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 | |
8966be90 | 107 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 108 | unsigned int hash) |
ceb5bdc2 | 109 | { |
6d7d1a0d | 110 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
99d263d4 | 111 | return dentry_hashtable + hash_32(hash, d_hash_shift); |
ceb5bdc2 NP |
112 | } |
113 | ||
94bdd655 AV |
114 | #define IN_LOOKUP_SHIFT 10 |
115 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
116 | ||
117 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
118 | unsigned int hash) | |
119 | { | |
120 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
121 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
122 | } | |
123 | ||
124 | ||
1da177e4 LT |
125 | /* Statistics gathering. */ |
126 | struct dentry_stat_t dentry_stat = { | |
127 | .age_limit = 45, | |
128 | }; | |
129 | ||
3942c07c | 130 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 131 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
312d3ca8 CH |
132 | |
133 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
62d36c77 DC |
134 | |
135 | /* | |
136 | * Here we resort to our own counters instead of using generic per-cpu counters | |
137 | * for consistency with what the vfs inode code does. We are expected to harvest | |
138 | * better code and performance by having our own specialized counters. | |
139 | * | |
140 | * Please note that the loop is done over all possible CPUs, not over all online | |
141 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
142 | * on and off. If one of them goes off, we will just keep their counters. | |
143 | * | |
144 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
145 | * please update all vfs counters to match. | |
146 | */ | |
3942c07c | 147 | static long get_nr_dentry(void) |
3e880fb5 NP |
148 | { |
149 | int i; | |
3942c07c | 150 | long sum = 0; |
3e880fb5 NP |
151 | for_each_possible_cpu(i) |
152 | sum += per_cpu(nr_dentry, i); | |
153 | return sum < 0 ? 0 : sum; | |
154 | } | |
155 | ||
62d36c77 DC |
156 | static long get_nr_dentry_unused(void) |
157 | { | |
158 | int i; | |
159 | long sum = 0; | |
160 | for_each_possible_cpu(i) | |
161 | sum += per_cpu(nr_dentry_unused, i); | |
162 | return sum < 0 ? 0 : sum; | |
163 | } | |
164 | ||
1f7e0616 | 165 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, |
312d3ca8 CH |
166 | size_t *lenp, loff_t *ppos) |
167 | { | |
3e880fb5 | 168 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 169 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
3942c07c | 170 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 CH |
171 | } |
172 | #endif | |
173 | ||
5483f18e LT |
174 | /* |
175 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
176 | * The strings are both count bytes long, and count is non-zero. | |
177 | */ | |
e419b4cc LT |
178 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
179 | ||
180 | #include <asm/word-at-a-time.h> | |
181 | /* | |
182 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
183 | * aligned allocation for this particular component. We don't | |
184 | * strictly need the load_unaligned_zeropad() safety, but it | |
185 | * doesn't hurt either. | |
186 | * | |
187 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
188 | * need the careful unaligned handling. | |
189 | */ | |
94753db5 | 190 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 191 | { |
bfcfaa77 | 192 | unsigned long a,b,mask; |
bfcfaa77 LT |
193 | |
194 | for (;;) { | |
12f8ad4b | 195 | a = *(unsigned long *)cs; |
e419b4cc | 196 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
197 | if (tcount < sizeof(unsigned long)) |
198 | break; | |
199 | if (unlikely(a != b)) | |
200 | return 1; | |
201 | cs += sizeof(unsigned long); | |
202 | ct += sizeof(unsigned long); | |
203 | tcount -= sizeof(unsigned long); | |
204 | if (!tcount) | |
205 | return 0; | |
206 | } | |
a5c21dce | 207 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 208 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
209 | } |
210 | ||
bfcfaa77 | 211 | #else |
e419b4cc | 212 | |
94753db5 | 213 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 214 | { |
5483f18e LT |
215 | do { |
216 | if (*cs != *ct) | |
217 | return 1; | |
218 | cs++; | |
219 | ct++; | |
220 | tcount--; | |
221 | } while (tcount); | |
222 | return 0; | |
223 | } | |
224 | ||
e419b4cc LT |
225 | #endif |
226 | ||
94753db5 LT |
227 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
228 | { | |
6326c71f | 229 | const unsigned char *cs; |
94753db5 LT |
230 | /* |
231 | * Be careful about RCU walk racing with rename: | |
232 | * use ACCESS_ONCE to fetch the name pointer. | |
233 | * | |
234 | * NOTE! Even if a rename will mean that the length | |
235 | * was not loaded atomically, we don't care. The | |
236 | * RCU walk will check the sequence count eventually, | |
237 | * and catch it. And we won't overrun the buffer, | |
238 | * because we're reading the name pointer atomically, | |
239 | * and a dentry name is guaranteed to be properly | |
240 | * terminated with a NUL byte. | |
241 | * | |
242 | * End result: even if 'len' is wrong, we'll exit | |
243 | * early because the data cannot match (there can | |
244 | * be no NUL in the ct/tcount data) | |
245 | */ | |
6326c71f LT |
246 | cs = ACCESS_ONCE(dentry->d_name.name); |
247 | smp_read_barrier_depends(); | |
248 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
249 | } |
250 | ||
8d85b484 AV |
251 | struct external_name { |
252 | union { | |
253 | atomic_t count; | |
254 | struct rcu_head head; | |
255 | } u; | |
256 | unsigned char name[]; | |
257 | }; | |
258 | ||
259 | static inline struct external_name *external_name(struct dentry *dentry) | |
260 | { | |
261 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
262 | } | |
263 | ||
9c82ab9c | 264 | static void __d_free(struct rcu_head *head) |
1da177e4 | 265 | { |
9c82ab9c CH |
266 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
267 | ||
8d85b484 AV |
268 | kmem_cache_free(dentry_cache, dentry); |
269 | } | |
270 | ||
271 | static void __d_free_external(struct rcu_head *head) | |
272 | { | |
273 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
8d85b484 | 274 | kfree(external_name(dentry)); |
1da177e4 LT |
275 | kmem_cache_free(dentry_cache, dentry); |
276 | } | |
277 | ||
810bb172 AV |
278 | static inline int dname_external(const struct dentry *dentry) |
279 | { | |
280 | return dentry->d_name.name != dentry->d_iname; | |
281 | } | |
282 | ||
4bf46a27 DH |
283 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
284 | struct inode *inode, | |
285 | unsigned type_flags) | |
286 | { | |
287 | unsigned flags; | |
288 | ||
289 | dentry->d_inode = inode; | |
4bf46a27 DH |
290 | flags = READ_ONCE(dentry->d_flags); |
291 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
292 | flags |= type_flags; | |
293 | WRITE_ONCE(dentry->d_flags, flags); | |
294 | } | |
295 | ||
4bf46a27 DH |
296 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
297 | { | |
298 | unsigned flags = READ_ONCE(dentry->d_flags); | |
299 | ||
300 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
301 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 DH |
302 | dentry->d_inode = NULL; |
303 | } | |
304 | ||
b4f0354e AV |
305 | static void dentry_free(struct dentry *dentry) |
306 | { | |
946e51f2 | 307 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
308 | if (unlikely(dname_external(dentry))) { |
309 | struct external_name *p = external_name(dentry); | |
310 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
311 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
312 | return; | |
313 | } | |
314 | } | |
b4f0354e AV |
315 | /* if dentry was never visible to RCU, immediate free is OK */ |
316 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
317 | __d_free(&dentry->d_u.d_rcu); | |
318 | else | |
319 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
320 | } | |
321 | ||
31e6b01f | 322 | /** |
a7c6f571 | 323 | * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups |
ff5fdb61 | 324 | * @dentry: the target dentry |
31e6b01f NP |
325 | * After this call, in-progress rcu-walk path lookup will fail. This |
326 | * should be called after unhashing, and after changing d_inode (if | |
327 | * the dentry has not already been unhashed). | |
328 | */ | |
a7c6f571 | 329 | static inline void dentry_rcuwalk_invalidate(struct dentry *dentry) |
31e6b01f | 330 | { |
a7c6f571 PZ |
331 | lockdep_assert_held(&dentry->d_lock); |
332 | /* Go through am invalidation barrier */ | |
333 | write_seqcount_invalidate(&dentry->d_seq); | |
31e6b01f NP |
334 | } |
335 | ||
1da177e4 LT |
336 | /* |
337 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
338 | * d_iput() operation if defined. Dentry has no refcount |
339 | * and is unhashed. | |
1da177e4 | 340 | */ |
858119e1 | 341 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 342 | __releases(dentry->d_lock) |
873feea0 | 343 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
344 | { |
345 | struct inode *inode = dentry->d_inode; | |
346 | if (inode) { | |
4bf46a27 | 347 | __d_clear_type_and_inode(dentry); |
946e51f2 | 348 | hlist_del_init(&dentry->d_u.d_alias); |
1da177e4 | 349 | spin_unlock(&dentry->d_lock); |
873feea0 | 350 | spin_unlock(&inode->i_lock); |
f805fbda LT |
351 | if (!inode->i_nlink) |
352 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
353 | if (dentry->d_op && dentry->d_op->d_iput) |
354 | dentry->d_op->d_iput(dentry, inode); | |
355 | else | |
356 | iput(inode); | |
357 | } else { | |
358 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
359 | } |
360 | } | |
361 | ||
31e6b01f NP |
362 | /* |
363 | * Release the dentry's inode, using the filesystem | |
364 | * d_iput() operation if defined. dentry remains in-use. | |
365 | */ | |
366 | static void dentry_unlink_inode(struct dentry * dentry) | |
367 | __releases(dentry->d_lock) | |
873feea0 | 368 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
369 | { |
370 | struct inode *inode = dentry->d_inode; | |
a528aca7 AV |
371 | |
372 | raw_write_seqcount_begin(&dentry->d_seq); | |
4bf46a27 | 373 | __d_clear_type_and_inode(dentry); |
946e51f2 | 374 | hlist_del_init(&dentry->d_u.d_alias); |
a528aca7 | 375 | raw_write_seqcount_end(&dentry->d_seq); |
31e6b01f | 376 | spin_unlock(&dentry->d_lock); |
873feea0 | 377 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
378 | if (!inode->i_nlink) |
379 | fsnotify_inoderemove(inode); | |
380 | if (dentry->d_op && dentry->d_op->d_iput) | |
381 | dentry->d_op->d_iput(dentry, inode); | |
382 | else | |
383 | iput(inode); | |
384 | } | |
385 | ||
89dc77bc LT |
386 | /* |
387 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
388 | * is in use - which includes both the "real" per-superblock | |
389 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
390 | * | |
391 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
392 | * on the shrink list (ie not on the superblock LRU list). | |
393 | * | |
394 | * The per-cpu "nr_dentry_unused" counters are updated with | |
395 | * the DCACHE_LRU_LIST bit. | |
396 | * | |
397 | * These helper functions make sure we always follow the | |
398 | * rules. d_lock must be held by the caller. | |
399 | */ | |
400 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
401 | static void d_lru_add(struct dentry *dentry) | |
402 | { | |
403 | D_FLAG_VERIFY(dentry, 0); | |
404 | dentry->d_flags |= DCACHE_LRU_LIST; | |
405 | this_cpu_inc(nr_dentry_unused); | |
406 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
407 | } | |
408 | ||
409 | static void d_lru_del(struct dentry *dentry) | |
410 | { | |
411 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
412 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
413 | this_cpu_dec(nr_dentry_unused); | |
414 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
415 | } | |
416 | ||
417 | static void d_shrink_del(struct dentry *dentry) | |
418 | { | |
419 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
420 | list_del_init(&dentry->d_lru); | |
421 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
422 | this_cpu_dec(nr_dentry_unused); | |
423 | } | |
424 | ||
425 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
426 | { | |
427 | D_FLAG_VERIFY(dentry, 0); | |
428 | list_add(&dentry->d_lru, list); | |
429 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
430 | this_cpu_inc(nr_dentry_unused); | |
431 | } | |
432 | ||
433 | /* | |
434 | * These can only be called under the global LRU lock, ie during the | |
435 | * callback for freeing the LRU list. "isolate" removes it from the | |
436 | * LRU lists entirely, while shrink_move moves it to the indicated | |
437 | * private list. | |
438 | */ | |
3f97b163 | 439 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
440 | { |
441 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
442 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
443 | this_cpu_dec(nr_dentry_unused); | |
3f97b163 | 444 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
445 | } |
446 | ||
3f97b163 VD |
447 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
448 | struct list_head *list) | |
89dc77bc LT |
449 | { |
450 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
451 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
3f97b163 | 452 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
453 | } |
454 | ||
da3bbdd4 | 455 | /* |
f6041567 | 456 | * dentry_lru_(add|del)_list) must be called with d_lock held. |
da3bbdd4 KM |
457 | */ |
458 | static void dentry_lru_add(struct dentry *dentry) | |
459 | { | |
89dc77bc LT |
460 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
461 | d_lru_add(dentry); | |
da3bbdd4 KM |
462 | } |
463 | ||
789680d1 NP |
464 | /** |
465 | * d_drop - drop a dentry | |
466 | * @dentry: dentry to drop | |
467 | * | |
468 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
469 | * be found through a VFS lookup any more. Note that this is different from | |
470 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
471 | * possible, giving a successful _negative_ lookup, while d_drop will | |
472 | * just make the cache lookup fail. | |
473 | * | |
474 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
475 | * reason (NFS timeouts or autofs deletes). | |
476 | * | |
477 | * __d_drop requires dentry->d_lock. | |
478 | */ | |
479 | void __d_drop(struct dentry *dentry) | |
480 | { | |
dea3667b | 481 | if (!d_unhashed(dentry)) { |
b61625d2 | 482 | struct hlist_bl_head *b; |
7632e465 BF |
483 | /* |
484 | * Hashed dentries are normally on the dentry hashtable, | |
485 | * with the exception of those newly allocated by | |
486 | * d_obtain_alias, which are always IS_ROOT: | |
487 | */ | |
488 | if (unlikely(IS_ROOT(dentry))) | |
b61625d2 AV |
489 | b = &dentry->d_sb->s_anon; |
490 | else | |
491 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
492 | ||
493 | hlist_bl_lock(b); | |
494 | __hlist_bl_del(&dentry->d_hash); | |
495 | dentry->d_hash.pprev = NULL; | |
496 | hlist_bl_unlock(b); | |
a7c6f571 | 497 | dentry_rcuwalk_invalidate(dentry); |
789680d1 NP |
498 | } |
499 | } | |
500 | EXPORT_SYMBOL(__d_drop); | |
501 | ||
502 | void d_drop(struct dentry *dentry) | |
503 | { | |
789680d1 NP |
504 | spin_lock(&dentry->d_lock); |
505 | __d_drop(dentry); | |
506 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
507 | } |
508 | EXPORT_SYMBOL(d_drop); | |
509 | ||
e55fd011 | 510 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 511 | { |
41edf278 AV |
512 | struct dentry *parent = NULL; |
513 | bool can_free = true; | |
41edf278 | 514 | if (!IS_ROOT(dentry)) |
77812a1e | 515 | parent = dentry->d_parent; |
31e6b01f | 516 | |
0d98439e LT |
517 | /* |
518 | * The dentry is now unrecoverably dead to the world. | |
519 | */ | |
520 | lockref_mark_dead(&dentry->d_lockref); | |
521 | ||
f0023bc6 | 522 | /* |
f0023bc6 SW |
523 | * inform the fs via d_prune that this dentry is about to be |
524 | * unhashed and destroyed. | |
525 | */ | |
29266201 | 526 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
527 | dentry->d_op->d_prune(dentry); |
528 | ||
01b60351 AV |
529 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
530 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
531 | d_lru_del(dentry); | |
01b60351 | 532 | } |
77812a1e NP |
533 | /* if it was on the hash then remove it */ |
534 | __d_drop(dentry); | |
ca5358ef | 535 | __list_del_entry(&dentry->d_child); |
03b3b889 AV |
536 | /* |
537 | * Inform d_walk() that we are no longer attached to the | |
538 | * dentry tree | |
539 | */ | |
540 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
541 | if (parent) | |
542 | spin_unlock(&parent->d_lock); | |
543 | dentry_iput(dentry); | |
544 | /* | |
545 | * dentry_iput drops the locks, at which point nobody (except | |
546 | * transient RCU lookups) can reach this dentry. | |
547 | */ | |
360f5479 | 548 | BUG_ON(dentry->d_lockref.count > 0); |
03b3b889 AV |
549 | this_cpu_dec(nr_dentry); |
550 | if (dentry->d_op && dentry->d_op->d_release) | |
551 | dentry->d_op->d_release(dentry); | |
552 | ||
41edf278 AV |
553 | spin_lock(&dentry->d_lock); |
554 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
555 | dentry->d_flags |= DCACHE_MAY_FREE; | |
556 | can_free = false; | |
557 | } | |
558 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
559 | if (likely(can_free)) |
560 | dentry_free(dentry); | |
e55fd011 AV |
561 | } |
562 | ||
563 | /* | |
564 | * Finish off a dentry we've decided to kill. | |
565 | * dentry->d_lock must be held, returns with it unlocked. | |
566 | * If ref is non-zero, then decrement the refcount too. | |
567 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
568 | */ | |
8cbf74da | 569 | static struct dentry *dentry_kill(struct dentry *dentry) |
e55fd011 AV |
570 | __releases(dentry->d_lock) |
571 | { | |
572 | struct inode *inode = dentry->d_inode; | |
573 | struct dentry *parent = NULL; | |
574 | ||
575 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
576 | goto failed; | |
577 | ||
578 | if (!IS_ROOT(dentry)) { | |
579 | parent = dentry->d_parent; | |
580 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
581 | if (inode) | |
582 | spin_unlock(&inode->i_lock); | |
583 | goto failed; | |
584 | } | |
585 | } | |
586 | ||
587 | __dentry_kill(dentry); | |
03b3b889 | 588 | return parent; |
e55fd011 AV |
589 | |
590 | failed: | |
8cbf74da AV |
591 | spin_unlock(&dentry->d_lock); |
592 | cpu_relax(); | |
e55fd011 | 593 | return dentry; /* try again with same dentry */ |
77812a1e NP |
594 | } |
595 | ||
046b961b AV |
596 | static inline struct dentry *lock_parent(struct dentry *dentry) |
597 | { | |
598 | struct dentry *parent = dentry->d_parent; | |
599 | if (IS_ROOT(dentry)) | |
600 | return NULL; | |
360f5479 | 601 | if (unlikely(dentry->d_lockref.count < 0)) |
c2338f2d | 602 | return NULL; |
046b961b AV |
603 | if (likely(spin_trylock(&parent->d_lock))) |
604 | return parent; | |
046b961b | 605 | rcu_read_lock(); |
c2338f2d | 606 | spin_unlock(&dentry->d_lock); |
046b961b AV |
607 | again: |
608 | parent = ACCESS_ONCE(dentry->d_parent); | |
609 | spin_lock(&parent->d_lock); | |
610 | /* | |
611 | * We can't blindly lock dentry until we are sure | |
612 | * that we won't violate the locking order. | |
613 | * Any changes of dentry->d_parent must have | |
614 | * been done with parent->d_lock held, so | |
615 | * spin_lock() above is enough of a barrier | |
616 | * for checking if it's still our child. | |
617 | */ | |
618 | if (unlikely(parent != dentry->d_parent)) { | |
619 | spin_unlock(&parent->d_lock); | |
620 | goto again; | |
621 | } | |
622 | rcu_read_unlock(); | |
623 | if (parent != dentry) | |
9f12600f | 624 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
046b961b AV |
625 | else |
626 | parent = NULL; | |
627 | return parent; | |
628 | } | |
629 | ||
360f5479 LT |
630 | /* |
631 | * Try to do a lockless dput(), and return whether that was successful. | |
632 | * | |
633 | * If unsuccessful, we return false, having already taken the dentry lock. | |
634 | * | |
635 | * The caller needs to hold the RCU read lock, so that the dentry is | |
636 | * guaranteed to stay around even if the refcount goes down to zero! | |
637 | */ | |
638 | static inline bool fast_dput(struct dentry *dentry) | |
639 | { | |
640 | int ret; | |
641 | unsigned int d_flags; | |
642 | ||
643 | /* | |
644 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 645 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
646 | */ |
647 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
648 | return lockref_put_or_lock(&dentry->d_lockref); | |
649 | ||
650 | /* | |
651 | * .. otherwise, we can try to just decrement the | |
652 | * lockref optimistically. | |
653 | */ | |
654 | ret = lockref_put_return(&dentry->d_lockref); | |
655 | ||
656 | /* | |
657 | * If the lockref_put_return() failed due to the lock being held | |
658 | * by somebody else, the fast path has failed. We will need to | |
659 | * get the lock, and then check the count again. | |
660 | */ | |
661 | if (unlikely(ret < 0)) { | |
662 | spin_lock(&dentry->d_lock); | |
663 | if (dentry->d_lockref.count > 1) { | |
664 | dentry->d_lockref.count--; | |
665 | spin_unlock(&dentry->d_lock); | |
666 | return 1; | |
667 | } | |
668 | return 0; | |
669 | } | |
670 | ||
671 | /* | |
672 | * If we weren't the last ref, we're done. | |
673 | */ | |
674 | if (ret) | |
675 | return 1; | |
676 | ||
677 | /* | |
678 | * Careful, careful. The reference count went down | |
679 | * to zero, but we don't hold the dentry lock, so | |
680 | * somebody else could get it again, and do another | |
681 | * dput(), and we need to not race with that. | |
682 | * | |
683 | * However, there is a very special and common case | |
684 | * where we don't care, because there is nothing to | |
685 | * do: the dentry is still hashed, it does not have | |
686 | * a 'delete' op, and it's referenced and already on | |
687 | * the LRU list. | |
688 | * | |
689 | * NOTE! Since we aren't locked, these values are | |
690 | * not "stable". However, it is sufficient that at | |
691 | * some point after we dropped the reference the | |
692 | * dentry was hashed and the flags had the proper | |
693 | * value. Other dentry users may have re-gotten | |
694 | * a reference to the dentry and change that, but | |
695 | * our work is done - we can leave the dentry | |
696 | * around with a zero refcount. | |
697 | */ | |
698 | smp_rmb(); | |
699 | d_flags = ACCESS_ONCE(dentry->d_flags); | |
75a6f82a | 700 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; |
360f5479 LT |
701 | |
702 | /* Nothing to do? Dropping the reference was all we needed? */ | |
703 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
704 | return 1; | |
705 | ||
706 | /* | |
707 | * Not the fast normal case? Get the lock. We've already decremented | |
708 | * the refcount, but we'll need to re-check the situation after | |
709 | * getting the lock. | |
710 | */ | |
711 | spin_lock(&dentry->d_lock); | |
712 | ||
713 | /* | |
714 | * Did somebody else grab a reference to it in the meantime, and | |
715 | * we're no longer the last user after all? Alternatively, somebody | |
716 | * else could have killed it and marked it dead. Either way, we | |
717 | * don't need to do anything else. | |
718 | */ | |
719 | if (dentry->d_lockref.count) { | |
720 | spin_unlock(&dentry->d_lock); | |
721 | return 1; | |
722 | } | |
723 | ||
724 | /* | |
725 | * Re-get the reference we optimistically dropped. We hold the | |
726 | * lock, and we just tested that it was zero, so we can just | |
727 | * set it to 1. | |
728 | */ | |
729 | dentry->d_lockref.count = 1; | |
730 | return 0; | |
731 | } | |
732 | ||
733 | ||
1da177e4 LT |
734 | /* |
735 | * This is dput | |
736 | * | |
737 | * This is complicated by the fact that we do not want to put | |
738 | * dentries that are no longer on any hash chain on the unused | |
739 | * list: we'd much rather just get rid of them immediately. | |
740 | * | |
741 | * However, that implies that we have to traverse the dentry | |
742 | * tree upwards to the parents which might _also_ now be | |
743 | * scheduled for deletion (it may have been only waiting for | |
744 | * its last child to go away). | |
745 | * | |
746 | * This tail recursion is done by hand as we don't want to depend | |
747 | * on the compiler to always get this right (gcc generally doesn't). | |
748 | * Real recursion would eat up our stack space. | |
749 | */ | |
750 | ||
751 | /* | |
752 | * dput - release a dentry | |
753 | * @dentry: dentry to release | |
754 | * | |
755 | * Release a dentry. This will drop the usage count and if appropriate | |
756 | * call the dentry unlink method as well as removing it from the queues and | |
757 | * releasing its resources. If the parent dentries were scheduled for release | |
758 | * they too may now get deleted. | |
1da177e4 | 759 | */ |
1da177e4 LT |
760 | void dput(struct dentry *dentry) |
761 | { | |
8aab6a27 | 762 | if (unlikely(!dentry)) |
1da177e4 LT |
763 | return; |
764 | ||
765 | repeat: | |
360f5479 LT |
766 | rcu_read_lock(); |
767 | if (likely(fast_dput(dentry))) { | |
768 | rcu_read_unlock(); | |
1da177e4 | 769 | return; |
360f5479 LT |
770 | } |
771 | ||
772 | /* Slow case: now with the dentry lock held */ | |
773 | rcu_read_unlock(); | |
1da177e4 | 774 | |
85c7f810 AV |
775 | WARN_ON(d_in_lookup(dentry)); |
776 | ||
8aab6a27 LT |
777 | /* Unreachable? Get rid of it */ |
778 | if (unlikely(d_unhashed(dentry))) | |
779 | goto kill_it; | |
780 | ||
75a6f82a AV |
781 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
782 | goto kill_it; | |
783 | ||
8aab6a27 | 784 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { |
1da177e4 | 785 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 786 | goto kill_it; |
1da177e4 | 787 | } |
265ac902 | 788 | |
358eec18 LT |
789 | if (!(dentry->d_flags & DCACHE_REFERENCED)) |
790 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 791 | dentry_lru_add(dentry); |
265ac902 | 792 | |
98474236 | 793 | dentry->d_lockref.count--; |
61f3dee4 | 794 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
795 | return; |
796 | ||
d52b9086 | 797 | kill_it: |
8cbf74da | 798 | dentry = dentry_kill(dentry); |
d52b9086 MS |
799 | if (dentry) |
800 | goto repeat; | |
1da177e4 | 801 | } |
ec4f8605 | 802 | EXPORT_SYMBOL(dput); |
1da177e4 | 803 | |
1da177e4 | 804 | |
b5c84bf6 | 805 | /* This must be called with d_lock held */ |
dc0474be | 806 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 807 | { |
98474236 | 808 | dentry->d_lockref.count++; |
23044507 NP |
809 | } |
810 | ||
dc0474be | 811 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 812 | { |
98474236 | 813 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
814 | } |
815 | ||
b7ab39f6 NP |
816 | struct dentry *dget_parent(struct dentry *dentry) |
817 | { | |
df3d0bbc | 818 | int gotref; |
b7ab39f6 NP |
819 | struct dentry *ret; |
820 | ||
df3d0bbc WL |
821 | /* |
822 | * Do optimistic parent lookup without any | |
823 | * locking. | |
824 | */ | |
825 | rcu_read_lock(); | |
826 | ret = ACCESS_ONCE(dentry->d_parent); | |
827 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
828 | rcu_read_unlock(); | |
829 | if (likely(gotref)) { | |
830 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
831 | return ret; | |
832 | dput(ret); | |
833 | } | |
834 | ||
b7ab39f6 | 835 | repeat: |
a734eb45 NP |
836 | /* |
837 | * Don't need rcu_dereference because we re-check it was correct under | |
838 | * the lock. | |
839 | */ | |
840 | rcu_read_lock(); | |
b7ab39f6 | 841 | ret = dentry->d_parent; |
a734eb45 NP |
842 | spin_lock(&ret->d_lock); |
843 | if (unlikely(ret != dentry->d_parent)) { | |
844 | spin_unlock(&ret->d_lock); | |
845 | rcu_read_unlock(); | |
b7ab39f6 NP |
846 | goto repeat; |
847 | } | |
a734eb45 | 848 | rcu_read_unlock(); |
98474236 WL |
849 | BUG_ON(!ret->d_lockref.count); |
850 | ret->d_lockref.count++; | |
b7ab39f6 | 851 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
852 | return ret; |
853 | } | |
854 | EXPORT_SYMBOL(dget_parent); | |
855 | ||
1da177e4 LT |
856 | /** |
857 | * d_find_alias - grab a hashed alias of inode | |
858 | * @inode: inode in question | |
1da177e4 LT |
859 | * |
860 | * If inode has a hashed alias, or is a directory and has any alias, | |
861 | * acquire the reference to alias and return it. Otherwise return NULL. | |
862 | * Notice that if inode is a directory there can be only one alias and | |
863 | * it can be unhashed only if it has no children, or if it is the root | |
3ccb354d EB |
864 | * of a filesystem, or if the directory was renamed and d_revalidate |
865 | * was the first vfs operation to notice. | |
1da177e4 | 866 | * |
21c0d8fd | 867 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
52ed46f0 | 868 | * any other hashed alias over that one. |
1da177e4 | 869 | */ |
52ed46f0 | 870 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 871 | { |
da502956 | 872 | struct dentry *alias, *discon_alias; |
1da177e4 | 873 | |
da502956 NP |
874 | again: |
875 | discon_alias = NULL; | |
946e51f2 | 876 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 877 | spin_lock(&alias->d_lock); |
1da177e4 | 878 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 879 | if (IS_ROOT(alias) && |
da502956 | 880 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 881 | discon_alias = alias; |
52ed46f0 | 882 | } else { |
dc0474be | 883 | __dget_dlock(alias); |
da502956 NP |
884 | spin_unlock(&alias->d_lock); |
885 | return alias; | |
886 | } | |
887 | } | |
888 | spin_unlock(&alias->d_lock); | |
889 | } | |
890 | if (discon_alias) { | |
891 | alias = discon_alias; | |
892 | spin_lock(&alias->d_lock); | |
893 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
8d80d7da BF |
894 | __dget_dlock(alias); |
895 | spin_unlock(&alias->d_lock); | |
896 | return alias; | |
1da177e4 | 897 | } |
da502956 NP |
898 | spin_unlock(&alias->d_lock); |
899 | goto again; | |
1da177e4 | 900 | } |
da502956 | 901 | return NULL; |
1da177e4 LT |
902 | } |
903 | ||
da502956 | 904 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 905 | { |
214fda1f DH |
906 | struct dentry *de = NULL; |
907 | ||
b3d9b7a3 | 908 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 909 | spin_lock(&inode->i_lock); |
52ed46f0 | 910 | de = __d_find_alias(inode); |
873feea0 | 911 | spin_unlock(&inode->i_lock); |
214fda1f | 912 | } |
1da177e4 LT |
913 | return de; |
914 | } | |
ec4f8605 | 915 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
916 | |
917 | /* | |
918 | * Try to kill dentries associated with this inode. | |
919 | * WARNING: you must own a reference to inode. | |
920 | */ | |
921 | void d_prune_aliases(struct inode *inode) | |
922 | { | |
0cdca3f9 | 923 | struct dentry *dentry; |
1da177e4 | 924 | restart: |
873feea0 | 925 | spin_lock(&inode->i_lock); |
946e51f2 | 926 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 927 | spin_lock(&dentry->d_lock); |
98474236 | 928 | if (!dentry->d_lockref.count) { |
29355c39 AV |
929 | struct dentry *parent = lock_parent(dentry); |
930 | if (likely(!dentry->d_lockref.count)) { | |
931 | __dentry_kill(dentry); | |
4a7795d3 | 932 | dput(parent); |
29355c39 AV |
933 | goto restart; |
934 | } | |
935 | if (parent) | |
936 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
937 | } |
938 | spin_unlock(&dentry->d_lock); | |
939 | } | |
873feea0 | 940 | spin_unlock(&inode->i_lock); |
1da177e4 | 941 | } |
ec4f8605 | 942 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 943 | |
3049cfe2 | 944 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 945 | { |
5c47e6d0 | 946 | struct dentry *dentry, *parent; |
da3bbdd4 | 947 | |
60942f2f | 948 | while (!list_empty(list)) { |
ff2fde99 | 949 | struct inode *inode; |
60942f2f | 950 | dentry = list_entry(list->prev, struct dentry, d_lru); |
ec33679d | 951 | spin_lock(&dentry->d_lock); |
046b961b AV |
952 | parent = lock_parent(dentry); |
953 | ||
dd1f6b2e DC |
954 | /* |
955 | * The dispose list is isolated and dentries are not accounted | |
956 | * to the LRU here, so we can simply remove it from the list | |
957 | * here regardless of whether it is referenced or not. | |
958 | */ | |
89dc77bc | 959 | d_shrink_del(dentry); |
dd1f6b2e | 960 | |
1da177e4 LT |
961 | /* |
962 | * We found an inuse dentry which was not removed from | |
dd1f6b2e | 963 | * the LRU because of laziness during lookup. Do not free it. |
1da177e4 | 964 | */ |
360f5479 | 965 | if (dentry->d_lockref.count > 0) { |
da3bbdd4 | 966 | spin_unlock(&dentry->d_lock); |
046b961b AV |
967 | if (parent) |
968 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
969 | continue; |
970 | } | |
77812a1e | 971 | |
64fd72e0 AV |
972 | |
973 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) { | |
974 | bool can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
975 | spin_unlock(&dentry->d_lock); | |
046b961b AV |
976 | if (parent) |
977 | spin_unlock(&parent->d_lock); | |
64fd72e0 AV |
978 | if (can_free) |
979 | dentry_free(dentry); | |
980 | continue; | |
981 | } | |
982 | ||
ff2fde99 AV |
983 | inode = dentry->d_inode; |
984 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
89dc77bc | 985 | d_shrink_add(dentry, list); |
dd1f6b2e | 986 | spin_unlock(&dentry->d_lock); |
046b961b AV |
987 | if (parent) |
988 | spin_unlock(&parent->d_lock); | |
5c47e6d0 | 989 | continue; |
dd1f6b2e | 990 | } |
ff2fde99 | 991 | |
ff2fde99 | 992 | __dentry_kill(dentry); |
046b961b | 993 | |
5c47e6d0 AV |
994 | /* |
995 | * We need to prune ancestors too. This is necessary to prevent | |
996 | * quadratic behavior of shrink_dcache_parent(), but is also | |
997 | * expected to be beneficial in reducing dentry cache | |
998 | * fragmentation. | |
999 | */ | |
1000 | dentry = parent; | |
b2b80195 AV |
1001 | while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) { |
1002 | parent = lock_parent(dentry); | |
1003 | if (dentry->d_lockref.count != 1) { | |
1004 | dentry->d_lockref.count--; | |
1005 | spin_unlock(&dentry->d_lock); | |
1006 | if (parent) | |
1007 | spin_unlock(&parent->d_lock); | |
1008 | break; | |
1009 | } | |
1010 | inode = dentry->d_inode; /* can't be NULL */ | |
1011 | if (unlikely(!spin_trylock(&inode->i_lock))) { | |
1012 | spin_unlock(&dentry->d_lock); | |
1013 | if (parent) | |
1014 | spin_unlock(&parent->d_lock); | |
1015 | cpu_relax(); | |
1016 | continue; | |
1017 | } | |
1018 | __dentry_kill(dentry); | |
1019 | dentry = parent; | |
1020 | } | |
da3bbdd4 | 1021 | } |
3049cfe2 CH |
1022 | } |
1023 | ||
3f97b163 VD |
1024 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1025 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1026 | { |
1027 | struct list_head *freeable = arg; | |
1028 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1029 | ||
1030 | ||
1031 | /* | |
1032 | * we are inverting the lru lock/dentry->d_lock here, | |
1033 | * so use a trylock. If we fail to get the lock, just skip | |
1034 | * it | |
1035 | */ | |
1036 | if (!spin_trylock(&dentry->d_lock)) | |
1037 | return LRU_SKIP; | |
1038 | ||
1039 | /* | |
1040 | * Referenced dentries are still in use. If they have active | |
1041 | * counts, just remove them from the LRU. Otherwise give them | |
1042 | * another pass through the LRU. | |
1043 | */ | |
1044 | if (dentry->d_lockref.count) { | |
3f97b163 | 1045 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1046 | spin_unlock(&dentry->d_lock); |
1047 | return LRU_REMOVED; | |
1048 | } | |
1049 | ||
1050 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1051 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1052 | spin_unlock(&dentry->d_lock); | |
1053 | ||
1054 | /* | |
1055 | * The list move itself will be made by the common LRU code. At | |
1056 | * this point, we've dropped the dentry->d_lock but keep the | |
1057 | * lru lock. This is safe to do, since every list movement is | |
1058 | * protected by the lru lock even if both locks are held. | |
1059 | * | |
1060 | * This is guaranteed by the fact that all LRU management | |
1061 | * functions are intermediated by the LRU API calls like | |
1062 | * list_lru_add and list_lru_del. List movement in this file | |
1063 | * only ever occur through this functions or through callbacks | |
1064 | * like this one, that are called from the LRU API. | |
1065 | * | |
1066 | * The only exceptions to this are functions like | |
1067 | * shrink_dentry_list, and code that first checks for the | |
1068 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1069 | * operating only with stack provided lists after they are | |
1070 | * properly isolated from the main list. It is thus, always a | |
1071 | * local access. | |
1072 | */ | |
1073 | return LRU_ROTATE; | |
1074 | } | |
1075 | ||
3f97b163 | 1076 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1077 | spin_unlock(&dentry->d_lock); |
1078 | ||
1079 | return LRU_REMOVED; | |
1080 | } | |
1081 | ||
3049cfe2 | 1082 | /** |
b48f03b3 DC |
1083 | * prune_dcache_sb - shrink the dcache |
1084 | * @sb: superblock | |
503c358c | 1085 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1086 | * |
503c358c VD |
1087 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1088 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1089 | * function. |
3049cfe2 | 1090 | * |
b48f03b3 DC |
1091 | * This function may fail to free any resources if all the dentries are in |
1092 | * use. | |
3049cfe2 | 1093 | */ |
503c358c | 1094 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1095 | { |
f6041567 DC |
1096 | LIST_HEAD(dispose); |
1097 | long freed; | |
3049cfe2 | 1098 | |
503c358c VD |
1099 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1100 | dentry_lru_isolate, &dispose); | |
f6041567 | 1101 | shrink_dentry_list(&dispose); |
0a234c6d | 1102 | return freed; |
da3bbdd4 | 1103 | } |
23044507 | 1104 | |
4e717f5c | 1105 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1106 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1107 | { |
4e717f5c GC |
1108 | struct list_head *freeable = arg; |
1109 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1110 | |
4e717f5c GC |
1111 | /* |
1112 | * we are inverting the lru lock/dentry->d_lock here, | |
1113 | * so use a trylock. If we fail to get the lock, just skip | |
1114 | * it | |
1115 | */ | |
1116 | if (!spin_trylock(&dentry->d_lock)) | |
1117 | return LRU_SKIP; | |
1118 | ||
3f97b163 | 1119 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1120 | spin_unlock(&dentry->d_lock); |
ec33679d | 1121 | |
4e717f5c | 1122 | return LRU_REMOVED; |
da3bbdd4 KM |
1123 | } |
1124 | ||
4e717f5c | 1125 | |
1da177e4 LT |
1126 | /** |
1127 | * shrink_dcache_sb - shrink dcache for a superblock | |
1128 | * @sb: superblock | |
1129 | * | |
3049cfe2 CH |
1130 | * Shrink the dcache for the specified super block. This is used to free |
1131 | * the dcache before unmounting a file system. | |
1da177e4 | 1132 | */ |
3049cfe2 | 1133 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1134 | { |
4e717f5c GC |
1135 | long freed; |
1136 | ||
1137 | do { | |
1138 | LIST_HEAD(dispose); | |
1139 | ||
1140 | freed = list_lru_walk(&sb->s_dentry_lru, | |
1141 | dentry_lru_isolate_shrink, &dispose, UINT_MAX); | |
3049cfe2 | 1142 | |
4e717f5c GC |
1143 | this_cpu_sub(nr_dentry_unused, freed); |
1144 | shrink_dentry_list(&dispose); | |
1145 | } while (freed > 0); | |
1da177e4 | 1146 | } |
ec4f8605 | 1147 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1148 | |
db14fc3a MS |
1149 | /** |
1150 | * enum d_walk_ret - action to talke during tree walk | |
1151 | * @D_WALK_CONTINUE: contrinue walk | |
1152 | * @D_WALK_QUIT: quit walk | |
1153 | * @D_WALK_NORETRY: quit when retry is needed | |
1154 | * @D_WALK_SKIP: skip this dentry and its children | |
1155 | */ | |
1156 | enum d_walk_ret { | |
1157 | D_WALK_CONTINUE, | |
1158 | D_WALK_QUIT, | |
1159 | D_WALK_NORETRY, | |
1160 | D_WALK_SKIP, | |
1161 | }; | |
c826cb7d | 1162 | |
1da177e4 | 1163 | /** |
db14fc3a MS |
1164 | * d_walk - walk the dentry tree |
1165 | * @parent: start of walk | |
1166 | * @data: data passed to @enter() and @finish() | |
1167 | * @enter: callback when first entering the dentry | |
1168 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1169 | * |
db14fc3a | 1170 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1171 | */ |
db14fc3a MS |
1172 | static void d_walk(struct dentry *parent, void *data, |
1173 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1174 | void (*finish)(void *)) | |
1da177e4 | 1175 | { |
949854d0 | 1176 | struct dentry *this_parent; |
1da177e4 | 1177 | struct list_head *next; |
48f5ec21 | 1178 | unsigned seq = 0; |
db14fc3a MS |
1179 | enum d_walk_ret ret; |
1180 | bool retry = true; | |
949854d0 | 1181 | |
58db63d0 | 1182 | again: |
48f5ec21 | 1183 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1184 | this_parent = parent; |
2fd6b7f5 | 1185 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1186 | |
1187 | ret = enter(data, this_parent); | |
1188 | switch (ret) { | |
1189 | case D_WALK_CONTINUE: | |
1190 | break; | |
1191 | case D_WALK_QUIT: | |
1192 | case D_WALK_SKIP: | |
1193 | goto out_unlock; | |
1194 | case D_WALK_NORETRY: | |
1195 | retry = false; | |
1196 | break; | |
1197 | } | |
1da177e4 LT |
1198 | repeat: |
1199 | next = this_parent->d_subdirs.next; | |
1200 | resume: | |
1201 | while (next != &this_parent->d_subdirs) { | |
1202 | struct list_head *tmp = next; | |
946e51f2 | 1203 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1204 | next = tmp->next; |
2fd6b7f5 NP |
1205 | |
1206 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1207 | |
1208 | ret = enter(data, dentry); | |
1209 | switch (ret) { | |
1210 | case D_WALK_CONTINUE: | |
1211 | break; | |
1212 | case D_WALK_QUIT: | |
2fd6b7f5 | 1213 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1214 | goto out_unlock; |
1215 | case D_WALK_NORETRY: | |
1216 | retry = false; | |
1217 | break; | |
1218 | case D_WALK_SKIP: | |
1219 | spin_unlock(&dentry->d_lock); | |
1220 | continue; | |
2fd6b7f5 | 1221 | } |
db14fc3a | 1222 | |
1da177e4 | 1223 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1224 | spin_unlock(&this_parent->d_lock); |
1225 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1226 | this_parent = dentry; |
2fd6b7f5 | 1227 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1228 | goto repeat; |
1229 | } | |
2fd6b7f5 | 1230 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1231 | } |
1232 | /* | |
1233 | * All done at this level ... ascend and resume the search. | |
1234 | */ | |
ca5358ef AV |
1235 | rcu_read_lock(); |
1236 | ascend: | |
1da177e4 | 1237 | if (this_parent != parent) { |
c826cb7d | 1238 | struct dentry *child = this_parent; |
31dec132 AV |
1239 | this_parent = child->d_parent; |
1240 | ||
31dec132 AV |
1241 | spin_unlock(&child->d_lock); |
1242 | spin_lock(&this_parent->d_lock); | |
1243 | ||
ca5358ef AV |
1244 | /* might go back up the wrong parent if we have had a rename. */ |
1245 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1246 | goto rename_retry; |
2159184e AV |
1247 | /* go into the first sibling still alive */ |
1248 | do { | |
1249 | next = child->d_child.next; | |
ca5358ef AV |
1250 | if (next == &this_parent->d_subdirs) |
1251 | goto ascend; | |
1252 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1253 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1254 | rcu_read_unlock(); |
1da177e4 LT |
1255 | goto resume; |
1256 | } | |
ca5358ef | 1257 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1258 | goto rename_retry; |
ca5358ef | 1259 | rcu_read_unlock(); |
db14fc3a MS |
1260 | if (finish) |
1261 | finish(data); | |
1262 | ||
1263 | out_unlock: | |
1264 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1265 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1266 | return; |
58db63d0 NP |
1267 | |
1268 | rename_retry: | |
ca5358ef AV |
1269 | spin_unlock(&this_parent->d_lock); |
1270 | rcu_read_unlock(); | |
1271 | BUG_ON(seq & 1); | |
db14fc3a MS |
1272 | if (!retry) |
1273 | return; | |
48f5ec21 | 1274 | seq = 1; |
58db63d0 | 1275 | goto again; |
1da177e4 | 1276 | } |
db14fc3a MS |
1277 | |
1278 | /* | |
1279 | * Search for at least 1 mount point in the dentry's subdirs. | |
1280 | * We descend to the next level whenever the d_subdirs | |
1281 | * list is non-empty and continue searching. | |
1282 | */ | |
1283 | ||
db14fc3a MS |
1284 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) |
1285 | { | |
1286 | int *ret = data; | |
1287 | if (d_mountpoint(dentry)) { | |
1288 | *ret = 1; | |
1289 | return D_WALK_QUIT; | |
1290 | } | |
1291 | return D_WALK_CONTINUE; | |
1292 | } | |
1293 | ||
69c88dc7 RD |
1294 | /** |
1295 | * have_submounts - check for mounts over a dentry | |
1296 | * @parent: dentry to check. | |
1297 | * | |
1298 | * Return true if the parent or its subdirectories contain | |
1299 | * a mount point | |
1300 | */ | |
db14fc3a MS |
1301 | int have_submounts(struct dentry *parent) |
1302 | { | |
1303 | int ret = 0; | |
1304 | ||
1305 | d_walk(parent, &ret, check_mount, NULL); | |
1306 | ||
1307 | return ret; | |
1308 | } | |
ec4f8605 | 1309 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1310 | |
eed81007 MS |
1311 | /* |
1312 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1313 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1314 | * subtree can become unreachable). | |
1315 | * | |
1ffe46d1 | 1316 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1317 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1318 | */ | |
1319 | int d_set_mounted(struct dentry *dentry) | |
1320 | { | |
1321 | struct dentry *p; | |
1322 | int ret = -ENOENT; | |
1323 | write_seqlock(&rename_lock); | |
1324 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1325 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1326 | spin_lock(&p->d_lock); |
1327 | if (unlikely(d_unhashed(p))) { | |
1328 | spin_unlock(&p->d_lock); | |
1329 | goto out; | |
1330 | } | |
1331 | spin_unlock(&p->d_lock); | |
1332 | } | |
1333 | spin_lock(&dentry->d_lock); | |
1334 | if (!d_unlinked(dentry)) { | |
1335 | dentry->d_flags |= DCACHE_MOUNTED; | |
1336 | ret = 0; | |
1337 | } | |
1338 | spin_unlock(&dentry->d_lock); | |
1339 | out: | |
1340 | write_sequnlock(&rename_lock); | |
1341 | return ret; | |
1342 | } | |
1343 | ||
1da177e4 | 1344 | /* |
fd517909 | 1345 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1346 | * and move any unused dentries to the end of the unused |
1347 | * list for prune_dcache(). We descend to the next level | |
1348 | * whenever the d_subdirs list is non-empty and continue | |
1349 | * searching. | |
1350 | * | |
1351 | * It returns zero iff there are no unused children, | |
1352 | * otherwise it returns the number of children moved to | |
1353 | * the end of the unused list. This may not be the total | |
1354 | * number of unused children, because select_parent can | |
1355 | * drop the lock and return early due to latency | |
1356 | * constraints. | |
1357 | */ | |
1da177e4 | 1358 | |
db14fc3a MS |
1359 | struct select_data { |
1360 | struct dentry *start; | |
1361 | struct list_head dispose; | |
1362 | int found; | |
1363 | }; | |
23044507 | 1364 | |
db14fc3a MS |
1365 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1366 | { | |
1367 | struct select_data *data = _data; | |
1368 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1369 | |
db14fc3a MS |
1370 | if (data->start == dentry) |
1371 | goto out; | |
2fd6b7f5 | 1372 | |
fe91522a | 1373 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1374 | data->found++; |
fe91522a AV |
1375 | } else { |
1376 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1377 | d_lru_del(dentry); | |
1378 | if (!dentry->d_lockref.count) { | |
1379 | d_shrink_add(dentry, &data->dispose); | |
1380 | data->found++; | |
1381 | } | |
1da177e4 | 1382 | } |
db14fc3a MS |
1383 | /* |
1384 | * We can return to the caller if we have found some (this | |
1385 | * ensures forward progress). We'll be coming back to find | |
1386 | * the rest. | |
1387 | */ | |
fe91522a AV |
1388 | if (!list_empty(&data->dispose)) |
1389 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1390 | out: |
db14fc3a | 1391 | return ret; |
1da177e4 LT |
1392 | } |
1393 | ||
1394 | /** | |
1395 | * shrink_dcache_parent - prune dcache | |
1396 | * @parent: parent of entries to prune | |
1397 | * | |
1398 | * Prune the dcache to remove unused children of the parent dentry. | |
1399 | */ | |
db14fc3a | 1400 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1401 | { |
db14fc3a MS |
1402 | for (;;) { |
1403 | struct select_data data; | |
1da177e4 | 1404 | |
db14fc3a MS |
1405 | INIT_LIST_HEAD(&data.dispose); |
1406 | data.start = parent; | |
1407 | data.found = 0; | |
1408 | ||
1409 | d_walk(parent, &data, select_collect, NULL); | |
1410 | if (!data.found) | |
1411 | break; | |
1412 | ||
1413 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1414 | cond_resched(); |
1415 | } | |
1da177e4 | 1416 | } |
ec4f8605 | 1417 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1418 | |
9c8c10e2 | 1419 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1420 | { |
9c8c10e2 AV |
1421 | /* it has busy descendents; complain about those instead */ |
1422 | if (!list_empty(&dentry->d_subdirs)) | |
1423 | return D_WALK_CONTINUE; | |
42c32608 | 1424 | |
9c8c10e2 AV |
1425 | /* root with refcount 1 is fine */ |
1426 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1427 | return D_WALK_CONTINUE; | |
1428 | ||
1429 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1430 | " still in use (%d) [unmount of %s %s]\n", | |
42c32608 AV |
1431 | dentry, |
1432 | dentry->d_inode ? | |
1433 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1434 | dentry, |
42c32608 AV |
1435 | dentry->d_lockref.count, |
1436 | dentry->d_sb->s_type->name, | |
1437 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1438 | WARN_ON(1); |
1439 | return D_WALK_CONTINUE; | |
1440 | } | |
1441 | ||
1442 | static void do_one_tree(struct dentry *dentry) | |
1443 | { | |
1444 | shrink_dcache_parent(dentry); | |
1445 | d_walk(dentry, dentry, umount_check, NULL); | |
1446 | d_drop(dentry); | |
1447 | dput(dentry); | |
42c32608 AV |
1448 | } |
1449 | ||
1450 | /* | |
1451 | * destroy the dentries attached to a superblock on unmounting | |
1452 | */ | |
1453 | void shrink_dcache_for_umount(struct super_block *sb) | |
1454 | { | |
1455 | struct dentry *dentry; | |
1456 | ||
9c8c10e2 | 1457 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1458 | |
1459 | dentry = sb->s_root; | |
1460 | sb->s_root = NULL; | |
9c8c10e2 | 1461 | do_one_tree(dentry); |
42c32608 AV |
1462 | |
1463 | while (!hlist_bl_empty(&sb->s_anon)) { | |
9c8c10e2 AV |
1464 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash)); |
1465 | do_one_tree(dentry); | |
42c32608 AV |
1466 | } |
1467 | } | |
1468 | ||
8ed936b5 EB |
1469 | struct detach_data { |
1470 | struct select_data select; | |
1471 | struct dentry *mountpoint; | |
1472 | }; | |
1473 | static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry) | |
848ac114 | 1474 | { |
8ed936b5 | 1475 | struct detach_data *data = _data; |
848ac114 MS |
1476 | |
1477 | if (d_mountpoint(dentry)) { | |
8ed936b5 EB |
1478 | __dget_dlock(dentry); |
1479 | data->mountpoint = dentry; | |
848ac114 MS |
1480 | return D_WALK_QUIT; |
1481 | } | |
1482 | ||
8ed936b5 | 1483 | return select_collect(&data->select, dentry); |
848ac114 MS |
1484 | } |
1485 | ||
1486 | static void check_and_drop(void *_data) | |
1487 | { | |
8ed936b5 | 1488 | struct detach_data *data = _data; |
848ac114 | 1489 | |
8ed936b5 EB |
1490 | if (!data->mountpoint && !data->select.found) |
1491 | __d_drop(data->select.start); | |
848ac114 MS |
1492 | } |
1493 | ||
1494 | /** | |
1ffe46d1 EB |
1495 | * d_invalidate - detach submounts, prune dcache, and drop |
1496 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1497 | * | |
1ffe46d1 | 1498 | * no dcache lock. |
848ac114 | 1499 | * |
8ed936b5 EB |
1500 | * The final d_drop is done as an atomic operation relative to |
1501 | * rename_lock ensuring there are no races with d_set_mounted. This | |
1502 | * ensures there are no unhashed dentries on the path to a mountpoint. | |
848ac114 | 1503 | */ |
5542aa2f | 1504 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1505 | { |
1ffe46d1 EB |
1506 | /* |
1507 | * If it's already been dropped, return OK. | |
1508 | */ | |
1509 | spin_lock(&dentry->d_lock); | |
1510 | if (d_unhashed(dentry)) { | |
1511 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1512 | return; |
1ffe46d1 EB |
1513 | } |
1514 | spin_unlock(&dentry->d_lock); | |
1515 | ||
848ac114 MS |
1516 | /* Negative dentries can be dropped without further checks */ |
1517 | if (!dentry->d_inode) { | |
1518 | d_drop(dentry); | |
5542aa2f | 1519 | return; |
848ac114 MS |
1520 | } |
1521 | ||
1522 | for (;;) { | |
8ed936b5 | 1523 | struct detach_data data; |
848ac114 | 1524 | |
8ed936b5 EB |
1525 | data.mountpoint = NULL; |
1526 | INIT_LIST_HEAD(&data.select.dispose); | |
1527 | data.select.start = dentry; | |
1528 | data.select.found = 0; | |
1529 | ||
1530 | d_walk(dentry, &data, detach_and_collect, check_and_drop); | |
848ac114 | 1531 | |
8ed936b5 EB |
1532 | if (data.select.found) |
1533 | shrink_dentry_list(&data.select.dispose); | |
848ac114 | 1534 | |
8ed936b5 EB |
1535 | if (data.mountpoint) { |
1536 | detach_mounts(data.mountpoint); | |
1537 | dput(data.mountpoint); | |
1538 | } | |
848ac114 | 1539 | |
8ed936b5 | 1540 | if (!data.mountpoint && !data.select.found) |
848ac114 MS |
1541 | break; |
1542 | ||
1543 | cond_resched(); | |
1544 | } | |
848ac114 | 1545 | } |
1ffe46d1 | 1546 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1547 | |
1da177e4 | 1548 | /** |
a4464dbc AV |
1549 | * __d_alloc - allocate a dcache entry |
1550 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1551 | * @name: qstr of the name |
1552 | * | |
1553 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1554 | * available. On a success the dentry is returned. The name passed in is | |
1555 | * copied and the copy passed in may be reused after this call. | |
1556 | */ | |
1557 | ||
a4464dbc | 1558 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1559 | { |
1560 | struct dentry *dentry; | |
1561 | char *dname; | |
1562 | ||
e12ba74d | 1563 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1564 | if (!dentry) |
1565 | return NULL; | |
1566 | ||
6326c71f LT |
1567 | /* |
1568 | * We guarantee that the inline name is always NUL-terminated. | |
1569 | * This way the memcpy() done by the name switching in rename | |
1570 | * will still always have a NUL at the end, even if we might | |
1571 | * be overwriting an internal NUL character | |
1572 | */ | |
1573 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd AV |
1574 | if (unlikely(!name)) { |
1575 | static const struct qstr anon = QSTR_INIT("/", 1); | |
1576 | name = &anon; | |
1577 | dname = dentry->d_iname; | |
1578 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1579 | size_t size = offsetof(struct external_name, name[1]); |
5d097056 VD |
1580 | struct external_name *p = kmalloc(size + name->len, |
1581 | GFP_KERNEL_ACCOUNT); | |
8d85b484 | 1582 | if (!p) { |
1da177e4 LT |
1583 | kmem_cache_free(dentry_cache, dentry); |
1584 | return NULL; | |
1585 | } | |
8d85b484 AV |
1586 | atomic_set(&p->u.count, 1); |
1587 | dname = p->name; | |
df4c0e36 AR |
1588 | if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS)) |
1589 | kasan_unpoison_shadow(dname, | |
1590 | round_up(name->len + 1, sizeof(unsigned long))); | |
1da177e4 LT |
1591 | } else { |
1592 | dname = dentry->d_iname; | |
1593 | } | |
1da177e4 LT |
1594 | |
1595 | dentry->d_name.len = name->len; | |
1596 | dentry->d_name.hash = name->hash; | |
1597 | memcpy(dname, name->name, name->len); | |
1598 | dname[name->len] = 0; | |
1599 | ||
6326c71f LT |
1600 | /* Make sure we always see the terminating NUL character */ |
1601 | smp_wmb(); | |
1602 | dentry->d_name.name = dname; | |
1603 | ||
98474236 | 1604 | dentry->d_lockref.count = 1; |
dea3667b | 1605 | dentry->d_flags = 0; |
1da177e4 | 1606 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1607 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1608 | dentry->d_inode = NULL; |
a4464dbc AV |
1609 | dentry->d_parent = dentry; |
1610 | dentry->d_sb = sb; | |
1da177e4 LT |
1611 | dentry->d_op = NULL; |
1612 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1613 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1614 | INIT_LIST_HEAD(&dentry->d_lru); |
1615 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1616 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1617 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1618 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1619 | |
3e880fb5 | 1620 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1621 | |
1da177e4 LT |
1622 | return dentry; |
1623 | } | |
a4464dbc AV |
1624 | |
1625 | /** | |
1626 | * d_alloc - allocate a dcache entry | |
1627 | * @parent: parent of entry to allocate | |
1628 | * @name: qstr of the name | |
1629 | * | |
1630 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1631 | * available. On a success the dentry is returned. The name passed in is | |
1632 | * copied and the copy passed in may be reused after this call. | |
1633 | */ | |
1634 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1635 | { | |
1636 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1637 | if (!dentry) | |
1638 | return NULL; | |
1639 | ||
1640 | spin_lock(&parent->d_lock); | |
1641 | /* | |
1642 | * don't need child lock because it is not subject | |
1643 | * to concurrency here | |
1644 | */ | |
1645 | __dget_dlock(parent); | |
1646 | dentry->d_parent = parent; | |
946e51f2 | 1647 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1648 | spin_unlock(&parent->d_lock); |
1649 | ||
1650 | return dentry; | |
1651 | } | |
ec4f8605 | 1652 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1653 | |
e1a24bb0 BF |
1654 | /** |
1655 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1656 | * @sb: the superblock | |
1657 | * @name: qstr of the name | |
1658 | * | |
1659 | * For a filesystem that just pins its dentries in memory and never | |
1660 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1661 | */ | |
4b936885 NP |
1662 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1663 | { | |
e1a24bb0 | 1664 | return __d_alloc(sb, name); |
4b936885 NP |
1665 | } |
1666 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1667 | ||
1da177e4 LT |
1668 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1669 | { | |
1670 | struct qstr q; | |
1671 | ||
1672 | q.name = name; | |
fcfd2fbf | 1673 | q.hash_len = hashlen_string(name); |
1da177e4 LT |
1674 | return d_alloc(parent, &q); |
1675 | } | |
ef26ca97 | 1676 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1677 | |
fb045adb NP |
1678 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1679 | { | |
6f7f7caa LT |
1680 | WARN_ON_ONCE(dentry->d_op); |
1681 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1682 | DCACHE_OP_COMPARE | |
1683 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1684 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1685 | DCACHE_OP_DELETE | |
d101a125 MS |
1686 | DCACHE_OP_SELECT_INODE | |
1687 | DCACHE_OP_REAL)); | |
fb045adb NP |
1688 | dentry->d_op = op; |
1689 | if (!op) | |
1690 | return; | |
1691 | if (op->d_hash) | |
1692 | dentry->d_flags |= DCACHE_OP_HASH; | |
1693 | if (op->d_compare) | |
1694 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1695 | if (op->d_revalidate) | |
1696 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1697 | if (op->d_weak_revalidate) |
1698 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1699 | if (op->d_delete) |
1700 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1701 | if (op->d_prune) |
1702 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
4bacc9c9 DH |
1703 | if (op->d_select_inode) |
1704 | dentry->d_flags |= DCACHE_OP_SELECT_INODE; | |
d101a125 MS |
1705 | if (op->d_real) |
1706 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1707 | |
1708 | } | |
1709 | EXPORT_SYMBOL(d_set_d_op); | |
1710 | ||
df1a085a DH |
1711 | |
1712 | /* | |
1713 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1714 | * @dentry - The dentry to mark | |
1715 | * | |
1716 | * Mark a dentry as falling through to the lower layer (as set with | |
1717 | * d_pin_lower()). This flag may be recorded on the medium. | |
1718 | */ | |
1719 | void d_set_fallthru(struct dentry *dentry) | |
1720 | { | |
1721 | spin_lock(&dentry->d_lock); | |
1722 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1723 | spin_unlock(&dentry->d_lock); | |
1724 | } | |
1725 | EXPORT_SYMBOL(d_set_fallthru); | |
1726 | ||
b18825a7 DH |
1727 | static unsigned d_flags_for_inode(struct inode *inode) |
1728 | { | |
44bdb5e5 | 1729 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1730 | |
1731 | if (!inode) | |
1732 | return DCACHE_MISS_TYPE; | |
1733 | ||
1734 | if (S_ISDIR(inode->i_mode)) { | |
1735 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1736 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1737 | if (unlikely(!inode->i_op->lookup)) | |
1738 | add_flags = DCACHE_AUTODIR_TYPE; | |
1739 | else | |
1740 | inode->i_opflags |= IOP_LOOKUP; | |
1741 | } | |
44bdb5e5 DH |
1742 | goto type_determined; |
1743 | } | |
1744 | ||
1745 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1746 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1747 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1748 | goto type_determined; |
1749 | } | |
1750 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1751 | } |
1752 | ||
44bdb5e5 DH |
1753 | if (unlikely(!S_ISREG(inode->i_mode))) |
1754 | add_flags = DCACHE_SPECIAL_TYPE; | |
1755 | ||
1756 | type_determined: | |
b18825a7 DH |
1757 | if (unlikely(IS_AUTOMOUNT(inode))) |
1758 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1759 | return add_flags; | |
1760 | } | |
1761 | ||
360da900 OH |
1762 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1763 | { | |
b18825a7 | 1764 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 1765 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 1766 | |
b23fb0a6 | 1767 | spin_lock(&dentry->d_lock); |
de689f5e | 1768 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 1769 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 1770 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 1771 | raw_write_seqcount_end(&dentry->d_seq); |
de689f5e | 1772 | __fsnotify_d_instantiate(dentry); |
b23fb0a6 | 1773 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1774 | } |
1775 | ||
1da177e4 LT |
1776 | /** |
1777 | * d_instantiate - fill in inode information for a dentry | |
1778 | * @entry: dentry to complete | |
1779 | * @inode: inode to attach to this dentry | |
1780 | * | |
1781 | * Fill in inode information in the entry. | |
1782 | * | |
1783 | * This turns negative dentries into productive full members | |
1784 | * of society. | |
1785 | * | |
1786 | * NOTE! This assumes that the inode count has been incremented | |
1787 | * (or otherwise set) by the caller to indicate that it is now | |
1788 | * in use by the dcache. | |
1789 | */ | |
1790 | ||
1791 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1792 | { | |
946e51f2 | 1793 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 1794 | if (inode) { |
b9680917 | 1795 | security_d_instantiate(entry, inode); |
873feea0 | 1796 | spin_lock(&inode->i_lock); |
de689f5e | 1797 | __d_instantiate(entry, inode); |
873feea0 | 1798 | spin_unlock(&inode->i_lock); |
de689f5e | 1799 | } |
1da177e4 | 1800 | } |
ec4f8605 | 1801 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 1802 | |
b70a80e7 MS |
1803 | /** |
1804 | * d_instantiate_no_diralias - instantiate a non-aliased dentry | |
1805 | * @entry: dentry to complete | |
1806 | * @inode: inode to attach to this dentry | |
1807 | * | |
1808 | * Fill in inode information in the entry. If a directory alias is found, then | |
1809 | * return an error (and drop inode). Together with d_materialise_unique() this | |
1810 | * guarantees that a directory inode may never have more than one alias. | |
1811 | */ | |
1812 | int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode) | |
1813 | { | |
946e51f2 | 1814 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
b70a80e7 | 1815 | |
b9680917 | 1816 | security_d_instantiate(entry, inode); |
b70a80e7 MS |
1817 | spin_lock(&inode->i_lock); |
1818 | if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) { | |
1819 | spin_unlock(&inode->i_lock); | |
1820 | iput(inode); | |
1821 | return -EBUSY; | |
1822 | } | |
1823 | __d_instantiate(entry, inode); | |
1824 | spin_unlock(&inode->i_lock); | |
b70a80e7 MS |
1825 | |
1826 | return 0; | |
1827 | } | |
1828 | EXPORT_SYMBOL(d_instantiate_no_diralias); | |
1829 | ||
adc0e91a AV |
1830 | struct dentry *d_make_root(struct inode *root_inode) |
1831 | { | |
1832 | struct dentry *res = NULL; | |
1833 | ||
1834 | if (root_inode) { | |
798434bd | 1835 | res = __d_alloc(root_inode->i_sb, NULL); |
adc0e91a AV |
1836 | if (res) |
1837 | d_instantiate(res, root_inode); | |
1838 | else | |
1839 | iput(root_inode); | |
1840 | } | |
1841 | return res; | |
1842 | } | |
1843 | EXPORT_SYMBOL(d_make_root); | |
1844 | ||
d891eedb BF |
1845 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1846 | { | |
1847 | struct dentry *alias; | |
1848 | ||
b3d9b7a3 | 1849 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1850 | return NULL; |
946e51f2 | 1851 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1852 | __dget(alias); |
1853 | return alias; | |
1854 | } | |
1855 | ||
46f72b34 SW |
1856 | /** |
1857 | * d_find_any_alias - find any alias for a given inode | |
1858 | * @inode: inode to find an alias for | |
1859 | * | |
1860 | * If any aliases exist for the given inode, take and return a | |
1861 | * reference for one of them. If no aliases exist, return %NULL. | |
1862 | */ | |
1863 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1864 | { |
1865 | struct dentry *de; | |
1866 | ||
1867 | spin_lock(&inode->i_lock); | |
1868 | de = __d_find_any_alias(inode); | |
1869 | spin_unlock(&inode->i_lock); | |
1870 | return de; | |
1871 | } | |
46f72b34 | 1872 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1873 | |
49c7dd28 | 1874 | static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected) |
4ea3ada2 | 1875 | { |
9308a612 CH |
1876 | struct dentry *tmp; |
1877 | struct dentry *res; | |
b18825a7 | 1878 | unsigned add_flags; |
4ea3ada2 CH |
1879 | |
1880 | if (!inode) | |
44003728 | 1881 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1882 | if (IS_ERR(inode)) |
1883 | return ERR_CAST(inode); | |
1884 | ||
d891eedb | 1885 | res = d_find_any_alias(inode); |
9308a612 CH |
1886 | if (res) |
1887 | goto out_iput; | |
1888 | ||
798434bd | 1889 | tmp = __d_alloc(inode->i_sb, NULL); |
9308a612 CH |
1890 | if (!tmp) { |
1891 | res = ERR_PTR(-ENOMEM); | |
1892 | goto out_iput; | |
4ea3ada2 | 1893 | } |
b5c84bf6 | 1894 | |
b9680917 | 1895 | security_d_instantiate(tmp, inode); |
873feea0 | 1896 | spin_lock(&inode->i_lock); |
d891eedb | 1897 | res = __d_find_any_alias(inode); |
9308a612 | 1898 | if (res) { |
873feea0 | 1899 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1900 | dput(tmp); |
1901 | goto out_iput; | |
1902 | } | |
1903 | ||
1904 | /* attach a disconnected dentry */ | |
1a0a397e BF |
1905 | add_flags = d_flags_for_inode(inode); |
1906 | ||
1907 | if (disconnected) | |
1908 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 1909 | |
9308a612 | 1910 | spin_lock(&tmp->d_lock); |
4bf46a27 | 1911 | __d_set_inode_and_type(tmp, inode, add_flags); |
946e51f2 | 1912 | hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry); |
1879fd6a | 1913 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1914 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1915 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1916 | spin_unlock(&tmp->d_lock); |
873feea0 | 1917 | spin_unlock(&inode->i_lock); |
9308a612 | 1918 | |
9308a612 CH |
1919 | return tmp; |
1920 | ||
1921 | out_iput: | |
1922 | iput(inode); | |
1923 | return res; | |
4ea3ada2 | 1924 | } |
1a0a397e BF |
1925 | |
1926 | /** | |
1927 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
1928 | * @inode: inode to allocate the dentry for | |
1929 | * | |
1930 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1931 | * similar open by handle operations. The returned dentry may be anonymous, | |
1932 | * or may have a full name (if the inode was already in the cache). | |
1933 | * | |
1934 | * When called on a directory inode, we must ensure that the inode only ever | |
1935 | * has one dentry. If a dentry is found, that is returned instead of | |
1936 | * allocating a new one. | |
1937 | * | |
1938 | * On successful return, the reference to the inode has been transferred | |
1939 | * to the dentry. In case of an error the reference on the inode is released. | |
1940 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1941 | * be passed in and the error will be propagated to the return value, | |
1942 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
1943 | */ | |
1944 | struct dentry *d_obtain_alias(struct inode *inode) | |
1945 | { | |
1946 | return __d_obtain_alias(inode, 1); | |
1947 | } | |
adc48720 | 1948 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 1949 | |
1a0a397e BF |
1950 | /** |
1951 | * d_obtain_root - find or allocate a dentry for a given inode | |
1952 | * @inode: inode to allocate the dentry for | |
1953 | * | |
1954 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
1955 | * | |
1956 | * We must ensure that directory inodes only ever have one dentry. If a | |
1957 | * dentry is found, that is returned instead of allocating a new one. | |
1958 | * | |
1959 | * On successful return, the reference to the inode has been transferred | |
1960 | * to the dentry. In case of an error the reference on the inode is | |
1961 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
1962 | * error will be propagate to the return value, with a %NULL @inode | |
1963 | * replaced by ERR_PTR(-ESTALE). | |
1964 | */ | |
1965 | struct dentry *d_obtain_root(struct inode *inode) | |
1966 | { | |
1967 | return __d_obtain_alias(inode, 0); | |
1968 | } | |
1969 | EXPORT_SYMBOL(d_obtain_root); | |
1970 | ||
9403540c BN |
1971 | /** |
1972 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1973 | * @inode: the inode case-insensitive lookup has found | |
1974 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1975 | * @name: the case-exact name to be associated with the returned dentry | |
1976 | * | |
1977 | * This is to avoid filling the dcache with case-insensitive names to the | |
1978 | * same inode, only the actual correct case is stored in the dcache for | |
1979 | * case-insensitive filesystems. | |
1980 | * | |
1981 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1982 | * already exists in in the dcache, use it and return it. | |
1983 | * | |
1984 | * If no entry exists with the exact case name, allocate new dentry with | |
1985 | * the exact case, and return the spliced entry. | |
1986 | */ | |
e45b590b | 1987 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1988 | struct qstr *name) |
1989 | { | |
d9171b93 | 1990 | struct dentry *found, *res; |
9403540c | 1991 | |
b6520c81 CH |
1992 | /* |
1993 | * First check if a dentry matching the name already exists, | |
1994 | * if not go ahead and create it now. | |
1995 | */ | |
9403540c | 1996 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
1997 | if (found) { |
1998 | iput(inode); | |
1999 | return found; | |
2000 | } | |
2001 | if (d_in_lookup(dentry)) { | |
2002 | found = d_alloc_parallel(dentry->d_parent, name, | |
2003 | dentry->d_wait); | |
2004 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2005 | iput(inode); | |
2006 | return found; | |
9403540c | 2007 | } |
d9171b93 AV |
2008 | } else { |
2009 | found = d_alloc(dentry->d_parent, name); | |
2010 | if (!found) { | |
2011 | iput(inode); | |
2012 | return ERR_PTR(-ENOMEM); | |
2013 | } | |
2014 | } | |
2015 | res = d_splice_alias(inode, found); | |
2016 | if (res) { | |
2017 | dput(found); | |
2018 | return res; | |
9403540c | 2019 | } |
4f522a24 | 2020 | return found; |
9403540c | 2021 | } |
ec4f8605 | 2022 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2023 | |
12f8ad4b LT |
2024 | /* |
2025 | * Do the slow-case of the dentry name compare. | |
2026 | * | |
2027 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 2028 | * load the name and length information, so that the |
12f8ad4b LT |
2029 | * filesystem can rely on them, and can use the 'name' and |
2030 | * 'len' information without worrying about walking off the | |
2031 | * end of memory etc. | |
2032 | * | |
2033 | * Thus the read_seqcount_retry() and the "duplicate" info | |
2034 | * in arguments (the low-level filesystem should not look | |
2035 | * at the dentry inode or name contents directly, since | |
2036 | * rename can change them while we're in RCU mode). | |
2037 | */ | |
2038 | enum slow_d_compare { | |
2039 | D_COMP_OK, | |
2040 | D_COMP_NOMATCH, | |
2041 | D_COMP_SEQRETRY, | |
2042 | }; | |
2043 | ||
2044 | static noinline enum slow_d_compare slow_dentry_cmp( | |
2045 | const struct dentry *parent, | |
12f8ad4b LT |
2046 | struct dentry *dentry, |
2047 | unsigned int seq, | |
2048 | const struct qstr *name) | |
2049 | { | |
2050 | int tlen = dentry->d_name.len; | |
2051 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
2052 | |
2053 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2054 | cpu_relax(); | |
2055 | return D_COMP_SEQRETRY; | |
2056 | } | |
da53be12 | 2057 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
2058 | return D_COMP_NOMATCH; |
2059 | return D_COMP_OK; | |
2060 | } | |
2061 | ||
31e6b01f NP |
2062 | /** |
2063 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2064 | * @parent: parent dentry | |
2065 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2066 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2067 | * Returns: dentry, or NULL |
2068 | * | |
2069 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2070 | * resolution (store-free path walking) design described in | |
2071 | * Documentation/filesystems/path-lookup.txt. | |
2072 | * | |
2073 | * This is not to be used outside core vfs. | |
2074 | * | |
2075 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2076 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2077 | * without taking d_lock and checking d_seq sequence count against @seq | |
2078 | * returned here. | |
2079 | * | |
15570086 | 2080 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2081 | * function. |
2082 | * | |
2083 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2084 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2085 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2086 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2087 | * |
2088 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2089 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2090 | */ |
8966be90 LT |
2091 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2092 | const struct qstr *name, | |
da53be12 | 2093 | unsigned *seqp) |
31e6b01f | 2094 | { |
26fe5750 | 2095 | u64 hashlen = name->hash_len; |
31e6b01f | 2096 | const unsigned char *str = name->name; |
26fe5750 | 2097 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 2098 | struct hlist_bl_node *node; |
31e6b01f NP |
2099 | struct dentry *dentry; |
2100 | ||
2101 | /* | |
2102 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2103 | * required to prevent single threaded performance regressions | |
2104 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2105 | * Keep the two functions in sync. | |
2106 | */ | |
2107 | ||
2108 | /* | |
2109 | * The hash list is protected using RCU. | |
2110 | * | |
2111 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2112 | * races with d_move(). | |
2113 | * | |
2114 | * It is possible that concurrent renames can mess up our list | |
2115 | * walk here and result in missing our dentry, resulting in the | |
2116 | * false-negative result. d_lookup() protects against concurrent | |
2117 | * renames using rename_lock seqlock. | |
2118 | * | |
b0a4bb83 | 2119 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2120 | */ |
b07ad996 | 2121 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2122 | unsigned seq; |
31e6b01f | 2123 | |
31e6b01f | 2124 | seqretry: |
12f8ad4b LT |
2125 | /* |
2126 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2127 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2128 | * |
2129 | * The caller must perform a seqcount check in order | |
da53be12 | 2130 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2131 | * |
2132 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2133 | * we don't wait for the sequence count to stabilize if it | |
2134 | * is in the middle of a sequence change. If we do the slow | |
2135 | * dentry compare, we will do seqretries until it is stable, | |
2136 | * and if we end up with a successful lookup, we actually | |
2137 | * want to exit RCU lookup anyway. | |
2138 | */ | |
2139 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2140 | if (dentry->d_parent != parent) |
2141 | continue; | |
2e321806 LT |
2142 | if (d_unhashed(dentry)) |
2143 | continue; | |
12f8ad4b | 2144 | |
830c0f0e | 2145 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
2146 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2147 | continue; | |
da53be12 LT |
2148 | *seqp = seq; |
2149 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
2150 | case D_COMP_OK: |
2151 | return dentry; | |
2152 | case D_COMP_NOMATCH: | |
31e6b01f | 2153 | continue; |
12f8ad4b LT |
2154 | default: |
2155 | goto seqretry; | |
2156 | } | |
31e6b01f | 2157 | } |
12f8ad4b | 2158 | |
26fe5750 | 2159 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2160 | continue; |
da53be12 | 2161 | *seqp = seq; |
26fe5750 | 2162 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2163 | return dentry; |
31e6b01f NP |
2164 | } |
2165 | return NULL; | |
2166 | } | |
2167 | ||
1da177e4 LT |
2168 | /** |
2169 | * d_lookup - search for a dentry | |
2170 | * @parent: parent dentry | |
2171 | * @name: qstr of name we wish to find | |
b04f784e | 2172 | * Returns: dentry, or NULL |
1da177e4 | 2173 | * |
b04f784e NP |
2174 | * d_lookup searches the children of the parent dentry for the name in |
2175 | * question. If the dentry is found its reference count is incremented and the | |
2176 | * dentry is returned. The caller must use dput to free the entry when it has | |
2177 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2178 | */ |
da2d8455 | 2179 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2180 | { |
31e6b01f | 2181 | struct dentry *dentry; |
949854d0 | 2182 | unsigned seq; |
1da177e4 | 2183 | |
b8314f93 DY |
2184 | do { |
2185 | seq = read_seqbegin(&rename_lock); | |
2186 | dentry = __d_lookup(parent, name); | |
2187 | if (dentry) | |
1da177e4 LT |
2188 | break; |
2189 | } while (read_seqretry(&rename_lock, seq)); | |
2190 | return dentry; | |
2191 | } | |
ec4f8605 | 2192 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2193 | |
31e6b01f | 2194 | /** |
b04f784e NP |
2195 | * __d_lookup - search for a dentry (racy) |
2196 | * @parent: parent dentry | |
2197 | * @name: qstr of name we wish to find | |
2198 | * Returns: dentry, or NULL | |
2199 | * | |
2200 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2201 | * false-negative result due to unrelated rename activity. | |
2202 | * | |
2203 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2204 | * however it must be used carefully, eg. with a following d_lookup in | |
2205 | * the case of failure. | |
2206 | * | |
2207 | * __d_lookup callers must be commented. | |
2208 | */ | |
a713ca2a | 2209 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2210 | { |
2211 | unsigned int len = name->len; | |
2212 | unsigned int hash = name->hash; | |
2213 | const unsigned char *str = name->name; | |
b07ad996 | 2214 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2215 | struct hlist_bl_node *node; |
31e6b01f | 2216 | struct dentry *found = NULL; |
665a7583 | 2217 | struct dentry *dentry; |
1da177e4 | 2218 | |
31e6b01f NP |
2219 | /* |
2220 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2221 | * required to prevent single threaded performance regressions | |
2222 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2223 | * Keep the two functions in sync. | |
2224 | */ | |
2225 | ||
b04f784e NP |
2226 | /* |
2227 | * The hash list is protected using RCU. | |
2228 | * | |
2229 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2230 | * with d_move(). | |
2231 | * | |
2232 | * It is possible that concurrent renames can mess up our list | |
2233 | * walk here and result in missing our dentry, resulting in the | |
2234 | * false-negative result. d_lookup() protects against concurrent | |
2235 | * renames using rename_lock seqlock. | |
2236 | * | |
b0a4bb83 | 2237 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2238 | */ |
1da177e4 LT |
2239 | rcu_read_lock(); |
2240 | ||
b07ad996 | 2241 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2242 | |
1da177e4 LT |
2243 | if (dentry->d_name.hash != hash) |
2244 | continue; | |
1da177e4 LT |
2245 | |
2246 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2247 | if (dentry->d_parent != parent) |
2248 | goto next; | |
d0185c08 LT |
2249 | if (d_unhashed(dentry)) |
2250 | goto next; | |
2251 | ||
1da177e4 LT |
2252 | /* |
2253 | * It is safe to compare names since d_move() cannot | |
2254 | * change the qstr (protected by d_lock). | |
2255 | */ | |
fb045adb | 2256 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2257 | int tlen = dentry->d_name.len; |
2258 | const char *tname = dentry->d_name.name; | |
da53be12 | 2259 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2260 | goto next; |
2261 | } else { | |
ee983e89 LT |
2262 | if (dentry->d_name.len != len) |
2263 | goto next; | |
12f8ad4b | 2264 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2265 | goto next; |
2266 | } | |
2267 | ||
98474236 | 2268 | dentry->d_lockref.count++; |
d0185c08 | 2269 | found = dentry; |
1da177e4 LT |
2270 | spin_unlock(&dentry->d_lock); |
2271 | break; | |
2272 | next: | |
2273 | spin_unlock(&dentry->d_lock); | |
2274 | } | |
2275 | rcu_read_unlock(); | |
2276 | ||
2277 | return found; | |
2278 | } | |
2279 | ||
3e7e241f EB |
2280 | /** |
2281 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2282 | * @dir: Directory to search in | |
2283 | * @name: qstr of name we wish to find | |
2284 | * | |
4f522a24 | 2285 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2286 | */ |
2287 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2288 | { | |
3e7e241f EB |
2289 | /* |
2290 | * Check for a fs-specific hash function. Note that we must | |
2291 | * calculate the standard hash first, as the d_op->d_hash() | |
2292 | * routine may choose to leave the hash value unchanged. | |
2293 | */ | |
2294 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2295 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2296 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2297 | if (unlikely(err < 0)) |
2298 | return ERR_PTR(err); | |
3e7e241f | 2299 | } |
4f522a24 | 2300 | return d_lookup(dir, name); |
3e7e241f | 2301 | } |
4f522a24 | 2302 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2303 | |
1da177e4 LT |
2304 | /* |
2305 | * When a file is deleted, we have two options: | |
2306 | * - turn this dentry into a negative dentry | |
2307 | * - unhash this dentry and free it. | |
2308 | * | |
2309 | * Usually, we want to just turn this into | |
2310 | * a negative dentry, but if anybody else is | |
2311 | * currently using the dentry or the inode | |
2312 | * we can't do that and we fall back on removing | |
2313 | * it from the hash queues and waiting for | |
2314 | * it to be deleted later when it has no users | |
2315 | */ | |
2316 | ||
2317 | /** | |
2318 | * d_delete - delete a dentry | |
2319 | * @dentry: The dentry to delete | |
2320 | * | |
2321 | * Turn the dentry into a negative dentry if possible, otherwise | |
2322 | * remove it from the hash queues so it can be deleted later | |
2323 | */ | |
2324 | ||
2325 | void d_delete(struct dentry * dentry) | |
2326 | { | |
873feea0 | 2327 | struct inode *inode; |
7a91bf7f | 2328 | int isdir = 0; |
1da177e4 LT |
2329 | /* |
2330 | * Are we the only user? | |
2331 | */ | |
357f8e65 | 2332 | again: |
1da177e4 | 2333 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2334 | inode = dentry->d_inode; |
2335 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2336 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2337 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2338 | spin_unlock(&dentry->d_lock); |
2339 | cpu_relax(); | |
2340 | goto again; | |
2341 | } | |
13e3c5e5 | 2342 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2343 | dentry_unlink_inode(dentry); |
7a91bf7f | 2344 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2345 | return; |
2346 | } | |
2347 | ||
2348 | if (!d_unhashed(dentry)) | |
2349 | __d_drop(dentry); | |
2350 | ||
2351 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2352 | |
2353 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2354 | } |
ec4f8605 | 2355 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2356 | |
b07ad996 | 2357 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2358 | { |
ceb5bdc2 | 2359 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2360 | hlist_bl_lock(b); |
dea3667b | 2361 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2362 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2363 | hlist_bl_unlock(b); |
1da177e4 LT |
2364 | } |
2365 | ||
770bfad8 DH |
2366 | static void _d_rehash(struct dentry * entry) |
2367 | { | |
2368 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2369 | } | |
2370 | ||
1da177e4 LT |
2371 | /** |
2372 | * d_rehash - add an entry back to the hash | |
2373 | * @entry: dentry to add to the hash | |
2374 | * | |
2375 | * Adds a dentry to the hash according to its name. | |
2376 | */ | |
2377 | ||
2378 | void d_rehash(struct dentry * entry) | |
2379 | { | |
1da177e4 | 2380 | spin_lock(&entry->d_lock); |
770bfad8 | 2381 | _d_rehash(entry); |
1da177e4 | 2382 | spin_unlock(&entry->d_lock); |
1da177e4 | 2383 | } |
ec4f8605 | 2384 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2385 | |
84e710da AV |
2386 | static inline unsigned start_dir_add(struct inode *dir) |
2387 | { | |
2388 | ||
2389 | for (;;) { | |
2390 | unsigned n = dir->i_dir_seq; | |
2391 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2392 | return n; | |
2393 | cpu_relax(); | |
2394 | } | |
2395 | } | |
2396 | ||
2397 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2398 | { | |
2399 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2400 | } | |
2401 | ||
d9171b93 AV |
2402 | static void d_wait_lookup(struct dentry *dentry) |
2403 | { | |
2404 | if (d_in_lookup(dentry)) { | |
2405 | DECLARE_WAITQUEUE(wait, current); | |
2406 | add_wait_queue(dentry->d_wait, &wait); | |
2407 | do { | |
2408 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2409 | spin_unlock(&dentry->d_lock); | |
2410 | schedule(); | |
2411 | spin_lock(&dentry->d_lock); | |
2412 | } while (d_in_lookup(dentry)); | |
2413 | } | |
2414 | } | |
2415 | ||
94bdd655 | 2416 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2417 | const struct qstr *name, |
2418 | wait_queue_head_t *wq) | |
94bdd655 AV |
2419 | { |
2420 | unsigned int len = name->len; | |
2421 | unsigned int hash = name->hash; | |
2422 | const unsigned char *str = name->name; | |
2423 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); | |
2424 | struct hlist_bl_node *node; | |
2425 | struct dentry *new = d_alloc(parent, name); | |
2426 | struct dentry *dentry; | |
2427 | unsigned seq, r_seq, d_seq; | |
2428 | ||
2429 | if (unlikely(!new)) | |
2430 | return ERR_PTR(-ENOMEM); | |
2431 | ||
2432 | retry: | |
2433 | rcu_read_lock(); | |
2434 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1; | |
2435 | r_seq = read_seqbegin(&rename_lock); | |
2436 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2437 | if (unlikely(dentry)) { | |
2438 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2439 | rcu_read_unlock(); | |
2440 | goto retry; | |
2441 | } | |
2442 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2443 | rcu_read_unlock(); | |
2444 | dput(dentry); | |
2445 | goto retry; | |
2446 | } | |
2447 | rcu_read_unlock(); | |
2448 | dput(new); | |
2449 | return dentry; | |
2450 | } | |
2451 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2452 | rcu_read_unlock(); | |
2453 | goto retry; | |
2454 | } | |
2455 | hlist_bl_lock(b); | |
2456 | if (unlikely(parent->d_inode->i_dir_seq != seq)) { | |
2457 | hlist_bl_unlock(b); | |
2458 | rcu_read_unlock(); | |
2459 | goto retry; | |
2460 | } | |
2461 | rcu_read_unlock(); | |
2462 | /* | |
2463 | * No changes for the parent since the beginning of d_lookup(). | |
2464 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2465 | * any potential in-lookup matches are going to stay here until | |
2466 | * we unlock the chain. All fields are stable in everything | |
2467 | * we encounter. | |
2468 | */ | |
2469 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2470 | if (dentry->d_name.hash != hash) | |
2471 | continue; | |
2472 | if (dentry->d_parent != parent) | |
2473 | continue; | |
2474 | if (d_unhashed(dentry)) | |
2475 | continue; | |
2476 | if (parent->d_flags & DCACHE_OP_COMPARE) { | |
2477 | int tlen = dentry->d_name.len; | |
2478 | const char *tname = dentry->d_name.name; | |
2479 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) | |
2480 | continue; | |
2481 | } else { | |
2482 | if (dentry->d_name.len != len) | |
2483 | continue; | |
2484 | if (dentry_cmp(dentry, str, len)) | |
2485 | continue; | |
2486 | } | |
2487 | dget(dentry); | |
2488 | hlist_bl_unlock(b); | |
d9171b93 AV |
2489 | /* somebody is doing lookup for it right now; wait for it */ |
2490 | spin_lock(&dentry->d_lock); | |
2491 | d_wait_lookup(dentry); | |
2492 | /* | |
2493 | * it's not in-lookup anymore; in principle we should repeat | |
2494 | * everything from dcache lookup, but it's likely to be what | |
2495 | * d_lookup() would've found anyway. If it is, just return it; | |
2496 | * otherwise we really have to repeat the whole thing. | |
2497 | */ | |
2498 | if (unlikely(dentry->d_name.hash != hash)) | |
2499 | goto mismatch; | |
2500 | if (unlikely(dentry->d_parent != parent)) | |
2501 | goto mismatch; | |
2502 | if (unlikely(d_unhashed(dentry))) | |
2503 | goto mismatch; | |
2504 | if (parent->d_flags & DCACHE_OP_COMPARE) { | |
2505 | int tlen = dentry->d_name.len; | |
2506 | const char *tname = dentry->d_name.name; | |
2507 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) | |
2508 | goto mismatch; | |
2509 | } else { | |
2510 | if (unlikely(dentry->d_name.len != len)) | |
2511 | goto mismatch; | |
2512 | if (unlikely(dentry_cmp(dentry, str, len))) | |
2513 | goto mismatch; | |
2514 | } | |
2515 | /* OK, it *is* a hashed match; return it */ | |
2516 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2517 | dput(new); |
2518 | return dentry; | |
2519 | } | |
2520 | /* we can't take ->d_lock here; it's OK, though. */ | |
2521 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2522 | new->d_wait = wq; |
94bdd655 AV |
2523 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2524 | hlist_bl_unlock(b); | |
2525 | return new; | |
d9171b93 AV |
2526 | mismatch: |
2527 | spin_unlock(&dentry->d_lock); | |
2528 | dput(dentry); | |
2529 | goto retry; | |
94bdd655 AV |
2530 | } |
2531 | EXPORT_SYMBOL(d_alloc_parallel); | |
2532 | ||
85c7f810 AV |
2533 | void __d_lookup_done(struct dentry *dentry) |
2534 | { | |
94bdd655 AV |
2535 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, |
2536 | dentry->d_name.hash); | |
2537 | hlist_bl_lock(b); | |
85c7f810 | 2538 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2539 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
d9171b93 AV |
2540 | wake_up_all(dentry->d_wait); |
2541 | dentry->d_wait = NULL; | |
94bdd655 AV |
2542 | hlist_bl_unlock(b); |
2543 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2544 | INIT_LIST_HEAD(&dentry->d_lru); |
85c7f810 AV |
2545 | } |
2546 | EXPORT_SYMBOL(__d_lookup_done); | |
ed782b5a AV |
2547 | |
2548 | /* inode->i_lock held if inode is non-NULL */ | |
2549 | ||
2550 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2551 | { | |
84e710da AV |
2552 | struct inode *dir = NULL; |
2553 | unsigned n; | |
0568d705 | 2554 | spin_lock(&dentry->d_lock); |
84e710da AV |
2555 | if (unlikely(d_in_lookup(dentry))) { |
2556 | dir = dentry->d_parent->d_inode; | |
2557 | n = start_dir_add(dir); | |
85c7f810 | 2558 | __d_lookup_done(dentry); |
84e710da | 2559 | } |
ed782b5a | 2560 | if (inode) { |
0568d705 AV |
2561 | unsigned add_flags = d_flags_for_inode(inode); |
2562 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2563 | raw_write_seqcount_begin(&dentry->d_seq); | |
2564 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2565 | raw_write_seqcount_end(&dentry->d_seq); | |
2566 | __fsnotify_d_instantiate(dentry); | |
ed782b5a | 2567 | } |
0568d705 | 2568 | _d_rehash(dentry); |
84e710da AV |
2569 | if (dir) |
2570 | end_dir_add(dir, n); | |
0568d705 AV |
2571 | spin_unlock(&dentry->d_lock); |
2572 | if (inode) | |
2573 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2574 | } |
2575 | ||
34d0d19d AV |
2576 | /** |
2577 | * d_add - add dentry to hash queues | |
2578 | * @entry: dentry to add | |
2579 | * @inode: The inode to attach to this dentry | |
2580 | * | |
2581 | * This adds the entry to the hash queues and initializes @inode. | |
2582 | * The entry was actually filled in earlier during d_alloc(). | |
2583 | */ | |
2584 | ||
2585 | void d_add(struct dentry *entry, struct inode *inode) | |
2586 | { | |
b9680917 AV |
2587 | if (inode) { |
2588 | security_d_instantiate(entry, inode); | |
ed782b5a | 2589 | spin_lock(&inode->i_lock); |
b9680917 | 2590 | } |
ed782b5a | 2591 | __d_add(entry, inode); |
34d0d19d AV |
2592 | } |
2593 | EXPORT_SYMBOL(d_add); | |
2594 | ||
668d0cd5 AV |
2595 | /** |
2596 | * d_exact_alias - find and hash an exact unhashed alias | |
2597 | * @entry: dentry to add | |
2598 | * @inode: The inode to go with this dentry | |
2599 | * | |
2600 | * If an unhashed dentry with the same name/parent and desired | |
2601 | * inode already exists, hash and return it. Otherwise, return | |
2602 | * NULL. | |
2603 | * | |
2604 | * Parent directory should be locked. | |
2605 | */ | |
2606 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2607 | { | |
2608 | struct dentry *alias; | |
2609 | int len = entry->d_name.len; | |
2610 | const char *name = entry->d_name.name; | |
2611 | unsigned int hash = entry->d_name.hash; | |
2612 | ||
2613 | spin_lock(&inode->i_lock); | |
2614 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2615 | /* | |
2616 | * Don't need alias->d_lock here, because aliases with | |
2617 | * d_parent == entry->d_parent are not subject to name or | |
2618 | * parent changes, because the parent inode i_mutex is held. | |
2619 | */ | |
2620 | if (alias->d_name.hash != hash) | |
2621 | continue; | |
2622 | if (alias->d_parent != entry->d_parent) | |
2623 | continue; | |
2624 | if (alias->d_name.len != len) | |
2625 | continue; | |
2626 | if (dentry_cmp(alias, name, len)) | |
2627 | continue; | |
2628 | spin_lock(&alias->d_lock); | |
2629 | if (!d_unhashed(alias)) { | |
2630 | spin_unlock(&alias->d_lock); | |
2631 | alias = NULL; | |
2632 | } else { | |
2633 | __dget_dlock(alias); | |
2634 | _d_rehash(alias); | |
2635 | spin_unlock(&alias->d_lock); | |
2636 | } | |
2637 | spin_unlock(&inode->i_lock); | |
2638 | return alias; | |
2639 | } | |
2640 | spin_unlock(&inode->i_lock); | |
2641 | return NULL; | |
2642 | } | |
2643 | EXPORT_SYMBOL(d_exact_alias); | |
2644 | ||
fb2d5b86 NP |
2645 | /** |
2646 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2647 | * @dentry: dentry to be updated | |
2648 | * @name: new name | |
2649 | * | |
2650 | * Update a case insensitive dentry with new case of name. | |
2651 | * | |
2652 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2653 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2654 | * lengths). | |
2655 | * | |
2656 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2657 | * keep renames and concurrent inserts, and readdir(2) away). | |
2658 | */ | |
2659 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2660 | { | |
5955102c | 2661 | BUG_ON(!inode_is_locked(dentry->d_parent->d_inode)); |
fb2d5b86 NP |
2662 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2663 | ||
fb2d5b86 | 2664 | spin_lock(&dentry->d_lock); |
31e6b01f | 2665 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2666 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2667 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2668 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2669 | } |
2670 | EXPORT_SYMBOL(dentry_update_name_case); | |
2671 | ||
8d85b484 | 2672 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2673 | { |
8d85b484 AV |
2674 | if (unlikely(dname_external(target))) { |
2675 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2676 | /* |
2677 | * Both external: swap the pointers | |
2678 | */ | |
9a8d5bb4 | 2679 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2680 | } else { |
2681 | /* | |
2682 | * dentry:internal, target:external. Steal target's | |
2683 | * storage and make target internal. | |
2684 | */ | |
321bcf92 BF |
2685 | memcpy(target->d_iname, dentry->d_name.name, |
2686 | dentry->d_name.len + 1); | |
1da177e4 LT |
2687 | dentry->d_name.name = target->d_name.name; |
2688 | target->d_name.name = target->d_iname; | |
2689 | } | |
2690 | } else { | |
8d85b484 | 2691 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2692 | /* |
2693 | * dentry:external, target:internal. Give dentry's | |
2694 | * storage to target and make dentry internal | |
2695 | */ | |
2696 | memcpy(dentry->d_iname, target->d_name.name, | |
2697 | target->d_name.len + 1); | |
2698 | target->d_name.name = dentry->d_name.name; | |
2699 | dentry->d_name.name = dentry->d_iname; | |
2700 | } else { | |
2701 | /* | |
da1ce067 | 2702 | * Both are internal. |
1da177e4 | 2703 | */ |
da1ce067 MS |
2704 | unsigned int i; |
2705 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
08d4f772 MP |
2706 | kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN); |
2707 | kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN); | |
da1ce067 MS |
2708 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { |
2709 | swap(((long *) &dentry->d_iname)[i], | |
2710 | ((long *) &target->d_iname)[i]); | |
2711 | } | |
1da177e4 LT |
2712 | } |
2713 | } | |
a28ddb87 | 2714 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2715 | } |
2716 | ||
8d85b484 AV |
2717 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2718 | { | |
2719 | struct external_name *old_name = NULL; | |
2720 | if (unlikely(dname_external(dentry))) | |
2721 | old_name = external_name(dentry); | |
2722 | if (unlikely(dname_external(target))) { | |
2723 | atomic_inc(&external_name(target)->u.count); | |
2724 | dentry->d_name = target->d_name; | |
2725 | } else { | |
2726 | memcpy(dentry->d_iname, target->d_name.name, | |
2727 | target->d_name.len + 1); | |
2728 | dentry->d_name.name = dentry->d_iname; | |
2729 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2730 | } | |
2731 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2732 | kfree_rcu(old_name, u.head); | |
2733 | } | |
2734 | ||
2fd6b7f5 NP |
2735 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2736 | { | |
2737 | /* | |
2738 | * XXXX: do we really need to take target->d_lock? | |
2739 | */ | |
2740 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2741 | spin_lock(&target->d_parent->d_lock); | |
2742 | else { | |
2743 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2744 | spin_lock(&dentry->d_parent->d_lock); | |
2745 | spin_lock_nested(&target->d_parent->d_lock, | |
2746 | DENTRY_D_LOCK_NESTED); | |
2747 | } else { | |
2748 | spin_lock(&target->d_parent->d_lock); | |
2749 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2750 | DENTRY_D_LOCK_NESTED); | |
2751 | } | |
2752 | } | |
2753 | if (target < dentry) { | |
2754 | spin_lock_nested(&target->d_lock, 2); | |
2755 | spin_lock_nested(&dentry->d_lock, 3); | |
2756 | } else { | |
2757 | spin_lock_nested(&dentry->d_lock, 2); | |
2758 | spin_lock_nested(&target->d_lock, 3); | |
2759 | } | |
2760 | } | |
2761 | ||
986c0194 | 2762 | static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target) |
2fd6b7f5 NP |
2763 | { |
2764 | if (target->d_parent != dentry->d_parent) | |
2765 | spin_unlock(&dentry->d_parent->d_lock); | |
2766 | if (target->d_parent != target) | |
2767 | spin_unlock(&target->d_parent->d_lock); | |
986c0194 AV |
2768 | spin_unlock(&target->d_lock); |
2769 | spin_unlock(&dentry->d_lock); | |
2fd6b7f5 NP |
2770 | } |
2771 | ||
1da177e4 | 2772 | /* |
2fd6b7f5 NP |
2773 | * When switching names, the actual string doesn't strictly have to |
2774 | * be preserved in the target - because we're dropping the target | |
2775 | * anyway. As such, we can just do a simple memcpy() to copy over | |
d2fa4a84 ME |
2776 | * the new name before we switch, unless we are going to rehash |
2777 | * it. Note that if we *do* unhash the target, we are not allowed | |
2778 | * to rehash it without giving it a new name/hash key - whether | |
2779 | * we swap or overwrite the names here, resulting name won't match | |
2780 | * the reality in filesystem; it's only there for d_path() purposes. | |
2781 | * Note that all of this is happening under rename_lock, so the | |
2782 | * any hash lookup seeing it in the middle of manipulations will | |
2783 | * be discarded anyway. So we do not care what happens to the hash | |
2784 | * key in that case. | |
1da177e4 | 2785 | */ |
9eaef27b | 2786 | /* |
18367501 | 2787 | * __d_move - move a dentry |
1da177e4 LT |
2788 | * @dentry: entry to move |
2789 | * @target: new dentry | |
da1ce067 | 2790 | * @exchange: exchange the two dentries |
1da177e4 LT |
2791 | * |
2792 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2793 | * dcache entries should not be moved in this way. Caller must hold |
2794 | * rename_lock, the i_mutex of the source and target directories, | |
2795 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2796 | */ |
da1ce067 MS |
2797 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2798 | bool exchange) | |
1da177e4 | 2799 | { |
84e710da AV |
2800 | struct inode *dir = NULL; |
2801 | unsigned n; | |
1da177e4 LT |
2802 | if (!dentry->d_inode) |
2803 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2804 | ||
2fd6b7f5 NP |
2805 | BUG_ON(d_ancestor(dentry, target)); |
2806 | BUG_ON(d_ancestor(target, dentry)); | |
2807 | ||
2fd6b7f5 | 2808 | dentry_lock_for_move(dentry, target); |
84e710da AV |
2809 | if (unlikely(d_in_lookup(target))) { |
2810 | dir = target->d_parent->d_inode; | |
2811 | n = start_dir_add(dir); | |
85c7f810 | 2812 | __d_lookup_done(target); |
84e710da | 2813 | } |
1da177e4 | 2814 | |
31e6b01f | 2815 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2816 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2817 | |
ceb5bdc2 NP |
2818 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2819 | ||
2820 | /* | |
2821 | * Move the dentry to the target hash queue. Don't bother checking | |
2822 | * for the same hash queue because of how unlikely it is. | |
2823 | */ | |
2824 | __d_drop(dentry); | |
789680d1 | 2825 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 | 2826 | |
da1ce067 MS |
2827 | /* |
2828 | * Unhash the target (d_delete() is not usable here). If exchanging | |
2829 | * the two dentries, then rehash onto the other's hash queue. | |
2830 | */ | |
1da177e4 | 2831 | __d_drop(target); |
da1ce067 MS |
2832 | if (exchange) { |
2833 | __d_rehash(target, | |
2834 | d_hash(dentry->d_parent, dentry->d_name.hash)); | |
2835 | } | |
1da177e4 | 2836 | |
1da177e4 | 2837 | /* Switch the names.. */ |
8d85b484 AV |
2838 | if (exchange) |
2839 | swap_names(dentry, target); | |
2840 | else | |
2841 | copy_name(dentry, target); | |
1da177e4 | 2842 | |
63cf427a | 2843 | /* ... and switch them in the tree */ |
1da177e4 | 2844 | if (IS_ROOT(dentry)) { |
63cf427a | 2845 | /* splicing a tree */ |
1da177e4 LT |
2846 | dentry->d_parent = target->d_parent; |
2847 | target->d_parent = target; | |
946e51f2 AV |
2848 | list_del_init(&target->d_child); |
2849 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
1da177e4 | 2850 | } else { |
63cf427a | 2851 | /* swapping two dentries */ |
9a8d5bb4 | 2852 | swap(dentry->d_parent, target->d_parent); |
946e51f2 AV |
2853 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
2854 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
63cf427a AV |
2855 | if (exchange) |
2856 | fsnotify_d_move(target); | |
2857 | fsnotify_d_move(dentry); | |
1da177e4 LT |
2858 | } |
2859 | ||
31e6b01f NP |
2860 | write_seqcount_end(&target->d_seq); |
2861 | write_seqcount_end(&dentry->d_seq); | |
2862 | ||
84e710da AV |
2863 | if (dir) |
2864 | end_dir_add(dir, n); | |
986c0194 | 2865 | dentry_unlock_for_move(dentry, target); |
18367501 AV |
2866 | } |
2867 | ||
2868 | /* | |
2869 | * d_move - move a dentry | |
2870 | * @dentry: entry to move | |
2871 | * @target: new dentry | |
2872 | * | |
2873 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2874 | * dcache entries should not be moved in this way. See the locking |
2875 | * requirements for __d_move. | |
18367501 AV |
2876 | */ |
2877 | void d_move(struct dentry *dentry, struct dentry *target) | |
2878 | { | |
2879 | write_seqlock(&rename_lock); | |
da1ce067 | 2880 | __d_move(dentry, target, false); |
1da177e4 | 2881 | write_sequnlock(&rename_lock); |
9eaef27b | 2882 | } |
ec4f8605 | 2883 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2884 | |
da1ce067 MS |
2885 | /* |
2886 | * d_exchange - exchange two dentries | |
2887 | * @dentry1: first dentry | |
2888 | * @dentry2: second dentry | |
2889 | */ | |
2890 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2891 | { | |
2892 | write_seqlock(&rename_lock); | |
2893 | ||
2894 | WARN_ON(!dentry1->d_inode); | |
2895 | WARN_ON(!dentry2->d_inode); | |
2896 | WARN_ON(IS_ROOT(dentry1)); | |
2897 | WARN_ON(IS_ROOT(dentry2)); | |
2898 | ||
2899 | __d_move(dentry1, dentry2, true); | |
2900 | ||
2901 | write_sequnlock(&rename_lock); | |
2902 | } | |
2903 | ||
e2761a11 OH |
2904 | /** |
2905 | * d_ancestor - search for an ancestor | |
2906 | * @p1: ancestor dentry | |
2907 | * @p2: child dentry | |
2908 | * | |
2909 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2910 | * an ancestor of p2, else NULL. | |
9eaef27b | 2911 | */ |
e2761a11 | 2912 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2913 | { |
2914 | struct dentry *p; | |
2915 | ||
871c0067 | 2916 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2917 | if (p->d_parent == p1) |
e2761a11 | 2918 | return p; |
9eaef27b | 2919 | } |
e2761a11 | 2920 | return NULL; |
9eaef27b TM |
2921 | } |
2922 | ||
2923 | /* | |
2924 | * This helper attempts to cope with remotely renamed directories | |
2925 | * | |
2926 | * It assumes that the caller is already holding | |
a03e283b | 2927 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
2928 | * |
2929 | * Note: If ever the locking in lock_rename() changes, then please | |
2930 | * remember to update this too... | |
9eaef27b | 2931 | */ |
b5ae6b15 | 2932 | static int __d_unalias(struct inode *inode, |
873feea0 | 2933 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 2934 | { |
9902af79 AV |
2935 | struct mutex *m1 = NULL; |
2936 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 2937 | int ret = -ESTALE; |
9eaef27b TM |
2938 | |
2939 | /* If alias and dentry share a parent, then no extra locks required */ | |
2940 | if (alias->d_parent == dentry->d_parent) | |
2941 | goto out_unalias; | |
2942 | ||
9eaef27b | 2943 | /* See lock_rename() */ |
9eaef27b TM |
2944 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2945 | goto out_err; | |
2946 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 2947 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 2948 | goto out_err; |
9902af79 | 2949 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 2950 | out_unalias: |
8ed936b5 | 2951 | __d_move(alias, dentry, false); |
b5ae6b15 | 2952 | ret = 0; |
9eaef27b | 2953 | out_err: |
9eaef27b | 2954 | if (m2) |
9902af79 | 2955 | up_read(m2); |
9eaef27b TM |
2956 | if (m1) |
2957 | mutex_unlock(m1); | |
2958 | return ret; | |
2959 | } | |
2960 | ||
3f70bd51 BF |
2961 | /** |
2962 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2963 | * @inode: the inode which may have a disconnected dentry | |
2964 | * @dentry: a negative dentry which we want to point to the inode. | |
2965 | * | |
da093a9b BF |
2966 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
2967 | * place of the given dentry and return it, else simply d_add the inode | |
2968 | * to the dentry and return NULL. | |
3f70bd51 | 2969 | * |
908790fa BF |
2970 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
2971 | * we should error out: directories can't have multiple aliases. | |
2972 | * | |
3f70bd51 BF |
2973 | * This is needed in the lookup routine of any filesystem that is exportable |
2974 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2975 | * | |
2976 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2977 | * is returned. This matches the expected return value of ->lookup. | |
2978 | * | |
2979 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2980 | * In that case, we know that the inode will be a regular file, and also this | |
2981 | * will only occur during atomic_open. So we need to check for the dentry | |
2982 | * being already hashed only in the final case. | |
2983 | */ | |
2984 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
2985 | { | |
3f70bd51 BF |
2986 | if (IS_ERR(inode)) |
2987 | return ERR_CAST(inode); | |
2988 | ||
770bfad8 DH |
2989 | BUG_ON(!d_unhashed(dentry)); |
2990 | ||
de689f5e | 2991 | if (!inode) |
b5ae6b15 | 2992 | goto out; |
de689f5e | 2993 | |
b9680917 | 2994 | security_d_instantiate(dentry, inode); |
873feea0 | 2995 | spin_lock(&inode->i_lock); |
9eaef27b | 2996 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
2997 | struct dentry *new = __d_find_any_alias(inode); |
2998 | if (unlikely(new)) { | |
a03e283b EB |
2999 | /* The reference to new ensures it remains an alias */ |
3000 | spin_unlock(&inode->i_lock); | |
18367501 | 3001 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3002 | if (unlikely(d_ancestor(new, dentry))) { |
3003 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3004 | dput(new); |
3005 | new = ERR_PTR(-ELOOP); | |
3006 | pr_warn_ratelimited( | |
3007 | "VFS: Lookup of '%s' in %s %s" | |
3008 | " would have caused loop\n", | |
3009 | dentry->d_name.name, | |
3010 | inode->i_sb->s_type->name, | |
3011 | inode->i_sb->s_id); | |
3012 | } else if (!IS_ROOT(new)) { | |
3013 | int err = __d_unalias(inode, dentry, new); | |
18367501 | 3014 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3015 | if (err) { |
3016 | dput(new); | |
3017 | new = ERR_PTR(err); | |
3018 | } | |
18367501 | 3019 | } else { |
b5ae6b15 AV |
3020 | __d_move(new, dentry, false); |
3021 | write_sequnlock(&rename_lock); | |
dd179946 | 3022 | } |
b5ae6b15 AV |
3023 | iput(inode); |
3024 | return new; | |
9eaef27b | 3025 | } |
770bfad8 | 3026 | } |
b5ae6b15 | 3027 | out: |
ed782b5a | 3028 | __d_add(dentry, inode); |
b5ae6b15 | 3029 | return NULL; |
770bfad8 | 3030 | } |
b5ae6b15 | 3031 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3032 | |
cdd16d02 | 3033 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
3034 | { |
3035 | *buflen -= namelen; | |
3036 | if (*buflen < 0) | |
3037 | return -ENAMETOOLONG; | |
3038 | *buffer -= namelen; | |
3039 | memcpy(*buffer, str, namelen); | |
3040 | return 0; | |
3041 | } | |
3042 | ||
232d2d60 WL |
3043 | /** |
3044 | * prepend_name - prepend a pathname in front of current buffer pointer | |
18129977 WL |
3045 | * @buffer: buffer pointer |
3046 | * @buflen: allocated length of the buffer | |
3047 | * @name: name string and length qstr structure | |
232d2d60 WL |
3048 | * |
3049 | * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to | |
3050 | * make sure that either the old or the new name pointer and length are | |
3051 | * fetched. However, there may be mismatch between length and pointer. | |
3052 | * The length cannot be trusted, we need to copy it byte-by-byte until | |
3053 | * the length is reached or a null byte is found. It also prepends "/" at | |
3054 | * the beginning of the name. The sequence number check at the caller will | |
3055 | * retry it again when a d_move() does happen. So any garbage in the buffer | |
3056 | * due to mismatched pointer and length will be discarded. | |
6d13f694 AV |
3057 | * |
3058 | * Data dependency barrier is needed to make sure that we see that terminating | |
3059 | * NUL. Alpha strikes again, film at 11... | |
232d2d60 | 3060 | */ |
cdd16d02 MS |
3061 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
3062 | { | |
232d2d60 WL |
3063 | const char *dname = ACCESS_ONCE(name->name); |
3064 | u32 dlen = ACCESS_ONCE(name->len); | |
3065 | char *p; | |
3066 | ||
6d13f694 AV |
3067 | smp_read_barrier_depends(); |
3068 | ||
232d2d60 | 3069 | *buflen -= dlen + 1; |
e825196d AV |
3070 | if (*buflen < 0) |
3071 | return -ENAMETOOLONG; | |
232d2d60 WL |
3072 | p = *buffer -= dlen + 1; |
3073 | *p++ = '/'; | |
3074 | while (dlen--) { | |
3075 | char c = *dname++; | |
3076 | if (!c) | |
3077 | break; | |
3078 | *p++ = c; | |
3079 | } | |
3080 | return 0; | |
cdd16d02 MS |
3081 | } |
3082 | ||
1da177e4 | 3083 | /** |
208898c1 | 3084 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 3085 | * @path: the dentry/vfsmount to report |
02125a82 | 3086 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
3087 | * @buffer: pointer to the end of the buffer |
3088 | * @buflen: pointer to buffer length | |
552ce544 | 3089 | * |
18129977 WL |
3090 | * The function will first try to write out the pathname without taking any |
3091 | * lock other than the RCU read lock to make sure that dentries won't go away. | |
3092 | * It only checks the sequence number of the global rename_lock as any change | |
3093 | * in the dentry's d_seq will be preceded by changes in the rename_lock | |
3094 | * sequence number. If the sequence number had been changed, it will restart | |
3095 | * the whole pathname back-tracing sequence again by taking the rename_lock. | |
3096 | * In this case, there is no need to take the RCU read lock as the recursive | |
3097 | * parent pointer references will keep the dentry chain alive as long as no | |
3098 | * rename operation is performed. | |
1da177e4 | 3099 | */ |
02125a82 AV |
3100 | static int prepend_path(const struct path *path, |
3101 | const struct path *root, | |
f2eb6575 | 3102 | char **buffer, int *buflen) |
1da177e4 | 3103 | { |
ede4cebc AV |
3104 | struct dentry *dentry; |
3105 | struct vfsmount *vfsmnt; | |
3106 | struct mount *mnt; | |
f2eb6575 | 3107 | int error = 0; |
48a066e7 | 3108 | unsigned seq, m_seq = 0; |
232d2d60 WL |
3109 | char *bptr; |
3110 | int blen; | |
6092d048 | 3111 | |
48f5ec21 | 3112 | rcu_read_lock(); |
48a066e7 AV |
3113 | restart_mnt: |
3114 | read_seqbegin_or_lock(&mount_lock, &m_seq); | |
3115 | seq = 0; | |
4ec6c2ae | 3116 | rcu_read_lock(); |
232d2d60 WL |
3117 | restart: |
3118 | bptr = *buffer; | |
3119 | blen = *buflen; | |
48a066e7 | 3120 | error = 0; |
ede4cebc AV |
3121 | dentry = path->dentry; |
3122 | vfsmnt = path->mnt; | |
3123 | mnt = real_mount(vfsmnt); | |
232d2d60 | 3124 | read_seqbegin_or_lock(&rename_lock, &seq); |
f2eb6575 | 3125 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
3126 | struct dentry * parent; |
3127 | ||
1da177e4 | 3128 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
48a066e7 | 3129 | struct mount *parent = ACCESS_ONCE(mnt->mnt_parent); |
cde93be4 EB |
3130 | /* Escaped? */ |
3131 | if (dentry != vfsmnt->mnt_root) { | |
3132 | bptr = *buffer; | |
3133 | blen = *buflen; | |
3134 | error = 3; | |
3135 | break; | |
3136 | } | |
552ce544 | 3137 | /* Global root? */ |
48a066e7 AV |
3138 | if (mnt != parent) { |
3139 | dentry = ACCESS_ONCE(mnt->mnt_mountpoint); | |
3140 | mnt = parent; | |
232d2d60 WL |
3141 | vfsmnt = &mnt->mnt; |
3142 | continue; | |
3143 | } | |
232d2d60 WL |
3144 | if (!error) |
3145 | error = is_mounted(vfsmnt) ? 1 : 2; | |
3146 | break; | |
1da177e4 LT |
3147 | } |
3148 | parent = dentry->d_parent; | |
3149 | prefetch(parent); | |
232d2d60 | 3150 | error = prepend_name(&bptr, &blen, &dentry->d_name); |
f2eb6575 MS |
3151 | if (error) |
3152 | break; | |
3153 | ||
1da177e4 LT |
3154 | dentry = parent; |
3155 | } | |
48f5ec21 AV |
3156 | if (!(seq & 1)) |
3157 | rcu_read_unlock(); | |
3158 | if (need_seqretry(&rename_lock, seq)) { | |
3159 | seq = 1; | |
232d2d60 | 3160 | goto restart; |
48f5ec21 AV |
3161 | } |
3162 | done_seqretry(&rename_lock, seq); | |
4ec6c2ae LZ |
3163 | |
3164 | if (!(m_seq & 1)) | |
3165 | rcu_read_unlock(); | |
48a066e7 AV |
3166 | if (need_seqretry(&mount_lock, m_seq)) { |
3167 | m_seq = 1; | |
3168 | goto restart_mnt; | |
3169 | } | |
3170 | done_seqretry(&mount_lock, m_seq); | |
1da177e4 | 3171 | |
232d2d60 WL |
3172 | if (error >= 0 && bptr == *buffer) { |
3173 | if (--blen < 0) | |
3174 | error = -ENAMETOOLONG; | |
3175 | else | |
3176 | *--bptr = '/'; | |
3177 | } | |
3178 | *buffer = bptr; | |
3179 | *buflen = blen; | |
7ea600b5 | 3180 | return error; |
f2eb6575 | 3181 | } |
be285c71 | 3182 | |
f2eb6575 MS |
3183 | /** |
3184 | * __d_path - return the path of a dentry | |
3185 | * @path: the dentry/vfsmount to report | |
02125a82 | 3186 | * @root: root vfsmnt/dentry |
cd956a1c | 3187 | * @buf: buffer to return value in |
f2eb6575 MS |
3188 | * @buflen: buffer length |
3189 | * | |
ffd1f4ed | 3190 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
3191 | * |
3192 | * Returns a pointer into the buffer or an error code if the | |
3193 | * path was too long. | |
3194 | * | |
be148247 | 3195 | * "buflen" should be positive. |
f2eb6575 | 3196 | * |
02125a82 | 3197 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 3198 | */ |
02125a82 AV |
3199 | char *__d_path(const struct path *path, |
3200 | const struct path *root, | |
f2eb6575 MS |
3201 | char *buf, int buflen) |
3202 | { | |
3203 | char *res = buf + buflen; | |
3204 | int error; | |
3205 | ||
3206 | prepend(&res, &buflen, "\0", 1); | |
f2eb6575 | 3207 | error = prepend_path(path, root, &res, &buflen); |
be148247 | 3208 | |
02125a82 AV |
3209 | if (error < 0) |
3210 | return ERR_PTR(error); | |
3211 | if (error > 0) | |
3212 | return NULL; | |
3213 | return res; | |
3214 | } | |
3215 | ||
3216 | char *d_absolute_path(const struct path *path, | |
3217 | char *buf, int buflen) | |
3218 | { | |
3219 | struct path root = {}; | |
3220 | char *res = buf + buflen; | |
3221 | int error; | |
3222 | ||
3223 | prepend(&res, &buflen, "\0", 1); | |
02125a82 | 3224 | error = prepend_path(path, &root, &res, &buflen); |
02125a82 AV |
3225 | |
3226 | if (error > 1) | |
3227 | error = -EINVAL; | |
3228 | if (error < 0) | |
f2eb6575 | 3229 | return ERR_PTR(error); |
f2eb6575 | 3230 | return res; |
1da177e4 LT |
3231 | } |
3232 | ||
ffd1f4ed MS |
3233 | /* |
3234 | * same as __d_path but appends "(deleted)" for unlinked files. | |
3235 | */ | |
02125a82 AV |
3236 | static int path_with_deleted(const struct path *path, |
3237 | const struct path *root, | |
3238 | char **buf, int *buflen) | |
ffd1f4ed MS |
3239 | { |
3240 | prepend(buf, buflen, "\0", 1); | |
3241 | if (d_unlinked(path->dentry)) { | |
3242 | int error = prepend(buf, buflen, " (deleted)", 10); | |
3243 | if (error) | |
3244 | return error; | |
3245 | } | |
3246 | ||
3247 | return prepend_path(path, root, buf, buflen); | |
3248 | } | |
3249 | ||
8df9d1a4 MS |
3250 | static int prepend_unreachable(char **buffer, int *buflen) |
3251 | { | |
3252 | return prepend(buffer, buflen, "(unreachable)", 13); | |
3253 | } | |
3254 | ||
68f0d9d9 LT |
3255 | static void get_fs_root_rcu(struct fs_struct *fs, struct path *root) |
3256 | { | |
3257 | unsigned seq; | |
3258 | ||
3259 | do { | |
3260 | seq = read_seqcount_begin(&fs->seq); | |
3261 | *root = fs->root; | |
3262 | } while (read_seqcount_retry(&fs->seq, seq)); | |
3263 | } | |
3264 | ||
a03a8a70 JB |
3265 | /** |
3266 | * d_path - return the path of a dentry | |
cf28b486 | 3267 | * @path: path to report |
a03a8a70 JB |
3268 | * @buf: buffer to return value in |
3269 | * @buflen: buffer length | |
3270 | * | |
3271 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
3272 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
3273 | * | |
52afeefb AV |
3274 | * Returns a pointer into the buffer or an error code if the path was |
3275 | * too long. Note: Callers should use the returned pointer, not the passed | |
3276 | * in buffer, to use the name! The implementation often starts at an offset | |
3277 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 3278 | * |
31f3e0b3 | 3279 | * "buflen" should be positive. |
a03a8a70 | 3280 | */ |
20d4fdc1 | 3281 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 3282 | { |
ffd1f4ed | 3283 | char *res = buf + buflen; |
6ac08c39 | 3284 | struct path root; |
ffd1f4ed | 3285 | int error; |
1da177e4 | 3286 | |
c23fbb6b ED |
3287 | /* |
3288 | * We have various synthetic filesystems that never get mounted. On | |
3289 | * these filesystems dentries are never used for lookup purposes, and | |
3290 | * thus don't need to be hashed. They also don't need a name until a | |
3291 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
3292 | * below allows us to generate a name for these objects on demand: | |
f48cfddc EB |
3293 | * |
3294 | * Some pseudo inodes are mountable. When they are mounted | |
3295 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
3296 | * and instead have d_path return the mounted path. | |
c23fbb6b | 3297 | */ |
f48cfddc EB |
3298 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
3299 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 3300 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 3301 | |
68f0d9d9 LT |
3302 | rcu_read_lock(); |
3303 | get_fs_root_rcu(current->fs, &root); | |
02125a82 | 3304 | error = path_with_deleted(path, &root, &res, &buflen); |
68f0d9d9 LT |
3305 | rcu_read_unlock(); |
3306 | ||
02125a82 | 3307 | if (error < 0) |
ffd1f4ed | 3308 | res = ERR_PTR(error); |
1da177e4 LT |
3309 | return res; |
3310 | } | |
ec4f8605 | 3311 | EXPORT_SYMBOL(d_path); |
1da177e4 | 3312 | |
c23fbb6b ED |
3313 | /* |
3314 | * Helper function for dentry_operations.d_dname() members | |
3315 | */ | |
3316 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
3317 | const char *fmt, ...) | |
3318 | { | |
3319 | va_list args; | |
3320 | char temp[64]; | |
3321 | int sz; | |
3322 | ||
3323 | va_start(args, fmt); | |
3324 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
3325 | va_end(args); | |
3326 | ||
3327 | if (sz > sizeof(temp) || sz > buflen) | |
3328 | return ERR_PTR(-ENAMETOOLONG); | |
3329 | ||
3330 | buffer += buflen - sz; | |
3331 | return memcpy(buffer, temp, sz); | |
3332 | } | |
3333 | ||
118b2302 AV |
3334 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
3335 | { | |
3336 | char *end = buffer + buflen; | |
3337 | /* these dentries are never renamed, so d_lock is not needed */ | |
3338 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
232d2d60 | 3339 | prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) || |
118b2302 AV |
3340 | prepend(&end, &buflen, "/", 1)) |
3341 | end = ERR_PTR(-ENAMETOOLONG); | |
232d2d60 | 3342 | return end; |
118b2302 | 3343 | } |
31bbe16f | 3344 | EXPORT_SYMBOL(simple_dname); |
118b2302 | 3345 | |
6092d048 RP |
3346 | /* |
3347 | * Write full pathname from the root of the filesystem into the buffer. | |
3348 | */ | |
f6500801 | 3349 | static char *__dentry_path(struct dentry *d, char *buf, int buflen) |
6092d048 | 3350 | { |
f6500801 | 3351 | struct dentry *dentry; |
232d2d60 WL |
3352 | char *end, *retval; |
3353 | int len, seq = 0; | |
3354 | int error = 0; | |
6092d048 | 3355 | |
f6500801 AV |
3356 | if (buflen < 2) |
3357 | goto Elong; | |
3358 | ||
48f5ec21 | 3359 | rcu_read_lock(); |
232d2d60 | 3360 | restart: |
f6500801 | 3361 | dentry = d; |
232d2d60 WL |
3362 | end = buf + buflen; |
3363 | len = buflen; | |
3364 | prepend(&end, &len, "\0", 1); | |
6092d048 RP |
3365 | /* Get '/' right */ |
3366 | retval = end-1; | |
3367 | *retval = '/'; | |
232d2d60 | 3368 | read_seqbegin_or_lock(&rename_lock, &seq); |
cdd16d02 MS |
3369 | while (!IS_ROOT(dentry)) { |
3370 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 3371 | |
6092d048 | 3372 | prefetch(parent); |
232d2d60 WL |
3373 | error = prepend_name(&end, &len, &dentry->d_name); |
3374 | if (error) | |
3375 | break; | |
6092d048 RP |
3376 | |
3377 | retval = end; | |
3378 | dentry = parent; | |
3379 | } | |
48f5ec21 AV |
3380 | if (!(seq & 1)) |
3381 | rcu_read_unlock(); | |
3382 | if (need_seqretry(&rename_lock, seq)) { | |
3383 | seq = 1; | |
232d2d60 | 3384 | goto restart; |
48f5ec21 AV |
3385 | } |
3386 | done_seqretry(&rename_lock, seq); | |
232d2d60 WL |
3387 | if (error) |
3388 | goto Elong; | |
c103135c AV |
3389 | return retval; |
3390 | Elong: | |
3391 | return ERR_PTR(-ENAMETOOLONG); | |
3392 | } | |
ec2447c2 NP |
3393 | |
3394 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
3395 | { | |
232d2d60 | 3396 | return __dentry_path(dentry, buf, buflen); |
ec2447c2 NP |
3397 | } |
3398 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
3399 | |
3400 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
3401 | { | |
3402 | char *p = NULL; | |
3403 | char *retval; | |
3404 | ||
c103135c AV |
3405 | if (d_unlinked(dentry)) { |
3406 | p = buf + buflen; | |
3407 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
3408 | goto Elong; | |
3409 | buflen++; | |
3410 | } | |
3411 | retval = __dentry_path(dentry, buf, buflen); | |
c103135c AV |
3412 | if (!IS_ERR(retval) && p) |
3413 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
3414 | return retval; |
3415 | Elong: | |
6092d048 RP |
3416 | return ERR_PTR(-ENAMETOOLONG); |
3417 | } | |
3418 | ||
8b19e341 LT |
3419 | static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root, |
3420 | struct path *pwd) | |
5762482f | 3421 | { |
8b19e341 LT |
3422 | unsigned seq; |
3423 | ||
3424 | do { | |
3425 | seq = read_seqcount_begin(&fs->seq); | |
3426 | *root = fs->root; | |
3427 | *pwd = fs->pwd; | |
3428 | } while (read_seqcount_retry(&fs->seq, seq)); | |
5762482f LT |
3429 | } |
3430 | ||
1da177e4 LT |
3431 | /* |
3432 | * NOTE! The user-level library version returns a | |
3433 | * character pointer. The kernel system call just | |
3434 | * returns the length of the buffer filled (which | |
3435 | * includes the ending '\0' character), or a negative | |
3436 | * error value. So libc would do something like | |
3437 | * | |
3438 | * char *getcwd(char * buf, size_t size) | |
3439 | * { | |
3440 | * int retval; | |
3441 | * | |
3442 | * retval = sys_getcwd(buf, size); | |
3443 | * if (retval >= 0) | |
3444 | * return buf; | |
3445 | * errno = -retval; | |
3446 | * return NULL; | |
3447 | * } | |
3448 | */ | |
3cdad428 | 3449 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 3450 | { |
552ce544 | 3451 | int error; |
6ac08c39 | 3452 | struct path pwd, root; |
3272c544 | 3453 | char *page = __getname(); |
1da177e4 LT |
3454 | |
3455 | if (!page) | |
3456 | return -ENOMEM; | |
3457 | ||
8b19e341 LT |
3458 | rcu_read_lock(); |
3459 | get_fs_root_and_pwd_rcu(current->fs, &root, &pwd); | |
1da177e4 | 3460 | |
552ce544 | 3461 | error = -ENOENT; |
f3da392e | 3462 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 3463 | unsigned long len; |
3272c544 LT |
3464 | char *cwd = page + PATH_MAX; |
3465 | int buflen = PATH_MAX; | |
1da177e4 | 3466 | |
8df9d1a4 | 3467 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 3468 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
ff812d72 | 3469 | rcu_read_unlock(); |
552ce544 | 3470 | |
02125a82 | 3471 | if (error < 0) |
552ce544 LT |
3472 | goto out; |
3473 | ||
8df9d1a4 | 3474 | /* Unreachable from current root */ |
02125a82 | 3475 | if (error > 0) { |
8df9d1a4 MS |
3476 | error = prepend_unreachable(&cwd, &buflen); |
3477 | if (error) | |
3478 | goto out; | |
3479 | } | |
3480 | ||
552ce544 | 3481 | error = -ERANGE; |
3272c544 | 3482 | len = PATH_MAX + page - cwd; |
552ce544 LT |
3483 | if (len <= size) { |
3484 | error = len; | |
3485 | if (copy_to_user(buf, cwd, len)) | |
3486 | error = -EFAULT; | |
3487 | } | |
949854d0 | 3488 | } else { |
ff812d72 | 3489 | rcu_read_unlock(); |
949854d0 | 3490 | } |
1da177e4 LT |
3491 | |
3492 | out: | |
3272c544 | 3493 | __putname(page); |
1da177e4 LT |
3494 | return error; |
3495 | } | |
3496 | ||
3497 | /* | |
3498 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3499 | * | |
3500 | * Trivially implemented using the dcache structure | |
3501 | */ | |
3502 | ||
3503 | /** | |
3504 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3505 | * @new_dentry: new dentry | |
3506 | * @old_dentry: old dentry | |
3507 | * | |
a6e5787f YB |
3508 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3509 | * Returns false otherwise. | |
1da177e4 LT |
3510 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3511 | */ | |
3512 | ||
a6e5787f | 3513 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3514 | { |
a6e5787f | 3515 | bool result; |
949854d0 | 3516 | unsigned seq; |
1da177e4 | 3517 | |
e2761a11 | 3518 | if (new_dentry == old_dentry) |
a6e5787f | 3519 | return true; |
e2761a11 | 3520 | |
e2761a11 | 3521 | do { |
1da177e4 | 3522 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3523 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3524 | /* |
3525 | * Need rcu_readlock to protect against the d_parent trashing | |
3526 | * due to d_move | |
3527 | */ | |
3528 | rcu_read_lock(); | |
e2761a11 | 3529 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3530 | result = true; |
e2761a11 | 3531 | else |
a6e5787f | 3532 | result = false; |
949854d0 | 3533 | rcu_read_unlock(); |
1da177e4 | 3534 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3535 | |
3536 | return result; | |
3537 | } | |
3538 | ||
db14fc3a | 3539 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3540 | { |
db14fc3a MS |
3541 | struct dentry *root = data; |
3542 | if (dentry != root) { | |
3543 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3544 | return D_WALK_SKIP; | |
1da177e4 | 3545 | |
01ddc4ed MS |
3546 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3547 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3548 | dentry->d_lockref.count--; | |
3549 | } | |
1da177e4 | 3550 | } |
db14fc3a MS |
3551 | return D_WALK_CONTINUE; |
3552 | } | |
58db63d0 | 3553 | |
db14fc3a MS |
3554 | void d_genocide(struct dentry *parent) |
3555 | { | |
3556 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3557 | } |
3558 | ||
60545d0d | 3559 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3560 | { |
60545d0d AV |
3561 | inode_dec_link_count(inode); |
3562 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
946e51f2 | 3563 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3564 | !d_unlinked(dentry)); |
3565 | spin_lock(&dentry->d_parent->d_lock); | |
3566 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3567 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3568 | (unsigned long long)inode->i_ino); | |
3569 | spin_unlock(&dentry->d_lock); | |
3570 | spin_unlock(&dentry->d_parent->d_lock); | |
3571 | d_instantiate(dentry, inode); | |
1da177e4 | 3572 | } |
60545d0d | 3573 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3574 | |
3575 | static __initdata unsigned long dhash_entries; | |
3576 | static int __init set_dhash_entries(char *str) | |
3577 | { | |
3578 | if (!str) | |
3579 | return 0; | |
3580 | dhash_entries = simple_strtoul(str, &str, 0); | |
3581 | return 1; | |
3582 | } | |
3583 | __setup("dhash_entries=", set_dhash_entries); | |
3584 | ||
3585 | static void __init dcache_init_early(void) | |
3586 | { | |
074b8517 | 3587 | unsigned int loop; |
1da177e4 LT |
3588 | |
3589 | /* If hashes are distributed across NUMA nodes, defer | |
3590 | * hash allocation until vmalloc space is available. | |
3591 | */ | |
3592 | if (hashdist) | |
3593 | return; | |
3594 | ||
3595 | dentry_hashtable = | |
3596 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3597 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3598 | dhash_entries, |
3599 | 13, | |
3600 | HASH_EARLY, | |
3601 | &d_hash_shift, | |
3602 | &d_hash_mask, | |
31fe62b9 | 3603 | 0, |
1da177e4 LT |
3604 | 0); |
3605 | ||
074b8517 | 3606 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3607 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3608 | } |
3609 | ||
74bf17cf | 3610 | static void __init dcache_init(void) |
1da177e4 | 3611 | { |
074b8517 | 3612 | unsigned int loop; |
1da177e4 LT |
3613 | |
3614 | /* | |
3615 | * A constructor could be added for stable state like the lists, | |
3616 | * but it is probably not worth it because of the cache nature | |
3617 | * of the dcache. | |
3618 | */ | |
0a31bd5f | 3619 | dentry_cache = KMEM_CACHE(dentry, |
5d097056 | 3620 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT); |
1da177e4 LT |
3621 | |
3622 | /* Hash may have been set up in dcache_init_early */ | |
3623 | if (!hashdist) | |
3624 | return; | |
3625 | ||
3626 | dentry_hashtable = | |
3627 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3628 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3629 | dhash_entries, |
3630 | 13, | |
3631 | 0, | |
3632 | &d_hash_shift, | |
3633 | &d_hash_mask, | |
31fe62b9 | 3634 | 0, |
1da177e4 LT |
3635 | 0); |
3636 | ||
074b8517 | 3637 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3638 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3639 | } |
3640 | ||
3641 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3642 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3643 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3644 | |
1da177e4 LT |
3645 | EXPORT_SYMBOL(d_genocide); |
3646 | ||
1da177e4 LT |
3647 | void __init vfs_caches_init_early(void) |
3648 | { | |
3649 | dcache_init_early(); | |
3650 | inode_init_early(); | |
3651 | } | |
3652 | ||
4248b0da | 3653 | void __init vfs_caches_init(void) |
1da177e4 | 3654 | { |
1da177e4 | 3655 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, |
20c2df83 | 3656 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3657 | |
74bf17cf DC |
3658 | dcache_init(); |
3659 | inode_init(); | |
4248b0da MG |
3660 | files_init(); |
3661 | files_maxfiles_init(); | |
74bf17cf | 3662 | mnt_init(); |
1da177e4 LT |
3663 | bdev_cache_init(); |
3664 | chrdev_init(); | |
3665 | } |