]>
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 | ||
31e6b01f | 319 | /** |
a7c6f571 | 320 | * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups |
ff5fdb61 | 321 | * @dentry: the target dentry |
31e6b01f NP |
322 | * After this call, in-progress rcu-walk path lookup will fail. This |
323 | * should be called after unhashing, and after changing d_inode (if | |
324 | * the dentry has not already been unhashed). | |
325 | */ | |
a7c6f571 | 326 | static inline void dentry_rcuwalk_invalidate(struct dentry *dentry) |
31e6b01f | 327 | { |
a7c6f571 PZ |
328 | lockdep_assert_held(&dentry->d_lock); |
329 | /* Go through am invalidation barrier */ | |
330 | write_seqcount_invalidate(&dentry->d_seq); | |
31e6b01f NP |
331 | } |
332 | ||
1da177e4 LT |
333 | /* |
334 | * Release the dentry's inode, using the filesystem | |
550dce01 | 335 | * d_iput() operation if defined. |
31e6b01f NP |
336 | */ |
337 | static void dentry_unlink_inode(struct dentry * dentry) | |
338 | __releases(dentry->d_lock) | |
873feea0 | 339 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
340 | { |
341 | struct inode *inode = dentry->d_inode; | |
550dce01 | 342 | bool hashed = !d_unhashed(dentry); |
a528aca7 | 343 | |
550dce01 AV |
344 | if (hashed) |
345 | raw_write_seqcount_begin(&dentry->d_seq); | |
4bf46a27 | 346 | __d_clear_type_and_inode(dentry); |
946e51f2 | 347 | hlist_del_init(&dentry->d_u.d_alias); |
550dce01 AV |
348 | if (hashed) |
349 | raw_write_seqcount_end(&dentry->d_seq); | |
31e6b01f | 350 | spin_unlock(&dentry->d_lock); |
873feea0 | 351 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
352 | if (!inode->i_nlink) |
353 | fsnotify_inoderemove(inode); | |
354 | if (dentry->d_op && dentry->d_op->d_iput) | |
355 | dentry->d_op->d_iput(dentry, inode); | |
356 | else | |
357 | iput(inode); | |
358 | } | |
359 | ||
89dc77bc LT |
360 | /* |
361 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
362 | * is in use - which includes both the "real" per-superblock | |
363 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
364 | * | |
365 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
366 | * on the shrink list (ie not on the superblock LRU list). | |
367 | * | |
368 | * The per-cpu "nr_dentry_unused" counters are updated with | |
369 | * the DCACHE_LRU_LIST bit. | |
370 | * | |
371 | * These helper functions make sure we always follow the | |
372 | * rules. d_lock must be held by the caller. | |
373 | */ | |
374 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
375 | static void d_lru_add(struct dentry *dentry) | |
376 | { | |
377 | D_FLAG_VERIFY(dentry, 0); | |
378 | dentry->d_flags |= DCACHE_LRU_LIST; | |
379 | this_cpu_inc(nr_dentry_unused); | |
380 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
381 | } | |
382 | ||
383 | static void d_lru_del(struct dentry *dentry) | |
384 | { | |
385 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
386 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
387 | this_cpu_dec(nr_dentry_unused); | |
388 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
389 | } | |
390 | ||
391 | static void d_shrink_del(struct dentry *dentry) | |
392 | { | |
393 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
394 | list_del_init(&dentry->d_lru); | |
395 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
396 | this_cpu_dec(nr_dentry_unused); | |
397 | } | |
398 | ||
399 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
400 | { | |
401 | D_FLAG_VERIFY(dentry, 0); | |
402 | list_add(&dentry->d_lru, list); | |
403 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
404 | this_cpu_inc(nr_dentry_unused); | |
405 | } | |
406 | ||
407 | /* | |
408 | * These can only be called under the global LRU lock, ie during the | |
409 | * callback for freeing the LRU list. "isolate" removes it from the | |
410 | * LRU lists entirely, while shrink_move moves it to the indicated | |
411 | * private list. | |
412 | */ | |
3f97b163 | 413 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
414 | { |
415 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
416 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
417 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 418 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
419 | } |
420 | ||
3f97b163 VD |
421 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
422 | struct list_head *list) | |
89dc77bc LT |
423 | { |
424 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
425 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 426 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
427 | } |
428 | ||
da3bbdd4 | 429 | /* |
f6041567 | 430 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
431 | */ |
432 | static void dentry_lru_add(struct dentry *dentry) | |
433 | { | |
89dc77bc LT |
434 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
435 | d_lru_add(dentry); | |
da3bbdd4 KM |
436 | } |
437 | ||
789680d1 NP |
438 | /** |
439 | * d_drop - drop a dentry | |
440 | * @dentry: dentry to drop | |
441 | * | |
442 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
443 | * be found through a VFS lookup any more. Note that this is different from | |
444 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
445 | * possible, giving a successful _negative_ lookup, while d_drop will | |
446 | * just make the cache lookup fail. | |
447 | * | |
448 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
449 | * reason (NFS timeouts or autofs deletes). | |
450 | * | |
451 | * __d_drop requires dentry->d_lock. | |
452 | */ | |
453 | void __d_drop(struct dentry *dentry) | |
454 | { | |
dea3667b | 455 | if (!d_unhashed(dentry)) { |
b61625d2 | 456 | struct hlist_bl_head *b; |
7632e465 BF |
457 | /* |
458 | * Hashed dentries are normally on the dentry hashtable, | |
459 | * with the exception of those newly allocated by | |
460 | * d_obtain_alias, which are always IS_ROOT: | |
461 | */ | |
462 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
463 | b = &dentry->d_sb->s_anon; |
464 | else | |
8387ff25 | 465 | b = d_hash(dentry->d_name.hash); |
b61625d2 AV |
466 | |
467 | hlist_bl_lock(b); | |
468 | __hlist_bl_del(&dentry->d_hash); | |
469 | dentry->d_hash.pprev = NULL; | |
470 | hlist_bl_unlock(b); | |
a7c6f571 | 471 | dentry_rcuwalk_invalidate(dentry); |
789680d1 NP |
472 | } |
473 | } | |
474 | EXPORT_SYMBOL(__d_drop); | |
475 | ||
476 | void d_drop(struct dentry *dentry) | |
477 | { | |
789680d1 NP |
478 | spin_lock(&dentry->d_lock); |
479 | __d_drop(dentry); | |
480 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
481 | } |
482 | EXPORT_SYMBOL(d_drop); | |
483 | ||
ba65dc5e AV |
484 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
485 | { | |
486 | struct dentry *next; | |
487 | /* | |
488 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
489 | * attached to the dentry tree | |
490 | */ | |
491 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
492 | if (unlikely(list_empty(&dentry->d_child))) | |
493 | return; | |
494 | __list_del_entry(&dentry->d_child); | |
495 | /* | |
496 | * Cursors can move around the list of children. While we'd been | |
497 | * a normal list member, it didn't matter - ->d_child.next would've | |
498 | * been updated. However, from now on it won't be and for the | |
499 | * things like d_walk() it might end up with a nasty surprise. | |
500 | * Normally d_walk() doesn't care about cursors moving around - | |
501 | * ->d_lock on parent prevents that and since a cursor has no children | |
502 | * of its own, we get through it without ever unlocking the parent. | |
503 | * There is one exception, though - if we ascend from a child that | |
504 | * gets killed as soon as we unlock it, the next sibling is found | |
505 | * using the value left in its ->d_child.next. And if _that_ | |
506 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
507 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
508 | * everything the cursor had been moved past. | |
509 | * | |
510 | * Solution: make sure that the pointer left behind in ->d_child.next | |
511 | * points to something that won't be moving around. I.e. skip the | |
512 | * cursors. | |
513 | */ | |
514 | while (dentry->d_child.next != &parent->d_subdirs) { | |
515 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
516 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
517 | break; | |
518 | dentry->d_child.next = next->d_child.next; | |
519 | } | |
520 | } | |
521 | ||
e55fd011 | 522 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 523 | { |
41edf278 AV |
524 | struct dentry *parent = NULL; |
525 | bool can_free = true; | |
41edf278 | 526 | if (!IS_ROOT(dentry)) |
77812a1e | 527 | parent = dentry->d_parent; |
31e6b01f | 528 | |
0d98439e LT |
529 | /* |
530 | * The dentry is now unrecoverably dead to the world. | |
531 | */ | |
532 | lockref_mark_dead(&dentry->d_lockref); | |
533 | ||
f0023bc6 | 534 | /* |
f0023bc6 SW |
535 | * inform the fs via d_prune that this dentry is about to be |
536 | * unhashed and destroyed. | |
537 | */ | |
29266201 | 538 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
539 | dentry->d_op->d_prune(dentry); |
540 | ||
01b60351 AV |
541 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
542 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
543 | d_lru_del(dentry); | |
01b60351 | 544 | } |
77812a1e NP |
545 | /* if it was on the hash then remove it */ |
546 | __d_drop(dentry); | |
ba65dc5e | 547 | dentry_unlist(dentry, parent); |
03b3b889 AV |
548 | if (parent) |
549 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
550 | if (dentry->d_inode) |
551 | dentry_unlink_inode(dentry); | |
552 | else | |
553 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
554 | this_cpu_dec(nr_dentry); |
555 | if (dentry->d_op && dentry->d_op->d_release) | |
556 | dentry->d_op->d_release(dentry); | |
557 | ||
41edf278 AV |
558 | spin_lock(&dentry->d_lock); |
559 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
560 | dentry->d_flags |= DCACHE_MAY_FREE; | |
561 | can_free = false; | |
562 | } | |
563 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
564 | if (likely(can_free)) |
565 | dentry_free(dentry); | |
e55fd011 AV |
566 | } |
567 | ||
568 | /* | |
569 | * Finish off a dentry we've decided to kill. | |
570 | * dentry->d_lock must be held, returns with it unlocked. | |
571 | * If ref is non-zero, then decrement the refcount too. | |
572 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
573 | */ | |
8cbf74da | 574 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
575 | __releases(dentry->d_lock) |
576 | { | |
577 | struct inode *inode = dentry->d_inode; | |
578 | struct dentry *parent = NULL; | |
579 | ||
580 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
581 | goto failed; | |
582 | ||
583 | if (!IS_ROOT(dentry)) { | |
584 | parent = dentry->d_parent; | |
585 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
586 | if (inode) | |
587 | spin_unlock(&inode->i_lock); | |
588 | goto failed; | |
589 | } | |
590 | } | |
591 | ||
592 | __dentry_kill(dentry); | |
03b3b889 | 593 | return parent; |
e55fd011 AV |
594 | |
595 | failed: | |
8cbf74da | 596 | spin_unlock(&dentry->d_lock); |
e55fd011 | 597 | return dentry; /* try again with same dentry */ |
77812a1e NP |
598 | } |
599 | ||
046b961b AV |
600 | static inline struct dentry *lock_parent(struct dentry *dentry) |
601 | { | |
602 | struct dentry *parent = dentry->d_parent; | |
603 | if (IS_ROOT(dentry)) | |
604 | return NULL; | |
360f5479 | 605 | if (unlikely(dentry->d_lockref.count < 0)) |
c2338f2d | 606 | return NULL; |
046b961b AV |
607 | if (likely(spin_trylock(&parent->d_lock))) |
608 | return parent; | |
046b961b | 609 | rcu_read_lock(); |
c2338f2d | 610 | spin_unlock(&dentry->d_lock); |
046b961b AV |
611 | again: |
612 | parent = ACCESS_ONCE(dentry->d_parent); | |
613 | spin_lock(&parent->d_lock); | |
614 | /* | |
615 | * We can't blindly lock dentry until we are sure | |
616 | * that we won't violate the locking order. | |
617 | * Any changes of dentry->d_parent must have | |
618 | * been done with parent->d_lock held, so | |
619 | * spin_lock() above is enough of a barrier | |
620 | * for checking if it's still our child. | |
621 | */ | |
622 | if (unlikely(parent != dentry->d_parent)) { | |
623 | spin_unlock(&parent->d_lock); | |
624 | goto again; | |
625 | } | |
626 | rcu_read_unlock(); | |
627 | if (parent != dentry) | |
9f12600f | 628 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
046b961b AV |
629 | else |
630 | parent = NULL; | |
631 | return parent; | |
632 | } | |
633 | ||
360f5479 LT |
634 | /* |
635 | * Try to do a lockless dput(), and return whether that was successful. | |
636 | * | |
637 | * If unsuccessful, we return false, having already taken the dentry lock. | |
638 | * | |
639 | * The caller needs to hold the RCU read lock, so that the dentry is | |
640 | * guaranteed to stay around even if the refcount goes down to zero! | |
641 | */ | |
642 | static inline bool fast_dput(struct dentry *dentry) | |
643 | { | |
644 | int ret; | |
645 | unsigned int d_flags; | |
646 | ||
647 | /* | |
648 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 649 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
650 | */ |
651 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
652 | return lockref_put_or_lock(&dentry->d_lockref); | |
653 | ||
654 | /* | |
655 | * .. otherwise, we can try to just decrement the | |
656 | * lockref optimistically. | |
657 | */ | |
658 | ret = lockref_put_return(&dentry->d_lockref); | |
659 | ||
660 | /* | |
661 | * If the lockref_put_return() failed due to the lock being held | |
662 | * by somebody else, the fast path has failed. We will need to | |
663 | * get the lock, and then check the count again. | |
664 | */ | |
665 | if (unlikely(ret < 0)) { | |
666 | spin_lock(&dentry->d_lock); | |
667 | if (dentry->d_lockref.count > 1) { | |
668 | dentry->d_lockref.count--; | |
669 | spin_unlock(&dentry->d_lock); | |
670 | return 1; | |
671 | } | |
672 | return 0; | |
673 | } | |
674 | ||
675 | /* | |
676 | * If we weren't the last ref, we're done. | |
677 | */ | |
678 | if (ret) | |
679 | return 1; | |
680 | ||
681 | /* | |
682 | * Careful, careful. The reference count went down | |
683 | * to zero, but we don't hold the dentry lock, so | |
684 | * somebody else could get it again, and do another | |
685 | * dput(), and we need to not race with that. | |
686 | * | |
687 | * However, there is a very special and common case | |
688 | * where we don't care, because there is nothing to | |
689 | * do: the dentry is still hashed, it does not have | |
690 | * a 'delete' op, and it's referenced and already on | |
691 | * the LRU list. | |
692 | * | |
693 | * NOTE! Since we aren't locked, these values are | |
694 | * not "stable". However, it is sufficient that at | |
695 | * some point after we dropped the reference the | |
696 | * dentry was hashed and the flags had the proper | |
697 | * value. Other dentry users may have re-gotten | |
698 | * a reference to the dentry and change that, but | |
699 | * our work is done - we can leave the dentry | |
700 | * around with a zero refcount. | |
701 | */ | |
702 | smp_rmb(); | |
703 | d_flags = ACCESS_ONCE(dentry->d_flags); | |
75a6f82a | 704 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
705 | |
706 | /* Nothing to do? Dropping the reference was all we needed? */ | |
707 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
708 | return 1; | |
709 | ||
710 | /* | |
711 | * Not the fast normal case? Get the lock. We've already decremented | |
712 | * the refcount, but we'll need to re-check the situation after | |
713 | * getting the lock. | |
714 | */ | |
715 | spin_lock(&dentry->d_lock); | |
716 | ||
717 | /* | |
718 | * Did somebody else grab a reference to it in the meantime, and | |
719 | * we're no longer the last user after all? Alternatively, somebody | |
720 | * else could have killed it and marked it dead. Either way, we | |
721 | * don't need to do anything else. | |
722 | */ | |
723 | if (dentry->d_lockref.count) { | |
724 | spin_unlock(&dentry->d_lock); | |
725 | return 1; | |
726 | } | |
727 | ||
728 | /* | |
729 | * Re-get the reference we optimistically dropped. We hold the | |
730 | * lock, and we just tested that it was zero, so we can just | |
731 | * set it to 1. | |
732 | */ | |
733 | dentry->d_lockref.count = 1; | |
734 | return 0; | |
735 | } | |
736 | ||
737 | ||
1da177e4 LT |
738 | /* |
739 | * This is dput | |
740 | * | |
741 | * This is complicated by the fact that we do not want to put | |
742 | * dentries that are no longer on any hash chain on the unused | |
743 | * list: we'd much rather just get rid of them immediately. | |
744 | * | |
745 | * However, that implies that we have to traverse the dentry | |
746 | * tree upwards to the parents which might _also_ now be | |
747 | * scheduled for deletion (it may have been only waiting for | |
748 | * its last child to go away). | |
749 | * | |
750 | * This tail recursion is done by hand as we don't want to depend | |
751 | * on the compiler to always get this right (gcc generally doesn't). | |
752 | * Real recursion would eat up our stack space. | |
753 | */ | |
754 | ||
755 | /* | |
756 | * dput - release a dentry | |
757 | * @dentry: dentry to release | |
758 | * | |
759 | * Release a dentry. This will drop the usage count and if appropriate | |
760 | * call the dentry unlink method as well as removing it from the queues and | |
761 | * releasing its resources. If the parent dentries were scheduled for release | |
762 | * they too may now get deleted. | |
1da177e4 | 763 | */ |
1da177e4 LT |
764 | void dput(struct dentry *dentry) |
765 | { | |
8aab6a27 | 766 | if (unlikely(!dentry)) |
1da177e4 LT |
767 | return; |
768 | ||
769 | repeat: | |
47be6184 WF |
770 | might_sleep(); |
771 | ||
360f5479 LT |
772 | rcu_read_lock(); |
773 | if (likely(fast_dput(dentry))) { | |
774 | rcu_read_unlock(); | |
1da177e4 | 775 | return; |
360f5479 LT |
776 | } |
777 | ||
778 | /* Slow case: now with the dentry lock held */ | |
779 | rcu_read_unlock(); | |
1da177e4 | 780 | |
85c7f810 AV |
781 | WARN_ON(d_in_lookup(dentry)); |
782 | ||
8aab6a27 LT |
783 | /* Unreachable? Get rid of it */ |
784 | if (unlikely(d_unhashed(dentry))) | |
785 | goto kill_it; | |
786 | ||
75a6f82a AV |
787 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
788 | goto kill_it; | |
789 | ||
8aab6a27 | 790 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { |
1da177e4 | 791 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 792 | goto kill_it; |
1da177e4 | 793 | } |
265ac902 | 794 | |
358eec18 LT |
795 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
796 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 797 | dentry_lru_add(dentry); |
265ac902 | 798 | |
98474236 | 799 | dentry->d_lockref.count--; |
61f3dee4 | 800 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
801 | return; |
802 | ||
d52b9086 | 803 | kill_it: |
8cbf74da | 804 | dentry = dentry_kill(dentry); |
47be6184 WF |
805 | if (dentry) { |
806 | cond_resched(); | |
d52b9086 | 807 | goto repeat; |
47be6184 | 808 | } |
1da177e4 | 809 | } |
ec4f8605 | 810 | EXPORT_SYMBOL(dput); |
1da177e4 | 811 | |
1da177e4 | 812 | |
b5c84bf6 | 813 | /* This must be called with d_lock held */ |
dc0474be | 814 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 815 | { |
98474236 | 816 | dentry->d_lockref.count++; |
23044507 NP |
817 | } |
818 | ||
dc0474be | 819 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 820 | { |
98474236 | 821 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
822 | } |
823 | ||
b7ab39f6 NP |
824 | struct dentry *dget_parent(struct dentry *dentry) |
825 | { | |
df3d0bbc | 826 | int gotref; |
b7ab39f6 NP |
827 | struct dentry *ret; |
828 | ||
df3d0bbc WL |
829 | /* |
830 | * Do optimistic parent lookup without any | |
831 | * locking. | |
832 | */ | |
833 | rcu_read_lock(); | |
834 | ret = ACCESS_ONCE(dentry->d_parent); | |
835 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
836 | rcu_read_unlock(); | |
837 | if (likely(gotref)) { | |
838 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
839 | return ret; | |
840 | dput(ret); | |
841 | } | |
842 | ||
b7ab39f6 | 843 | repeat: |
a734eb45 NP |
844 | /* |
845 | * Don't need rcu_dereference because we re-check it was correct under | |
846 | * the lock. | |
847 | */ | |
848 | rcu_read_lock(); | |
b7ab39f6 | 849 | ret = dentry->d_parent; |
a734eb45 NP |
850 | spin_lock(&ret->d_lock); |
851 | if (unlikely(ret != dentry->d_parent)) { | |
852 | spin_unlock(&ret->d_lock); | |
853 | rcu_read_unlock(); | |
b7ab39f6 NP |
854 | goto repeat; |
855 | } | |
a734eb45 | 856 | rcu_read_unlock(); |
98474236 WL |
857 | BUG_ON(!ret->d_lockref.count); |
858 | ret->d_lockref.count++; | |
b7ab39f6 | 859 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
860 | return ret; |
861 | } | |
862 | EXPORT_SYMBOL(dget_parent); | |
863 | ||
1da177e4 LT |
864 | /** |
865 | * d_find_alias - grab a hashed alias of inode | |
866 | * @inode: inode in question | |
1da177e4 LT |
867 | * |
868 | * If inode has a hashed alias, or is a directory and has any alias, | |
869 | * acquire the reference to alias and return it. Otherwise return NULL. | |
870 | * Notice that if inode is a directory there can be only one alias and | |
871 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
872 | * of a filesystem, or if the directory was renamed and d_revalidate |
873 | * was the first vfs operation to notice. | |
1da177e4 | 874 | * |
21c0d8fd | 875 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 876 | * any other hashed alias over that one. |
1da177e4 | 877 | */ |
52ed46f0 | 878 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 879 | { |
da502956 | 880 | struct dentry *alias, *discon_alias; |
1da177e4 | 881 | |
da502956 NP |
882 | again: |
883 | discon_alias = NULL; | |
946e51f2 | 884 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 885 | spin_lock(&alias->d_lock); |
1da177e4 | 886 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 887 | if (IS_ROOT(alias) && |
da502956 | 888 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 889 | discon_alias = alias; |
52ed46f0 | 890 | } else { |
dc0474be | 891 | __dget_dlock(alias); |
da502956 NP |
892 | spin_unlock(&alias->d_lock); |
893 | return alias; | |
894 | } | |
895 | } | |
896 | spin_unlock(&alias->d_lock); | |
897 | } | |
898 | if (discon_alias) { | |
899 | alias = discon_alias; | |
900 | spin_lock(&alias->d_lock); | |
901 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
902 | __dget_dlock(alias); |
903 | spin_unlock(&alias->d_lock); | |
904 | return alias; | |
1da177e4 | 905 | } |
da502956 NP |
906 | spin_unlock(&alias->d_lock); |
907 | goto again; | |
1da177e4 | 908 | } |
da502956 | 909 | return NULL; |
1da177e4 LT |
910 | } |
911 | ||
da502956 | 912 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 913 | { |
214fda1f DH |
914 | struct dentry *de = NULL; |
915 | ||
b3d9b7a3 | 916 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 917 | spin_lock(&inode->i_lock); |
52ed46f0 | 918 | de = __d_find_alias(inode); |
873feea0 | 919 | spin_unlock(&inode->i_lock); |
214fda1f | 920 | } |
1da177e4 LT |
921 | return de; |
922 | } | |
ec4f8605 | 923 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
924 | |
925 | /* | |
926 | * Try to kill dentries associated with this inode. | |
927 | * WARNING: you must own a reference to inode. | |
928 | */ | |
929 | void d_prune_aliases(struct inode *inode) | |
930 | { | |
0cdca3f9 | 931 | struct dentry *dentry; |
1da177e4 | 932 | restart: |
873feea0 | 933 | spin_lock(&inode->i_lock); |
946e51f2 | 934 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 935 | spin_lock(&dentry->d_lock); |
98474236 | 936 | if (!dentry->d_lockref.count) { |
29355c39 AV |
937 | struct dentry *parent = lock_parent(dentry); |
938 | if (likely(!dentry->d_lockref.count)) { | |
939 | __dentry_kill(dentry); | |
4a7795d3 | 940 | dput(parent); |
29355c39 AV |
941 | goto restart; |
942 | } | |
943 | if (parent) | |
944 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
945 | } |
946 | spin_unlock(&dentry->d_lock); | |
947 | } | |
873feea0 | 948 | spin_unlock(&inode->i_lock); |
1da177e4 | 949 | } |
ec4f8605 | 950 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 951 | |
3049cfe2 | 952 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 953 | { |
5c47e6d0 | 954 | struct dentry *dentry, *parent; |
da3bbdd4 | 955 | |
60942f2f | 956 | while (!list_empty(list)) { |
ff2fde99 | 957 | struct inode *inode; |
60942f2f | 958 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 959 | spin_lock(&dentry->d_lock); |
046b961b AV |
960 | parent = lock_parent(dentry); |
961 | ||
dd1f6b2e DC |
962 | /* |
963 | * The dispose list is isolated and dentries are not accounted | |
964 | * to the LRU here, so we can simply remove it from the list | |
965 | * here regardless of whether it is referenced or not. | |
966 | */ | |
89dc77bc | 967 | d_shrink_del(dentry); |
dd1f6b2e | 968 | |
1da177e4 LT |
969 | /* |
970 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 971 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 972 | */ |
360f5479 | 973 | if (dentry->d_lockref.count > 0) { |
da3bbdd4 | 974 | spin_unlock(&dentry->d_lock); |
046b961b AV |
975 | if (parent) |
976 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
977 | continue; |
978 | } | |
77812a1e | 979 | |
64fd72e0 AV |
980 | |
981 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
982 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
983 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
984 | if (parent) |
985 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
986 | if (can_free) |
987 | dentry_free(dentry); | |
988 | continue; | |
989 | } | |
990 | ||
ff2fde99 AV |
991 | inode = dentry->d_inode; |
992 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 993 | d_shrink_add(dentry, list); |
dd1f6b2e | 994 | spin_unlock(&dentry->d_lock); |
046b961b AV |
995 | if (parent) |
996 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 997 | continue; |
dd1f6b2e | 998 | } |
ff2fde99 | 999 | |
ff2fde99 | 1000 | __dentry_kill(dentry); |
046b961b | 1001 | |
5c47e6d0 AV |
1002 | /* |
1003 | * We need to prune ancestors too. This is necessary to prevent | |
1004 | * quadratic behavior of shrink_dcache_parent(), but is also | |
1005 | * expected to be beneficial in reducing dentry cache | |
1006 | * fragmentation. | |
1007 | */ | |
1008 | dentry = parent; | |
b2b80195 AV |
1009 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
1010 | parent = lock_parent(dentry); | |
1011 | if (dentry->d_lockref.count != 1) { | |
1012 | dentry->d_lockref.count--; | |
1013 | spin_unlock(&dentry->d_lock); | |
1014 | if (parent) | |
1015 | spin_unlock(&parent->d_lock); | |
1016 | break; | |
1017 | } | |
1018 | inode = dentry->d_inode; /* can't be NULL */ | |
1019 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
1020 | spin_unlock(&dentry->d_lock); | |
1021 | if (parent) | |
1022 | spin_unlock(&parent->d_lock); | |
1023 | cpu_relax(); | |
1024 | continue; | |
1025 | } | |
1026 | __dentry_kill(dentry); | |
1027 | dentry = parent; | |
1028 | } | |
da3bbdd4 | 1029 | } |
3049cfe2 CH |
1030 | } |
1031 | ||
3f97b163 VD |
1032 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1033 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1034 | { |
1035 | struct list_head *freeable = arg; | |
1036 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1037 | ||
1038 | ||
1039 | /* | |
1040 | * we are inverting the lru lock/dentry->d_lock here, | |
1041 | * so use a trylock. If we fail to get the lock, just skip | |
1042 | * it | |
1043 | */ | |
1044 | if (!spin_trylock(&dentry->d_lock)) | |
1045 | return LRU_SKIP; | |
1046 | ||
1047 | /* | |
1048 | * Referenced dentries are still in use. If they have active | |
1049 | * counts, just remove them from the LRU. Otherwise give them | |
1050 | * another pass through the LRU. | |
1051 | */ | |
1052 | if (dentry->d_lockref.count) { | |
3f97b163 | 1053 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1054 | spin_unlock(&dentry->d_lock); |
1055 | return LRU_REMOVED; | |
1056 | } | |
1057 | ||
1058 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1059 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1060 | spin_unlock(&dentry->d_lock); | |
1061 | ||
1062 | /* | |
1063 | * The list move itself will be made by the common LRU code. At | |
1064 | * this point, we've dropped the dentry->d_lock but keep the | |
1065 | * lru lock. This is safe to do, since every list movement is | |
1066 | * protected by the lru lock even if both locks are held. | |
1067 | * | |
1068 | * This is guaranteed by the fact that all LRU management | |
1069 | * functions are intermediated by the LRU API calls like | |
1070 | * list_lru_add and list_lru_del. List movement in this file | |
1071 | * only ever occur through this functions or through callbacks | |
1072 | * like this one, that are called from the LRU API. | |
1073 | * | |
1074 | * The only exceptions to this are functions like | |
1075 | * shrink_dentry_list, and code that first checks for the | |
1076 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1077 | * operating only with stack provided lists after they are | |
1078 | * properly isolated from the main list. It is thus, always a | |
1079 | * local access. | |
1080 | */ | |
1081 | return LRU_ROTATE; | |
1082 | } | |
1083 | ||
3f97b163 | 1084 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1085 | spin_unlock(&dentry->d_lock); |
1086 | ||
1087 | return LRU_REMOVED; | |
1088 | } | |
1089 | ||
3049cfe2 | 1090 | /** |
b48f03b3 DC |
1091 | * prune_dcache_sb - shrink the dcache |
1092 | * @sb: superblock | |
503c358c | 1093 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1094 | * |
503c358c VD |
1095 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1096 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1097 | * function. |
3049cfe2 | 1098 | * |
b48f03b3 DC |
1099 | * This function may fail to free any resources if all the dentries are in |
1100 | * use. | |
3049cfe2 | 1101 | */ |
503c358c | 1102 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1103 | { |
f6041567 DC |
1104 | LIST_HEAD(dispose); |
1105 | long freed; | |
3049cfe2 | 1106 | |
503c358c VD |
1107 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1108 | dentry_lru_isolate, &dispose); | |
f6041567 | 1109 | shrink_dentry_list(&dispose); |
0a234c6d | 1110 | return freed; |
da3bbdd4 | 1111 | } |
23044507 | 1112 | |
4e717f5c | 1113 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1114 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1115 | { |
4e717f5c GC |
1116 | struct list_head *freeable = arg; |
1117 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1118 | |
4e717f5c GC |
1119 | /* |
1120 | * we are inverting the lru lock/dentry->d_lock here, | |
1121 | * so use a trylock. If we fail to get the lock, just skip | |
1122 | * it | |
1123 | */ | |
1124 | if (!spin_trylock(&dentry->d_lock)) | |
1125 | return LRU_SKIP; | |
1126 | ||
3f97b163 | 1127 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1128 | spin_unlock(&dentry->d_lock); |
ec33679d | 1129 | |
4e717f5c | 1130 | return LRU_REMOVED; |
da3bbdd4 KM |
1131 | } |
1132 | ||
4e717f5c | 1133 | |
1da177e4 LT |
1134 | /** |
1135 | * shrink_dcache_sb - shrink dcache for a superblock | |
1136 | * @sb: superblock | |
1137 | * | |
3049cfe2 CH |
1138 | * Shrink the dcache for the specified super block. This is used to free |
1139 | * the dcache before unmounting a file system. | |
1da177e4 | 1140 | */ |
3049cfe2 | 1141 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1142 | { |
4e717f5c GC |
1143 | long freed; |
1144 | ||
1145 | do { | |
1146 | LIST_HEAD(dispose); | |
1147 | ||
1148 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1149 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1150 | |
4e717f5c GC |
1151 | this_cpu_sub(nr_dentry_unused, freed); |
1152 | shrink_dentry_list(&dispose); | |
1153 | } while (freed > 0); | |
1da177e4 | 1154 | } |
ec4f8605 | 1155 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1156 | |
db14fc3a MS |
1157 | /** |
1158 | * enum d_walk_ret - action to talke during tree walk | |
1159 | * @D_WALK_CONTINUE: contrinue walk | |
1160 | * @D_WALK_QUIT: quit walk | |
1161 | * @D_WALK_NORETRY: quit when retry is needed | |
1162 | * @D_WALK_SKIP: skip this dentry and its children | |
1163 | */ | |
1164 | enum d_walk_ret { | |
1165 | D_WALK_CONTINUE, | |
1166 | D_WALK_QUIT, | |
1167 | D_WALK_NORETRY, | |
1168 | D_WALK_SKIP, | |
1169 | }; | |
c826cb7d | 1170 | |
1da177e4 | 1171 | /** |
db14fc3a MS |
1172 | * d_walk - walk the dentry tree |
1173 | * @parent: start of walk | |
1174 | * @data: data passed to @enter() and @finish() | |
1175 | * @enter: callback when first entering the dentry | |
1176 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1177 | * |
db14fc3a | 1178 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1179 | */ |
db14fc3a MS |
1180 | static void d_walk(struct dentry *parent, void *data, |
1181 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1182 | void (*finish)(void *)) | |
1da177e4 | 1183 | { |
949854d0 | 1184 | struct dentry *this_parent; |
1da177e4 | 1185 | struct list_head *next; |
48f5ec21 | 1186 | unsigned seq = 0; |
db14fc3a MS |
1187 | enum d_walk_ret ret; |
1188 | bool retry = true; | |
949854d0 | 1189 | |
58db63d0 | 1190 | again: |
48f5ec21 | 1191 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1192 | this_parent = parent; |
2fd6b7f5 | 1193 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1194 | |
1195 | ret = enter(data, this_parent); | |
1196 | switch (ret) { | |
1197 | case D_WALK_CONTINUE: | |
1198 | break; | |
1199 | case D_WALK_QUIT: | |
1200 | case D_WALK_SKIP: | |
1201 | goto out_unlock; | |
1202 | case D_WALK_NORETRY: | |
1203 | retry = false; | |
1204 | break; | |
1205 | } | |
1da177e4 LT |
1206 | repeat: |
1207 | next = this_parent->d_subdirs.next; | |
1208 | resume: | |
1209 | while (next != &this_parent->d_subdirs) { | |
1210 | struct list_head *tmp = next; | |
946e51f2 | 1211 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1212 | next = tmp->next; |
2fd6b7f5 | 1213 | |
ba65dc5e AV |
1214 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1215 | continue; | |
1216 | ||
2fd6b7f5 | 1217 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1218 | |
1219 | ret = enter(data, dentry); | |
1220 | switch (ret) { | |
1221 | case D_WALK_CONTINUE: | |
1222 | break; | |
1223 | case D_WALK_QUIT: | |
2fd6b7f5 | 1224 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1225 | goto out_unlock; |
1226 | case D_WALK_NORETRY: | |
1227 | retry = false; | |
1228 | break; | |
1229 | case D_WALK_SKIP: | |
1230 | spin_unlock(&dentry->d_lock); | |
1231 | continue; | |
2fd6b7f5 | 1232 | } |
db14fc3a | 1233 | |
1da177e4 | 1234 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1235 | spin_unlock(&this_parent->d_lock); |
1236 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1237 | this_parent = dentry; |
2fd6b7f5 | 1238 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1239 | goto repeat; |
1240 | } | |
2fd6b7f5 | 1241 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1242 | } |
1243 | /* | |
1244 | * All done at this level ... ascend and resume the search. | |
1245 | */ | |
ca5358ef AV |
1246 | rcu_read_lock(); |
1247 | ascend: | |
1da177e4 | 1248 | if (this_parent != parent) { |
c826cb7d | 1249 | struct dentry *child = this_parent; |
31dec132 AV |
1250 | this_parent = child->d_parent; |
1251 | ||
31dec132 AV |
1252 | spin_unlock(&child->d_lock); |
1253 | spin_lock(&this_parent->d_lock); | |
1254 | ||
ca5358ef AV |
1255 | /* might go back up the wrong parent if we have had a rename. */ |
1256 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1257 | goto rename_retry; |
2159184e AV |
1258 | /* go into the first sibling still alive */ |
1259 | do { | |
1260 | next = child->d_child.next; | |
ca5358ef AV |
1261 | if (next == &this_parent->d_subdirs) |
1262 | goto ascend; | |
1263 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1264 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1265 | rcu_read_unlock(); |
1da177e4 LT |
1266 | goto resume; |
1267 | } | |
ca5358ef | 1268 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1269 | goto rename_retry; |
ca5358ef | 1270 | rcu_read_unlock(); |
db14fc3a MS |
1271 | if (finish) |
1272 | finish(data); | |
1273 | ||
1274 | out_unlock: | |
1275 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1276 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1277 | return; |
58db63d0 NP |
1278 | |
1279 | rename_retry: | |
ca5358ef AV |
1280 | spin_unlock(&this_parent->d_lock); |
1281 | rcu_read_unlock(); | |
1282 | BUG_ON(seq & 1); | |
db14fc3a MS |
1283 | if (!retry) |
1284 | return; | |
48f5ec21 | 1285 | seq = 1; |
58db63d0 | 1286 | goto again; |
1da177e4 | 1287 | } |
db14fc3a MS |
1288 | |
1289 | /* | |
1290 | * Search for at least 1 mount point in the dentry's subdirs. | |
1291 | * We descend to the next level whenever the d_subdirs | |
1292 | * list is non-empty and continue searching. | |
1293 | */ | |
1294 | ||
db14fc3a MS |
1295 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1296 | { | |
1297 | int *ret = data; | |
1298 | if (d_mountpoint(dentry)) { | |
1299 | *ret = 1; | |
1300 | return D_WALK_QUIT; | |
1301 | } | |
1302 | return D_WALK_CONTINUE; | |
1303 | } | |
1304 | ||
69c88dc7 RD |
1305 | /** |
1306 | * have_submounts - check for mounts over a dentry | |
1307 | * @parent: dentry to check. | |
1308 | * | |
1309 | * Return true if the parent or its subdirectories contain | |
1310 | * a mount point | |
1311 | */ | |
db14fc3a MS |
1312 | int have_submounts(struct dentry *parent) |
1313 | { | |
1314 | int ret = 0; | |
1315 | ||
1316 | d_walk(parent, &ret, check_mount, NULL); | |
1317 | ||
1318 | return ret; | |
1319 | } | |
ec4f8605 | 1320 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1321 | |
eed81007 MS |
1322 | /* |
1323 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1324 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1325 | * subtree can become unreachable). | |
1326 | * | |
1ffe46d1 | 1327 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1328 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1329 | */ | |
1330 | int d_set_mounted(struct dentry *dentry) | |
1331 | { | |
1332 | struct dentry *p; | |
1333 | int ret = -ENOENT; | |
1334 | write_seqlock(&rename_lock); | |
1335 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1336 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1337 | spin_lock(&p->d_lock); |
1338 | if (unlikely(d_unhashed(p))) { | |
1339 | spin_unlock(&p->d_lock); | |
1340 | goto out; | |
1341 | } | |
1342 | spin_unlock(&p->d_lock); | |
1343 | } | |
1344 | spin_lock(&dentry->d_lock); | |
1345 | if (!d_unlinked(dentry)) { | |
1346 | dentry->d_flags |= DCACHE_MOUNTED; | |
1347 | ret = 0; | |
1348 | } | |
1349 | spin_unlock(&dentry->d_lock); | |
1350 | out: | |
1351 | write_sequnlock(&rename_lock); | |
1352 | return ret; | |
1353 | } | |
1354 | ||
1da177e4 | 1355 | /* |
fd517909 | 1356 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1357 | * and move any unused dentries to the end of the unused |
1358 | * list for prune_dcache(). We descend to the next level | |
1359 | * whenever the d_subdirs list is non-empty and continue | |
1360 | * searching. | |
1361 | * | |
1362 | * It returns zero iff there are no unused children, | |
1363 | * otherwise it returns the number of children moved to | |
1364 | * the end of the unused list. This may not be the total | |
1365 | * number of unused children, because select_parent can | |
1366 | * drop the lock and return early due to latency | |
1367 | * constraints. | |
1368 | */ | |
1da177e4 | 1369 | |
db14fc3a MS |
1370 | struct select_data { |
1371 | struct dentry *start; | |
1372 | struct list_head dispose; | |
1373 | int found; | |
1374 | }; | |
23044507 | 1375 | |
db14fc3a MS |
1376 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1377 | { | |
1378 | struct select_data *data = _data; | |
1379 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1380 | |
db14fc3a MS |
1381 | if (data->start == dentry) |
1382 | goto out; | |
2fd6b7f5 | 1383 | |
fe91522a | 1384 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1385 | data->found++; |
fe91522a AV |
1386 | } else { |
1387 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1388 | d_lru_del(dentry); | |
1389 | if (!dentry->d_lockref.count) { | |
1390 | d_shrink_add(dentry, &data->dispose); | |
1391 | data->found++; | |
1392 | } | |
1da177e4 | 1393 | } |
db14fc3a MS |
1394 | /* |
1395 | * We can return to the caller if we have found some (this | |
1396 | * ensures forward progress). We'll be coming back to find | |
1397 | * the rest. | |
1398 | */ | |
fe91522a AV |
1399 | if (!list_empty(&data->dispose)) |
1400 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1401 | out: |
db14fc3a | 1402 | return ret; |
1da177e4 LT |
1403 | } |
1404 | ||
1405 | /** | |
1406 | * shrink_dcache_parent - prune dcache | |
1407 | * @parent: parent of entries to prune | |
1408 | * | |
1409 | * Prune the dcache to remove unused children of the parent dentry. | |
1410 | */ | |
db14fc3a | 1411 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1412 | { |
db14fc3a MS |
1413 | for (;;) { |
1414 | struct select_data data; | |
1da177e4 | 1415 | |
db14fc3a MS |
1416 | INIT_LIST_HEAD(&data.dispose); |
1417 | data.start = parent; | |
1418 | data.found = 0; | |
1419 | ||
1420 | d_walk(parent, &data, select_collect, NULL); | |
1421 | if (!data.found) | |
1422 | break; | |
1423 | ||
1424 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1425 | cond_resched(); |
1426 | } | |
1da177e4 | 1427 | } |
ec4f8605 | 1428 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1429 | |
9c8c10e2 | 1430 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1431 | { |
9c8c10e2 AV |
1432 | /* it has busy descendents; complain about those instead */ |
1433 | if (!list_empty(&dentry->d_subdirs)) | |
1434 | return D_WALK_CONTINUE; | |
42c32608 | 1435 | |
9c8c10e2 AV |
1436 | /* root with refcount 1 is fine */ |
1437 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1438 | return D_WALK_CONTINUE; | |
1439 | ||
1440 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1441 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1442 | dentry, |
1443 | dentry->d_inode ? | |
1444 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1445 | dentry, |
42c32608 AV |
1446 | dentry->d_lockref.count, |
1447 | dentry->d_sb->s_type->name, | |
1448 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1449 | WARN_ON(1); |
1450 | return D_WALK_CONTINUE; | |
1451 | } | |
1452 | ||
1453 | static void do_one_tree(struct dentry *dentry) | |
1454 | { | |
1455 | shrink_dcache_parent(dentry); | |
1456 | d_walk(dentry, dentry, umount_check, NULL); | |
1457 | d_drop(dentry); | |
1458 | dput(dentry); | |
42c32608 AV |
1459 | } |
1460 | ||
1461 | /* | |
1462 | * destroy the dentries attached to a superblock on unmounting | |
1463 | */ | |
1464 | void shrink_dcache_for_umount(struct super_block *sb) | |
1465 | { | |
1466 | struct dentry *dentry; | |
1467 | ||
9c8c10e2 | 1468 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1469 | |
1470 | dentry = sb->s_root; | |
1471 | sb->s_root = NULL; | |
9c8c10e2 | 1472 | do_one_tree(dentry); |
42c32608 AV |
1473 | |
1474 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1475 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1476 | do_one_tree(dentry); | |
42c32608 AV |
1477 | } |
1478 | } | |
1479 | ||
8ed936b5 EB |
1480 | struct detach_data { |
1481 | struct select_data select; | |
1482 | struct dentry *mountpoint; | |
1483 | }; | |
1484 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1485 | { |
8ed936b5 | 1486 | struct detach_data *data = _data; |
848ac114 MS |
1487 | |
1488 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1489 | __dget_dlock(dentry); |
1490 | data->mountpoint = dentry; | |
848ac114 MS |
1491 | return D_WALK_QUIT; |
1492 | } | |
1493 | ||
8ed936b5 | 1494 | return select_collect(&data->select, dentry); |
848ac114 MS |
1495 | } |
1496 | ||
1497 | static void check_and_drop(void *_data) | |
1498 | { | |
8ed936b5 | 1499 | struct detach_data *data = _data; |
848ac114 | 1500 | |
8ed936b5 EB |
1501 | if (!data->mountpoint && !data->select.found) |
1502 | __d_drop(data->select.start); | |
848ac114 MS |
1503 | } |
1504 | ||
1505 | /** | |
1ffe46d1 EB |
1506 | * d_invalidate - detach submounts, prune dcache, and drop |
1507 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1508 | * | |
1ffe46d1 | 1509 | * no dcache lock. |
848ac114 | 1510 | * |
8ed936b5 EB |
1511 | * The final d_drop is done as an atomic operation relative to |
1512 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1513 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1514 | */ |
5542aa2f | 1515 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1516 | { |
1ffe46d1 EB |
1517 | /* |
1518 | * If it's already been dropped, return OK. | |
1519 | */ | |
1520 | spin_lock(&dentry->d_lock); | |
1521 | if (d_unhashed(dentry)) { | |
1522 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1523 | return; |
1ffe46d1 EB |
1524 | } |
1525 | spin_unlock(&dentry->d_lock); | |
1526 | ||
848ac114 MS |
1527 | /* Negative dentries can be dropped without further checks */ |
1528 | if (!dentry->d_inode) { | |
1529 | d_drop(dentry); | |
5542aa2f | 1530 | return; |
848ac114 MS |
1531 | } |
1532 | ||
1533 | for (;;) { | |
8ed936b5 | 1534 | struct detach_data data; |
848ac114 | 1535 | |
8ed936b5 EB |
1536 | data.mountpoint = NULL; |
1537 | INIT_LIST_HEAD(&data.select.dispose); | |
1538 | data.select.start = dentry; | |
1539 | data.select.found = 0; | |
1540 | ||
1541 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1542 | |
8ed936b5 EB |
1543 | if (data.select.found) |
1544 | shrink_dentry_list(&data.select.dispose); | |
848ac114 | 1545 | |
8ed936b5 EB |
1546 | if (data.mountpoint) { |
1547 | detach_mounts(data.mountpoint); | |
1548 | dput(data.mountpoint); | |
1549 | } | |
848ac114 | 1550 | |
8ed936b5 | 1551 | if (!data.mountpoint && !data.select.found) |
848ac114 MS |
1552 | break; |
1553 | ||
1554 | cond_resched(); | |
1555 | } | |
848ac114 | 1556 | } |
1ffe46d1 | 1557 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1558 | |
1da177e4 | 1559 | /** |
a4464dbc AV |
1560 | * __d_alloc - allocate a dcache entry |
1561 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1562 | * @name: qstr of the name |
1563 | * | |
1564 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1565 | * available. On a success the dentry is returned. The name passed in is | |
1566 | * copied and the copy passed in may be reused after this call. | |
1567 | */ | |
1568 | ||
a4464dbc | 1569 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1570 | { |
1571 | struct dentry *dentry; | |
1572 | char *dname; | |
285b102d | 1573 | int err; |
1da177e4 | 1574 | |
e12ba74d | 1575 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1576 | if (!dentry) |
1577 | return NULL; | |
1578 | ||
6326c71f LT |
1579 | /* |
1580 | * We guarantee that the inline name is always NUL-terminated. | |
1581 | * This way the memcpy() done by the name switching in rename | |
1582 | * will still always have a NUL at the end, even if we might | |
1583 | * be overwriting an internal NUL character | |
1584 | */ | |
1585 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd AV |
1586 | if (unlikely(!name)) { |
1587 | static const struct qstr anon = QSTR_INIT("/", 1); | |
1588 | name = &anon; | |
1589 | dname = dentry->d_iname; | |
1590 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1591 | size_t size = offsetof(struct external_name, name[1]); |
5d097056 VD |
1592 | struct external_name *p = kmalloc(size + name->len, |
1593 | GFP_KERNEL_ACCOUNT); | |
8d85b484 | 1594 | if (!p) { |
1da177e4 LT |
1595 | kmem_cache_free(dentry_cache, dentry); |
1596 | return NULL; | |
1597 | } | |
8d85b484 AV |
1598 | atomic_set(&p->u.count, 1); |
1599 | dname = p->name; | |
df4c0e36 AR |
1600 | if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS)) |
1601 | kasan_unpoison_shadow(dname, | |
1602 | round_up(name->len + 1, sizeof(unsigned long))); | |
1da177e4 LT |
1603 | } else { |
1604 | dname = dentry->d_iname; | |
1605 | } | |
1da177e4 LT |
1606 | |
1607 | dentry->d_name.len = name->len; | |
1608 | dentry->d_name.hash = name->hash; | |
1609 | memcpy(dname, name->name, name->len); | |
1610 | dname[name->len] = 0; | |
1611 | ||
6326c71f LT |
1612 | /* Make sure we always see the terminating NUL character */ |
1613 | smp_wmb(); | |
1614 | dentry->d_name.name = dname; | |
1615 | ||
98474236 | 1616 | dentry->d_lockref.count = 1; |
dea3667b | 1617 | dentry->d_flags = 0; |
1da177e4 | 1618 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1619 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1620 | dentry->d_inode = NULL; |
a4464dbc AV |
1621 | dentry->d_parent = dentry; |
1622 | dentry->d_sb = sb; | |
1da177e4 LT |
1623 | dentry->d_op = NULL; |
1624 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1625 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1626 | INIT_LIST_HEAD(&dentry->d_lru); |
1627 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1628 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1629 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1630 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1631 | |
285b102d MS |
1632 | if (dentry->d_op && dentry->d_op->d_init) { |
1633 | err = dentry->d_op->d_init(dentry); | |
1634 | if (err) { | |
1635 | if (dname_external(dentry)) | |
1636 | kfree(external_name(dentry)); | |
1637 | kmem_cache_free(dentry_cache, dentry); | |
1638 | return NULL; | |
1639 | } | |
1640 | } | |
1641 | ||
3e880fb5 | 1642 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1643 | |
1da177e4 LT |
1644 | return dentry; |
1645 | } | |
a4464dbc AV |
1646 | |
1647 | /** | |
1648 | * d_alloc - allocate a dcache entry | |
1649 | * @parent: parent of entry to allocate | |
1650 | * @name: qstr of the name | |
1651 | * | |
1652 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1653 | * available. On a success the dentry is returned. The name passed in is | |
1654 | * copied and the copy passed in may be reused after this call. | |
1655 | */ | |
1656 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1657 | { | |
1658 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1659 | if (!dentry) | |
1660 | return NULL; | |
3d56c25e | 1661 | dentry->d_flags |= DCACHE_RCUACCESS; |
a4464dbc AV |
1662 | spin_lock(&parent->d_lock); |
1663 | /* | |
1664 | * don't need child lock because it is not subject | |
1665 | * to concurrency here | |
1666 | */ | |
1667 | __dget_dlock(parent); | |
1668 | dentry->d_parent = parent; | |
946e51f2 | 1669 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1670 | spin_unlock(&parent->d_lock); |
1671 | ||
1672 | return dentry; | |
1673 | } | |
ec4f8605 | 1674 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1675 | |
ba65dc5e AV |
1676 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1677 | { | |
1678 | struct dentry *dentry = __d_alloc(parent->d_sb, NULL); | |
1679 | if (dentry) { | |
1680 | dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR; | |
1681 | dentry->d_parent = dget(parent); | |
1682 | } | |
1683 | return dentry; | |
1684 | } | |
1685 | ||
e1a24bb0 BF |
1686 | /** |
1687 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1688 | * @sb: the superblock | |
1689 | * @name: qstr of the name | |
1690 | * | |
1691 | * For a filesystem that just pins its dentries in memory and never | |
1692 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1693 | */ | |
4b936885 NP |
1694 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1695 | { | |
e1a24bb0 | 1696 | return __d_alloc(sb, name); |
4b936885 NP |
1697 | } |
1698 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1699 | ||
1da177e4 LT |
1700 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1701 | { | |
1702 | struct qstr q; | |
1703 | ||
1704 | q.name = name; | |
8387ff25 | 1705 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1706 | return d_alloc(parent, &q); |
1707 | } | |
ef26ca97 | 1708 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1709 | |
fb045adb NP |
1710 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1711 | { | |
6f7f7caa LT |
1712 | WARN_ON_ONCE(dentry->d_op); |
1713 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1714 | DCACHE_OP_COMPARE | |
1715 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1716 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1717 | DCACHE_OP_DELETE | |
d101a125 | 1718 | DCACHE_OP_REAL)); |
fb045adb NP |
1719 | dentry->d_op = op; |
1720 | if (!op) | |
1721 | return; | |
1722 | if (op->d_hash) | |
1723 | dentry->d_flags |= DCACHE_OP_HASH; | |
1724 | if (op->d_compare) | |
1725 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1726 | if (op->d_revalidate) | |
1727 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1728 | if (op->d_weak_revalidate) |
1729 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1730 | if (op->d_delete) |
1731 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1732 | if (op->d_prune) |
1733 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1734 | if (op->d_real) |
1735 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1736 | |
1737 | } | |
1738 | EXPORT_SYMBOL(d_set_d_op); | |
1739 | ||
df1a085a DH |
1740 | |
1741 | /* | |
1742 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1743 | * @dentry - The dentry to mark | |
1744 | * | |
1745 | * Mark a dentry as falling through to the lower layer (as set with | |
1746 | * d_pin_lower()). This flag may be recorded on the medium. | |
1747 | */ | |
1748 | void d_set_fallthru(struct dentry *dentry) | |
1749 | { | |
1750 | spin_lock(&dentry->d_lock); | |
1751 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1752 | spin_unlock(&dentry->d_lock); | |
1753 | } | |
1754 | EXPORT_SYMBOL(d_set_fallthru); | |
1755 | ||
b18825a7 DH |
1756 | static unsigned d_flags_for_inode(struct inode *inode) |
1757 | { | |
44bdb5e5 | 1758 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1759 | |
1760 | if (!inode) | |
1761 | return DCACHE_MISS_TYPE; | |
1762 | ||
1763 | if (S_ISDIR(inode->i_mode)) { | |
1764 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1765 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1766 | if (unlikely(!inode->i_op->lookup)) | |
1767 | add_flags = DCACHE_AUTODIR_TYPE; | |
1768 | else | |
1769 | inode->i_opflags |= IOP_LOOKUP; | |
1770 | } | |
44bdb5e5 DH |
1771 | goto type_determined; |
1772 | } | |
1773 | ||
1774 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1775 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1776 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1777 | goto type_determined; |
1778 | } | |
1779 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1780 | } |
1781 | ||
44bdb5e5 DH |
1782 | if (unlikely(!S_ISREG(inode->i_mode))) |
1783 | add_flags = DCACHE_SPECIAL_TYPE; | |
1784 | ||
1785 | type_determined: | |
b18825a7 DH |
1786 | if (unlikely(IS_AUTOMOUNT(inode))) |
1787 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1788 | return add_flags; | |
1789 | } | |
1790 | ||
360da900 OH |
1791 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1792 | { | |
b18825a7 | 1793 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1794 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1795 | |
b23fb0a6 | 1796 | spin_lock(&dentry->d_lock); |
de689f5e | 1797 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1798 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1799 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1800 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 1801 | fsnotify_update_flags(dentry); |
b23fb0a6 | 1802 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1803 | } |
1804 | ||
1da177e4 LT |
1805 | /** |
1806 | * d_instantiate - fill in inode information for a dentry | |
1807 | * @entry: dentry to complete | |
1808 | * @inode: inode to attach to this dentry | |
1809 | * | |
1810 | * Fill in inode information in the entry. | |
1811 | * | |
1812 | * This turns negative dentries into productive full members | |
1813 | * of society. | |
1814 | * | |
1815 | * NOTE! This assumes that the inode count has been incremented | |
1816 | * (or otherwise set) by the caller to indicate that it is now | |
1817 | * in use by the dcache. | |
1818 | */ | |
1819 | ||
1820 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1821 | { | |
946e51f2 | 1822 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1823 | if (inode) { |
b9680917 | 1824 | security_d_instantiate(entry, inode); |
873feea0 | 1825 | spin_lock(&inode->i_lock); |
de689f5e | 1826 | __d_instantiate(entry, inode); |
873feea0 | 1827 | spin_unlock(&inode->i_lock); |
de689f5e | 1828 | } |
1da177e4 | 1829 | } |
ec4f8605 | 1830 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1831 | |
b70a80e7 MS |
1832 | /** |
1833 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1834 | * @entry: dentry to complete | |
1835 | * @inode: inode to attach to this dentry | |
1836 | * | |
1837 | * Fill in inode information in the entry. If a directory alias is found, then | |
1838 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1839 | * guarantees that a directory inode may never have more than one alias. | |
1840 | */ | |
1841 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1842 | { | |
946e51f2 | 1843 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 | 1844 | |
b9680917 | 1845 | security_d_instantiate(entry, inode); |
b70a80e7 MS |
1846 | spin_lock(&inode->i_lock); |
1847 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1848 | spin_unlock(&inode->i_lock); | |
1849 | iput(inode); | |
1850 | return -EBUSY; | |
1851 | } | |
1852 | __d_instantiate(entry, inode); | |
1853 | spin_unlock(&inode->i_lock); | |
b70a80e7 MS |
1854 | |
1855 | return 0; | |
1856 | } | |
1857 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1858 | ||
adc0e91a AV |
1859 | struct dentry *d_make_root(struct inode *root_inode) |
1860 | { | |
1861 | struct dentry *res = NULL; | |
1862 | ||
1863 | if (root_inode) { | |
798434bd | 1864 | res = __d_alloc(root_inode->i_sb, NULL); |
adc0e91a AV |
1865 | if (res) |
1866 | d_instantiate(res, root_inode); | |
1867 | else | |
1868 | iput(root_inode); | |
1869 | } | |
1870 | return res; | |
1871 | } | |
1872 | EXPORT_SYMBOL(d_make_root); | |
1873 | ||
d891eedb BF |
1874 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1875 | { | |
1876 | struct dentry *alias; | |
1877 | ||
b3d9b7a3 | 1878 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1879 | return NULL; |
946e51f2 | 1880 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1881 | __dget(alias); |
1882 | return alias; | |
1883 | } | |
1884 | ||
46f72b34 SW |
1885 | /** |
1886 | * d_find_any_alias - find any alias for a given inode | |
1887 | * @inode: inode to find an alias for | |
1888 | * | |
1889 | * If any aliases exist for the given inode, take and return a | |
1890 | * reference for one of them. If no aliases exist, return %NULL. | |
1891 | */ | |
1892 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1893 | { |
1894 | struct dentry *de; | |
1895 | ||
1896 | spin_lock(&inode->i_lock); | |
1897 | de = __d_find_any_alias(inode); | |
1898 | spin_unlock(&inode->i_lock); | |
1899 | return de; | |
1900 | } | |
46f72b34 | 1901 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1902 | |
49c7dd28 | 1903 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1904 | { |
9308a612 CH |
1905 | struct dentry *tmp; |
1906 | struct dentry *res; | |
b18825a7 | 1907 | unsigned add_flags; |
4ea3ada2 CH |
1908 | |
1909 | if (!inode) | |
44003728 | 1910 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1911 | if (IS_ERR(inode)) |
1912 | return ERR_CAST(inode); | |
1913 | ||
d891eedb | 1914 | res = d_find_any_alias(inode); |
9308a612 CH |
1915 | if (res) |
1916 | goto out_iput; | |
1917 | ||
798434bd | 1918 | tmp = __d_alloc(inode->i_sb, NULL); |
9308a612 CH |
1919 | if (!tmp) { |
1920 | res = ERR_PTR(-ENOMEM); | |
1921 | goto out_iput; | |
4ea3ada2 | 1922 | } |
b5c84bf6 | 1923 | |
b9680917 | 1924 | security_d_instantiate(tmp, inode); |
873feea0 | 1925 | spin_lock(&inode->i_lock); |
d891eedb | 1926 | res = __d_find_any_alias(inode); |
9308a612 | 1927 | if (res) { |
873feea0 | 1928 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1929 | dput(tmp); |
1930 | goto out_iput; | |
1931 | } | |
1932 | ||
1933 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1934 | add_flags = d_flags_for_inode(inode); |
1935 | ||
1936 | if (disconnected) | |
1937 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1938 | |
9308a612 | 1939 | spin_lock(&tmp->d_lock); |
4bf46a27 | 1940 | __d_set_inode_and_type(tmp, inode, add_flags); |
946e51f2 | 1941 | hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry); |
1879fd6a | 1942 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1943 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1944 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1945 | spin_unlock(&tmp->d_lock); |
873feea0 | 1946 | spin_unlock(&inode->i_lock); |
9308a612 | 1947 | |
9308a612 CH |
1948 | return tmp; |
1949 | ||
1950 | out_iput: | |
1951 | iput(inode); | |
1952 | return res; | |
4ea3ada2 | 1953 | } |
1a0a397e BF |
1954 | |
1955 | /** | |
1956 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1957 | * @inode: inode to allocate the dentry for | |
1958 | * | |
1959 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1960 | * similar open by handle operations. The returned dentry may be anonymous, | |
1961 | * or may have a full name (if the inode was already in the cache). | |
1962 | * | |
1963 | * When called on a directory inode, we must ensure that the inode only ever | |
1964 | * has one dentry. If a dentry is found, that is returned instead of | |
1965 | * allocating a new one. | |
1966 | * | |
1967 | * On successful return, the reference to the inode has been transferred | |
1968 | * to the dentry. In case of an error the reference on the inode is released. | |
1969 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1970 | * be passed in and the error will be propagated to the return value, | |
1971 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1972 | */ | |
1973 | struct dentry *d_obtain_alias(struct inode *inode) | |
1974 | { | |
1975 | return __d_obtain_alias(inode, 1); | |
1976 | } | |
adc48720 | 1977 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1978 | |
1a0a397e BF |
1979 | /** |
1980 | * d_obtain_root - find or allocate a dentry for a given inode | |
1981 | * @inode: inode to allocate the dentry for | |
1982 | * | |
1983 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1984 | * | |
1985 | * We must ensure that directory inodes only ever have one dentry. If a | |
1986 | * dentry is found, that is returned instead of allocating a new one. | |
1987 | * | |
1988 | * On successful return, the reference to the inode has been transferred | |
1989 | * to the dentry. In case of an error the reference on the inode is | |
1990 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1991 | * error will be propagate to the return value, with a %NULL @inode | |
1992 | * replaced by ERR_PTR(-ESTALE). | |
1993 | */ | |
1994 | struct dentry *d_obtain_root(struct inode *inode) | |
1995 | { | |
1996 | return __d_obtain_alias(inode, 0); | |
1997 | } | |
1998 | EXPORT_SYMBOL(d_obtain_root); | |
1999 | ||
9403540c BN |
2000 | /** |
2001 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2002 | * @inode: the inode case-insensitive lookup has found | |
2003 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2004 | * @name: the case-exact name to be associated with the returned dentry | |
2005 | * | |
2006 | * This is to avoid filling the dcache with case-insensitive names to the | |
2007 | * same inode, only the actual correct case is stored in the dcache for | |
2008 | * case-insensitive filesystems. | |
2009 | * | |
2010 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2011 | * already exists in in the dcache, use it and return it. | |
2012 | * | |
2013 | * If no entry exists with the exact case name, allocate new dentry with | |
2014 | * the exact case, and return the spliced entry. | |
2015 | */ | |
e45b590b | 2016 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2017 | struct qstr *name) |
2018 | { | |
d9171b93 | 2019 | struct dentry *found, *res; |
9403540c | 2020 | |
b6520c81 CH |
2021 | /* |
2022 | * First check if a dentry matching the name already exists, | |
2023 | * if not go ahead and create it now. | |
2024 | */ | |
9403540c | 2025 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2026 | if (found) { |
2027 | iput(inode); | |
2028 | return found; | |
2029 | } | |
2030 | if (d_in_lookup(dentry)) { | |
2031 | found = d_alloc_parallel(dentry->d_parent, name, | |
2032 | dentry->d_wait); | |
2033 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2034 | iput(inode); | |
2035 | return found; | |
9403540c | 2036 | } |
d9171b93 AV |
2037 | } else { |
2038 | found = d_alloc(dentry->d_parent, name); | |
2039 | if (!found) { | |
2040 | iput(inode); | |
2041 | return ERR_PTR(-ENOMEM); | |
2042 | } | |
2043 | } | |
2044 | res = d_splice_alias(inode, found); | |
2045 | if (res) { | |
2046 | dput(found); | |
2047 | return res; | |
9403540c | 2048 | } |
4f522a24 | 2049 | return found; |
9403540c | 2050 | } |
ec4f8605 | 2051 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2052 | |
12f8ad4b | 2053 | |
d4c91a8f AV |
2054 | static inline bool d_same_name(const struct dentry *dentry, |
2055 | const struct dentry *parent, | |
2056 | const struct qstr *name) | |
12f8ad4b | 2057 | { |
d4c91a8f AV |
2058 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2059 | if (dentry->d_name.len != name->len) | |
2060 | return false; | |
2061 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2062 | } |
d4c91a8f AV |
2063 | return parent->d_op->d_compare(parent, dentry, |
2064 | dentry->d_name.len, dentry->d_name.name, | |
2065 | name) == 0; | |
12f8ad4b LT |
2066 | } |
2067 | ||
31e6b01f NP |
2068 | /** |
2069 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2070 | * @parent: parent dentry | |
2071 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2072 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2073 | * Returns: dentry, or NULL |
2074 | * | |
2075 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2076 | * resolution (store-free path walking) design described in | |
2077 | * Documentation/filesystems/path-lookup.txt. | |
2078 | * | |
2079 | * This is not to be used outside core vfs. | |
2080 | * | |
2081 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2082 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2083 | * without taking d_lock and checking d_seq sequence count against @seq | |
2084 | * returned here. | |
2085 | * | |
15570086 | 2086 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2087 | * function. |
2088 | * | |
2089 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2090 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2091 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2092 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2093 | * |
2094 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2095 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2096 | */ |
8966be90 LT |
2097 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2098 | const struct qstr *name, | |
da53be12 | 2099 | unsigned *seqp) |
31e6b01f | 2100 | { |
26fe5750 | 2101 | u64 hashlen = name->hash_len; |
31e6b01f | 2102 | const unsigned char *str = name->name; |
8387ff25 | 2103 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2104 | struct hlist_bl_node *node; |
31e6b01f NP |
2105 | struct dentry *dentry; |
2106 | ||
2107 | /* | |
2108 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2109 | * required to prevent single threaded performance regressions | |
2110 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2111 | * Keep the two functions in sync. | |
2112 | */ | |
2113 | ||
2114 | /* | |
2115 | * The hash list is protected using RCU. | |
2116 | * | |
2117 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2118 | * races with d_move(). | |
2119 | * | |
2120 | * It is possible that concurrent renames can mess up our list | |
2121 | * walk here and result in missing our dentry, resulting in the | |
2122 | * false-negative result. d_lookup() protects against concurrent | |
2123 | * renames using rename_lock seqlock. | |
2124 | * | |
b0a4bb83 | 2125 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2126 | */ |
b07ad996 | 2127 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2128 | unsigned seq; |
31e6b01f | 2129 | |
31e6b01f | 2130 | seqretry: |
12f8ad4b LT |
2131 | /* |
2132 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2133 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2134 | * |
2135 | * The caller must perform a seqcount check in order | |
da53be12 | 2136 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2137 | * |
2138 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2139 | * we don't wait for the sequence count to stabilize if it | |
2140 | * is in the middle of a sequence change. If we do the slow | |
2141 | * dentry compare, we will do seqretries until it is stable, | |
2142 | * and if we end up with a successful lookup, we actually | |
2143 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2144 | * |
2145 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2146 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2147 | */ |
2148 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2149 | if (dentry->d_parent != parent) |
2150 | continue; | |
2e321806 LT |
2151 | if (d_unhashed(dentry)) |
2152 | continue; | |
12f8ad4b | 2153 | |
830c0f0e | 2154 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
d4c91a8f AV |
2155 | int tlen; |
2156 | const char *tname; | |
26fe5750 LT |
2157 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2158 | continue; | |
d4c91a8f AV |
2159 | tlen = dentry->d_name.len; |
2160 | tname = dentry->d_name.name; | |
2161 | /* we want a consistent (name,len) pair */ | |
2162 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2163 | cpu_relax(); | |
12f8ad4b LT |
2164 | goto seqretry; |
2165 | } | |
d4c91a8f AV |
2166 | if (parent->d_op->d_compare(parent, dentry, |
2167 | tlen, tname, name) != 0) | |
2168 | continue; | |
2169 | } else { | |
2170 | if (dentry->d_name.hash_len != hashlen) | |
2171 | continue; | |
2172 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2173 | continue; | |
31e6b01f | 2174 | } |
da53be12 | 2175 | *seqp = seq; |
d4c91a8f | 2176 | return dentry; |
31e6b01f NP |
2177 | } |
2178 | return NULL; | |
2179 | } | |
2180 | ||
1da177e4 LT |
2181 | /** |
2182 | * d_lookup - search for a dentry | |
2183 | * @parent: parent dentry | |
2184 | * @name: qstr of name we wish to find | |
b04f784e | 2185 | * Returns: dentry, or NULL |
1da177e4 | 2186 | * |
b04f784e NP |
2187 | * d_lookup searches the children of the parent dentry for the name in |
2188 | * question. If the dentry is found its reference count is incremented and the | |
2189 | * dentry is returned. The caller must use dput to free the entry when it has | |
2190 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2191 | */ |
da2d8455 | 2192 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2193 | { |
31e6b01f | 2194 | struct dentry *dentry; |
949854d0 | 2195 | unsigned seq; |
1da177e4 | 2196 | |
b8314f93 DY |
2197 | do { |
2198 | seq = read_seqbegin(&rename_lock); | |
2199 | dentry = __d_lookup(parent, name); | |
2200 | if (dentry) | |
1da177e4 LT |
2201 | break; |
2202 | } while (read_seqretry(&rename_lock, seq)); | |
2203 | return dentry; | |
2204 | } | |
ec4f8605 | 2205 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2206 | |
31e6b01f | 2207 | /** |
b04f784e NP |
2208 | * __d_lookup - search for a dentry (racy) |
2209 | * @parent: parent dentry | |
2210 | * @name: qstr of name we wish to find | |
2211 | * Returns: dentry, or NULL | |
2212 | * | |
2213 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2214 | * false-negative result due to unrelated rename activity. | |
2215 | * | |
2216 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2217 | * however it must be used carefully, eg. with a following d_lookup in | |
2218 | * the case of failure. | |
2219 | * | |
2220 | * __d_lookup callers must be commented. | |
2221 | */ | |
a713ca2a | 2222 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2223 | { |
1da177e4 | 2224 | unsigned int hash = name->hash; |
8387ff25 | 2225 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2226 | struct hlist_bl_node *node; |
31e6b01f | 2227 | struct dentry *found = NULL; |
665a7583 | 2228 | struct dentry *dentry; |
1da177e4 | 2229 | |
31e6b01f NP |
2230 | /* |
2231 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2232 | * required to prevent single threaded performance regressions | |
2233 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2234 | * Keep the two functions in sync. | |
2235 | */ | |
2236 | ||
b04f784e NP |
2237 | /* |
2238 | * The hash list is protected using RCU. | |
2239 | * | |
2240 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2241 | * with d_move(). | |
2242 | * | |
2243 | * It is possible that concurrent renames can mess up our list | |
2244 | * walk here and result in missing our dentry, resulting in the | |
2245 | * false-negative result. d_lookup() protects against concurrent | |
2246 | * renames using rename_lock seqlock. | |
2247 | * | |
b0a4bb83 | 2248 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2249 | */ |
1da177e4 LT |
2250 | rcu_read_lock(); |
2251 | ||
b07ad996 | 2252 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2253 | |
1da177e4 LT |
2254 | if (dentry->d_name.hash != hash) |
2255 | continue; | |
1da177e4 LT |
2256 | |
2257 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2258 | if (dentry->d_parent != parent) |
2259 | goto next; | |
d0185c08 LT |
2260 | if (d_unhashed(dentry)) |
2261 | goto next; | |
2262 | ||
d4c91a8f AV |
2263 | if (!d_same_name(dentry, parent, name)) |
2264 | goto next; | |
1da177e4 | 2265 | |
98474236 | 2266 | dentry->d_lockref.count++; |
d0185c08 | 2267 | found = dentry; |
1da177e4 LT |
2268 | spin_unlock(&dentry->d_lock); |
2269 | break; | |
2270 | next: | |
2271 | spin_unlock(&dentry->d_lock); | |
2272 | } | |
2273 | rcu_read_unlock(); | |
2274 | ||
2275 | return found; | |
2276 | } | |
2277 | ||
3e7e241f EB |
2278 | /** |
2279 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2280 | * @dir: Directory to search in | |
2281 | * @name: qstr of name we wish to find | |
2282 | * | |
4f522a24 | 2283 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2284 | */ |
2285 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2286 | { | |
3e7e241f EB |
2287 | /* |
2288 | * Check for a fs-specific hash function. Note that we must | |
2289 | * calculate the standard hash first, as the d_op->d_hash() | |
2290 | * routine may choose to leave the hash value unchanged. | |
2291 | */ | |
8387ff25 | 2292 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2293 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2294 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2295 | if (unlikely(err < 0)) |
2296 | return ERR_PTR(err); | |
3e7e241f | 2297 | } |
4f522a24 | 2298 | return d_lookup(dir, name); |
3e7e241f | 2299 | } |
4f522a24 | 2300 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2301 | |
1da177e4 LT |
2302 | /* |
2303 | * When a file is deleted, we have two options: | |
2304 | * - turn this dentry into a negative dentry | |
2305 | * - unhash this dentry and free it. | |
2306 | * | |
2307 | * Usually, we want to just turn this into | |
2308 | * a negative dentry, but if anybody else is | |
2309 | * currently using the dentry or the inode | |
2310 | * we can't do that and we fall back on removing | |
2311 | * it from the hash queues and waiting for | |
2312 | * it to be deleted later when it has no users | |
2313 | */ | |
2314 | ||
2315 | /** | |
2316 | * d_delete - delete a dentry | |
2317 | * @dentry: The dentry to delete | |
2318 | * | |
2319 | * Turn the dentry into a negative dentry if possible, otherwise | |
2320 | * remove it from the hash queues so it can be deleted later | |
2321 | */ | |
2322 | ||
2323 | void d_delete(struct dentry * dentry) | |
2324 | { | |
873feea0 | 2325 | struct inode *inode; |
7a91bf7f | 2326 | int isdir = 0; |
1da177e4 LT |
2327 | /* |
2328 | * Are we the only user? | |
2329 | */ | |
357f8e65 | 2330 | again: |
1da177e4 | 2331 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2332 | inode = dentry->d_inode; |
2333 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2334 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2335 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2336 | spin_unlock(&dentry->d_lock); |
2337 | cpu_relax(); | |
2338 | goto again; | |
2339 | } | |
13e3c5e5 | 2340 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2341 | dentry_unlink_inode(dentry); |
7a91bf7f | 2342 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2343 | return; |
2344 | } | |
2345 | ||
2346 | if (!d_unhashed(dentry)) | |
2347 | __d_drop(dentry); | |
2348 | ||
2349 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2350 | |
2351 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2352 | } |
ec4f8605 | 2353 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2354 | |
b07ad996 | 2355 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2356 | { |
ceb5bdc2 | 2357 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2358 | hlist_bl_lock(b); |
b07ad996 | 2359 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2360 | hlist_bl_unlock(b); |
1da177e4 LT |
2361 | } |
2362 | ||
770bfad8 DH |
2363 | static void _d_rehash(struct dentry * entry) |
2364 | { | |
8387ff25 | 2365 | __d_rehash(entry, d_hash(entry->d_name.hash)); |
770bfad8 DH |
2366 | } |
2367 | ||
1da177e4 LT |
2368 | /** |
2369 | * d_rehash - add an entry back to the hash | |
2370 | * @entry: dentry to add to the hash | |
2371 | * | |
2372 | * Adds a dentry to the hash according to its name. | |
2373 | */ | |
2374 | ||
2375 | void d_rehash(struct dentry * entry) | |
2376 | { | |
1da177e4 | 2377 | spin_lock(&entry->d_lock); |
770bfad8 | 2378 | _d_rehash(entry); |
1da177e4 | 2379 | spin_unlock(&entry->d_lock); |
1da177e4 | 2380 | } |
ec4f8605 | 2381 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2382 | |
84e710da AV |
2383 | static inline unsigned start_dir_add(struct inode *dir) |
2384 | { | |
2385 | ||
2386 | for (;;) { | |
2387 | unsigned n = dir->i_dir_seq; | |
2388 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2389 | return n; | |
2390 | cpu_relax(); | |
2391 | } | |
2392 | } | |
2393 | ||
2394 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2395 | { | |
2396 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2397 | } | |
2398 | ||
d9171b93 AV |
2399 | static void d_wait_lookup(struct dentry *dentry) |
2400 | { | |
2401 | if (d_in_lookup(dentry)) { | |
2402 | DECLARE_WAITQUEUE(wait, current); | |
2403 | add_wait_queue(dentry->d_wait, &wait); | |
2404 | do { | |
2405 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2406 | spin_unlock(&dentry->d_lock); | |
2407 | schedule(); | |
2408 | spin_lock(&dentry->d_lock); | |
2409 | } while (d_in_lookup(dentry)); | |
2410 | } | |
2411 | } | |
2412 | ||
94bdd655 | 2413 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2414 | const struct qstr *name, |
2415 | wait_queue_head_t *wq) | |
94bdd655 | 2416 | { |
94bdd655 | 2417 | unsigned int hash = name->hash; |
94bdd655 AV |
2418 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2419 | struct hlist_bl_node *node; | |
2420 | struct dentry *new = d_alloc(parent, name); | |
2421 | struct dentry *dentry; | |
2422 | unsigned seq, r_seq, d_seq; | |
2423 | ||
2424 | if (unlikely(!new)) | |
2425 | return ERR_PTR(-ENOMEM); | |
2426 | ||
2427 | retry: | |
2428 | rcu_read_lock(); | |
2429 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1; | |
2430 | r_seq = read_seqbegin(&rename_lock); | |
2431 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2432 | if (unlikely(dentry)) { | |
2433 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2434 | rcu_read_unlock(); | |
2435 | goto retry; | |
2436 | } | |
2437 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2438 | rcu_read_unlock(); | |
2439 | dput(dentry); | |
2440 | goto retry; | |
2441 | } | |
2442 | rcu_read_unlock(); | |
2443 | dput(new); | |
2444 | return dentry; | |
2445 | } | |
2446 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2447 | rcu_read_unlock(); | |
2448 | goto retry; | |
2449 | } | |
2450 | hlist_bl_lock(b); | |
2451 | if (unlikely(parent->d_inode->i_dir_seq != seq)) { | |
2452 | hlist_bl_unlock(b); | |
2453 | rcu_read_unlock(); | |
2454 | goto retry; | |
2455 | } | |
94bdd655 AV |
2456 | /* |
2457 | * No changes for the parent since the beginning of d_lookup(). | |
2458 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2459 | * any potential in-lookup matches are going to stay here until | |
2460 | * we unlock the chain. All fields are stable in everything | |
2461 | * we encounter. | |
2462 | */ | |
2463 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2464 | if (dentry->d_name.hash != hash) | |
2465 | continue; | |
2466 | if (dentry->d_parent != parent) | |
2467 | continue; | |
d4c91a8f AV |
2468 | if (!d_same_name(dentry, parent, name)) |
2469 | continue; | |
94bdd655 | 2470 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2471 | /* now we can try to grab a reference */ |
2472 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2473 | rcu_read_unlock(); | |
2474 | goto retry; | |
2475 | } | |
2476 | ||
2477 | rcu_read_unlock(); | |
2478 | /* | |
2479 | * somebody is likely to be still doing lookup for it; | |
2480 | * wait for them to finish | |
2481 | */ | |
d9171b93 AV |
2482 | spin_lock(&dentry->d_lock); |
2483 | d_wait_lookup(dentry); | |
2484 | /* | |
2485 | * it's not in-lookup anymore; in principle we should repeat | |
2486 | * everything from dcache lookup, but it's likely to be what | |
2487 | * d_lookup() would've found anyway. If it is, just return it; | |
2488 | * otherwise we really have to repeat the whole thing. | |
2489 | */ | |
2490 | if (unlikely(dentry->d_name.hash != hash)) | |
2491 | goto mismatch; | |
2492 | if (unlikely(dentry->d_parent != parent)) | |
2493 | goto mismatch; | |
2494 | if (unlikely(d_unhashed(dentry))) | |
2495 | goto mismatch; | |
d4c91a8f AV |
2496 | if (unlikely(!d_same_name(dentry, parent, name))) |
2497 | goto mismatch; | |
d9171b93 AV |
2498 | /* OK, it *is* a hashed match; return it */ |
2499 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2500 | dput(new); |
2501 | return dentry; | |
2502 | } | |
e7d6ef97 | 2503 | rcu_read_unlock(); |
94bdd655 AV |
2504 | /* we can't take ->d_lock here; it's OK, though. */ |
2505 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2506 | new->d_wait = wq; |
94bdd655 AV |
2507 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2508 | hlist_bl_unlock(b); | |
2509 | return new; | |
d9171b93 AV |
2510 | mismatch: |
2511 | spin_unlock(&dentry->d_lock); | |
2512 | dput(dentry); | |
2513 | goto retry; | |
94bdd655 AV |
2514 | } |
2515 | EXPORT_SYMBOL(d_alloc_parallel); | |
2516 | ||
85c7f810 AV |
2517 | void __d_lookup_done(struct dentry *dentry) |
2518 | { | |
94bdd655 AV |
2519 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2520 | dentry->d_name.hash); | |
2521 | hlist_bl_lock(b); | |
85c7f810 | 2522 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2523 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2524 | wake_up_all(dentry->d_wait); |
2525 | dentry->d_wait = NULL; | |
94bdd655 AV |
2526 | hlist_bl_unlock(b); |
2527 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2528 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2529 | } |
2530 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2531 | |
2532 | /* inode->i_lock held if inode is non-NULL */ | |
2533 | ||
2534 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2535 | { | |
84e710da AV |
2536 | struct inode *dir = NULL; |
2537 | unsigned n; | |
0568d705 | 2538 | spin_lock(&dentry->d_lock); |
84e710da AV |
2539 | if (unlikely(d_in_lookup(dentry))) { |
2540 | dir = dentry->d_parent->d_inode; | |
2541 | n = start_dir_add(dir); | |
85c7f810 | 2542 | __d_lookup_done(dentry); |
84e710da | 2543 | } |
ed782b5a | 2544 | if (inode) { |
0568d705 AV |
2545 | unsigned add_flags = d_flags_for_inode(inode); |
2546 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2547 | raw_write_seqcount_begin(&dentry->d_seq); | |
2548 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2549 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2550 | fsnotify_update_flags(dentry); |
ed782b5a | 2551 | } |
0568d705 | 2552 | _d_rehash(dentry); |
84e710da AV |
2553 | if (dir) |
2554 | end_dir_add(dir, n); | |
0568d705 AV |
2555 | spin_unlock(&dentry->d_lock); |
2556 | if (inode) | |
2557 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2558 | } |
2559 | ||
34d0d19d AV |
2560 | /** |
2561 | * d_add - add dentry to hash queues | |
2562 | * @entry: dentry to add | |
2563 | * @inode: The inode to attach to this dentry | |
2564 | * | |
2565 | * This adds the entry to the hash queues and initializes @inode. | |
2566 | * The entry was actually filled in earlier during d_alloc(). | |
2567 | */ | |
2568 | ||
2569 | void d_add(struct dentry *entry, struct inode *inode) | |
2570 | { | |
b9680917 AV |
2571 | if (inode) { |
2572 | security_d_instantiate(entry, inode); | |
ed782b5a | 2573 | spin_lock(&inode->i_lock); |
b9680917 | 2574 | } |
ed782b5a | 2575 | __d_add(entry, inode); |
34d0d19d AV |
2576 | } |
2577 | EXPORT_SYMBOL(d_add); | |
2578 | ||
668d0cd5 AV |
2579 | /** |
2580 | * d_exact_alias - find and hash an exact unhashed alias | |
2581 | * @entry: dentry to add | |
2582 | * @inode: The inode to go with this dentry | |
2583 | * | |
2584 | * If an unhashed dentry with the same name/parent and desired | |
2585 | * inode already exists, hash and return it. Otherwise, return | |
2586 | * NULL. | |
2587 | * | |
2588 | * Parent directory should be locked. | |
2589 | */ | |
2590 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2591 | { | |
2592 | struct dentry *alias; | |
668d0cd5 AV |
2593 | unsigned int hash = entry->d_name.hash; |
2594 | ||
2595 | spin_lock(&inode->i_lock); | |
2596 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2597 | /* | |
2598 | * Don't need alias->d_lock here, because aliases with | |
2599 | * d_parent == entry->d_parent are not subject to name or | |
2600 | * parent changes, because the parent inode i_mutex is held. | |
2601 | */ | |
2602 | if (alias->d_name.hash != hash) | |
2603 | continue; | |
2604 | if (alias->d_parent != entry->d_parent) | |
2605 | continue; | |
d4c91a8f | 2606 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2607 | continue; |
2608 | spin_lock(&alias->d_lock); | |
2609 | if (!d_unhashed(alias)) { | |
2610 | spin_unlock(&alias->d_lock); | |
2611 | alias = NULL; | |
2612 | } else { | |
2613 | __dget_dlock(alias); | |
2614 | _d_rehash(alias); | |
2615 | spin_unlock(&alias->d_lock); | |
2616 | } | |
2617 | spin_unlock(&inode->i_lock); | |
2618 | return alias; | |
2619 | } | |
2620 | spin_unlock(&inode->i_lock); | |
2621 | return NULL; | |
2622 | } | |
2623 | EXPORT_SYMBOL(d_exact_alias); | |
2624 | ||
fb2d5b86 NP |
2625 | /** |
2626 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2627 | * @dentry: dentry to be updated | |
2628 | * @name: new name | |
2629 | * | |
2630 | * Update a case insensitive dentry with new case of name. | |
2631 | * | |
2632 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2633 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2634 | * lengths). | |
2635 | * | |
2636 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2637 | * keep renames and concurrent inserts, and readdir(2) away). | |
2638 | */ | |
2639 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2640 | { | |
5955102c | 2641 | BUG_ON(!inode_is_locked(dentry->d_parent->d_inode)); |
fb2d5b86 NP |
2642 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2643 | ||
fb2d5b86 | 2644 | spin_lock(&dentry->d_lock); |
31e6b01f | 2645 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2646 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2647 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2648 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2649 | } |
2650 | EXPORT_SYMBOL(dentry_update_name_case); | |
2651 | ||
8d85b484 | 2652 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2653 | { |
8d85b484 AV |
2654 | if (unlikely(dname_external(target))) { |
2655 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2656 | /* |
2657 | * Both external: swap the pointers | |
2658 | */ | |
9a8d5bb4 | 2659 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2660 | } else { |
2661 | /* | |
2662 | * dentry:internal, target:external. Steal target's | |
2663 | * storage and make target internal. | |
2664 | */ | |
321bcf92 BF |
2665 | memcpy(target->d_iname, dentry->d_name.name, |
2666 | dentry->d_name.len + 1); | |
1da177e4 LT |
2667 | dentry->d_name.name = target->d_name.name; |
2668 | target->d_name.name = target->d_iname; | |
2669 | } | |
2670 | } else { | |
8d85b484 | 2671 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2672 | /* |
2673 | * dentry:external, target:internal. Give dentry's | |
2674 | * storage to target and make dentry internal | |
2675 | */ | |
2676 | memcpy(dentry->d_iname, target->d_name.name, | |
2677 | target->d_name.len + 1); | |
2678 | target->d_name.name = dentry->d_name.name; | |
2679 | dentry->d_name.name = dentry->d_iname; | |
2680 | } else { | |
2681 | /* | |
da1ce067 | 2682 | * Both are internal. |
1da177e4 | 2683 | */ |
da1ce067 MS |
2684 | unsigned int i; |
2685 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
08d4f772 MP |
2686 | kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN); |
2687 | kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN); | |
da1ce067 MS |
2688 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { |
2689 | swap(((long *) &dentry->d_iname)[i], | |
2690 | ((long *) &target->d_iname)[i]); | |
2691 | } | |
1da177e4 LT |
2692 | } |
2693 | } | |
a28ddb87 | 2694 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2695 | } |
2696 | ||
8d85b484 AV |
2697 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2698 | { | |
2699 | struct external_name *old_name = NULL; | |
2700 | if (unlikely(dname_external(dentry))) | |
2701 | old_name = external_name(dentry); | |
2702 | if (unlikely(dname_external(target))) { | |
2703 | atomic_inc(&external_name(target)->u.count); | |
2704 | dentry->d_name = target->d_name; | |
2705 | } else { | |
2706 | memcpy(dentry->d_iname, target->d_name.name, | |
2707 | target->d_name.len + 1); | |
2708 | dentry->d_name.name = dentry->d_iname; | |
2709 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2710 | } | |
2711 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2712 | kfree_rcu(old_name, u.head); | |
2713 | } | |
2714 | ||
2fd6b7f5 NP |
2715 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2716 | { | |
2717 | /* | |
2718 | * XXXX: do we really need to take target->d_lock? | |
2719 | */ | |
2720 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2721 | spin_lock(&target->d_parent->d_lock); | |
2722 | else { | |
2723 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2724 | spin_lock(&dentry->d_parent->d_lock); | |
2725 | spin_lock_nested(&target->d_parent->d_lock, | |
2726 | DENTRY_D_LOCK_NESTED); | |
2727 | } else { | |
2728 | spin_lock(&target->d_parent->d_lock); | |
2729 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2730 | DENTRY_D_LOCK_NESTED); | |
2731 | } | |
2732 | } | |
2733 | if (target < dentry) { | |
2734 | spin_lock_nested(&target->d_lock, 2); | |
2735 | spin_lock_nested(&dentry->d_lock, 3); | |
2736 | } else { | |
2737 | spin_lock_nested(&dentry->d_lock, 2); | |
2738 | spin_lock_nested(&target->d_lock, 3); | |
2739 | } | |
2740 | } | |
2741 | ||
986c0194 | 2742 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2743 | { |
2744 | if (target->d_parent != dentry->d_parent) | |
2745 | spin_unlock(&dentry->d_parent->d_lock); | |
2746 | if (target->d_parent != target) | |
2747 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2748 | spin_unlock(&target->d_lock); |
2749 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2750 | } |
2751 | ||
1da177e4 | 2752 | /* |
2fd6b7f5 NP |
2753 | * When switching names, the actual string doesn't strictly have to |
2754 | * be preserved in the target - because we're dropping the target | |
2755 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2756 | * the new name before we switch, unless we are going to rehash |
2757 | * it. Note that if we *do* unhash the target, we are not allowed | |
2758 | * to rehash it without giving it a new name/hash key - whether | |
2759 | * we swap or overwrite the names here, resulting name won't match | |
2760 | * the reality in filesystem; it's only there for d_path() purposes. | |
2761 | * Note that all of this is happening under rename_lock, so the | |
2762 | * any hash lookup seeing it in the middle of manipulations will | |
2763 | * be discarded anyway. So we do not care what happens to the hash | |
2764 | * key in that case. | |
1da177e4 | 2765 | */ |
9eaef27b | 2766 | /* |
18367501 | 2767 | * __d_move - move a dentry |
1da177e4 LT |
2768 | * @dentry: entry to move |
2769 | * @target: new dentry | |
da1ce067 | 2770 | * @exchange: exchange the two dentries |
1da177e4 LT |
2771 | * |
2772 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2773 | * dcache entries should not be moved in this way. Caller must hold |
2774 | * rename_lock, the i_mutex of the source and target directories, | |
2775 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2776 | */ |
da1ce067 MS |
2777 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2778 | bool exchange) | |
1da177e4 | 2779 | { |
84e710da AV |
2780 | struct inode *dir = NULL; |
2781 | unsigned n; | |
1da177e4 LT |
2782 | if (!dentry->d_inode) |
2783 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2784 | ||
2fd6b7f5 NP |
2785 | BUG_ON(d_ancestor(dentry, target)); |
2786 | BUG_ON(d_ancestor(target, dentry)); | |
2787 | ||
2fd6b7f5 | 2788 | dentry_lock_for_move(dentry, target); |
84e710da AV |
2789 | if (unlikely(d_in_lookup(target))) { |
2790 | dir = target->d_parent->d_inode; | |
2791 | n = start_dir_add(dir); | |
85c7f810 | 2792 | __d_lookup_done(target); |
84e710da | 2793 | } |
1da177e4 | 2794 | |
31e6b01f | 2795 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2796 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2797 | |
ceb5bdc2 NP |
2798 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2799 | ||
2800 | /* | |
2801 | * Move the dentry to the target hash queue. Don't bother checking | |
2802 | * for the same hash queue because of how unlikely it is. | |
2803 | */ | |
2804 | __d_drop(dentry); | |
8387ff25 | 2805 | __d_rehash(dentry, d_hash(target->d_name.hash)); |
1da177e4 | 2806 | |
da1ce067 MS |
2807 | /* |
2808 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2809 | * the two dentries, then rehash onto the other's hash queue. | |
2810 | */ | |
1da177e4 | 2811 | __d_drop(target); |
da1ce067 | 2812 | if (exchange) { |
8387ff25 | 2813 | __d_rehash(target, d_hash(dentry->d_name.hash)); |
da1ce067 | 2814 | } |
1da177e4 | 2815 | |
1da177e4 | 2816 | /* Switch the names.. */ |
8d85b484 AV |
2817 | if (exchange) |
2818 | swap_names(dentry, target); | |
2819 | else | |
2820 | copy_name(dentry, target); | |
1da177e4 | 2821 | |
63cf427a | 2822 | /* ... and switch them in the tree */ |
1da177e4 | 2823 | if (IS_ROOT(dentry)) { |
63cf427a | 2824 | /* splicing a tree */ |
3d56c25e | 2825 | dentry->d_flags |= DCACHE_RCUACCESS; |
1da177e4 LT |
2826 | dentry->d_parent = target->d_parent; |
2827 | target->d_parent = target; | |
946e51f2 AV |
2828 | list_del_init(&target->d_child); |
2829 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
1da177e4 | 2830 | } else { |
63cf427a | 2831 | /* swapping two dentries */ |
9a8d5bb4 | 2832 | swap(dentry->d_parent, target->d_parent); |
946e51f2 AV |
2833 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
2834 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
63cf427a | 2835 | if (exchange) |
affda484 AV |
2836 | fsnotify_update_flags(target); |
2837 | fsnotify_update_flags(dentry); | |
1da177e4 LT |
2838 | } |
2839 | ||
31e6b01f NP |
2840 | write_seqcount_end(&target->d_seq); |
2841 | write_seqcount_end(&dentry->d_seq); | |
2842 | ||
84e710da AV |
2843 | if (dir) |
2844 | end_dir_add(dir, n); | |
986c0194 | 2845 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2846 | } |
2847 | ||
2848 | /* | |
2849 | * d_move - move a dentry | |
2850 | * @dentry: entry to move | |
2851 | * @target: new dentry | |
2852 | * | |
2853 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2854 | * dcache entries should not be moved in this way. See the locking |
2855 | * requirements for __d_move. | |
18367501 AV |
2856 | */ |
2857 | void d_move(struct dentry *dentry, struct dentry *target) | |
2858 | { | |
2859 | write_seqlock(&rename_lock); | |
da1ce067 | 2860 | __d_move(dentry, target, false); |
1da177e4 | 2861 | write_sequnlock(&rename_lock); |
9eaef27b | 2862 | } |
ec4f8605 | 2863 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2864 | |
da1ce067 MS |
2865 | /* |
2866 | * d_exchange - exchange two dentries | |
2867 | * @dentry1: first dentry | |
2868 | * @dentry2: second dentry | |
2869 | */ | |
2870 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2871 | { | |
2872 | write_seqlock(&rename_lock); | |
2873 | ||
2874 | WARN_ON(!dentry1->d_inode); | |
2875 | WARN_ON(!dentry2->d_inode); | |
2876 | WARN_ON(IS_ROOT(dentry1)); | |
2877 | WARN_ON(IS_ROOT(dentry2)); | |
2878 | ||
2879 | __d_move(dentry1, dentry2, true); | |
2880 | ||
2881 | write_sequnlock(&rename_lock); | |
2882 | } | |
2883 | ||
e2761a11 OH |
2884 | /** |
2885 | * d_ancestor - search for an ancestor | |
2886 | * @p1: ancestor dentry | |
2887 | * @p2: child dentry | |
2888 | * | |
2889 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2890 | * an ancestor of p2, else NULL. | |
9eaef27b | 2891 | */ |
e2761a11 | 2892 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2893 | { |
2894 | struct dentry *p; | |
2895 | ||
871c0067 | 2896 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2897 | if (p->d_parent == p1) |
e2761a11 | 2898 | return p; |
9eaef27b | 2899 | } |
e2761a11 | 2900 | return NULL; |
9eaef27b TM |
2901 | } |
2902 | ||
2903 | /* | |
2904 | * This helper attempts to cope with remotely renamed directories | |
2905 | * | |
2906 | * It assumes that the caller is already holding | |
a03e283b | 2907 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2908 | * |
2909 | * Note: If ever the locking in lock_rename() changes, then please | |
2910 | * remember to update this too... | |
9eaef27b | 2911 | */ |
b5ae6b15 | 2912 | static int __d_unalias(struct inode *inode, |
873feea0 | 2913 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2914 | { |
9902af79 AV |
2915 | struct mutex *m1 = NULL; |
2916 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2917 | int ret = -ESTALE; |
9eaef27b TM |
2918 | |
2919 | /* If alias and dentry share a parent, then no extra locks required */ | |
2920 | if (alias->d_parent == dentry->d_parent) | |
2921 | goto out_unalias; | |
2922 | ||
9eaef27b | 2923 | /* See lock_rename() */ |
9eaef27b TM |
2924 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2925 | goto out_err; | |
2926 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2927 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2928 | goto out_err; |
9902af79 | 2929 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 2930 | out_unalias: |
8ed936b5 | 2931 | __d_move(alias, dentry, false); |
b5ae6b15 | 2932 | ret = 0; |
9eaef27b | 2933 | out_err: |
9eaef27b | 2934 | if (m2) |
9902af79 | 2935 | up_read(m2); |
9eaef27b TM |
2936 | if (m1) |
2937 | mutex_unlock(m1); | |
2938 | return ret; | |
2939 | } | |
2940 | ||
3f70bd51 BF |
2941 | /** |
2942 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2943 | * @inode: the inode which may have a disconnected dentry | |
2944 | * @dentry: a negative dentry which we want to point to the inode. | |
2945 | * | |
da093a9b BF |
2946 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2947 | * place of the given dentry and return it, else simply d_add the inode | |
2948 | * to the dentry and return NULL. | |
3f70bd51 | 2949 | * |
908790fa BF |
2950 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2951 | * we should error out: directories can't have multiple aliases. | |
2952 | * | |
3f70bd51 BF |
2953 | * This is needed in the lookup routine of any filesystem that is exportable |
2954 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2955 | * | |
2956 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2957 | * is returned. This matches the expected return value of ->lookup. | |
2958 | * | |
2959 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2960 | * In that case, we know that the inode will be a regular file, and also this | |
2961 | * will only occur during atomic_open. So we need to check for the dentry | |
2962 | * being already hashed only in the final case. | |
2963 | */ | |
2964 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2965 | { | |
3f70bd51 BF |
2966 | if (IS_ERR(inode)) |
2967 | return ERR_CAST(inode); | |
2968 | ||
770bfad8 DH |
2969 | BUG_ON(!d_unhashed(dentry)); |
2970 | ||
de689f5e | 2971 | if (!inode) |
b5ae6b15 | 2972 | goto out; |
de689f5e | 2973 | |
b9680917 | 2974 | security_d_instantiate(dentry, inode); |
873feea0 | 2975 | spin_lock(&inode->i_lock); |
9eaef27b | 2976 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
2977 | struct dentry *new = __d_find_any_alias(inode); |
2978 | if (unlikely(new)) { | |
a03e283b EB |
2979 | /* The reference to new ensures it remains an alias */ |
2980 | spin_unlock(&inode->i_lock); | |
18367501 | 2981 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
2982 | if (unlikely(d_ancestor(new, dentry))) { |
2983 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
2984 | dput(new); |
2985 | new = ERR_PTR(-ELOOP); | |
2986 | pr_warn_ratelimited( | |
2987 | "VFS: Lookup of '%s' in %s %s" | |
2988 | " would have caused loop\n", | |
2989 | dentry->d_name.name, | |
2990 | inode->i_sb->s_type->name, | |
2991 | inode->i_sb->s_id); | |
2992 | } else if (!IS_ROOT(new)) { | |
2993 | int err = __d_unalias(inode, dentry, new); | |
18367501 | 2994 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
2995 | if (err) { |
2996 | dput(new); | |
2997 | new = ERR_PTR(err); | |
2998 | } | |
18367501 | 2999 | } else { |
b5ae6b15 AV |
3000 | __d_move(new, dentry, false); |
3001 | write_sequnlock(&rename_lock); | |
dd179946 | 3002 | } |
b5ae6b15 AV |
3003 | iput(inode); |
3004 | return new; | |
9eaef27b | 3005 | } |
770bfad8 | 3006 | } |
b5ae6b15 | 3007 | out: |
ed782b5a | 3008 | __d_add(dentry, inode); |
b5ae6b15 | 3009 | return NULL; |
770bfad8 | 3010 | } |
b5ae6b15 | 3011 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3012 | |
cdd16d02 | 3013 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
3014 | { |
3015 | *buflen -= namelen; | |
3016 | if (*buflen < 0) | |
3017 | return -ENAMETOOLONG; | |
3018 | *buffer -= namelen; | |
3019 | memcpy(*buffer, str, namelen); | |
3020 | return 0; | |
3021 | } | |
3022 | ||
232d2d60 WL |
3023 | /** |
3024 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
3025 | * @buffer: buffer pointer |
3026 | * @buflen: allocated length of the buffer | |
3027 | * @name: name string and length qstr structure | |
232d2d60 WL |
3028 | * |
3029 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
3030 | * make sure that either the old or the new name pointer and length are | |
3031 | * fetched. However, there may be mismatch between length and pointer. | |
3032 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
3033 | * the length is reached or a null byte is found. It also prepends "/" at | |
3034 | * the beginning of the name. The sequence number check at the caller will | |
3035 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
3036 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
3037 | * |
3038 | * Data dependency barrier is needed to make sure that we see that terminating | |
3039 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 3040 | */ |
cdd16d02 MS |
3041 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
3042 | { | |
232d2d60 WL |
3043 | const char *dname = ACCESS_ONCE(name->name); |
3044 | u32 dlen = ACCESS_ONCE(name->len); | |
3045 | char *p; | |
3046 | ||
6d13f694 AV |
3047 | smp_read_barrier_depends(); |
3048 | ||
232d2d60 | 3049 | *buflen -= dlen + 1; |
e825196d AV |
3050 | if (*buflen < 0) |
3051 | return -ENAMETOOLONG; | |
232d2d60 WL |
3052 | p = *buffer -= dlen + 1; |
3053 | *p++ = '/'; | |
3054 | while (dlen--) { | |
3055 | char c = *dname++; | |
3056 | if (!c) | |
3057 | break; | |
3058 | *p++ = c; | |
3059 | } | |
3060 | return 0; | |
cdd16d02 MS |
3061 | } |
3062 | ||
1da177e4 | 3063 | /** |
208898c1 | 3064 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 3065 | * @path: the dentry/vfsmount to report |
02125a82 | 3066 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
3067 | * @buffer: pointer to the end of the buffer |
3068 | * @buflen: pointer to buffer length | |
552ce544 | 3069 | * |
18129977 WL |
3070 | * The function will first try to write out the pathname without taking any |
3071 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
3072 | * It only checks the sequence number of the global rename_lock as any change | |
3073 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
3074 | * sequence number. If the sequence number had been changed, it will restart | |
3075 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
3076 | * In this case, there is no need to take the RCU read lock as the recursive | |
3077 | * parent pointer references will keep the dentry chain alive as long as no | |
3078 | * rename operation is performed. | |
1da177e4 | 3079 | */ |
02125a82 AV |
3080 | static int prepend_path(const struct path *path, |
3081 | const struct path *root, | |
f2eb6575 | 3082 | char **buffer, int *buflen) |
1da177e4 | 3083 | { |
ede4cebc AV |
3084 | struct dentry *dentry; |
3085 | struct vfsmount *vfsmnt; | |
3086 | struct mount *mnt; | |
f2eb6575 | 3087 | int error = 0; |
48a066e7 | 3088 | unsigned seq, m_seq = 0; |
232d2d60 WL |
3089 | char *bptr; |
3090 | int blen; | |
6092d048 | 3091 | |
48f5ec21 | 3092 | rcu_read_lock(); |
48a066e7 AV |
3093 | restart_mnt: |
3094 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
3095 | seq = 0; | |
4ec6c2ae | 3096 | rcu_read_lock(); |
232d2d60 WL |
3097 | restart: |
3098 | bptr = *buffer; | |
3099 | blen = *buflen; | |
48a066e7 | 3100 | error = 0; |
ede4cebc AV |
3101 | dentry = path->dentry; |
3102 | vfsmnt = path->mnt; | |
3103 | mnt = real_mount(vfsmnt); | |
232d2d60 | 3104 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 3105 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
3106 | struct dentry * parent; |
3107 | ||
1da177e4 | 3108 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 3109 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
cde93be4 EB |
3110 | /* Escaped? */ |
3111 | if (dentry != vfsmnt->mnt_root) { | |
3112 | bptr = *buffer; | |
3113 | blen = *buflen; | |
3114 | error = 3; | |
3115 | break; | |
3116 | } | |
552ce544 | 3117 | /* Global root? */ |
48a066e7 AV |
3118 | if (mnt != parent) { |
3119 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
3120 | mnt = parent; | |
232d2d60 WL |
3121 | vfsmnt = &mnt->mnt; |
3122 | continue; | |
3123 | } | |
232d2d60 WL |
3124 | if (!error) |
3125 | error = is_mounted(vfsmnt) ? 1 : 2; | |
3126 | break; | |
1da177e4 LT |
3127 | } |
3128 | parent = dentry->d_parent; | |
3129 | prefetch(parent); | |
232d2d60 | 3130 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
3131 | if (error) |
3132 | break; | |
3133 | ||
1da177e4 LT |
3134 | dentry = parent; |
3135 | } | |
48f5ec21 AV |
3136 | if (!(seq & 1)) |
3137 | rcu_read_unlock(); | |
3138 | if (need_seqretry(&rename_lock, seq)) { | |
3139 | seq = 1; | |
232d2d60 | 3140 | goto restart; |
48f5ec21 AV |
3141 | } |
3142 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
3143 | |
3144 | if (!(m_seq & 1)) | |
3145 | rcu_read_unlock(); | |
48a066e7 AV |
3146 | if (need_seqretry(&mount_lock, m_seq)) { |
3147 | m_seq = 1; | |
3148 | goto restart_mnt; | |
3149 | } | |
3150 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 3151 | |
232d2d60 WL |
3152 | if (error >= 0 && bptr == *buffer) { |
3153 | if (--blen < 0) | |
3154 | error = -ENAMETOOLONG; | |
3155 | else | |
3156 | *--bptr = '/'; | |
3157 | } | |
3158 | *buffer = bptr; | |
3159 | *buflen = blen; | |
7ea600b5 | 3160 | return error; |
f2eb6575 | 3161 | } |
be285c71 | 3162 | |
f2eb6575 MS |
3163 | /** |
3164 | * __d_path - return the path of a dentry | |
3165 | * @path: the dentry/vfsmount to report | |
02125a82 | 3166 | * @root: root vfsmnt/dentry |
cd956a1c | 3167 | * @buf: buffer to return value in |
f2eb6575 MS |
3168 | * @buflen: buffer length |
3169 | * | |
ffd1f4ed | 3170 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
3171 | * |
3172 | * Returns a pointer into the buffer or an error code if the | |
3173 | * path was too long. | |
3174 | * | |
be148247 | 3175 | * "buflen" should be positive. |
f2eb6575 | 3176 | * |
02125a82 | 3177 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 3178 | */ |
02125a82 AV |
3179 | char *__d_path(const struct path *path, |
3180 | const struct path *root, | |
f2eb6575 MS |
3181 | char *buf, int buflen) |
3182 | { | |
3183 | char *res = buf + buflen; | |
3184 | int error; | |
3185 | ||
3186 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 3187 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 3188 | |
02125a82 AV |
3189 | if (error < 0) |
3190 | return ERR_PTR(error); | |
3191 | if (error > 0) | |
3192 | return NULL; | |
3193 | return res; | |
3194 | } | |
3195 | ||
3196 | char *d_absolute_path(const struct path *path, | |
3197 | char *buf, int buflen) | |
3198 | { | |
3199 | struct path root = {}; | |
3200 | char *res = buf + buflen; | |
3201 | int error; | |
3202 | ||
3203 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3204 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3205 | |
3206 | if (error > 1) | |
3207 | error = -EINVAL; | |
3208 | if (error < 0) | |
f2eb6575 | 3209 | return ERR_PTR(error); |
f2eb6575 | 3210 | return res; |
1da177e4 LT |
3211 | } |
3212 | ||
ffd1f4ed MS |
3213 | /* |
3214 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3215 | */ | |
02125a82 AV |
3216 | static int path_with_deleted(const struct path *path, |
3217 | const struct path *root, | |
3218 | char **buf, int *buflen) | |
ffd1f4ed MS |
3219 | { |
3220 | prepend(buf, buflen, "\0", 1); | |
3221 | if (d_unlinked(path->dentry)) { | |
3222 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3223 | if (error) | |
3224 | return error; | |
3225 | } | |
3226 | ||
3227 | return prepend_path(path, root, buf, buflen); | |
3228 | } | |
3229 | ||
8df9d1a4 MS |
3230 | static int prepend_unreachable(char **buffer, int *buflen) |
3231 | { | |
3232 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3233 | } | |
3234 | ||
68f0d9d9 LT |
3235 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3236 | { | |
3237 | unsigned seq; | |
3238 | ||
3239 | do { | |
3240 | seq = read_seqcount_begin(&fs->seq); | |
3241 | *root = fs->root; | |
3242 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3243 | } | |
3244 | ||
a03a8a70 JB |
3245 | /** |
3246 | * d_path - return the path of a dentry | |
cf28b486 | 3247 | * @path: path to report |
a03a8a70 JB |
3248 | * @buf: buffer to return value in |
3249 | * @buflen: buffer length | |
3250 | * | |
3251 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3252 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3253 | * | |
52afeefb AV |
3254 | * Returns a pointer into the buffer or an error code if the path was |
3255 | * too long. Note: Callers should use the returned pointer, not the passed | |
3256 | * in buffer, to use the name! The implementation often starts at an offset | |
3257 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3258 | * |
31f3e0b3 | 3259 | * "buflen" should be positive. |
a03a8a70 | 3260 | */ |
20d4fdc1 | 3261 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3262 | { |
ffd1f4ed | 3263 | char *res = buf + buflen; |
6ac08c39 | 3264 | struct path root; |
ffd1f4ed | 3265 | int error; |
1da177e4 | 3266 | |
c23fbb6b ED |
3267 | /* |
3268 | * We have various synthetic filesystems that never get mounted. On | |
3269 | * these filesystems dentries are never used for lookup purposes, and | |
3270 | * thus don't need to be hashed. They also don't need a name until a | |
3271 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3272 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3273 | * |
3274 | * Some pseudo inodes are mountable. When they are mounted | |
3275 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3276 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3277 | */ |
f48cfddc EB |
3278 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3279 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3280 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3281 | |
68f0d9d9 LT |
3282 | rcu_read_lock(); |
3283 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3284 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3285 | rcu_read_unlock(); |
3286 | ||
02125a82 | 3287 | if (error < 0) |
ffd1f4ed | 3288 | res = ERR_PTR(error); |
1da177e4 LT |
3289 | return res; |
3290 | } | |
ec4f8605 | 3291 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3292 | |
c23fbb6b ED |
3293 | /* |
3294 | * Helper function for dentry_operations.d_dname() members | |
3295 | */ | |
3296 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3297 | const char *fmt, ...) | |
3298 | { | |
3299 | va_list args; | |
3300 | char temp[64]; | |
3301 | int sz; | |
3302 | ||
3303 | va_start(args, fmt); | |
3304 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3305 | va_end(args); | |
3306 | ||
3307 | if (sz > sizeof(temp) || sz > buflen) | |
3308 | return ERR_PTR(-ENAMETOOLONG); | |
3309 | ||
3310 | buffer += buflen - sz; | |
3311 | return memcpy(buffer, temp, sz); | |
3312 | } | |
3313 | ||
118b2302 AV |
3314 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3315 | { | |
3316 | char *end = buffer + buflen; | |
3317 | /* these dentries are never renamed, so d_lock is not needed */ | |
3318 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3319 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3320 | prepend(&end, &buflen, "/", 1)) |
3321 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3322 | return end; |
118b2302 | 3323 | } |
31bbe16f | 3324 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3325 | |
6092d048 RP |
3326 | /* |
3327 | * Write full pathname from the root of the filesystem into the buffer. | |
3328 | */ | |
f6500801 | 3329 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3330 | { |
f6500801 | 3331 | struct dentry *dentry; |
232d2d60 WL |
3332 | char *end, *retval; |
3333 | int len, seq = 0; | |
3334 | int error = 0; | |
6092d048 | 3335 | |
f6500801 AV |
3336 | if (buflen < 2) |
3337 | goto Elong; | |
3338 | ||
48f5ec21 | 3339 | rcu_read_lock(); |
232d2d60 | 3340 | restart: |
f6500801 | 3341 | dentry = d; |
232d2d60 WL |
3342 | end = buf + buflen; |
3343 | len = buflen; | |
3344 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3345 | /* Get '/' right */ |
3346 | retval = end-1; | |
3347 | *retval = '/'; | |
232d2d60 | 3348 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3349 | while (!IS_ROOT(dentry)) { |
3350 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3351 | |
6092d048 | 3352 | prefetch(parent); |
232d2d60 WL |
3353 | error = prepend_name(&end, &len, &dentry->d_name); |
3354 | if (error) | |
3355 | break; | |
6092d048 RP |
3356 | |
3357 | retval = end; | |
3358 | dentry = parent; | |
3359 | } | |
48f5ec21 AV |
3360 | if (!(seq & 1)) |
3361 | rcu_read_unlock(); | |
3362 | if (need_seqretry(&rename_lock, seq)) { | |
3363 | seq = 1; | |
232d2d60 | 3364 | goto restart; |
48f5ec21 AV |
3365 | } |
3366 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3367 | if (error) |
3368 | goto Elong; | |
c103135c AV |
3369 | return retval; |
3370 | Elong: | |
3371 | return ERR_PTR(-ENAMETOOLONG); | |
3372 | } | |
ec2447c2 NP |
3373 | |
3374 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3375 | { | |
232d2d60 | 3376 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3377 | } |
3378 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3379 | |
3380 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3381 | { | |
3382 | char *p = NULL; | |
3383 | char *retval; | |
3384 | ||
c103135c AV |
3385 | if (d_unlinked(dentry)) { |
3386 | p = buf + buflen; | |
3387 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3388 | goto Elong; | |
3389 | buflen++; | |
3390 | } | |
3391 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3392 | if (!IS_ERR(retval) && p) |
3393 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3394 | return retval; |
3395 | Elong: | |
6092d048 RP |
3396 | return ERR_PTR(-ENAMETOOLONG); |
3397 | } | |
3398 | ||
8b19e341 LT |
3399 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3400 | struct path *pwd) | |
5762482f | 3401 | { |
8b19e341 LT |
3402 | unsigned seq; |
3403 | ||
3404 | do { | |
3405 | seq = read_seqcount_begin(&fs->seq); | |
3406 | *root = fs->root; | |
3407 | *pwd = fs->pwd; | |
3408 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3409 | } |
3410 | ||
1da177e4 LT |
3411 | /* |
3412 | * NOTE! The user-level library version returns a | |
3413 | * character pointer. The kernel system call just | |
3414 | * returns the length of the buffer filled (which | |
3415 | * includes the ending '\0' character), or a negative | |
3416 | * error value. So libc would do something like | |
3417 | * | |
3418 | * char *getcwd(char * buf, size_t size) | |
3419 | * { | |
3420 | * int retval; | |
3421 | * | |
3422 | * retval = sys_getcwd(buf, size); | |
3423 | * if (retval >= 0) | |
3424 | * return buf; | |
3425 | * errno = -retval; | |
3426 | * return NULL; | |
3427 | * } | |
3428 | */ | |
3cdad428 | 3429 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3430 | { |
552ce544 | 3431 | int error; |
6ac08c39 | 3432 | struct path pwd, root; |
3272c544 | 3433 | char *page = __getname(); |
1da177e4 LT |
3434 | |
3435 | if (!page) | |
3436 | return -ENOMEM; | |
3437 | ||
8b19e341 LT |
3438 | rcu_read_lock(); |
3439 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3440 | |
552ce544 | 3441 | error = -ENOENT; |
f3da392e | 3442 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3443 | unsigned long len; |
3272c544 LT |
3444 | char *cwd = page + PATH_MAX; |
3445 | int buflen = PATH_MAX; | |
1da177e4 | 3446 | |
8df9d1a4 | 3447 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3448 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3449 | rcu_read_unlock(); |
552ce544 | 3450 | |
02125a82 | 3451 | if (error < 0) |
552ce544 LT |
3452 | goto out; |
3453 | ||
8df9d1a4 | 3454 | /* Unreachable from current root */ |
02125a82 | 3455 | if (error > 0) { |
8df9d1a4 MS |
3456 | error = prepend_unreachable(&cwd, &buflen); |
3457 | if (error) | |
3458 | goto out; | |
3459 | } | |
3460 | ||
552ce544 | 3461 | error = -ERANGE; |
3272c544 | 3462 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3463 | if (len <= size) { |
3464 | error = len; | |
3465 | if (copy_to_user(buf, cwd, len)) | |
3466 | error = -EFAULT; | |
3467 | } | |
949854d0 | 3468 | } else { |
ff812d72 | 3469 | rcu_read_unlock(); |
949854d0 | 3470 | } |
1da177e4 LT |
3471 | |
3472 | out: | |
3272c544 | 3473 | __putname(page); |
1da177e4 LT |
3474 | return error; |
3475 | } | |
3476 | ||
3477 | /* | |
3478 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3479 | * | |
3480 | * Trivially implemented using the dcache structure | |
3481 | */ | |
3482 | ||
3483 | /** | |
3484 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3485 | * @new_dentry: new dentry | |
3486 | * @old_dentry: old dentry | |
3487 | * | |
a6e5787f YB |
3488 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3489 | * Returns false otherwise. | |
1da177e4 LT |
3490 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3491 | */ | |
3492 | ||
a6e5787f | 3493 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3494 | { |
a6e5787f | 3495 | bool result; |
949854d0 | 3496 | unsigned seq; |
1da177e4 | 3497 | |
e2761a11 | 3498 | if (new_dentry == old_dentry) |
a6e5787f | 3499 | return true; |
e2761a11 | 3500 | |
e2761a11 | 3501 | do { |
1da177e4 | 3502 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3503 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3504 | /* |
3505 | * Need rcu_readlock to protect against the d_parent trashing | |
3506 | * due to d_move | |
3507 | */ | |
3508 | rcu_read_lock(); | |
e2761a11 | 3509 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3510 | result = true; |
e2761a11 | 3511 | else |
a6e5787f | 3512 | result = false; |
949854d0 | 3513 | rcu_read_unlock(); |
1da177e4 | 3514 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3515 | |
3516 | return result; | |
3517 | } | |
3518 | ||
db14fc3a | 3519 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3520 | { |
db14fc3a MS |
3521 | struct dentry *root = data; |
3522 | if (dentry != root) { | |
3523 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3524 | return D_WALK_SKIP; | |
1da177e4 | 3525 | |
01ddc4ed MS |
3526 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3527 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3528 | dentry->d_lockref.count--; | |
3529 | } | |
1da177e4 | 3530 | } |
db14fc3a MS |
3531 | return D_WALK_CONTINUE; |
3532 | } | |
58db63d0 | 3533 | |
db14fc3a MS |
3534 | void d_genocide(struct dentry *parent) |
3535 | { | |
3536 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3537 | } |
3538 | ||
60545d0d | 3539 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3540 | { |
60545d0d AV |
3541 | inode_dec_link_count(inode); |
3542 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3543 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3544 | !d_unlinked(dentry)); |
3545 | spin_lock(&dentry->d_parent->d_lock); | |
3546 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3547 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3548 | (unsigned long long)inode->i_ino); | |
3549 | spin_unlock(&dentry->d_lock); | |
3550 | spin_unlock(&dentry->d_parent->d_lock); | |
3551 | d_instantiate(dentry, inode); | |
1da177e4 | 3552 | } |
60545d0d | 3553 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3554 | |
3555 | static __initdata unsigned long dhash_entries; | |
3556 | static int __init set_dhash_entries(char *str) | |
3557 | { | |
3558 | if (!str) | |
3559 | return 0; | |
3560 | dhash_entries = simple_strtoul(str, &str, 0); | |
3561 | return 1; | |
3562 | } | |
3563 | __setup("dhash_entries=", set_dhash_entries); | |
3564 | ||
3565 | static void __init dcache_init_early(void) | |
3566 | { | |
074b8517 | 3567 | unsigned int loop; |
1da177e4 LT |
3568 | |
3569 | /* If hashes are distributed across NUMA nodes, defer | |
3570 | * hash allocation until vmalloc space is available. | |
3571 | */ | |
3572 | if (hashdist) | |
3573 | return; | |
3574 | ||
3575 | dentry_hashtable = | |
3576 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3577 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3578 | dhash_entries, |
3579 | 13, | |
3580 | HASH_EARLY, | |
3581 | &d_hash_shift, | |
3582 | &d_hash_mask, | |
31fe62b9 | 3583 | 0, |
1da177e4 LT |
3584 | 0); |
3585 | ||
074b8517 | 3586 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3587 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3588 | } |
3589 | ||
74bf17cf | 3590 | static void __init dcache_init(void) |
1da177e4 | 3591 | { |
074b8517 | 3592 | unsigned int loop; |
1da177e4 LT |
3593 | |
3594 | /* | |
3595 | * A constructor could be added for stable state like the lists, | |
3596 | * but it is probably not worth it because of the cache nature | |
3597 | * of the dcache. | |
3598 | */ | |
0a31bd5f | 3599 | dentry_cache = KMEM_CACHE(dentry, |
5d097056 | 3600 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT); |
1da177e4 LT |
3601 | |
3602 | /* Hash may have been set up in dcache_init_early */ | |
3603 | if (!hashdist) | |
3604 | return; | |
3605 | ||
3606 | dentry_hashtable = | |
3607 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3608 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3609 | dhash_entries, |
3610 | 13, | |
3611 | 0, | |
3612 | &d_hash_shift, | |
3613 | &d_hash_mask, | |
31fe62b9 | 3614 | 0, |
1da177e4 LT |
3615 | 0); |
3616 | ||
074b8517 | 3617 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3618 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3619 | } |
3620 | ||
3621 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3622 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3623 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3624 | |
1da177e4 LT |
3625 | EXPORT_SYMBOL(d_genocide); |
3626 | ||
1da177e4 LT |
3627 | void __init vfs_caches_init_early(void) |
3628 | { | |
3629 | dcache_init_early(); | |
3630 | inode_init_early(); | |
3631 | } | |
3632 | ||
4248b0da | 3633 | void __init vfs_caches_init(void) |
1da177e4 | 3634 | { |
1da177e4 | 3635 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, |
20c2df83 | 3636 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3637 | |
74bf17cf DC |
3638 | dcache_init(); |
3639 | inode_init(); | |
4248b0da MG |
3640 | files_init(); |
3641 | files_maxfiles_init(); | |
74bf17cf | 3642 | mnt_init(); |
1da177e4 LT |
3643 | bdev_cache_init(); |
3644 | chrdev_init(); | |
3645 | } |