; | |
2963 | ||
9eaef27b | 2964 | /* See lock_rename() */ |
9eaef27b TM |
2965 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2966 | goto out_err; | |
2967 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2968 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2969 | goto out_err; |
9902af79 | 2970 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 2971 | out_unalias: |
8ed936b5 | 2972 | __d_move(alias, dentry, false); |
b5ae6b15 | 2973 | ret = 0; |
9eaef27b | 2974 | out_err: |
9eaef27b | 2975 | if (m2) |
9902af79 | 2976 | up_read(m2); |
9eaef27b TM |
2977 | if (m1) |
2978 | mutex_unlock(m1); | |
2979 | return ret; | |
2980 | } | |
2981 | ||
3f70bd51 BF |
2982 | /** |
2983 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2984 | * @inode: the inode which may have a disconnected dentry | |
2985 | * @dentry: a negative dentry which we want to point to the inode. | |
2986 | * | |
da093a9b BF |
2987 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2988 | * place of the given dentry and return it, else simply d_add the inode | |
2989 | * to the dentry and return NULL. | |
3f70bd51 | 2990 | * |
908790fa BF |
2991 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2992 | * we should error out: directories can't have multiple aliases. | |
2993 | * | |
3f70bd51 BF |
2994 | * This is needed in the lookup routine of any filesystem that is exportable |
2995 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2996 | * | |
2997 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2998 | * is returned. This matches the expected return value of ->lookup. | |
2999 | * | |
3000 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
3001 | * In that case, we know that the inode will be a regular file, and also this | |
3002 | * will only occur during atomic_open. So we need to check for the dentry | |
3003 | * being already hashed only in the final case. | |
3004 | */ | |
3005 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3006 | { | |
3f70bd51 BF |
3007 | if (IS_ERR(inode)) |
3008 | return ERR_CAST(inode); | |
3009 | ||
770bfad8 DH |
3010 | BUG_ON(!d_unhashed(dentry)); |
3011 | ||
de689f5e | 3012 | if (!inode) |
b5ae6b15 | 3013 | goto out; |
de689f5e | 3014 | |
b9680917 | 3015 | security_d_instantiate(dentry, inode); |
873feea0 | 3016 | spin_lock(&inode->i_lock); |
9eaef27b | 3017 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
3018 | struct dentry *new = __d_find_any_alias(inode); |
3019 | if (unlikely(new)) { | |
a03e283b EB |
3020 | /* The reference to new ensures it remains an alias */ |
3021 | spin_unlock(&inode->i_lock); | |
18367501 | 3022 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3023 | if (unlikely(d_ancestor(new, dentry))) { |
3024 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3025 | dput(new); |
3026 | new = ERR_PTR(-ELOOP); | |
3027 | pr_warn_ratelimited( | |
3028 | "VFS: Lookup of '%s' in %s %s" | |
3029 | " would have caused loop\n", | |
3030 | dentry->d_name.name, | |
3031 | inode->i_sb->s_type->name, | |
3032 | inode->i_sb->s_id); | |
3033 | } else if (!IS_ROOT(new)) { | |
076515fc | 3034 | struct dentry *old_parent = dget(new->d_parent); |
b5ae6b15 | 3035 | int err = __d_unalias(inode, dentry, new); |
18367501 | 3036 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3037 | if (err) { |
3038 | dput(new); | |
3039 | new = ERR_PTR(err); | |
3040 | } | |
076515fc | 3041 | dput(old_parent); |
18367501 | 3042 | } else { |
b5ae6b15 AV |
3043 | __d_move(new, dentry, false); |
3044 | write_sequnlock(&rename_lock); | |
dd179946 | 3045 | } |
b5ae6b15 AV |
3046 | iput(inode); |
3047 | return new; | |
9eaef27b | 3048 | } |
770bfad8 | 3049 | } |
b5ae6b15 | 3050 | out: |
ed782b5a | 3051 | __d_add(dentry, inode); |
b5ae6b15 | 3052 | return NULL; |
770bfad8 | 3053 | } |
b5ae6b15 | 3054 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3055 | |
1da177e4 LT |
3056 | /* |
3057 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3058 | * | |
3059 | * Trivially implemented using the dcache structure | |
3060 | */ | |
3061 | ||
3062 | /** | |
3063 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3064 | * @new_dentry: new dentry | |
3065 | * @old_dentry: old dentry | |
3066 | * | |
a6e5787f YB |
3067 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3068 | * Returns false otherwise. | |
1da177e4 LT |
3069 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3070 | */ | |
3071 | ||
a6e5787f | 3072 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3073 | { |
a6e5787f | 3074 | bool result; |
949854d0 | 3075 | unsigned seq; |
1da177e4 | 3076 | |
e2761a11 | 3077 | if (new_dentry == old_dentry) |
a6e5787f | 3078 | return true; |
e2761a11 | 3079 | |
e2761a11 | 3080 | do { |
1da177e4 | 3081 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3082 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3083 | /* |
3084 | * Need rcu_readlock to protect against the d_parent trashing | |
3085 | * due to d_move | |
3086 | */ | |
3087 | rcu_read_lock(); | |
e2761a11 | 3088 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3089 | result = true; |
e2761a11 | 3090 | else |
a6e5787f | 3091 | result = false; |
949854d0 | 3092 | rcu_read_unlock(); |
1da177e4 | 3093 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3094 | |
3095 | return result; | |
3096 | } | |
e8f9e5b7 | 3097 | EXPORT_SYMBOL(is_subdir); |
1da177e4 | 3098 | |
db14fc3a | 3099 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3100 | { |
db14fc3a MS |
3101 | struct dentry *root = data; |
3102 | if (dentry != root) { | |
3103 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3104 | return D_WALK_SKIP; | |
1da177e4 | 3105 | |
01ddc4ed MS |
3106 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3107 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3108 | dentry->d_lockref.count--; | |
3109 | } | |
1da177e4 | 3110 | } |
db14fc3a MS |
3111 | return D_WALK_CONTINUE; |
3112 | } | |
58db63d0 | 3113 | |
db14fc3a MS |
3114 | void d_genocide(struct dentry *parent) |
3115 | { | |
3116 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3117 | } |
3118 | ||
cbd4a5bc AV |
3119 | EXPORT_SYMBOL(d_genocide); |
3120 | ||
60545d0d | 3121 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3122 | { |
60545d0d AV |
3123 | inode_dec_link_count(inode); |
3124 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3125 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3126 | !d_unlinked(dentry)); |
3127 | spin_lock(&dentry->d_parent->d_lock); | |
3128 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3129 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3130 | (unsigned long long)inode->i_ino); | |
3131 | spin_unlock(&dentry->d_lock); | |
3132 | spin_unlock(&dentry->d_parent->d_lock); | |
3133 | d_instantiate(dentry, inode); | |
1da177e4 | 3134 | } |
60545d0d | 3135 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3136 | |
3137 | static __initdata unsigned long dhash_entries; | |
3138 | static int __init set_dhash_entries(char *str) | |
3139 | { | |
3140 | if (!str) | |
3141 | return 0; | |
3142 | dhash_entries = simple_strtoul(str, &str, 0); | |
3143 | return 1; | |
3144 | } | |
3145 | __setup("dhash_entries=", set_dhash_entries); | |
3146 | ||
3147 | static void __init dcache_init_early(void) | |
3148 | { | |
1da177e4 LT |
3149 | /* If hashes are distributed across NUMA nodes, defer |
3150 | * hash allocation until vmalloc space is available. | |
3151 | */ | |
3152 | if (hashdist) | |
3153 | return; | |
3154 | ||
3155 | dentry_hashtable = | |
3156 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3157 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3158 | dhash_entries, |
3159 | 13, | |
3d375d78 | 3160 | HASH_EARLY | HASH_ZERO, |
1da177e4 | 3161 | &d_hash_shift, |
b35d786b | 3162 | NULL, |
31fe62b9 | 3163 | 0, |
1da177e4 | 3164 | 0); |
854d3e63 | 3165 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3166 | } |
3167 | ||
74bf17cf | 3168 | static void __init dcache_init(void) |
1da177e4 | 3169 | { |
3d375d78 | 3170 | /* |
1da177e4 LT |
3171 | * A constructor could be added for stable state like the lists, |
3172 | * but it is probably not worth it because of the cache nature | |
3d375d78 | 3173 | * of the dcache. |
1da177e4 | 3174 | */ |
80344266 DW |
3175 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, |
3176 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, | |
3177 | d_iname); | |
1da177e4 LT |
3178 | |
3179 | /* Hash may have been set up in dcache_init_early */ | |
3180 | if (!hashdist) | |
3181 | return; | |
3182 | ||
3183 | dentry_hashtable = | |
3184 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3185 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3186 | dhash_entries, |
3187 | 13, | |
3d375d78 | 3188 | HASH_ZERO, |
1da177e4 | 3189 | &d_hash_shift, |
b35d786b | 3190 | NULL, |
31fe62b9 | 3191 | 0, |
1da177e4 | 3192 | 0); |
854d3e63 | 3193 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3194 | } |
3195 | ||
3196 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3197 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3198 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3199 | |
1da177e4 LT |
3200 | void __init vfs_caches_init_early(void) |
3201 | { | |
6916363f SAS |
3202 | int i; |
3203 | ||
3204 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3205 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3206 | ||
1da177e4 LT |
3207 | dcache_init_early(); |
3208 | inode_init_early(); | |
3209 | } | |
3210 | ||
4248b0da | 3211 | void __init vfs_caches_init(void) |
1da177e4 | 3212 | { |
6a9b8820 DW |
3213 | names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0, |
3214 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL); | |
1da177e4 | 3215 | |
74bf17cf DC |
3216 | dcache_init(); |
3217 | inode_init(); | |
4248b0da MG |
3218 | files_init(); |
3219 | files_maxfiles_init(); | |
74bf17cf | 3220 | mnt_init(); |
1da177e4 LT |
3221 | bdev_cache_init(); |
3222 | chrdev_init(); | |
3223 | } |