]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0-only | |
2 | /* | |
3 | * fs/dcache.c | |
4 | * | |
5 | * Complete reimplementation | |
6 | * (C) 1997 Thomas Schoebel-Theuer, | |
7 | * with heavy changes by Linus Torvalds | |
8 | */ | |
9 | ||
10 | /* | |
11 | * Notes on the allocation strategy: | |
12 | * | |
13 | * The dcache is a master of the icache - whenever a dcache entry | |
14 | * exists, the inode will always exist. "iput()" is done either when | |
15 | * the dcache entry is deleted or garbage collected. | |
16 | */ | |
17 | ||
18 | #include <linux/ratelimit.h> | |
19 | #include <linux/string.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/fscrypt.h> | |
23 | #include <linux/fsnotify.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/init.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/cache.h> | |
28 | #include <linux/export.h> | |
29 | #include <linux/security.h> | |
30 | #include <linux/seqlock.h> | |
31 | #include <linux/memblock.h> | |
32 | #include <linux/bit_spinlock.h> | |
33 | #include <linux/rculist_bl.h> | |
34 | #include <linux/list_lru.h> | |
35 | #include "internal.h" | |
36 | #include "mount.h" | |
37 | ||
38 | /* | |
39 | * Usage: | |
40 | * dcache->d_inode->i_lock protects: | |
41 | * - i_dentry, d_u.d_alias, d_inode of aliases | |
42 | * dcache_hash_bucket lock protects: | |
43 | * - the dcache hash table | |
44 | * s_roots bl list spinlock protects: | |
45 | * - the s_roots list (see __d_drop) | |
46 | * dentry->d_sb->s_dentry_lru_lock protects: | |
47 | * - the dcache lru lists and counters | |
48 | * d_lock protects: | |
49 | * - d_flags | |
50 | * - d_name | |
51 | * - d_lru | |
52 | * - d_count | |
53 | * - d_unhashed() | |
54 | * - d_parent and d_subdirs | |
55 | * - childrens' d_child and d_parent | |
56 | * - d_u.d_alias, d_inode | |
57 | * | |
58 | * Ordering: | |
59 | * dentry->d_inode->i_lock | |
60 | * dentry->d_lock | |
61 | * dentry->d_sb->s_dentry_lru_lock | |
62 | * dcache_hash_bucket lock | |
63 | * s_roots lock | |
64 | * | |
65 | * If there is an ancestor relationship: | |
66 | * dentry->d_parent->...->d_parent->d_lock | |
67 | * ... | |
68 | * dentry->d_parent->d_lock | |
69 | * dentry->d_lock | |
70 | * | |
71 | * If no ancestor relationship: | |
72 | * arbitrary, since it's serialized on rename_lock | |
73 | */ | |
74 | int sysctl_vfs_cache_pressure __read_mostly = 100; | |
75 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); | |
76 | ||
77 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); | |
78 | ||
79 | EXPORT_SYMBOL(rename_lock); | |
80 | ||
81 | static struct kmem_cache *dentry_cache __read_mostly; | |
82 | ||
83 | const struct qstr empty_name = QSTR_INIT("", 0); | |
84 | EXPORT_SYMBOL(empty_name); | |
85 | const struct qstr slash_name = QSTR_INIT("/", 1); | |
86 | EXPORT_SYMBOL(slash_name); | |
87 | ||
88 | /* | |
89 | * This is the single most critical data structure when it comes | |
90 | * to the dcache: the hashtable for lookups. Somebody should try | |
91 | * to make this good - I've just made it work. | |
92 | * | |
93 | * This hash-function tries to avoid losing too many bits of hash | |
94 | * information, yet avoid using a prime hash-size or similar. | |
95 | */ | |
96 | ||
97 | static unsigned int d_hash_shift __read_mostly; | |
98 | ||
99 | static struct hlist_bl_head *dentry_hashtable __read_mostly; | |
100 | ||
101 | static inline struct hlist_bl_head *d_hash(unsigned int hash) | |
102 | { | |
103 | return dentry_hashtable + (hash >> d_hash_shift); | |
104 | } | |
105 | ||
106 | #define IN_LOOKUP_SHIFT 10 | |
107 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
108 | ||
109 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
110 | unsigned int hash) | |
111 | { | |
112 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
113 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
114 | } | |
115 | ||
116 | ||
117 | /* Statistics gathering. */ | |
118 | struct dentry_stat_t dentry_stat = { | |
119 | .age_limit = 45, | |
120 | }; | |
121 | ||
122 | static DEFINE_PER_CPU(long, nr_dentry); | |
123 | static DEFINE_PER_CPU(long, nr_dentry_unused); | |
124 | static DEFINE_PER_CPU(long, nr_dentry_negative); | |
125 | ||
126 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
127 | ||
128 | /* | |
129 | * Here we resort to our own counters instead of using generic per-cpu counters | |
130 | * for consistency with what the vfs inode code does. We are expected to harvest | |
131 | * better code and performance by having our own specialized counters. | |
132 | * | |
133 | * Please note that the loop is done over all possible CPUs, not over all online | |
134 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
135 | * on and off. If one of them goes off, we will just keep their counters. | |
136 | * | |
137 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
138 | * please update all vfs counters to match. | |
139 | */ | |
140 | static long get_nr_dentry(void) | |
141 | { | |
142 | int i; | |
143 | long sum = 0; | |
144 | for_each_possible_cpu(i) | |
145 | sum += per_cpu(nr_dentry, i); | |
146 | return sum < 0 ? 0 : sum; | |
147 | } | |
148 | ||
149 | static long get_nr_dentry_unused(void) | |
150 | { | |
151 | int i; | |
152 | long sum = 0; | |
153 | for_each_possible_cpu(i) | |
154 | sum += per_cpu(nr_dentry_unused, i); | |
155 | return sum < 0 ? 0 : sum; | |
156 | } | |
157 | ||
158 | static long get_nr_dentry_negative(void) | |
159 | { | |
160 | int i; | |
161 | long sum = 0; | |
162 | ||
163 | for_each_possible_cpu(i) | |
164 | sum += per_cpu(nr_dentry_negative, i); | |
165 | return sum < 0 ? 0 : sum; | |
166 | } | |
167 | ||
168 | int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer, | |
169 | size_t *lenp, loff_t *ppos) | |
170 | { | |
171 | dentry_stat.nr_dentry = get_nr_dentry(); | |
172 | dentry_stat.nr_unused = get_nr_dentry_unused(); | |
173 | dentry_stat.nr_negative = get_nr_dentry_negative(); | |
174 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); | |
175 | } | |
176 | #endif | |
177 | ||
178 | /* | |
179 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
180 | * The strings are both count bytes long, and count is non-zero. | |
181 | */ | |
182 | #ifdef CONFIG_DCACHE_WORD_ACCESS | |
183 | ||
184 | #include <asm/word-at-a-time.h> | |
185 | /* | |
186 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
187 | * aligned allocation for this particular component. We don't | |
188 | * strictly need the load_unaligned_zeropad() safety, but it | |
189 | * doesn't hurt either. | |
190 | * | |
191 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
192 | * need the careful unaligned handling. | |
193 | */ | |
194 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) | |
195 | { | |
196 | unsigned long a,b,mask; | |
197 | ||
198 | for (;;) { | |
199 | a = read_word_at_a_time(cs); | |
200 | b = load_unaligned_zeropad(ct); | |
201 | if (tcount < sizeof(unsigned long)) | |
202 | break; | |
203 | if (unlikely(a != b)) | |
204 | return 1; | |
205 | cs += sizeof(unsigned long); | |
206 | ct += sizeof(unsigned long); | |
207 | tcount -= sizeof(unsigned long); | |
208 | if (!tcount) | |
209 | return 0; | |
210 | } | |
211 | mask = bytemask_from_count(tcount); | |
212 | return unlikely(!!((a ^ b) & mask)); | |
213 | } | |
214 | ||
215 | #else | |
216 | ||
217 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) | |
218 | { | |
219 | do { | |
220 | if (*cs != *ct) | |
221 | return 1; | |
222 | cs++; | |
223 | ct++; | |
224 | tcount--; | |
225 | } while (tcount); | |
226 | return 0; | |
227 | } | |
228 | ||
229 | #endif | |
230 | ||
231 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) | |
232 | { | |
233 | /* | |
234 | * Be careful about RCU walk racing with rename: | |
235 | * use 'READ_ONCE' to fetch the name pointer. | |
236 | * | |
237 | * NOTE! Even if a rename will mean that the length | |
238 | * was not loaded atomically, we don't care. The | |
239 | * RCU walk will check the sequence count eventually, | |
240 | * and catch it. And we won't overrun the buffer, | |
241 | * because we're reading the name pointer atomically, | |
242 | * and a dentry name is guaranteed to be properly | |
243 | * terminated with a NUL byte. | |
244 | * | |
245 | * End result: even if 'len' is wrong, we'll exit | |
246 | * early because the data cannot match (there can | |
247 | * be no NUL in the ct/tcount data) | |
248 | */ | |
249 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); | |
250 | ||
251 | return dentry_string_cmp(cs, ct, tcount); | |
252 | } | |
253 | ||
254 | struct external_name { | |
255 | union { | |
256 | atomic_t count; | |
257 | struct rcu_head head; | |
258 | } u; | |
259 | unsigned char name[]; | |
260 | }; | |
261 | ||
262 | static inline struct external_name *external_name(struct dentry *dentry) | |
263 | { | |
264 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
265 | } | |
266 | ||
267 | static void __d_free(struct rcu_head *head) | |
268 | { | |
269 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
270 | ||
271 | kmem_cache_free(dentry_cache, dentry); | |
272 | } | |
273 | ||
274 | static void __d_free_external(struct rcu_head *head) | |
275 | { | |
276 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
277 | kfree(external_name(dentry)); | |
278 | kmem_cache_free(dentry_cache, dentry); | |
279 | } | |
280 | ||
281 | static inline int dname_external(const struct dentry *dentry) | |
282 | { | |
283 | return dentry->d_name.name != dentry->d_iname; | |
284 | } | |
285 | ||
286 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) | |
287 | { | |
288 | spin_lock(&dentry->d_lock); | |
289 | name->name = dentry->d_name; | |
290 | if (unlikely(dname_external(dentry))) { | |
291 | atomic_inc(&external_name(dentry)->u.count); | |
292 | } else { | |
293 | memcpy(name->inline_name, dentry->d_iname, | |
294 | dentry->d_name.len + 1); | |
295 | name->name.name = name->inline_name; | |
296 | } | |
297 | spin_unlock(&dentry->d_lock); | |
298 | } | |
299 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
300 | ||
301 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
302 | { | |
303 | if (unlikely(name->name.name != name->inline_name)) { | |
304 | struct external_name *p; | |
305 | p = container_of(name->name.name, struct external_name, name[0]); | |
306 | if (unlikely(atomic_dec_and_test(&p->u.count))) | |
307 | kfree_rcu(p, u.head); | |
308 | } | |
309 | } | |
310 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
311 | ||
312 | static inline void __d_set_inode_and_type(struct dentry *dentry, | |
313 | struct inode *inode, | |
314 | unsigned type_flags) | |
315 | { | |
316 | unsigned flags; | |
317 | ||
318 | dentry->d_inode = inode; | |
319 | flags = READ_ONCE(dentry->d_flags); | |
320 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
321 | flags |= type_flags; | |
322 | WRITE_ONCE(dentry->d_flags, flags); | |
323 | } | |
324 | ||
325 | static inline void __d_clear_type_and_inode(struct dentry *dentry) | |
326 | { | |
327 | unsigned flags = READ_ONCE(dentry->d_flags); | |
328 | ||
329 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
330 | WRITE_ONCE(dentry->d_flags, flags); | |
331 | dentry->d_inode = NULL; | |
332 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
333 | this_cpu_inc(nr_dentry_negative); | |
334 | } | |
335 | ||
336 | static void dentry_free(struct dentry *dentry) | |
337 | { | |
338 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); | |
339 | if (unlikely(dname_external(dentry))) { | |
340 | struct external_name *p = external_name(dentry); | |
341 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
342 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
343 | return; | |
344 | } | |
345 | } | |
346 | /* if dentry was never visible to RCU, immediate free is OK */ | |
347 | if (dentry->d_flags & DCACHE_NORCU) | |
348 | __d_free(&dentry->d_u.d_rcu); | |
349 | else | |
350 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
351 | } | |
352 | ||
353 | /* | |
354 | * Release the dentry's inode, using the filesystem | |
355 | * d_iput() operation if defined. | |
356 | */ | |
357 | static void dentry_unlink_inode(struct dentry * dentry) | |
358 | __releases(dentry->d_lock) | |
359 | __releases(dentry->d_inode->i_lock) | |
360 | { | |
361 | struct inode *inode = dentry->d_inode; | |
362 | ||
363 | raw_write_seqcount_begin(&dentry->d_seq); | |
364 | __d_clear_type_and_inode(dentry); | |
365 | hlist_del_init(&dentry->d_u.d_alias); | |
366 | raw_write_seqcount_end(&dentry->d_seq); | |
367 | spin_unlock(&dentry->d_lock); | |
368 | spin_unlock(&inode->i_lock); | |
369 | if (!inode->i_nlink) | |
370 | fsnotify_inoderemove(inode); | |
371 | if (dentry->d_op && dentry->d_op->d_iput) | |
372 | dentry->d_op->d_iput(dentry, inode); | |
373 | else | |
374 | iput(inode); | |
375 | } | |
376 | ||
377 | /* | |
378 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
379 | * is in use - which includes both the "real" per-superblock | |
380 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
381 | * | |
382 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
383 | * on the shrink list (ie not on the superblock LRU list). | |
384 | * | |
385 | * The per-cpu "nr_dentry_unused" counters are updated with | |
386 | * the DCACHE_LRU_LIST bit. | |
387 | * | |
388 | * The per-cpu "nr_dentry_negative" counters are only updated | |
389 | * when deleted from or added to the per-superblock LRU list, not | |
390 | * from/to the shrink list. That is to avoid an unneeded dec/inc | |
391 | * pair when moving from LRU to shrink list in select_collect(). | |
392 | * | |
393 | * These helper functions make sure we always follow the | |
394 | * rules. d_lock must be held by the caller. | |
395 | */ | |
396 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
397 | static void d_lru_add(struct dentry *dentry) | |
398 | { | |
399 | D_FLAG_VERIFY(dentry, 0); | |
400 | dentry->d_flags |= DCACHE_LRU_LIST; | |
401 | this_cpu_inc(nr_dentry_unused); | |
402 | if (d_is_negative(dentry)) | |
403 | this_cpu_inc(nr_dentry_negative); | |
404 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
405 | } | |
406 | ||
407 | static void d_lru_del(struct dentry *dentry) | |
408 | { | |
409 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
410 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
411 | this_cpu_dec(nr_dentry_unused); | |
412 | if (d_is_negative(dentry)) | |
413 | this_cpu_dec(nr_dentry_negative); | |
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 | */ | |
439 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) | |
440 | { | |
441 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
442 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
443 | this_cpu_dec(nr_dentry_unused); | |
444 | if (d_is_negative(dentry)) | |
445 | this_cpu_dec(nr_dentry_negative); | |
446 | list_lru_isolate(lru, &dentry->d_lru); | |
447 | } | |
448 | ||
449 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, | |
450 | struct list_head *list) | |
451 | { | |
452 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
453 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
454 | if (d_is_negative(dentry)) | |
455 | this_cpu_dec(nr_dentry_negative); | |
456 | list_lru_isolate_move(lru, &dentry->d_lru, list); | |
457 | } | |
458 | ||
459 | /** | |
460 | * d_drop - drop a dentry | |
461 | * @dentry: dentry to drop | |
462 | * | |
463 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
464 | * be found through a VFS lookup any more. Note that this is different from | |
465 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
466 | * possible, giving a successful _negative_ lookup, while d_drop will | |
467 | * just make the cache lookup fail. | |
468 | * | |
469 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
470 | * reason (NFS timeouts or autofs deletes). | |
471 | * | |
472 | * __d_drop requires dentry->d_lock | |
473 | * ___d_drop doesn't mark dentry as "unhashed" | |
474 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
475 | */ | |
476 | static void ___d_drop(struct dentry *dentry) | |
477 | { | |
478 | struct hlist_bl_head *b; | |
479 | /* | |
480 | * Hashed dentries are normally on the dentry hashtable, | |
481 | * with the exception of those newly allocated by | |
482 | * d_obtain_root, which are always IS_ROOT: | |
483 | */ | |
484 | if (unlikely(IS_ROOT(dentry))) | |
485 | b = &dentry->d_sb->s_roots; | |
486 | else | |
487 | b = d_hash(dentry->d_name.hash); | |
488 | ||
489 | hlist_bl_lock(b); | |
490 | __hlist_bl_del(&dentry->d_hash); | |
491 | hlist_bl_unlock(b); | |
492 | } | |
493 | ||
494 | void __d_drop(struct dentry *dentry) | |
495 | { | |
496 | if (!d_unhashed(dentry)) { | |
497 | ___d_drop(dentry); | |
498 | dentry->d_hash.pprev = NULL; | |
499 | write_seqcount_invalidate(&dentry->d_seq); | |
500 | } | |
501 | } | |
502 | EXPORT_SYMBOL(__d_drop); | |
503 | ||
504 | void d_drop(struct dentry *dentry) | |
505 | { | |
506 | spin_lock(&dentry->d_lock); | |
507 | __d_drop(dentry); | |
508 | spin_unlock(&dentry->d_lock); | |
509 | } | |
510 | EXPORT_SYMBOL(d_drop); | |
511 | ||
512 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) | |
513 | { | |
514 | struct dentry *next; | |
515 | /* | |
516 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
517 | * attached to the dentry tree | |
518 | */ | |
519 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
520 | if (unlikely(list_empty(&dentry->d_child))) | |
521 | return; | |
522 | __list_del_entry(&dentry->d_child); | |
523 | /* | |
524 | * Cursors can move around the list of children. While we'd been | |
525 | * a normal list member, it didn't matter - ->d_child.next would've | |
526 | * been updated. However, from now on it won't be and for the | |
527 | * things like d_walk() it might end up with a nasty surprise. | |
528 | * Normally d_walk() doesn't care about cursors moving around - | |
529 | * ->d_lock on parent prevents that and since a cursor has no children | |
530 | * of its own, we get through it without ever unlocking the parent. | |
531 | * There is one exception, though - if we ascend from a child that | |
532 | * gets killed as soon as we unlock it, the next sibling is found | |
533 | * using the value left in its ->d_child.next. And if _that_ | |
534 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
535 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
536 | * everything the cursor had been moved past. | |
537 | * | |
538 | * Solution: make sure that the pointer left behind in ->d_child.next | |
539 | * points to something that won't be moving around. I.e. skip the | |
540 | * cursors. | |
541 | */ | |
542 | while (dentry->d_child.next != &parent->d_subdirs) { | |
543 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
544 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
545 | break; | |
546 | dentry->d_child.next = next->d_child.next; | |
547 | } | |
548 | } | |
549 | ||
550 | static void __dentry_kill(struct dentry *dentry) | |
551 | { | |
552 | struct dentry *parent = NULL; | |
553 | bool can_free = true; | |
554 | if (!IS_ROOT(dentry)) | |
555 | parent = dentry->d_parent; | |
556 | ||
557 | /* | |
558 | * The dentry is now unrecoverably dead to the world. | |
559 | */ | |
560 | lockref_mark_dead(&dentry->d_lockref); | |
561 | ||
562 | /* | |
563 | * inform the fs via d_prune that this dentry is about to be | |
564 | * unhashed and destroyed. | |
565 | */ | |
566 | if (dentry->d_flags & DCACHE_OP_PRUNE) | |
567 | dentry->d_op->d_prune(dentry); | |
568 | ||
569 | if (dentry->d_flags & DCACHE_LRU_LIST) { | |
570 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
571 | d_lru_del(dentry); | |
572 | } | |
573 | /* if it was on the hash then remove it */ | |
574 | __d_drop(dentry); | |
575 | dentry_unlist(dentry, parent); | |
576 | if (parent) | |
577 | spin_unlock(&parent->d_lock); | |
578 | if (dentry->d_inode) | |
579 | dentry_unlink_inode(dentry); | |
580 | else | |
581 | spin_unlock(&dentry->d_lock); | |
582 | this_cpu_dec(nr_dentry); | |
583 | if (dentry->d_op && dentry->d_op->d_release) | |
584 | dentry->d_op->d_release(dentry); | |
585 | ||
586 | spin_lock(&dentry->d_lock); | |
587 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
588 | dentry->d_flags |= DCACHE_MAY_FREE; | |
589 | can_free = false; | |
590 | } | |
591 | spin_unlock(&dentry->d_lock); | |
592 | if (likely(can_free)) | |
593 | dentry_free(dentry); | |
594 | cond_resched(); | |
595 | } | |
596 | ||
597 | static struct dentry *__lock_parent(struct dentry *dentry) | |
598 | { | |
599 | struct dentry *parent; | |
600 | rcu_read_lock(); | |
601 | spin_unlock(&dentry->d_lock); | |
602 | again: | |
603 | parent = READ_ONCE(dentry->d_parent); | |
604 | spin_lock(&parent->d_lock); | |
605 | /* | |
606 | * We can't blindly lock dentry until we are sure | |
607 | * that we won't violate the locking order. | |
608 | * Any changes of dentry->d_parent must have | |
609 | * been done with parent->d_lock held, so | |
610 | * spin_lock() above is enough of a barrier | |
611 | * for checking if it's still our child. | |
612 | */ | |
613 | if (unlikely(parent != dentry->d_parent)) { | |
614 | spin_unlock(&parent->d_lock); | |
615 | goto again; | |
616 | } | |
617 | rcu_read_unlock(); | |
618 | if (parent != dentry) | |
619 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
620 | else | |
621 | parent = NULL; | |
622 | return parent; | |
623 | } | |
624 | ||
625 | static inline struct dentry *lock_parent(struct dentry *dentry) | |
626 | { | |
627 | struct dentry *parent = dentry->d_parent; | |
628 | if (IS_ROOT(dentry)) | |
629 | return NULL; | |
630 | if (likely(spin_trylock(&parent->d_lock))) | |
631 | return parent; | |
632 | return __lock_parent(dentry); | |
633 | } | |
634 | ||
635 | static inline bool retain_dentry(struct dentry *dentry) | |
636 | { | |
637 | WARN_ON(d_in_lookup(dentry)); | |
638 | ||
639 | /* Unreachable? Get rid of it */ | |
640 | if (unlikely(d_unhashed(dentry))) | |
641 | return false; | |
642 | ||
643 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
644 | return false; | |
645 | ||
646 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
647 | if (dentry->d_op->d_delete(dentry)) | |
648 | return false; | |
649 | } | |
650 | /* retain; LRU fodder */ | |
651 | dentry->d_lockref.count--; | |
652 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | |
653 | d_lru_add(dentry); | |
654 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) | |
655 | dentry->d_flags |= DCACHE_REFERENCED; | |
656 | return true; | |
657 | } | |
658 | ||
659 | /* | |
660 | * Finish off a dentry we've decided to kill. | |
661 | * dentry->d_lock must be held, returns with it unlocked. | |
662 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
663 | */ | |
664 | static struct dentry *dentry_kill(struct dentry *dentry) | |
665 | __releases(dentry->d_lock) | |
666 | { | |
667 | struct inode *inode = dentry->d_inode; | |
668 | struct dentry *parent = NULL; | |
669 | ||
670 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
671 | goto slow_positive; | |
672 | ||
673 | if (!IS_ROOT(dentry)) { | |
674 | parent = dentry->d_parent; | |
675 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
676 | parent = __lock_parent(dentry); | |
677 | if (likely(inode || !dentry->d_inode)) | |
678 | goto got_locks; | |
679 | /* negative that became positive */ | |
680 | if (parent) | |
681 | spin_unlock(&parent->d_lock); | |
682 | inode = dentry->d_inode; | |
683 | goto slow_positive; | |
684 | } | |
685 | } | |
686 | __dentry_kill(dentry); | |
687 | return parent; | |
688 | ||
689 | slow_positive: | |
690 | spin_unlock(&dentry->d_lock); | |
691 | spin_lock(&inode->i_lock); | |
692 | spin_lock(&dentry->d_lock); | |
693 | parent = lock_parent(dentry); | |
694 | got_locks: | |
695 | if (unlikely(dentry->d_lockref.count != 1)) { | |
696 | dentry->d_lockref.count--; | |
697 | } else if (likely(!retain_dentry(dentry))) { | |
698 | __dentry_kill(dentry); | |
699 | return parent; | |
700 | } | |
701 | /* we are keeping it, after all */ | |
702 | if (inode) | |
703 | spin_unlock(&inode->i_lock); | |
704 | if (parent) | |
705 | spin_unlock(&parent->d_lock); | |
706 | spin_unlock(&dentry->d_lock); | |
707 | return NULL; | |
708 | } | |
709 | ||
710 | /* | |
711 | * Try to do a lockless dput(), and return whether that was successful. | |
712 | * | |
713 | * If unsuccessful, we return false, having already taken the dentry lock. | |
714 | * | |
715 | * The caller needs to hold the RCU read lock, so that the dentry is | |
716 | * guaranteed to stay around even if the refcount goes down to zero! | |
717 | */ | |
718 | static inline bool fast_dput(struct dentry *dentry) | |
719 | { | |
720 | int ret; | |
721 | unsigned int d_flags; | |
722 | ||
723 | /* | |
724 | * If we have a d_op->d_delete() operation, we sould not | |
725 | * let the dentry count go to zero, so use "put_or_lock". | |
726 | */ | |
727 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
728 | return lockref_put_or_lock(&dentry->d_lockref); | |
729 | ||
730 | /* | |
731 | * .. otherwise, we can try to just decrement the | |
732 | * lockref optimistically. | |
733 | */ | |
734 | ret = lockref_put_return(&dentry->d_lockref); | |
735 | ||
736 | /* | |
737 | * If the lockref_put_return() failed due to the lock being held | |
738 | * by somebody else, the fast path has failed. We will need to | |
739 | * get the lock, and then check the count again. | |
740 | */ | |
741 | if (unlikely(ret < 0)) { | |
742 | spin_lock(&dentry->d_lock); | |
743 | if (dentry->d_lockref.count > 1) { | |
744 | dentry->d_lockref.count--; | |
745 | spin_unlock(&dentry->d_lock); | |
746 | return true; | |
747 | } | |
748 | return false; | |
749 | } | |
750 | ||
751 | /* | |
752 | * If we weren't the last ref, we're done. | |
753 | */ | |
754 | if (ret) | |
755 | return true; | |
756 | ||
757 | /* | |
758 | * Careful, careful. The reference count went down | |
759 | * to zero, but we don't hold the dentry lock, so | |
760 | * somebody else could get it again, and do another | |
761 | * dput(), and we need to not race with that. | |
762 | * | |
763 | * However, there is a very special and common case | |
764 | * where we don't care, because there is nothing to | |
765 | * do: the dentry is still hashed, it does not have | |
766 | * a 'delete' op, and it's referenced and already on | |
767 | * the LRU list. | |
768 | * | |
769 | * NOTE! Since we aren't locked, these values are | |
770 | * not "stable". However, it is sufficient that at | |
771 | * some point after we dropped the reference the | |
772 | * dentry was hashed and the flags had the proper | |
773 | * value. Other dentry users may have re-gotten | |
774 | * a reference to the dentry and change that, but | |
775 | * our work is done - we can leave the dentry | |
776 | * around with a zero refcount. | |
777 | */ | |
778 | smp_rmb(); | |
779 | d_flags = READ_ONCE(dentry->d_flags); | |
780 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED; | |
781 | ||
782 | /* Nothing to do? Dropping the reference was all we needed? */ | |
783 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
784 | return true; | |
785 | ||
786 | /* | |
787 | * Not the fast normal case? Get the lock. We've already decremented | |
788 | * the refcount, but we'll need to re-check the situation after | |
789 | * getting the lock. | |
790 | */ | |
791 | spin_lock(&dentry->d_lock); | |
792 | ||
793 | /* | |
794 | * Did somebody else grab a reference to it in the meantime, and | |
795 | * we're no longer the last user after all? Alternatively, somebody | |
796 | * else could have killed it and marked it dead. Either way, we | |
797 | * don't need to do anything else. | |
798 | */ | |
799 | if (dentry->d_lockref.count) { | |
800 | spin_unlock(&dentry->d_lock); | |
801 | return true; | |
802 | } | |
803 | ||
804 | /* | |
805 | * Re-get the reference we optimistically dropped. We hold the | |
806 | * lock, and we just tested that it was zero, so we can just | |
807 | * set it to 1. | |
808 | */ | |
809 | dentry->d_lockref.count = 1; | |
810 | return false; | |
811 | } | |
812 | ||
813 | ||
814 | /* | |
815 | * This is dput | |
816 | * | |
817 | * This is complicated by the fact that we do not want to put | |
818 | * dentries that are no longer on any hash chain on the unused | |
819 | * list: we'd much rather just get rid of them immediately. | |
820 | * | |
821 | * However, that implies that we have to traverse the dentry | |
822 | * tree upwards to the parents which might _also_ now be | |
823 | * scheduled for deletion (it may have been only waiting for | |
824 | * its last child to go away). | |
825 | * | |
826 | * This tail recursion is done by hand as we don't want to depend | |
827 | * on the compiler to always get this right (gcc generally doesn't). | |
828 | * Real recursion would eat up our stack space. | |
829 | */ | |
830 | ||
831 | /* | |
832 | * dput - release a dentry | |
833 | * @dentry: dentry to release | |
834 | * | |
835 | * Release a dentry. This will drop the usage count and if appropriate | |
836 | * call the dentry unlink method as well as removing it from the queues and | |
837 | * releasing its resources. If the parent dentries were scheduled for release | |
838 | * they too may now get deleted. | |
839 | */ | |
840 | void dput(struct dentry *dentry) | |
841 | { | |
842 | while (dentry) { | |
843 | might_sleep(); | |
844 | ||
845 | rcu_read_lock(); | |
846 | if (likely(fast_dput(dentry))) { | |
847 | rcu_read_unlock(); | |
848 | return; | |
849 | } | |
850 | ||
851 | /* Slow case: now with the dentry lock held */ | |
852 | rcu_read_unlock(); | |
853 | ||
854 | if (likely(retain_dentry(dentry))) { | |
855 | spin_unlock(&dentry->d_lock); | |
856 | return; | |
857 | } | |
858 | ||
859 | dentry = dentry_kill(dentry); | |
860 | } | |
861 | } | |
862 | EXPORT_SYMBOL(dput); | |
863 | ||
864 | static void __dput_to_list(struct dentry *dentry, struct list_head *list) | |
865 | __must_hold(&dentry->d_lock) | |
866 | { | |
867 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
868 | /* let the owner of the list it's on deal with it */ | |
869 | --dentry->d_lockref.count; | |
870 | } else { | |
871 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
872 | d_lru_del(dentry); | |
873 | if (!--dentry->d_lockref.count) | |
874 | d_shrink_add(dentry, list); | |
875 | } | |
876 | } | |
877 | ||
878 | void dput_to_list(struct dentry *dentry, struct list_head *list) | |
879 | { | |
880 | rcu_read_lock(); | |
881 | if (likely(fast_dput(dentry))) { | |
882 | rcu_read_unlock(); | |
883 | return; | |
884 | } | |
885 | rcu_read_unlock(); | |
886 | if (!retain_dentry(dentry)) | |
887 | __dput_to_list(dentry, list); | |
888 | spin_unlock(&dentry->d_lock); | |
889 | } | |
890 | ||
891 | /* This must be called with d_lock held */ | |
892 | static inline void __dget_dlock(struct dentry *dentry) | |
893 | { | |
894 | dentry->d_lockref.count++; | |
895 | } | |
896 | ||
897 | static inline void __dget(struct dentry *dentry) | |
898 | { | |
899 | lockref_get(&dentry->d_lockref); | |
900 | } | |
901 | ||
902 | struct dentry *dget_parent(struct dentry *dentry) | |
903 | { | |
904 | int gotref; | |
905 | struct dentry *ret; | |
906 | unsigned seq; | |
907 | ||
908 | /* | |
909 | * Do optimistic parent lookup without any | |
910 | * locking. | |
911 | */ | |
912 | rcu_read_lock(); | |
913 | seq = raw_seqcount_begin(&dentry->d_seq); | |
914 | ret = READ_ONCE(dentry->d_parent); | |
915 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
916 | rcu_read_unlock(); | |
917 | if (likely(gotref)) { | |
918 | if (!read_seqcount_retry(&dentry->d_seq, seq)) | |
919 | return ret; | |
920 | dput(ret); | |
921 | } | |
922 | ||
923 | repeat: | |
924 | /* | |
925 | * Don't need rcu_dereference because we re-check it was correct under | |
926 | * the lock. | |
927 | */ | |
928 | rcu_read_lock(); | |
929 | ret = dentry->d_parent; | |
930 | spin_lock(&ret->d_lock); | |
931 | if (unlikely(ret != dentry->d_parent)) { | |
932 | spin_unlock(&ret->d_lock); | |
933 | rcu_read_unlock(); | |
934 | goto repeat; | |
935 | } | |
936 | rcu_read_unlock(); | |
937 | BUG_ON(!ret->d_lockref.count); | |
938 | ret->d_lockref.count++; | |
939 | spin_unlock(&ret->d_lock); | |
940 | return ret; | |
941 | } | |
942 | EXPORT_SYMBOL(dget_parent); | |
943 | ||
944 | static struct dentry * __d_find_any_alias(struct inode *inode) | |
945 | { | |
946 | struct dentry *alias; | |
947 | ||
948 | if (hlist_empty(&inode->i_dentry)) | |
949 | return NULL; | |
950 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); | |
951 | __dget(alias); | |
952 | return alias; | |
953 | } | |
954 | ||
955 | /** | |
956 | * d_find_any_alias - find any alias for a given inode | |
957 | * @inode: inode to find an alias for | |
958 | * | |
959 | * If any aliases exist for the given inode, take and return a | |
960 | * reference for one of them. If no aliases exist, return %NULL. | |
961 | */ | |
962 | struct dentry *d_find_any_alias(struct inode *inode) | |
963 | { | |
964 | struct dentry *de; | |
965 | ||
966 | spin_lock(&inode->i_lock); | |
967 | de = __d_find_any_alias(inode); | |
968 | spin_unlock(&inode->i_lock); | |
969 | return de; | |
970 | } | |
971 | EXPORT_SYMBOL(d_find_any_alias); | |
972 | ||
973 | /** | |
974 | * d_find_alias - grab a hashed alias of inode | |
975 | * @inode: inode in question | |
976 | * | |
977 | * If inode has a hashed alias, or is a directory and has any alias, | |
978 | * acquire the reference to alias and return it. Otherwise return NULL. | |
979 | * Notice that if inode is a directory there can be only one alias and | |
980 | * it can be unhashed only if it has no children, or if it is the root | |
981 | * of a filesystem, or if the directory was renamed and d_revalidate | |
982 | * was the first vfs operation to notice. | |
983 | * | |
984 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer | |
985 | * any other hashed alias over that one. | |
986 | */ | |
987 | static struct dentry *__d_find_alias(struct inode *inode) | |
988 | { | |
989 | struct dentry *alias; | |
990 | ||
991 | if (S_ISDIR(inode->i_mode)) | |
992 | return __d_find_any_alias(inode); | |
993 | ||
994 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
995 | spin_lock(&alias->d_lock); | |
996 | if (!d_unhashed(alias)) { | |
997 | __dget_dlock(alias); | |
998 | spin_unlock(&alias->d_lock); | |
999 | return alias; | |
1000 | } | |
1001 | spin_unlock(&alias->d_lock); | |
1002 | } | |
1003 | return NULL; | |
1004 | } | |
1005 | ||
1006 | struct dentry *d_find_alias(struct inode *inode) | |
1007 | { | |
1008 | struct dentry *de = NULL; | |
1009 | ||
1010 | if (!hlist_empty(&inode->i_dentry)) { | |
1011 | spin_lock(&inode->i_lock); | |
1012 | de = __d_find_alias(inode); | |
1013 | spin_unlock(&inode->i_lock); | |
1014 | } | |
1015 | return de; | |
1016 | } | |
1017 | EXPORT_SYMBOL(d_find_alias); | |
1018 | ||
1019 | /* | |
1020 | * Try to kill dentries associated with this inode. | |
1021 | * WARNING: you must own a reference to inode. | |
1022 | */ | |
1023 | void d_prune_aliases(struct inode *inode) | |
1024 | { | |
1025 | struct dentry *dentry; | |
1026 | restart: | |
1027 | spin_lock(&inode->i_lock); | |
1028 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { | |
1029 | spin_lock(&dentry->d_lock); | |
1030 | if (!dentry->d_lockref.count) { | |
1031 | struct dentry *parent = lock_parent(dentry); | |
1032 | if (likely(!dentry->d_lockref.count)) { | |
1033 | __dentry_kill(dentry); | |
1034 | dput(parent); | |
1035 | goto restart; | |
1036 | } | |
1037 | if (parent) | |
1038 | spin_unlock(&parent->d_lock); | |
1039 | } | |
1040 | spin_unlock(&dentry->d_lock); | |
1041 | } | |
1042 | spin_unlock(&inode->i_lock); | |
1043 | } | |
1044 | EXPORT_SYMBOL(d_prune_aliases); | |
1045 | ||
1046 | /* | |
1047 | * Lock a dentry from shrink list. | |
1048 | * Called under rcu_read_lock() and dentry->d_lock; the former | |
1049 | * guarantees that nothing we access will be freed under us. | |
1050 | * Note that dentry is *not* protected from concurrent dentry_kill(), | |
1051 | * d_delete(), etc. | |
1052 | * | |
1053 | * Return false if dentry has been disrupted or grabbed, leaving | |
1054 | * the caller to kick it off-list. Otherwise, return true and have | |
1055 | * that dentry's inode and parent both locked. | |
1056 | */ | |
1057 | static bool shrink_lock_dentry(struct dentry *dentry) | |
1058 | { | |
1059 | struct inode *inode; | |
1060 | struct dentry *parent; | |
1061 | ||
1062 | if (dentry->d_lockref.count) | |
1063 | return false; | |
1064 | ||
1065 | inode = dentry->d_inode; | |
1066 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
1067 | spin_unlock(&dentry->d_lock); | |
1068 | spin_lock(&inode->i_lock); | |
1069 | spin_lock(&dentry->d_lock); | |
1070 | if (unlikely(dentry->d_lockref.count)) | |
1071 | goto out; | |
1072 | /* changed inode means that somebody had grabbed it */ | |
1073 | if (unlikely(inode != dentry->d_inode)) | |
1074 | goto out; | |
1075 | } | |
1076 | ||
1077 | parent = dentry->d_parent; | |
1078 | if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) | |
1079 | return true; | |
1080 | ||
1081 | spin_unlock(&dentry->d_lock); | |
1082 | spin_lock(&parent->d_lock); | |
1083 | if (unlikely(parent != dentry->d_parent)) { | |
1084 | spin_unlock(&parent->d_lock); | |
1085 | spin_lock(&dentry->d_lock); | |
1086 | goto out; | |
1087 | } | |
1088 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1089 | if (likely(!dentry->d_lockref.count)) | |
1090 | return true; | |
1091 | spin_unlock(&parent->d_lock); | |
1092 | out: | |
1093 | if (inode) | |
1094 | spin_unlock(&inode->i_lock); | |
1095 | return false; | |
1096 | } | |
1097 | ||
1098 | void shrink_dentry_list(struct list_head *list) | |
1099 | { | |
1100 | while (!list_empty(list)) { | |
1101 | struct dentry *dentry, *parent; | |
1102 | ||
1103 | dentry = list_entry(list->prev, struct dentry, d_lru); | |
1104 | spin_lock(&dentry->d_lock); | |
1105 | rcu_read_lock(); | |
1106 | if (!shrink_lock_dentry(dentry)) { | |
1107 | bool can_free = false; | |
1108 | rcu_read_unlock(); | |
1109 | d_shrink_del(dentry); | |
1110 | if (dentry->d_lockref.count < 0) | |
1111 | can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
1112 | spin_unlock(&dentry->d_lock); | |
1113 | if (can_free) | |
1114 | dentry_free(dentry); | |
1115 | continue; | |
1116 | } | |
1117 | rcu_read_unlock(); | |
1118 | d_shrink_del(dentry); | |
1119 | parent = dentry->d_parent; | |
1120 | if (parent != dentry) | |
1121 | __dput_to_list(parent, list); | |
1122 | __dentry_kill(dentry); | |
1123 | } | |
1124 | } | |
1125 | ||
1126 | static enum lru_status dentry_lru_isolate(struct list_head *item, | |
1127 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
1128 | { | |
1129 | struct list_head *freeable = arg; | |
1130 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1131 | ||
1132 | ||
1133 | /* | |
1134 | * we are inverting the lru lock/dentry->d_lock here, | |
1135 | * so use a trylock. If we fail to get the lock, just skip | |
1136 | * it | |
1137 | */ | |
1138 | if (!spin_trylock(&dentry->d_lock)) | |
1139 | return LRU_SKIP; | |
1140 | ||
1141 | /* | |
1142 | * Referenced dentries are still in use. If they have active | |
1143 | * counts, just remove them from the LRU. Otherwise give them | |
1144 | * another pass through the LRU. | |
1145 | */ | |
1146 | if (dentry->d_lockref.count) { | |
1147 | d_lru_isolate(lru, dentry); | |
1148 | spin_unlock(&dentry->d_lock); | |
1149 | return LRU_REMOVED; | |
1150 | } | |
1151 | ||
1152 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1153 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1154 | spin_unlock(&dentry->d_lock); | |
1155 | ||
1156 | /* | |
1157 | * The list move itself will be made by the common LRU code. At | |
1158 | * this point, we've dropped the dentry->d_lock but keep the | |
1159 | * lru lock. This is safe to do, since every list movement is | |
1160 | * protected by the lru lock even if both locks are held. | |
1161 | * | |
1162 | * This is guaranteed by the fact that all LRU management | |
1163 | * functions are intermediated by the LRU API calls like | |
1164 | * list_lru_add and list_lru_del. List movement in this file | |
1165 | * only ever occur through this functions or through callbacks | |
1166 | * like this one, that are called from the LRU API. | |
1167 | * | |
1168 | * The only exceptions to this are functions like | |
1169 | * shrink_dentry_list, and code that first checks for the | |
1170 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1171 | * operating only with stack provided lists after they are | |
1172 | * properly isolated from the main list. It is thus, always a | |
1173 | * local access. | |
1174 | */ | |
1175 | return LRU_ROTATE; | |
1176 | } | |
1177 | ||
1178 | d_lru_shrink_move(lru, dentry, freeable); | |
1179 | spin_unlock(&dentry->d_lock); | |
1180 | ||
1181 | return LRU_REMOVED; | |
1182 | } | |
1183 | ||
1184 | /** | |
1185 | * prune_dcache_sb - shrink the dcache | |
1186 | * @sb: superblock | |
1187 | * @sc: shrink control, passed to list_lru_shrink_walk() | |
1188 | * | |
1189 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This | |
1190 | * is done when we need more memory and called from the superblock shrinker | |
1191 | * function. | |
1192 | * | |
1193 | * This function may fail to free any resources if all the dentries are in | |
1194 | * use. | |
1195 | */ | |
1196 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) | |
1197 | { | |
1198 | LIST_HEAD(dispose); | |
1199 | long freed; | |
1200 | ||
1201 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, | |
1202 | dentry_lru_isolate, &dispose); | |
1203 | shrink_dentry_list(&dispose); | |
1204 | return freed; | |
1205 | } | |
1206 | ||
1207 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, | |
1208 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
1209 | { | |
1210 | struct list_head *freeable = arg; | |
1211 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1212 | ||
1213 | /* | |
1214 | * we are inverting the lru lock/dentry->d_lock here, | |
1215 | * so use a trylock. If we fail to get the lock, just skip | |
1216 | * it | |
1217 | */ | |
1218 | if (!spin_trylock(&dentry->d_lock)) | |
1219 | return LRU_SKIP; | |
1220 | ||
1221 | d_lru_shrink_move(lru, dentry, freeable); | |
1222 | spin_unlock(&dentry->d_lock); | |
1223 | ||
1224 | return LRU_REMOVED; | |
1225 | } | |
1226 | ||
1227 | ||
1228 | /** | |
1229 | * shrink_dcache_sb - shrink dcache for a superblock | |
1230 | * @sb: superblock | |
1231 | * | |
1232 | * Shrink the dcache for the specified super block. This is used to free | |
1233 | * the dcache before unmounting a file system. | |
1234 | */ | |
1235 | void shrink_dcache_sb(struct super_block *sb) | |
1236 | { | |
1237 | do { | |
1238 | LIST_HEAD(dispose); | |
1239 | ||
1240 | list_lru_walk(&sb->s_dentry_lru, | |
1241 | dentry_lru_isolate_shrink, &dispose, 1024); | |
1242 | shrink_dentry_list(&dispose); | |
1243 | } while (list_lru_count(&sb->s_dentry_lru) > 0); | |
1244 | } | |
1245 | EXPORT_SYMBOL(shrink_dcache_sb); | |
1246 | ||
1247 | /** | |
1248 | * enum d_walk_ret - action to talke during tree walk | |
1249 | * @D_WALK_CONTINUE: contrinue walk | |
1250 | * @D_WALK_QUIT: quit walk | |
1251 | * @D_WALK_NORETRY: quit when retry is needed | |
1252 | * @D_WALK_SKIP: skip this dentry and its children | |
1253 | */ | |
1254 | enum d_walk_ret { | |
1255 | D_WALK_CONTINUE, | |
1256 | D_WALK_QUIT, | |
1257 | D_WALK_NORETRY, | |
1258 | D_WALK_SKIP, | |
1259 | }; | |
1260 | ||
1261 | /** | |
1262 | * d_walk - walk the dentry tree | |
1263 | * @parent: start of walk | |
1264 | * @data: data passed to @enter() and @finish() | |
1265 | * @enter: callback when first entering the dentry | |
1266 | * | |
1267 | * The @enter() callbacks are called with d_lock held. | |
1268 | */ | |
1269 | void d_walk(struct dentry *parent, void *data, | |
1270 | enum d_walk_ret (*enter)(void *, struct dentry *)) | |
1271 | { | |
1272 | struct dentry *this_parent; | |
1273 | struct list_head *next; | |
1274 | unsigned seq = 0; | |
1275 | enum d_walk_ret ret; | |
1276 | bool retry = true; | |
1277 | ||
1278 | again: | |
1279 | read_seqbegin_or_lock(&rename_lock, &seq); | |
1280 | this_parent = parent; | |
1281 | spin_lock(&this_parent->d_lock); | |
1282 | ||
1283 | ret = enter(data, this_parent); | |
1284 | switch (ret) { | |
1285 | case D_WALK_CONTINUE: | |
1286 | break; | |
1287 | case D_WALK_QUIT: | |
1288 | case D_WALK_SKIP: | |
1289 | goto out_unlock; | |
1290 | case D_WALK_NORETRY: | |
1291 | retry = false; | |
1292 | break; | |
1293 | } | |
1294 | repeat: | |
1295 | next = this_parent->d_subdirs.next; | |
1296 | resume: | |
1297 | while (next != &this_parent->d_subdirs) { | |
1298 | struct list_head *tmp = next; | |
1299 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); | |
1300 | next = tmp->next; | |
1301 | ||
1302 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) | |
1303 | continue; | |
1304 | ||
1305 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1306 | ||
1307 | ret = enter(data, dentry); | |
1308 | switch (ret) { | |
1309 | case D_WALK_CONTINUE: | |
1310 | break; | |
1311 | case D_WALK_QUIT: | |
1312 | spin_unlock(&dentry->d_lock); | |
1313 | goto out_unlock; | |
1314 | case D_WALK_NORETRY: | |
1315 | retry = false; | |
1316 | break; | |
1317 | case D_WALK_SKIP: | |
1318 | spin_unlock(&dentry->d_lock); | |
1319 | continue; | |
1320 | } | |
1321 | ||
1322 | if (!list_empty(&dentry->d_subdirs)) { | |
1323 | spin_unlock(&this_parent->d_lock); | |
1324 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1325 | this_parent = dentry; | |
1326 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); | |
1327 | goto repeat; | |
1328 | } | |
1329 | spin_unlock(&dentry->d_lock); | |
1330 | } | |
1331 | /* | |
1332 | * All done at this level ... ascend and resume the search. | |
1333 | */ | |
1334 | rcu_read_lock(); | |
1335 | ascend: | |
1336 | if (this_parent != parent) { | |
1337 | struct dentry *child = this_parent; | |
1338 | this_parent = child->d_parent; | |
1339 | ||
1340 | spin_unlock(&child->d_lock); | |
1341 | spin_lock(&this_parent->d_lock); | |
1342 | ||
1343 | /* might go back up the wrong parent if we have had a rename. */ | |
1344 | if (need_seqretry(&rename_lock, seq)) | |
1345 | goto rename_retry; | |
1346 | /* go into the first sibling still alive */ | |
1347 | do { | |
1348 | next = child->d_child.next; | |
1349 | if (next == &this_parent->d_subdirs) | |
1350 | goto ascend; | |
1351 | child = list_entry(next, struct dentry, d_child); | |
1352 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); | |
1353 | rcu_read_unlock(); | |
1354 | goto resume; | |
1355 | } | |
1356 | if (need_seqretry(&rename_lock, seq)) | |
1357 | goto rename_retry; | |
1358 | rcu_read_unlock(); | |
1359 | ||
1360 | out_unlock: | |
1361 | spin_unlock(&this_parent->d_lock); | |
1362 | done_seqretry(&rename_lock, seq); | |
1363 | return; | |
1364 | ||
1365 | rename_retry: | |
1366 | spin_unlock(&this_parent->d_lock); | |
1367 | rcu_read_unlock(); | |
1368 | BUG_ON(seq & 1); | |
1369 | if (!retry) | |
1370 | return; | |
1371 | seq = 1; | |
1372 | goto again; | |
1373 | } | |
1374 | EXPORT_SYMBOL_GPL(d_walk); | |
1375 | ||
1376 | struct check_mount { | |
1377 | struct vfsmount *mnt; | |
1378 | unsigned int mounted; | |
1379 | }; | |
1380 | ||
1381 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1382 | { | |
1383 | struct check_mount *info = data; | |
1384 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1385 | ||
1386 | if (likely(!d_mountpoint(dentry))) | |
1387 | return D_WALK_CONTINUE; | |
1388 | if (__path_is_mountpoint(&path)) { | |
1389 | info->mounted = 1; | |
1390 | return D_WALK_QUIT; | |
1391 | } | |
1392 | return D_WALK_CONTINUE; | |
1393 | } | |
1394 | ||
1395 | /** | |
1396 | * path_has_submounts - check for mounts over a dentry in the | |
1397 | * current namespace. | |
1398 | * @parent: path to check. | |
1399 | * | |
1400 | * Return true if the parent or its subdirectories contain | |
1401 | * a mount point in the current namespace. | |
1402 | */ | |
1403 | int path_has_submounts(const struct path *parent) | |
1404 | { | |
1405 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1406 | ||
1407 | read_seqlock_excl(&mount_lock); | |
1408 | d_walk(parent->dentry, &data, path_check_mount); | |
1409 | read_sequnlock_excl(&mount_lock); | |
1410 | ||
1411 | return data.mounted; | |
1412 | } | |
1413 | EXPORT_SYMBOL(path_has_submounts); | |
1414 | ||
1415 | /* | |
1416 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1417 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1418 | * subtree can become unreachable). | |
1419 | * | |
1420 | * Only one of d_invalidate() and d_set_mounted() must succeed. For | |
1421 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1422 | */ | |
1423 | int d_set_mounted(struct dentry *dentry) | |
1424 | { | |
1425 | struct dentry *p; | |
1426 | int ret = -ENOENT; | |
1427 | write_seqlock(&rename_lock); | |
1428 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1429 | /* Need exclusion wrt. d_invalidate() */ | |
1430 | spin_lock(&p->d_lock); | |
1431 | if (unlikely(d_unhashed(p))) { | |
1432 | spin_unlock(&p->d_lock); | |
1433 | goto out; | |
1434 | } | |
1435 | spin_unlock(&p->d_lock); | |
1436 | } | |
1437 | spin_lock(&dentry->d_lock); | |
1438 | if (!d_unlinked(dentry)) { | |
1439 | ret = -EBUSY; | |
1440 | if (!d_mountpoint(dentry)) { | |
1441 | dentry->d_flags |= DCACHE_MOUNTED; | |
1442 | ret = 0; | |
1443 | } | |
1444 | } | |
1445 | spin_unlock(&dentry->d_lock); | |
1446 | out: | |
1447 | write_sequnlock(&rename_lock); | |
1448 | return ret; | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * Search the dentry child list of the specified parent, | |
1453 | * and move any unused dentries to the end of the unused | |
1454 | * list for prune_dcache(). We descend to the next level | |
1455 | * whenever the d_subdirs list is non-empty and continue | |
1456 | * searching. | |
1457 | * | |
1458 | * It returns zero iff there are no unused children, | |
1459 | * otherwise it returns the number of children moved to | |
1460 | * the end of the unused list. This may not be the total | |
1461 | * number of unused children, because select_parent can | |
1462 | * drop the lock and return early due to latency | |
1463 | * constraints. | |
1464 | */ | |
1465 | ||
1466 | struct select_data { | |
1467 | struct dentry *start; | |
1468 | union { | |
1469 | long found; | |
1470 | struct dentry *victim; | |
1471 | }; | |
1472 | struct list_head dispose; | |
1473 | }; | |
1474 | ||
1475 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) | |
1476 | { | |
1477 | struct select_data *data = _data; | |
1478 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1479 | ||
1480 | if (data->start == dentry) | |
1481 | goto out; | |
1482 | ||
1483 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
1484 | data->found++; | |
1485 | } else { | |
1486 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1487 | d_lru_del(dentry); | |
1488 | if (!dentry->d_lockref.count) { | |
1489 | d_shrink_add(dentry, &data->dispose); | |
1490 | data->found++; | |
1491 | } | |
1492 | } | |
1493 | /* | |
1494 | * We can return to the caller if we have found some (this | |
1495 | * ensures forward progress). We'll be coming back to find | |
1496 | * the rest. | |
1497 | */ | |
1498 | if (!list_empty(&data->dispose)) | |
1499 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1500 | out: | |
1501 | return ret; | |
1502 | } | |
1503 | ||
1504 | static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry) | |
1505 | { | |
1506 | struct select_data *data = _data; | |
1507 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1508 | ||
1509 | if (data->start == dentry) | |
1510 | goto out; | |
1511 | ||
1512 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
1513 | if (!dentry->d_lockref.count) { | |
1514 | rcu_read_lock(); | |
1515 | data->victim = dentry; | |
1516 | return D_WALK_QUIT; | |
1517 | } | |
1518 | } else { | |
1519 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1520 | d_lru_del(dentry); | |
1521 | if (!dentry->d_lockref.count) | |
1522 | d_shrink_add(dentry, &data->dispose); | |
1523 | } | |
1524 | /* | |
1525 | * We can return to the caller if we have found some (this | |
1526 | * ensures forward progress). We'll be coming back to find | |
1527 | * the rest. | |
1528 | */ | |
1529 | if (!list_empty(&data->dispose)) | |
1530 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1531 | out: | |
1532 | return ret; | |
1533 | } | |
1534 | ||
1535 | /** | |
1536 | * shrink_dcache_parent - prune dcache | |
1537 | * @parent: parent of entries to prune | |
1538 | * | |
1539 | * Prune the dcache to remove unused children of the parent dentry. | |
1540 | */ | |
1541 | void shrink_dcache_parent(struct dentry *parent) | |
1542 | { | |
1543 | for (;;) { | |
1544 | struct select_data data = {.start = parent}; | |
1545 | ||
1546 | INIT_LIST_HEAD(&data.dispose); | |
1547 | d_walk(parent, &data, select_collect); | |
1548 | ||
1549 | if (!list_empty(&data.dispose)) { | |
1550 | shrink_dentry_list(&data.dispose); | |
1551 | continue; | |
1552 | } | |
1553 | ||
1554 | cond_resched(); | |
1555 | if (!data.found) | |
1556 | break; | |
1557 | data.victim = NULL; | |
1558 | d_walk(parent, &data, select_collect2); | |
1559 | if (data.victim) { | |
1560 | struct dentry *parent; | |
1561 | spin_lock(&data.victim->d_lock); | |
1562 | if (!shrink_lock_dentry(data.victim)) { | |
1563 | spin_unlock(&data.victim->d_lock); | |
1564 | rcu_read_unlock(); | |
1565 | } else { | |
1566 | rcu_read_unlock(); | |
1567 | parent = data.victim->d_parent; | |
1568 | if (parent != data.victim) | |
1569 | __dput_to_list(parent, &data.dispose); | |
1570 | __dentry_kill(data.victim); | |
1571 | } | |
1572 | } | |
1573 | if (!list_empty(&data.dispose)) | |
1574 | shrink_dentry_list(&data.dispose); | |
1575 | } | |
1576 | } | |
1577 | EXPORT_SYMBOL(shrink_dcache_parent); | |
1578 | ||
1579 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) | |
1580 | { | |
1581 | /* it has busy descendents; complain about those instead */ | |
1582 | if (!list_empty(&dentry->d_subdirs)) | |
1583 | return D_WALK_CONTINUE; | |
1584 | ||
1585 | /* root with refcount 1 is fine */ | |
1586 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1587 | return D_WALK_CONTINUE; | |
1588 | ||
1589 | printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} " | |
1590 | " still in use (%d) [unmount of %s %s]\n", | |
1591 | dentry, | |
1592 | dentry->d_inode ? | |
1593 | dentry->d_inode->i_ino : 0UL, | |
1594 | dentry, | |
1595 | dentry->d_lockref.count, | |
1596 | dentry->d_sb->s_type->name, | |
1597 | dentry->d_sb->s_id); | |
1598 | WARN_ON(1); | |
1599 | return D_WALK_CONTINUE; | |
1600 | } | |
1601 | ||
1602 | static void do_one_tree(struct dentry *dentry) | |
1603 | { | |
1604 | shrink_dcache_parent(dentry); | |
1605 | d_walk(dentry, dentry, umount_check); | |
1606 | d_drop(dentry); | |
1607 | dput(dentry); | |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * destroy the dentries attached to a superblock on unmounting | |
1612 | */ | |
1613 | void shrink_dcache_for_umount(struct super_block *sb) | |
1614 | { | |
1615 | struct dentry *dentry; | |
1616 | ||
1617 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); | |
1618 | ||
1619 | dentry = sb->s_root; | |
1620 | sb->s_root = NULL; | |
1621 | do_one_tree(dentry); | |
1622 | ||
1623 | while (!hlist_bl_empty(&sb->s_roots)) { | |
1624 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); | |
1625 | do_one_tree(dentry); | |
1626 | } | |
1627 | } | |
1628 | ||
1629 | static enum d_walk_ret find_submount(void *_data, struct dentry *dentry) | |
1630 | { | |
1631 | struct dentry **victim = _data; | |
1632 | if (d_mountpoint(dentry)) { | |
1633 | __dget_dlock(dentry); | |
1634 | *victim = dentry; | |
1635 | return D_WALK_QUIT; | |
1636 | } | |
1637 | return D_WALK_CONTINUE; | |
1638 | } | |
1639 | ||
1640 | /** | |
1641 | * d_invalidate - detach submounts, prune dcache, and drop | |
1642 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
1643 | */ | |
1644 | void d_invalidate(struct dentry *dentry) | |
1645 | { | |
1646 | bool had_submounts = false; | |
1647 | spin_lock(&dentry->d_lock); | |
1648 | if (d_unhashed(dentry)) { | |
1649 | spin_unlock(&dentry->d_lock); | |
1650 | return; | |
1651 | } | |
1652 | __d_drop(dentry); | |
1653 | spin_unlock(&dentry->d_lock); | |
1654 | ||
1655 | /* Negative dentries can be dropped without further checks */ | |
1656 | if (!dentry->d_inode) | |
1657 | return; | |
1658 | ||
1659 | shrink_dcache_parent(dentry); | |
1660 | for (;;) { | |
1661 | struct dentry *victim = NULL; | |
1662 | d_walk(dentry, &victim, find_submount); | |
1663 | if (!victim) { | |
1664 | if (had_submounts) | |
1665 | shrink_dcache_parent(dentry); | |
1666 | return; | |
1667 | } | |
1668 | had_submounts = true; | |
1669 | detach_mounts(victim); | |
1670 | dput(victim); | |
1671 | } | |
1672 | } | |
1673 | EXPORT_SYMBOL(d_invalidate); | |
1674 | ||
1675 | /** | |
1676 | * __d_alloc - allocate a dcache entry | |
1677 | * @sb: filesystem it will belong to | |
1678 | * @name: qstr of the name | |
1679 | * | |
1680 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1681 | * available. On a success the dentry is returned. The name passed in is | |
1682 | * copied and the copy passed in may be reused after this call. | |
1683 | */ | |
1684 | ||
1685 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) | |
1686 | { | |
1687 | struct dentry *dentry; | |
1688 | char *dname; | |
1689 | int err; | |
1690 | ||
1691 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); | |
1692 | if (!dentry) | |
1693 | return NULL; | |
1694 | ||
1695 | /* | |
1696 | * We guarantee that the inline name is always NUL-terminated. | |
1697 | * This way the memcpy() done by the name switching in rename | |
1698 | * will still always have a NUL at the end, even if we might | |
1699 | * be overwriting an internal NUL character | |
1700 | */ | |
1701 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1702 | if (unlikely(!name)) { | |
1703 | name = &slash_name; | |
1704 | dname = dentry->d_iname; | |
1705 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
1706 | size_t size = offsetof(struct external_name, name[1]); | |
1707 | struct external_name *p = kmalloc(size + name->len, | |
1708 | GFP_KERNEL_ACCOUNT | | |
1709 | __GFP_RECLAIMABLE); | |
1710 | if (!p) { | |
1711 | kmem_cache_free(dentry_cache, dentry); | |
1712 | return NULL; | |
1713 | } | |
1714 | atomic_set(&p->u.count, 1); | |
1715 | dname = p->name; | |
1716 | } else { | |
1717 | dname = dentry->d_iname; | |
1718 | } | |
1719 | ||
1720 | dentry->d_name.len = name->len; | |
1721 | dentry->d_name.hash = name->hash; | |
1722 | memcpy(dname, name->name, name->len); | |
1723 | dname[name->len] = 0; | |
1724 | ||
1725 | /* Make sure we always see the terminating NUL character */ | |
1726 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ | |
1727 | ||
1728 | dentry->d_lockref.count = 1; | |
1729 | dentry->d_flags = 0; | |
1730 | spin_lock_init(&dentry->d_lock); | |
1731 | seqcount_init(&dentry->d_seq); | |
1732 | dentry->d_inode = NULL; | |
1733 | dentry->d_parent = dentry; | |
1734 | dentry->d_sb = sb; | |
1735 | dentry->d_op = NULL; | |
1736 | dentry->d_fsdata = NULL; | |
1737 | INIT_HLIST_BL_NODE(&dentry->d_hash); | |
1738 | INIT_LIST_HEAD(&dentry->d_lru); | |
1739 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1740 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
1741 | INIT_LIST_HEAD(&dentry->d_child); | |
1742 | d_set_d_op(dentry, dentry->d_sb->s_d_op); | |
1743 | ||
1744 | if (dentry->d_op && dentry->d_op->d_init) { | |
1745 | err = dentry->d_op->d_init(dentry); | |
1746 | if (err) { | |
1747 | if (dname_external(dentry)) | |
1748 | kfree(external_name(dentry)); | |
1749 | kmem_cache_free(dentry_cache, dentry); | |
1750 | return NULL; | |
1751 | } | |
1752 | } | |
1753 | ||
1754 | this_cpu_inc(nr_dentry); | |
1755 | ||
1756 | return dentry; | |
1757 | } | |
1758 | ||
1759 | /** | |
1760 | * d_alloc - allocate a dcache entry | |
1761 | * @parent: parent of entry to allocate | |
1762 | * @name: qstr of the name | |
1763 | * | |
1764 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1765 | * available. On a success the dentry is returned. The name passed in is | |
1766 | * copied and the copy passed in may be reused after this call. | |
1767 | */ | |
1768 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1769 | { | |
1770 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1771 | if (!dentry) | |
1772 | return NULL; | |
1773 | spin_lock(&parent->d_lock); | |
1774 | /* | |
1775 | * don't need child lock because it is not subject | |
1776 | * to concurrency here | |
1777 | */ | |
1778 | __dget_dlock(parent); | |
1779 | dentry->d_parent = parent; | |
1780 | list_add(&dentry->d_child, &parent->d_subdirs); | |
1781 | spin_unlock(&parent->d_lock); | |
1782 | ||
1783 | return dentry; | |
1784 | } | |
1785 | EXPORT_SYMBOL(d_alloc); | |
1786 | ||
1787 | struct dentry *d_alloc_anon(struct super_block *sb) | |
1788 | { | |
1789 | return __d_alloc(sb, NULL); | |
1790 | } | |
1791 | EXPORT_SYMBOL(d_alloc_anon); | |
1792 | ||
1793 | struct dentry *d_alloc_cursor(struct dentry * parent) | |
1794 | { | |
1795 | struct dentry *dentry = d_alloc_anon(parent->d_sb); | |
1796 | if (dentry) { | |
1797 | dentry->d_flags |= DCACHE_DENTRY_CURSOR; | |
1798 | dentry->d_parent = dget(parent); | |
1799 | } | |
1800 | return dentry; | |
1801 | } | |
1802 | ||
1803 | /** | |
1804 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1805 | * @sb: the superblock | |
1806 | * @name: qstr of the name | |
1807 | * | |
1808 | * For a filesystem that just pins its dentries in memory and never | |
1809 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
1810 | * This is used for pipes, sockets et.al. - the stuff that should | |
1811 | * never be anyone's children or parents. Unlike all other | |
1812 | * dentries, these will not have RCU delay between dropping the | |
1813 | * last reference and freeing them. | |
1814 | * | |
1815 | * The only user is alloc_file_pseudo() and that's what should | |
1816 | * be considered a public interface. Don't use directly. | |
1817 | */ | |
1818 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) | |
1819 | { | |
1820 | struct dentry *dentry = __d_alloc(sb, name); | |
1821 | if (likely(dentry)) | |
1822 | dentry->d_flags |= DCACHE_NORCU; | |
1823 | return dentry; | |
1824 | } | |
1825 | ||
1826 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
1827 | { | |
1828 | struct qstr q; | |
1829 | ||
1830 | q.name = name; | |
1831 | q.hash_len = hashlen_string(parent, name); | |
1832 | return d_alloc(parent, &q); | |
1833 | } | |
1834 | EXPORT_SYMBOL(d_alloc_name); | |
1835 | ||
1836 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) | |
1837 | { | |
1838 | WARN_ON_ONCE(dentry->d_op); | |
1839 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
1840 | DCACHE_OP_COMPARE | | |
1841 | DCACHE_OP_REVALIDATE | | |
1842 | DCACHE_OP_WEAK_REVALIDATE | | |
1843 | DCACHE_OP_DELETE | | |
1844 | DCACHE_OP_REAL)); | |
1845 | dentry->d_op = op; | |
1846 | if (!op) | |
1847 | return; | |
1848 | if (op->d_hash) | |
1849 | dentry->d_flags |= DCACHE_OP_HASH; | |
1850 | if (op->d_compare) | |
1851 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1852 | if (op->d_revalidate) | |
1853 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1854 | if (op->d_weak_revalidate) | |
1855 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
1856 | if (op->d_delete) | |
1857 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1858 | if (op->d_prune) | |
1859 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
1860 | if (op->d_real) | |
1861 | dentry->d_flags |= DCACHE_OP_REAL; | |
1862 | ||
1863 | } | |
1864 | EXPORT_SYMBOL(d_set_d_op); | |
1865 | ||
1866 | ||
1867 | /* | |
1868 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1869 | * @dentry - The dentry to mark | |
1870 | * | |
1871 | * Mark a dentry as falling through to the lower layer (as set with | |
1872 | * d_pin_lower()). This flag may be recorded on the medium. | |
1873 | */ | |
1874 | void d_set_fallthru(struct dentry *dentry) | |
1875 | { | |
1876 | spin_lock(&dentry->d_lock); | |
1877 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1878 | spin_unlock(&dentry->d_lock); | |
1879 | } | |
1880 | EXPORT_SYMBOL(d_set_fallthru); | |
1881 | ||
1882 | static unsigned d_flags_for_inode(struct inode *inode) | |
1883 | { | |
1884 | unsigned add_flags = DCACHE_REGULAR_TYPE; | |
1885 | ||
1886 | if (!inode) | |
1887 | return DCACHE_MISS_TYPE; | |
1888 | ||
1889 | if (S_ISDIR(inode->i_mode)) { | |
1890 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1891 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1892 | if (unlikely(!inode->i_op->lookup)) | |
1893 | add_flags = DCACHE_AUTODIR_TYPE; | |
1894 | else | |
1895 | inode->i_opflags |= IOP_LOOKUP; | |
1896 | } | |
1897 | goto type_determined; | |
1898 | } | |
1899 | ||
1900 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
1901 | if (unlikely(inode->i_op->get_link)) { | |
1902 | add_flags = DCACHE_SYMLINK_TYPE; | |
1903 | goto type_determined; | |
1904 | } | |
1905 | inode->i_opflags |= IOP_NOFOLLOW; | |
1906 | } | |
1907 | ||
1908 | if (unlikely(!S_ISREG(inode->i_mode))) | |
1909 | add_flags = DCACHE_SPECIAL_TYPE; | |
1910 | ||
1911 | type_determined: | |
1912 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1913 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1914 | return add_flags; | |
1915 | } | |
1916 | ||
1917 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
1918 | { | |
1919 | unsigned add_flags = d_flags_for_inode(inode); | |
1920 | WARN_ON(d_in_lookup(dentry)); | |
1921 | ||
1922 | spin_lock(&dentry->d_lock); | |
1923 | /* | |
1924 | * Decrement negative dentry count if it was in the LRU list. | |
1925 | */ | |
1926 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1927 | this_cpu_dec(nr_dentry_negative); | |
1928 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
1929 | raw_write_seqcount_begin(&dentry->d_seq); | |
1930 | __d_set_inode_and_type(dentry, inode, add_flags); | |
1931 | raw_write_seqcount_end(&dentry->d_seq); | |
1932 | fsnotify_update_flags(dentry); | |
1933 | spin_unlock(&dentry->d_lock); | |
1934 | } | |
1935 | ||
1936 | /** | |
1937 | * d_instantiate - fill in inode information for a dentry | |
1938 | * @entry: dentry to complete | |
1939 | * @inode: inode to attach to this dentry | |
1940 | * | |
1941 | * Fill in inode information in the entry. | |
1942 | * | |
1943 | * This turns negative dentries into productive full members | |
1944 | * of society. | |
1945 | * | |
1946 | * NOTE! This assumes that the inode count has been incremented | |
1947 | * (or otherwise set) by the caller to indicate that it is now | |
1948 | * in use by the dcache. | |
1949 | */ | |
1950 | ||
1951 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1952 | { | |
1953 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); | |
1954 | if (inode) { | |
1955 | security_d_instantiate(entry, inode); | |
1956 | spin_lock(&inode->i_lock); | |
1957 | __d_instantiate(entry, inode); | |
1958 | spin_unlock(&inode->i_lock); | |
1959 | } | |
1960 | } | |
1961 | EXPORT_SYMBOL(d_instantiate); | |
1962 | ||
1963 | /* | |
1964 | * This should be equivalent to d_instantiate() + unlock_new_inode(), | |
1965 | * with lockdep-related part of unlock_new_inode() done before | |
1966 | * anything else. Use that instead of open-coding d_instantiate()/ | |
1967 | * unlock_new_inode() combinations. | |
1968 | */ | |
1969 | void d_instantiate_new(struct dentry *entry, struct inode *inode) | |
1970 | { | |
1971 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); | |
1972 | BUG_ON(!inode); | |
1973 | lockdep_annotate_inode_mutex_key(inode); | |
1974 | security_d_instantiate(entry, inode); | |
1975 | spin_lock(&inode->i_lock); | |
1976 | __d_instantiate(entry, inode); | |
1977 | WARN_ON(!(inode->i_state & I_NEW)); | |
1978 | inode->i_state &= ~I_NEW & ~I_CREATING; | |
1979 | smp_mb(); | |
1980 | wake_up_bit(&inode->i_state, __I_NEW); | |
1981 | spin_unlock(&inode->i_lock); | |
1982 | } | |
1983 | EXPORT_SYMBOL(d_instantiate_new); | |
1984 | ||
1985 | struct dentry *d_make_root(struct inode *root_inode) | |
1986 | { | |
1987 | struct dentry *res = NULL; | |
1988 | ||
1989 | if (root_inode) { | |
1990 | res = d_alloc_anon(root_inode->i_sb); | |
1991 | if (res) | |
1992 | d_instantiate(res, root_inode); | |
1993 | else | |
1994 | iput(root_inode); | |
1995 | } | |
1996 | return res; | |
1997 | } | |
1998 | EXPORT_SYMBOL(d_make_root); | |
1999 | ||
2000 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, | |
2001 | struct inode *inode, | |
2002 | bool disconnected) | |
2003 | { | |
2004 | struct dentry *res; | |
2005 | unsigned add_flags; | |
2006 | ||
2007 | security_d_instantiate(dentry, inode); | |
2008 | spin_lock(&inode->i_lock); | |
2009 | res = __d_find_any_alias(inode); | |
2010 | if (res) { | |
2011 | spin_unlock(&inode->i_lock); | |
2012 | dput(dentry); | |
2013 | goto out_iput; | |
2014 | } | |
2015 | ||
2016 | /* attach a disconnected dentry */ | |
2017 | add_flags = d_flags_for_inode(inode); | |
2018 | ||
2019 | if (disconnected) | |
2020 | add_flags |= DCACHE_DISCONNECTED; | |
2021 | ||
2022 | spin_lock(&dentry->d_lock); | |
2023 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2024 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2025 | if (!disconnected) { | |
2026 | hlist_bl_lock(&dentry->d_sb->s_roots); | |
2027 | hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots); | |
2028 | hlist_bl_unlock(&dentry->d_sb->s_roots); | |
2029 | } | |
2030 | spin_unlock(&dentry->d_lock); | |
2031 | spin_unlock(&inode->i_lock); | |
2032 | ||
2033 | return dentry; | |
2034 | ||
2035 | out_iput: | |
2036 | iput(inode); | |
2037 | return res; | |
2038 | } | |
2039 | ||
2040 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) | |
2041 | { | |
2042 | return __d_instantiate_anon(dentry, inode, true); | |
2043 | } | |
2044 | EXPORT_SYMBOL(d_instantiate_anon); | |
2045 | ||
2046 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) | |
2047 | { | |
2048 | struct dentry *tmp; | |
2049 | struct dentry *res; | |
2050 | ||
2051 | if (!inode) | |
2052 | return ERR_PTR(-ESTALE); | |
2053 | if (IS_ERR(inode)) | |
2054 | return ERR_CAST(inode); | |
2055 | ||
2056 | res = d_find_any_alias(inode); | |
2057 | if (res) | |
2058 | goto out_iput; | |
2059 | ||
2060 | tmp = d_alloc_anon(inode->i_sb); | |
2061 | if (!tmp) { | |
2062 | res = ERR_PTR(-ENOMEM); | |
2063 | goto out_iput; | |
2064 | } | |
2065 | ||
2066 | return __d_instantiate_anon(tmp, inode, disconnected); | |
2067 | ||
2068 | out_iput: | |
2069 | iput(inode); | |
2070 | return res; | |
2071 | } | |
2072 | ||
2073 | /** | |
2074 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2075 | * @inode: inode to allocate the dentry for | |
2076 | * | |
2077 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2078 | * similar open by handle operations. The returned dentry may be anonymous, | |
2079 | * or may have a full name (if the inode was already in the cache). | |
2080 | * | |
2081 | * When called on a directory inode, we must ensure that the inode only ever | |
2082 | * has one dentry. If a dentry is found, that is returned instead of | |
2083 | * allocating a new one. | |
2084 | * | |
2085 | * On successful return, the reference to the inode has been transferred | |
2086 | * to the dentry. In case of an error the reference on the inode is released. | |
2087 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2088 | * be passed in and the error will be propagated to the return value, | |
2089 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2090 | */ | |
2091 | struct dentry *d_obtain_alias(struct inode *inode) | |
2092 | { | |
2093 | return __d_obtain_alias(inode, true); | |
2094 | } | |
2095 | EXPORT_SYMBOL(d_obtain_alias); | |
2096 | ||
2097 | /** | |
2098 | * d_obtain_root - find or allocate a dentry for a given inode | |
2099 | * @inode: inode to allocate the dentry for | |
2100 | * | |
2101 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2102 | * | |
2103 | * We must ensure that directory inodes only ever have one dentry. If a | |
2104 | * dentry is found, that is returned instead of allocating a new one. | |
2105 | * | |
2106 | * On successful return, the reference to the inode has been transferred | |
2107 | * to the dentry. In case of an error the reference on the inode is | |
2108 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2109 | * error will be propagate to the return value, with a %NULL @inode | |
2110 | * replaced by ERR_PTR(-ESTALE). | |
2111 | */ | |
2112 | struct dentry *d_obtain_root(struct inode *inode) | |
2113 | { | |
2114 | return __d_obtain_alias(inode, false); | |
2115 | } | |
2116 | EXPORT_SYMBOL(d_obtain_root); | |
2117 | ||
2118 | /** | |
2119 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2120 | * @inode: the inode case-insensitive lookup has found | |
2121 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2122 | * @name: the case-exact name to be associated with the returned dentry | |
2123 | * | |
2124 | * This is to avoid filling the dcache with case-insensitive names to the | |
2125 | * same inode, only the actual correct case is stored in the dcache for | |
2126 | * case-insensitive filesystems. | |
2127 | * | |
2128 | * For a case-insensitive lookup match and if the the case-exact dentry | |
2129 | * already exists in in the dcache, use it and return it. | |
2130 | * | |
2131 | * If no entry exists with the exact case name, allocate new dentry with | |
2132 | * the exact case, and return the spliced entry. | |
2133 | */ | |
2134 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, | |
2135 | struct qstr *name) | |
2136 | { | |
2137 | struct dentry *found, *res; | |
2138 | ||
2139 | /* | |
2140 | * First check if a dentry matching the name already exists, | |
2141 | * if not go ahead and create it now. | |
2142 | */ | |
2143 | found = d_hash_and_lookup(dentry->d_parent, name); | |
2144 | if (found) { | |
2145 | iput(inode); | |
2146 | return found; | |
2147 | } | |
2148 | if (d_in_lookup(dentry)) { | |
2149 | found = d_alloc_parallel(dentry->d_parent, name, | |
2150 | dentry->d_wait); | |
2151 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2152 | iput(inode); | |
2153 | return found; | |
2154 | } | |
2155 | } else { | |
2156 | found = d_alloc(dentry->d_parent, name); | |
2157 | if (!found) { | |
2158 | iput(inode); | |
2159 | return ERR_PTR(-ENOMEM); | |
2160 | } | |
2161 | } | |
2162 | res = d_splice_alias(inode, found); | |
2163 | if (res) { | |
2164 | dput(found); | |
2165 | return res; | |
2166 | } | |
2167 | return found; | |
2168 | } | |
2169 | EXPORT_SYMBOL(d_add_ci); | |
2170 | ||
2171 | ||
2172 | static inline bool d_same_name(const struct dentry *dentry, | |
2173 | const struct dentry *parent, | |
2174 | const struct qstr *name) | |
2175 | { | |
2176 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { | |
2177 | if (dentry->d_name.len != name->len) | |
2178 | return false; | |
2179 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
2180 | } | |
2181 | return parent->d_op->d_compare(dentry, | |
2182 | dentry->d_name.len, dentry->d_name.name, | |
2183 | name) == 0; | |
2184 | } | |
2185 | ||
2186 | /** | |
2187 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2188 | * @parent: parent dentry | |
2189 | * @name: qstr of name we wish to find | |
2190 | * @seqp: returns d_seq value at the point where the dentry was found | |
2191 | * Returns: dentry, or NULL | |
2192 | * | |
2193 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2194 | * resolution (store-free path walking) design described in | |
2195 | * Documentation/filesystems/path-lookup.txt. | |
2196 | * | |
2197 | * This is not to be used outside core vfs. | |
2198 | * | |
2199 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2200 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2201 | * without taking d_lock and checking d_seq sequence count against @seq | |
2202 | * returned here. | |
2203 | * | |
2204 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount | |
2205 | * function. | |
2206 | * | |
2207 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2208 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2209 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2210 | * is formed, giving integrity down the path walk. | |
2211 | * | |
2212 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2213 | * number we've returned before using any of the resulting dentry state! | |
2214 | */ | |
2215 | struct dentry *__d_lookup_rcu(const struct dentry *parent, | |
2216 | const struct qstr *name, | |
2217 | unsigned *seqp) | |
2218 | { | |
2219 | u64 hashlen = name->hash_len; | |
2220 | const unsigned char *str = name->name; | |
2221 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); | |
2222 | struct hlist_bl_node *node; | |
2223 | struct dentry *dentry; | |
2224 | ||
2225 | /* | |
2226 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2227 | * required to prevent single threaded performance regressions | |
2228 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2229 | * Keep the two functions in sync. | |
2230 | */ | |
2231 | ||
2232 | /* | |
2233 | * The hash list is protected using RCU. | |
2234 | * | |
2235 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2236 | * races with d_move(). | |
2237 | * | |
2238 | * It is possible that concurrent renames can mess up our list | |
2239 | * walk here and result in missing our dentry, resulting in the | |
2240 | * false-negative result. d_lookup() protects against concurrent | |
2241 | * renames using rename_lock seqlock. | |
2242 | * | |
2243 | * See Documentation/filesystems/path-lookup.txt for more details. | |
2244 | */ | |
2245 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | |
2246 | unsigned seq; | |
2247 | ||
2248 | seqretry: | |
2249 | /* | |
2250 | * The dentry sequence count protects us from concurrent | |
2251 | * renames, and thus protects parent and name fields. | |
2252 | * | |
2253 | * The caller must perform a seqcount check in order | |
2254 | * to do anything useful with the returned dentry. | |
2255 | * | |
2256 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2257 | * we don't wait for the sequence count to stabilize if it | |
2258 | * is in the middle of a sequence change. If we do the slow | |
2259 | * dentry compare, we will do seqretries until it is stable, | |
2260 | * and if we end up with a successful lookup, we actually | |
2261 | * want to exit RCU lookup anyway. | |
2262 | * | |
2263 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2264 | * we are still guaranteed NUL-termination of ->d_name.name. | |
2265 | */ | |
2266 | seq = raw_seqcount_begin(&dentry->d_seq); | |
2267 | if (dentry->d_parent != parent) | |
2268 | continue; | |
2269 | if (d_unhashed(dentry)) | |
2270 | continue; | |
2271 | ||
2272 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { | |
2273 | int tlen; | |
2274 | const char *tname; | |
2275 | if (dentry->d_name.hash != hashlen_hash(hashlen)) | |
2276 | continue; | |
2277 | tlen = dentry->d_name.len; | |
2278 | tname = dentry->d_name.name; | |
2279 | /* we want a consistent (name,len) pair */ | |
2280 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2281 | cpu_relax(); | |
2282 | goto seqretry; | |
2283 | } | |
2284 | if (parent->d_op->d_compare(dentry, | |
2285 | tlen, tname, name) != 0) | |
2286 | continue; | |
2287 | } else { | |
2288 | if (dentry->d_name.hash_len != hashlen) | |
2289 | continue; | |
2290 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2291 | continue; | |
2292 | } | |
2293 | *seqp = seq; | |
2294 | return dentry; | |
2295 | } | |
2296 | return NULL; | |
2297 | } | |
2298 | ||
2299 | /** | |
2300 | * d_lookup - search for a dentry | |
2301 | * @parent: parent dentry | |
2302 | * @name: qstr of name we wish to find | |
2303 | * Returns: dentry, or NULL | |
2304 | * | |
2305 | * d_lookup searches the children of the parent dentry for the name in | |
2306 | * question. If the dentry is found its reference count is incremented and the | |
2307 | * dentry is returned. The caller must use dput to free the entry when it has | |
2308 | * finished using it. %NULL is returned if the dentry does not exist. | |
2309 | */ | |
2310 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) | |
2311 | { | |
2312 | struct dentry *dentry; | |
2313 | unsigned seq; | |
2314 | ||
2315 | do { | |
2316 | seq = read_seqbegin(&rename_lock); | |
2317 | dentry = __d_lookup(parent, name); | |
2318 | if (dentry) | |
2319 | break; | |
2320 | } while (read_seqretry(&rename_lock, seq)); | |
2321 | return dentry; | |
2322 | } | |
2323 | EXPORT_SYMBOL(d_lookup); | |
2324 | ||
2325 | /** | |
2326 | * __d_lookup - search for a dentry (racy) | |
2327 | * @parent: parent dentry | |
2328 | * @name: qstr of name we wish to find | |
2329 | * Returns: dentry, or NULL | |
2330 | * | |
2331 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2332 | * false-negative result due to unrelated rename activity. | |
2333 | * | |
2334 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2335 | * however it must be used carefully, eg. with a following d_lookup in | |
2336 | * the case of failure. | |
2337 | * | |
2338 | * __d_lookup callers must be commented. | |
2339 | */ | |
2340 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) | |
2341 | { | |
2342 | unsigned int hash = name->hash; | |
2343 | struct hlist_bl_head *b = d_hash(hash); | |
2344 | struct hlist_bl_node *node; | |
2345 | struct dentry *found = NULL; | |
2346 | struct dentry *dentry; | |
2347 | ||
2348 | /* | |
2349 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2350 | * required to prevent single threaded performance regressions | |
2351 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2352 | * Keep the two functions in sync. | |
2353 | */ | |
2354 | ||
2355 | /* | |
2356 | * The hash list is protected using RCU. | |
2357 | * | |
2358 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2359 | * with d_move(). | |
2360 | * | |
2361 | * It is possible that concurrent renames can mess up our list | |
2362 | * walk here and result in missing our dentry, resulting in the | |
2363 | * false-negative result. d_lookup() protects against concurrent | |
2364 | * renames using rename_lock seqlock. | |
2365 | * | |
2366 | * See Documentation/filesystems/path-lookup.txt for more details. | |
2367 | */ | |
2368 | rcu_read_lock(); | |
2369 | ||
2370 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | |
2371 | ||
2372 | if (dentry->d_name.hash != hash) | |
2373 | continue; | |
2374 | ||
2375 | spin_lock(&dentry->d_lock); | |
2376 | if (dentry->d_parent != parent) | |
2377 | goto next; | |
2378 | if (d_unhashed(dentry)) | |
2379 | goto next; | |
2380 | ||
2381 | if (!d_same_name(dentry, parent, name)) | |
2382 | goto next; | |
2383 | ||
2384 | dentry->d_lockref.count++; | |
2385 | found = dentry; | |
2386 | spin_unlock(&dentry->d_lock); | |
2387 | break; | |
2388 | next: | |
2389 | spin_unlock(&dentry->d_lock); | |
2390 | } | |
2391 | rcu_read_unlock(); | |
2392 | ||
2393 | return found; | |
2394 | } | |
2395 | ||
2396 | /** | |
2397 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2398 | * @dir: Directory to search in | |
2399 | * @name: qstr of name we wish to find | |
2400 | * | |
2401 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) | |
2402 | */ | |
2403 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2404 | { | |
2405 | /* | |
2406 | * Check for a fs-specific hash function. Note that we must | |
2407 | * calculate the standard hash first, as the d_op->d_hash() | |
2408 | * routine may choose to leave the hash value unchanged. | |
2409 | */ | |
2410 | name->hash = full_name_hash(dir, name->name, name->len); | |
2411 | if (dir->d_flags & DCACHE_OP_HASH) { | |
2412 | int err = dir->d_op->d_hash(dir, name); | |
2413 | if (unlikely(err < 0)) | |
2414 | return ERR_PTR(err); | |
2415 | } | |
2416 | return d_lookup(dir, name); | |
2417 | } | |
2418 | EXPORT_SYMBOL(d_hash_and_lookup); | |
2419 | ||
2420 | /* | |
2421 | * When a file is deleted, we have two options: | |
2422 | * - turn this dentry into a negative dentry | |
2423 | * - unhash this dentry and free it. | |
2424 | * | |
2425 | * Usually, we want to just turn this into | |
2426 | * a negative dentry, but if anybody else is | |
2427 | * currently using the dentry or the inode | |
2428 | * we can't do that and we fall back on removing | |
2429 | * it from the hash queues and waiting for | |
2430 | * it to be deleted later when it has no users | |
2431 | */ | |
2432 | ||
2433 | /** | |
2434 | * d_delete - delete a dentry | |
2435 | * @dentry: The dentry to delete | |
2436 | * | |
2437 | * Turn the dentry into a negative dentry if possible, otherwise | |
2438 | * remove it from the hash queues so it can be deleted later | |
2439 | */ | |
2440 | ||
2441 | void d_delete(struct dentry * dentry) | |
2442 | { | |
2443 | struct inode *inode = dentry->d_inode; | |
2444 | ||
2445 | spin_lock(&inode->i_lock); | |
2446 | spin_lock(&dentry->d_lock); | |
2447 | /* | |
2448 | * Are we the only user? | |
2449 | */ | |
2450 | if (dentry->d_lockref.count == 1) { | |
2451 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; | |
2452 | dentry_unlink_inode(dentry); | |
2453 | } else { | |
2454 | __d_drop(dentry); | |
2455 | spin_unlock(&dentry->d_lock); | |
2456 | spin_unlock(&inode->i_lock); | |
2457 | } | |
2458 | } | |
2459 | EXPORT_SYMBOL(d_delete); | |
2460 | ||
2461 | static void __d_rehash(struct dentry *entry) | |
2462 | { | |
2463 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); | |
2464 | ||
2465 | hlist_bl_lock(b); | |
2466 | hlist_bl_add_head_rcu(&entry->d_hash, b); | |
2467 | hlist_bl_unlock(b); | |
2468 | } | |
2469 | ||
2470 | /** | |
2471 | * d_rehash - add an entry back to the hash | |
2472 | * @entry: dentry to add to the hash | |
2473 | * | |
2474 | * Adds a dentry to the hash according to its name. | |
2475 | */ | |
2476 | ||
2477 | void d_rehash(struct dentry * entry) | |
2478 | { | |
2479 | spin_lock(&entry->d_lock); | |
2480 | __d_rehash(entry); | |
2481 | spin_unlock(&entry->d_lock); | |
2482 | } | |
2483 | EXPORT_SYMBOL(d_rehash); | |
2484 | ||
2485 | static inline unsigned start_dir_add(struct inode *dir) | |
2486 | { | |
2487 | ||
2488 | for (;;) { | |
2489 | unsigned n = dir->i_dir_seq; | |
2490 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2491 | return n; | |
2492 | cpu_relax(); | |
2493 | } | |
2494 | } | |
2495 | ||
2496 | static inline void end_dir_add(struct inode *dir, unsigned n) | |
2497 | { | |
2498 | smp_store_release(&dir->i_dir_seq, n + 2); | |
2499 | } | |
2500 | ||
2501 | static void d_wait_lookup(struct dentry *dentry) | |
2502 | { | |
2503 | if (d_in_lookup(dentry)) { | |
2504 | DECLARE_WAITQUEUE(wait, current); | |
2505 | add_wait_queue(dentry->d_wait, &wait); | |
2506 | do { | |
2507 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2508 | spin_unlock(&dentry->d_lock); | |
2509 | schedule(); | |
2510 | spin_lock(&dentry->d_lock); | |
2511 | } while (d_in_lookup(dentry)); | |
2512 | } | |
2513 | } | |
2514 | ||
2515 | struct dentry *d_alloc_parallel(struct dentry *parent, | |
2516 | const struct qstr *name, | |
2517 | wait_queue_head_t *wq) | |
2518 | { | |
2519 | unsigned int hash = name->hash; | |
2520 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); | |
2521 | struct hlist_bl_node *node; | |
2522 | struct dentry *new = d_alloc(parent, name); | |
2523 | struct dentry *dentry; | |
2524 | unsigned seq, r_seq, d_seq; | |
2525 | ||
2526 | if (unlikely(!new)) | |
2527 | return ERR_PTR(-ENOMEM); | |
2528 | ||
2529 | retry: | |
2530 | rcu_read_lock(); | |
2531 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); | |
2532 | r_seq = read_seqbegin(&rename_lock); | |
2533 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2534 | if (unlikely(dentry)) { | |
2535 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2536 | rcu_read_unlock(); | |
2537 | goto retry; | |
2538 | } | |
2539 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2540 | rcu_read_unlock(); | |
2541 | dput(dentry); | |
2542 | goto retry; | |
2543 | } | |
2544 | rcu_read_unlock(); | |
2545 | dput(new); | |
2546 | return dentry; | |
2547 | } | |
2548 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2549 | rcu_read_unlock(); | |
2550 | goto retry; | |
2551 | } | |
2552 | ||
2553 | if (unlikely(seq & 1)) { | |
2554 | rcu_read_unlock(); | |
2555 | goto retry; | |
2556 | } | |
2557 | ||
2558 | hlist_bl_lock(b); | |
2559 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { | |
2560 | hlist_bl_unlock(b); | |
2561 | rcu_read_unlock(); | |
2562 | goto retry; | |
2563 | } | |
2564 | /* | |
2565 | * No changes for the parent since the beginning of d_lookup(). | |
2566 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2567 | * any potential in-lookup matches are going to stay here until | |
2568 | * we unlock the chain. All fields are stable in everything | |
2569 | * we encounter. | |
2570 | */ | |
2571 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2572 | if (dentry->d_name.hash != hash) | |
2573 | continue; | |
2574 | if (dentry->d_parent != parent) | |
2575 | continue; | |
2576 | if (!d_same_name(dentry, parent, name)) | |
2577 | continue; | |
2578 | hlist_bl_unlock(b); | |
2579 | /* now we can try to grab a reference */ | |
2580 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2581 | rcu_read_unlock(); | |
2582 | goto retry; | |
2583 | } | |
2584 | ||
2585 | rcu_read_unlock(); | |
2586 | /* | |
2587 | * somebody is likely to be still doing lookup for it; | |
2588 | * wait for them to finish | |
2589 | */ | |
2590 | spin_lock(&dentry->d_lock); | |
2591 | d_wait_lookup(dentry); | |
2592 | /* | |
2593 | * it's not in-lookup anymore; in principle we should repeat | |
2594 | * everything from dcache lookup, but it's likely to be what | |
2595 | * d_lookup() would've found anyway. If it is, just return it; | |
2596 | * otherwise we really have to repeat the whole thing. | |
2597 | */ | |
2598 | if (unlikely(dentry->d_name.hash != hash)) | |
2599 | goto mismatch; | |
2600 | if (unlikely(dentry->d_parent != parent)) | |
2601 | goto mismatch; | |
2602 | if (unlikely(d_unhashed(dentry))) | |
2603 | goto mismatch; | |
2604 | if (unlikely(!d_same_name(dentry, parent, name))) | |
2605 | goto mismatch; | |
2606 | /* OK, it *is* a hashed match; return it */ | |
2607 | spin_unlock(&dentry->d_lock); | |
2608 | dput(new); | |
2609 | return dentry; | |
2610 | } | |
2611 | rcu_read_unlock(); | |
2612 | /* we can't take ->d_lock here; it's OK, though. */ | |
2613 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
2614 | new->d_wait = wq; | |
2615 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); | |
2616 | hlist_bl_unlock(b); | |
2617 | return new; | |
2618 | mismatch: | |
2619 | spin_unlock(&dentry->d_lock); | |
2620 | dput(dentry); | |
2621 | goto retry; | |
2622 | } | |
2623 | EXPORT_SYMBOL(d_alloc_parallel); | |
2624 | ||
2625 | void __d_lookup_done(struct dentry *dentry) | |
2626 | { | |
2627 | struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent, | |
2628 | dentry->d_name.hash); | |
2629 | hlist_bl_lock(b); | |
2630 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; | |
2631 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); | |
2632 | wake_up_all(dentry->d_wait); | |
2633 | dentry->d_wait = NULL; | |
2634 | hlist_bl_unlock(b); | |
2635 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
2636 | INIT_LIST_HEAD(&dentry->d_lru); | |
2637 | } | |
2638 | EXPORT_SYMBOL(__d_lookup_done); | |
2639 | ||
2640 | /* inode->i_lock held if inode is non-NULL */ | |
2641 | ||
2642 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2643 | { | |
2644 | struct inode *dir = NULL; | |
2645 | unsigned n; | |
2646 | spin_lock(&dentry->d_lock); | |
2647 | if (unlikely(d_in_lookup(dentry))) { | |
2648 | dir = dentry->d_parent->d_inode; | |
2649 | n = start_dir_add(dir); | |
2650 | __d_lookup_done(dentry); | |
2651 | } | |
2652 | if (inode) { | |
2653 | unsigned add_flags = d_flags_for_inode(inode); | |
2654 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2655 | raw_write_seqcount_begin(&dentry->d_seq); | |
2656 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2657 | raw_write_seqcount_end(&dentry->d_seq); | |
2658 | fsnotify_update_flags(dentry); | |
2659 | } | |
2660 | __d_rehash(dentry); | |
2661 | if (dir) | |
2662 | end_dir_add(dir, n); | |
2663 | spin_unlock(&dentry->d_lock); | |
2664 | if (inode) | |
2665 | spin_unlock(&inode->i_lock); | |
2666 | } | |
2667 | ||
2668 | /** | |
2669 | * d_add - add dentry to hash queues | |
2670 | * @entry: dentry to add | |
2671 | * @inode: The inode to attach to this dentry | |
2672 | * | |
2673 | * This adds the entry to the hash queues and initializes @inode. | |
2674 | * The entry was actually filled in earlier during d_alloc(). | |
2675 | */ | |
2676 | ||
2677 | void d_add(struct dentry *entry, struct inode *inode) | |
2678 | { | |
2679 | if (inode) { | |
2680 | security_d_instantiate(entry, inode); | |
2681 | spin_lock(&inode->i_lock); | |
2682 | } | |
2683 | __d_add(entry, inode); | |
2684 | } | |
2685 | EXPORT_SYMBOL(d_add); | |
2686 | ||
2687 | /** | |
2688 | * d_exact_alias - find and hash an exact unhashed alias | |
2689 | * @entry: dentry to add | |
2690 | * @inode: The inode to go with this dentry | |
2691 | * | |
2692 | * If an unhashed dentry with the same name/parent and desired | |
2693 | * inode already exists, hash and return it. Otherwise, return | |
2694 | * NULL. | |
2695 | * | |
2696 | * Parent directory should be locked. | |
2697 | */ | |
2698 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2699 | { | |
2700 | struct dentry *alias; | |
2701 | unsigned int hash = entry->d_name.hash; | |
2702 | ||
2703 | spin_lock(&inode->i_lock); | |
2704 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2705 | /* | |
2706 | * Don't need alias->d_lock here, because aliases with | |
2707 | * d_parent == entry->d_parent are not subject to name or | |
2708 | * parent changes, because the parent inode i_mutex is held. | |
2709 | */ | |
2710 | if (alias->d_name.hash != hash) | |
2711 | continue; | |
2712 | if (alias->d_parent != entry->d_parent) | |
2713 | continue; | |
2714 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) | |
2715 | continue; | |
2716 | spin_lock(&alias->d_lock); | |
2717 | if (!d_unhashed(alias)) { | |
2718 | spin_unlock(&alias->d_lock); | |
2719 | alias = NULL; | |
2720 | } else { | |
2721 | __dget_dlock(alias); | |
2722 | __d_rehash(alias); | |
2723 | spin_unlock(&alias->d_lock); | |
2724 | } | |
2725 | spin_unlock(&inode->i_lock); | |
2726 | return alias; | |
2727 | } | |
2728 | spin_unlock(&inode->i_lock); | |
2729 | return NULL; | |
2730 | } | |
2731 | EXPORT_SYMBOL(d_exact_alias); | |
2732 | ||
2733 | static void swap_names(struct dentry *dentry, struct dentry *target) | |
2734 | { | |
2735 | if (unlikely(dname_external(target))) { | |
2736 | if (unlikely(dname_external(dentry))) { | |
2737 | /* | |
2738 | * Both external: swap the pointers | |
2739 | */ | |
2740 | swap(target->d_name.name, dentry->d_name.name); | |
2741 | } else { | |
2742 | /* | |
2743 | * dentry:internal, target:external. Steal target's | |
2744 | * storage and make target internal. | |
2745 | */ | |
2746 | memcpy(target->d_iname, dentry->d_name.name, | |
2747 | dentry->d_name.len + 1); | |
2748 | dentry->d_name.name = target->d_name.name; | |
2749 | target->d_name.name = target->d_iname; | |
2750 | } | |
2751 | } else { | |
2752 | if (unlikely(dname_external(dentry))) { | |
2753 | /* | |
2754 | * dentry:external, target:internal. Give dentry's | |
2755 | * storage to target and make dentry internal | |
2756 | */ | |
2757 | memcpy(dentry->d_iname, target->d_name.name, | |
2758 | target->d_name.len + 1); | |
2759 | target->d_name.name = dentry->d_name.name; | |
2760 | dentry->d_name.name = dentry->d_iname; | |
2761 | } else { | |
2762 | /* | |
2763 | * Both are internal. | |
2764 | */ | |
2765 | unsigned int i; | |
2766 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2767 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2768 | swap(((long *) &dentry->d_iname)[i], | |
2769 | ((long *) &target->d_iname)[i]); | |
2770 | } | |
2771 | } | |
2772 | } | |
2773 | swap(dentry->d_name.hash_len, target->d_name.hash_len); | |
2774 | } | |
2775 | ||
2776 | static void copy_name(struct dentry *dentry, struct dentry *target) | |
2777 | { | |
2778 | struct external_name *old_name = NULL; | |
2779 | if (unlikely(dname_external(dentry))) | |
2780 | old_name = external_name(dentry); | |
2781 | if (unlikely(dname_external(target))) { | |
2782 | atomic_inc(&external_name(target)->u.count); | |
2783 | dentry->d_name = target->d_name; | |
2784 | } else { | |
2785 | memcpy(dentry->d_iname, target->d_name.name, | |
2786 | target->d_name.len + 1); | |
2787 | dentry->d_name.name = dentry->d_iname; | |
2788 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2789 | } | |
2790 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2791 | kfree_rcu(old_name, u.head); | |
2792 | } | |
2793 | ||
2794 | /* | |
2795 | * __d_move - move a dentry | |
2796 | * @dentry: entry to move | |
2797 | * @target: new dentry | |
2798 | * @exchange: exchange the two dentries | |
2799 | * | |
2800 | * Update the dcache to reflect the move of a file name. Negative | |
2801 | * dcache entries should not be moved in this way. Caller must hold | |
2802 | * rename_lock, the i_mutex of the source and target directories, | |
2803 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
2804 | */ | |
2805 | static void __d_move(struct dentry *dentry, struct dentry *target, | |
2806 | bool exchange) | |
2807 | { | |
2808 | struct dentry *old_parent, *p; | |
2809 | struct inode *dir = NULL; | |
2810 | unsigned n; | |
2811 | ||
2812 | WARN_ON(!dentry->d_inode); | |
2813 | if (WARN_ON(dentry == target)) | |
2814 | return; | |
2815 | ||
2816 | BUG_ON(d_ancestor(target, dentry)); | |
2817 | old_parent = dentry->d_parent; | |
2818 | p = d_ancestor(old_parent, target); | |
2819 | if (IS_ROOT(dentry)) { | |
2820 | BUG_ON(p); | |
2821 | spin_lock(&target->d_parent->d_lock); | |
2822 | } else if (!p) { | |
2823 | /* target is not a descendent of dentry->d_parent */ | |
2824 | spin_lock(&target->d_parent->d_lock); | |
2825 | spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED); | |
2826 | } else { | |
2827 | BUG_ON(p == dentry); | |
2828 | spin_lock(&old_parent->d_lock); | |
2829 | if (p != target) | |
2830 | spin_lock_nested(&target->d_parent->d_lock, | |
2831 | DENTRY_D_LOCK_NESTED); | |
2832 | } | |
2833 | spin_lock_nested(&dentry->d_lock, 2); | |
2834 | spin_lock_nested(&target->d_lock, 3); | |
2835 | ||
2836 | if (unlikely(d_in_lookup(target))) { | |
2837 | dir = target->d_parent->d_inode; | |
2838 | n = start_dir_add(dir); | |
2839 | __d_lookup_done(target); | |
2840 | } | |
2841 | ||
2842 | write_seqcount_begin(&dentry->d_seq); | |
2843 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); | |
2844 | ||
2845 | /* unhash both */ | |
2846 | if (!d_unhashed(dentry)) | |
2847 | ___d_drop(dentry); | |
2848 | if (!d_unhashed(target)) | |
2849 | ___d_drop(target); | |
2850 | ||
2851 | /* ... and switch them in the tree */ | |
2852 | dentry->d_parent = target->d_parent; | |
2853 | if (!exchange) { | |
2854 | copy_name(dentry, target); | |
2855 | target->d_hash.pprev = NULL; | |
2856 | dentry->d_parent->d_lockref.count++; | |
2857 | if (dentry != old_parent) /* wasn't IS_ROOT */ | |
2858 | WARN_ON(!--old_parent->d_lockref.count); | |
2859 | } else { | |
2860 | target->d_parent = old_parent; | |
2861 | swap_names(dentry, target); | |
2862 | list_move(&target->d_child, &target->d_parent->d_subdirs); | |
2863 | __d_rehash(target); | |
2864 | fsnotify_update_flags(target); | |
2865 | } | |
2866 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); | |
2867 | __d_rehash(dentry); | |
2868 | fsnotify_update_flags(dentry); | |
2869 | fscrypt_handle_d_move(dentry); | |
2870 | ||
2871 | write_seqcount_end(&target->d_seq); | |
2872 | write_seqcount_end(&dentry->d_seq); | |
2873 | ||
2874 | if (dir) | |
2875 | end_dir_add(dir, n); | |
2876 | ||
2877 | if (dentry->d_parent != old_parent) | |
2878 | spin_unlock(&dentry->d_parent->d_lock); | |
2879 | if (dentry != old_parent) | |
2880 | spin_unlock(&old_parent->d_lock); | |
2881 | spin_unlock(&target->d_lock); | |
2882 | spin_unlock(&dentry->d_lock); | |
2883 | } | |
2884 | ||
2885 | /* | |
2886 | * d_move - move a dentry | |
2887 | * @dentry: entry to move | |
2888 | * @target: new dentry | |
2889 | * | |
2890 | * Update the dcache to reflect the move of a file name. Negative | |
2891 | * dcache entries should not be moved in this way. See the locking | |
2892 | * requirements for __d_move. | |
2893 | */ | |
2894 | void d_move(struct dentry *dentry, struct dentry *target) | |
2895 | { | |
2896 | write_seqlock(&rename_lock); | |
2897 | __d_move(dentry, target, false); | |
2898 | write_sequnlock(&rename_lock); | |
2899 | } | |
2900 | EXPORT_SYMBOL(d_move); | |
2901 | ||
2902 | /* | |
2903 | * d_exchange - exchange two dentries | |
2904 | * @dentry1: first dentry | |
2905 | * @dentry2: second dentry | |
2906 | */ | |
2907 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
2908 | { | |
2909 | write_seqlock(&rename_lock); | |
2910 | ||
2911 | WARN_ON(!dentry1->d_inode); | |
2912 | WARN_ON(!dentry2->d_inode); | |
2913 | WARN_ON(IS_ROOT(dentry1)); | |
2914 | WARN_ON(IS_ROOT(dentry2)); | |
2915 | ||
2916 | __d_move(dentry1, dentry2, true); | |
2917 | ||
2918 | write_sequnlock(&rename_lock); | |
2919 | } | |
2920 | EXPORT_SYMBOL_GPL(d_exchange); | |
2921 | ||
2922 | /** | |
2923 | * d_ancestor - search for an ancestor | |
2924 | * @p1: ancestor dentry | |
2925 | * @p2: child dentry | |
2926 | * | |
2927 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2928 | * an ancestor of p2, else NULL. | |
2929 | */ | |
2930 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) | |
2931 | { | |
2932 | struct dentry *p; | |
2933 | ||
2934 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { | |
2935 | if (p->d_parent == p1) | |
2936 | return p; | |
2937 | } | |
2938 | return NULL; | |
2939 | } | |
2940 | ||
2941 | /* | |
2942 | * This helper attempts to cope with remotely renamed directories | |
2943 | * | |
2944 | * It assumes that the caller is already holding | |
2945 | * dentry->d_parent->d_inode->i_mutex, and rename_lock | |
2946 | * | |
2947 | * Note: If ever the locking in lock_rename() changes, then please | |
2948 | * remember to update this too... | |
2949 | */ | |
2950 | static int __d_unalias(struct inode *inode, | |
2951 | struct dentry *dentry, struct dentry *alias) | |
2952 | { | |
2953 | struct mutex *m1 = NULL; | |
2954 | struct rw_semaphore *m2 = NULL; | |
2955 | int ret = -ESTALE; | |
2956 | ||
2957 | /* If alias and dentry share a parent, then no extra locks required */ | |
2958 | if (alias->d_parent == dentry->d_parent) | |
2959 | goto out_unalias; | |
2960 | ||
2961 | /* See lock_rename() */ | |
2962 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2963 | goto out_err; | |
2964 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2965 | if (!inode_trylock_shared(alias->d_parent->d_inode)) | |
2966 | goto out_err; | |
2967 | m2 = &alias->d_parent->d_inode->i_rwsem; | |
2968 | out_unalias: | |
2969 | __d_move(alias, dentry, false); | |
2970 | ret = 0; | |
2971 | out_err: | |
2972 | if (m2) | |
2973 | up_read(m2); | |
2974 | if (m1) | |
2975 | mutex_unlock(m1); | |
2976 | return ret; | |
2977 | } | |
2978 | ||
2979 | /** | |
2980 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
2981 | * @inode: the inode which may have a disconnected dentry | |
2982 | * @dentry: a negative dentry which we want to point to the inode. | |
2983 | * | |
2984 | * If inode is a directory and has an IS_ROOT alias, then d_move that in | |
2985 | * place of the given dentry and return it, else simply d_add the inode | |
2986 | * to the dentry and return NULL. | |
2987 | * | |
2988 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and | |
2989 | * we should error out: directories can't have multiple aliases. | |
2990 | * | |
2991 | * This is needed in the lookup routine of any filesystem that is exportable | |
2992 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
2993 | * | |
2994 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
2995 | * is returned. This matches the expected return value of ->lookup. | |
2996 | * | |
2997 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
2998 | * In that case, we know that the inode will be a regular file, and also this | |
2999 | * will only occur during atomic_open. So we need to check for the dentry | |
3000 | * being already hashed only in the final case. | |
3001 | */ | |
3002 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3003 | { | |
3004 | if (IS_ERR(inode)) | |
3005 | return ERR_CAST(inode); | |
3006 | ||
3007 | BUG_ON(!d_unhashed(dentry)); | |
3008 | ||
3009 | if (!inode) | |
3010 | goto out; | |
3011 | ||
3012 | security_d_instantiate(dentry, inode); | |
3013 | spin_lock(&inode->i_lock); | |
3014 | if (S_ISDIR(inode->i_mode)) { | |
3015 | struct dentry *new = __d_find_any_alias(inode); | |
3016 | if (unlikely(new)) { | |
3017 | /* The reference to new ensures it remains an alias */ | |
3018 | spin_unlock(&inode->i_lock); | |
3019 | write_seqlock(&rename_lock); | |
3020 | if (unlikely(d_ancestor(new, dentry))) { | |
3021 | write_sequnlock(&rename_lock); | |
3022 | dput(new); | |
3023 | new = ERR_PTR(-ELOOP); | |
3024 | pr_warn_ratelimited( | |
3025 | "VFS: Lookup of '%s' in %s %s" | |
3026 | " would have caused loop\n", | |
3027 | dentry->d_name.name, | |
3028 | inode->i_sb->s_type->name, | |
3029 | inode->i_sb->s_id); | |
3030 | } else if (!IS_ROOT(new)) { | |
3031 | struct dentry *old_parent = dget(new->d_parent); | |
3032 | int err = __d_unalias(inode, dentry, new); | |
3033 | write_sequnlock(&rename_lock); | |
3034 | if (err) { | |
3035 | dput(new); | |
3036 | new = ERR_PTR(err); | |
3037 | } | |
3038 | dput(old_parent); | |
3039 | } else { | |
3040 | __d_move(new, dentry, false); | |
3041 | write_sequnlock(&rename_lock); | |
3042 | } | |
3043 | iput(inode); | |
3044 | return new; | |
3045 | } | |
3046 | } | |
3047 | out: | |
3048 | __d_add(dentry, inode); | |
3049 | return NULL; | |
3050 | } | |
3051 | EXPORT_SYMBOL(d_splice_alias); | |
3052 | ||
3053 | /* | |
3054 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3055 | * | |
3056 | * Trivially implemented using the dcache structure | |
3057 | */ | |
3058 | ||
3059 | /** | |
3060 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3061 | * @new_dentry: new dentry | |
3062 | * @old_dentry: old dentry | |
3063 | * | |
3064 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). | |
3065 | * Returns false otherwise. | |
3066 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3067 | */ | |
3068 | ||
3069 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) | |
3070 | { | |
3071 | bool result; | |
3072 | unsigned seq; | |
3073 | ||
3074 | if (new_dentry == old_dentry) | |
3075 | return true; | |
3076 | ||
3077 | do { | |
3078 | /* for restarting inner loop in case of seq retry */ | |
3079 | seq = read_seqbegin(&rename_lock); | |
3080 | /* | |
3081 | * Need rcu_readlock to protect against the d_parent trashing | |
3082 | * due to d_move | |
3083 | */ | |
3084 | rcu_read_lock(); | |
3085 | if (d_ancestor(old_dentry, new_dentry)) | |
3086 | result = true; | |
3087 | else | |
3088 | result = false; | |
3089 | rcu_read_unlock(); | |
3090 | } while (read_seqretry(&rename_lock, seq)); | |
3091 | ||
3092 | return result; | |
3093 | } | |
3094 | EXPORT_SYMBOL(is_subdir); | |
3095 | ||
3096 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) | |
3097 | { | |
3098 | struct dentry *root = data; | |
3099 | if (dentry != root) { | |
3100 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3101 | return D_WALK_SKIP; | |
3102 | ||
3103 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { | |
3104 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3105 | dentry->d_lockref.count--; | |
3106 | } | |
3107 | } | |
3108 | return D_WALK_CONTINUE; | |
3109 | } | |
3110 | ||
3111 | void d_genocide(struct dentry *parent) | |
3112 | { | |
3113 | d_walk(parent, parent, d_genocide_kill); | |
3114 | } | |
3115 | ||
3116 | EXPORT_SYMBOL(d_genocide); | |
3117 | ||
3118 | void d_tmpfile(struct dentry *dentry, struct inode *inode) | |
3119 | { | |
3120 | inode_dec_link_count(inode); | |
3121 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3122 | !hlist_unhashed(&dentry->d_u.d_alias) || | |
3123 | !d_unlinked(dentry)); | |
3124 | spin_lock(&dentry->d_parent->d_lock); | |
3125 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3126 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3127 | (unsigned long long)inode->i_ino); | |
3128 | spin_unlock(&dentry->d_lock); | |
3129 | spin_unlock(&dentry->d_parent->d_lock); | |
3130 | d_instantiate(dentry, inode); | |
3131 | } | |
3132 | EXPORT_SYMBOL(d_tmpfile); | |
3133 | ||
3134 | static __initdata unsigned long dhash_entries; | |
3135 | static int __init set_dhash_entries(char *str) | |
3136 | { | |
3137 | if (!str) | |
3138 | return 0; | |
3139 | dhash_entries = simple_strtoul(str, &str, 0); | |
3140 | return 1; | |
3141 | } | |
3142 | __setup("dhash_entries=", set_dhash_entries); | |
3143 | ||
3144 | static void __init dcache_init_early(void) | |
3145 | { | |
3146 | /* If hashes are distributed across NUMA nodes, defer | |
3147 | * hash allocation until vmalloc space is available. | |
3148 | */ | |
3149 | if (hashdist) | |
3150 | return; | |
3151 | ||
3152 | dentry_hashtable = | |
3153 | alloc_large_system_hash("Dentry cache", | |
3154 | sizeof(struct hlist_bl_head), | |
3155 | dhash_entries, | |
3156 | 13, | |
3157 | HASH_EARLY | HASH_ZERO, | |
3158 | &d_hash_shift, | |
3159 | NULL, | |
3160 | 0, | |
3161 | 0); | |
3162 | d_hash_shift = 32 - d_hash_shift; | |
3163 | } | |
3164 | ||
3165 | static void __init dcache_init(void) | |
3166 | { | |
3167 | /* | |
3168 | * A constructor could be added for stable state like the lists, | |
3169 | * but it is probably not worth it because of the cache nature | |
3170 | * of the dcache. | |
3171 | */ | |
3172 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, | |
3173 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, | |
3174 | d_iname); | |
3175 | ||
3176 | /* Hash may have been set up in dcache_init_early */ | |
3177 | if (!hashdist) | |
3178 | return; | |
3179 | ||
3180 | dentry_hashtable = | |
3181 | alloc_large_system_hash("Dentry cache", | |
3182 | sizeof(struct hlist_bl_head), | |
3183 | dhash_entries, | |
3184 | 13, | |
3185 | HASH_ZERO, | |
3186 | &d_hash_shift, | |
3187 | NULL, | |
3188 | 0, | |
3189 | 0); | |
3190 | d_hash_shift = 32 - d_hash_shift; | |
3191 | } | |
3192 | ||
3193 | /* SLAB cache for __getname() consumers */ | |
3194 | struct kmem_cache *names_cachep __read_mostly; | |
3195 | EXPORT_SYMBOL(names_cachep); | |
3196 | ||
3197 | void __init vfs_caches_init_early(void) | |
3198 | { | |
3199 | int i; | |
3200 | ||
3201 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3202 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3203 | ||
3204 | dcache_init_early(); | |
3205 | inode_init_early(); | |
3206 | } | |
3207 | ||
3208 | void __init vfs_caches_init(void) | |
3209 | { | |
3210 | names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0, | |
3211 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL); | |
3212 | ||
3213 | dcache_init(); | |
3214 | inode_init(); | |
3215 | files_init(); | |
3216 | files_maxfiles_init(); | |
3217 | mnt_init(); | |
3218 | bdev_cache_init(); | |
3219 | chrdev_init(); | |
3220 | } |