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 __ro_after_init; |
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 | const struct qstr dotdot_name = QSTR_INIT(".." , 2); |
88 | EXPORT_SYMBOL(dotdot_name); |
89 | |
90 | /* |
91 | * This is the single most critical data structure when it comes |
92 | * to the dcache: the hashtable for lookups. Somebody should try |
93 | * to make this good - I've just made it work. |
94 | * |
95 | * This hash-function tries to avoid losing too many bits of hash |
96 | * information, yet avoid using a prime hash-size or similar. |
97 | */ |
98 | |
99 | static unsigned int d_hash_shift __ro_after_init; |
100 | |
101 | static struct hlist_bl_head *dentry_hashtable __ro_after_init; |
102 | |
103 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
104 | { |
105 | return dentry_hashtable + (hash >> d_hash_shift); |
106 | } |
107 | |
108 | #define IN_LOOKUP_SHIFT 10 |
109 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; |
110 | |
111 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, |
112 | unsigned int hash) |
113 | { |
114 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
115 | return in_lookup_hashtable + hash_32(val: hash, IN_LOOKUP_SHIFT); |
116 | } |
117 | |
118 | struct dentry_stat_t { |
119 | long nr_dentry; |
120 | long nr_unused; |
121 | long age_limit; /* age in seconds */ |
122 | long want_pages; /* pages requested by system */ |
123 | long nr_negative; /* # of unused negative dentries */ |
124 | long dummy; /* Reserved for future use */ |
125 | }; |
126 | |
127 | static DEFINE_PER_CPU(long, nr_dentry); |
128 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
129 | static DEFINE_PER_CPU(long, nr_dentry_negative); |
130 | |
131 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) |
132 | /* Statistics gathering. */ |
133 | static struct dentry_stat_t dentry_stat = { |
134 | .age_limit = 45, |
135 | }; |
136 | |
137 | /* |
138 | * Here we resort to our own counters instead of using generic per-cpu counters |
139 | * for consistency with what the vfs inode code does. We are expected to harvest |
140 | * better code and performance by having our own specialized counters. |
141 | * |
142 | * Please note that the loop is done over all possible CPUs, not over all online |
143 | * CPUs. The reason for this is that we don't want to play games with CPUs going |
144 | * on and off. If one of them goes off, we will just keep their counters. |
145 | * |
146 | * glommer: See cffbc8a for details, and if you ever intend to change this, |
147 | * please update all vfs counters to match. |
148 | */ |
149 | static long get_nr_dentry(void) |
150 | { |
151 | int i; |
152 | long sum = 0; |
153 | for_each_possible_cpu(i) |
154 | sum += per_cpu(nr_dentry, i); |
155 | return sum < 0 ? 0 : sum; |
156 | } |
157 | |
158 | static long get_nr_dentry_unused(void) |
159 | { |
160 | int i; |
161 | long sum = 0; |
162 | for_each_possible_cpu(i) |
163 | sum += per_cpu(nr_dentry_unused, i); |
164 | return sum < 0 ? 0 : sum; |
165 | } |
166 | |
167 | static long get_nr_dentry_negative(void) |
168 | { |
169 | int i; |
170 | long sum = 0; |
171 | |
172 | for_each_possible_cpu(i) |
173 | sum += per_cpu(nr_dentry_negative, i); |
174 | return sum < 0 ? 0 : sum; |
175 | } |
176 | |
177 | static int proc_nr_dentry(struct ctl_table *table, int write, void *buffer, |
178 | size_t *lenp, loff_t *ppos) |
179 | { |
180 | dentry_stat.nr_dentry = get_nr_dentry(); |
181 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
182 | dentry_stat.nr_negative = get_nr_dentry_negative(); |
183 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
184 | } |
185 | |
186 | static struct ctl_table fs_dcache_sysctls[] = { |
187 | { |
188 | .procname = "dentry-state" , |
189 | .data = &dentry_stat, |
190 | .maxlen = 6*sizeof(long), |
191 | .mode = 0444, |
192 | .proc_handler = proc_nr_dentry, |
193 | }, |
194 | { } |
195 | }; |
196 | |
197 | static int __init init_fs_dcache_sysctls(void) |
198 | { |
199 | register_sysctl_init("fs" , fs_dcache_sysctls); |
200 | return 0; |
201 | } |
202 | fs_initcall(init_fs_dcache_sysctls); |
203 | #endif |
204 | |
205 | /* |
206 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. |
207 | * The strings are both count bytes long, and count is non-zero. |
208 | */ |
209 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
210 | |
211 | #include <asm/word-at-a-time.h> |
212 | /* |
213 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a |
214 | * aligned allocation for this particular component. We don't |
215 | * strictly need the load_unaligned_zeropad() safety, but it |
216 | * doesn't hurt either. |
217 | * |
218 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do |
219 | * need the careful unaligned handling. |
220 | */ |
221 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
222 | { |
223 | unsigned long a,b,mask; |
224 | |
225 | for (;;) { |
226 | a = read_word_at_a_time(addr: cs); |
227 | b = load_unaligned_zeropad(addr: ct); |
228 | if (tcount < sizeof(unsigned long)) |
229 | break; |
230 | if (unlikely(a != b)) |
231 | return 1; |
232 | cs += sizeof(unsigned long); |
233 | ct += sizeof(unsigned long); |
234 | tcount -= sizeof(unsigned long); |
235 | if (!tcount) |
236 | return 0; |
237 | } |
238 | mask = bytemask_from_count(tcount); |
239 | return unlikely(!!((a ^ b) & mask)); |
240 | } |
241 | |
242 | #else |
243 | |
244 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
245 | { |
246 | do { |
247 | if (*cs != *ct) |
248 | return 1; |
249 | cs++; |
250 | ct++; |
251 | tcount--; |
252 | } while (tcount); |
253 | return 0; |
254 | } |
255 | |
256 | #endif |
257 | |
258 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
259 | { |
260 | /* |
261 | * Be careful about RCU walk racing with rename: |
262 | * use 'READ_ONCE' to fetch the name pointer. |
263 | * |
264 | * NOTE! Even if a rename will mean that the length |
265 | * was not loaded atomically, we don't care. The |
266 | * RCU walk will check the sequence count eventually, |
267 | * and catch it. And we won't overrun the buffer, |
268 | * because we're reading the name pointer atomically, |
269 | * and a dentry name is guaranteed to be properly |
270 | * terminated with a NUL byte. |
271 | * |
272 | * End result: even if 'len' is wrong, we'll exit |
273 | * early because the data cannot match (there can |
274 | * be no NUL in the ct/tcount data) |
275 | */ |
276 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
277 | |
278 | return dentry_string_cmp(cs, ct, tcount); |
279 | } |
280 | |
281 | struct external_name { |
282 | union { |
283 | atomic_t count; |
284 | struct rcu_head head; |
285 | } u; |
286 | unsigned char name[]; |
287 | }; |
288 | |
289 | static inline struct external_name *external_name(struct dentry *dentry) |
290 | { |
291 | return container_of(dentry->d_name.name, struct external_name, name[0]); |
292 | } |
293 | |
294 | static void __d_free(struct rcu_head *head) |
295 | { |
296 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
297 | |
298 | kmem_cache_free(s: dentry_cache, objp: dentry); |
299 | } |
300 | |
301 | static void __d_free_external(struct rcu_head *head) |
302 | { |
303 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
304 | kfree(objp: external_name(dentry)); |
305 | kmem_cache_free(s: dentry_cache, objp: dentry); |
306 | } |
307 | |
308 | static inline int dname_external(const struct dentry *dentry) |
309 | { |
310 | return dentry->d_name.name != dentry->d_iname; |
311 | } |
312 | |
313 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
314 | { |
315 | spin_lock(lock: &dentry->d_lock); |
316 | name->name = dentry->d_name; |
317 | if (unlikely(dname_external(dentry))) { |
318 | atomic_inc(v: &external_name(dentry)->u.count); |
319 | } else { |
320 | memcpy(name->inline_name, dentry->d_iname, |
321 | dentry->d_name.len + 1); |
322 | name->name.name = name->inline_name; |
323 | } |
324 | spin_unlock(lock: &dentry->d_lock); |
325 | } |
326 | EXPORT_SYMBOL(take_dentry_name_snapshot); |
327 | |
328 | void release_dentry_name_snapshot(struct name_snapshot *name) |
329 | { |
330 | if (unlikely(name->name.name != name->inline_name)) { |
331 | struct external_name *p; |
332 | p = container_of(name->name.name, struct external_name, name[0]); |
333 | if (unlikely(atomic_dec_and_test(&p->u.count))) |
334 | kfree_rcu(p, u.head); |
335 | } |
336 | } |
337 | EXPORT_SYMBOL(release_dentry_name_snapshot); |
338 | |
339 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
340 | struct inode *inode, |
341 | unsigned type_flags) |
342 | { |
343 | unsigned flags; |
344 | |
345 | dentry->d_inode = inode; |
346 | flags = READ_ONCE(dentry->d_flags); |
347 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); |
348 | flags |= type_flags; |
349 | smp_store_release(&dentry->d_flags, flags); |
350 | } |
351 | |
352 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
353 | { |
354 | unsigned flags = READ_ONCE(dentry->d_flags); |
355 | |
356 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); |
357 | WRITE_ONCE(dentry->d_flags, flags); |
358 | dentry->d_inode = NULL; |
359 | if (dentry->d_flags & DCACHE_LRU_LIST) |
360 | this_cpu_inc(nr_dentry_negative); |
361 | } |
362 | |
363 | static void dentry_free(struct dentry *dentry) |
364 | { |
365 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
366 | if (unlikely(dname_external(dentry))) { |
367 | struct external_name *p = external_name(dentry); |
368 | if (likely(atomic_dec_and_test(&p->u.count))) { |
369 | call_rcu(head: &dentry->d_u.d_rcu, func: __d_free_external); |
370 | return; |
371 | } |
372 | } |
373 | /* if dentry was never visible to RCU, immediate free is OK */ |
374 | if (dentry->d_flags & DCACHE_NORCU) |
375 | __d_free(head: &dentry->d_u.d_rcu); |
376 | else |
377 | call_rcu(head: &dentry->d_u.d_rcu, func: __d_free); |
378 | } |
379 | |
380 | /* |
381 | * Release the dentry's inode, using the filesystem |
382 | * d_iput() operation if defined. |
383 | */ |
384 | static void dentry_unlink_inode(struct dentry * dentry) |
385 | __releases(dentry->d_lock) |
386 | __releases(dentry->d_inode->i_lock) |
387 | { |
388 | struct inode *inode = dentry->d_inode; |
389 | |
390 | raw_write_seqcount_begin(&dentry->d_seq); |
391 | __d_clear_type_and_inode(dentry); |
392 | hlist_del_init(n: &dentry->d_u.d_alias); |
393 | raw_write_seqcount_end(&dentry->d_seq); |
394 | spin_unlock(lock: &dentry->d_lock); |
395 | spin_unlock(lock: &inode->i_lock); |
396 | if (!inode->i_nlink) |
397 | fsnotify_inoderemove(inode); |
398 | if (dentry->d_op && dentry->d_op->d_iput) |
399 | dentry->d_op->d_iput(dentry, inode); |
400 | else |
401 | iput(inode); |
402 | } |
403 | |
404 | /* |
405 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry |
406 | * is in use - which includes both the "real" per-superblock |
407 | * LRU list _and_ the DCACHE_SHRINK_LIST use. |
408 | * |
409 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is |
410 | * on the shrink list (ie not on the superblock LRU list). |
411 | * |
412 | * The per-cpu "nr_dentry_unused" counters are updated with |
413 | * the DCACHE_LRU_LIST bit. |
414 | * |
415 | * The per-cpu "nr_dentry_negative" counters are only updated |
416 | * when deleted from or added to the per-superblock LRU list, not |
417 | * from/to the shrink list. That is to avoid an unneeded dec/inc |
418 | * pair when moving from LRU to shrink list in select_collect(). |
419 | * |
420 | * These helper functions make sure we always follow the |
421 | * rules. d_lock must be held by the caller. |
422 | */ |
423 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) |
424 | static void d_lru_add(struct dentry *dentry) |
425 | { |
426 | D_FLAG_VERIFY(dentry, 0); |
427 | dentry->d_flags |= DCACHE_LRU_LIST; |
428 | this_cpu_inc(nr_dentry_unused); |
429 | if (d_is_negative(dentry)) |
430 | this_cpu_inc(nr_dentry_negative); |
431 | WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
432 | } |
433 | |
434 | static void d_lru_del(struct dentry *dentry) |
435 | { |
436 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); |
437 | dentry->d_flags &= ~DCACHE_LRU_LIST; |
438 | this_cpu_dec(nr_dentry_unused); |
439 | if (d_is_negative(dentry)) |
440 | this_cpu_dec(nr_dentry_negative); |
441 | WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru)); |
442 | } |
443 | |
444 | static void d_shrink_del(struct dentry *dentry) |
445 | { |
446 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); |
447 | list_del_init(entry: &dentry->d_lru); |
448 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); |
449 | this_cpu_dec(nr_dentry_unused); |
450 | } |
451 | |
452 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) |
453 | { |
454 | D_FLAG_VERIFY(dentry, 0); |
455 | list_add(new: &dentry->d_lru, head: list); |
456 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; |
457 | this_cpu_inc(nr_dentry_unused); |
458 | } |
459 | |
460 | /* |
461 | * These can only be called under the global LRU lock, ie during the |
462 | * callback for freeing the LRU list. "isolate" removes it from the |
463 | * LRU lists entirely, while shrink_move moves it to the indicated |
464 | * private list. |
465 | */ |
466 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
467 | { |
468 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); |
469 | dentry->d_flags &= ~DCACHE_LRU_LIST; |
470 | this_cpu_dec(nr_dentry_unused); |
471 | if (d_is_negative(dentry)) |
472 | this_cpu_dec(nr_dentry_negative); |
473 | list_lru_isolate(list: lru, item: &dentry->d_lru); |
474 | } |
475 | |
476 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
477 | struct list_head *list) |
478 | { |
479 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); |
480 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
481 | if (d_is_negative(dentry)) |
482 | this_cpu_dec(nr_dentry_negative); |
483 | list_lru_isolate_move(list: lru, item: &dentry->d_lru, head: list); |
484 | } |
485 | |
486 | static void ___d_drop(struct dentry *dentry) |
487 | { |
488 | struct hlist_bl_head *b; |
489 | /* |
490 | * Hashed dentries are normally on the dentry hashtable, |
491 | * with the exception of those newly allocated by |
492 | * d_obtain_root, which are always IS_ROOT: |
493 | */ |
494 | if (unlikely(IS_ROOT(dentry))) |
495 | b = &dentry->d_sb->s_roots; |
496 | else |
497 | b = d_hash(hash: dentry->d_name.hash); |
498 | |
499 | hlist_bl_lock(b); |
500 | __hlist_bl_del(n: &dentry->d_hash); |
501 | hlist_bl_unlock(b); |
502 | } |
503 | |
504 | void __d_drop(struct dentry *dentry) |
505 | { |
506 | if (!d_unhashed(dentry)) { |
507 | ___d_drop(dentry); |
508 | dentry->d_hash.pprev = NULL; |
509 | write_seqcount_invalidate(&dentry->d_seq); |
510 | } |
511 | } |
512 | EXPORT_SYMBOL(__d_drop); |
513 | |
514 | /** |
515 | * d_drop - drop a dentry |
516 | * @dentry: dentry to drop |
517 | * |
518 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't |
519 | * be found through a VFS lookup any more. Note that this is different from |
520 | * deleting the dentry - d_delete will try to mark the dentry negative if |
521 | * possible, giving a successful _negative_ lookup, while d_drop will |
522 | * just make the cache lookup fail. |
523 | * |
524 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some |
525 | * reason (NFS timeouts or autofs deletes). |
526 | * |
527 | * __d_drop requires dentry->d_lock |
528 | * |
529 | * ___d_drop doesn't mark dentry as "unhashed" |
530 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). |
531 | */ |
532 | void d_drop(struct dentry *dentry) |
533 | { |
534 | spin_lock(lock: &dentry->d_lock); |
535 | __d_drop(dentry); |
536 | spin_unlock(lock: &dentry->d_lock); |
537 | } |
538 | EXPORT_SYMBOL(d_drop); |
539 | |
540 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
541 | { |
542 | struct dentry *next; |
543 | /* |
544 | * Inform d_walk() and shrink_dentry_list() that we are no longer |
545 | * attached to the dentry tree |
546 | */ |
547 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
548 | if (unlikely(list_empty(&dentry->d_child))) |
549 | return; |
550 | __list_del_entry(entry: &dentry->d_child); |
551 | /* |
552 | * Cursors can move around the list of children. While we'd been |
553 | * a normal list member, it didn't matter - ->d_child.next would've |
554 | * been updated. However, from now on it won't be and for the |
555 | * things like d_walk() it might end up with a nasty surprise. |
556 | * Normally d_walk() doesn't care about cursors moving around - |
557 | * ->d_lock on parent prevents that and since a cursor has no children |
558 | * of its own, we get through it without ever unlocking the parent. |
559 | * There is one exception, though - if we ascend from a child that |
560 | * gets killed as soon as we unlock it, the next sibling is found |
561 | * using the value left in its ->d_child.next. And if _that_ |
562 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) |
563 | * before d_walk() regains parent->d_lock, we'll end up skipping |
564 | * everything the cursor had been moved past. |
565 | * |
566 | * Solution: make sure that the pointer left behind in ->d_child.next |
567 | * points to something that won't be moving around. I.e. skip the |
568 | * cursors. |
569 | */ |
570 | while (dentry->d_child.next != &parent->d_subdirs) { |
571 | next = list_entry(dentry->d_child.next, struct dentry, d_child); |
572 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) |
573 | break; |
574 | dentry->d_child.next = next->d_child.next; |
575 | } |
576 | } |
577 | |
578 | static void __dentry_kill(struct dentry *dentry) |
579 | { |
580 | struct dentry *parent = NULL; |
581 | bool can_free = true; |
582 | if (!IS_ROOT(dentry)) |
583 | parent = dentry->d_parent; |
584 | |
585 | /* |
586 | * The dentry is now unrecoverably dead to the world. |
587 | */ |
588 | lockref_mark_dead(&dentry->d_lockref); |
589 | |
590 | /* |
591 | * inform the fs via d_prune that this dentry is about to be |
592 | * unhashed and destroyed. |
593 | */ |
594 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
595 | dentry->d_op->d_prune(dentry); |
596 | |
597 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
598 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) |
599 | d_lru_del(dentry); |
600 | } |
601 | /* if it was on the hash then remove it */ |
602 | __d_drop(dentry); |
603 | dentry_unlist(dentry, parent); |
604 | if (parent) |
605 | spin_unlock(lock: &parent->d_lock); |
606 | if (dentry->d_inode) |
607 | dentry_unlink_inode(dentry); |
608 | else |
609 | spin_unlock(lock: &dentry->d_lock); |
610 | this_cpu_dec(nr_dentry); |
611 | if (dentry->d_op && dentry->d_op->d_release) |
612 | dentry->d_op->d_release(dentry); |
613 | |
614 | spin_lock(lock: &dentry->d_lock); |
615 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
616 | dentry->d_flags |= DCACHE_MAY_FREE; |
617 | can_free = false; |
618 | } |
619 | spin_unlock(lock: &dentry->d_lock); |
620 | if (likely(can_free)) |
621 | dentry_free(dentry); |
622 | cond_resched(); |
623 | } |
624 | |
625 | static struct dentry *__lock_parent(struct dentry *dentry) |
626 | { |
627 | struct dentry *parent; |
628 | rcu_read_lock(); |
629 | spin_unlock(lock: &dentry->d_lock); |
630 | again: |
631 | parent = READ_ONCE(dentry->d_parent); |
632 | spin_lock(lock: &parent->d_lock); |
633 | /* |
634 | * We can't blindly lock dentry until we are sure |
635 | * that we won't violate the locking order. |
636 | * Any changes of dentry->d_parent must have |
637 | * been done with parent->d_lock held, so |
638 | * spin_lock() above is enough of a barrier |
639 | * for checking if it's still our child. |
640 | */ |
641 | if (unlikely(parent != dentry->d_parent)) { |
642 | spin_unlock(lock: &parent->d_lock); |
643 | goto again; |
644 | } |
645 | rcu_read_unlock(); |
646 | if (parent != dentry) |
647 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
648 | else |
649 | parent = NULL; |
650 | return parent; |
651 | } |
652 | |
653 | static inline struct dentry *lock_parent(struct dentry *dentry) |
654 | { |
655 | struct dentry *parent = dentry->d_parent; |
656 | if (IS_ROOT(dentry)) |
657 | return NULL; |
658 | if (likely(spin_trylock(&parent->d_lock))) |
659 | return parent; |
660 | return __lock_parent(dentry); |
661 | } |
662 | |
663 | static inline bool retain_dentry(struct dentry *dentry) |
664 | { |
665 | WARN_ON(d_in_lookup(dentry)); |
666 | |
667 | /* Unreachable? Get rid of it */ |
668 | if (unlikely(d_unhashed(dentry))) |
669 | return false; |
670 | |
671 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
672 | return false; |
673 | |
674 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { |
675 | if (dentry->d_op->d_delete(dentry)) |
676 | return false; |
677 | } |
678 | |
679 | if (unlikely(dentry->d_flags & DCACHE_DONTCACHE)) |
680 | return false; |
681 | |
682 | /* retain; LRU fodder */ |
683 | dentry->d_lockref.count--; |
684 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) |
685 | d_lru_add(dentry); |
686 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) |
687 | dentry->d_flags |= DCACHE_REFERENCED; |
688 | return true; |
689 | } |
690 | |
691 | void d_mark_dontcache(struct inode *inode) |
692 | { |
693 | struct dentry *de; |
694 | |
695 | spin_lock(lock: &inode->i_lock); |
696 | hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) { |
697 | spin_lock(lock: &de->d_lock); |
698 | de->d_flags |= DCACHE_DONTCACHE; |
699 | spin_unlock(lock: &de->d_lock); |
700 | } |
701 | inode->i_state |= I_DONTCACHE; |
702 | spin_unlock(lock: &inode->i_lock); |
703 | } |
704 | EXPORT_SYMBOL(d_mark_dontcache); |
705 | |
706 | /* |
707 | * Finish off a dentry we've decided to kill. |
708 | * dentry->d_lock must be held, returns with it unlocked. |
709 | * Returns dentry requiring refcount drop, or NULL if we're done. |
710 | */ |
711 | static struct dentry *dentry_kill(struct dentry *dentry) |
712 | __releases(dentry->d_lock) |
713 | { |
714 | struct inode *inode = dentry->d_inode; |
715 | struct dentry *parent = NULL; |
716 | |
717 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) |
718 | goto slow_positive; |
719 | |
720 | if (!IS_ROOT(dentry)) { |
721 | parent = dentry->d_parent; |
722 | if (unlikely(!spin_trylock(&parent->d_lock))) { |
723 | parent = __lock_parent(dentry); |
724 | if (likely(inode || !dentry->d_inode)) |
725 | goto got_locks; |
726 | /* negative that became positive */ |
727 | if (parent) |
728 | spin_unlock(lock: &parent->d_lock); |
729 | inode = dentry->d_inode; |
730 | goto slow_positive; |
731 | } |
732 | } |
733 | __dentry_kill(dentry); |
734 | return parent; |
735 | |
736 | slow_positive: |
737 | spin_unlock(lock: &dentry->d_lock); |
738 | spin_lock(lock: &inode->i_lock); |
739 | spin_lock(lock: &dentry->d_lock); |
740 | parent = lock_parent(dentry); |
741 | got_locks: |
742 | if (unlikely(dentry->d_lockref.count != 1)) { |
743 | dentry->d_lockref.count--; |
744 | } else if (likely(!retain_dentry(dentry))) { |
745 | __dentry_kill(dentry); |
746 | return parent; |
747 | } |
748 | /* we are keeping it, after all */ |
749 | if (inode) |
750 | spin_unlock(lock: &inode->i_lock); |
751 | if (parent) |
752 | spin_unlock(lock: &parent->d_lock); |
753 | spin_unlock(lock: &dentry->d_lock); |
754 | return NULL; |
755 | } |
756 | |
757 | /* |
758 | * Try to do a lockless dput(), and return whether that was successful. |
759 | * |
760 | * If unsuccessful, we return false, having already taken the dentry lock. |
761 | * |
762 | * The caller needs to hold the RCU read lock, so that the dentry is |
763 | * guaranteed to stay around even if the refcount goes down to zero! |
764 | */ |
765 | static inline bool fast_dput(struct dentry *dentry) |
766 | { |
767 | int ret; |
768 | unsigned int d_flags; |
769 | |
770 | /* |
771 | * If we have a d_op->d_delete() operation, we sould not |
772 | * let the dentry count go to zero, so use "put_or_lock". |
773 | */ |
774 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) |
775 | return lockref_put_or_lock(&dentry->d_lockref); |
776 | |
777 | /* |
778 | * .. otherwise, we can try to just decrement the |
779 | * lockref optimistically. |
780 | */ |
781 | ret = lockref_put_return(&dentry->d_lockref); |
782 | |
783 | /* |
784 | * If the lockref_put_return() failed due to the lock being held |
785 | * by somebody else, the fast path has failed. We will need to |
786 | * get the lock, and then check the count again. |
787 | */ |
788 | if (unlikely(ret < 0)) { |
789 | spin_lock(lock: &dentry->d_lock); |
790 | if (dentry->d_lockref.count > 1) { |
791 | dentry->d_lockref.count--; |
792 | spin_unlock(lock: &dentry->d_lock); |
793 | return true; |
794 | } |
795 | return false; |
796 | } |
797 | |
798 | /* |
799 | * If we weren't the last ref, we're done. |
800 | */ |
801 | if (ret) |
802 | return true; |
803 | |
804 | /* |
805 | * Careful, careful. The reference count went down |
806 | * to zero, but we don't hold the dentry lock, so |
807 | * somebody else could get it again, and do another |
808 | * dput(), and we need to not race with that. |
809 | * |
810 | * However, there is a very special and common case |
811 | * where we don't care, because there is nothing to |
812 | * do: the dentry is still hashed, it does not have |
813 | * a 'delete' op, and it's referenced and already on |
814 | * the LRU list. |
815 | * |
816 | * NOTE! Since we aren't locked, these values are |
817 | * not "stable". However, it is sufficient that at |
818 | * some point after we dropped the reference the |
819 | * dentry was hashed and the flags had the proper |
820 | * value. Other dentry users may have re-gotten |
821 | * a reference to the dentry and change that, but |
822 | * our work is done - we can leave the dentry |
823 | * around with a zero refcount. |
824 | * |
825 | * Nevertheless, there are two cases that we should kill |
826 | * the dentry anyway. |
827 | * 1. free disconnected dentries as soon as their refcount |
828 | * reached zero. |
829 | * 2. free dentries if they should not be cached. |
830 | */ |
831 | smp_rmb(); |
832 | d_flags = READ_ONCE(dentry->d_flags); |
833 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | |
834 | DCACHE_DISCONNECTED | DCACHE_DONTCACHE; |
835 | |
836 | /* Nothing to do? Dropping the reference was all we needed? */ |
837 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) |
838 | return true; |
839 | |
840 | /* |
841 | * Not the fast normal case? Get the lock. We've already decremented |
842 | * the refcount, but we'll need to re-check the situation after |
843 | * getting the lock. |
844 | */ |
845 | spin_lock(lock: &dentry->d_lock); |
846 | |
847 | /* |
848 | * Did somebody else grab a reference to it in the meantime, and |
849 | * we're no longer the last user after all? Alternatively, somebody |
850 | * else could have killed it and marked it dead. Either way, we |
851 | * don't need to do anything else. |
852 | */ |
853 | if (dentry->d_lockref.count) { |
854 | spin_unlock(lock: &dentry->d_lock); |
855 | return true; |
856 | } |
857 | |
858 | /* |
859 | * Re-get the reference we optimistically dropped. We hold the |
860 | * lock, and we just tested that it was zero, so we can just |
861 | * set it to 1. |
862 | */ |
863 | dentry->d_lockref.count = 1; |
864 | return false; |
865 | } |
866 | |
867 | |
868 | /* |
869 | * This is dput |
870 | * |
871 | * This is complicated by the fact that we do not want to put |
872 | * dentries that are no longer on any hash chain on the unused |
873 | * list: we'd much rather just get rid of them immediately. |
874 | * |
875 | * However, that implies that we have to traverse the dentry |
876 | * tree upwards to the parents which might _also_ now be |
877 | * scheduled for deletion (it may have been only waiting for |
878 | * its last child to go away). |
879 | * |
880 | * This tail recursion is done by hand as we don't want to depend |
881 | * on the compiler to always get this right (gcc generally doesn't). |
882 | * Real recursion would eat up our stack space. |
883 | */ |
884 | |
885 | /* |
886 | * dput - release a dentry |
887 | * @dentry: dentry to release |
888 | * |
889 | * Release a dentry. This will drop the usage count and if appropriate |
890 | * call the dentry unlink method as well as removing it from the queues and |
891 | * releasing its resources. If the parent dentries were scheduled for release |
892 | * they too may now get deleted. |
893 | */ |
894 | void dput(struct dentry *dentry) |
895 | { |
896 | while (dentry) { |
897 | might_sleep(); |
898 | |
899 | rcu_read_lock(); |
900 | if (likely(fast_dput(dentry))) { |
901 | rcu_read_unlock(); |
902 | return; |
903 | } |
904 | |
905 | /* Slow case: now with the dentry lock held */ |
906 | rcu_read_unlock(); |
907 | |
908 | if (likely(retain_dentry(dentry))) { |
909 | spin_unlock(lock: &dentry->d_lock); |
910 | return; |
911 | } |
912 | |
913 | dentry = dentry_kill(dentry); |
914 | } |
915 | } |
916 | EXPORT_SYMBOL(dput); |
917 | |
918 | static void __dput_to_list(struct dentry *dentry, struct list_head *list) |
919 | __must_hold(&dentry->d_lock) |
920 | { |
921 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
922 | /* let the owner of the list it's on deal with it */ |
923 | --dentry->d_lockref.count; |
924 | } else { |
925 | if (dentry->d_flags & DCACHE_LRU_LIST) |
926 | d_lru_del(dentry); |
927 | if (!--dentry->d_lockref.count) |
928 | d_shrink_add(dentry, list); |
929 | } |
930 | } |
931 | |
932 | void dput_to_list(struct dentry *dentry, struct list_head *list) |
933 | { |
934 | rcu_read_lock(); |
935 | if (likely(fast_dput(dentry))) { |
936 | rcu_read_unlock(); |
937 | return; |
938 | } |
939 | rcu_read_unlock(); |
940 | if (!retain_dentry(dentry)) |
941 | __dput_to_list(dentry, list); |
942 | spin_unlock(lock: &dentry->d_lock); |
943 | } |
944 | |
945 | /* This must be called with d_lock held */ |
946 | static inline void __dget_dlock(struct dentry *dentry) |
947 | { |
948 | dentry->d_lockref.count++; |
949 | } |
950 | |
951 | static inline void __dget(struct dentry *dentry) |
952 | { |
953 | lockref_get(&dentry->d_lockref); |
954 | } |
955 | |
956 | struct dentry *dget_parent(struct dentry *dentry) |
957 | { |
958 | int gotref; |
959 | struct dentry *ret; |
960 | unsigned seq; |
961 | |
962 | /* |
963 | * Do optimistic parent lookup without any |
964 | * locking. |
965 | */ |
966 | rcu_read_lock(); |
967 | seq = raw_seqcount_begin(&dentry->d_seq); |
968 | ret = READ_ONCE(dentry->d_parent); |
969 | gotref = lockref_get_not_zero(&ret->d_lockref); |
970 | rcu_read_unlock(); |
971 | if (likely(gotref)) { |
972 | if (!read_seqcount_retry(&dentry->d_seq, seq)) |
973 | return ret; |
974 | dput(ret); |
975 | } |
976 | |
977 | repeat: |
978 | /* |
979 | * Don't need rcu_dereference because we re-check it was correct under |
980 | * the lock. |
981 | */ |
982 | rcu_read_lock(); |
983 | ret = dentry->d_parent; |
984 | spin_lock(lock: &ret->d_lock); |
985 | if (unlikely(ret != dentry->d_parent)) { |
986 | spin_unlock(lock: &ret->d_lock); |
987 | rcu_read_unlock(); |
988 | goto repeat; |
989 | } |
990 | rcu_read_unlock(); |
991 | BUG_ON(!ret->d_lockref.count); |
992 | ret->d_lockref.count++; |
993 | spin_unlock(lock: &ret->d_lock); |
994 | return ret; |
995 | } |
996 | EXPORT_SYMBOL(dget_parent); |
997 | |
998 | static struct dentry * __d_find_any_alias(struct inode *inode) |
999 | { |
1000 | struct dentry *alias; |
1001 | |
1002 | if (hlist_empty(h: &inode->i_dentry)) |
1003 | return NULL; |
1004 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
1005 | __dget(dentry: alias); |
1006 | return alias; |
1007 | } |
1008 | |
1009 | /** |
1010 | * d_find_any_alias - find any alias for a given inode |
1011 | * @inode: inode to find an alias for |
1012 | * |
1013 | * If any aliases exist for the given inode, take and return a |
1014 | * reference for one of them. If no aliases exist, return %NULL. |
1015 | */ |
1016 | struct dentry *d_find_any_alias(struct inode *inode) |
1017 | { |
1018 | struct dentry *de; |
1019 | |
1020 | spin_lock(lock: &inode->i_lock); |
1021 | de = __d_find_any_alias(inode); |
1022 | spin_unlock(lock: &inode->i_lock); |
1023 | return de; |
1024 | } |
1025 | EXPORT_SYMBOL(d_find_any_alias); |
1026 | |
1027 | static struct dentry *__d_find_alias(struct inode *inode) |
1028 | { |
1029 | struct dentry *alias; |
1030 | |
1031 | if (S_ISDIR(inode->i_mode)) |
1032 | return __d_find_any_alias(inode); |
1033 | |
1034 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
1035 | spin_lock(lock: &alias->d_lock); |
1036 | if (!d_unhashed(dentry: alias)) { |
1037 | __dget_dlock(dentry: alias); |
1038 | spin_unlock(lock: &alias->d_lock); |
1039 | return alias; |
1040 | } |
1041 | spin_unlock(lock: &alias->d_lock); |
1042 | } |
1043 | return NULL; |
1044 | } |
1045 | |
1046 | /** |
1047 | * d_find_alias - grab a hashed alias of inode |
1048 | * @inode: inode in question |
1049 | * |
1050 | * If inode has a hashed alias, or is a directory and has any alias, |
1051 | * acquire the reference to alias and return it. Otherwise return NULL. |
1052 | * Notice that if inode is a directory there can be only one alias and |
1053 | * it can be unhashed only if it has no children, or if it is the root |
1054 | * of a filesystem, or if the directory was renamed and d_revalidate |
1055 | * was the first vfs operation to notice. |
1056 | * |
1057 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1058 | * any other hashed alias over that one. |
1059 | */ |
1060 | struct dentry *d_find_alias(struct inode *inode) |
1061 | { |
1062 | struct dentry *de = NULL; |
1063 | |
1064 | if (!hlist_empty(h: &inode->i_dentry)) { |
1065 | spin_lock(lock: &inode->i_lock); |
1066 | de = __d_find_alias(inode); |
1067 | spin_unlock(lock: &inode->i_lock); |
1068 | } |
1069 | return de; |
1070 | } |
1071 | EXPORT_SYMBOL(d_find_alias); |
1072 | |
1073 | /* |
1074 | * Caller MUST be holding rcu_read_lock() and be guaranteed |
1075 | * that inode won't get freed until rcu_read_unlock(). |
1076 | */ |
1077 | struct dentry *d_find_alias_rcu(struct inode *inode) |
1078 | { |
1079 | struct hlist_head *l = &inode->i_dentry; |
1080 | struct dentry *de = NULL; |
1081 | |
1082 | spin_lock(lock: &inode->i_lock); |
1083 | // ->i_dentry and ->i_rcu are colocated, but the latter won't be |
1084 | // used without having I_FREEING set, which means no aliases left |
1085 | if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) { |
1086 | if (S_ISDIR(inode->i_mode)) { |
1087 | de = hlist_entry(l->first, struct dentry, d_u.d_alias); |
1088 | } else { |
1089 | hlist_for_each_entry(de, l, d_u.d_alias) |
1090 | if (!d_unhashed(dentry: de)) |
1091 | break; |
1092 | } |
1093 | } |
1094 | spin_unlock(lock: &inode->i_lock); |
1095 | return de; |
1096 | } |
1097 | |
1098 | /* |
1099 | * Try to kill dentries associated with this inode. |
1100 | * WARNING: you must own a reference to inode. |
1101 | */ |
1102 | void d_prune_aliases(struct inode *inode) |
1103 | { |
1104 | struct dentry *dentry; |
1105 | restart: |
1106 | spin_lock(lock: &inode->i_lock); |
1107 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1108 | spin_lock(lock: &dentry->d_lock); |
1109 | if (!dentry->d_lockref.count) { |
1110 | struct dentry *parent = lock_parent(dentry); |
1111 | if (likely(!dentry->d_lockref.count)) { |
1112 | __dentry_kill(dentry); |
1113 | dput(parent); |
1114 | goto restart; |
1115 | } |
1116 | if (parent) |
1117 | spin_unlock(lock: &parent->d_lock); |
1118 | } |
1119 | spin_unlock(lock: &dentry->d_lock); |
1120 | } |
1121 | spin_unlock(lock: &inode->i_lock); |
1122 | } |
1123 | EXPORT_SYMBOL(d_prune_aliases); |
1124 | |
1125 | /* |
1126 | * Lock a dentry from shrink list. |
1127 | * Called under rcu_read_lock() and dentry->d_lock; the former |
1128 | * guarantees that nothing we access will be freed under us. |
1129 | * Note that dentry is *not* protected from concurrent dentry_kill(), |
1130 | * d_delete(), etc. |
1131 | * |
1132 | * Return false if dentry has been disrupted or grabbed, leaving |
1133 | * the caller to kick it off-list. Otherwise, return true and have |
1134 | * that dentry's inode and parent both locked. |
1135 | */ |
1136 | static bool shrink_lock_dentry(struct dentry *dentry) |
1137 | { |
1138 | struct inode *inode; |
1139 | struct dentry *parent; |
1140 | |
1141 | if (dentry->d_lockref.count) |
1142 | return false; |
1143 | |
1144 | inode = dentry->d_inode; |
1145 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { |
1146 | spin_unlock(lock: &dentry->d_lock); |
1147 | spin_lock(lock: &inode->i_lock); |
1148 | spin_lock(lock: &dentry->d_lock); |
1149 | if (unlikely(dentry->d_lockref.count)) |
1150 | goto out; |
1151 | /* changed inode means that somebody had grabbed it */ |
1152 | if (unlikely(inode != dentry->d_inode)) |
1153 | goto out; |
1154 | } |
1155 | |
1156 | parent = dentry->d_parent; |
1157 | if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) |
1158 | return true; |
1159 | |
1160 | spin_unlock(lock: &dentry->d_lock); |
1161 | spin_lock(lock: &parent->d_lock); |
1162 | if (unlikely(parent != dentry->d_parent)) { |
1163 | spin_unlock(lock: &parent->d_lock); |
1164 | spin_lock(lock: &dentry->d_lock); |
1165 | goto out; |
1166 | } |
1167 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1168 | if (likely(!dentry->d_lockref.count)) |
1169 | return true; |
1170 | spin_unlock(lock: &parent->d_lock); |
1171 | out: |
1172 | if (inode) |
1173 | spin_unlock(lock: &inode->i_lock); |
1174 | return false; |
1175 | } |
1176 | |
1177 | void shrink_dentry_list(struct list_head *list) |
1178 | { |
1179 | while (!list_empty(head: list)) { |
1180 | struct dentry *dentry, *parent; |
1181 | |
1182 | dentry = list_entry(list->prev, struct dentry, d_lru); |
1183 | spin_lock(lock: &dentry->d_lock); |
1184 | rcu_read_lock(); |
1185 | if (!shrink_lock_dentry(dentry)) { |
1186 | bool can_free = false; |
1187 | rcu_read_unlock(); |
1188 | d_shrink_del(dentry); |
1189 | if (dentry->d_lockref.count < 0) |
1190 | can_free = dentry->d_flags & DCACHE_MAY_FREE; |
1191 | spin_unlock(lock: &dentry->d_lock); |
1192 | if (can_free) |
1193 | dentry_free(dentry); |
1194 | continue; |
1195 | } |
1196 | rcu_read_unlock(); |
1197 | d_shrink_del(dentry); |
1198 | parent = dentry->d_parent; |
1199 | if (parent != dentry) |
1200 | __dput_to_list(dentry: parent, list); |
1201 | __dentry_kill(dentry); |
1202 | } |
1203 | } |
1204 | |
1205 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1206 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
1207 | { |
1208 | struct list_head *freeable = arg; |
1209 | struct dentry *dentry = container_of(item, struct dentry, d_lru); |
1210 | |
1211 | |
1212 | /* |
1213 | * we are inverting the lru lock/dentry->d_lock here, |
1214 | * so use a trylock. If we fail to get the lock, just skip |
1215 | * it |
1216 | */ |
1217 | if (!spin_trylock(lock: &dentry->d_lock)) |
1218 | return LRU_SKIP; |
1219 | |
1220 | /* |
1221 | * Referenced dentries are still in use. If they have active |
1222 | * counts, just remove them from the LRU. Otherwise give them |
1223 | * another pass through the LRU. |
1224 | */ |
1225 | if (dentry->d_lockref.count) { |
1226 | d_lru_isolate(lru, dentry); |
1227 | spin_unlock(lock: &dentry->d_lock); |
1228 | return LRU_REMOVED; |
1229 | } |
1230 | |
1231 | if (dentry->d_flags & DCACHE_REFERENCED) { |
1232 | dentry->d_flags &= ~DCACHE_REFERENCED; |
1233 | spin_unlock(lock: &dentry->d_lock); |
1234 | |
1235 | /* |
1236 | * The list move itself will be made by the common LRU code. At |
1237 | * this point, we've dropped the dentry->d_lock but keep the |
1238 | * lru lock. This is safe to do, since every list movement is |
1239 | * protected by the lru lock even if both locks are held. |
1240 | * |
1241 | * This is guaranteed by the fact that all LRU management |
1242 | * functions are intermediated by the LRU API calls like |
1243 | * list_lru_add and list_lru_del. List movement in this file |
1244 | * only ever occur through this functions or through callbacks |
1245 | * like this one, that are called from the LRU API. |
1246 | * |
1247 | * The only exceptions to this are functions like |
1248 | * shrink_dentry_list, and code that first checks for the |
1249 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be |
1250 | * operating only with stack provided lists after they are |
1251 | * properly isolated from the main list. It is thus, always a |
1252 | * local access. |
1253 | */ |
1254 | return LRU_ROTATE; |
1255 | } |
1256 | |
1257 | d_lru_shrink_move(lru, dentry, list: freeable); |
1258 | spin_unlock(lock: &dentry->d_lock); |
1259 | |
1260 | return LRU_REMOVED; |
1261 | } |
1262 | |
1263 | /** |
1264 | * prune_dcache_sb - shrink the dcache |
1265 | * @sb: superblock |
1266 | * @sc: shrink control, passed to list_lru_shrink_walk() |
1267 | * |
1268 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1269 | * is done when we need more memory and called from the superblock shrinker |
1270 | * function. |
1271 | * |
1272 | * This function may fail to free any resources if all the dentries are in |
1273 | * use. |
1274 | */ |
1275 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
1276 | { |
1277 | LIST_HEAD(dispose); |
1278 | long freed; |
1279 | |
1280 | freed = list_lru_shrink_walk(lru: &sb->s_dentry_lru, sc, |
1281 | isolate: dentry_lru_isolate, cb_arg: &dispose); |
1282 | shrink_dentry_list(list: &dispose); |
1283 | return freed; |
1284 | } |
1285 | |
1286 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
1287 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
1288 | { |
1289 | struct list_head *freeable = arg; |
1290 | struct dentry *dentry = container_of(item, struct dentry, d_lru); |
1291 | |
1292 | /* |
1293 | * we are inverting the lru lock/dentry->d_lock here, |
1294 | * so use a trylock. If we fail to get the lock, just skip |
1295 | * it |
1296 | */ |
1297 | if (!spin_trylock(lock: &dentry->d_lock)) |
1298 | return LRU_SKIP; |
1299 | |
1300 | d_lru_shrink_move(lru, dentry, list: freeable); |
1301 | spin_unlock(lock: &dentry->d_lock); |
1302 | |
1303 | return LRU_REMOVED; |
1304 | } |
1305 | |
1306 | |
1307 | /** |
1308 | * shrink_dcache_sb - shrink dcache for a superblock |
1309 | * @sb: superblock |
1310 | * |
1311 | * Shrink the dcache for the specified super block. This is used to free |
1312 | * the dcache before unmounting a file system. |
1313 | */ |
1314 | void shrink_dcache_sb(struct super_block *sb) |
1315 | { |
1316 | do { |
1317 | LIST_HEAD(dispose); |
1318 | |
1319 | list_lru_walk(lru: &sb->s_dentry_lru, |
1320 | isolate: dentry_lru_isolate_shrink, cb_arg: &dispose, nr_to_walk: 1024); |
1321 | shrink_dentry_list(list: &dispose); |
1322 | } while (list_lru_count(lru: &sb->s_dentry_lru) > 0); |
1323 | } |
1324 | EXPORT_SYMBOL(shrink_dcache_sb); |
1325 | |
1326 | /** |
1327 | * enum d_walk_ret - action to talke during tree walk |
1328 | * @D_WALK_CONTINUE: contrinue walk |
1329 | * @D_WALK_QUIT: quit walk |
1330 | * @D_WALK_NORETRY: quit when retry is needed |
1331 | * @D_WALK_SKIP: skip this dentry and its children |
1332 | */ |
1333 | enum d_walk_ret { |
1334 | D_WALK_CONTINUE, |
1335 | D_WALK_QUIT, |
1336 | D_WALK_NORETRY, |
1337 | D_WALK_SKIP, |
1338 | }; |
1339 | |
1340 | /** |
1341 | * d_walk - walk the dentry tree |
1342 | * @parent: start of walk |
1343 | * @data: data passed to @enter() and @finish() |
1344 | * @enter: callback when first entering the dentry |
1345 | * |
1346 | * The @enter() callbacks are called with d_lock held. |
1347 | */ |
1348 | static void d_walk(struct dentry *parent, void *data, |
1349 | enum d_walk_ret (*enter)(void *, struct dentry *)) |
1350 | { |
1351 | struct dentry *this_parent; |
1352 | struct list_head *next; |
1353 | unsigned seq = 0; |
1354 | enum d_walk_ret ret; |
1355 | bool retry = true; |
1356 | |
1357 | again: |
1358 | read_seqbegin_or_lock(lock: &rename_lock, seq: &seq); |
1359 | this_parent = parent; |
1360 | spin_lock(lock: &this_parent->d_lock); |
1361 | |
1362 | ret = enter(data, this_parent); |
1363 | switch (ret) { |
1364 | case D_WALK_CONTINUE: |
1365 | break; |
1366 | case D_WALK_QUIT: |
1367 | case D_WALK_SKIP: |
1368 | goto out_unlock; |
1369 | case D_WALK_NORETRY: |
1370 | retry = false; |
1371 | break; |
1372 | } |
1373 | repeat: |
1374 | next = this_parent->d_subdirs.next; |
1375 | resume: |
1376 | while (next != &this_parent->d_subdirs) { |
1377 | struct list_head *tmp = next; |
1378 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1379 | next = tmp->next; |
1380 | |
1381 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1382 | continue; |
1383 | |
1384 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1385 | |
1386 | ret = enter(data, dentry); |
1387 | switch (ret) { |
1388 | case D_WALK_CONTINUE: |
1389 | break; |
1390 | case D_WALK_QUIT: |
1391 | spin_unlock(lock: &dentry->d_lock); |
1392 | goto out_unlock; |
1393 | case D_WALK_NORETRY: |
1394 | retry = false; |
1395 | break; |
1396 | case D_WALK_SKIP: |
1397 | spin_unlock(lock: &dentry->d_lock); |
1398 | continue; |
1399 | } |
1400 | |
1401 | if (!list_empty(head: &dentry->d_subdirs)) { |
1402 | spin_unlock(lock: &this_parent->d_lock); |
1403 | spin_release(&dentry->d_lock.dep_map, _RET_IP_); |
1404 | this_parent = dentry; |
1405 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1406 | goto repeat; |
1407 | } |
1408 | spin_unlock(lock: &dentry->d_lock); |
1409 | } |
1410 | /* |
1411 | * All done at this level ... ascend and resume the search. |
1412 | */ |
1413 | rcu_read_lock(); |
1414 | ascend: |
1415 | if (this_parent != parent) { |
1416 | struct dentry *child = this_parent; |
1417 | this_parent = child->d_parent; |
1418 | |
1419 | spin_unlock(lock: &child->d_lock); |
1420 | spin_lock(lock: &this_parent->d_lock); |
1421 | |
1422 | /* might go back up the wrong parent if we have had a rename. */ |
1423 | if (need_seqretry(lock: &rename_lock, seq)) |
1424 | goto rename_retry; |
1425 | /* go into the first sibling still alive */ |
1426 | do { |
1427 | next = child->d_child.next; |
1428 | if (next == &this_parent->d_subdirs) |
1429 | goto ascend; |
1430 | child = list_entry(next, struct dentry, d_child); |
1431 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
1432 | rcu_read_unlock(); |
1433 | goto resume; |
1434 | } |
1435 | if (need_seqretry(lock: &rename_lock, seq)) |
1436 | goto rename_retry; |
1437 | rcu_read_unlock(); |
1438 | |
1439 | out_unlock: |
1440 | spin_unlock(lock: &this_parent->d_lock); |
1441 | done_seqretry(lock: &rename_lock, seq); |
1442 | return; |
1443 | |
1444 | rename_retry: |
1445 | spin_unlock(lock: &this_parent->d_lock); |
1446 | rcu_read_unlock(); |
1447 | BUG_ON(seq & 1); |
1448 | if (!retry) |
1449 | return; |
1450 | seq = 1; |
1451 | goto again; |
1452 | } |
1453 | |
1454 | struct check_mount { |
1455 | struct vfsmount *mnt; |
1456 | unsigned int mounted; |
1457 | }; |
1458 | |
1459 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) |
1460 | { |
1461 | struct check_mount *info = data; |
1462 | struct path path = { .mnt = info->mnt, .dentry = dentry }; |
1463 | |
1464 | if (likely(!d_mountpoint(dentry))) |
1465 | return D_WALK_CONTINUE; |
1466 | if (__path_is_mountpoint(path: &path)) { |
1467 | info->mounted = 1; |
1468 | return D_WALK_QUIT; |
1469 | } |
1470 | return D_WALK_CONTINUE; |
1471 | } |
1472 | |
1473 | /** |
1474 | * path_has_submounts - check for mounts over a dentry in the |
1475 | * current namespace. |
1476 | * @parent: path to check. |
1477 | * |
1478 | * Return true if the parent or its subdirectories contain |
1479 | * a mount point in the current namespace. |
1480 | */ |
1481 | int path_has_submounts(const struct path *parent) |
1482 | { |
1483 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; |
1484 | |
1485 | read_seqlock_excl(sl: &mount_lock); |
1486 | d_walk(parent: parent->dentry, data: &data, enter: path_check_mount); |
1487 | read_sequnlock_excl(sl: &mount_lock); |
1488 | |
1489 | return data.mounted; |
1490 | } |
1491 | EXPORT_SYMBOL(path_has_submounts); |
1492 | |
1493 | /* |
1494 | * Called by mount code to set a mountpoint and check if the mountpoint is |
1495 | * reachable (e.g. NFS can unhash a directory dentry and then the complete |
1496 | * subtree can become unreachable). |
1497 | * |
1498 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
1499 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1500 | */ |
1501 | int d_set_mounted(struct dentry *dentry) |
1502 | { |
1503 | struct dentry *p; |
1504 | int ret = -ENOENT; |
1505 | write_seqlock(sl: &rename_lock); |
1506 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { |
1507 | /* Need exclusion wrt. d_invalidate() */ |
1508 | spin_lock(lock: &p->d_lock); |
1509 | if (unlikely(d_unhashed(p))) { |
1510 | spin_unlock(lock: &p->d_lock); |
1511 | goto out; |
1512 | } |
1513 | spin_unlock(lock: &p->d_lock); |
1514 | } |
1515 | spin_lock(lock: &dentry->d_lock); |
1516 | if (!d_unlinked(dentry)) { |
1517 | ret = -EBUSY; |
1518 | if (!d_mountpoint(dentry)) { |
1519 | dentry->d_flags |= DCACHE_MOUNTED; |
1520 | ret = 0; |
1521 | } |
1522 | } |
1523 | spin_unlock(lock: &dentry->d_lock); |
1524 | out: |
1525 | write_sequnlock(sl: &rename_lock); |
1526 | return ret; |
1527 | } |
1528 | |
1529 | /* |
1530 | * Search the dentry child list of the specified parent, |
1531 | * and move any unused dentries to the end of the unused |
1532 | * list for prune_dcache(). We descend to the next level |
1533 | * whenever the d_subdirs list is non-empty and continue |
1534 | * searching. |
1535 | * |
1536 | * It returns zero iff there are no unused children, |
1537 | * otherwise it returns the number of children moved to |
1538 | * the end of the unused list. This may not be the total |
1539 | * number of unused children, because select_parent can |
1540 | * drop the lock and return early due to latency |
1541 | * constraints. |
1542 | */ |
1543 | |
1544 | struct select_data { |
1545 | struct dentry *start; |
1546 | union { |
1547 | long found; |
1548 | struct dentry *victim; |
1549 | }; |
1550 | struct list_head dispose; |
1551 | }; |
1552 | |
1553 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1554 | { |
1555 | struct select_data *data = _data; |
1556 | enum d_walk_ret ret = D_WALK_CONTINUE; |
1557 | |
1558 | if (data->start == dentry) |
1559 | goto out; |
1560 | |
1561 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
1562 | data->found++; |
1563 | } else { |
1564 | if (dentry->d_flags & DCACHE_LRU_LIST) |
1565 | d_lru_del(dentry); |
1566 | if (!dentry->d_lockref.count) { |
1567 | d_shrink_add(dentry, list: &data->dispose); |
1568 | data->found++; |
1569 | } |
1570 | } |
1571 | /* |
1572 | * We can return to the caller if we have found some (this |
1573 | * ensures forward progress). We'll be coming back to find |
1574 | * the rest. |
1575 | */ |
1576 | if (!list_empty(head: &data->dispose)) |
1577 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; |
1578 | out: |
1579 | return ret; |
1580 | } |
1581 | |
1582 | static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry) |
1583 | { |
1584 | struct select_data *data = _data; |
1585 | enum d_walk_ret ret = D_WALK_CONTINUE; |
1586 | |
1587 | if (data->start == dentry) |
1588 | goto out; |
1589 | |
1590 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
1591 | if (!dentry->d_lockref.count) { |
1592 | rcu_read_lock(); |
1593 | data->victim = dentry; |
1594 | return D_WALK_QUIT; |
1595 | } |
1596 | } else { |
1597 | if (dentry->d_flags & DCACHE_LRU_LIST) |
1598 | d_lru_del(dentry); |
1599 | if (!dentry->d_lockref.count) |
1600 | d_shrink_add(dentry, list: &data->dispose); |
1601 | } |
1602 | /* |
1603 | * We can return to the caller if we have found some (this |
1604 | * ensures forward progress). We'll be coming back to find |
1605 | * the rest. |
1606 | */ |
1607 | if (!list_empty(head: &data->dispose)) |
1608 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; |
1609 | out: |
1610 | return ret; |
1611 | } |
1612 | |
1613 | /** |
1614 | * shrink_dcache_parent - prune dcache |
1615 | * @parent: parent of entries to prune |
1616 | * |
1617 | * Prune the dcache to remove unused children of the parent dentry. |
1618 | */ |
1619 | void shrink_dcache_parent(struct dentry *parent) |
1620 | { |
1621 | for (;;) { |
1622 | struct select_data data = {.start = parent}; |
1623 | |
1624 | INIT_LIST_HEAD(list: &data.dispose); |
1625 | d_walk(parent, data: &data, enter: select_collect); |
1626 | |
1627 | if (!list_empty(head: &data.dispose)) { |
1628 | shrink_dentry_list(list: &data.dispose); |
1629 | continue; |
1630 | } |
1631 | |
1632 | cond_resched(); |
1633 | if (!data.found) |
1634 | break; |
1635 | data.victim = NULL; |
1636 | d_walk(parent, data: &data, enter: select_collect2); |
1637 | if (data.victim) { |
1638 | struct dentry *parent; |
1639 | spin_lock(lock: &data.victim->d_lock); |
1640 | if (!shrink_lock_dentry(dentry: data.victim)) { |
1641 | spin_unlock(lock: &data.victim->d_lock); |
1642 | rcu_read_unlock(); |
1643 | } else { |
1644 | rcu_read_unlock(); |
1645 | parent = data.victim->d_parent; |
1646 | if (parent != data.victim) |
1647 | __dput_to_list(dentry: parent, list: &data.dispose); |
1648 | __dentry_kill(dentry: data.victim); |
1649 | } |
1650 | } |
1651 | if (!list_empty(head: &data.dispose)) |
1652 | shrink_dentry_list(list: &data.dispose); |
1653 | } |
1654 | } |
1655 | EXPORT_SYMBOL(shrink_dcache_parent); |
1656 | |
1657 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
1658 | { |
1659 | /* it has busy descendents; complain about those instead */ |
1660 | if (!list_empty(head: &dentry->d_subdirs)) |
1661 | return D_WALK_CONTINUE; |
1662 | |
1663 | /* root with refcount 1 is fine */ |
1664 | if (dentry == _data && dentry->d_lockref.count == 1) |
1665 | return D_WALK_CONTINUE; |
1666 | |
1667 | WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} " |
1668 | " still in use (%d) [unmount of %s %s]\n" , |
1669 | dentry, |
1670 | dentry->d_inode ? |
1671 | dentry->d_inode->i_ino : 0UL, |
1672 | dentry, |
1673 | dentry->d_lockref.count, |
1674 | dentry->d_sb->s_type->name, |
1675 | dentry->d_sb->s_id); |
1676 | return D_WALK_CONTINUE; |
1677 | } |
1678 | |
1679 | static void do_one_tree(struct dentry *dentry) |
1680 | { |
1681 | shrink_dcache_parent(dentry); |
1682 | d_walk(parent: dentry, data: dentry, enter: umount_check); |
1683 | d_drop(dentry); |
1684 | dput(dentry); |
1685 | } |
1686 | |
1687 | /* |
1688 | * destroy the dentries attached to a superblock on unmounting |
1689 | */ |
1690 | void shrink_dcache_for_umount(struct super_block *sb) |
1691 | { |
1692 | struct dentry *dentry; |
1693 | |
1694 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked" ); |
1695 | |
1696 | dentry = sb->s_root; |
1697 | sb->s_root = NULL; |
1698 | do_one_tree(dentry); |
1699 | |
1700 | while (!hlist_bl_empty(h: &sb->s_roots)) { |
1701 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); |
1702 | do_one_tree(dentry); |
1703 | } |
1704 | } |
1705 | |
1706 | static enum d_walk_ret find_submount(void *_data, struct dentry *dentry) |
1707 | { |
1708 | struct dentry **victim = _data; |
1709 | if (d_mountpoint(dentry)) { |
1710 | __dget_dlock(dentry); |
1711 | *victim = dentry; |
1712 | return D_WALK_QUIT; |
1713 | } |
1714 | return D_WALK_CONTINUE; |
1715 | } |
1716 | |
1717 | /** |
1718 | * d_invalidate - detach submounts, prune dcache, and drop |
1719 | * @dentry: dentry to invalidate (aka detach, prune and drop) |
1720 | */ |
1721 | void d_invalidate(struct dentry *dentry) |
1722 | { |
1723 | bool had_submounts = false; |
1724 | spin_lock(lock: &dentry->d_lock); |
1725 | if (d_unhashed(dentry)) { |
1726 | spin_unlock(lock: &dentry->d_lock); |
1727 | return; |
1728 | } |
1729 | __d_drop(dentry); |
1730 | spin_unlock(lock: &dentry->d_lock); |
1731 | |
1732 | /* Negative dentries can be dropped without further checks */ |
1733 | if (!dentry->d_inode) |
1734 | return; |
1735 | |
1736 | shrink_dcache_parent(dentry); |
1737 | for (;;) { |
1738 | struct dentry *victim = NULL; |
1739 | d_walk(parent: dentry, data: &victim, enter: find_submount); |
1740 | if (!victim) { |
1741 | if (had_submounts) |
1742 | shrink_dcache_parent(dentry); |
1743 | return; |
1744 | } |
1745 | had_submounts = true; |
1746 | detach_mounts(dentry: victim); |
1747 | dput(victim); |
1748 | } |
1749 | } |
1750 | EXPORT_SYMBOL(d_invalidate); |
1751 | |
1752 | /** |
1753 | * __d_alloc - allocate a dcache entry |
1754 | * @sb: filesystem it will belong to |
1755 | * @name: qstr of the name |
1756 | * |
1757 | * Allocates a dentry. It returns %NULL if there is insufficient memory |
1758 | * available. On a success the dentry is returned. The name passed in is |
1759 | * copied and the copy passed in may be reused after this call. |
1760 | */ |
1761 | |
1762 | static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1763 | { |
1764 | struct dentry *dentry; |
1765 | char *dname; |
1766 | int err; |
1767 | |
1768 | dentry = kmem_cache_alloc_lru(s: dentry_cache, lru: &sb->s_dentry_lru, |
1769 | GFP_KERNEL); |
1770 | if (!dentry) |
1771 | return NULL; |
1772 | |
1773 | /* |
1774 | * We guarantee that the inline name is always NUL-terminated. |
1775 | * This way the memcpy() done by the name switching in rename |
1776 | * will still always have a NUL at the end, even if we might |
1777 | * be overwriting an internal NUL character |
1778 | */ |
1779 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; |
1780 | if (unlikely(!name)) { |
1781 | name = &slash_name; |
1782 | dname = dentry->d_iname; |
1783 | } else if (name->len > DNAME_INLINE_LEN-1) { |
1784 | size_t size = offsetof(struct external_name, name[1]); |
1785 | struct external_name *p = kmalloc(size: size + name->len, |
1786 | GFP_KERNEL_ACCOUNT | |
1787 | __GFP_RECLAIMABLE); |
1788 | if (!p) { |
1789 | kmem_cache_free(s: dentry_cache, objp: dentry); |
1790 | return NULL; |
1791 | } |
1792 | atomic_set(v: &p->u.count, i: 1); |
1793 | dname = p->name; |
1794 | } else { |
1795 | dname = dentry->d_iname; |
1796 | } |
1797 | |
1798 | dentry->d_name.len = name->len; |
1799 | dentry->d_name.hash = name->hash; |
1800 | memcpy(dname, name->name, name->len); |
1801 | dname[name->len] = 0; |
1802 | |
1803 | /* Make sure we always see the terminating NUL character */ |
1804 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ |
1805 | |
1806 | dentry->d_lockref.count = 1; |
1807 | dentry->d_flags = 0; |
1808 | spin_lock_init(&dentry->d_lock); |
1809 | seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock); |
1810 | dentry->d_inode = NULL; |
1811 | dentry->d_parent = dentry; |
1812 | dentry->d_sb = sb; |
1813 | dentry->d_op = NULL; |
1814 | dentry->d_fsdata = NULL; |
1815 | INIT_HLIST_BL_NODE(h: &dentry->d_hash); |
1816 | INIT_LIST_HEAD(list: &dentry->d_lru); |
1817 | INIT_LIST_HEAD(list: &dentry->d_subdirs); |
1818 | INIT_HLIST_NODE(h: &dentry->d_u.d_alias); |
1819 | INIT_LIST_HEAD(list: &dentry->d_child); |
1820 | d_set_d_op(dentry, op: dentry->d_sb->s_d_op); |
1821 | |
1822 | if (dentry->d_op && dentry->d_op->d_init) { |
1823 | err = dentry->d_op->d_init(dentry); |
1824 | if (err) { |
1825 | if (dname_external(dentry)) |
1826 | kfree(objp: external_name(dentry)); |
1827 | kmem_cache_free(s: dentry_cache, objp: dentry); |
1828 | return NULL; |
1829 | } |
1830 | } |
1831 | |
1832 | this_cpu_inc(nr_dentry); |
1833 | |
1834 | return dentry; |
1835 | } |
1836 | |
1837 | /** |
1838 | * d_alloc - allocate a dcache entry |
1839 | * @parent: parent of entry to allocate |
1840 | * @name: qstr of the name |
1841 | * |
1842 | * Allocates a dentry. It returns %NULL if there is insufficient memory |
1843 | * available. On a success the dentry is returned. The name passed in is |
1844 | * copied and the copy passed in may be reused after this call. |
1845 | */ |
1846 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) |
1847 | { |
1848 | struct dentry *dentry = __d_alloc(sb: parent->d_sb, name); |
1849 | if (!dentry) |
1850 | return NULL; |
1851 | spin_lock(lock: &parent->d_lock); |
1852 | /* |
1853 | * don't need child lock because it is not subject |
1854 | * to concurrency here |
1855 | */ |
1856 | __dget_dlock(dentry: parent); |
1857 | dentry->d_parent = parent; |
1858 | list_add(new: &dentry->d_child, head: &parent->d_subdirs); |
1859 | spin_unlock(lock: &parent->d_lock); |
1860 | |
1861 | return dentry; |
1862 | } |
1863 | EXPORT_SYMBOL(d_alloc); |
1864 | |
1865 | struct dentry *d_alloc_anon(struct super_block *sb) |
1866 | { |
1867 | return __d_alloc(sb, NULL); |
1868 | } |
1869 | EXPORT_SYMBOL(d_alloc_anon); |
1870 | |
1871 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1872 | { |
1873 | struct dentry *dentry = d_alloc_anon(parent->d_sb); |
1874 | if (dentry) { |
1875 | dentry->d_flags |= DCACHE_DENTRY_CURSOR; |
1876 | dentry->d_parent = dget(dentry: parent); |
1877 | } |
1878 | return dentry; |
1879 | } |
1880 | |
1881 | /** |
1882 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) |
1883 | * @sb: the superblock |
1884 | * @name: qstr of the name |
1885 | * |
1886 | * For a filesystem that just pins its dentries in memory and never |
1887 | * performs lookups at all, return an unhashed IS_ROOT dentry. |
1888 | * This is used for pipes, sockets et.al. - the stuff that should |
1889 | * never be anyone's children or parents. Unlike all other |
1890 | * dentries, these will not have RCU delay between dropping the |
1891 | * last reference and freeing them. |
1892 | * |
1893 | * The only user is alloc_file_pseudo() and that's what should |
1894 | * be considered a public interface. Don't use directly. |
1895 | */ |
1896 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1897 | { |
1898 | struct dentry *dentry = __d_alloc(sb, name); |
1899 | if (likely(dentry)) |
1900 | dentry->d_flags |= DCACHE_NORCU; |
1901 | return dentry; |
1902 | } |
1903 | |
1904 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1905 | { |
1906 | struct qstr q; |
1907 | |
1908 | q.name = name; |
1909 | q.hash_len = hashlen_string(salt: parent, name); |
1910 | return d_alloc(parent, &q); |
1911 | } |
1912 | EXPORT_SYMBOL(d_alloc_name); |
1913 | |
1914 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1915 | { |
1916 | WARN_ON_ONCE(dentry->d_op); |
1917 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | |
1918 | DCACHE_OP_COMPARE | |
1919 | DCACHE_OP_REVALIDATE | |
1920 | DCACHE_OP_WEAK_REVALIDATE | |
1921 | DCACHE_OP_DELETE | |
1922 | DCACHE_OP_REAL)); |
1923 | dentry->d_op = op; |
1924 | if (!op) |
1925 | return; |
1926 | if (op->d_hash) |
1927 | dentry->d_flags |= DCACHE_OP_HASH; |
1928 | if (op->d_compare) |
1929 | dentry->d_flags |= DCACHE_OP_COMPARE; |
1930 | if (op->d_revalidate) |
1931 | dentry->d_flags |= DCACHE_OP_REVALIDATE; |
1932 | if (op->d_weak_revalidate) |
1933 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; |
1934 | if (op->d_delete) |
1935 | dentry->d_flags |= DCACHE_OP_DELETE; |
1936 | if (op->d_prune) |
1937 | dentry->d_flags |= DCACHE_OP_PRUNE; |
1938 | if (op->d_real) |
1939 | dentry->d_flags |= DCACHE_OP_REAL; |
1940 | |
1941 | } |
1942 | EXPORT_SYMBOL(d_set_d_op); |
1943 | |
1944 | |
1945 | /* |
1946 | * d_set_fallthru - Mark a dentry as falling through to a lower layer |
1947 | * @dentry - The dentry to mark |
1948 | * |
1949 | * Mark a dentry as falling through to the lower layer (as set with |
1950 | * d_pin_lower()). This flag may be recorded on the medium. |
1951 | */ |
1952 | void d_set_fallthru(struct dentry *dentry) |
1953 | { |
1954 | spin_lock(lock: &dentry->d_lock); |
1955 | dentry->d_flags |= DCACHE_FALLTHRU; |
1956 | spin_unlock(lock: &dentry->d_lock); |
1957 | } |
1958 | EXPORT_SYMBOL(d_set_fallthru); |
1959 | |
1960 | static unsigned d_flags_for_inode(struct inode *inode) |
1961 | { |
1962 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
1963 | |
1964 | if (!inode) |
1965 | return DCACHE_MISS_TYPE; |
1966 | |
1967 | if (S_ISDIR(inode->i_mode)) { |
1968 | add_flags = DCACHE_DIRECTORY_TYPE; |
1969 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { |
1970 | if (unlikely(!inode->i_op->lookup)) |
1971 | add_flags = DCACHE_AUTODIR_TYPE; |
1972 | else |
1973 | inode->i_opflags |= IOP_LOOKUP; |
1974 | } |
1975 | goto type_determined; |
1976 | } |
1977 | |
1978 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { |
1979 | if (unlikely(inode->i_op->get_link)) { |
1980 | add_flags = DCACHE_SYMLINK_TYPE; |
1981 | goto type_determined; |
1982 | } |
1983 | inode->i_opflags |= IOP_NOFOLLOW; |
1984 | } |
1985 | |
1986 | if (unlikely(!S_ISREG(inode->i_mode))) |
1987 | add_flags = DCACHE_SPECIAL_TYPE; |
1988 | |
1989 | type_determined: |
1990 | if (unlikely(IS_AUTOMOUNT(inode))) |
1991 | add_flags |= DCACHE_NEED_AUTOMOUNT; |
1992 | return add_flags; |
1993 | } |
1994 | |
1995 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1996 | { |
1997 | unsigned add_flags = d_flags_for_inode(inode); |
1998 | WARN_ON(d_in_lookup(dentry)); |
1999 | |
2000 | spin_lock(lock: &dentry->d_lock); |
2001 | /* |
2002 | * Decrement negative dentry count if it was in the LRU list. |
2003 | */ |
2004 | if (dentry->d_flags & DCACHE_LRU_LIST) |
2005 | this_cpu_dec(nr_dentry_negative); |
2006 | hlist_add_head(n: &dentry->d_u.d_alias, h: &inode->i_dentry); |
2007 | raw_write_seqcount_begin(&dentry->d_seq); |
2008 | __d_set_inode_and_type(dentry, inode, type_flags: add_flags); |
2009 | raw_write_seqcount_end(&dentry->d_seq); |
2010 | fsnotify_update_flags(dentry); |
2011 | spin_unlock(lock: &dentry->d_lock); |
2012 | } |
2013 | |
2014 | /** |
2015 | * d_instantiate - fill in inode information for a dentry |
2016 | * @entry: dentry to complete |
2017 | * @inode: inode to attach to this dentry |
2018 | * |
2019 | * Fill in inode information in the entry. |
2020 | * |
2021 | * This turns negative dentries into productive full members |
2022 | * of society. |
2023 | * |
2024 | * NOTE! This assumes that the inode count has been incremented |
2025 | * (or otherwise set) by the caller to indicate that it is now |
2026 | * in use by the dcache. |
2027 | */ |
2028 | |
2029 | void d_instantiate(struct dentry *entry, struct inode * inode) |
2030 | { |
2031 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
2032 | if (inode) { |
2033 | security_d_instantiate(dentry: entry, inode); |
2034 | spin_lock(lock: &inode->i_lock); |
2035 | __d_instantiate(dentry: entry, inode); |
2036 | spin_unlock(lock: &inode->i_lock); |
2037 | } |
2038 | } |
2039 | EXPORT_SYMBOL(d_instantiate); |
2040 | |
2041 | /* |
2042 | * This should be equivalent to d_instantiate() + unlock_new_inode(), |
2043 | * with lockdep-related part of unlock_new_inode() done before |
2044 | * anything else. Use that instead of open-coding d_instantiate()/ |
2045 | * unlock_new_inode() combinations. |
2046 | */ |
2047 | void d_instantiate_new(struct dentry *entry, struct inode *inode) |
2048 | { |
2049 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
2050 | BUG_ON(!inode); |
2051 | lockdep_annotate_inode_mutex_key(inode); |
2052 | security_d_instantiate(dentry: entry, inode); |
2053 | spin_lock(lock: &inode->i_lock); |
2054 | __d_instantiate(dentry: entry, inode); |
2055 | WARN_ON(!(inode->i_state & I_NEW)); |
2056 | inode->i_state &= ~I_NEW & ~I_CREATING; |
2057 | smp_mb(); |
2058 | wake_up_bit(word: &inode->i_state, __I_NEW); |
2059 | spin_unlock(lock: &inode->i_lock); |
2060 | } |
2061 | EXPORT_SYMBOL(d_instantiate_new); |
2062 | |
2063 | struct dentry *d_make_root(struct inode *root_inode) |
2064 | { |
2065 | struct dentry *res = NULL; |
2066 | |
2067 | if (root_inode) { |
2068 | res = d_alloc_anon(root_inode->i_sb); |
2069 | if (res) |
2070 | d_instantiate(res, root_inode); |
2071 | else |
2072 | iput(root_inode); |
2073 | } |
2074 | return res; |
2075 | } |
2076 | EXPORT_SYMBOL(d_make_root); |
2077 | |
2078 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, |
2079 | struct inode *inode, |
2080 | bool disconnected) |
2081 | { |
2082 | struct dentry *res; |
2083 | unsigned add_flags; |
2084 | |
2085 | security_d_instantiate(dentry, inode); |
2086 | spin_lock(lock: &inode->i_lock); |
2087 | res = __d_find_any_alias(inode); |
2088 | if (res) { |
2089 | spin_unlock(lock: &inode->i_lock); |
2090 | dput(dentry); |
2091 | goto out_iput; |
2092 | } |
2093 | |
2094 | /* attach a disconnected dentry */ |
2095 | add_flags = d_flags_for_inode(inode); |
2096 | |
2097 | if (disconnected) |
2098 | add_flags |= DCACHE_DISCONNECTED; |
2099 | |
2100 | spin_lock(lock: &dentry->d_lock); |
2101 | __d_set_inode_and_type(dentry, inode, type_flags: add_flags); |
2102 | hlist_add_head(n: &dentry->d_u.d_alias, h: &inode->i_dentry); |
2103 | if (!disconnected) { |
2104 | hlist_bl_lock(b: &dentry->d_sb->s_roots); |
2105 | hlist_bl_add_head(n: &dentry->d_hash, h: &dentry->d_sb->s_roots); |
2106 | hlist_bl_unlock(b: &dentry->d_sb->s_roots); |
2107 | } |
2108 | spin_unlock(lock: &dentry->d_lock); |
2109 | spin_unlock(lock: &inode->i_lock); |
2110 | |
2111 | return dentry; |
2112 | |
2113 | out_iput: |
2114 | iput(inode); |
2115 | return res; |
2116 | } |
2117 | |
2118 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) |
2119 | { |
2120 | return __d_instantiate_anon(dentry, inode, disconnected: true); |
2121 | } |
2122 | EXPORT_SYMBOL(d_instantiate_anon); |
2123 | |
2124 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) |
2125 | { |
2126 | struct dentry *tmp; |
2127 | struct dentry *res; |
2128 | |
2129 | if (!inode) |
2130 | return ERR_PTR(error: -ESTALE); |
2131 | if (IS_ERR(ptr: inode)) |
2132 | return ERR_CAST(ptr: inode); |
2133 | |
2134 | res = d_find_any_alias(inode); |
2135 | if (res) |
2136 | goto out_iput; |
2137 | |
2138 | tmp = d_alloc_anon(inode->i_sb); |
2139 | if (!tmp) { |
2140 | res = ERR_PTR(error: -ENOMEM); |
2141 | goto out_iput; |
2142 | } |
2143 | |
2144 | return __d_instantiate_anon(dentry: tmp, inode, disconnected); |
2145 | |
2146 | out_iput: |
2147 | iput(inode); |
2148 | return res; |
2149 | } |
2150 | |
2151 | /** |
2152 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode |
2153 | * @inode: inode to allocate the dentry for |
2154 | * |
2155 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or |
2156 | * similar open by handle operations. The returned dentry may be anonymous, |
2157 | * or may have a full name (if the inode was already in the cache). |
2158 | * |
2159 | * When called on a directory inode, we must ensure that the inode only ever |
2160 | * has one dentry. If a dentry is found, that is returned instead of |
2161 | * allocating a new one. |
2162 | * |
2163 | * On successful return, the reference to the inode has been transferred |
2164 | * to the dentry. In case of an error the reference on the inode is released. |
2165 | * To make it easier to use in export operations a %NULL or IS_ERR inode may |
2166 | * be passed in and the error will be propagated to the return value, |
2167 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). |
2168 | */ |
2169 | struct dentry *d_obtain_alias(struct inode *inode) |
2170 | { |
2171 | return __d_obtain_alias(inode, disconnected: true); |
2172 | } |
2173 | EXPORT_SYMBOL(d_obtain_alias); |
2174 | |
2175 | /** |
2176 | * d_obtain_root - find or allocate a dentry for a given inode |
2177 | * @inode: inode to allocate the dentry for |
2178 | * |
2179 | * Obtain an IS_ROOT dentry for the root of a filesystem. |
2180 | * |
2181 | * We must ensure that directory inodes only ever have one dentry. If a |
2182 | * dentry is found, that is returned instead of allocating a new one. |
2183 | * |
2184 | * On successful return, the reference to the inode has been transferred |
2185 | * to the dentry. In case of an error the reference on the inode is |
2186 | * released. A %NULL or IS_ERR inode may be passed in and will be the |
2187 | * error will be propagate to the return value, with a %NULL @inode |
2188 | * replaced by ERR_PTR(-ESTALE). |
2189 | */ |
2190 | struct dentry *d_obtain_root(struct inode *inode) |
2191 | { |
2192 | return __d_obtain_alias(inode, disconnected: false); |
2193 | } |
2194 | EXPORT_SYMBOL(d_obtain_root); |
2195 | |
2196 | /** |
2197 | * d_add_ci - lookup or allocate new dentry with case-exact name |
2198 | * @inode: the inode case-insensitive lookup has found |
2199 | * @dentry: the negative dentry that was passed to the parent's lookup func |
2200 | * @name: the case-exact name to be associated with the returned dentry |
2201 | * |
2202 | * This is to avoid filling the dcache with case-insensitive names to the |
2203 | * same inode, only the actual correct case is stored in the dcache for |
2204 | * case-insensitive filesystems. |
2205 | * |
2206 | * For a case-insensitive lookup match and if the case-exact dentry |
2207 | * already exists in the dcache, use it and return it. |
2208 | * |
2209 | * If no entry exists with the exact case name, allocate new dentry with |
2210 | * the exact case, and return the spliced entry. |
2211 | */ |
2212 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
2213 | struct qstr *name) |
2214 | { |
2215 | struct dentry *found, *res; |
2216 | |
2217 | /* |
2218 | * First check if a dentry matching the name already exists, |
2219 | * if not go ahead and create it now. |
2220 | */ |
2221 | found = d_hash_and_lookup(dentry->d_parent, name); |
2222 | if (found) { |
2223 | iput(inode); |
2224 | return found; |
2225 | } |
2226 | if (d_in_lookup(dentry)) { |
2227 | found = d_alloc_parallel(dentry->d_parent, name, |
2228 | dentry->d_wait); |
2229 | if (IS_ERR(ptr: found) || !d_in_lookup(dentry: found)) { |
2230 | iput(inode); |
2231 | return found; |
2232 | } |
2233 | } else { |
2234 | found = d_alloc(dentry->d_parent, name); |
2235 | if (!found) { |
2236 | iput(inode); |
2237 | return ERR_PTR(error: -ENOMEM); |
2238 | } |
2239 | } |
2240 | res = d_splice_alias(inode, found); |
2241 | if (res) { |
2242 | d_lookup_done(dentry: found); |
2243 | dput(found); |
2244 | return res; |
2245 | } |
2246 | return found; |
2247 | } |
2248 | EXPORT_SYMBOL(d_add_ci); |
2249 | |
2250 | /** |
2251 | * d_same_name - compare dentry name with case-exact name |
2252 | * @parent: parent dentry |
2253 | * @dentry: the negative dentry that was passed to the parent's lookup func |
2254 | * @name: the case-exact name to be associated with the returned dentry |
2255 | * |
2256 | * Return: true if names are same, or false |
2257 | */ |
2258 | bool d_same_name(const struct dentry *dentry, const struct dentry *parent, |
2259 | const struct qstr *name) |
2260 | { |
2261 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2262 | if (dentry->d_name.len != name->len) |
2263 | return false; |
2264 | return dentry_cmp(dentry, ct: name->name, tcount: name->len) == 0; |
2265 | } |
2266 | return parent->d_op->d_compare(dentry, |
2267 | dentry->d_name.len, dentry->d_name.name, |
2268 | name) == 0; |
2269 | } |
2270 | EXPORT_SYMBOL_GPL(d_same_name); |
2271 | |
2272 | /* |
2273 | * This is __d_lookup_rcu() when the parent dentry has |
2274 | * DCACHE_OP_COMPARE, which makes things much nastier. |
2275 | */ |
2276 | static noinline struct dentry *__d_lookup_rcu_op_compare( |
2277 | const struct dentry *parent, |
2278 | const struct qstr *name, |
2279 | unsigned *seqp) |
2280 | { |
2281 | u64 hashlen = name->hash_len; |
2282 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
2283 | struct hlist_bl_node *node; |
2284 | struct dentry *dentry; |
2285 | |
2286 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
2287 | int tlen; |
2288 | const char *tname; |
2289 | unsigned seq; |
2290 | |
2291 | seqretry: |
2292 | seq = raw_seqcount_begin(&dentry->d_seq); |
2293 | if (dentry->d_parent != parent) |
2294 | continue; |
2295 | if (d_unhashed(dentry)) |
2296 | continue; |
2297 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
2298 | continue; |
2299 | tlen = dentry->d_name.len; |
2300 | tname = dentry->d_name.name; |
2301 | /* we want a consistent (name,len) pair */ |
2302 | if (read_seqcount_retry(&dentry->d_seq, seq)) { |
2303 | cpu_relax(); |
2304 | goto seqretry; |
2305 | } |
2306 | if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0) |
2307 | continue; |
2308 | *seqp = seq; |
2309 | return dentry; |
2310 | } |
2311 | return NULL; |
2312 | } |
2313 | |
2314 | /** |
2315 | * __d_lookup_rcu - search for a dentry (racy, store-free) |
2316 | * @parent: parent dentry |
2317 | * @name: qstr of name we wish to find |
2318 | * @seqp: returns d_seq value at the point where the dentry was found |
2319 | * Returns: dentry, or NULL |
2320 | * |
2321 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name |
2322 | * resolution (store-free path walking) design described in |
2323 | * Documentation/filesystems/path-lookup.txt. |
2324 | * |
2325 | * This is not to be used outside core vfs. |
2326 | * |
2327 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock |
2328 | * held, and rcu_read_lock held. The returned dentry must not be stored into |
2329 | * without taking d_lock and checking d_seq sequence count against @seq |
2330 | * returned here. |
2331 | * |
2332 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
2333 | * function. |
2334 | * |
2335 | * Alternatively, __d_lookup_rcu may be called again to look up the child of |
2336 | * the returned dentry, so long as its parent's seqlock is checked after the |
2337 | * child is looked up. Thus, an interlocking stepping of sequence lock checks |
2338 | * is formed, giving integrity down the path walk. |
2339 | * |
2340 | * NOTE! The caller *has* to check the resulting dentry against the sequence |
2341 | * number we've returned before using any of the resulting dentry state! |
2342 | */ |
2343 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2344 | const struct qstr *name, |
2345 | unsigned *seqp) |
2346 | { |
2347 | u64 hashlen = name->hash_len; |
2348 | const unsigned char *str = name->name; |
2349 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
2350 | struct hlist_bl_node *node; |
2351 | struct dentry *dentry; |
2352 | |
2353 | /* |
2354 | * Note: There is significant duplication with __d_lookup_rcu which is |
2355 | * required to prevent single threaded performance regressions |
2356 | * especially on architectures where smp_rmb (in seqcounts) are costly. |
2357 | * Keep the two functions in sync. |
2358 | */ |
2359 | |
2360 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) |
2361 | return __d_lookup_rcu_op_compare(parent, name, seqp); |
2362 | |
2363 | /* |
2364 | * The hash list is protected using RCU. |
2365 | * |
2366 | * Carefully use d_seq when comparing a candidate dentry, to avoid |
2367 | * races with d_move(). |
2368 | * |
2369 | * It is possible that concurrent renames can mess up our list |
2370 | * walk here and result in missing our dentry, resulting in the |
2371 | * false-negative result. d_lookup() protects against concurrent |
2372 | * renames using rename_lock seqlock. |
2373 | * |
2374 | * See Documentation/filesystems/path-lookup.txt for more details. |
2375 | */ |
2376 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
2377 | unsigned seq; |
2378 | |
2379 | /* |
2380 | * The dentry sequence count protects us from concurrent |
2381 | * renames, and thus protects parent and name fields. |
2382 | * |
2383 | * The caller must perform a seqcount check in order |
2384 | * to do anything useful with the returned dentry. |
2385 | * |
2386 | * NOTE! We do a "raw" seqcount_begin here. That means that |
2387 | * we don't wait for the sequence count to stabilize if it |
2388 | * is in the middle of a sequence change. If we do the slow |
2389 | * dentry compare, we will do seqretries until it is stable, |
2390 | * and if we end up with a successful lookup, we actually |
2391 | * want to exit RCU lookup anyway. |
2392 | * |
2393 | * Note that raw_seqcount_begin still *does* smp_rmb(), so |
2394 | * we are still guaranteed NUL-termination of ->d_name.name. |
2395 | */ |
2396 | seq = raw_seqcount_begin(&dentry->d_seq); |
2397 | if (dentry->d_parent != parent) |
2398 | continue; |
2399 | if (d_unhashed(dentry)) |
2400 | continue; |
2401 | if (dentry->d_name.hash_len != hashlen) |
2402 | continue; |
2403 | if (dentry_cmp(dentry, ct: str, hashlen_len(hashlen)) != 0) |
2404 | continue; |
2405 | *seqp = seq; |
2406 | return dentry; |
2407 | } |
2408 | return NULL; |
2409 | } |
2410 | |
2411 | /** |
2412 | * d_lookup - search for a dentry |
2413 | * @parent: parent dentry |
2414 | * @name: qstr of name we wish to find |
2415 | * Returns: dentry, or NULL |
2416 | * |
2417 | * d_lookup searches the children of the parent dentry for the name in |
2418 | * question. If the dentry is found its reference count is incremented and the |
2419 | * dentry is returned. The caller must use dput to free the entry when it has |
2420 | * finished using it. %NULL is returned if the dentry does not exist. |
2421 | */ |
2422 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
2423 | { |
2424 | struct dentry *dentry; |
2425 | unsigned seq; |
2426 | |
2427 | do { |
2428 | seq = read_seqbegin(sl: &rename_lock); |
2429 | dentry = __d_lookup(parent, name); |
2430 | if (dentry) |
2431 | break; |
2432 | } while (read_seqretry(sl: &rename_lock, start: seq)); |
2433 | return dentry; |
2434 | } |
2435 | EXPORT_SYMBOL(d_lookup); |
2436 | |
2437 | /** |
2438 | * __d_lookup - search for a dentry (racy) |
2439 | * @parent: parent dentry |
2440 | * @name: qstr of name we wish to find |
2441 | * Returns: dentry, or NULL |
2442 | * |
2443 | * __d_lookup is like d_lookup, however it may (rarely) return a |
2444 | * false-negative result due to unrelated rename activity. |
2445 | * |
2446 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, |
2447 | * however it must be used carefully, eg. with a following d_lookup in |
2448 | * the case of failure. |
2449 | * |
2450 | * __d_lookup callers must be commented. |
2451 | */ |
2452 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
2453 | { |
2454 | unsigned int hash = name->hash; |
2455 | struct hlist_bl_head *b = d_hash(hash); |
2456 | struct hlist_bl_node *node; |
2457 | struct dentry *found = NULL; |
2458 | struct dentry *dentry; |
2459 | |
2460 | /* |
2461 | * Note: There is significant duplication with __d_lookup_rcu which is |
2462 | * required to prevent single threaded performance regressions |
2463 | * especially on architectures where smp_rmb (in seqcounts) are costly. |
2464 | * Keep the two functions in sync. |
2465 | */ |
2466 | |
2467 | /* |
2468 | * The hash list is protected using RCU. |
2469 | * |
2470 | * Take d_lock when comparing a candidate dentry, to avoid races |
2471 | * with d_move(). |
2472 | * |
2473 | * It is possible that concurrent renames can mess up our list |
2474 | * walk here and result in missing our dentry, resulting in the |
2475 | * false-negative result. d_lookup() protects against concurrent |
2476 | * renames using rename_lock seqlock. |
2477 | * |
2478 | * See Documentation/filesystems/path-lookup.txt for more details. |
2479 | */ |
2480 | rcu_read_lock(); |
2481 | |
2482 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
2483 | |
2484 | if (dentry->d_name.hash != hash) |
2485 | continue; |
2486 | |
2487 | spin_lock(lock: &dentry->d_lock); |
2488 | if (dentry->d_parent != parent) |
2489 | goto next; |
2490 | if (d_unhashed(dentry)) |
2491 | goto next; |
2492 | |
2493 | if (!d_same_name(dentry, parent, name)) |
2494 | goto next; |
2495 | |
2496 | dentry->d_lockref.count++; |
2497 | found = dentry; |
2498 | spin_unlock(lock: &dentry->d_lock); |
2499 | break; |
2500 | next: |
2501 | spin_unlock(lock: &dentry->d_lock); |
2502 | } |
2503 | rcu_read_unlock(); |
2504 | |
2505 | return found; |
2506 | } |
2507 | |
2508 | /** |
2509 | * d_hash_and_lookup - hash the qstr then search for a dentry |
2510 | * @dir: Directory to search in |
2511 | * @name: qstr of name we wish to find |
2512 | * |
2513 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
2514 | */ |
2515 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) |
2516 | { |
2517 | /* |
2518 | * Check for a fs-specific hash function. Note that we must |
2519 | * calculate the standard hash first, as the d_op->d_hash() |
2520 | * routine may choose to leave the hash value unchanged. |
2521 | */ |
2522 | name->hash = full_name_hash(salt: dir, name->name, name->len); |
2523 | if (dir->d_flags & DCACHE_OP_HASH) { |
2524 | int err = dir->d_op->d_hash(dir, name); |
2525 | if (unlikely(err < 0)) |
2526 | return ERR_PTR(error: err); |
2527 | } |
2528 | return d_lookup(dir, name); |
2529 | } |
2530 | EXPORT_SYMBOL(d_hash_and_lookup); |
2531 | |
2532 | /* |
2533 | * When a file is deleted, we have two options: |
2534 | * - turn this dentry into a negative dentry |
2535 | * - unhash this dentry and free it. |
2536 | * |
2537 | * Usually, we want to just turn this into |
2538 | * a negative dentry, but if anybody else is |
2539 | * currently using the dentry or the inode |
2540 | * we can't do that and we fall back on removing |
2541 | * it from the hash queues and waiting for |
2542 | * it to be deleted later when it has no users |
2543 | */ |
2544 | |
2545 | /** |
2546 | * d_delete - delete a dentry |
2547 | * @dentry: The dentry to delete |
2548 | * |
2549 | * Turn the dentry into a negative dentry if possible, otherwise |
2550 | * remove it from the hash queues so it can be deleted later |
2551 | */ |
2552 | |
2553 | void d_delete(struct dentry * dentry) |
2554 | { |
2555 | struct inode *inode = dentry->d_inode; |
2556 | |
2557 | spin_lock(lock: &inode->i_lock); |
2558 | spin_lock(lock: &dentry->d_lock); |
2559 | /* |
2560 | * Are we the only user? |
2561 | */ |
2562 | if (dentry->d_lockref.count == 1) { |
2563 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
2564 | dentry_unlink_inode(dentry); |
2565 | } else { |
2566 | __d_drop(dentry); |
2567 | spin_unlock(lock: &dentry->d_lock); |
2568 | spin_unlock(lock: &inode->i_lock); |
2569 | } |
2570 | } |
2571 | EXPORT_SYMBOL(d_delete); |
2572 | |
2573 | static void __d_rehash(struct dentry *entry) |
2574 | { |
2575 | struct hlist_bl_head *b = d_hash(hash: entry->d_name.hash); |
2576 | |
2577 | hlist_bl_lock(b); |
2578 | hlist_bl_add_head_rcu(n: &entry->d_hash, h: b); |
2579 | hlist_bl_unlock(b); |
2580 | } |
2581 | |
2582 | /** |
2583 | * d_rehash - add an entry back to the hash |
2584 | * @entry: dentry to add to the hash |
2585 | * |
2586 | * Adds a dentry to the hash according to its name. |
2587 | */ |
2588 | |
2589 | void d_rehash(struct dentry * entry) |
2590 | { |
2591 | spin_lock(lock: &entry->d_lock); |
2592 | __d_rehash(entry); |
2593 | spin_unlock(lock: &entry->d_lock); |
2594 | } |
2595 | EXPORT_SYMBOL(d_rehash); |
2596 | |
2597 | static inline unsigned start_dir_add(struct inode *dir) |
2598 | { |
2599 | preempt_disable_nested(); |
2600 | for (;;) { |
2601 | unsigned n = dir->i_dir_seq; |
2602 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) |
2603 | return n; |
2604 | cpu_relax(); |
2605 | } |
2606 | } |
2607 | |
2608 | static inline void end_dir_add(struct inode *dir, unsigned int n, |
2609 | wait_queue_head_t *d_wait) |
2610 | { |
2611 | smp_store_release(&dir->i_dir_seq, n + 2); |
2612 | preempt_enable_nested(); |
2613 | wake_up_all(d_wait); |
2614 | } |
2615 | |
2616 | static void d_wait_lookup(struct dentry *dentry) |
2617 | { |
2618 | if (d_in_lookup(dentry)) { |
2619 | DECLARE_WAITQUEUE(wait, current); |
2620 | add_wait_queue(wq_head: dentry->d_wait, wq_entry: &wait); |
2621 | do { |
2622 | set_current_state(TASK_UNINTERRUPTIBLE); |
2623 | spin_unlock(lock: &dentry->d_lock); |
2624 | schedule(); |
2625 | spin_lock(lock: &dentry->d_lock); |
2626 | } while (d_in_lookup(dentry)); |
2627 | } |
2628 | } |
2629 | |
2630 | struct dentry *d_alloc_parallel(struct dentry *parent, |
2631 | const struct qstr *name, |
2632 | wait_queue_head_t *wq) |
2633 | { |
2634 | unsigned int hash = name->hash; |
2635 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2636 | struct hlist_bl_node *node; |
2637 | struct dentry *new = d_alloc(parent, name); |
2638 | struct dentry *dentry; |
2639 | unsigned seq, r_seq, d_seq; |
2640 | |
2641 | if (unlikely(!new)) |
2642 | return ERR_PTR(error: -ENOMEM); |
2643 | |
2644 | retry: |
2645 | rcu_read_lock(); |
2646 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
2647 | r_seq = read_seqbegin(sl: &rename_lock); |
2648 | dentry = __d_lookup_rcu(parent, name, seqp: &d_seq); |
2649 | if (unlikely(dentry)) { |
2650 | if (!lockref_get_not_dead(&dentry->d_lockref)) { |
2651 | rcu_read_unlock(); |
2652 | goto retry; |
2653 | } |
2654 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { |
2655 | rcu_read_unlock(); |
2656 | dput(dentry); |
2657 | goto retry; |
2658 | } |
2659 | rcu_read_unlock(); |
2660 | dput(new); |
2661 | return dentry; |
2662 | } |
2663 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { |
2664 | rcu_read_unlock(); |
2665 | goto retry; |
2666 | } |
2667 | |
2668 | if (unlikely(seq & 1)) { |
2669 | rcu_read_unlock(); |
2670 | goto retry; |
2671 | } |
2672 | |
2673 | hlist_bl_lock(b); |
2674 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
2675 | hlist_bl_unlock(b); |
2676 | rcu_read_unlock(); |
2677 | goto retry; |
2678 | } |
2679 | /* |
2680 | * No changes for the parent since the beginning of d_lookup(). |
2681 | * Since all removals from the chain happen with hlist_bl_lock(), |
2682 | * any potential in-lookup matches are going to stay here until |
2683 | * we unlock the chain. All fields are stable in everything |
2684 | * we encounter. |
2685 | */ |
2686 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { |
2687 | if (dentry->d_name.hash != hash) |
2688 | continue; |
2689 | if (dentry->d_parent != parent) |
2690 | continue; |
2691 | if (!d_same_name(dentry, parent, name)) |
2692 | continue; |
2693 | hlist_bl_unlock(b); |
2694 | /* now we can try to grab a reference */ |
2695 | if (!lockref_get_not_dead(&dentry->d_lockref)) { |
2696 | rcu_read_unlock(); |
2697 | goto retry; |
2698 | } |
2699 | |
2700 | rcu_read_unlock(); |
2701 | /* |
2702 | * somebody is likely to be still doing lookup for it; |
2703 | * wait for them to finish |
2704 | */ |
2705 | spin_lock(lock: &dentry->d_lock); |
2706 | d_wait_lookup(dentry); |
2707 | /* |
2708 | * it's not in-lookup anymore; in principle we should repeat |
2709 | * everything from dcache lookup, but it's likely to be what |
2710 | * d_lookup() would've found anyway. If it is, just return it; |
2711 | * otherwise we really have to repeat the whole thing. |
2712 | */ |
2713 | if (unlikely(dentry->d_name.hash != hash)) |
2714 | goto mismatch; |
2715 | if (unlikely(dentry->d_parent != parent)) |
2716 | goto mismatch; |
2717 | if (unlikely(d_unhashed(dentry))) |
2718 | goto mismatch; |
2719 | if (unlikely(!d_same_name(dentry, parent, name))) |
2720 | goto mismatch; |
2721 | /* OK, it *is* a hashed match; return it */ |
2722 | spin_unlock(lock: &dentry->d_lock); |
2723 | dput(new); |
2724 | return dentry; |
2725 | } |
2726 | rcu_read_unlock(); |
2727 | /* we can't take ->d_lock here; it's OK, though. */ |
2728 | new->d_flags |= DCACHE_PAR_LOOKUP; |
2729 | new->d_wait = wq; |
2730 | hlist_bl_add_head_rcu(n: &new->d_u.d_in_lookup_hash, h: b); |
2731 | hlist_bl_unlock(b); |
2732 | return new; |
2733 | mismatch: |
2734 | spin_unlock(lock: &dentry->d_lock); |
2735 | dput(dentry); |
2736 | goto retry; |
2737 | } |
2738 | EXPORT_SYMBOL(d_alloc_parallel); |
2739 | |
2740 | /* |
2741 | * - Unhash the dentry |
2742 | * - Retrieve and clear the waitqueue head in dentry |
2743 | * - Return the waitqueue head |
2744 | */ |
2745 | static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry) |
2746 | { |
2747 | wait_queue_head_t *d_wait; |
2748 | struct hlist_bl_head *b; |
2749 | |
2750 | lockdep_assert_held(&dentry->d_lock); |
2751 | |
2752 | b = in_lookup_hash(parent: dentry->d_parent, hash: dentry->d_name.hash); |
2753 | hlist_bl_lock(b); |
2754 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
2755 | __hlist_bl_del(n: &dentry->d_u.d_in_lookup_hash); |
2756 | d_wait = dentry->d_wait; |
2757 | dentry->d_wait = NULL; |
2758 | hlist_bl_unlock(b); |
2759 | INIT_HLIST_NODE(h: &dentry->d_u.d_alias); |
2760 | INIT_LIST_HEAD(list: &dentry->d_lru); |
2761 | return d_wait; |
2762 | } |
2763 | |
2764 | void __d_lookup_unhash_wake(struct dentry *dentry) |
2765 | { |
2766 | spin_lock(lock: &dentry->d_lock); |
2767 | wake_up_all(__d_lookup_unhash(dentry)); |
2768 | spin_unlock(lock: &dentry->d_lock); |
2769 | } |
2770 | EXPORT_SYMBOL(__d_lookup_unhash_wake); |
2771 | |
2772 | /* inode->i_lock held if inode is non-NULL */ |
2773 | |
2774 | static inline void __d_add(struct dentry *dentry, struct inode *inode) |
2775 | { |
2776 | wait_queue_head_t *d_wait; |
2777 | struct inode *dir = NULL; |
2778 | unsigned n; |
2779 | spin_lock(lock: &dentry->d_lock); |
2780 | if (unlikely(d_in_lookup(dentry))) { |
2781 | dir = dentry->d_parent->d_inode; |
2782 | n = start_dir_add(dir); |
2783 | d_wait = __d_lookup_unhash(dentry); |
2784 | } |
2785 | if (inode) { |
2786 | unsigned add_flags = d_flags_for_inode(inode); |
2787 | hlist_add_head(n: &dentry->d_u.d_alias, h: &inode->i_dentry); |
2788 | raw_write_seqcount_begin(&dentry->d_seq); |
2789 | __d_set_inode_and_type(dentry, inode, type_flags: add_flags); |
2790 | raw_write_seqcount_end(&dentry->d_seq); |
2791 | fsnotify_update_flags(dentry); |
2792 | } |
2793 | __d_rehash(entry: dentry); |
2794 | if (dir) |
2795 | end_dir_add(dir, n, d_wait); |
2796 | spin_unlock(lock: &dentry->d_lock); |
2797 | if (inode) |
2798 | spin_unlock(lock: &inode->i_lock); |
2799 | } |
2800 | |
2801 | /** |
2802 | * d_add - add dentry to hash queues |
2803 | * @entry: dentry to add |
2804 | * @inode: The inode to attach to this dentry |
2805 | * |
2806 | * This adds the entry to the hash queues and initializes @inode. |
2807 | * The entry was actually filled in earlier during d_alloc(). |
2808 | */ |
2809 | |
2810 | void d_add(struct dentry *entry, struct inode *inode) |
2811 | { |
2812 | if (inode) { |
2813 | security_d_instantiate(dentry: entry, inode); |
2814 | spin_lock(lock: &inode->i_lock); |
2815 | } |
2816 | __d_add(dentry: entry, inode); |
2817 | } |
2818 | EXPORT_SYMBOL(d_add); |
2819 | |
2820 | /** |
2821 | * d_exact_alias - find and hash an exact unhashed alias |
2822 | * @entry: dentry to add |
2823 | * @inode: The inode to go with this dentry |
2824 | * |
2825 | * If an unhashed dentry with the same name/parent and desired |
2826 | * inode already exists, hash and return it. Otherwise, return |
2827 | * NULL. |
2828 | * |
2829 | * Parent directory should be locked. |
2830 | */ |
2831 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) |
2832 | { |
2833 | struct dentry *alias; |
2834 | unsigned int hash = entry->d_name.hash; |
2835 | |
2836 | spin_lock(lock: &inode->i_lock); |
2837 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
2838 | /* |
2839 | * Don't need alias->d_lock here, because aliases with |
2840 | * d_parent == entry->d_parent are not subject to name or |
2841 | * parent changes, because the parent inode i_mutex is held. |
2842 | */ |
2843 | if (alias->d_name.hash != hash) |
2844 | continue; |
2845 | if (alias->d_parent != entry->d_parent) |
2846 | continue; |
2847 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
2848 | continue; |
2849 | spin_lock(lock: &alias->d_lock); |
2850 | if (!d_unhashed(dentry: alias)) { |
2851 | spin_unlock(lock: &alias->d_lock); |
2852 | alias = NULL; |
2853 | } else { |
2854 | __dget_dlock(dentry: alias); |
2855 | __d_rehash(entry: alias); |
2856 | spin_unlock(lock: &alias->d_lock); |
2857 | } |
2858 | spin_unlock(lock: &inode->i_lock); |
2859 | return alias; |
2860 | } |
2861 | spin_unlock(lock: &inode->i_lock); |
2862 | return NULL; |
2863 | } |
2864 | EXPORT_SYMBOL(d_exact_alias); |
2865 | |
2866 | static void swap_names(struct dentry *dentry, struct dentry *target) |
2867 | { |
2868 | if (unlikely(dname_external(target))) { |
2869 | if (unlikely(dname_external(dentry))) { |
2870 | /* |
2871 | * Both external: swap the pointers |
2872 | */ |
2873 | swap(target->d_name.name, dentry->d_name.name); |
2874 | } else { |
2875 | /* |
2876 | * dentry:internal, target:external. Steal target's |
2877 | * storage and make target internal. |
2878 | */ |
2879 | memcpy(target->d_iname, dentry->d_name.name, |
2880 | dentry->d_name.len + 1); |
2881 | dentry->d_name.name = target->d_name.name; |
2882 | target->d_name.name = target->d_iname; |
2883 | } |
2884 | } else { |
2885 | if (unlikely(dname_external(dentry))) { |
2886 | /* |
2887 | * dentry:external, target:internal. Give dentry's |
2888 | * storage to target and make dentry internal |
2889 | */ |
2890 | memcpy(dentry->d_iname, target->d_name.name, |
2891 | target->d_name.len + 1); |
2892 | target->d_name.name = dentry->d_name.name; |
2893 | dentry->d_name.name = dentry->d_iname; |
2894 | } else { |
2895 | /* |
2896 | * Both are internal. |
2897 | */ |
2898 | unsigned int i; |
2899 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); |
2900 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { |
2901 | swap(((long *) &dentry->d_iname)[i], |
2902 | ((long *) &target->d_iname)[i]); |
2903 | } |
2904 | } |
2905 | } |
2906 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
2907 | } |
2908 | |
2909 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2910 | { |
2911 | struct external_name *old_name = NULL; |
2912 | if (unlikely(dname_external(dentry))) |
2913 | old_name = external_name(dentry); |
2914 | if (unlikely(dname_external(target))) { |
2915 | atomic_inc(v: &external_name(dentry: target)->u.count); |
2916 | dentry->d_name = target->d_name; |
2917 | } else { |
2918 | memcpy(dentry->d_iname, target->d_name.name, |
2919 | target->d_name.len + 1); |
2920 | dentry->d_name.name = dentry->d_iname; |
2921 | dentry->d_name.hash_len = target->d_name.hash_len; |
2922 | } |
2923 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) |
2924 | kfree_rcu(old_name, u.head); |
2925 | } |
2926 | |
2927 | /* |
2928 | * __d_move - move a dentry |
2929 | * @dentry: entry to move |
2930 | * @target: new dentry |
2931 | * @exchange: exchange the two dentries |
2932 | * |
2933 | * Update the dcache to reflect the move of a file name. Negative |
2934 | * dcache entries should not be moved in this way. Caller must hold |
2935 | * rename_lock, the i_mutex of the source and target directories, |
2936 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). |
2937 | */ |
2938 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2939 | bool exchange) |
2940 | { |
2941 | struct dentry *old_parent, *p; |
2942 | wait_queue_head_t *d_wait; |
2943 | struct inode *dir = NULL; |
2944 | unsigned n; |
2945 | |
2946 | WARN_ON(!dentry->d_inode); |
2947 | if (WARN_ON(dentry == target)) |
2948 | return; |
2949 | |
2950 | BUG_ON(d_ancestor(target, dentry)); |
2951 | old_parent = dentry->d_parent; |
2952 | p = d_ancestor(old_parent, target); |
2953 | if (IS_ROOT(dentry)) { |
2954 | BUG_ON(p); |
2955 | spin_lock(lock: &target->d_parent->d_lock); |
2956 | } else if (!p) { |
2957 | /* target is not a descendent of dentry->d_parent */ |
2958 | spin_lock(lock: &target->d_parent->d_lock); |
2959 | spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED); |
2960 | } else { |
2961 | BUG_ON(p == dentry); |
2962 | spin_lock(lock: &old_parent->d_lock); |
2963 | if (p != target) |
2964 | spin_lock_nested(&target->d_parent->d_lock, |
2965 | DENTRY_D_LOCK_NESTED); |
2966 | } |
2967 | spin_lock_nested(&dentry->d_lock, 2); |
2968 | spin_lock_nested(&target->d_lock, 3); |
2969 | |
2970 | if (unlikely(d_in_lookup(target))) { |
2971 | dir = target->d_parent->d_inode; |
2972 | n = start_dir_add(dir); |
2973 | d_wait = __d_lookup_unhash(dentry: target); |
2974 | } |
2975 | |
2976 | write_seqcount_begin(&dentry->d_seq); |
2977 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
2978 | |
2979 | /* unhash both */ |
2980 | if (!d_unhashed(dentry)) |
2981 | ___d_drop(dentry); |
2982 | if (!d_unhashed(dentry: target)) |
2983 | ___d_drop(dentry: target); |
2984 | |
2985 | /* ... and switch them in the tree */ |
2986 | dentry->d_parent = target->d_parent; |
2987 | if (!exchange) { |
2988 | copy_name(dentry, target); |
2989 | target->d_hash.pprev = NULL; |
2990 | dentry->d_parent->d_lockref.count++; |
2991 | if (dentry != old_parent) /* wasn't IS_ROOT */ |
2992 | WARN_ON(!--old_parent->d_lockref.count); |
2993 | } else { |
2994 | target->d_parent = old_parent; |
2995 | swap_names(dentry, target); |
2996 | list_move(list: &target->d_child, head: &target->d_parent->d_subdirs); |
2997 | __d_rehash(entry: target); |
2998 | fsnotify_update_flags(dentry: target); |
2999 | } |
3000 | list_move(list: &dentry->d_child, head: &dentry->d_parent->d_subdirs); |
3001 | __d_rehash(entry: dentry); |
3002 | fsnotify_update_flags(dentry); |
3003 | fscrypt_handle_d_move(dentry); |
3004 | |
3005 | write_seqcount_end(&target->d_seq); |
3006 | write_seqcount_end(&dentry->d_seq); |
3007 | |
3008 | if (dir) |
3009 | end_dir_add(dir, n, d_wait); |
3010 | |
3011 | if (dentry->d_parent != old_parent) |
3012 | spin_unlock(lock: &dentry->d_parent->d_lock); |
3013 | if (dentry != old_parent) |
3014 | spin_unlock(lock: &old_parent->d_lock); |
3015 | spin_unlock(lock: &target->d_lock); |
3016 | spin_unlock(lock: &dentry->d_lock); |
3017 | } |
3018 | |
3019 | /* |
3020 | * d_move - move a dentry |
3021 | * @dentry: entry to move |
3022 | * @target: new dentry |
3023 | * |
3024 | * Update the dcache to reflect the move of a file name. Negative |
3025 | * dcache entries should not be moved in this way. See the locking |
3026 | * requirements for __d_move. |
3027 | */ |
3028 | void d_move(struct dentry *dentry, struct dentry *target) |
3029 | { |
3030 | write_seqlock(sl: &rename_lock); |
3031 | __d_move(dentry, target, exchange: false); |
3032 | write_sequnlock(sl: &rename_lock); |
3033 | } |
3034 | EXPORT_SYMBOL(d_move); |
3035 | |
3036 | /* |
3037 | * d_exchange - exchange two dentries |
3038 | * @dentry1: first dentry |
3039 | * @dentry2: second dentry |
3040 | */ |
3041 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) |
3042 | { |
3043 | write_seqlock(sl: &rename_lock); |
3044 | |
3045 | WARN_ON(!dentry1->d_inode); |
3046 | WARN_ON(!dentry2->d_inode); |
3047 | WARN_ON(IS_ROOT(dentry1)); |
3048 | WARN_ON(IS_ROOT(dentry2)); |
3049 | |
3050 | __d_move(dentry: dentry1, target: dentry2, exchange: true); |
3051 | |
3052 | write_sequnlock(sl: &rename_lock); |
3053 | } |
3054 | |
3055 | /** |
3056 | * d_ancestor - search for an ancestor |
3057 | * @p1: ancestor dentry |
3058 | * @p2: child dentry |
3059 | * |
3060 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is |
3061 | * an ancestor of p2, else NULL. |
3062 | */ |
3063 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
3064 | { |
3065 | struct dentry *p; |
3066 | |
3067 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
3068 | if (p->d_parent == p1) |
3069 | return p; |
3070 | } |
3071 | return NULL; |
3072 | } |
3073 | |
3074 | /* |
3075 | * This helper attempts to cope with remotely renamed directories |
3076 | * |
3077 | * It assumes that the caller is already holding |
3078 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
3079 | * |
3080 | * Note: If ever the locking in lock_rename() changes, then please |
3081 | * remember to update this too... |
3082 | */ |
3083 | static int __d_unalias(struct inode *inode, |
3084 | struct dentry *dentry, struct dentry *alias) |
3085 | { |
3086 | struct mutex *m1 = NULL; |
3087 | struct rw_semaphore *m2 = NULL; |
3088 | int ret = -ESTALE; |
3089 | |
3090 | /* If alias and dentry share a parent, then no extra locks required */ |
3091 | if (alias->d_parent == dentry->d_parent) |
3092 | goto out_unalias; |
3093 | |
3094 | /* See lock_rename() */ |
3095 | if (!mutex_trylock(lock: &dentry->d_sb->s_vfs_rename_mutex)) |
3096 | goto out_err; |
3097 | m1 = &dentry->d_sb->s_vfs_rename_mutex; |
3098 | if (!inode_trylock_shared(inode: alias->d_parent->d_inode)) |
3099 | goto out_err; |
3100 | m2 = &alias->d_parent->d_inode->i_rwsem; |
3101 | out_unalias: |
3102 | __d_move(dentry: alias, target: dentry, exchange: false); |
3103 | ret = 0; |
3104 | out_err: |
3105 | if (m2) |
3106 | up_read(sem: m2); |
3107 | if (m1) |
3108 | mutex_unlock(lock: m1); |
3109 | return ret; |
3110 | } |
3111 | |
3112 | /** |
3113 | * d_splice_alias - splice a disconnected dentry into the tree if one exists |
3114 | * @inode: the inode which may have a disconnected dentry |
3115 | * @dentry: a negative dentry which we want to point to the inode. |
3116 | * |
3117 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
3118 | * place of the given dentry and return it, else simply d_add the inode |
3119 | * to the dentry and return NULL. |
3120 | * |
3121 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
3122 | * we should error out: directories can't have multiple aliases. |
3123 | * |
3124 | * This is needed in the lookup routine of any filesystem that is exportable |
3125 | * (via knfsd) so that we can build dcache paths to directories effectively. |
3126 | * |
3127 | * If a dentry was found and moved, then it is returned. Otherwise NULL |
3128 | * is returned. This matches the expected return value of ->lookup. |
3129 | * |
3130 | * Cluster filesystems may call this function with a negative, hashed dentry. |
3131 | * In that case, we know that the inode will be a regular file, and also this |
3132 | * will only occur during atomic_open. So we need to check for the dentry |
3133 | * being already hashed only in the final case. |
3134 | */ |
3135 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) |
3136 | { |
3137 | if (IS_ERR(ptr: inode)) |
3138 | return ERR_CAST(ptr: inode); |
3139 | |
3140 | BUG_ON(!d_unhashed(dentry)); |
3141 | |
3142 | if (!inode) |
3143 | goto out; |
3144 | |
3145 | security_d_instantiate(dentry, inode); |
3146 | spin_lock(lock: &inode->i_lock); |
3147 | if (S_ISDIR(inode->i_mode)) { |
3148 | struct dentry *new = __d_find_any_alias(inode); |
3149 | if (unlikely(new)) { |
3150 | /* The reference to new ensures it remains an alias */ |
3151 | spin_unlock(lock: &inode->i_lock); |
3152 | write_seqlock(sl: &rename_lock); |
3153 | if (unlikely(d_ancestor(new, dentry))) { |
3154 | write_sequnlock(sl: &rename_lock); |
3155 | dput(new); |
3156 | new = ERR_PTR(error: -ELOOP); |
3157 | pr_warn_ratelimited( |
3158 | "VFS: Lookup of '%s' in %s %s" |
3159 | " would have caused loop\n" , |
3160 | dentry->d_name.name, |
3161 | inode->i_sb->s_type->name, |
3162 | inode->i_sb->s_id); |
3163 | } else if (!IS_ROOT(new)) { |
3164 | struct dentry *old_parent = dget(dentry: new->d_parent); |
3165 | int err = __d_unalias(inode, dentry, alias: new); |
3166 | write_sequnlock(sl: &rename_lock); |
3167 | if (err) { |
3168 | dput(new); |
3169 | new = ERR_PTR(error: err); |
3170 | } |
3171 | dput(old_parent); |
3172 | } else { |
3173 | __d_move(dentry: new, target: dentry, exchange: false); |
3174 | write_sequnlock(sl: &rename_lock); |
3175 | } |
3176 | iput(inode); |
3177 | return new; |
3178 | } |
3179 | } |
3180 | out: |
3181 | __d_add(dentry, inode); |
3182 | return NULL; |
3183 | } |
3184 | EXPORT_SYMBOL(d_splice_alias); |
3185 | |
3186 | /* |
3187 | * Test whether new_dentry is a subdirectory of old_dentry. |
3188 | * |
3189 | * Trivially implemented using the dcache structure |
3190 | */ |
3191 | |
3192 | /** |
3193 | * is_subdir - is new dentry a subdirectory of old_dentry |
3194 | * @new_dentry: new dentry |
3195 | * @old_dentry: old dentry |
3196 | * |
3197 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3198 | * Returns false otherwise. |
3199 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3200 | */ |
3201 | |
3202 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
3203 | { |
3204 | bool result; |
3205 | unsigned seq; |
3206 | |
3207 | if (new_dentry == old_dentry) |
3208 | return true; |
3209 | |
3210 | do { |
3211 | /* for restarting inner loop in case of seq retry */ |
3212 | seq = read_seqbegin(sl: &rename_lock); |
3213 | /* |
3214 | * Need rcu_readlock to protect against the d_parent trashing |
3215 | * due to d_move |
3216 | */ |
3217 | rcu_read_lock(); |
3218 | if (d_ancestor(p1: old_dentry, p2: new_dentry)) |
3219 | result = true; |
3220 | else |
3221 | result = false; |
3222 | rcu_read_unlock(); |
3223 | } while (read_seqretry(sl: &rename_lock, start: seq)); |
3224 | |
3225 | return result; |
3226 | } |
3227 | EXPORT_SYMBOL(is_subdir); |
3228 | |
3229 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
3230 | { |
3231 | struct dentry *root = data; |
3232 | if (dentry != root) { |
3233 | if (d_unhashed(dentry) || !dentry->d_inode) |
3234 | return D_WALK_SKIP; |
3235 | |
3236 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3237 | dentry->d_flags |= DCACHE_GENOCIDE; |
3238 | dentry->d_lockref.count--; |
3239 | } |
3240 | } |
3241 | return D_WALK_CONTINUE; |
3242 | } |
3243 | |
3244 | void d_genocide(struct dentry *parent) |
3245 | { |
3246 | d_walk(parent, data: parent, enter: d_genocide_kill); |
3247 | } |
3248 | |
3249 | void d_mark_tmpfile(struct file *file, struct inode *inode) |
3250 | { |
3251 | struct dentry *dentry = file->f_path.dentry; |
3252 | |
3253 | BUG_ON(dentry->d_name.name != dentry->d_iname || |
3254 | !hlist_unhashed(&dentry->d_u.d_alias) || |
3255 | !d_unlinked(dentry)); |
3256 | spin_lock(lock: &dentry->d_parent->d_lock); |
3257 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
3258 | dentry->d_name.len = sprintf(buf: dentry->d_iname, fmt: "#%llu" , |
3259 | (unsigned long long)inode->i_ino); |
3260 | spin_unlock(lock: &dentry->d_lock); |
3261 | spin_unlock(lock: &dentry->d_parent->d_lock); |
3262 | } |
3263 | EXPORT_SYMBOL(d_mark_tmpfile); |
3264 | |
3265 | void d_tmpfile(struct file *file, struct inode *inode) |
3266 | { |
3267 | struct dentry *dentry = file->f_path.dentry; |
3268 | |
3269 | inode_dec_link_count(inode); |
3270 | d_mark_tmpfile(file, inode); |
3271 | d_instantiate(dentry, inode); |
3272 | } |
3273 | EXPORT_SYMBOL(d_tmpfile); |
3274 | |
3275 | static __initdata unsigned long dhash_entries; |
3276 | static int __init set_dhash_entries(char *str) |
3277 | { |
3278 | if (!str) |
3279 | return 0; |
3280 | dhash_entries = simple_strtoul(str, &str, 0); |
3281 | return 1; |
3282 | } |
3283 | __setup("dhash_entries=" , set_dhash_entries); |
3284 | |
3285 | static void __init dcache_init_early(void) |
3286 | { |
3287 | /* If hashes are distributed across NUMA nodes, defer |
3288 | * hash allocation until vmalloc space is available. |
3289 | */ |
3290 | if (hashdist) |
3291 | return; |
3292 | |
3293 | dentry_hashtable = |
3294 | alloc_large_system_hash(tablename: "Dentry cache" , |
3295 | bucketsize: sizeof(struct hlist_bl_head), |
3296 | numentries: dhash_entries, |
3297 | scale: 13, |
3298 | HASH_EARLY | HASH_ZERO, |
3299 | hash_shift: &d_hash_shift, |
3300 | NULL, |
3301 | low_limit: 0, |
3302 | high_limit: 0); |
3303 | d_hash_shift = 32 - d_hash_shift; |
3304 | } |
3305 | |
3306 | static void __init dcache_init(void) |
3307 | { |
3308 | /* |
3309 | * A constructor could be added for stable state like the lists, |
3310 | * but it is probably not worth it because of the cache nature |
3311 | * of the dcache. |
3312 | */ |
3313 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, |
3314 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, |
3315 | d_iname); |
3316 | |
3317 | /* Hash may have been set up in dcache_init_early */ |
3318 | if (!hashdist) |
3319 | return; |
3320 | |
3321 | dentry_hashtable = |
3322 | alloc_large_system_hash(tablename: "Dentry cache" , |
3323 | bucketsize: sizeof(struct hlist_bl_head), |
3324 | numentries: dhash_entries, |
3325 | scale: 13, |
3326 | HASH_ZERO, |
3327 | hash_shift: &d_hash_shift, |
3328 | NULL, |
3329 | low_limit: 0, |
3330 | high_limit: 0); |
3331 | d_hash_shift = 32 - d_hash_shift; |
3332 | } |
3333 | |
3334 | /* SLAB cache for __getname() consumers */ |
3335 | struct kmem_cache *names_cachep __ro_after_init; |
3336 | EXPORT_SYMBOL(names_cachep); |
3337 | |
3338 | void __init vfs_caches_init_early(void) |
3339 | { |
3340 | int i; |
3341 | |
3342 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) |
3343 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); |
3344 | |
3345 | dcache_init_early(); |
3346 | inode_init_early(); |
3347 | } |
3348 | |
3349 | void __init vfs_caches_init(void) |
3350 | { |
3351 | names_cachep = kmem_cache_create_usercopy(name: "names_cache" , PATH_MAX, align: 0, |
3352 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, useroffset: 0, PATH_MAX, NULL); |
3353 | |
3354 | dcache_init(); |
3355 | inode_init(); |
3356 | files_init(); |
3357 | files_maxfiles_init(); |
3358 | mnt_init(); |
3359 | bdev_cache_init(); |
3360 | chrdev_init(); |
3361 | } |
3362 | |