]>
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 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
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/mount.h> |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
dd179946 | 39 | #include <linux/ratelimit.h> |
f6041567 | 40 | #include <linux/list_lru.h> |
df4c0e36 AR |
41 | #include <linux/kasan.h> |
42 | ||
07f3f05c | 43 | #include "internal.h" |
b2dba1af | 44 | #include "mount.h" |
1da177e4 | 45 | |
789680d1 NP |
46 | /* |
47 | * Usage: | |
873feea0 | 48 | * dcache->d_inode->i_lock protects: |
946e51f2 | 49 | * - i_dentry, d_u.d_alias, d_inode of aliases |
ceb5bdc2 NP |
50 | * dcache_hash_bucket lock protects: |
51 | * - the dcache hash table | |
52 | * s_anon bl list spinlock protects: | |
53 | * - the s_anon list (see __d_drop) | |
19156840 | 54 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
55 | * - the dcache lru lists and counters |
56 | * d_lock protects: | |
57 | * - d_flags | |
58 | * - d_name | |
59 | * - d_lru | |
b7ab39f6 | 60 | * - d_count |
da502956 | 61 | * - d_unhashed() |
2fd6b7f5 NP |
62 | * - d_parent and d_subdirs |
63 | * - childrens' d_child and d_parent | |
946e51f2 | 64 | * - d_u.d_alias, d_inode |
789680d1 NP |
65 | * |
66 | * Ordering: | |
873feea0 | 67 | * dentry->d_inode->i_lock |
b5c84bf6 | 68 | * dentry->d_lock |
19156840 | 69 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 NP |
70 | * dcache_hash_bucket lock |
71 | * s_anon lock | |
789680d1 | 72 | * |
da502956 NP |
73 | * If there is an ancestor relationship: |
74 | * dentry->d_parent->...->d_parent->d_lock | |
75 | * ... | |
76 | * dentry->d_parent->d_lock | |
77 | * dentry->d_lock | |
78 | * | |
79 | * If no ancestor relationship: | |
789680d1 NP |
80 | * if (dentry1 < dentry2) |
81 | * dentry1->d_lock | |
82 | * dentry2->d_lock | |
83 | */ | |
fa3536cc | 84 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
85 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
86 | ||
74c3cbe3 | 87 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 88 | |
949854d0 | 89 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 90 | |
e18b890b | 91 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 92 | |
1da177e4 LT |
93 | /* |
94 | * This is the single most critical data structure when it comes | |
95 | * to the dcache: the hashtable for lookups. Somebody should try | |
96 | * to make this good - I've just made it work. | |
97 | * | |
98 | * This hash-function tries to avoid losing too many bits of hash | |
99 | * information, yet avoid using a prime hash-size or similar. | |
100 | */ | |
1da177e4 | 101 | |
fa3536cc ED |
102 | static unsigned int d_hash_mask __read_mostly; |
103 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 104 | |
b07ad996 | 105 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 106 | |
8387ff25 | 107 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 108 | { |
703b5faf | 109 | return dentry_hashtable + (hash >> (32 - d_hash_shift)); |
ceb5bdc2 NP |
110 | } |
111 | ||
94bdd655 AV |
112 | #define IN_LOOKUP_SHIFT 10 |
113 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
114 | ||
115 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
116 | unsigned int hash) | |
117 | { | |
118 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
119 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
120 | } | |
121 | ||
122 | ||
1da177e4 LT |
123 | /* Statistics gathering. */ |
124 | struct dentry_stat_t dentry_stat = { | |
125 | .age_limit = 45, | |
126 | }; | |
127 | ||
3942c07c | 128 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 129 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
130 | |
131 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
132 | |
133 | /* | |
134 | * Here we resort to our own counters instead of using generic per-cpu counters | |
135 | * for consistency with what the vfs inode code does. We are expected to harvest | |
136 | * better code and performance by having our own specialized counters. | |
137 | * | |
138 | * Please note that the loop is done over all possible CPUs, not over all online | |
139 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
140 | * on and off. If one of them goes off, we will just keep their counters. | |
141 | * | |
142 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
143 | * please update all vfs counters to match. | |
144 | */ | |
3942c07c | 145 | static long get_nr_dentry(void) |
3e880fb5 NP |
146 | { |
147 | int i; | |
3942c07c | 148 | long sum = 0; |
3e880fb5 NP |
149 | for_each_possible_cpu(i) |
150 | sum += per_cpu(nr_dentry, i); | |
151 | return sum < 0 ? 0 : sum; | |
152 | } | |
153 | ||
62d36c77 DC |
154 | static long get_nr_dentry_unused(void) |
155 | { | |
156 | int i; | |
157 | long sum = 0; | |
158 | for_each_possible_cpu(i) | |
159 | sum += per_cpu(nr_dentry_unused, i); | |
160 | return sum < 0 ? 0 : sum; | |
161 | } | |
162 | ||
1f7e0616 | 163 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
164 | size_t *lenp, loff_t *ppos) |
165 | { | |
3e880fb5 | 166 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 167 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 168 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
169 | } |
170 | #endif | |
171 | ||
5483f18e LT |
172 | /* |
173 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
174 | * The strings are both count bytes long, and count is non-zero. | |
175 | */ | |
e419b4cc LT |
176 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
177 | ||
178 | #include <asm/word-at-a-time.h> | |
179 | /* | |
180 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
181 | * aligned allocation for this particular component. We don't | |
182 | * strictly need the load_unaligned_zeropad() safety, but it | |
183 | * doesn't hurt either. | |
184 | * | |
185 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
186 | * need the careful unaligned handling. | |
187 | */ | |
94753db5 | 188 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 189 | { |
bfcfaa77 | 190 | unsigned long a,b,mask; |
bfcfaa77 LT |
191 | |
192 | for (;;) { | |
12f8ad4b | 193 | a = *(unsigned long *)cs; |
e419b4cc | 194 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
195 | if (tcount < sizeof(unsigned long)) |
196 | break; | |
197 | if (unlikely(a != b)) | |
198 | return 1; | |
199 | cs += sizeof(unsigned long); | |
200 | ct += sizeof(unsigned long); | |
201 | tcount -= sizeof(unsigned long); | |
202 | if (!tcount) | |
203 | return 0; | |
204 | } | |
a5c21dce | 205 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 206 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
207 | } |
208 | ||
bfcfaa77 | 209 | #else |
e419b4cc | 210 | |
94753db5 | 211 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 212 | { |
5483f18e LT |
213 | do { |
214 | if (*cs != *ct) | |
215 | return 1; | |
216 | cs++; | |
217 | ct++; | |
218 | tcount--; | |
219 | } while (tcount); | |
220 | return 0; | |
221 | } | |
222 | ||
e419b4cc LT |
223 | #endif |
224 | ||
94753db5 LT |
225 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
226 | { | |
94753db5 LT |
227 | /* |
228 | * Be careful about RCU walk racing with rename: | |
ae0a843c | 229 | * use 'lockless_dereference' to fetch the name pointer. |
94753db5 LT |
230 | * |
231 | * NOTE! Even if a rename will mean that the length | |
232 | * was not loaded atomically, we don't care. The | |
233 | * RCU walk will check the sequence count eventually, | |
234 | * and catch it. And we won't overrun the buffer, | |
235 | * because we're reading the name pointer atomically, | |
236 | * and a dentry name is guaranteed to be properly | |
237 | * terminated with a NUL byte. | |
238 | * | |
239 | * End result: even if 'len' is wrong, we'll exit | |
240 | * early because the data cannot match (there can | |
241 | * be no NUL in the ct/tcount data) | |
242 | */ | |
ae0a843c HK |
243 | const unsigned char *cs = lockless_dereference(dentry->d_name.name); |
244 | ||
6326c71f | 245 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
246 | } |
247 | ||
8d85b484 AV |
248 | struct external_name { |
249 | union { | |
250 | atomic_t count; | |
251 | struct rcu_head head; | |
252 | } u; | |
253 | unsigned char name[]; | |
254 | }; | |
255 | ||
256 | static inline struct external_name *external_name(struct dentry *dentry) | |
257 | { | |
258 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
259 | } | |
260 | ||
9c82ab9c | 261 | static void __d_free(struct rcu_head *head) |
1da177e4 | 262 | { |
9c82ab9c CH |
263 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
264 | ||
8d85b484 AV |
265 | kmem_cache_free(dentry_cache, dentry); |
266 | } | |
267 | ||
268 | static void __d_free_external(struct rcu_head *head) | |
269 | { | |
270 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
8d85b484 | 271 | kfree(external_name(dentry)); |
1da177e4 LT |
272 | kmem_cache_free(dentry_cache, dentry); |
273 | } | |
274 | ||
810bb172 AV |
275 | static inline int dname_external(const struct dentry *dentry) |
276 | { | |
277 | return dentry->d_name.name != dentry->d_iname; | |
278 | } | |
279 | ||
4bf46a27 DH |
280 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
281 | struct inode *inode, | |
282 | unsigned type_flags) | |
283 | { | |
284 | unsigned flags; | |
285 | ||
286 | dentry->d_inode = inode; | |
4bf46a27 DH |
287 | flags = READ_ONCE(dentry->d_flags); |
288 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
289 | flags |= type_flags; | |
290 | WRITE_ONCE(dentry->d_flags, flags); | |
291 | } | |
292 | ||
4bf46a27 DH |
293 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
294 | { | |
295 | unsigned flags = READ_ONCE(dentry->d_flags); | |
296 | ||
297 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
298 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 DH |
299 | dentry->d_inode = NULL; |
300 | } | |
301 | ||
b4f0354e AV |
302 | static void dentry_free(struct dentry *dentry) |
303 | { | |
946e51f2 | 304 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
305 | if (unlikely(dname_external(dentry))) { |
306 | struct external_name *p = external_name(dentry); | |
307 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
308 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
309 | return; | |
310 | } | |
311 | } | |
b4f0354e AV |
312 | /* if dentry was never visible to RCU, immediate free is OK */ |
313 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
314 | __d_free(&dentry->d_u.d_rcu); | |
315 | else | |
316 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
317 | } | |
318 | ||
1da177e4 LT |
319 | /* |
320 | * Release the dentry's inode, using the filesystem | |
550dce01 | 321 | * d_iput() operation if defined. |
31e6b01f NP |
322 | */ |
323 | static void dentry_unlink_inode(struct dentry * dentry) | |
324 | __releases(dentry->d_lock) | |
873feea0 | 325 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
326 | { |
327 | struct inode *inode = dentry->d_inode; | |
550dce01 | 328 | bool hashed = !d_unhashed(dentry); |
a528aca7 | 329 | |
550dce01 AV |
330 | if (hashed) |
331 | raw_write_seqcount_begin(&dentry->d_seq); | |
4bf46a27 | 332 | __d_clear_type_and_inode(dentry); |
946e51f2 | 333 | hlist_del_init(&dentry->d_u.d_alias); |
550dce01 AV |
334 | if (hashed) |
335 | raw_write_seqcount_end(&dentry->d_seq); | |
31e6b01f | 336 | spin_unlock(&dentry->d_lock); |
873feea0 | 337 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
338 | if (!inode->i_nlink) |
339 | fsnotify_inoderemove(inode); | |
340 | if (dentry->d_op && dentry->d_op->d_iput) | |
341 | dentry->d_op->d_iput(dentry, inode); | |
342 | else | |
343 | iput(inode); | |
344 | } | |
345 | ||
89dc77bc LT |
346 | /* |
347 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
348 | * is in use - which includes both the "real" per-superblock | |
349 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
350 | * | |
351 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
352 | * on the shrink list (ie not on the superblock LRU list). | |
353 | * | |
354 | * The per-cpu "nr_dentry_unused" counters are updated with | |
355 | * the DCACHE_LRU_LIST bit. | |
356 | * | |
357 | * These helper functions make sure we always follow the | |
358 | * rules. d_lock must be held by the caller. | |
359 | */ | |
360 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
361 | static void d_lru_add(struct dentry *dentry) | |
362 | { | |
363 | D_FLAG_VERIFY(dentry, 0); | |
364 | dentry->d_flags |= DCACHE_LRU_LIST; | |
365 | this_cpu_inc(nr_dentry_unused); | |
366 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
367 | } | |
368 | ||
369 | static void d_lru_del(struct dentry *dentry) | |
370 | { | |
371 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
372 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
373 | this_cpu_dec(nr_dentry_unused); | |
374 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
375 | } | |
376 | ||
377 | static void d_shrink_del(struct dentry *dentry) | |
378 | { | |
379 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
380 | list_del_init(&dentry->d_lru); | |
381 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
382 | this_cpu_dec(nr_dentry_unused); | |
383 | } | |
384 | ||
385 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
386 | { | |
387 | D_FLAG_VERIFY(dentry, 0); | |
388 | list_add(&dentry->d_lru, list); | |
389 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
390 | this_cpu_inc(nr_dentry_unused); | |
391 | } | |
392 | ||
393 | /* | |
394 | * These can only be called under the global LRU lock, ie during the | |
395 | * callback for freeing the LRU list. "isolate" removes it from the | |
396 | * LRU lists entirely, while shrink_move moves it to the indicated | |
397 | * private list. | |
398 | */ | |
3f97b163 | 399 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
400 | { |
401 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
402 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
403 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 404 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
405 | } |
406 | ||
3f97b163 VD |
407 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
408 | struct list_head *list) | |
89dc77bc LT |
409 | { |
410 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
411 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 412 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
413 | } |
414 | ||
da3bbdd4 | 415 | /* |
f6041567 | 416 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
417 | */ |
418 | static void dentry_lru_add(struct dentry *dentry) | |
419 | { | |
89dc77bc LT |
420 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
421 | d_lru_add(dentry); | |
da3bbdd4 KM |
422 | } |
423 | ||
789680d1 NP |
424 | /** |
425 | * d_drop - drop a dentry | |
426 | * @dentry: dentry to drop | |
427 | * | |
428 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
429 | * be found through a VFS lookup any more. Note that this is different from | |
430 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
431 | * possible, giving a successful _negative_ lookup, while d_drop will | |
432 | * just make the cache lookup fail. | |
433 | * | |
434 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
435 | * reason (NFS timeouts or autofs deletes). | |
436 | * | |
437 | * __d_drop requires dentry->d_lock. | |
438 | */ | |
439 | void __d_drop(struct dentry *dentry) | |
440 | { | |
dea3667b | 441 | if (!d_unhashed(dentry)) { |
b61625d2 | 442 | struct hlist_bl_head *b; |
7632e465 BF |
443 | /* |
444 | * Hashed dentries are normally on the dentry hashtable, | |
445 | * with the exception of those newly allocated by | |
446 | * d_obtain_alias, which are always IS_ROOT: | |
447 | */ | |
448 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
449 | b = &dentry->d_sb->s_anon; |
450 | else | |
8387ff25 | 451 | b = d_hash(dentry->d_name.hash); |
b61625d2 AV |
452 | |
453 | hlist_bl_lock(b); | |
454 | __hlist_bl_del(&dentry->d_hash); | |
455 | dentry->d_hash.pprev = NULL; | |
456 | hlist_bl_unlock(b); | |
d614146d AV |
457 | /* After this call, in-progress rcu-walk path lookup will fail. */ |
458 | write_seqcount_invalidate(&dentry->d_seq); | |
789680d1 NP |
459 | } |
460 | } | |
461 | EXPORT_SYMBOL(__d_drop); | |
462 | ||
463 | void d_drop(struct dentry *dentry) | |
464 | { | |
789680d1 NP |
465 | spin_lock(&dentry->d_lock); |
466 | __d_drop(dentry); | |
467 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
468 | } |
469 | EXPORT_SYMBOL(d_drop); | |
470 | ||
ba65dc5e AV |
471 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
472 | { | |
473 | struct dentry *next; | |
474 | /* | |
475 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
476 | * attached to the dentry tree | |
477 | */ | |
478 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
479 | if (unlikely(list_empty(&dentry->d_child))) | |
480 | return; | |
481 | __list_del_entry(&dentry->d_child); | |
482 | /* | |
483 | * Cursors can move around the list of children. While we'd been | |
484 | * a normal list member, it didn't matter - ->d_child.next would've | |
485 | * been updated. However, from now on it won't be and for the | |
486 | * things like d_walk() it might end up with a nasty surprise. | |
487 | * Normally d_walk() doesn't care about cursors moving around - | |
488 | * ->d_lock on parent prevents that and since a cursor has no children | |
489 | * of its own, we get through it without ever unlocking the parent. | |
490 | * There is one exception, though - if we ascend from a child that | |
491 | * gets killed as soon as we unlock it, the next sibling is found | |
492 | * using the value left in its ->d_child.next. And if _that_ | |
493 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
494 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
495 | * everything the cursor had been moved past. | |
496 | * | |
497 | * Solution: make sure that the pointer left behind in ->d_child.next | |
498 | * points to something that won't be moving around. I.e. skip the | |
499 | * cursors. | |
500 | */ | |
501 | while (dentry->d_child.next != &parent->d_subdirs) { | |
502 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
503 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
504 | break; | |
505 | dentry->d_child.next = next->d_child.next; | |
506 | } | |
507 | } | |
508 | ||
e55fd011 | 509 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 510 | { |
41edf278 AV |
511 | struct dentry *parent = NULL; |
512 | bool can_free = true; | |
41edf278 | 513 | if (!IS_ROOT(dentry)) |
77812a1e | 514 | parent = dentry->d_parent; |
31e6b01f | 515 | |
0d98439e LT |
516 | /* |
517 | * The dentry is now unrecoverably dead to the world. | |
518 | */ | |
519 | lockref_mark_dead(&dentry->d_lockref); | |
520 | ||
f0023bc6 | 521 | /* |
f0023bc6 SW |
522 | * inform the fs via d_prune that this dentry is about to be |
523 | * unhashed and destroyed. | |
524 | */ | |
29266201 | 525 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
526 | dentry->d_op->d_prune(dentry); |
527 | ||
01b60351 AV |
528 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
529 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
530 | d_lru_del(dentry); | |
01b60351 | 531 | } |
77812a1e NP |
532 | /* if it was on the hash then remove it */ |
533 | __d_drop(dentry); | |
ba65dc5e | 534 | dentry_unlist(dentry, parent); |
03b3b889 AV |
535 | if (parent) |
536 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
537 | if (dentry->d_inode) |
538 | dentry_unlink_inode(dentry); | |
539 | else | |
540 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
541 | this_cpu_dec(nr_dentry); |
542 | if (dentry->d_op && dentry->d_op->d_release) | |
543 | dentry->d_op->d_release(dentry); | |
544 | ||
41edf278 AV |
545 | spin_lock(&dentry->d_lock); |
546 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
547 | dentry->d_flags |= DCACHE_MAY_FREE; | |
548 | can_free = false; | |
549 | } | |
550 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
551 | if (likely(can_free)) |
552 | dentry_free(dentry); | |
e55fd011 AV |
553 | } |
554 | ||
555 | /* | |
556 | * Finish off a dentry we've decided to kill. | |
557 | * dentry->d_lock must be held, returns with it unlocked. | |
558 | * If ref is non-zero, then decrement the refcount too. | |
559 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
560 | */ | |
8cbf74da | 561 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
562 | __releases(dentry->d_lock) |
563 | { | |
564 | struct inode *inode = dentry->d_inode; | |
565 | struct dentry *parent = NULL; | |
566 | ||
567 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
568 | goto failed; | |
569 | ||
570 | if (!IS_ROOT(dentry)) { | |
571 | parent = dentry->d_parent; | |
572 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
573 | if (inode) | |
574 | spin_unlock(&inode->i_lock); | |
575 | goto failed; | |
576 | } | |
577 | } | |
578 | ||
579 | __dentry_kill(dentry); | |
03b3b889 | 580 | return parent; |
e55fd011 AV |
581 | |
582 | failed: | |
8cbf74da | 583 | spin_unlock(&dentry->d_lock); |
e55fd011 | 584 | return dentry; /* try again with same dentry */ |
77812a1e NP |
585 | } |
586 | ||
046b961b AV |
587 | static inline struct dentry *lock_parent(struct dentry *dentry) |
588 | { | |
589 | struct dentry *parent = dentry->d_parent; | |
590 | if (IS_ROOT(dentry)) | |
591 | return NULL; | |
360f5479 | 592 | if (unlikely(dentry->d_lockref.count < 0)) |
c2338f2d | 593 | return NULL; |
046b961b AV |
594 | if (likely(spin_trylock(&parent->d_lock))) |
595 | return parent; | |
046b961b | 596 | rcu_read_lock(); |
c2338f2d | 597 | spin_unlock(&dentry->d_lock); |
046b961b AV |
598 | again: |
599 | parent = ACCESS_ONCE(dentry->d_parent); | |
600 | spin_lock(&parent->d_lock); | |
601 | /* | |
602 | * We can't blindly lock dentry until we are sure | |
603 | * that we won't violate the locking order. | |
604 | * Any changes of dentry->d_parent must have | |
605 | * been done with parent->d_lock held, so | |
606 | * spin_lock() above is enough of a barrier | |
607 | * for checking if it's still our child. | |
608 | */ | |
609 | if (unlikely(parent != dentry->d_parent)) { | |
610 | spin_unlock(&parent->d_lock); | |
611 | goto again; | |
612 | } | |
613 | rcu_read_unlock(); | |
614 | if (parent != dentry) | |
9f12600f | 615 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
046b961b AV |
616 | else |
617 | parent = NULL; | |
618 | return parent; | |
619 | } | |
620 | ||
360f5479 LT |
621 | /* |
622 | * Try to do a lockless dput(), and return whether that was successful. | |
623 | * | |
624 | * If unsuccessful, we return false, having already taken the dentry lock. | |
625 | * | |
626 | * The caller needs to hold the RCU read lock, so that the dentry is | |
627 | * guaranteed to stay around even if the refcount goes down to zero! | |
628 | */ | |
629 | static inline bool fast_dput(struct dentry *dentry) | |
630 | { | |
631 | int ret; | |
632 | unsigned int d_flags; | |
633 | ||
634 | /* | |
635 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 636 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
637 | */ |
638 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
639 | return lockref_put_or_lock(&dentry->d_lockref); | |
640 | ||
641 | /* | |
642 | * .. otherwise, we can try to just decrement the | |
643 | * lockref optimistically. | |
644 | */ | |
645 | ret = lockref_put_return(&dentry->d_lockref); | |
646 | ||
647 | /* | |
648 | * If the lockref_put_return() failed due to the lock being held | |
649 | * by somebody else, the fast path has failed. We will need to | |
650 | * get the lock, and then check the count again. | |
651 | */ | |
652 | if (unlikely(ret < 0)) { | |
653 | spin_lock(&dentry->d_lock); | |
654 | if (dentry->d_lockref.count > 1) { | |
655 | dentry->d_lockref.count--; | |
656 | spin_unlock(&dentry->d_lock); | |
657 | return 1; | |
658 | } | |
659 | return 0; | |
660 | } | |
661 | ||
662 | /* | |
663 | * If we weren't the last ref, we're done. | |
664 | */ | |
665 | if (ret) | |
666 | return 1; | |
667 | ||
668 | /* | |
669 | * Careful, careful. The reference count went down | |
670 | * to zero, but we don't hold the dentry lock, so | |
671 | * somebody else could get it again, and do another | |
672 | * dput(), and we need to not race with that. | |
673 | * | |
674 | * However, there is a very special and common case | |
675 | * where we don't care, because there is nothing to | |
676 | * do: the dentry is still hashed, it does not have | |
677 | * a 'delete' op, and it's referenced and already on | |
678 | * the LRU list. | |
679 | * | |
680 | * NOTE! Since we aren't locked, these values are | |
681 | * not "stable". However, it is sufficient that at | |
682 | * some point after we dropped the reference the | |
683 | * dentry was hashed and the flags had the proper | |
684 | * value. Other dentry users may have re-gotten | |
685 | * a reference to the dentry and change that, but | |
686 | * our work is done - we can leave the dentry | |
687 | * around with a zero refcount. | |
688 | */ | |
689 | smp_rmb(); | |
690 | d_flags = ACCESS_ONCE(dentry->d_flags); | |
75a6f82a | 691 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
692 | |
693 | /* Nothing to do? Dropping the reference was all we needed? */ | |
694 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
695 | return 1; | |
696 | ||
697 | /* | |
698 | * Not the fast normal case? Get the lock. We've already decremented | |
699 | * the refcount, but we'll need to re-check the situation after | |
700 | * getting the lock. | |
701 | */ | |
702 | spin_lock(&dentry->d_lock); | |
703 | ||
704 | /* | |
705 | * Did somebody else grab a reference to it in the meantime, and | |
706 | * we're no longer the last user after all? Alternatively, somebody | |
707 | * else could have killed it and marked it dead. Either way, we | |
708 | * don't need to do anything else. | |
709 | */ | |
710 | if (dentry->d_lockref.count) { | |
711 | spin_unlock(&dentry->d_lock); | |
712 | return 1; | |
713 | } | |
714 | ||
715 | /* | |
716 | * Re-get the reference we optimistically dropped. We hold the | |
717 | * lock, and we just tested that it was zero, so we can just | |
718 | * set it to 1. | |
719 | */ | |
720 | dentry->d_lockref.count = 1; | |
721 | return 0; | |
722 | } | |
723 | ||
724 | ||
1da177e4 LT |
725 | /* |
726 | * This is dput | |
727 | * | |
728 | * This is complicated by the fact that we do not want to put | |
729 | * dentries that are no longer on any hash chain on the unused | |
730 | * list: we'd much rather just get rid of them immediately. | |
731 | * | |
732 | * However, that implies that we have to traverse the dentry | |
733 | * tree upwards to the parents which might _also_ now be | |
734 | * scheduled for deletion (it may have been only waiting for | |
735 | * its last child to go away). | |
736 | * | |
737 | * This tail recursion is done by hand as we don't want to depend | |
738 | * on the compiler to always get this right (gcc generally doesn't). | |
739 | * Real recursion would eat up our stack space. | |
740 | */ | |
741 | ||
742 | /* | |
743 | * dput - release a dentry | |
744 | * @dentry: dentry to release | |
745 | * | |
746 | * Release a dentry. This will drop the usage count and if appropriate | |
747 | * call the dentry unlink method as well as removing it from the queues and | |
748 | * releasing its resources. If the parent dentries were scheduled for release | |
749 | * they too may now get deleted. | |
1da177e4 | 750 | */ |
1da177e4 LT |
751 | void dput(struct dentry *dentry) |
752 | { | |
8aab6a27 | 753 | if (unlikely(!dentry)) |
1da177e4 LT |
754 | return; |
755 | ||
756 | repeat: | |
47be6184 WF |
757 | might_sleep(); |
758 | ||
360f5479 LT |
759 | rcu_read_lock(); |
760 | if (likely(fast_dput(dentry))) { | |
761 | rcu_read_unlock(); | |
1da177e4 | 762 | return; |
360f5479 LT |
763 | } |
764 | ||
765 | /* Slow case: now with the dentry lock held */ | |
766 | rcu_read_unlock(); | |
1da177e4 | 767 | |
85c7f810 AV |
768 | WARN_ON(d_in_lookup(dentry)); |
769 | ||
8aab6a27 LT |
770 | /* Unreachable? Get rid of it */ |
771 | if (unlikely(d_unhashed(dentry))) | |
772 | goto kill_it; | |
773 | ||
75a6f82a AV |
774 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
775 | goto kill_it; | |
776 | ||
8aab6a27 | 777 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { |
1da177e4 | 778 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 779 | goto kill_it; |
1da177e4 | 780 | } |
265ac902 | 781 | |
358eec18 LT |
782 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
783 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 784 | dentry_lru_add(dentry); |
265ac902 | 785 | |
98474236 | 786 | dentry->d_lockref.count--; |
61f3dee4 | 787 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
788 | return; |
789 | ||
d52b9086 | 790 | kill_it: |
8cbf74da | 791 | dentry = dentry_kill(dentry); |
47be6184 WF |
792 | if (dentry) { |
793 | cond_resched(); | |
d52b9086 | 794 | goto repeat; |
47be6184 | 795 | } |
1da177e4 | 796 | } |
ec4f8605 | 797 | EXPORT_SYMBOL(dput); |
1da177e4 | 798 | |
1da177e4 | 799 | |
b5c84bf6 | 800 | /* This must be called with d_lock held */ |
dc0474be | 801 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 802 | { |
98474236 | 803 | dentry->d_lockref.count++; |
23044507 NP |
804 | } |
805 | ||
dc0474be | 806 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 807 | { |
98474236 | 808 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
809 | } |
810 | ||
b7ab39f6 NP |
811 | struct dentry *dget_parent(struct dentry *dentry) |
812 | { | |
df3d0bbc | 813 | int gotref; |
b7ab39f6 NP |
814 | struct dentry *ret; |
815 | ||
df3d0bbc WL |
816 | /* |
817 | * Do optimistic parent lookup without any | |
818 | * locking. | |
819 | */ | |
820 | rcu_read_lock(); | |
821 | ret = ACCESS_ONCE(dentry->d_parent); | |
822 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
823 | rcu_read_unlock(); | |
824 | if (likely(gotref)) { | |
825 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
826 | return ret; | |
827 | dput(ret); | |
828 | } | |
829 | ||
b7ab39f6 | 830 | repeat: |
a734eb45 NP |
831 | /* |
832 | * Don't need rcu_dereference because we re-check it was correct under | |
833 | * the lock. | |
834 | */ | |
835 | rcu_read_lock(); | |
b7ab39f6 | 836 | ret = dentry->d_parent; |
a734eb45 NP |
837 | spin_lock(&ret->d_lock); |
838 | if (unlikely(ret != dentry->d_parent)) { | |
839 | spin_unlock(&ret->d_lock); | |
840 | rcu_read_unlock(); | |
b7ab39f6 NP |
841 | goto repeat; |
842 | } | |
a734eb45 | 843 | rcu_read_unlock(); |
98474236 WL |
844 | BUG_ON(!ret->d_lockref.count); |
845 | ret->d_lockref.count++; | |
b7ab39f6 | 846 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
847 | return ret; |
848 | } | |
849 | EXPORT_SYMBOL(dget_parent); | |
850 | ||
1da177e4 LT |
851 | /** |
852 | * d_find_alias - grab a hashed alias of inode | |
853 | * @inode: inode in question | |
1da177e4 LT |
854 | * |
855 | * If inode has a hashed alias, or is a directory and has any alias, | |
856 | * acquire the reference to alias and return it. Otherwise return NULL. | |
857 | * Notice that if inode is a directory there can be only one alias and | |
858 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
859 | * of a filesystem, or if the directory was renamed and d_revalidate |
860 | * was the first vfs operation to notice. | |
1da177e4 | 861 | * |
21c0d8fd | 862 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 863 | * any other hashed alias over that one. |
1da177e4 | 864 | */ |
52ed46f0 | 865 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 866 | { |
da502956 | 867 | struct dentry *alias, *discon_alias; |
1da177e4 | 868 | |
da502956 NP |
869 | again: |
870 | discon_alias = NULL; | |
946e51f2 | 871 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 872 | spin_lock(&alias->d_lock); |
1da177e4 | 873 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 874 | if (IS_ROOT(alias) && |
da502956 | 875 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 876 | discon_alias = alias; |
52ed46f0 | 877 | } else { |
dc0474be | 878 | __dget_dlock(alias); |
da502956 NP |
879 | spin_unlock(&alias->d_lock); |
880 | return alias; | |
881 | } | |
882 | } | |
883 | spin_unlock(&alias->d_lock); | |
884 | } | |
885 | if (discon_alias) { | |
886 | alias = discon_alias; | |
887 | spin_lock(&alias->d_lock); | |
888 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
889 | __dget_dlock(alias); |
890 | spin_unlock(&alias->d_lock); | |
891 | return alias; | |
1da177e4 | 892 | } |
da502956 NP |
893 | spin_unlock(&alias->d_lock); |
894 | goto again; | |
1da177e4 | 895 | } |
da502956 | 896 | return NULL; |
1da177e4 LT |
897 | } |
898 | ||
da502956 | 899 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 900 | { |
214fda1f DH |
901 | struct dentry *de = NULL; |
902 | ||
b3d9b7a3 | 903 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 904 | spin_lock(&inode->i_lock); |
52ed46f0 | 905 | de = __d_find_alias(inode); |
873feea0 | 906 | spin_unlock(&inode->i_lock); |
214fda1f | 907 | } |
1da177e4 LT |
908 | return de; |
909 | } | |
ec4f8605 | 910 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
911 | |
912 | /* | |
913 | * Try to kill dentries associated with this inode. | |
914 | * WARNING: you must own a reference to inode. | |
915 | */ | |
916 | void d_prune_aliases(struct inode *inode) | |
917 | { | |
0cdca3f9 | 918 | struct dentry *dentry; |
1da177e4 | 919 | restart: |
873feea0 | 920 | spin_lock(&inode->i_lock); |
946e51f2 | 921 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 922 | spin_lock(&dentry->d_lock); |
98474236 | 923 | if (!dentry->d_lockref.count) { |
29355c39 AV |
924 | struct dentry *parent = lock_parent(dentry); |
925 | if (likely(!dentry->d_lockref.count)) { | |
926 | __dentry_kill(dentry); | |
4a7795d3 | 927 | dput(parent); |
29355c39 AV |
928 | goto restart; |
929 | } | |
930 | if (parent) | |
931 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
932 | } |
933 | spin_unlock(&dentry->d_lock); | |
934 | } | |
873feea0 | 935 | spin_unlock(&inode->i_lock); |
1da177e4 | 936 | } |
ec4f8605 | 937 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 938 | |
3049cfe2 | 939 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 940 | { |
5c47e6d0 | 941 | struct dentry *dentry, *parent; |
da3bbdd4 | 942 | |
60942f2f | 943 | while (!list_empty(list)) { |
ff2fde99 | 944 | struct inode *inode; |
60942f2f | 945 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 946 | spin_lock(&dentry->d_lock); |
046b961b AV |
947 | parent = lock_parent(dentry); |
948 | ||
dd1f6b2e DC |
949 | /* |
950 | * The dispose list is isolated and dentries are not accounted | |
951 | * to the LRU here, so we can simply remove it from the list | |
952 | * here regardless of whether it is referenced or not. | |
953 | */ | |
89dc77bc | 954 | d_shrink_del(dentry); |
dd1f6b2e | 955 | |
1da177e4 LT |
956 | /* |
957 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 958 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 959 | */ |
360f5479 | 960 | if (dentry->d_lockref.count > 0) { |
da3bbdd4 | 961 | spin_unlock(&dentry->d_lock); |
046b961b AV |
962 | if (parent) |
963 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
964 | continue; |
965 | } | |
77812a1e | 966 | |
64fd72e0 AV |
967 | |
968 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
969 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
970 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
971 | if (parent) |
972 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
973 | if (can_free) |
974 | dentry_free(dentry); | |
975 | continue; | |
976 | } | |
977 | ||
ff2fde99 AV |
978 | inode = dentry->d_inode; |
979 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 980 | d_shrink_add(dentry, list); |
dd1f6b2e | 981 | spin_unlock(&dentry->d_lock); |
046b961b AV |
982 | if (parent) |
983 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 984 | continue; |
dd1f6b2e | 985 | } |
ff2fde99 | 986 | |
ff2fde99 | 987 | __dentry_kill(dentry); |
046b961b | 988 | |
5c47e6d0 AV |
989 | /* |
990 | * We need to prune ancestors too. This is necessary to prevent | |
991 | * quadratic behavior of shrink_dcache_parent(), but is also | |
992 | * expected to be beneficial in reducing dentry cache | |
993 | * fragmentation. | |
994 | */ | |
995 | dentry = parent; | |
b2b80195 AV |
996 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
997 | parent = lock_parent(dentry); | |
998 | if (dentry->d_lockref.count != 1) { | |
999 | dentry->d_lockref.count--; | |
1000 | spin_unlock(&dentry->d_lock); | |
1001 | if (parent) | |
1002 | spin_unlock(&parent->d_lock); | |
1003 | break; | |
1004 | } | |
1005 | inode = dentry->d_inode; /* can't be NULL */ | |
1006 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
1007 | spin_unlock(&dentry->d_lock); | |
1008 | if (parent) | |
1009 | spin_unlock(&parent->d_lock); | |
1010 | cpu_relax(); | |
1011 | continue; | |
1012 | } | |
1013 | __dentry_kill(dentry); | |
1014 | dentry = parent; | |
1015 | } | |
da3bbdd4 | 1016 | } |
3049cfe2 CH |
1017 | } |
1018 | ||
3f97b163 VD |
1019 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1020 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1021 | { |
1022 | struct list_head *freeable = arg; | |
1023 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1024 | ||
1025 | ||
1026 | /* | |
1027 | * we are inverting the lru lock/dentry->d_lock here, | |
1028 | * so use a trylock. If we fail to get the lock, just skip | |
1029 | * it | |
1030 | */ | |
1031 | if (!spin_trylock(&dentry->d_lock)) | |
1032 | return LRU_SKIP; | |
1033 | ||
1034 | /* | |
1035 | * Referenced dentries are still in use. If they have active | |
1036 | * counts, just remove them from the LRU. Otherwise give them | |
1037 | * another pass through the LRU. | |
1038 | */ | |
1039 | if (dentry->d_lockref.count) { | |
3f97b163 | 1040 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1041 | spin_unlock(&dentry->d_lock); |
1042 | return LRU_REMOVED; | |
1043 | } | |
1044 | ||
1045 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1046 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1047 | spin_unlock(&dentry->d_lock); | |
1048 | ||
1049 | /* | |
1050 | * The list move itself will be made by the common LRU code. At | |
1051 | * this point, we've dropped the dentry->d_lock but keep the | |
1052 | * lru lock. This is safe to do, since every list movement is | |
1053 | * protected by the lru lock even if both locks are held. | |
1054 | * | |
1055 | * This is guaranteed by the fact that all LRU management | |
1056 | * functions are intermediated by the LRU API calls like | |
1057 | * list_lru_add and list_lru_del. List movement in this file | |
1058 | * only ever occur through this functions or through callbacks | |
1059 | * like this one, that are called from the LRU API. | |
1060 | * | |
1061 | * The only exceptions to this are functions like | |
1062 | * shrink_dentry_list, and code that first checks for the | |
1063 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1064 | * operating only with stack provided lists after they are | |
1065 | * properly isolated from the main list. It is thus, always a | |
1066 | * local access. | |
1067 | */ | |
1068 | return LRU_ROTATE; | |
1069 | } | |
1070 | ||
3f97b163 | 1071 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1072 | spin_unlock(&dentry->d_lock); |
1073 | ||
1074 | return LRU_REMOVED; | |
1075 | } | |
1076 | ||
3049cfe2 | 1077 | /** |
b48f03b3 DC |
1078 | * prune_dcache_sb - shrink the dcache |
1079 | * @sb: superblock | |
503c358c | 1080 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1081 | * |
503c358c VD |
1082 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1083 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1084 | * function. |
3049cfe2 | 1085 | * |
b48f03b3 DC |
1086 | * This function may fail to free any resources if all the dentries are in |
1087 | * use. | |
3049cfe2 | 1088 | */ |
503c358c | 1089 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1090 | { |
f6041567 DC |
1091 | LIST_HEAD(dispose); |
1092 | long freed; | |
3049cfe2 | 1093 | |
503c358c VD |
1094 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1095 | dentry_lru_isolate, &dispose); | |
f6041567 | 1096 | shrink_dentry_list(&dispose); |
0a234c6d | 1097 | return freed; |
da3bbdd4 | 1098 | } |
23044507 | 1099 | |
4e717f5c | 1100 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1101 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1102 | { |
4e717f5c GC |
1103 | struct list_head *freeable = arg; |
1104 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1105 | |
4e717f5c GC |
1106 | /* |
1107 | * we are inverting the lru lock/dentry->d_lock here, | |
1108 | * so use a trylock. If we fail to get the lock, just skip | |
1109 | * it | |
1110 | */ | |
1111 | if (!spin_trylock(&dentry->d_lock)) | |
1112 | return LRU_SKIP; | |
1113 | ||
3f97b163 | 1114 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1115 | spin_unlock(&dentry->d_lock); |
ec33679d | 1116 | |
4e717f5c | 1117 | return LRU_REMOVED; |
da3bbdd4 KM |
1118 | } |
1119 | ||
4e717f5c | 1120 | |
1da177e4 LT |
1121 | /** |
1122 | * shrink_dcache_sb - shrink dcache for a superblock | |
1123 | * @sb: superblock | |
1124 | * | |
3049cfe2 CH |
1125 | * Shrink the dcache for the specified super block. This is used to free |
1126 | * the dcache before unmounting a file system. | |
1da177e4 | 1127 | */ |
3049cfe2 | 1128 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1129 | { |
4e717f5c GC |
1130 | long freed; |
1131 | ||
1132 | do { | |
1133 | LIST_HEAD(dispose); | |
1134 | ||
1135 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1136 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1137 | |
4e717f5c GC |
1138 | this_cpu_sub(nr_dentry_unused, freed); |
1139 | shrink_dentry_list(&dispose); | |
1140 | } while (freed > 0); | |
1da177e4 | 1141 | } |
ec4f8605 | 1142 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1143 | |
db14fc3a MS |
1144 | /** |
1145 | * enum d_walk_ret - action to talke during tree walk | |
1146 | * @D_WALK_CONTINUE: contrinue walk | |
1147 | * @D_WALK_QUIT: quit walk | |
1148 | * @D_WALK_NORETRY: quit when retry is needed | |
1149 | * @D_WALK_SKIP: skip this dentry and its children | |
1150 | */ | |
1151 | enum d_walk_ret { | |
1152 | D_WALK_CONTINUE, | |
1153 | D_WALK_QUIT, | |
1154 | D_WALK_NORETRY, | |
1155 | D_WALK_SKIP, | |
1156 | }; | |
c826cb7d | 1157 | |
1da177e4 | 1158 | /** |
db14fc3a MS |
1159 | * d_walk - walk the dentry tree |
1160 | * @parent: start of walk | |
1161 | * @data: data passed to @enter() and @finish() | |
1162 | * @enter: callback when first entering the dentry | |
1163 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1164 | * |
db14fc3a | 1165 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1166 | */ |
db14fc3a MS |
1167 | static void d_walk(struct dentry *parent, void *data, |
1168 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1169 | void (*finish)(void *)) | |
1da177e4 | 1170 | { |
949854d0 | 1171 | struct dentry *this_parent; |
1da177e4 | 1172 | struct list_head *next; |
48f5ec21 | 1173 | unsigned seq = 0; |
db14fc3a MS |
1174 | enum d_walk_ret ret; |
1175 | bool retry = true; | |
949854d0 | 1176 | |
58db63d0 | 1177 | again: |
48f5ec21 | 1178 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1179 | this_parent = parent; |
2fd6b7f5 | 1180 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1181 | |
1182 | ret = enter(data, this_parent); | |
1183 | switch (ret) { | |
1184 | case D_WALK_CONTINUE: | |
1185 | break; | |
1186 | case D_WALK_QUIT: | |
1187 | case D_WALK_SKIP: | |
1188 | goto out_unlock; | |
1189 | case D_WALK_NORETRY: | |
1190 | retry = false; | |
1191 | break; | |
1192 | } | |
1da177e4 LT |
1193 | repeat: |
1194 | next = this_parent->d_subdirs.next; | |
1195 | resume: | |
1196 | while (next != &this_parent->d_subdirs) { | |
1197 | struct list_head *tmp = next; | |
946e51f2 | 1198 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1199 | next = tmp->next; |
2fd6b7f5 | 1200 | |
ba65dc5e AV |
1201 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1202 | continue; | |
1203 | ||
2fd6b7f5 | 1204 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1205 | |
1206 | ret = enter(data, dentry); | |
1207 | switch (ret) { | |
1208 | case D_WALK_CONTINUE: | |
1209 | break; | |
1210 | case D_WALK_QUIT: | |
2fd6b7f5 | 1211 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1212 | goto out_unlock; |
1213 | case D_WALK_NORETRY: | |
1214 | retry = false; | |
1215 | break; | |
1216 | case D_WALK_SKIP: | |
1217 | spin_unlock(&dentry->d_lock); | |
1218 | continue; | |
2fd6b7f5 | 1219 | } |
db14fc3a | 1220 | |
1da177e4 | 1221 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1222 | spin_unlock(&this_parent->d_lock); |
1223 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1224 | this_parent = dentry; |
2fd6b7f5 | 1225 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1226 | goto repeat; |
1227 | } | |
2fd6b7f5 | 1228 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1229 | } |
1230 | /* | |
1231 | * All done at this level ... ascend and resume the search. | |
1232 | */ | |
ca5358ef AV |
1233 | rcu_read_lock(); |
1234 | ascend: | |
1da177e4 | 1235 | if (this_parent != parent) { |
c826cb7d | 1236 | struct dentry *child = this_parent; |
31dec132 AV |
1237 | this_parent = child->d_parent; |
1238 | ||
31dec132 AV |
1239 | spin_unlock(&child->d_lock); |
1240 | spin_lock(&this_parent->d_lock); | |
1241 | ||
ca5358ef AV |
1242 | /* might go back up the wrong parent if we have had a rename. */ |
1243 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1244 | goto rename_retry; |
2159184e AV |
1245 | /* go into the first sibling still alive */ |
1246 | do { | |
1247 | next = child->d_child.next; | |
ca5358ef AV |
1248 | if (next == &this_parent->d_subdirs) |
1249 | goto ascend; | |
1250 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1251 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1252 | rcu_read_unlock(); |
1da177e4 LT |
1253 | goto resume; |
1254 | } | |
ca5358ef | 1255 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1256 | goto rename_retry; |
ca5358ef | 1257 | rcu_read_unlock(); |
db14fc3a MS |
1258 | if (finish) |
1259 | finish(data); | |
1260 | ||
1261 | out_unlock: | |
1262 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1263 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1264 | return; |
58db63d0 NP |
1265 | |
1266 | rename_retry: | |
ca5358ef AV |
1267 | spin_unlock(&this_parent->d_lock); |
1268 | rcu_read_unlock(); | |
1269 | BUG_ON(seq & 1); | |
db14fc3a MS |
1270 | if (!retry) |
1271 | return; | |
48f5ec21 | 1272 | seq = 1; |
58db63d0 | 1273 | goto again; |
1da177e4 | 1274 | } |
db14fc3a MS |
1275 | |
1276 | /* | |
1277 | * Search for at least 1 mount point in the dentry's subdirs. | |
1278 | * We descend to the next level whenever the d_subdirs | |
1279 | * list is non-empty and continue searching. | |
1280 | */ | |
1281 | ||
db14fc3a MS |
1282 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1283 | { | |
1284 | int *ret = data; | |
1285 | if (d_mountpoint(dentry)) { | |
1286 | *ret = 1; | |
1287 | return D_WALK_QUIT; | |
1288 | } | |
1289 | return D_WALK_CONTINUE; | |
1290 | } | |
1291 | ||
69c88dc7 RD |
1292 | /** |
1293 | * have_submounts - check for mounts over a dentry | |
1294 | * @parent: dentry to check. | |
1295 | * | |
1296 | * Return true if the parent or its subdirectories contain | |
1297 | * a mount point | |
1298 | */ | |
db14fc3a MS |
1299 | int have_submounts(struct dentry *parent) |
1300 | { | |
1301 | int ret = 0; | |
1302 | ||
1303 | d_walk(parent, &ret, check_mount, NULL); | |
1304 | ||
1305 | return ret; | |
1306 | } | |
ec4f8605 | 1307 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1308 | |
eed81007 MS |
1309 | /* |
1310 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1311 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1312 | * subtree can become unreachable). | |
1313 | * | |
1ffe46d1 | 1314 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1315 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1316 | */ | |
1317 | int d_set_mounted(struct dentry *dentry) | |
1318 | { | |
1319 | struct dentry *p; | |
1320 | int ret = -ENOENT; | |
1321 | write_seqlock(&rename_lock); | |
1322 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1323 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1324 | spin_lock(&p->d_lock); |
1325 | if (unlikely(d_unhashed(p))) { | |
1326 | spin_unlock(&p->d_lock); | |
1327 | goto out; | |
1328 | } | |
1329 | spin_unlock(&p->d_lock); | |
1330 | } | |
1331 | spin_lock(&dentry->d_lock); | |
1332 | if (!d_unlinked(dentry)) { | |
1333 | dentry->d_flags |= DCACHE_MOUNTED; | |
1334 | ret = 0; | |
1335 | } | |
1336 | spin_unlock(&dentry->d_lock); | |
1337 | out: | |
1338 | write_sequnlock(&rename_lock); | |
1339 | return ret; | |
1340 | } | |
1341 | ||
1da177e4 | 1342 | /* |
fd517909 | 1343 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1344 | * and move any unused dentries to the end of the unused |
1345 | * list for prune_dcache(). We descend to the next level | |
1346 | * whenever the d_subdirs list is non-empty and continue | |
1347 | * searching. | |
1348 | * | |
1349 | * It returns zero iff there are no unused children, | |
1350 | * otherwise it returns the number of children moved to | |
1351 | * the end of the unused list. This may not be the total | |
1352 | * number of unused children, because select_parent can | |
1353 | * drop the lock and return early due to latency | |
1354 | * constraints. | |
1355 | */ | |
1da177e4 | 1356 | |
db14fc3a MS |
1357 | struct select_data { |
1358 | struct dentry *start; | |
1359 | struct list_head dispose; | |
1360 | int found; | |
1361 | }; | |
23044507 | 1362 | |
db14fc3a MS |
1363 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1364 | { | |
1365 | struct select_data *data = _data; | |
1366 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1367 | |
db14fc3a MS |
1368 | if (data->start == dentry) |
1369 | goto out; | |
2fd6b7f5 | 1370 | |
fe91522a | 1371 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1372 | data->found++; |
fe91522a AV |
1373 | } else { |
1374 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1375 | d_lru_del(dentry); | |
1376 | if (!dentry->d_lockref.count) { | |
1377 | d_shrink_add(dentry, &data->dispose); | |
1378 | data->found++; | |
1379 | } | |
1da177e4 | 1380 | } |
db14fc3a MS |
1381 | /* |
1382 | * We can return to the caller if we have found some (this | |
1383 | * ensures forward progress). We'll be coming back to find | |
1384 | * the rest. | |
1385 | */ | |
fe91522a AV |
1386 | if (!list_empty(&data->dispose)) |
1387 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1388 | out: |
db14fc3a | 1389 | return ret; |
1da177e4 LT |
1390 | } |
1391 | ||
1392 | /** | |
1393 | * shrink_dcache_parent - prune dcache | |
1394 | * @parent: parent of entries to prune | |
1395 | * | |
1396 | * Prune the dcache to remove unused children of the parent dentry. | |
1397 | */ | |
db14fc3a | 1398 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1399 | { |
db14fc3a MS |
1400 | for (;;) { |
1401 | struct select_data data; | |
1da177e4 | 1402 | |
db14fc3a MS |
1403 | INIT_LIST_HEAD(&data.dispose); |
1404 | data.start = parent; | |
1405 | data.found = 0; | |
1406 | ||
1407 | d_walk(parent, &data, select_collect, NULL); | |
1408 | if (!data.found) | |
1409 | break; | |
1410 | ||
1411 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1412 | cond_resched(); |
1413 | } | |
1da177e4 | 1414 | } |
ec4f8605 | 1415 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1416 | |
9c8c10e2 | 1417 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1418 | { |
9c8c10e2 AV |
1419 | /* it has busy descendents; complain about those instead */ |
1420 | if (!list_empty(&dentry->d_subdirs)) | |
1421 | return D_WALK_CONTINUE; | |
42c32608 | 1422 | |
9c8c10e2 AV |
1423 | /* root with refcount 1 is fine */ |
1424 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1425 | return D_WALK_CONTINUE; | |
1426 | ||
1427 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1428 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1429 | dentry, |
1430 | dentry->d_inode ? | |
1431 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1432 | dentry, |
42c32608 AV |
1433 | dentry->d_lockref.count, |
1434 | dentry->d_sb->s_type->name, | |
1435 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1436 | WARN_ON(1); |
1437 | return D_WALK_CONTINUE; | |
1438 | } | |
1439 | ||
1440 | static void do_one_tree(struct dentry *dentry) | |
1441 | { | |
1442 | shrink_dcache_parent(dentry); | |
1443 | d_walk(dentry, dentry, umount_check, NULL); | |
1444 | d_drop(dentry); | |
1445 | dput(dentry); | |
42c32608 AV |
1446 | } |
1447 | ||
1448 | /* | |
1449 | * destroy the dentries attached to a superblock on unmounting | |
1450 | */ | |
1451 | void shrink_dcache_for_umount(struct super_block *sb) | |
1452 | { | |
1453 | struct dentry *dentry; | |
1454 | ||
9c8c10e2 | 1455 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1456 | |
1457 | dentry = sb->s_root; | |
1458 | sb->s_root = NULL; | |
9c8c10e2 | 1459 | do_one_tree(dentry); |
42c32608 AV |
1460 | |
1461 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1462 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1463 | do_one_tree(dentry); | |
42c32608 AV |
1464 | } |
1465 | } | |
1466 | ||
8ed936b5 EB |
1467 | struct detach_data { |
1468 | struct select_data select; | |
1469 | struct dentry *mountpoint; | |
1470 | }; | |
1471 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1472 | { |
8ed936b5 | 1473 | struct detach_data *data = _data; |
848ac114 MS |
1474 | |
1475 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1476 | __dget_dlock(dentry); |
1477 | data->mountpoint = dentry; | |
848ac114 MS |
1478 | return D_WALK_QUIT; |
1479 | } | |
1480 | ||
8ed936b5 | 1481 | return select_collect(&data->select, dentry); |
848ac114 MS |
1482 | } |
1483 | ||
1484 | static void check_and_drop(void *_data) | |
1485 | { | |
8ed936b5 | 1486 | struct detach_data *data = _data; |
848ac114 | 1487 | |
8ed936b5 EB |
1488 | if (!data->mountpoint && !data->select.found) |
1489 | __d_drop(data->select.start); | |
848ac114 MS |
1490 | } |
1491 | ||
1492 | /** | |
1ffe46d1 EB |
1493 | * d_invalidate - detach submounts, prune dcache, and drop |
1494 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1495 | * | |
1ffe46d1 | 1496 | * no dcache lock. |
848ac114 | 1497 | * |
8ed936b5 EB |
1498 | * The final d_drop is done as an atomic operation relative to |
1499 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1500 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1501 | */ |
5542aa2f | 1502 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1503 | { |
1ffe46d1 EB |
1504 | /* |
1505 | * If it's already been dropped, return OK. | |
1506 | */ | |
1507 | spin_lock(&dentry->d_lock); | |
1508 | if (d_unhashed(dentry)) { | |
1509 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1510 | return; |
1ffe46d1 EB |
1511 | } |
1512 | spin_unlock(&dentry->d_lock); | |
1513 | ||
848ac114 MS |
1514 | /* Negative dentries can be dropped without further checks */ |
1515 | if (!dentry->d_inode) { | |
1516 | d_drop(dentry); | |
5542aa2f | 1517 | return; |
848ac114 MS |
1518 | } |
1519 | ||
1520 | for (;;) { | |
8ed936b5 | 1521 | struct detach_data data; |
848ac114 | 1522 | |
8ed936b5 EB |
1523 | data.mountpoint = NULL; |
1524 | INIT_LIST_HEAD(&data.select.dispose); | |
1525 | data.select.start = dentry; | |
1526 | data.select.found = 0; | |
1527 | ||
1528 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1529 | |
8ed936b5 EB |
1530 | if (data.select.found) |
1531 | shrink_dentry_list(&data.select.dispose); | |
848ac114 | 1532 | |
8ed936b5 EB |
1533 | if (data.mountpoint) { |
1534 | detach_mounts(data.mountpoint); | |
1535 | dput(data.mountpoint); | |
1536 | } | |
848ac114 | 1537 | |
8ed936b5 | 1538 | if (!data.mountpoint && !data.select.found) |
848ac114 MS |
1539 | break; |
1540 | ||
1541 | cond_resched(); | |
1542 | } | |
848ac114 | 1543 | } |
1ffe46d1 | 1544 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1545 | |
1da177e4 | 1546 | /** |
a4464dbc AV |
1547 | * __d_alloc - allocate a dcache entry |
1548 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1549 | * @name: qstr of the name |
1550 | * | |
1551 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1552 | * available. On a success the dentry is returned. The name passed in is | |
1553 | * copied and the copy passed in may be reused after this call. | |
1554 | */ | |
1555 | ||
a4464dbc | 1556 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1557 | { |
1558 | struct dentry *dentry; | |
1559 | char *dname; | |
285b102d | 1560 | int err; |
1da177e4 | 1561 | |
e12ba74d | 1562 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1563 | if (!dentry) |
1564 | return NULL; | |
1565 | ||
6326c71f LT |
1566 | /* |
1567 | * We guarantee that the inline name is always NUL-terminated. | |
1568 | * This way the memcpy() done by the name switching in rename | |
1569 | * will still always have a NUL at the end, even if we might | |
1570 | * be overwriting an internal NUL character | |
1571 | */ | |
1572 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd AV |
1573 | if (unlikely(!name)) { |
1574 | static const struct qstr anon = QSTR_INIT("/", 1); | |
1575 | name = &anon; | |
1576 | dname = dentry->d_iname; | |
1577 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1578 | size_t size = offsetof(struct external_name, name[1]); |
5d097056 VD |
1579 | struct external_name *p = kmalloc(size + name->len, |
1580 | GFP_KERNEL_ACCOUNT); | |
8d85b484 | 1581 | if (!p) { |
1da177e4 LT |
1582 | kmem_cache_free(dentry_cache, dentry); |
1583 | return NULL; | |
1584 | } | |
8d85b484 AV |
1585 | atomic_set(&p->u.count, 1); |
1586 | dname = p->name; | |
df4c0e36 AR |
1587 | if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS)) |
1588 | kasan_unpoison_shadow(dname, | |
1589 | round_up(name->len + 1, sizeof(unsigned long))); | |
1da177e4 LT |
1590 | } else { |
1591 | dname = dentry->d_iname; | |
1592 | } | |
1da177e4 LT |
1593 | |
1594 | dentry->d_name.len = name->len; | |
1595 | dentry->d_name.hash = name->hash; | |
1596 | memcpy(dname, name->name, name->len); | |
1597 | dname[name->len] = 0; | |
1598 | ||
6326c71f LT |
1599 | /* Make sure we always see the terminating NUL character */ |
1600 | smp_wmb(); | |
1601 | dentry->d_name.name = dname; | |
1602 | ||
98474236 | 1603 | dentry->d_lockref.count = 1; |
dea3667b | 1604 | dentry->d_flags = 0; |
1da177e4 | 1605 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1606 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1607 | dentry->d_inode = NULL; |
a4464dbc AV |
1608 | dentry->d_parent = dentry; |
1609 | dentry->d_sb = sb; | |
1da177e4 LT |
1610 | dentry->d_op = NULL; |
1611 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1612 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1613 | INIT_LIST_HEAD(&dentry->d_lru); |
1614 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1615 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1616 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1617 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1618 | |
285b102d MS |
1619 | if (dentry->d_op && dentry->d_op->d_init) { |
1620 | err = dentry->d_op->d_init(dentry); | |
1621 | if (err) { | |
1622 | if (dname_external(dentry)) | |
1623 | kfree(external_name(dentry)); | |
1624 | kmem_cache_free(dentry_cache, dentry); | |
1625 | return NULL; | |
1626 | } | |
1627 | } | |
1628 | ||
3e880fb5 | 1629 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1630 | |
1da177e4 LT |
1631 | return dentry; |
1632 | } | |
a4464dbc AV |
1633 | |
1634 | /** | |
1635 | * d_alloc - allocate a dcache entry | |
1636 | * @parent: parent of entry to allocate | |
1637 | * @name: qstr of the name | |
1638 | * | |
1639 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1640 | * available. On a success the dentry is returned. The name passed in is | |
1641 | * copied and the copy passed in may be reused after this call. | |
1642 | */ | |
1643 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1644 | { | |
1645 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1646 | if (!dentry) | |
1647 | return NULL; | |
3d56c25e | 1648 | dentry->d_flags |= DCACHE_RCUACCESS; |
a4464dbc AV |
1649 | spin_lock(&parent->d_lock); |
1650 | /* | |
1651 | * don't need child lock because it is not subject | |
1652 | * to concurrency here | |
1653 | */ | |
1654 | __dget_dlock(parent); | |
1655 | dentry->d_parent = parent; | |
946e51f2 | 1656 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1657 | spin_unlock(&parent->d_lock); |
1658 | ||
1659 | return dentry; | |
1660 | } | |
ec4f8605 | 1661 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1662 | |
ba65dc5e AV |
1663 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1664 | { | |
1665 | struct dentry *dentry = __d_alloc(parent->d_sb, NULL); | |
1666 | if (dentry) { | |
1667 | dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR; | |
1668 | dentry->d_parent = dget(parent); | |
1669 | } | |
1670 | return dentry; | |
1671 | } | |
1672 | ||
e1a24bb0 BF |
1673 | /** |
1674 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1675 | * @sb: the superblock | |
1676 | * @name: qstr of the name | |
1677 | * | |
1678 | * For a filesystem that just pins its dentries in memory and never | |
1679 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1680 | */ | |
4b936885 NP |
1681 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1682 | { | |
e1a24bb0 | 1683 | return __d_alloc(sb, name); |
4b936885 NP |
1684 | } |
1685 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1686 | ||
1da177e4 LT |
1687 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1688 | { | |
1689 | struct qstr q; | |
1690 | ||
1691 | q.name = name; | |
8387ff25 | 1692 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1693 | return d_alloc(parent, &q); |
1694 | } | |
ef26ca97 | 1695 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1696 | |
fb045adb NP |
1697 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1698 | { | |
6f7f7caa LT |
1699 | WARN_ON_ONCE(dentry->d_op); |
1700 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1701 | DCACHE_OP_COMPARE | |
1702 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1703 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1704 | DCACHE_OP_DELETE | |
d101a125 | 1705 | DCACHE_OP_REAL)); |
fb045adb NP |
1706 | dentry->d_op = op; |
1707 | if (!op) | |
1708 | return; | |
1709 | if (op->d_hash) | |
1710 | dentry->d_flags |= DCACHE_OP_HASH; | |
1711 | if (op->d_compare) | |
1712 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1713 | if (op->d_revalidate) | |
1714 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1715 | if (op->d_weak_revalidate) |
1716 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1717 | if (op->d_delete) |
1718 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1719 | if (op->d_prune) |
1720 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1721 | if (op->d_real) |
1722 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1723 | |
1724 | } | |
1725 | EXPORT_SYMBOL(d_set_d_op); | |
1726 | ||
df1a085a DH |
1727 | |
1728 | /* | |
1729 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1730 | * @dentry - The dentry to mark | |
1731 | * | |
1732 | * Mark a dentry as falling through to the lower layer (as set with | |
1733 | * d_pin_lower()). This flag may be recorded on the medium. | |
1734 | */ | |
1735 | void d_set_fallthru(struct dentry *dentry) | |
1736 | { | |
1737 | spin_lock(&dentry->d_lock); | |
1738 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1739 | spin_unlock(&dentry->d_lock); | |
1740 | } | |
1741 | EXPORT_SYMBOL(d_set_fallthru); | |
1742 | ||
b18825a7 DH |
1743 | static unsigned d_flags_for_inode(struct inode *inode) |
1744 | { | |
44bdb5e5 | 1745 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1746 | |
1747 | if (!inode) | |
1748 | return DCACHE_MISS_TYPE; | |
1749 | ||
1750 | if (S_ISDIR(inode->i_mode)) { | |
1751 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1752 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1753 | if (unlikely(!inode->i_op->lookup)) | |
1754 | add_flags = DCACHE_AUTODIR_TYPE; | |
1755 | else | |
1756 | inode->i_opflags |= IOP_LOOKUP; | |
1757 | } | |
44bdb5e5 DH |
1758 | goto type_determined; |
1759 | } | |
1760 | ||
1761 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1762 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1763 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1764 | goto type_determined; |
1765 | } | |
1766 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1767 | } |
1768 | ||
44bdb5e5 DH |
1769 | if (unlikely(!S_ISREG(inode->i_mode))) |
1770 | add_flags = DCACHE_SPECIAL_TYPE; | |
1771 | ||
1772 | type_determined: | |
b18825a7 DH |
1773 | if (unlikely(IS_AUTOMOUNT(inode))) |
1774 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1775 | return add_flags; | |
1776 | } | |
1777 | ||
360da900 OH |
1778 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1779 | { | |
b18825a7 | 1780 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1781 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1782 | |
b23fb0a6 | 1783 | spin_lock(&dentry->d_lock); |
de689f5e | 1784 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1785 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1786 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1787 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1788 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1789 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1790 | } |
1791 | ||
1da177e4 LT |
1792 | /** |
1793 | * d_instantiate - fill in inode information for a dentry | |
1794 | * @entry: dentry to complete | |
1795 | * @inode: inode to attach to this dentry | |
1796 | * | |
1797 | * Fill in inode information in the entry. | |
1798 | * | |
1799 | * This turns negative dentries into productive full members | |
1800 | * of society. | |
1801 | * | |
1802 | * NOTE! This assumes that the inode count has been incremented | |
1803 | * (or otherwise set) by the caller to indicate that it is now | |
1804 | * in use by the dcache. | |
1805 | */ | |
1806 | ||
1807 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1808 | { | |
946e51f2 | 1809 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1810 | if (inode) { |
b9680917 | 1811 | security_d_instantiate(entry, inode); |
873feea0 | 1812 | spin_lock(&inode->i_lock); |
de689f5e | 1813 | __d_instantiate(entry, inode); |
873feea0 | 1814 | spin_unlock(&inode->i_lock); |
de689f5e | 1815 | } |
1da177e4 | 1816 | } |
ec4f8605 | 1817 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1818 | |
b70a80e7 MS |
1819 | /** |
1820 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1821 | * @entry: dentry to complete | |
1822 | * @inode: inode to attach to this dentry | |
1823 | * | |
1824 | * Fill in inode information in the entry. If a directory alias is found, then | |
1825 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1826 | * guarantees that a directory inode may never have more than one alias. | |
1827 | */ | |
1828 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1829 | { | |
946e51f2 | 1830 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 | 1831 | |
b9680917 | 1832 | security_d_instantiate(entry, inode); |
b70a80e7 MS |
1833 | spin_lock(&inode->i_lock); |
1834 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1835 | spin_unlock(&inode->i_lock); | |
1836 | iput(inode); | |
1837 | return -EBUSY; | |
1838 | } | |
1839 | __d_instantiate(entry, inode); | |
1840 | spin_unlock(&inode->i_lock); | |
b70a80e7 MS |
1841 | |
1842 | return 0; | |
1843 | } | |
1844 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1845 | ||
adc0e91a AV |
1846 | struct dentry *d_make_root(struct inode *root_inode) |
1847 | { | |
1848 | struct dentry *res = NULL; | |
1849 | ||
1850 | if (root_inode) { | |
798434bd | 1851 | res = __d_alloc(root_inode->i_sb, NULL); |
adc0e91a AV |
1852 | if (res) |
1853 | d_instantiate(res, root_inode); | |
1854 | else | |
1855 | iput(root_inode); | |
1856 | } | |
1857 | return res; | |
1858 | } | |
1859 | EXPORT_SYMBOL(d_make_root); | |
1860 | ||
d891eedb BF |
1861 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1862 | { | |
1863 | struct dentry *alias; | |
1864 | ||
b3d9b7a3 | 1865 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1866 | return NULL; |
946e51f2 | 1867 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1868 | __dget(alias); |
1869 | return alias; | |
1870 | } | |
1871 | ||
46f72b34 SW |
1872 | /** |
1873 | * d_find_any_alias - find any alias for a given inode | |
1874 | * @inode: inode to find an alias for | |
1875 | * | |
1876 | * If any aliases exist for the given inode, take and return a | |
1877 | * reference for one of them. If no aliases exist, return %NULL. | |
1878 | */ | |
1879 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1880 | { |
1881 | struct dentry *de; | |
1882 | ||
1883 | spin_lock(&inode->i_lock); | |
1884 | de = __d_find_any_alias(inode); | |
1885 | spin_unlock(&inode->i_lock); | |
1886 | return de; | |
1887 | } | |
46f72b34 | 1888 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1889 | |
49c7dd28 | 1890 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1891 | { |
9308a612 CH |
1892 | struct dentry *tmp; |
1893 | struct dentry *res; | |
b18825a7 | 1894 | unsigned add_flags; |
4ea3ada2 CH |
1895 | |
1896 | if (!inode) | |
44003728 | 1897 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1898 | if (IS_ERR(inode)) |
1899 | return ERR_CAST(inode); | |
1900 | ||
d891eedb | 1901 | res = d_find_any_alias(inode); |
9308a612 CH |
1902 | if (res) |
1903 | goto out_iput; | |
1904 | ||
798434bd | 1905 | tmp = __d_alloc(inode->i_sb, NULL); |
9308a612 CH |
1906 | if (!tmp) { |
1907 | res = ERR_PTR(-ENOMEM); | |
1908 | goto out_iput; | |
4ea3ada2 | 1909 | } |
b5c84bf6 | 1910 | |
b9680917 | 1911 | security_d_instantiate(tmp, inode); |
873feea0 | 1912 | spin_lock(&inode->i_lock); |
d891eedb | 1913 | res = __d_find_any_alias(inode); |
9308a612 | 1914 | if (res) { |
873feea0 | 1915 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1916 | dput(tmp); |
1917 | goto out_iput; | |
1918 | } | |
1919 | ||
1920 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1921 | add_flags = d_flags_for_inode(inode); |
1922 | ||
1923 | if (disconnected) | |
1924 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1925 | |
9308a612 | 1926 | spin_lock(&tmp->d_lock); |
4bf46a27 | 1927 | __d_set_inode_and_type(tmp, inode, add_flags); |
946e51f2 | 1928 | hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry); |
1879fd6a | 1929 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1930 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1931 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1932 | spin_unlock(&tmp->d_lock); |
873feea0 | 1933 | spin_unlock(&inode->i_lock); |
9308a612 | 1934 | |
9308a612 CH |
1935 | return tmp; |
1936 | ||
1937 | out_iput: | |
1938 | iput(inode); | |
1939 | return res; | |
4ea3ada2 | 1940 | } |
1a0a397e BF |
1941 | |
1942 | /** | |
1943 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1944 | * @inode: inode to allocate the dentry for | |
1945 | * | |
1946 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1947 | * similar open by handle operations. The returned dentry may be anonymous, | |
1948 | * or may have a full name (if the inode was already in the cache). | |
1949 | * | |
1950 | * When called on a directory inode, we must ensure that the inode only ever | |
1951 | * has one dentry. If a dentry is found, that is returned instead of | |
1952 | * allocating a new one. | |
1953 | * | |
1954 | * On successful return, the reference to the inode has been transferred | |
1955 | * to the dentry. In case of an error the reference on the inode is released. | |
1956 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1957 | * be passed in and the error will be propagated to the return value, | |
1958 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1959 | */ | |
1960 | struct dentry *d_obtain_alias(struct inode *inode) | |
1961 | { | |
1962 | return __d_obtain_alias(inode, 1); | |
1963 | } | |
adc48720 | 1964 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1965 | |
1a0a397e BF |
1966 | /** |
1967 | * d_obtain_root - find or allocate a dentry for a given inode | |
1968 | * @inode: inode to allocate the dentry for | |
1969 | * | |
1970 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1971 | * | |
1972 | * We must ensure that directory inodes only ever have one dentry. If a | |
1973 | * dentry is found, that is returned instead of allocating a new one. | |
1974 | * | |
1975 | * On successful return, the reference to the inode has been transferred | |
1976 | * to the dentry. In case of an error the reference on the inode is | |
1977 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1978 | * error will be propagate to the return value, with a %NULL @inode | |
1979 | * replaced by ERR_PTR(-ESTALE). | |
1980 | */ | |
1981 | struct dentry *d_obtain_root(struct inode *inode) | |
1982 | { | |
1983 | return __d_obtain_alias(inode, 0); | |
1984 | } | |
1985 | EXPORT_SYMBOL(d_obtain_root); | |
1986 | ||
9403540c BN |
1987 | /** |
1988 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1989 | * @inode: the inode case-insensitive lookup has found | |
1990 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1991 | * @name: the case-exact name to be associated with the returned dentry | |
1992 | * | |
1993 | * This is to avoid filling the dcache with case-insensitive names to the | |
1994 | * same inode, only the actual correct case is stored in the dcache for | |
1995 | * case-insensitive filesystems. | |
1996 | * | |
1997 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1998 | * already exists in in the dcache, use it and return it. | |
1999 | * | |
2000 | * If no entry exists with the exact case name, allocate new dentry with | |
2001 | * the exact case, and return the spliced entry. | |
2002 | */ | |
e45b590b | 2003 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2004 | struct qstr *name) |
2005 | { | |
d9171b93 | 2006 | struct dentry *found, *res; |
9403540c | 2007 | |
b6520c81 CH |
2008 | /* |
2009 | * First check if a dentry matching the name already exists, | |
2010 | * if not go ahead and create it now. | |
2011 | */ | |
9403540c | 2012 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2013 | if (found) { |
2014 | iput(inode); | |
2015 | return found; | |
2016 | } | |
2017 | if (d_in_lookup(dentry)) { | |
2018 | found = d_alloc_parallel(dentry->d_parent, name, | |
2019 | dentry->d_wait); | |
2020 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2021 | iput(inode); | |
2022 | return found; | |
9403540c | 2023 | } |
d9171b93 AV |
2024 | } else { |
2025 | found = d_alloc(dentry->d_parent, name); | |
2026 | if (!found) { | |
2027 | iput(inode); | |
2028 | return ERR_PTR(-ENOMEM); | |
2029 | } | |
2030 | } | |
2031 | res = d_splice_alias(inode, found); | |
2032 | if (res) { | |
2033 | dput(found); | |
2034 | return res; | |
9403540c | 2035 | } |
4f522a24 | 2036 | return found; |
9403540c | 2037 | } |
ec4f8605 | 2038 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2039 | |
12f8ad4b | 2040 | |
d4c91a8f AV |
2041 | static inline bool d_same_name(const struct dentry *dentry, |
2042 | const struct dentry *parent, | |
2043 | const struct qstr *name) | |
12f8ad4b | 2044 | { |
d4c91a8f AV |
2045 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2046 | if (dentry->d_name.len != name->len) | |
2047 | return false; | |
2048 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2049 | } |
d4c91a8f AV |
2050 | return parent->d_op->d_compare(parent, dentry, |
2051 | dentry->d_name.len, dentry->d_name.name, | |
2052 | name) == 0; | |
12f8ad4b LT |
2053 | } |
2054 | ||
31e6b01f NP |
2055 | /** |
2056 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2057 | * @parent: parent dentry | |
2058 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2059 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2060 | * Returns: dentry, or NULL |
2061 | * | |
2062 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2063 | * resolution (store-free path walking) design described in | |
2064 | * Documentation/filesystems/path-lookup.txt. | |
2065 | * | |
2066 | * This is not to be used outside core vfs. | |
2067 | * | |
2068 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2069 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2070 | * without taking d_lock and checking d_seq sequence count against @seq | |
2071 | * returned here. | |
2072 | * | |
15570086 | 2073 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2074 | * function. |
2075 | * | |
2076 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2077 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2078 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2079 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2080 | * |
2081 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2082 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2083 | */ |
8966be90 LT |
2084 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2085 | const struct qstr *name, | |
da53be12 | 2086 | unsigned *seqp) |
31e6b01f | 2087 | { |
26fe5750 | 2088 | u64 hashlen = name->hash_len; |
31e6b01f | 2089 | const unsigned char *str = name->name; |
8387ff25 | 2090 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2091 | struct hlist_bl_node *node; |
31e6b01f NP |
2092 | struct dentry *dentry; |
2093 | ||
2094 | /* | |
2095 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2096 | * required to prevent single threaded performance regressions | |
2097 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2098 | * Keep the two functions in sync. | |
2099 | */ | |
2100 | ||
2101 | /* | |
2102 | * The hash list is protected using RCU. | |
2103 | * | |
2104 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2105 | * races with d_move(). | |
2106 | * | |
2107 | * It is possible that concurrent renames can mess up our list | |
2108 | * walk here and result in missing our dentry, resulting in the | |
2109 | * false-negative result. d_lookup() protects against concurrent | |
2110 | * renames using rename_lock seqlock. | |
2111 | * | |
b0a4bb83 | 2112 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2113 | */ |
b07ad996 | 2114 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2115 | unsigned seq; |
31e6b01f | 2116 | |
31e6b01f | 2117 | seqretry: |
12f8ad4b LT |
2118 | /* |
2119 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2120 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2121 | * |
2122 | * The caller must perform a seqcount check in order | |
da53be12 | 2123 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2124 | * |
2125 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2126 | * we don't wait for the sequence count to stabilize if it | |
2127 | * is in the middle of a sequence change. If we do the slow | |
2128 | * dentry compare, we will do seqretries until it is stable, | |
2129 | * and if we end up with a successful lookup, we actually | |
2130 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2131 | * |
2132 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2133 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2134 | */ |
2135 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2136 | if (dentry->d_parent != parent) |
2137 | continue; | |
2e321806 LT |
2138 | if (d_unhashed(dentry)) |
2139 | continue; | |
12f8ad4b | 2140 | |
830c0f0e | 2141 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
d4c91a8f AV |
2142 | int tlen; |
2143 | const char *tname; | |
26fe5750 LT |
2144 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2145 | continue; | |
d4c91a8f AV |
2146 | tlen = dentry->d_name.len; |
2147 | tname = dentry->d_name.name; | |
2148 | /* we want a consistent (name,len) pair */ | |
2149 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2150 | cpu_relax(); | |
12f8ad4b LT |
2151 | goto seqretry; |
2152 | } | |
d4c91a8f AV |
2153 | if (parent->d_op->d_compare(parent, dentry, |
2154 | tlen, tname, name) != 0) | |
2155 | continue; | |
2156 | } else { | |
2157 | if (dentry->d_name.hash_len != hashlen) | |
2158 | continue; | |
2159 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2160 | continue; | |
31e6b01f | 2161 | } |
da53be12 | 2162 | *seqp = seq; |
d4c91a8f | 2163 | return dentry; |
31e6b01f NP |
2164 | } |
2165 | return NULL; | |
2166 | } | |
2167 | ||
1da177e4 LT |
2168 | /** |
2169 | * d_lookup - search for a dentry | |
2170 | * @parent: parent dentry | |
2171 | * @name: qstr of name we wish to find | |
b04f784e | 2172 | * Returns: dentry, or NULL |
1da177e4 | 2173 | * |
b04f784e NP |
2174 | * d_lookup searches the children of the parent dentry for the name in |
2175 | * question. If the dentry is found its reference count is incremented and the | |
2176 | * dentry is returned. The caller must use dput to free the entry when it has | |
2177 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2178 | */ |
da2d8455 | 2179 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2180 | { |
31e6b01f | 2181 | struct dentry *dentry; |
949854d0 | 2182 | unsigned seq; |
1da177e4 | 2183 | |
b8314f93 DY |
2184 | do { |
2185 | seq = read_seqbegin(&rename_lock); | |
2186 | dentry = __d_lookup(parent, name); | |
2187 | if (dentry) | |
1da177e4 LT |
2188 | break; |
2189 | } while (read_seqretry(&rename_lock, seq)); | |
2190 | return dentry; | |
2191 | } | |
ec4f8605 | 2192 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2193 | |
31e6b01f | 2194 | /** |
b04f784e NP |
2195 | * __d_lookup - search for a dentry (racy) |
2196 | * @parent: parent dentry | |
2197 | * @name: qstr of name we wish to find | |
2198 | * Returns: dentry, or NULL | |
2199 | * | |
2200 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2201 | * false-negative result due to unrelated rename activity. | |
2202 | * | |
2203 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2204 | * however it must be used carefully, eg. with a following d_lookup in | |
2205 | * the case of failure. | |
2206 | * | |
2207 | * __d_lookup callers must be commented. | |
2208 | */ | |
a713ca2a | 2209 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2210 | { |
1da177e4 | 2211 | unsigned int hash = name->hash; |
8387ff25 | 2212 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2213 | struct hlist_bl_node *node; |
31e6b01f | 2214 | struct dentry *found = NULL; |
665a7583 | 2215 | struct dentry *dentry; |
1da177e4 | 2216 | |
31e6b01f NP |
2217 | /* |
2218 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2219 | * required to prevent single threaded performance regressions | |
2220 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2221 | * Keep the two functions in sync. | |
2222 | */ | |
2223 | ||
b04f784e NP |
2224 | /* |
2225 | * The hash list is protected using RCU. | |
2226 | * | |
2227 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2228 | * with d_move(). | |
2229 | * | |
2230 | * It is possible that concurrent renames can mess up our list | |
2231 | * walk here and result in missing our dentry, resulting in the | |
2232 | * false-negative result. d_lookup() protects against concurrent | |
2233 | * renames using rename_lock seqlock. | |
2234 | * | |
b0a4bb83 | 2235 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2236 | */ |
1da177e4 LT |
2237 | rcu_read_lock(); |
2238 | ||
b07ad996 | 2239 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2240 | |
1da177e4 LT |
2241 | if (dentry->d_name.hash != hash) |
2242 | continue; | |
1da177e4 LT |
2243 | |
2244 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2245 | if (dentry->d_parent != parent) |
2246 | goto next; | |
d0185c08 LT |
2247 | if (d_unhashed(dentry)) |
2248 | goto next; | |
2249 | ||
d4c91a8f AV |
2250 | if (!d_same_name(dentry, parent, name)) |
2251 | goto next; | |
1da177e4 | 2252 | |
98474236 | 2253 | dentry->d_lockref.count++; |
d0185c08 | 2254 | found = dentry; |
1da177e4 LT |
2255 | spin_unlock(&dentry->d_lock); |
2256 | break; | |
2257 | next: | |
2258 | spin_unlock(&dentry->d_lock); | |
2259 | } | |
2260 | rcu_read_unlock(); | |
2261 | ||
2262 | return found; | |
2263 | } | |
2264 | ||
3e7e241f EB |
2265 | /** |
2266 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2267 | * @dir: Directory to search in | |
2268 | * @name: qstr of name we wish to find | |
2269 | * | |
4f522a24 | 2270 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2271 | */ |
2272 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2273 | { | |
3e7e241f EB |
2274 | /* |
2275 | * Check for a fs-specific hash function. Note that we must | |
2276 | * calculate the standard hash first, as the d_op->d_hash() | |
2277 | * routine may choose to leave the hash value unchanged. | |
2278 | */ | |
8387ff25 | 2279 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2280 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2281 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2282 | if (unlikely(err < 0)) |
2283 | return ERR_PTR(err); | |
3e7e241f | 2284 | } |
4f522a24 | 2285 | return d_lookup(dir, name); |
3e7e241f | 2286 | } |
4f522a24 | 2287 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2288 | |
1da177e4 LT |
2289 | /* |
2290 | * When a file is deleted, we have two options: | |
2291 | * - turn this dentry into a negative dentry | |
2292 | * - unhash this dentry and free it. | |
2293 | * | |
2294 | * Usually, we want to just turn this into | |
2295 | * a negative dentry, but if anybody else is | |
2296 | * currently using the dentry or the inode | |
2297 | * we can't do that and we fall back on removing | |
2298 | * it from the hash queues and waiting for | |
2299 | * it to be deleted later when it has no users | |
2300 | */ | |
2301 | ||
2302 | /** | |
2303 | * d_delete - delete a dentry | |
2304 | * @dentry: The dentry to delete | |
2305 | * | |
2306 | * Turn the dentry into a negative dentry if possible, otherwise | |
2307 | * remove it from the hash queues so it can be deleted later | |
2308 | */ | |
2309 | ||
2310 | void d_delete(struct dentry * dentry) | |
2311 | { | |
873feea0 | 2312 | struct inode *inode; |
7a91bf7f | 2313 | int isdir = 0; |
1da177e4 LT |
2314 | /* |
2315 | * Are we the only user? | |
2316 | */ | |
357f8e65 | 2317 | again: |
1da177e4 | 2318 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2319 | inode = dentry->d_inode; |
2320 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2321 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2322 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2323 | spin_unlock(&dentry->d_lock); |
2324 | cpu_relax(); | |
2325 | goto again; | |
2326 | } | |
13e3c5e5 | 2327 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2328 | dentry_unlink_inode(dentry); |
7a91bf7f | 2329 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2330 | return; |
2331 | } | |
2332 | ||
2333 | if (!d_unhashed(dentry)) | |
2334 | __d_drop(dentry); | |
2335 | ||
2336 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2337 | |
2338 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2339 | } |
ec4f8605 | 2340 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2341 | |
15d3c589 | 2342 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2343 | { |
15d3c589 | 2344 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
ceb5bdc2 | 2345 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2346 | hlist_bl_lock(b); |
b07ad996 | 2347 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2348 | hlist_bl_unlock(b); |
1da177e4 LT |
2349 | } |
2350 | ||
2351 | /** | |
2352 | * d_rehash - add an entry back to the hash | |
2353 | * @entry: dentry to add to the hash | |
2354 | * | |
2355 | * Adds a dentry to the hash according to its name. | |
2356 | */ | |
2357 | ||
2358 | void d_rehash(struct dentry * entry) | |
2359 | { | |
1da177e4 | 2360 | spin_lock(&entry->d_lock); |
15d3c589 | 2361 | __d_rehash(entry); |
1da177e4 | 2362 | spin_unlock(&entry->d_lock); |
1da177e4 | 2363 | } |
ec4f8605 | 2364 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2365 | |
84e710da AV |
2366 | static inline unsigned start_dir_add(struct inode *dir) |
2367 | { | |
2368 | ||
2369 | for (;;) { | |
2370 | unsigned n = dir->i_dir_seq; | |
2371 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2372 | return n; | |
2373 | cpu_relax(); | |
2374 | } | |
2375 | } | |
2376 | ||
2377 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2378 | { | |
2379 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2380 | } | |
2381 | ||
d9171b93 AV |
2382 | static void d_wait_lookup(struct dentry *dentry) |
2383 | { | |
2384 | if (d_in_lookup(dentry)) { | |
2385 | DECLARE_WAITQUEUE(wait, current); | |
2386 | add_wait_queue(dentry->d_wait, &wait); | |
2387 | do { | |
2388 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2389 | spin_unlock(&dentry->d_lock); | |
2390 | schedule(); | |
2391 | spin_lock(&dentry->d_lock); | |
2392 | } while (d_in_lookup(dentry)); | |
2393 | } | |
2394 | } | |
2395 | ||
94bdd655 | 2396 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2397 | const struct qstr *name, |
2398 | wait_queue_head_t *wq) | |
94bdd655 | 2399 | { |
94bdd655 | 2400 | unsigned int hash = name->hash; |
94bdd655 AV |
2401 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2402 | struct hlist_bl_node *node; | |
2403 | struct dentry *new = d_alloc(parent, name); | |
2404 | struct dentry *dentry; | |
2405 | unsigned seq, r_seq, d_seq; | |
2406 | ||
2407 | if (unlikely(!new)) | |
2408 | return ERR_PTR(-ENOMEM); | |
2409 | ||
2410 | retry: | |
2411 | rcu_read_lock(); | |
2412 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1; | |
2413 | r_seq = read_seqbegin(&rename_lock); | |
2414 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2415 | if (unlikely(dentry)) { | |
2416 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2417 | rcu_read_unlock(); | |
2418 | goto retry; | |
2419 | } | |
2420 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2421 | rcu_read_unlock(); | |
2422 | dput(dentry); | |
2423 | goto retry; | |
2424 | } | |
2425 | rcu_read_unlock(); | |
2426 | dput(new); | |
2427 | return dentry; | |
2428 | } | |
2429 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2430 | rcu_read_unlock(); | |
2431 | goto retry; | |
2432 | } | |
2433 | hlist_bl_lock(b); | |
2434 | if (unlikely(parent->d_inode->i_dir_seq != seq)) { | |
2435 | hlist_bl_unlock(b); | |
2436 | rcu_read_unlock(); | |
2437 | goto retry; | |
2438 | } | |
94bdd655 AV |
2439 | /* |
2440 | * No changes for the parent since the beginning of d_lookup(). | |
2441 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2442 | * any potential in-lookup matches are going to stay here until | |
2443 | * we unlock the chain. All fields are stable in everything | |
2444 | * we encounter. | |
2445 | */ | |
2446 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2447 | if (dentry->d_name.hash != hash) | |
2448 | continue; | |
2449 | if (dentry->d_parent != parent) | |
2450 | continue; | |
d4c91a8f AV |
2451 | if (!d_same_name(dentry, parent, name)) |
2452 | continue; | |
94bdd655 | 2453 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2454 | /* now we can try to grab a reference */ |
2455 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2456 | rcu_read_unlock(); | |
2457 | goto retry; | |
2458 | } | |
2459 | ||
2460 | rcu_read_unlock(); | |
2461 | /* | |
2462 | * somebody is likely to be still doing lookup for it; | |
2463 | * wait for them to finish | |
2464 | */ | |
d9171b93 AV |
2465 | spin_lock(&dentry->d_lock); |
2466 | d_wait_lookup(dentry); | |
2467 | /* | |
2468 | * it's not in-lookup anymore; in principle we should repeat | |
2469 | * everything from dcache lookup, but it's likely to be what | |
2470 | * d_lookup() would've found anyway. If it is, just return it; | |
2471 | * otherwise we really have to repeat the whole thing. | |
2472 | */ | |
2473 | if (unlikely(dentry->d_name.hash != hash)) | |
2474 | goto mismatch; | |
2475 | if (unlikely(dentry->d_parent != parent)) | |
2476 | goto mismatch; | |
2477 | if (unlikely(d_unhashed(dentry))) | |
2478 | goto mismatch; | |
d4c91a8f AV |
2479 | if (unlikely(!d_same_name(dentry, parent, name))) |
2480 | goto mismatch; | |
d9171b93 AV |
2481 | /* OK, it *is* a hashed match; return it */ |
2482 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2483 | dput(new); |
2484 | return dentry; | |
2485 | } | |
e7d6ef97 | 2486 | rcu_read_unlock(); |
94bdd655 AV |
2487 | /* we can't take ->d_lock here; it's OK, though. */ |
2488 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2489 | new->d_wait = wq; |
94bdd655 AV |
2490 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2491 | hlist_bl_unlock(b); | |
2492 | return new; | |
d9171b93 AV |
2493 | mismatch: |
2494 | spin_unlock(&dentry->d_lock); | |
2495 | dput(dentry); | |
2496 | goto retry; | |
94bdd655 AV |
2497 | } |
2498 | EXPORT_SYMBOL(d_alloc_parallel); | |
2499 | ||
85c7f810 AV |
2500 | void __d_lookup_done(struct dentry *dentry) |
2501 | { | |
94bdd655 AV |
2502 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2503 | dentry->d_name.hash); | |
2504 | hlist_bl_lock(b); | |
85c7f810 | 2505 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2506 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2507 | wake_up_all(dentry->d_wait); |
2508 | dentry->d_wait = NULL; | |
94bdd655 AV |
2509 | hlist_bl_unlock(b); |
2510 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2511 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2512 | } |
2513 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2514 | |
2515 | /* inode->i_lock held if inode is non-NULL */ | |
2516 | ||
2517 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2518 | { | |
84e710da AV |
2519 | struct inode *dir = NULL; |
2520 | unsigned n; | |
0568d705 | 2521 | spin_lock(&dentry->d_lock); |
84e710da AV |
2522 | if (unlikely(d_in_lookup(dentry))) { |
2523 | dir = dentry->d_parent->d_inode; | |
2524 | n = start_dir_add(dir); | |
85c7f810 | 2525 | __d_lookup_done(dentry); |
84e710da | 2526 | } |
ed782b5a | 2527 | if (inode) { |
0568d705 AV |
2528 | unsigned add_flags = d_flags_for_inode(inode); |
2529 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2530 | raw_write_seqcount_begin(&dentry->d_seq); | |
2531 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2532 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2533 | fsnotify_update_flags(dentry); |
ed782b5a | 2534 | } |
15d3c589 | 2535 | __d_rehash(dentry); |
84e710da AV |
2536 | if (dir) |
2537 | end_dir_add(dir, n); | |
0568d705 AV |
2538 | spin_unlock(&dentry->d_lock); |
2539 | if (inode) | |
2540 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2541 | } |
2542 | ||
34d0d19d AV |
2543 | /** |
2544 | * d_add - add dentry to hash queues | |
2545 | * @entry: dentry to add | |
2546 | * @inode: The inode to attach to this dentry | |
2547 | * | |
2548 | * This adds the entry to the hash queues and initializes @inode. | |
2549 | * The entry was actually filled in earlier during d_alloc(). | |
2550 | */ | |
2551 | ||
2552 | void d_add(struct dentry *entry, struct inode *inode) | |
2553 | { | |
b9680917 AV |
2554 | if (inode) { |
2555 | security_d_instantiate(entry, inode); | |
ed782b5a | 2556 | spin_lock(&inode->i_lock); |
b9680917 | 2557 | } |
ed782b5a | 2558 | __d_add(entry, inode); |
34d0d19d AV |
2559 | } |
2560 | EXPORT_SYMBOL(d_add); | |
2561 | ||
668d0cd5 AV |
2562 | /** |
2563 | * d_exact_alias - find and hash an exact unhashed alias | |
2564 | * @entry: dentry to add | |
2565 | * @inode: The inode to go with this dentry | |
2566 | * | |
2567 | * If an unhashed dentry with the same name/parent and desired | |
2568 | * inode already exists, hash and return it. Otherwise, return | |
2569 | * NULL. | |
2570 | * | |
2571 | * Parent directory should be locked. | |
2572 | */ | |
2573 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2574 | { | |
2575 | struct dentry *alias; | |
668d0cd5 AV |
2576 | unsigned int hash = entry->d_name.hash; |
2577 | ||
2578 | spin_lock(&inode->i_lock); | |
2579 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2580 | /* | |
2581 | * Don't need alias->d_lock here, because aliases with | |
2582 | * d_parent == entry->d_parent are not subject to name or | |
2583 | * parent changes, because the parent inode i_mutex is held. | |
2584 | */ | |
2585 | if (alias->d_name.hash != hash) | |
2586 | continue; | |
2587 | if (alias->d_parent != entry->d_parent) | |
2588 | continue; | |
d4c91a8f | 2589 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2590 | continue; |
2591 | spin_lock(&alias->d_lock); | |
2592 | if (!d_unhashed(alias)) { | |
2593 | spin_unlock(&alias->d_lock); | |
2594 | alias = NULL; | |
2595 | } else { | |
2596 | __dget_dlock(alias); | |
15d3c589 | 2597 | __d_rehash(alias); |
668d0cd5 AV |
2598 | spin_unlock(&alias->d_lock); |
2599 | } | |
2600 | spin_unlock(&inode->i_lock); | |
2601 | return alias; | |
2602 | } | |
2603 | spin_unlock(&inode->i_lock); | |
2604 | return NULL; | |
2605 | } | |
2606 | EXPORT_SYMBOL(d_exact_alias); | |
2607 | ||
fb2d5b86 NP |
2608 | /** |
2609 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2610 | * @dentry: dentry to be updated | |
2611 | * @name: new name | |
2612 | * | |
2613 | * Update a case insensitive dentry with new case of name. | |
2614 | * | |
2615 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2616 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2617 | * lengths). | |
2618 | * | |
2619 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2620 | * keep renames and concurrent inserts, and readdir(2) away). | |
2621 | */ | |
2622 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2623 | { | |
5955102c | 2624 | BUG_ON(!inode_is_locked(dentry->d_parent->d_inode)); |
fb2d5b86 NP |
2625 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2626 | ||
fb2d5b86 | 2627 | spin_lock(&dentry->d_lock); |
31e6b01f | 2628 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2629 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2630 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2631 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2632 | } |
2633 | EXPORT_SYMBOL(dentry_update_name_case); | |
2634 | ||
8d85b484 | 2635 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2636 | { |
8d85b484 AV |
2637 | if (unlikely(dname_external(target))) { |
2638 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2639 | /* |
2640 | * Both external: swap the pointers | |
2641 | */ | |
9a8d5bb4 | 2642 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2643 | } else { |
2644 | /* | |
2645 | * dentry:internal, target:external. Steal target's | |
2646 | * storage and make target internal. | |
2647 | */ | |
321bcf92 BF |
2648 | memcpy(target->d_iname, dentry->d_name.name, |
2649 | dentry->d_name.len + 1); | |
1da177e4 LT |
2650 | dentry->d_name.name = target->d_name.name; |
2651 | target->d_name.name = target->d_iname; | |
2652 | } | |
2653 | } else { | |
8d85b484 | 2654 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2655 | /* |
2656 | * dentry:external, target:internal. Give dentry's | |
2657 | * storage to target and make dentry internal | |
2658 | */ | |
2659 | memcpy(dentry->d_iname, target->d_name.name, | |
2660 | target->d_name.len + 1); | |
2661 | target->d_name.name = dentry->d_name.name; | |
2662 | dentry->d_name.name = dentry->d_iname; | |
2663 | } else { | |
2664 | /* | |
da1ce067 | 2665 | * Both are internal. |
1da177e4 | 2666 | */ |
da1ce067 MS |
2667 | unsigned int i; |
2668 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
08d4f772 MP |
2669 | kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN); |
2670 | kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN); | |
da1ce067 MS |
2671 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { |
2672 | swap(((long *) &dentry->d_iname)[i], | |
2673 | ((long *) &target->d_iname)[i]); | |
2674 | } | |
1da177e4 LT |
2675 | } |
2676 | } | |
a28ddb87 | 2677 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2678 | } |
2679 | ||
8d85b484 AV |
2680 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2681 | { | |
2682 | struct external_name *old_name = NULL; | |
2683 | if (unlikely(dname_external(dentry))) | |
2684 | old_name = external_name(dentry); | |
2685 | if (unlikely(dname_external(target))) { | |
2686 | atomic_inc(&external_name(target)->u.count); | |
2687 | dentry->d_name = target->d_name; | |
2688 | } else { | |
2689 | memcpy(dentry->d_iname, target->d_name.name, | |
2690 | target->d_name.len + 1); | |
2691 | dentry->d_name.name = dentry->d_iname; | |
2692 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2693 | } | |
2694 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2695 | kfree_rcu(old_name, u.head); | |
2696 | } | |
2697 | ||
2fd6b7f5 NP |
2698 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2699 | { | |
2700 | /* | |
2701 | * XXXX: do we really need to take target->d_lock? | |
2702 | */ | |
2703 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2704 | spin_lock(&target->d_parent->d_lock); | |
2705 | else { | |
2706 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2707 | spin_lock(&dentry->d_parent->d_lock); | |
2708 | spin_lock_nested(&target->d_parent->d_lock, | |
2709 | DENTRY_D_LOCK_NESTED); | |
2710 | } else { | |
2711 | spin_lock(&target->d_parent->d_lock); | |
2712 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2713 | DENTRY_D_LOCK_NESTED); | |
2714 | } | |
2715 | } | |
2716 | if (target < dentry) { | |
2717 | spin_lock_nested(&target->d_lock, 2); | |
2718 | spin_lock_nested(&dentry->d_lock, 3); | |
2719 | } else { | |
2720 | spin_lock_nested(&dentry->d_lock, 2); | |
2721 | spin_lock_nested(&target->d_lock, 3); | |
2722 | } | |
2723 | } | |
2724 | ||
986c0194 | 2725 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2726 | { |
2727 | if (target->d_parent != dentry->d_parent) | |
2728 | spin_unlock(&dentry->d_parent->d_lock); | |
2729 | if (target->d_parent != target) | |
2730 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2731 | spin_unlock(&target->d_lock); |
2732 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2733 | } |
2734 | ||
1da177e4 | 2735 | /* |
2fd6b7f5 NP |
2736 | * When switching names, the actual string doesn't strictly have to |
2737 | * be preserved in the target - because we're dropping the target | |
2738 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2739 | * the new name before we switch, unless we are going to rehash |
2740 | * it. Note that if we *do* unhash the target, we are not allowed | |
2741 | * to rehash it without giving it a new name/hash key - whether | |
2742 | * we swap or overwrite the names here, resulting name won't match | |
2743 | * the reality in filesystem; it's only there for d_path() purposes. | |
2744 | * Note that all of this is happening under rename_lock, so the | |
2745 | * any hash lookup seeing it in the middle of manipulations will | |
2746 | * be discarded anyway. So we do not care what happens to the hash | |
2747 | * key in that case. | |
1da177e4 | 2748 | */ |
9eaef27b | 2749 | /* |
18367501 | 2750 | * __d_move - move a dentry |
1da177e4 LT |
2751 | * @dentry: entry to move |
2752 | * @target: new dentry | |
da1ce067 | 2753 | * @exchange: exchange the two dentries |
1da177e4 LT |
2754 | * |
2755 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2756 | * dcache entries should not be moved in this way. Caller must hold |
2757 | * rename_lock, the i_mutex of the source and target directories, | |
2758 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2759 | */ |
da1ce067 MS |
2760 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2761 | bool exchange) | |
1da177e4 | 2762 | { |
84e710da AV |
2763 | struct inode *dir = NULL; |
2764 | unsigned n; | |
1da177e4 LT |
2765 | if (!dentry->d_inode) |
2766 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2767 | ||
2fd6b7f5 NP |
2768 | BUG_ON(d_ancestor(dentry, target)); |
2769 | BUG_ON(d_ancestor(target, dentry)); | |
2770 | ||
2fd6b7f5 | 2771 | dentry_lock_for_move(dentry, target); |
84e710da AV |
2772 | if (unlikely(d_in_lookup(target))) { |
2773 | dir = target->d_parent->d_inode; | |
2774 | n = start_dir_add(dir); | |
85c7f810 | 2775 | __d_lookup_done(target); |
84e710da | 2776 | } |
1da177e4 | 2777 | |
31e6b01f | 2778 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2779 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2780 | |
15d3c589 | 2781 | /* unhash both */ |
ceb5bdc2 | 2782 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
ceb5bdc2 | 2783 | __d_drop(dentry); |
1da177e4 LT |
2784 | __d_drop(target); |
2785 | ||
1da177e4 | 2786 | /* Switch the names.. */ |
8d85b484 AV |
2787 | if (exchange) |
2788 | swap_names(dentry, target); | |
2789 | else | |
2790 | copy_name(dentry, target); | |
1da177e4 | 2791 | |
15d3c589 AV |
2792 | /* rehash in new place(s) */ |
2793 | __d_rehash(dentry); | |
2794 | if (exchange) | |
2795 | __d_rehash(target); | |
2796 | ||
63cf427a | 2797 | /* ... and switch them in the tree */ |
1da177e4 | 2798 | if (IS_ROOT(dentry)) { |
63cf427a | 2799 | /* splicing a tree */ |
3d56c25e | 2800 | dentry->d_flags |= DCACHE_RCUACCESS; |
1da177e4 LT |
2801 | dentry->d_parent = target->d_parent; |
2802 | target->d_parent = target; | |
946e51f2 AV |
2803 | list_del_init(&target->d_child); |
2804 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
1da177e4 | 2805 | } else { |
63cf427a | 2806 | /* swapping two dentries */ |
9a8d5bb4 | 2807 | swap(dentry->d_parent, target->d_parent); |
946e51f2 AV |
2808 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
2809 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
63cf427a | 2810 | if (exchange) |
affda484 AV |
2811 | fsnotify_update_flags(target); |
2812 | fsnotify_update_flags(dentry); | |
1da177e4 LT |
2813 | } |
2814 | ||
31e6b01f NP |
2815 | write_seqcount_end(&target->d_seq); |
2816 | write_seqcount_end(&dentry->d_seq); | |
2817 | ||
84e710da AV |
2818 | if (dir) |
2819 | end_dir_add(dir, n); | |
986c0194 | 2820 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2821 | } |
2822 | ||
2823 | /* | |
2824 | * d_move - move a dentry | |
2825 | * @dentry: entry to move | |
2826 | * @target: new dentry | |
2827 | * | |
2828 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2829 | * dcache entries should not be moved in this way. See the locking |
2830 | * requirements for __d_move. | |
18367501 AV |
2831 | */ |
2832 | void d_move(struct dentry *dentry, struct dentry *target) | |
2833 | { | |
2834 | write_seqlock(&rename_lock); | |
da1ce067 | 2835 | __d_move(dentry, target, false); |
1da177e4 | 2836 | write_sequnlock(&rename_lock); |
9eaef27b | 2837 | } |
ec4f8605 | 2838 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2839 | |
da1ce067 MS |
2840 | /* |
2841 | * d_exchange - exchange two dentries | |
2842 | * @dentry1: first dentry | |
2843 | * @dentry2: second dentry | |
2844 | */ | |
2845 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2846 | { | |
2847 | write_seqlock(&rename_lock); | |
2848 | ||
2849 | WARN_ON(!dentry1->d_inode); | |
2850 | WARN_ON(!dentry2->d_inode); | |
2851 | WARN_ON(IS_ROOT(dentry1)); | |
2852 | WARN_ON(IS_ROOT(dentry2)); | |
2853 | ||
2854 | __d_move(dentry1, dentry2, true); | |
2855 | ||
2856 | write_sequnlock(&rename_lock); | |
2857 | } | |
2858 | ||
e2761a11 OH |
2859 | /** |
2860 | * d_ancestor - search for an ancestor | |
2861 | * @p1: ancestor dentry | |
2862 | * @p2: child dentry | |
2863 | * | |
2864 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2865 | * an ancestor of p2, else NULL. | |
9eaef27b | 2866 | */ |
e2761a11 | 2867 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2868 | { |
2869 | struct dentry *p; | |
2870 | ||
871c0067 | 2871 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2872 | if (p->d_parent == p1) |
e2761a11 | 2873 | return p; |
9eaef27b | 2874 | } |
e2761a11 | 2875 | return NULL; |
9eaef27b TM |
2876 | } |
2877 | ||
2878 | /* | |
2879 | * This helper attempts to cope with remotely renamed directories | |
2880 | * | |
2881 | * It assumes that the caller is already holding | |
a03e283b | 2882 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2883 | * |
2884 | * Note: If ever the locking in lock_rename() changes, then please | |
2885 | * remember to update this too... | |
9eaef27b | 2886 | */ |
b5ae6b15 | 2887 | static int __d_unalias(struct inode *inode, |
873feea0 | 2888 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2889 | { |
9902af79 AV |
2890 | struct mutex *m1 = NULL; |
2891 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2892 | int ret = -ESTALE; |
9eaef27b TM |
2893 | |
2894 | /* If alias and dentry share a parent, then no extra locks required */ | |
2895 | if (alias->d_parent == dentry->d_parent) | |
2896 | goto out_unalias; | |
2897 | ||
9eaef27b | 2898 | /* See lock_rename() */ |
9eaef27b TM |
2899 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2900 | goto out_err; | |
2901 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2902 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2903 | goto out_err; |
9902af79 | 2904 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 2905 | out_unalias: |
8ed936b5 | 2906 | __d_move(alias, dentry, false); |
b5ae6b15 | 2907 | ret = 0; |
9eaef27b | 2908 | out_err: |
9eaef27b | 2909 | if (m2) |
9902af79 | 2910 | up_read(m2); |
9eaef27b TM |
2911 | if (m1) |
2912 | mutex_unlock(m1); | |
2913 | return ret; | |
2914 | } | |
2915 | ||
3f70bd51 BF |
2916 | /** |
2917 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2918 | * @inode: the inode which may have a disconnected dentry | |
2919 | * @dentry: a negative dentry which we want to point to the inode. | |
2920 | * | |
da093a9b BF |
2921 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2922 | * place of the given dentry and return it, else simply d_add the inode | |
2923 | * to the dentry and return NULL. | |
3f70bd51 | 2924 | * |
908790fa BF |
2925 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2926 | * we should error out: directories can't have multiple aliases. | |
2927 | * | |
3f70bd51 BF |
2928 | * This is needed in the lookup routine of any filesystem that is exportable |
2929 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2930 | * | |
2931 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2932 | * is returned. This matches the expected return value of ->lookup. | |
2933 | * | |
2934 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2935 | * In that case, we know that the inode will be a regular file, and also this | |
2936 | * will only occur during atomic_open. So we need to check for the dentry | |
2937 | * being already hashed only in the final case. | |
2938 | */ | |
2939 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2940 | { | |
3f70bd51 BF |
2941 | if (IS_ERR(inode)) |
2942 | return ERR_CAST(inode); | |
2943 | ||
770bfad8 DH |
2944 | BUG_ON(!d_unhashed(dentry)); |
2945 | ||
de689f5e | 2946 | if (!inode) |
b5ae6b15 | 2947 | goto out; |
de689f5e | 2948 | |
b9680917 | 2949 | security_d_instantiate(dentry, inode); |
873feea0 | 2950 | spin_lock(&inode->i_lock); |
9eaef27b | 2951 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
2952 | struct dentry *new = __d_find_any_alias(inode); |
2953 | if (unlikely(new)) { | |
a03e283b EB |
2954 | /* The reference to new ensures it remains an alias */ |
2955 | spin_unlock(&inode->i_lock); | |
18367501 | 2956 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
2957 | if (unlikely(d_ancestor(new, dentry))) { |
2958 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
2959 | dput(new); |
2960 | new = ERR_PTR(-ELOOP); | |
2961 | pr_warn_ratelimited( | |
2962 | "VFS: Lookup of '%s' in %s %s" | |
2963 | " would have caused loop\n", | |
2964 | dentry->d_name.name, | |
2965 | inode->i_sb->s_type->name, | |
2966 | inode->i_sb->s_id); | |
2967 | } else if (!IS_ROOT(new)) { | |
2968 | int err = __d_unalias(inode, dentry, new); | |
18367501 | 2969 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
2970 | if (err) { |
2971 | dput(new); | |
2972 | new = ERR_PTR(err); | |
2973 | } | |
18367501 | 2974 | } else { |
b5ae6b15 AV |
2975 | __d_move(new, dentry, false); |
2976 | write_sequnlock(&rename_lock); | |
dd179946 | 2977 | } |
b5ae6b15 AV |
2978 | iput(inode); |
2979 | return new; | |
9eaef27b | 2980 | } |
770bfad8 | 2981 | } |
b5ae6b15 | 2982 | out: |
ed782b5a | 2983 | __d_add(dentry, inode); |
b5ae6b15 | 2984 | return NULL; |
770bfad8 | 2985 | } |
b5ae6b15 | 2986 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 2987 | |
cdd16d02 | 2988 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2989 | { |
2990 | *buflen -= namelen; | |
2991 | if (*buflen < 0) | |
2992 | return -ENAMETOOLONG; | |
2993 | *buffer -= namelen; | |
2994 | memcpy(*buffer, str, namelen); | |
2995 | return 0; | |
2996 | } | |
2997 | ||
232d2d60 WL |
2998 | /** |
2999 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
3000 | * @buffer: buffer pointer |
3001 | * @buflen: allocated length of the buffer | |
3002 | * @name: name string and length qstr structure | |
232d2d60 WL |
3003 | * |
3004 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
3005 | * make sure that either the old or the new name pointer and length are | |
3006 | * fetched. However, there may be mismatch between length and pointer. | |
3007 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
3008 | * the length is reached or a null byte is found. It also prepends "/" at | |
3009 | * the beginning of the name. The sequence number check at the caller will | |
3010 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
3011 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
3012 | * |
3013 | * Data dependency barrier is needed to make sure that we see that terminating | |
3014 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 3015 | */ |
cdd16d02 MS |
3016 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
3017 | { | |
232d2d60 WL |
3018 | const char *dname = ACCESS_ONCE(name->name); |
3019 | u32 dlen = ACCESS_ONCE(name->len); | |
3020 | char *p; | |
3021 | ||
6d13f694 AV |
3022 | smp_read_barrier_depends(); |
3023 | ||
232d2d60 | 3024 | *buflen -= dlen + 1; |
e825196d AV |
3025 | if (*buflen < 0) |
3026 | return -ENAMETOOLONG; | |
232d2d60 WL |
3027 | p = *buffer -= dlen + 1; |
3028 | *p++ = '/'; | |
3029 | while (dlen--) { | |
3030 | char c = *dname++; | |
3031 | if (!c) | |
3032 | break; | |
3033 | *p++ = c; | |
3034 | } | |
3035 | return 0; | |
cdd16d02 MS |
3036 | } |
3037 | ||
1da177e4 | 3038 | /** |
208898c1 | 3039 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 3040 | * @path: the dentry/vfsmount to report |
02125a82 | 3041 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
3042 | * @buffer: pointer to the end of the buffer |
3043 | * @buflen: pointer to buffer length | |
552ce544 | 3044 | * |
18129977 WL |
3045 | * The function will first try to write out the pathname without taking any |
3046 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
3047 | * It only checks the sequence number of the global rename_lock as any change | |
3048 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
3049 | * sequence number. If the sequence number had been changed, it will restart | |
3050 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
3051 | * In this case, there is no need to take the RCU read lock as the recursive | |
3052 | * parent pointer references will keep the dentry chain alive as long as no | |
3053 | * rename operation is performed. | |
1da177e4 | 3054 | */ |
02125a82 AV |
3055 | static int prepend_path(const struct path *path, |
3056 | const struct path *root, | |
f2eb6575 | 3057 | char **buffer, int *buflen) |
1da177e4 | 3058 | { |
ede4cebc AV |
3059 | struct dentry *dentry; |
3060 | struct vfsmount *vfsmnt; | |
3061 | struct mount *mnt; | |
f2eb6575 | 3062 | int error = 0; |
48a066e7 | 3063 | unsigned seq, m_seq = 0; |
232d2d60 WL |
3064 | char *bptr; |
3065 | int blen; | |
6092d048 | 3066 | |
48f5ec21 | 3067 | rcu_read_lock(); |
48a066e7 AV |
3068 | restart_mnt: |
3069 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
3070 | seq = 0; | |
4ec6c2ae | 3071 | rcu_read_lock(); |
232d2d60 WL |
3072 | restart: |
3073 | bptr = *buffer; | |
3074 | blen = *buflen; | |
48a066e7 | 3075 | error = 0; |
ede4cebc AV |
3076 | dentry = path->dentry; |
3077 | vfsmnt = path->mnt; | |
3078 | mnt = real_mount(vfsmnt); | |
232d2d60 | 3079 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 3080 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
3081 | struct dentry * parent; |
3082 | ||
1da177e4 | 3083 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 3084 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
cde93be4 EB |
3085 | /* Escaped? */ |
3086 | if (dentry != vfsmnt->mnt_root) { | |
3087 | bptr = *buffer; | |
3088 | blen = *buflen; | |
3089 | error = 3; | |
3090 | break; | |
3091 | } | |
552ce544 | 3092 | /* Global root? */ |
48a066e7 AV |
3093 | if (mnt != parent) { |
3094 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
3095 | mnt = parent; | |
232d2d60 WL |
3096 | vfsmnt = &mnt->mnt; |
3097 | continue; | |
3098 | } | |
232d2d60 WL |
3099 | if (!error) |
3100 | error = is_mounted(vfsmnt) ? 1 : 2; | |
3101 | break; | |
1da177e4 LT |
3102 | } |
3103 | parent = dentry->d_parent; | |
3104 | prefetch(parent); | |
232d2d60 | 3105 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
3106 | if (error) |
3107 | break; | |
3108 | ||
1da177e4 LT |
3109 | dentry = parent; |
3110 | } | |
48f5ec21 AV |
3111 | if (!(seq & 1)) |
3112 | rcu_read_unlock(); | |
3113 | if (need_seqretry(&rename_lock, seq)) { | |
3114 | seq = 1; | |
232d2d60 | 3115 | goto restart; |
48f5ec21 AV |
3116 | } |
3117 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
3118 | |
3119 | if (!(m_seq & 1)) | |
3120 | rcu_read_unlock(); | |
48a066e7 AV |
3121 | if (need_seqretry(&mount_lock, m_seq)) { |
3122 | m_seq = 1; | |
3123 | goto restart_mnt; | |
3124 | } | |
3125 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 3126 | |
232d2d60 WL |
3127 | if (error >= 0 && bptr == *buffer) { |
3128 | if (--blen < 0) | |
3129 | error = -ENAMETOOLONG; | |
3130 | else | |
3131 | *--bptr = '/'; | |
3132 | } | |
3133 | *buffer = bptr; | |
3134 | *buflen = blen; | |
7ea600b5 | 3135 | return error; |
f2eb6575 | 3136 | } |
be285c71 | 3137 | |
f2eb6575 MS |
3138 | /** |
3139 | * __d_path - return the path of a dentry | |
3140 | * @path: the dentry/vfsmount to report | |
02125a82 | 3141 | * @root: root vfsmnt/dentry |
cd956a1c | 3142 | * @buf: buffer to return value in |
f2eb6575 MS |
3143 | * @buflen: buffer length |
3144 | * | |
ffd1f4ed | 3145 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
3146 | * |
3147 | * Returns a pointer into the buffer or an error code if the | |
3148 | * path was too long. | |
3149 | * | |
be148247 | 3150 | * "buflen" should be positive. |
f2eb6575 | 3151 | * |
02125a82 | 3152 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 3153 | */ |
02125a82 AV |
3154 | char *__d_path(const struct path *path, |
3155 | const struct path *root, | |
f2eb6575 MS |
3156 | char *buf, int buflen) |
3157 | { | |
3158 | char *res = buf + buflen; | |
3159 | int error; | |
3160 | ||
3161 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 3162 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 3163 | |
02125a82 AV |
3164 | if (error < 0) |
3165 | return ERR_PTR(error); | |
3166 | if (error > 0) | |
3167 | return NULL; | |
3168 | return res; | |
3169 | } | |
3170 | ||
3171 | char *d_absolute_path(const struct path *path, | |
3172 | char *buf, int buflen) | |
3173 | { | |
3174 | struct path root = {}; | |
3175 | char *res = buf + buflen; | |
3176 | int error; | |
3177 | ||
3178 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3179 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3180 | |
3181 | if (error > 1) | |
3182 | error = -EINVAL; | |
3183 | if (error < 0) | |
f2eb6575 | 3184 | return ERR_PTR(error); |
f2eb6575 | 3185 | return res; |
1da177e4 LT |
3186 | } |
3187 | ||
ffd1f4ed MS |
3188 | /* |
3189 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3190 | */ | |
02125a82 AV |
3191 | static int path_with_deleted(const struct path *path, |
3192 | const struct path *root, | |
3193 | char **buf, int *buflen) | |
ffd1f4ed MS |
3194 | { |
3195 | prepend(buf, buflen, "\0", 1); | |
3196 | if (d_unlinked(path->dentry)) { | |
3197 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3198 | if (error) | |
3199 | return error; | |
3200 | } | |
3201 | ||
3202 | return prepend_path(path, root, buf, buflen); | |
3203 | } | |
3204 | ||
8df9d1a4 MS |
3205 | static int prepend_unreachable(char **buffer, int *buflen) |
3206 | { | |
3207 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3208 | } | |
3209 | ||
68f0d9d9 LT |
3210 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3211 | { | |
3212 | unsigned seq; | |
3213 | ||
3214 | do { | |
3215 | seq = read_seqcount_begin(&fs->seq); | |
3216 | *root = fs->root; | |
3217 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3218 | } | |
3219 | ||
a03a8a70 JB |
3220 | /** |
3221 | * d_path - return the path of a dentry | |
cf28b486 | 3222 | * @path: path to report |
a03a8a70 JB |
3223 | * @buf: buffer to return value in |
3224 | * @buflen: buffer length | |
3225 | * | |
3226 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3227 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3228 | * | |
52afeefb AV |
3229 | * Returns a pointer into the buffer or an error code if the path was |
3230 | * too long. Note: Callers should use the returned pointer, not the passed | |
3231 | * in buffer, to use the name! The implementation often starts at an offset | |
3232 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3233 | * |
31f3e0b3 | 3234 | * "buflen" should be positive. |
a03a8a70 | 3235 | */ |
20d4fdc1 | 3236 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3237 | { |
ffd1f4ed | 3238 | char *res = buf + buflen; |
6ac08c39 | 3239 | struct path root; |
ffd1f4ed | 3240 | int error; |
1da177e4 | 3241 | |
c23fbb6b ED |
3242 | /* |
3243 | * We have various synthetic filesystems that never get mounted. On | |
3244 | * these filesystems dentries are never used for lookup purposes, and | |
3245 | * thus don't need to be hashed. They also don't need a name until a | |
3246 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3247 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3248 | * |
3249 | * Some pseudo inodes are mountable. When they are mounted | |
3250 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3251 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3252 | */ |
f48cfddc EB |
3253 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3254 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3255 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3256 | |
68f0d9d9 LT |
3257 | rcu_read_lock(); |
3258 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3259 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3260 | rcu_read_unlock(); |
3261 | ||
02125a82 | 3262 | if (error < 0) |
ffd1f4ed | 3263 | res = ERR_PTR(error); |
1da177e4 LT |
3264 | return res; |
3265 | } | |
ec4f8605 | 3266 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3267 | |
c23fbb6b ED |
3268 | /* |
3269 | * Helper function for dentry_operations.d_dname() members | |
3270 | */ | |
3271 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3272 | const char *fmt, ...) | |
3273 | { | |
3274 | va_list args; | |
3275 | char temp[64]; | |
3276 | int sz; | |
3277 | ||
3278 | va_start(args, fmt); | |
3279 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3280 | va_end(args); | |
3281 | ||
3282 | if (sz > sizeof(temp) || sz > buflen) | |
3283 | return ERR_PTR(-ENAMETOOLONG); | |
3284 | ||
3285 | buffer += buflen - sz; | |
3286 | return memcpy(buffer, temp, sz); | |
3287 | } | |
3288 | ||
118b2302 AV |
3289 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3290 | { | |
3291 | char *end = buffer + buflen; | |
3292 | /* these dentries are never renamed, so d_lock is not needed */ | |
3293 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3294 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3295 | prepend(&end, &buflen, "/", 1)) |
3296 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3297 | return end; |
118b2302 | 3298 | } |
31bbe16f | 3299 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3300 | |
6092d048 RP |
3301 | /* |
3302 | * Write full pathname from the root of the filesystem into the buffer. | |
3303 | */ | |
f6500801 | 3304 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3305 | { |
f6500801 | 3306 | struct dentry *dentry; |
232d2d60 WL |
3307 | char *end, *retval; |
3308 | int len, seq = 0; | |
3309 | int error = 0; | |
6092d048 | 3310 | |
f6500801 AV |
3311 | if (buflen < 2) |
3312 | goto Elong; | |
3313 | ||
48f5ec21 | 3314 | rcu_read_lock(); |
232d2d60 | 3315 | restart: |
f6500801 | 3316 | dentry = d; |
232d2d60 WL |
3317 | end = buf + buflen; |
3318 | len = buflen; | |
3319 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3320 | /* Get '/' right */ |
3321 | retval = end-1; | |
3322 | *retval = '/'; | |
232d2d60 | 3323 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3324 | while (!IS_ROOT(dentry)) { |
3325 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3326 | |
6092d048 | 3327 | prefetch(parent); |
232d2d60 WL |
3328 | error = prepend_name(&end, &len, &dentry->d_name); |
3329 | if (error) | |
3330 | break; | |
6092d048 RP |
3331 | |
3332 | retval = end; | |
3333 | dentry = parent; | |
3334 | } | |
48f5ec21 AV |
3335 | if (!(seq & 1)) |
3336 | rcu_read_unlock(); | |
3337 | if (need_seqretry(&rename_lock, seq)) { | |
3338 | seq = 1; | |
232d2d60 | 3339 | goto restart; |
48f5ec21 AV |
3340 | } |
3341 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3342 | if (error) |
3343 | goto Elong; | |
c103135c AV |
3344 | return retval; |
3345 | Elong: | |
3346 | return ERR_PTR(-ENAMETOOLONG); | |
3347 | } | |
ec2447c2 NP |
3348 | |
3349 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3350 | { | |
232d2d60 | 3351 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3352 | } |
3353 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3354 | |
3355 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3356 | { | |
3357 | char *p = NULL; | |
3358 | char *retval; | |
3359 | ||
c103135c AV |
3360 | if (d_unlinked(dentry)) { |
3361 | p = buf + buflen; | |
3362 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3363 | goto Elong; | |
3364 | buflen++; | |
3365 | } | |
3366 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3367 | if (!IS_ERR(retval) && p) |
3368 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3369 | return retval; |
3370 | Elong: | |
6092d048 RP |
3371 | return ERR_PTR(-ENAMETOOLONG); |
3372 | } | |
3373 | ||
8b19e341 LT |
3374 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3375 | struct path *pwd) | |
5762482f | 3376 | { |
8b19e341 LT |
3377 | unsigned seq; |
3378 | ||
3379 | do { | |
3380 | seq = read_seqcount_begin(&fs->seq); | |
3381 | *root = fs->root; | |
3382 | *pwd = fs->pwd; | |
3383 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3384 | } |
3385 | ||
1da177e4 LT |
3386 | /* |
3387 | * NOTE! The user-level library version returns a | |
3388 | * character pointer. The kernel system call just | |
3389 | * returns the length of the buffer filled (which | |
3390 | * includes the ending '\0' character), or a negative | |
3391 | * error value. So libc would do something like | |
3392 | * | |
3393 | * char *getcwd(char * buf, size_t size) | |
3394 | * { | |
3395 | * int retval; | |
3396 | * | |
3397 | * retval = sys_getcwd(buf, size); | |
3398 | * if (retval >= 0) | |
3399 | * return buf; | |
3400 | * errno = -retval; | |
3401 | * return NULL; | |
3402 | * } | |
3403 | */ | |
3cdad428 | 3404 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3405 | { |
552ce544 | 3406 | int error; |
6ac08c39 | 3407 | struct path pwd, root; |
3272c544 | 3408 | char *page = __getname(); |
1da177e4 LT |
3409 | |
3410 | if (!page) | |
3411 | return -ENOMEM; | |
3412 | ||
8b19e341 LT |
3413 | rcu_read_lock(); |
3414 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3415 | |
552ce544 | 3416 | error = -ENOENT; |
f3da392e | 3417 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3418 | unsigned long len; |
3272c544 LT |
3419 | char *cwd = page + PATH_MAX; |
3420 | int buflen = PATH_MAX; | |
1da177e4 | 3421 | |
8df9d1a4 | 3422 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3423 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3424 | rcu_read_unlock(); |
552ce544 | 3425 | |
02125a82 | 3426 | if (error < 0) |
552ce544 LT |
3427 | goto out; |
3428 | ||
8df9d1a4 | 3429 | /* Unreachable from current root */ |
02125a82 | 3430 | if (error > 0) { |
8df9d1a4 MS |
3431 | error = prepend_unreachable(&cwd, &buflen); |
3432 | if (error) | |
3433 | goto out; | |
3434 | } | |
3435 | ||
552ce544 | 3436 | error = -ERANGE; |
3272c544 | 3437 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3438 | if (len <= size) { |
3439 | error = len; | |
3440 | if (copy_to_user(buf, cwd, len)) | |
3441 | error = -EFAULT; | |
3442 | } | |
949854d0 | 3443 | } else { |
ff812d72 | 3444 | rcu_read_unlock(); |
949854d0 | 3445 | } |
1da177e4 LT |
3446 | |
3447 | out: | |
3272c544 | 3448 | __putname(page); |
1da177e4 LT |
3449 | return error; |
3450 | } | |
3451 | ||
3452 | /* | |
3453 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3454 | * | |
3455 | * Trivially implemented using the dcache structure | |
3456 | */ | |
3457 | ||
3458 | /** | |
3459 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3460 | * @new_dentry: new dentry | |
3461 | * @old_dentry: old dentry | |
3462 | * | |
a6e5787f YB |
3463 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3464 | * Returns false otherwise. | |
1da177e4 LT |
3465 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3466 | */ | |
3467 | ||
a6e5787f | 3468 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3469 | { |
a6e5787f | 3470 | bool result; |
949854d0 | 3471 | unsigned seq; |
1da177e4 | 3472 | |
e2761a11 | 3473 | if (new_dentry == old_dentry) |
a6e5787f | 3474 | return true; |
e2761a11 | 3475 | |
e2761a11 | 3476 | do { |
1da177e4 | 3477 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3478 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3479 | /* |
3480 | * Need rcu_readlock to protect against the d_parent trashing | |
3481 | * due to d_move | |
3482 | */ | |
3483 | rcu_read_lock(); | |
e2761a11 | 3484 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3485 | result = true; |
e2761a11 | 3486 | else |
a6e5787f | 3487 | result = false; |
949854d0 | 3488 | rcu_read_unlock(); |
1da177e4 | 3489 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3490 | |
3491 | return result; | |
3492 | } | |
3493 | ||
db14fc3a | 3494 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3495 | { |
db14fc3a MS |
3496 | struct dentry *root = data; |
3497 | if (dentry != root) { | |
3498 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3499 | return D_WALK_SKIP; | |
1da177e4 | 3500 | |
01ddc4ed MS |
3501 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3502 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3503 | dentry->d_lockref.count--; | |
3504 | } | |
1da177e4 | 3505 | } |
db14fc3a MS |
3506 | return D_WALK_CONTINUE; |
3507 | } | |
58db63d0 | 3508 | |
db14fc3a MS |
3509 | void d_genocide(struct dentry *parent) |
3510 | { | |
3511 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3512 | } |
3513 | ||
60545d0d | 3514 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3515 | { |
60545d0d AV |
3516 | inode_dec_link_count(inode); |
3517 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3518 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3519 | !d_unlinked(dentry)); |
3520 | spin_lock(&dentry->d_parent->d_lock); | |
3521 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3522 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3523 | (unsigned long long)inode->i_ino); | |
3524 | spin_unlock(&dentry->d_lock); | |
3525 | spin_unlock(&dentry->d_parent->d_lock); | |
3526 | d_instantiate(dentry, inode); | |
1da177e4 | 3527 | } |
60545d0d | 3528 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3529 | |
3530 | static __initdata unsigned long dhash_entries; | |
3531 | static int __init set_dhash_entries(char *str) | |
3532 | { | |
3533 | if (!str) | |
3534 | return 0; | |
3535 | dhash_entries = simple_strtoul(str, &str, 0); | |
3536 | return 1; | |
3537 | } | |
3538 | __setup("dhash_entries=", set_dhash_entries); | |
3539 | ||
3540 | static void __init dcache_init_early(void) | |
3541 | { | |
074b8517 | 3542 | unsigned int loop; |
1da177e4 LT |
3543 | |
3544 | /* If hashes are distributed across NUMA nodes, defer | |
3545 | * hash allocation until vmalloc space is available. | |
3546 | */ | |
3547 | if (hashdist) | |
3548 | return; | |
3549 | ||
3550 | dentry_hashtable = | |
3551 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3552 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3553 | dhash_entries, |
3554 | 13, | |
3555 | HASH_EARLY, | |
3556 | &d_hash_shift, | |
3557 | &d_hash_mask, | |
31fe62b9 | 3558 | 0, |
1da177e4 LT |
3559 | 0); |
3560 | ||
074b8517 | 3561 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3562 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3563 | } |
3564 | ||
74bf17cf | 3565 | static void __init dcache_init(void) |
1da177e4 | 3566 | { |
074b8517 | 3567 | unsigned int loop; |
1da177e4 LT |
3568 | |
3569 | /* | |
3570 | * A constructor could be added for stable state like the lists, | |
3571 | * but it is probably not worth it because of the cache nature | |
3572 | * of the dcache. | |
3573 | */ | |
0a31bd5f | 3574 | dentry_cache = KMEM_CACHE(dentry, |
5d097056 | 3575 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT); |
1da177e4 LT |
3576 | |
3577 | /* Hash may have been set up in dcache_init_early */ | |
3578 | if (!hashdist) | |
3579 | return; | |
3580 | ||
3581 | dentry_hashtable = | |
3582 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3583 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3584 | dhash_entries, |
3585 | 13, | |
3586 | 0, | |
3587 | &d_hash_shift, | |
3588 | &d_hash_mask, | |
31fe62b9 | 3589 | 0, |
1da177e4 LT |
3590 | 0); |
3591 | ||
074b8517 | 3592 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3593 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3594 | } |
3595 | ||
3596 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3597 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3598 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3599 | |
1da177e4 LT |
3600 | EXPORT_SYMBOL(d_genocide); |
3601 | ||
1da177e4 LT |
3602 | void __init vfs_caches_init_early(void) |
3603 | { | |
3604 | dcache_init_early(); | |
3605 | inode_init_early(); | |
3606 | } | |
3607 | ||
4248b0da | 3608 | void __init vfs_caches_init(void) |
1da177e4 | 3609 | { |
1da177e4 | 3610 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, |
20c2df83 | 3611 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3612 | |
74bf17cf DC |
3613 | dcache_init(); |
3614 | inode_init(); | |
4248b0da MG |
3615 | files_init(); |
3616 | files_maxfiles_init(); | |
74bf17cf | 3617 | mnt_init(); |
1da177e4 LT |
3618 | bdev_cache_init(); |
3619 | chrdev_init(); | |
3620 | } |