Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * fs/dcache.c
4 : *
5 : * Complete reimplementation
6 : * (C) 1997 Thomas Schoebel-Theuer,
7 : * with heavy changes by Linus Torvalds
8 : */
9 :
10 : /*
11 : * Notes on the allocation strategy:
12 : *
13 : * The dcache is a master of the icache - whenever a dcache entry
14 : * exists, the inode will always exist. "iput()" is done either when
15 : * the dcache entry is deleted or garbage collected.
16 : */
17 :
18 : #include <linux/ratelimit.h>
19 : #include <linux/string.h>
20 : #include <linux/mm.h>
21 : #include <linux/fs.h>
22 : #include <linux/fscrypt.h>
23 : #include <linux/fsnotify.h>
24 : #include <linux/slab.h>
25 : #include <linux/init.h>
26 : #include <linux/hash.h>
27 : #include <linux/cache.h>
28 : #include <linux/export.h>
29 : #include <linux/security.h>
30 : #include <linux/seqlock.h>
31 : #include <linux/memblock.h>
32 : #include <linux/bit_spinlock.h>
33 : #include <linux/rculist_bl.h>
34 : #include <linux/list_lru.h>
35 : #include "internal.h"
36 : #include "mount.h"
37 :
38 : /*
39 : * Usage:
40 : * dcache->d_inode->i_lock protects:
41 : * - i_dentry, d_u.d_alias, d_inode of aliases
42 : * dcache_hash_bucket lock protects:
43 : * - the dcache hash table
44 : * s_roots bl list spinlock protects:
45 : * - the s_roots list (see __d_drop)
46 : * dentry->d_sb->s_dentry_lru_lock protects:
47 : * - the dcache lru lists and counters
48 : * d_lock protects:
49 : * - d_flags
50 : * - d_name
51 : * - d_lru
52 : * - d_count
53 : * - d_unhashed()
54 : * - d_parent and d_subdirs
55 : * - childrens' d_child and d_parent
56 : * - d_u.d_alias, d_inode
57 : *
58 : * Ordering:
59 : * dentry->d_inode->i_lock
60 : * dentry->d_lock
61 : * dentry->d_sb->s_dentry_lru_lock
62 : * dcache_hash_bucket lock
63 : * s_roots lock
64 : *
65 : * If there is an ancestor relationship:
66 : * dentry->d_parent->...->d_parent->d_lock
67 : * ...
68 : * dentry->d_parent->d_lock
69 : * dentry->d_lock
70 : *
71 : * If no ancestor relationship:
72 : * arbitrary, since it's serialized on rename_lock
73 : */
74 : int sysctl_vfs_cache_pressure __read_mostly = 100;
75 : EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
76 :
77 : __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
78 :
79 : EXPORT_SYMBOL(rename_lock);
80 :
81 : static struct kmem_cache *dentry_cache __read_mostly;
82 :
83 : const struct qstr empty_name = QSTR_INIT("", 0);
84 : EXPORT_SYMBOL(empty_name);
85 : const struct qstr slash_name = QSTR_INIT("/", 1);
86 : EXPORT_SYMBOL(slash_name);
87 : 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 __read_mostly;
100 :
101 : static struct hlist_bl_head *dentry_hashtable __read_mostly;
102 :
103 77086832710 : static inline struct hlist_bl_head *d_hash(unsigned int hash)
104 : {
105 77086832710 : 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 2234628010 : hash += (unsigned long) parent / L1_CACHE_BYTES;
115 2234628010 : return in_lookup_hashtable + hash_32(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 1120 : static long get_nr_dentry(void)
150 : {
151 1120 : int i;
152 1120 : long sum = 0;
153 5600 : for_each_possible_cpu(i)
154 4480 : sum += per_cpu(nr_dentry, i);
155 1120 : return sum < 0 ? 0 : sum;
156 : }
157 :
158 1120 : static long get_nr_dentry_unused(void)
159 : {
160 1120 : int i;
161 1120 : long sum = 0;
162 5600 : for_each_possible_cpu(i)
163 4480 : sum += per_cpu(nr_dentry_unused, i);
164 1120 : return sum < 0 ? 0 : sum;
165 : }
166 :
167 1120 : static long get_nr_dentry_negative(void)
168 : {
169 1120 : int i;
170 1120 : long sum = 0;
171 :
172 5600 : for_each_possible_cpu(i)
173 4480 : sum += per_cpu(nr_dentry_negative, i);
174 1120 : return sum < 0 ? 0 : sum;
175 : }
176 :
177 1120 : static int proc_nr_dentry(struct ctl_table *table, int write, void *buffer,
178 : size_t *lenp, loff_t *ppos)
179 : {
180 1120 : dentry_stat.nr_dentry = get_nr_dentry();
181 1120 : dentry_stat.nr_unused = get_nr_dentry_unused();
182 1120 : dentry_stat.nr_negative = get_nr_dentry_negative();
183 1120 : 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 0 : static int __init init_fs_dcache_sysctls(void)
198 : {
199 0 : register_sysctl_init("fs", fs_dcache_sysctls);
200 0 : 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 72119944377 : static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
222 : {
223 73923352166 : unsigned long a,b,mask;
224 :
225 73923352166 : for (;;) {
226 73923352166 : a = read_word_at_a_time(cs);
227 73923352166 : b = load_unaligned_zeropad(ct);
228 73964219869 : if (tcount < sizeof(unsigned long))
229 : break;
230 2385374348 : if (unlikely(a != b))
231 : return 1;
232 2385374203 : cs += sizeof(unsigned long);
233 2385374203 : ct += sizeof(unsigned long);
234 2385374203 : tcount -= sizeof(unsigned long);
235 2385374203 : if (!tcount)
236 : return 0;
237 : }
238 71578845521 : mask = bytemask_from_count(tcount);
239 71578845521 : 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 72156301355 : const unsigned char *cs = READ_ONCE(dentry->d_name.name);
277 :
278 72156301355 : 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 358680853 : return container_of(dentry->d_name.name, struct external_name, name[0]);
292 : }
293 :
294 1392591767 : static void __d_free(struct rcu_head *head)
295 : {
296 1433685783 : struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
297 :
298 1392591767 : kmem_cache_free(dentry_cache, dentry);
299 41100122 : }
300 :
301 357389634 : static void __d_free_external(struct rcu_head *head)
302 : {
303 357389634 : struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
304 357389634 : kfree(external_name(dentry));
305 357071349 : kmem_cache_free(dentry_cache, dentry);
306 357605968 : }
307 :
308 : static inline int dname_external(const struct dentry *dentry)
309 : {
310 1970935532 : return dentry->d_name.name != dentry->d_iname;
311 : }
312 :
313 55222206 : void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
314 : {
315 55222206 : spin_lock(&dentry->d_lock);
316 55223037 : name->name = dentry->d_name;
317 55223037 : if (unlikely(dname_external(dentry))) {
318 1292137 : atomic_inc(&external_name(dentry)->u.count);
319 : } else {
320 107861800 : memcpy(name->inline_name, dentry->d_iname,
321 : dentry->d_name.len + 1);
322 53930900 : name->name.name = name->inline_name;
323 : }
324 55223353 : spin_unlock(&dentry->d_lock);
325 55222858 : }
326 : EXPORT_SYMBOL(take_dentry_name_snapshot);
327 :
328 55224093 : void release_dentry_name_snapshot(struct name_snapshot *name)
329 : {
330 55224093 : if (unlikely(name->name.name != name->inline_name)) {
331 1292912 : struct external_name *p;
332 1292912 : p = container_of(name->name.name, struct external_name, name[0]);
333 1292912 : if (unlikely(atomic_dec_and_test(&p->u.count)))
334 1291203 : kfree_rcu(p, u.head);
335 : }
336 55224041 : }
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 749299390 : unsigned flags;
344 :
345 749299390 : dentry->d_inode = inode;
346 749299390 : flags = READ_ONCE(dentry->d_flags);
347 749299390 : flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
348 749299390 : flags |= type_flags;
349 1498580230 : smp_store_release(&dentry->d_flags, flags);
350 : }
351 :
352 : static inline void __d_clear_type_and_inode(struct dentry *dentry)
353 : {
354 746774325 : unsigned flags = READ_ONCE(dentry->d_flags);
355 :
356 746774325 : flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
357 746774325 : WRITE_ONCE(dentry->d_flags, flags);
358 746774325 : dentry->d_inode = NULL;
359 746774325 : if (dentry->d_flags & DCACHE_LRU_LIST)
360 62194767 : this_cpu_inc(nr_dentry_negative);
361 : }
362 :
363 1801383507 : static void dentry_free(struct dentry *dentry)
364 : {
365 1801383507 : WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
366 1801383507 : if (unlikely(dname_external(dentry))) {
367 360431225 : struct external_name *p = external_name(dentry);
368 360431225 : if (likely(atomic_dec_and_test(&p->u.count))) {
369 359559006 : call_rcu(&dentry->d_u.d_rcu, __d_free_external);
370 359559006 : return;
371 : }
372 : }
373 : /* if dentry was never visible to RCU, immediate free is OK */
374 1442164845 : if (dentry->d_flags & DCACHE_NORCU)
375 41094016 : __d_free(&dentry->d_u.d_rcu);
376 : else
377 1401070829 : call_rcu(&dentry->d_u.d_rcu, __d_free);
378 : }
379 :
380 : /*
381 : * Release the dentry's inode, using the filesystem
382 : * d_iput() operation if defined.
383 : */
384 748394447 : static void dentry_unlink_inode(struct dentry * dentry)
385 : __releases(dentry->d_lock)
386 : __releases(dentry->d_inode->i_lock)
387 : {
388 748394447 : struct inode *inode = dentry->d_inode;
389 :
390 748394447 : raw_write_seqcount_begin(&dentry->d_seq);
391 746774325 : __d_clear_type_and_inode(dentry);
392 746965795 : hlist_del_init(&dentry->d_u.d_alias);
393 746965795 : raw_write_seqcount_end(&dentry->d_seq);
394 746528163 : spin_unlock(&dentry->d_lock);
395 751422609 : spin_unlock(&inode->i_lock);
396 751749785 : if (!inode->i_nlink)
397 434985740 : fsnotify_inoderemove(inode);
398 750558871 : if (dentry->d_op && dentry->d_op->d_iput)
399 838728 : dentry->d_op->d_iput(dentry, inode);
400 : else
401 749720143 : iput(inode);
402 749915640 : }
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 193625305 : static void d_lru_add(struct dentry *dentry)
425 : {
426 193625305 : D_FLAG_VERIFY(dentry, 0);
427 193625305 : dentry->d_flags |= DCACHE_LRU_LIST;
428 193625305 : this_cpu_inc(nr_dentry_unused);
429 193453495 : if (d_is_negative(dentry))
430 42310206 : this_cpu_inc(nr_dentry_negative);
431 193451725 : WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
432 194593248 : }
433 :
434 149338295 : static void d_lru_del(struct dentry *dentry)
435 : {
436 149338295 : D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
437 149338295 : dentry->d_flags &= ~DCACHE_LRU_LIST;
438 149338295 : this_cpu_dec(nr_dentry_unused);
439 149338188 : if (d_is_negative(dentry))
440 73456273 : this_cpu_dec(nr_dentry_negative);
441 149338163 : WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
442 149339262 : }
443 :
444 201682719 : static void d_shrink_del(struct dentry *dentry)
445 : {
446 201682719 : D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
447 201682719 : list_del_init(&dentry->d_lru);
448 201681118 : dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
449 201681118 : this_cpu_dec(nr_dentry_unused);
450 201681135 : }
451 :
452 158054741 : static void d_shrink_add(struct dentry *dentry, struct list_head *list)
453 : {
454 158054741 : D_FLAG_VERIFY(dentry, 0);
455 158054741 : list_add(&dentry->d_lru, list);
456 158054652 : dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
457 158054652 : this_cpu_inc(nr_dentry_unused);
458 158054704 : }
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 1798348 : static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
467 : {
468 1798348 : D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
469 1798348 : dentry->d_flags &= ~DCACHE_LRU_LIST;
470 1798348 : this_cpu_dec(nr_dentry_unused);
471 1798348 : if (d_is_negative(dentry))
472 0 : this_cpu_dec(nr_dentry_negative);
473 1798348 : list_lru_isolate(lru, &dentry->d_lru);
474 1798348 : }
475 :
476 43629446 : static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
477 : struct list_head *list)
478 : {
479 43629446 : D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
480 43629446 : dentry->d_flags |= DCACHE_SHRINK_LIST;
481 43629446 : if (d_is_negative(dentry))
482 4110886 : this_cpu_dec(nr_dentry_negative);
483 43629446 : list_lru_isolate_move(lru, &dentry->d_lru, list);
484 43629446 : }
485 :
486 1313312039 : static void ___d_drop(struct dentry *dentry)
487 : {
488 1313312039 : 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 1313312039 : if (unlikely(IS_ROOT(dentry)))
495 0 : b = &dentry->d_sb->s_roots;
496 : else
497 1313312039 : b = d_hash(dentry->d_name.hash);
498 :
499 1313301787 : hlist_bl_lock(b);
500 1313309584 : __hlist_bl_del(&dentry->d_hash);
501 1313312009 : hlist_bl_unlock(b);
502 1313304057 : }
503 :
504 1818159518 : void __d_drop(struct dentry *dentry)
505 : {
506 1818159518 : if (!d_unhashed(dentry)) {
507 1204228178 : ___d_drop(dentry);
508 1204216011 : dentry->d_hash.pprev = NULL;
509 1204216011 : write_seqcount_invalidate(&dentry->d_seq);
510 : }
511 1818146975 : }
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 211429 : void d_drop(struct dentry *dentry)
533 : {
534 211429 : spin_lock(&dentry->d_lock);
535 211427 : __d_drop(dentry);
536 211427 : spin_unlock(&dentry->d_lock);
537 211428 : }
538 : EXPORT_SYMBOL(d_drop);
539 :
540 1798553990 : static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
541 : {
542 1798553990 : 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 1798553990 : dentry->d_flags |= DCACHE_DENTRY_KILLED;
548 1798553990 : if (unlikely(list_empty(&dentry->d_child)))
549 : return;
550 1267919936 : __list_del_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 1267919936 : while (dentry->d_child.next != &parent->d_subdirs) {
571 1188592722 : next = list_entry(dentry->d_child.next, struct dentry, d_child);
572 1188592722 : if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
573 : break;
574 0 : dentry->d_child.next = next->d_child.next;
575 : }
576 : }
577 :
578 1802663581 : static void __dentry_kill(struct dentry *dentry)
579 : {
580 1802663581 : struct dentry *parent = NULL;
581 1802663581 : bool can_free = true;
582 1802663581 : if (!IS_ROOT(dentry))
583 1276271738 : parent = dentry->d_parent;
584 :
585 : /*
586 : * The dentry is now unrecoverably dead to the world.
587 : */
588 1802663581 : 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 1800313421 : if (dentry->d_flags & DCACHE_OP_PRUNE)
595 6811 : dentry->d_op->d_prune(dentry);
596 :
597 1800313420 : if (dentry->d_flags & DCACHE_LRU_LIST) {
598 1993859 : if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
599 1993849 : d_lru_del(dentry);
600 : }
601 : /* if it was on the hash then remove it */
602 1800313469 : __d_drop(dentry);
603 1799096300 : dentry_unlist(dentry, parent);
604 1798781211 : if (parent)
605 1276244539 : spin_unlock(&parent->d_lock);
606 1798812196 : if (dentry->d_inode)
607 675467602 : dentry_unlink_inode(dentry);
608 : else
609 1123344594 : spin_unlock(&dentry->d_lock);
610 1798451870 : this_cpu_dec(nr_dentry);
611 1798633450 : if (dentry->d_op && dentry->d_op->d_release)
612 3458164 : dentry->d_op->d_release(dentry);
613 :
614 1798632944 : spin_lock(&dentry->d_lock);
615 1801387701 : if (dentry->d_flags & DCACHE_SHRINK_LIST) {
616 0 : dentry->d_flags |= DCACHE_MAY_FREE;
617 0 : can_free = false;
618 : }
619 1801387701 : spin_unlock(&dentry->d_lock);
620 1800916262 : if (likely(can_free))
621 1800916262 : dentry_free(dentry);
622 1800643807 : cond_resched();
623 1800063929 : }
624 :
625 3352936 : static struct dentry *__lock_parent(struct dentry *dentry)
626 : {
627 3352936 : struct dentry *parent;
628 3352936 : rcu_read_lock();
629 3353413 : spin_unlock(&dentry->d_lock);
630 3364372 : again:
631 3364372 : parent = READ_ONCE(dentry->d_parent);
632 3364372 : spin_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 3365170 : if (unlikely(parent != dentry->d_parent)) {
642 0 : spin_unlock(&parent->d_lock);
643 0 : goto again;
644 : }
645 3365170 : rcu_read_unlock();
646 3365170 : if (parent != dentry)
647 3365170 : spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
648 : else
649 : parent = NULL;
650 3365170 : return parent;
651 : }
652 :
653 16216 : static inline struct dentry *lock_parent(struct dentry *dentry)
654 : {
655 16216 : struct dentry *parent = dentry->d_parent;
656 16216 : if (IS_ROOT(dentry))
657 : return NULL;
658 6 : if (likely(spin_trylock(&parent->d_lock)))
659 : return parent;
660 0 : return __lock_parent(dentry);
661 : }
662 :
663 4639070594 : static inline bool retain_dentry(struct dentry *dentry)
664 : {
665 4639070594 : WARN_ON(d_in_lookup(dentry));
666 :
667 : /* Unreachable? Get rid of it */
668 4639070594 : if (unlikely(d_unhashed(dentry)))
669 : return false;
670 :
671 4025607138 : if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
672 : return false;
673 :
674 4025607138 : if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
675 1273749382 : if (dentry->d_op->d_delete(dentry))
676 : return false;
677 : }
678 :
679 3036553058 : if (unlikely(dentry->d_flags & DCACHE_DONTCACHE))
680 : return false;
681 :
682 : /* retain; LRU fodder */
683 3036553034 : dentry->d_lockref.count--;
684 3036553034 : if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
685 193811466 : d_lru_add(dentry);
686 2842741568 : else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
687 163706761 : dentry->d_flags |= DCACHE_REFERENCED;
688 : return true;
689 : }
690 :
691 391386595 : void d_mark_dontcache(struct inode *inode)
692 : {
693 391386595 : struct dentry *de;
694 :
695 391386595 : spin_lock(&inode->i_lock);
696 782837222 : hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
697 372 : spin_lock(&de->d_lock);
698 24 : de->d_flags |= DCACHE_DONTCACHE;
699 24 : spin_unlock(&de->d_lock);
700 : }
701 391418263 : inode->i_state |= I_DONTCACHE;
702 391418263 : spin_unlock(&inode->i_lock);
703 391430483 : }
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 1599563004 : static struct dentry *dentry_kill(struct dentry *dentry)
712 : __releases(dentry->d_lock)
713 : {
714 1599563004 : struct inode *inode = dentry->d_inode;
715 1599563004 : struct dentry *parent = NULL;
716 :
717 2154509787 : if (inode && unlikely(!spin_trylock(&inode->i_lock)))
718 15841 : goto slow_positive;
719 :
720 1601927523 : if (!IS_ROOT(dentry)) {
721 1074824349 : parent = dentry->d_parent;
722 1074824349 : if (unlikely(!spin_trylock(&parent->d_lock))) {
723 3358061 : parent = __lock_parent(dentry);
724 3365170 : if (likely(inode || !dentry->d_inode))
725 3365170 : goto got_locks;
726 : /* negative that became positive */
727 0 : if (parent)
728 0 : spin_unlock(&parent->d_lock);
729 0 : inode = dentry->d_inode;
730 0 : goto slow_positive;
731 : }
732 : }
733 1598833627 : __dentry_kill(dentry);
734 1598833627 : return parent;
735 :
736 15841 : slow_positive:
737 15841 : spin_unlock(&dentry->d_lock);
738 15753 : spin_lock(&inode->i_lock);
739 16216 : spin_lock(&dentry->d_lock);
740 16216 : parent = lock_parent(dentry);
741 3381386 : got_locks:
742 3381386 : if (unlikely(dentry->d_lockref.count != 1)) {
743 382332 : dentry->d_lockref.count--;
744 2999054 : } else if (likely(!retain_dentry(dentry))) {
745 2999054 : __dentry_kill(dentry);
746 2999054 : return parent;
747 : }
748 : /* we are keeping it, after all */
749 382332 : if (inode)
750 0 : spin_unlock(&inode->i_lock);
751 382332 : if (parent)
752 382332 : spin_unlock(&parent->d_lock);
753 382332 : spin_unlock(&dentry->d_lock);
754 382332 : 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 21360903326 : static inline bool fast_dput(struct dentry *dentry)
766 : {
767 21360903326 : int ret;
768 21360903326 : 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 21360903326 : if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
775 9593339210 : return lockref_put_or_lock(&dentry->d_lockref);
776 :
777 : /*
778 : * .. otherwise, we can try to just decrement the
779 : * lockref optimistically.
780 : */
781 11767564116 : 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 11767738988 : if (unlikely(ret < 0)) {
789 11767738988 : spin_lock(&dentry->d_lock);
790 11780586952 : if (dentry->d_lockref.count > 1) {
791 8428955743 : dentry->d_lockref.count--;
792 8428955743 : spin_unlock(&dentry->d_lock);
793 8428955743 : return true;
794 : }
795 : return false;
796 : }
797 :
798 : /*
799 : * If we weren't the last ref, we're done.
800 : */
801 0 : 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 0 : smp_rmb();
832 0 : d_flags = READ_ONCE(dentry->d_flags);
833 0 : 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 0 : 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 0 : spin_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 0 : if (dentry->d_lockref.count) {
854 0 : spin_unlock(&dentry->d_lock);
855 0 : 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 0 : dentry->d_lockref.count = 1;
864 0 : 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 29156392559 : void dput(struct dentry *dentry)
895 : {
896 30754859258 : while (dentry) {
897 21342719625 : might_sleep();
898 :
899 21354514745 : rcu_read_lock();
900 21348592184 : if (likely(fast_dput(dentry))) {
901 16732342597 : rcu_read_unlock();
902 16732342597 : return;
903 : }
904 :
905 : /* Slow case: now with the dentry lock held */
906 4634065218 : rcu_read_unlock();
907 :
908 4634246134 : if (likely(retain_dentry(dentry))) {
909 3036581096 : spin_unlock(&dentry->d_lock);
910 3036581096 : return;
911 : }
912 :
913 1599649090 : dentry = dentry_kill(dentry);
914 : }
915 : }
916 : EXPORT_SYMBOL(dput);
917 :
918 201534997 : static void __dput_to_list(struct dentry *dentry, struct list_head *list)
919 : __must_hold(&dentry->d_lock)
920 : {
921 201534997 : if (dentry->d_flags & DCACHE_SHRINK_LIST) {
922 : /* let the owner of the list it's on deal with it */
923 40733 : --dentry->d_lockref.count;
924 : } else {
925 201494264 : if (dentry->d_flags & DCACHE_LRU_LIST)
926 728486 : d_lru_del(dentry);
927 201494499 : if (!--dentry->d_lockref.count)
928 20240016 : d_shrink_add(dentry, list);
929 : }
930 201535233 : }
931 :
932 389935 : void dput_to_list(struct dentry *dentry, struct list_head *list)
933 : {
934 389935 : rcu_read_lock();
935 389935 : if (likely(fast_dput(dentry))) {
936 289702 : rcu_read_unlock();
937 289702 : return;
938 : }
939 100233 : rcu_read_unlock();
940 100233 : if (!retain_dentry(dentry))
941 0 : __dput_to_list(dentry, list);
942 100233 : spin_unlock(&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 1264552185 : dentry->d_lockref.count++;
949 : }
950 :
951 : static inline void __dget(struct dentry *dentry)
952 : {
953 236377129 : lockref_get(&dentry->d_lockref);
954 : }
955 :
956 1011815410 : struct dentry *dget_parent(struct dentry *dentry)
957 : {
958 1011815410 : int gotref;
959 1011815410 : struct dentry *ret;
960 1011815410 : unsigned seq;
961 :
962 : /*
963 : * Do optimistic parent lookup without any
964 : * locking.
965 : */
966 1011815410 : rcu_read_lock();
967 1011170440 : seq = raw_seqcount_begin(&dentry->d_seq);
968 1011168453 : ret = READ_ONCE(dentry->d_parent);
969 1011168453 : gotref = lockref_get_not_zero(&ret->d_lockref);
970 1012779208 : rcu_read_unlock();
971 1012780128 : if (likely(gotref)) {
972 1012780128 : if (!read_seqcount_retry(&dentry->d_seq, seq))
973 : return ret;
974 5 : dput(ret);
975 : }
976 :
977 0 : repeat:
978 : /*
979 : * Don't need rcu_dereference because we re-check it was correct under
980 : * the lock.
981 : */
982 5 : rcu_read_lock();
983 5 : ret = dentry->d_parent;
984 5 : spin_lock(&ret->d_lock);
985 5 : if (unlikely(ret != dentry->d_parent)) {
986 0 : spin_unlock(&ret->d_lock);
987 0 : rcu_read_unlock();
988 0 : goto repeat;
989 : }
990 5 : rcu_read_unlock();
991 5 : BUG_ON(!ret->d_lockref.count);
992 5 : ret->d_lockref.count++;
993 5 : spin_unlock(&ret->d_lock);
994 5 : return ret;
995 : }
996 : EXPORT_SYMBOL(dget_parent);
997 :
998 : static struct dentry * __d_find_any_alias(struct inode *inode)
999 : {
1000 523302666 : struct dentry *alias;
1001 :
1002 523302666 : if (hlist_empty(&inode->i_dentry))
1003 : return NULL;
1004 236377129 : alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1005 118 : __dget(alias);
1006 236377011 : 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 350208763 : struct dentry *d_find_any_alias(struct inode *inode)
1017 : {
1018 350208763 : struct dentry *de;
1019 :
1020 350208763 : spin_lock(&inode->i_lock);
1021 360361328 : de = __d_find_any_alias(inode);
1022 359630805 : spin_unlock(&inode->i_lock);
1023 358553330 : return de;
1024 : }
1025 : EXPORT_SYMBOL(d_find_any_alias);
1026 :
1027 5256713 : static struct dentry *__d_find_alias(struct inode *inode)
1028 : {
1029 5256713 : struct dentry *alias;
1030 :
1031 5256713 : if (S_ISDIR(inode->i_mode))
1032 10390329 : return __d_find_any_alias(inode);
1033 :
1034 123098 : hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1035 61549 : spin_lock(&alias->d_lock);
1036 61550 : if (!d_unhashed(alias)) {
1037 61550 : __dget_dlock(alias);
1038 61550 : spin_unlock(&alias->d_lock);
1039 61550 : return alias;
1040 : }
1041 0 : spin_unlock(&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 5256178 : struct dentry *d_find_alias(struct inode *inode)
1061 : {
1062 5256178 : struct dentry *de = NULL;
1063 :
1064 5256178 : if (!hlist_empty(&inode->i_dentry)) {
1065 5255467 : spin_lock(&inode->i_lock);
1066 5256721 : de = __d_find_alias(inode);
1067 5256718 : spin_unlock(&inode->i_lock);
1068 : }
1069 5257436 : 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 0 : struct dentry *d_find_alias_rcu(struct inode *inode)
1078 : {
1079 0 : struct hlist_head *l = &inode->i_dentry;
1080 0 : struct dentry *de = NULL;
1081 :
1082 0 : spin_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 0 : if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) {
1086 0 : if (S_ISDIR(inode->i_mode)) {
1087 0 : de = hlist_entry(l->first, struct dentry, d_u.d_alias);
1088 : } else {
1089 0 : hlist_for_each_entry(de, l, d_u.d_alias)
1090 0 : if (!d_unhashed(de))
1091 : break;
1092 : }
1093 : }
1094 0 : spin_unlock(&inode->i_lock);
1095 0 : 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 170 : void d_prune_aliases(struct inode *inode)
1103 : {
1104 170 : struct dentry *dentry;
1105 170 : restart:
1106 170 : spin_lock(&inode->i_lock);
1107 510 : hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1108 170 : spin_lock(&dentry->d_lock);
1109 170 : if (!dentry->d_lockref.count) {
1110 0 : struct dentry *parent = lock_parent(dentry);
1111 0 : if (likely(!dentry->d_lockref.count)) {
1112 0 : __dentry_kill(dentry);
1113 0 : dput(parent);
1114 0 : goto restart;
1115 : }
1116 0 : if (parent)
1117 0 : spin_unlock(&parent->d_lock);
1118 : }
1119 170 : spin_unlock(&dentry->d_lock);
1120 : }
1121 170 : spin_unlock(&inode->i_lock);
1122 170 : }
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 201683003 : static bool shrink_lock_dentry(struct dentry *dentry)
1137 : {
1138 201683003 : struct inode *inode;
1139 201683003 : struct dentry *parent;
1140 :
1141 201683003 : if (dentry->d_lockref.count)
1142 : return false;
1143 :
1144 201536737 : inode = dentry->d_inode;
1145 326612912 : if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1146 36 : spin_unlock(&dentry->d_lock);
1147 36 : spin_lock(&inode->i_lock);
1148 36 : spin_lock(&dentry->d_lock);
1149 36 : if (unlikely(dentry->d_lockref.count))
1150 0 : goto out;
1151 : /* changed inode means that somebody had grabbed it */
1152 36 : if (unlikely(inode != dentry->d_inode))
1153 0 : goto out;
1154 : }
1155 :
1156 201537008 : parent = dentry->d_parent;
1157 403074326 : if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1158 201536231 : return true;
1159 :
1160 1087 : spin_unlock(&dentry->d_lock);
1161 1087 : spin_lock(&parent->d_lock);
1162 1087 : if (unlikely(parent != dentry->d_parent)) {
1163 0 : spin_unlock(&parent->d_lock);
1164 0 : spin_lock(&dentry->d_lock);
1165 0 : goto out;
1166 : }
1167 1087 : spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1168 1087 : if (likely(!dentry->d_lockref.count))
1169 : return true;
1170 0 : spin_unlock(&parent->d_lock);
1171 0 : out:
1172 0 : if (inode)
1173 0 : spin_unlock(&inode->i_lock);
1174 : return false;
1175 : }
1176 :
1177 17601894 : void shrink_dentry_list(struct list_head *list)
1178 : {
1179 219281460 : while (!list_empty(list)) {
1180 201683893 : struct dentry *dentry, *parent;
1181 :
1182 201683893 : dentry = list_entry(list->prev, struct dentry, d_lru);
1183 201683893 : spin_lock(&dentry->d_lock);
1184 201683569 : rcu_read_lock();
1185 201683526 : if (!shrink_lock_dentry(dentry)) {
1186 146266 : bool can_free = false;
1187 146266 : rcu_read_unlock();
1188 146266 : d_shrink_del(dentry);
1189 146266 : if (dentry->d_lockref.count < 0)
1190 0 : can_free = dentry->d_flags & DCACHE_MAY_FREE;
1191 146266 : spin_unlock(&dentry->d_lock);
1192 146266 : if (can_free)
1193 0 : dentry_free(dentry);
1194 146266 : continue;
1195 : }
1196 201537199 : rcu_read_unlock();
1197 201537185 : d_shrink_del(dentry);
1198 201535202 : parent = dentry->d_parent;
1199 201535202 : if (parent != dentry)
1200 201535202 : __dput_to_list(parent, list);
1201 201535096 : __dentry_kill(dentry);
1202 : }
1203 17597567 : }
1204 :
1205 99561252 : 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 99561252 : struct list_head *freeable = arg;
1209 99561252 : 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 99561252 : if (!spin_trylock(&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 99561251 : if (dentry->d_lockref.count) {
1226 1798348 : d_lru_isolate(lru, dentry);
1227 1798348 : spin_unlock(&dentry->d_lock);
1228 1798348 : return LRU_REMOVED;
1229 : }
1230 :
1231 97762903 : if (dentry->d_flags & DCACHE_REFERENCED) {
1232 55149261 : dentry->d_flags &= ~DCACHE_REFERENCED;
1233 55149261 : spin_unlock(&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 55149261 : return LRU_ROTATE;
1255 : }
1256 :
1257 42613642 : d_lru_shrink_move(lru, dentry, freeable);
1258 42613642 : spin_unlock(&dentry->d_lock);
1259 :
1260 42613642 : 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 383258 : long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1276 : {
1277 383258 : LIST_HEAD(dispose);
1278 383258 : long freed;
1279 :
1280 383258 : freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1281 : dentry_lru_isolate, &dispose);
1282 383258 : shrink_dentry_list(&dispose);
1283 383258 : return freed;
1284 : }
1285 :
1286 1015805 : 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 1015805 : struct list_head *freeable = arg;
1290 1015805 : 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 1015805 : if (!spin_trylock(&dentry->d_lock))
1298 : return LRU_SKIP;
1299 :
1300 1015804 : d_lru_shrink_move(lru, dentry, freeable);
1301 1015804 : spin_unlock(&dentry->d_lock);
1302 :
1303 1015804 : 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 13293 : void shrink_dcache_sb(struct super_block *sb)
1315 : {
1316 14089 : do {
1317 14089 : LIST_HEAD(dispose);
1318 :
1319 14089 : list_lru_walk(&sb->s_dentry_lru,
1320 : dentry_lru_isolate_shrink, &dispose, 1024);
1321 14089 : shrink_dentry_list(&dispose);
1322 14089 : } while (list_lru_count(&sb->s_dentry_lru) > 0);
1323 13293 : }
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 45444514 : static void d_walk(struct dentry *parent, void *data,
1349 : enum d_walk_ret (*enter)(void *, struct dentry *))
1350 : {
1351 45444514 : struct dentry *this_parent;
1352 45444514 : struct list_head *next;
1353 45444514 : unsigned seq = 0;
1354 45444514 : enum d_walk_ret ret;
1355 45444514 : bool retry = true;
1356 :
1357 45444930 : again:
1358 45444930 : read_seqbegin_or_lock(&rename_lock, &seq);
1359 45444357 : this_parent = parent;
1360 45444357 : spin_lock(&this_parent->d_lock);
1361 :
1362 45451323 : ret = enter(data, this_parent);
1363 45444238 : switch (ret) {
1364 : case D_WALK_CONTINUE:
1365 : break;
1366 0 : case D_WALK_QUIT:
1367 : case D_WALK_SKIP:
1368 0 : goto out_unlock;
1369 0 : case D_WALK_NORETRY:
1370 0 : retry = false;
1371 0 : break;
1372 : }
1373 : repeat:
1374 56283932 : next = this_parent->d_subdirs.next;
1375 : resume:
1376 202408769 : while (next != &this_parent->d_subdirs) {
1377 148659537 : struct list_head *tmp = next;
1378 148659537 : struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1379 148659537 : next = tmp->next;
1380 :
1381 148659537 : if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1382 0 : continue;
1383 :
1384 148659537 : spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1385 :
1386 148660408 : ret = enter(data, dentry);
1387 148660537 : switch (ret) {
1388 : case D_WALK_CONTINUE:
1389 : break;
1390 32480 : case D_WALK_QUIT:
1391 32480 : spin_unlock(&dentry->d_lock);
1392 32480 : goto out_unlock;
1393 148379233 : case D_WALK_NORETRY:
1394 148379233 : retry = false;
1395 148379233 : break;
1396 0 : case D_WALK_SKIP:
1397 0 : spin_unlock(&dentry->d_lock);
1398 0 : continue;
1399 : }
1400 :
1401 148628057 : if (!list_empty(&dentry->d_subdirs)) {
1402 10839694 : spin_unlock(&this_parent->d_lock);
1403 10839694 : spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1404 10839694 : this_parent = dentry;
1405 10839694 : spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1406 10839694 : goto repeat;
1407 : }
1408 137788363 : spin_unlock(&dentry->d_lock);
1409 : }
1410 : /*
1411 : * All done at this level ... ascend and resume the search.
1412 : */
1413 53749232 : rcu_read_lock();
1414 56181642 : ascend:
1415 56181642 : if (this_parent != parent) {
1416 10770189 : struct dentry *child = this_parent;
1417 10770189 : this_parent = child->d_parent;
1418 :
1419 10770189 : spin_unlock(&child->d_lock);
1420 10769723 : spin_lock(&this_parent->d_lock);
1421 :
1422 : /* might go back up the wrong parent if we have had a rename. */
1423 10769723 : if (need_seqretry(&rename_lock, seq))
1424 95 : goto rename_retry;
1425 : /* go into the first sibling still alive */
1426 10769628 : do {
1427 10769628 : next = child->d_child.next;
1428 10769628 : if (next == &this_parent->d_subdirs)
1429 2433382 : goto ascend;
1430 8336246 : child = list_entry(next, struct dentry, d_child);
1431 8336246 : } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1432 8336246 : rcu_read_unlock();
1433 8336246 : goto resume;
1434 : }
1435 45411453 : if (need_seqretry(&rename_lock, seq))
1436 2254 : goto rename_retry;
1437 45409729 : rcu_read_unlock();
1438 :
1439 45442859 : out_unlock:
1440 45442859 : spin_unlock(&this_parent->d_lock);
1441 45449515 : done_seqretry(&rename_lock, seq);
1442 : return;
1443 :
1444 2349 : rename_retry:
1445 2349 : spin_unlock(&this_parent->d_lock);
1446 2349 : rcu_read_unlock();
1447 2349 : BUG_ON(seq & 1);
1448 2349 : if (!retry)
1449 : return;
1450 416 : seq = 1;
1451 416 : goto again;
1452 : }
1453 :
1454 : struct check_mount {
1455 : struct vfsmount *mnt;
1456 : unsigned int mounted;
1457 : };
1458 :
1459 0 : static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1460 : {
1461 0 : struct check_mount *info = data;
1462 0 : struct path path = { .mnt = info->mnt, .dentry = dentry };
1463 :
1464 0 : if (likely(!d_mountpoint(dentry)))
1465 : return D_WALK_CONTINUE;
1466 0 : if (__path_is_mountpoint(&path)) {
1467 0 : info->mounted = 1;
1468 0 : 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 0 : int path_has_submounts(const struct path *parent)
1482 : {
1483 0 : struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1484 :
1485 0 : read_seqlock_excl(&mount_lock);
1486 0 : d_walk(parent->dentry, &data, path_check_mount);
1487 0 : read_sequnlock_excl(&mount_lock);
1488 :
1489 0 : 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 389935 : int d_set_mounted(struct dentry *dentry)
1502 : {
1503 389935 : struct dentry *p;
1504 389935 : int ret = -ENOENT;
1505 389935 : write_seqlock(&rename_lock);
1506 392270 : for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1507 : /* Need exclusion wrt. d_invalidate() */
1508 2335 : spin_lock(&p->d_lock);
1509 2335 : if (unlikely(d_unhashed(p))) {
1510 0 : spin_unlock(&p->d_lock);
1511 0 : goto out;
1512 : }
1513 2335 : spin_unlock(&p->d_lock);
1514 : }
1515 389935 : spin_lock(&dentry->d_lock);
1516 389935 : if (!d_unlinked(dentry)) {
1517 389935 : ret = -EBUSY;
1518 389935 : if (!d_mountpoint(dentry)) {
1519 389935 : dentry->d_flags |= DCACHE_MOUNTED;
1520 389935 : ret = 0;
1521 : }
1522 : }
1523 389935 : spin_unlock(&dentry->d_lock);
1524 389935 : out:
1525 389935 : write_sequnlock(&rename_lock);
1526 389935 : 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 178788295 : static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1554 : {
1555 178788295 : struct select_data *data = _data;
1556 178788295 : enum d_walk_ret ret = D_WALK_CONTINUE;
1557 :
1558 178788295 : if (data->start == dentry)
1559 30128606 : goto out;
1560 :
1561 148659689 : if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1562 0 : data->found++;
1563 : } else {
1564 148659689 : if (dentry->d_flags & DCACHE_LRU_LIST)
1565 146616356 : d_lru_del(dentry);
1566 148660050 : if (!dentry->d_lockref.count) {
1567 137814795 : d_shrink_add(dentry, &data->dispose);
1568 137814598 : 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 148659853 : if (!list_empty(&data->dispose))
1577 148411683 : ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1578 248170 : out:
1579 178788459 : return ret;
1580 : }
1581 :
1582 0 : static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1583 : {
1584 0 : struct select_data *data = _data;
1585 0 : enum d_walk_ret ret = D_WALK_CONTINUE;
1586 :
1587 0 : if (data->start == dentry)
1588 0 : goto out;
1589 :
1590 0 : if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1591 0 : if (!dentry->d_lockref.count) {
1592 0 : rcu_read_lock();
1593 0 : data->victim = dentry;
1594 0 : return D_WALK_QUIT;
1595 : }
1596 : } else {
1597 0 : if (dentry->d_flags & DCACHE_LRU_LIST)
1598 0 : d_lru_del(dentry);
1599 0 : if (!dentry->d_lockref.count)
1600 0 : d_shrink_add(dentry, &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 0 : if (!list_empty(&data->dispose))
1608 0 : ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1609 0 : 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 17038747 : void shrink_dcache_parent(struct dentry *parent)
1620 : {
1621 30128349 : for (;;) {
1622 30128349 : struct select_data data = {.start = parent};
1623 :
1624 30128349 : INIT_LIST_HEAD(&data.dispose);
1625 30128349 : d_walk(parent, &data, select_collect);
1626 :
1627 30128879 : if (!list_empty(&data.dispose)) {
1628 13092990 : shrink_dentry_list(&data.dispose);
1629 13089602 : continue;
1630 : }
1631 :
1632 17035889 : cond_resched();
1633 17034123 : if (!data.found)
1634 : break;
1635 0 : data.victim = NULL;
1636 0 : d_walk(parent, &data, select_collect2);
1637 0 : if (data.victim) {
1638 0 : struct dentry *parent;
1639 0 : spin_lock(&data.victim->d_lock);
1640 0 : if (!shrink_lock_dentry(data.victim)) {
1641 0 : spin_unlock(&data.victim->d_lock);
1642 0 : rcu_read_unlock();
1643 : } else {
1644 0 : rcu_read_unlock();
1645 0 : parent = data.victim->d_parent;
1646 0 : if (parent != data.victim)
1647 0 : __dput_to_list(parent, &data.dispose);
1648 0 : __dentry_kill(data.victim);
1649 : }
1650 : }
1651 0 : if (!list_empty(&data.dispose))
1652 0 : shrink_dentry_list(&data.dispose);
1653 : }
1654 17034123 : }
1655 : EXPORT_SYMBOL(shrink_dcache_parent);
1656 :
1657 124613 : static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1658 : {
1659 : /* it has busy descendents; complain about those instead */
1660 124613 : if (!list_empty(&dentry->d_subdirs))
1661 : return D_WALK_CONTINUE;
1662 :
1663 : /* root with refcount 1 is fine */
1664 124612 : if (dentry == _data && dentry->d_lockref.count == 1)
1665 : return D_WALK_CONTINUE;
1666 :
1667 0 : printk(KERN_ERR "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 0 : WARN_ON(1);
1677 : return D_WALK_CONTINUE;
1678 : }
1679 :
1680 124610 : static void do_one_tree(struct dentry *dentry)
1681 : {
1682 124610 : shrink_dcache_parent(dentry);
1683 124610 : d_walk(dentry, dentry, umount_check);
1684 124609 : d_drop(dentry);
1685 124608 : dput(dentry);
1686 124610 : }
1687 :
1688 : /*
1689 : * destroy the dentries attached to a superblock on unmounting
1690 : */
1691 124609 : void shrink_dcache_for_umount(struct super_block *sb)
1692 : {
1693 124609 : struct dentry *dentry;
1694 :
1695 124609 : WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1696 :
1697 124606 : dentry = sb->s_root;
1698 124606 : sb->s_root = NULL;
1699 124606 : do_one_tree(dentry);
1700 :
1701 124606 : while (!hlist_bl_empty(&sb->s_roots)) {
1702 0 : dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1703 0 : do_one_tree(dentry);
1704 : }
1705 124610 : }
1706 :
1707 15199230 : static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1708 : {
1709 15199230 : struct dentry **victim = _data;
1710 15199230 : if (d_mountpoint(dentry)) {
1711 0 : __dget_dlock(dentry);
1712 0 : *victim = dentry;
1713 0 : return D_WALK_QUIT;
1714 : }
1715 : return D_WALK_CONTINUE;
1716 : }
1717 :
1718 : /**
1719 : * d_invalidate - detach submounts, prune dcache, and drop
1720 : * @dentry: dentry to invalidate (aka detach, prune and drop)
1721 : */
1722 16098846 : void d_invalidate(struct dentry *dentry)
1723 : {
1724 16098846 : bool had_submounts = false;
1725 16098846 : spin_lock(&dentry->d_lock);
1726 16100463 : if (d_unhashed(dentry)) {
1727 8917 : spin_unlock(&dentry->d_lock);
1728 8917 : return;
1729 : }
1730 16091546 : __d_drop(dentry);
1731 16091051 : spin_unlock(&dentry->d_lock);
1732 :
1733 : /* Negative dentries can be dropped without further checks */
1734 16091470 : if (!dentry->d_inode)
1735 : return;
1736 :
1737 15202678 : shrink_dcache_parent(dentry);
1738 0 : for (;;) {
1739 15198479 : struct dentry *victim = NULL;
1740 15198479 : d_walk(dentry, &victim, find_submount);
1741 15196088 : if (!victim) {
1742 15196088 : if (had_submounts)
1743 0 : shrink_dcache_parent(dentry);
1744 15196088 : return;
1745 : }
1746 0 : had_submounts = true;
1747 0 : detach_mounts(victim);
1748 0 : dput(victim);
1749 : }
1750 : }
1751 : EXPORT_SYMBOL(d_invalidate);
1752 :
1753 : /**
1754 : * __d_alloc - allocate a dcache entry
1755 : * @sb: filesystem it will belong to
1756 : * @name: qstr of the name
1757 : *
1758 : * Allocates a dentry. It returns %NULL if there is insufficient memory
1759 : * available. On a success the dentry is returned. The name passed in is
1760 : * copied and the copy passed in may be reused after this call.
1761 : */
1762 :
1763 1796328783 : static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1764 : {
1765 1796328783 : struct dentry *dentry;
1766 1796328783 : char *dname;
1767 1796328783 : int err;
1768 :
1769 1796328783 : dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru,
1770 : GFP_KERNEL);
1771 1796352543 : if (!dentry)
1772 : return NULL;
1773 :
1774 : /*
1775 : * We guarantee that the inline name is always NUL-terminated.
1776 : * This way the memcpy() done by the name switching in rename
1777 : * will still always have a NUL at the end, even if we might
1778 : * be overwriting an internal NUL character
1779 : */
1780 1796352543 : dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1781 1796352543 : if (unlikely(!name)) {
1782 136285197 : name = &slash_name;
1783 136285197 : dname = dentry->d_iname;
1784 1660067346 : } else if (name->len > DNAME_INLINE_LEN-1) {
1785 360367800 : size_t size = offsetof(struct external_name, name[1]);
1786 360367800 : struct external_name *p = kmalloc(size + name->len,
1787 : GFP_KERNEL_ACCOUNT |
1788 : __GFP_RECLAIMABLE);
1789 360870830 : if (!p) {
1790 0 : kmem_cache_free(dentry_cache, dentry);
1791 0 : return NULL;
1792 : }
1793 360870830 : atomic_set(&p->u.count, 1);
1794 360870830 : dname = p->name;
1795 : } else {
1796 1299699546 : dname = dentry->d_iname;
1797 : }
1798 :
1799 1796855573 : dentry->d_name.len = name->len;
1800 1796855573 : dentry->d_name.hash = name->hash;
1801 3593711146 : memcpy(dname, name->name, name->len);
1802 1796855573 : dname[name->len] = 0;
1803 :
1804 : /* Make sure we always see the terminating NUL character */
1805 1796855573 : smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1806 :
1807 1795557031 : dentry->d_lockref.count = 1;
1808 1795557031 : dentry->d_flags = 0;
1809 1795557031 : spin_lock_init(&dentry->d_lock);
1810 1788654150 : seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1811 1788654150 : dentry->d_inode = NULL;
1812 1788654150 : dentry->d_parent = dentry;
1813 1788654150 : dentry->d_sb = sb;
1814 1788654150 : dentry->d_op = NULL;
1815 1788654150 : dentry->d_fsdata = NULL;
1816 1788654150 : INIT_HLIST_BL_NODE(&dentry->d_hash);
1817 1788654150 : INIT_LIST_HEAD(&dentry->d_lru);
1818 1788654150 : INIT_LIST_HEAD(&dentry->d_subdirs);
1819 1788654150 : INIT_HLIST_NODE(&dentry->d_u.d_alias);
1820 1788654150 : INIT_LIST_HEAD(&dentry->d_child);
1821 1788654150 : d_set_d_op(dentry, dentry->d_sb->s_d_op);
1822 :
1823 1795021860 : if (dentry->d_op && dentry->d_op->d_init) {
1824 0 : err = dentry->d_op->d_init(dentry);
1825 0 : if (err) {
1826 0 : if (dname_external(dentry))
1827 0 : kfree(external_name(dentry));
1828 0 : kmem_cache_free(dentry_cache, dentry);
1829 0 : return NULL;
1830 : }
1831 : }
1832 :
1833 1795021860 : this_cpu_inc(nr_dentry);
1834 :
1835 1795021860 : return dentry;
1836 : }
1837 :
1838 : /**
1839 : * d_alloc - allocate a dcache entry
1840 : * @parent: parent of entry to allocate
1841 : * @name: qstr of the name
1842 : *
1843 : * Allocates a dentry. It returns %NULL if there is insufficient memory
1844 : * available. On a success the dentry is returned. The name passed in is
1845 : * copied and the copy passed in may be reused after this call.
1846 : */
1847 1263524418 : struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1848 : {
1849 1263524418 : struct dentry *dentry = __d_alloc(parent->d_sb, name);
1850 1262978375 : if (!dentry)
1851 : return NULL;
1852 1262978375 : spin_lock(&parent->d_lock);
1853 : /*
1854 : * don't need child lock because it is not subject
1855 : * to concurrency here
1856 : */
1857 1264490635 : __dget_dlock(parent);
1858 1264490635 : dentry->d_parent = parent;
1859 1264490635 : list_add(&dentry->d_child, &parent->d_subdirs);
1860 1264434274 : spin_unlock(&parent->d_lock);
1861 :
1862 1264434274 : return dentry;
1863 : }
1864 : EXPORT_SYMBOL(d_alloc);
1865 :
1866 6811 : struct dentry *d_alloc_anon(struct super_block *sb)
1867 : {
1868 6811 : return __d_alloc(sb, NULL);
1869 : }
1870 : EXPORT_SYMBOL(d_alloc_anon);
1871 :
1872 8334193 : struct dentry *d_alloc_cursor(struct dentry * parent)
1873 : {
1874 8334193 : struct dentry *dentry = d_alloc_anon(parent->d_sb);
1875 8333169 : if (dentry) {
1876 8333169 : dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1877 16667214 : dentry->d_parent = dget(parent);
1878 : }
1879 8334045 : return dentry;
1880 : }
1881 :
1882 : /**
1883 : * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1884 : * @sb: the superblock
1885 : * @name: qstr of the name
1886 : *
1887 : * For a filesystem that just pins its dentries in memory and never
1888 : * performs lookups at all, return an unhashed IS_ROOT dentry.
1889 : * This is used for pipes, sockets et.al. - the stuff that should
1890 : * never be anyone's children or parents. Unlike all other
1891 : * dentries, these will not have RCU delay between dropping the
1892 : * last reference and freeing them.
1893 : *
1894 : * The only user is alloc_file_pseudo() and that's what should
1895 : * be considered a public interface. Don't use directly.
1896 : */
1897 397376167 : struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1898 : {
1899 397376167 : struct dentry *dentry = __d_alloc(sb, name);
1900 396987617 : if (likely(dentry))
1901 396987617 : dentry->d_flags |= DCACHE_NORCU;
1902 396987617 : return dentry;
1903 : }
1904 :
1905 165 : struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1906 : {
1907 165 : struct qstr q;
1908 :
1909 165 : q.name = name;
1910 165 : q.hash_len = hashlen_string(parent, name);
1911 165 : return d_alloc(parent, &q);
1912 : }
1913 : EXPORT_SYMBOL(d_alloc_name);
1914 :
1915 3169380000 : void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1916 : {
1917 3169380000 : WARN_ON_ONCE(dentry->d_op);
1918 3169380000 : WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1919 : DCACHE_OP_COMPARE |
1920 : DCACHE_OP_REVALIDATE |
1921 : DCACHE_OP_WEAK_REVALIDATE |
1922 : DCACHE_OP_DELETE |
1923 : DCACHE_OP_REAL));
1924 3169380000 : dentry->d_op = op;
1925 3169380000 : if (!op)
1926 : return;
1927 1424091806 : if (op->d_hash)
1928 0 : dentry->d_flags |= DCACHE_OP_HASH;
1929 1424091806 : if (op->d_compare)
1930 3119 : dentry->d_flags |= DCACHE_OP_COMPARE;
1931 1424091806 : if (op->d_revalidate)
1932 31964976 : dentry->d_flags |= DCACHE_OP_REVALIDATE;
1933 1424091806 : if (op->d_weak_revalidate)
1934 1620132 : dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1935 1424091806 : if (op->d_delete)
1936 1021669267 : dentry->d_flags |= DCACHE_OP_DELETE;
1937 1424091806 : if (op->d_prune)
1938 8476 : dentry->d_flags |= DCACHE_OP_PRUNE;
1939 1424091806 : if (op->d_real)
1940 246607 : dentry->d_flags |= DCACHE_OP_REAL;
1941 :
1942 : }
1943 : EXPORT_SYMBOL(d_set_d_op);
1944 :
1945 :
1946 : /*
1947 : * d_set_fallthru - Mark a dentry as falling through to a lower layer
1948 : * @dentry - The dentry to mark
1949 : *
1950 : * Mark a dentry as falling through to the lower layer (as set with
1951 : * d_pin_lower()). This flag may be recorded on the medium.
1952 : */
1953 0 : void d_set_fallthru(struct dentry *dentry)
1954 : {
1955 0 : spin_lock(&dentry->d_lock);
1956 0 : dentry->d_flags |= DCACHE_FALLTHRU;
1957 0 : spin_unlock(&dentry->d_lock);
1958 0 : }
1959 : EXPORT_SYMBOL(d_set_fallthru);
1960 :
1961 747509749 : static unsigned d_flags_for_inode(struct inode *inode)
1962 : {
1963 747509749 : unsigned add_flags = DCACHE_REGULAR_TYPE;
1964 :
1965 747509749 : if (!inode)
1966 : return DCACHE_MISS_TYPE;
1967 :
1968 747509749 : if (S_ISDIR(inode->i_mode)) {
1969 40592651 : add_flags = DCACHE_DIRECTORY_TYPE;
1970 40592651 : if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1971 32057424 : if (unlikely(!inode->i_op->lookup))
1972 : add_flags = DCACHE_AUTODIR_TYPE;
1973 : else
1974 32057424 : inode->i_opflags |= IOP_LOOKUP;
1975 : }
1976 40592651 : goto type_determined;
1977 : }
1978 :
1979 706917098 : if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1980 571713739 : if (unlikely(inode->i_op->get_link)) {
1981 40559659 : add_flags = DCACHE_SYMLINK_TYPE;
1982 40559659 : goto type_determined;
1983 : }
1984 531154080 : inode->i_opflags |= IOP_NOFOLLOW;
1985 : }
1986 :
1987 666357439 : if (unlikely(!S_ISREG(inode->i_mode)))
1988 81086890 : add_flags = DCACHE_SPECIAL_TYPE;
1989 :
1990 585270549 : type_determined:
1991 747509749 : if (unlikely(IS_AUTOMOUNT(inode)))
1992 0 : add_flags |= DCACHE_NEED_AUTOMOUNT;
1993 : return add_flags;
1994 : }
1995 :
1996 536482060 : static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1997 : {
1998 536482060 : unsigned add_flags = d_flags_for_inode(inode);
1999 535023365 : WARN_ON(d_in_lookup(dentry));
2000 :
2001 535023365 : spin_lock(&dentry->d_lock);
2002 : /*
2003 : * Decrement negative dentry count if it was in the LRU list.
2004 : */
2005 536752679 : if (dentry->d_flags & DCACHE_LRU_LIST)
2006 26860706 : this_cpu_dec(nr_dentry_negative);
2007 536657082 : hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2008 536657082 : raw_write_seqcount_begin(&dentry->d_seq);
2009 535807930 : __d_set_inode_and_type(dentry, inode, add_flags);
2010 535782749 : raw_write_seqcount_end(&dentry->d_seq);
2011 535744939 : fsnotify_update_flags(dentry);
2012 535631732 : spin_unlock(&dentry->d_lock);
2013 537205982 : }
2014 :
2015 : /**
2016 : * d_instantiate - fill in inode information for a dentry
2017 : * @entry: dentry to complete
2018 : * @inode: inode to attach to this dentry
2019 : *
2020 : * Fill in inode information in the entry.
2021 : *
2022 : * This turns negative dentries into productive full members
2023 : * of society.
2024 : *
2025 : * NOTE! This assumes that the inode count has been incremented
2026 : * (or otherwise set) by the caller to indicate that it is now
2027 : * in use by the dcache.
2028 : */
2029 :
2030 530948553 : void d_instantiate(struct dentry *entry, struct inode * inode)
2031 : {
2032 530948553 : BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
2033 530948553 : if (inode) {
2034 530948553 : security_d_instantiate(entry, inode);
2035 530948553 : spin_lock(&inode->i_lock);
2036 531998158 : __d_instantiate(entry, inode);
2037 532045896 : spin_unlock(&inode->i_lock);
2038 : }
2039 532332330 : }
2040 : EXPORT_SYMBOL(d_instantiate);
2041 :
2042 : /*
2043 : * This should be equivalent to d_instantiate() + unlock_new_inode(),
2044 : * with lockdep-related part of unlock_new_inode() done before
2045 : * anything else. Use that instead of open-coding d_instantiate()/
2046 : * unlock_new_inode() combinations.
2047 : */
2048 5127581 : void d_instantiate_new(struct dentry *entry, struct inode *inode)
2049 : {
2050 5127581 : BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
2051 5127581 : BUG_ON(!inode);
2052 5127581 : lockdep_annotate_inode_mutex_key(inode);
2053 5127581 : security_d_instantiate(entry, inode);
2054 5127581 : spin_lock(&inode->i_lock);
2055 5128211 : __d_instantiate(entry, inode);
2056 5126623 : WARN_ON(!(inode->i_state & I_NEW));
2057 5126623 : inode->i_state &= ~I_NEW & ~I_CREATING;
2058 5126623 : smp_mb();
2059 5128092 : wake_up_bit(&inode->i_state, __I_NEW);
2060 5127290 : spin_unlock(&inode->i_lock);
2061 5127482 : }
2062 : EXPORT_SYMBOL(d_instantiate_new);
2063 :
2064 124562 : struct dentry *d_make_root(struct inode *root_inode)
2065 : {
2066 124562 : struct dentry *res = NULL;
2067 :
2068 124562 : if (root_inode) {
2069 124562 : res = d_alloc_anon(root_inode->i_sb);
2070 124549 : if (res)
2071 124549 : d_instantiate(res, root_inode);
2072 : else
2073 0 : iput(root_inode);
2074 : }
2075 124580 : return res;
2076 : }
2077 : EXPORT_SYMBOL(d_make_root);
2078 :
2079 127427051 : static struct dentry *__d_instantiate_anon(struct dentry *dentry,
2080 : struct inode *inode,
2081 : bool disconnected)
2082 : {
2083 127427051 : struct dentry *res;
2084 127427051 : unsigned add_flags;
2085 :
2086 127427051 : security_d_instantiate(dentry, inode);
2087 127427051 : spin_lock(&inode->i_lock);
2088 131595113 : res = __d_find_any_alias(inode);
2089 118 : if (res) {
2090 118 : spin_unlock(&inode->i_lock);
2091 118 : dput(dentry);
2092 118 : goto out_iput;
2093 : }
2094 :
2095 : /* attach a disconnected dentry */
2096 131594995 : add_flags = d_flags_for_inode(inode);
2097 :
2098 128690550 : if (disconnected)
2099 128176898 : add_flags |= DCACHE_DISCONNECTED;
2100 :
2101 128690550 : spin_lock(&dentry->d_lock);
2102 131674012 : __d_set_inode_and_type(dentry, inode, add_flags);
2103 131680654 : hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2104 131680654 : if (!disconnected) {
2105 0 : hlist_bl_lock(&dentry->d_sb->s_roots);
2106 0 : hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
2107 0 : hlist_bl_unlock(&dentry->d_sb->s_roots);
2108 : }
2109 131680654 : spin_unlock(&dentry->d_lock);
2110 130632161 : spin_unlock(&inode->i_lock);
2111 :
2112 130632161 : return dentry;
2113 :
2114 : out_iput:
2115 118 : iput(inode);
2116 118 : return res;
2117 : }
2118 :
2119 0 : struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2120 : {
2121 0 : return __d_instantiate_anon(dentry, inode, true);
2122 : }
2123 : EXPORT_SYMBOL(d_instantiate_anon);
2124 :
2125 341207285 : static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2126 : {
2127 341207285 : struct dentry *tmp;
2128 341207285 : struct dentry *res;
2129 :
2130 341207285 : if (!inode)
2131 : return ERR_PTR(-ESTALE);
2132 341207285 : if (IS_ERR(inode))
2133 : return ERR_CAST(inode);
2134 :
2135 341183024 : res = d_find_any_alias(inode);
2136 344672581 : if (res)
2137 215534543 : goto out_iput;
2138 :
2139 129138038 : tmp = d_alloc_anon(inode->i_sb);
2140 128073046 : if (!tmp) {
2141 0 : res = ERR_PTR(-ENOMEM);
2142 0 : goto out_iput;
2143 : }
2144 :
2145 128073046 : return __d_instantiate_anon(tmp, inode, disconnected);
2146 :
2147 215534543 : out_iput:
2148 215534543 : iput(inode);
2149 215534543 : return res;
2150 : }
2151 :
2152 : /**
2153 : * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2154 : * @inode: inode to allocate the dentry for
2155 : *
2156 : * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2157 : * similar open by handle operations. The returned dentry may be anonymous,
2158 : * or may have a full name (if the inode was already in the cache).
2159 : *
2160 : * When called on a directory inode, we must ensure that the inode only ever
2161 : * has one dentry. If a dentry is found, that is returned instead of
2162 : * allocating a new one.
2163 : *
2164 : * On successful return, the reference to the inode has been transferred
2165 : * to the dentry. In case of an error the reference on the inode is released.
2166 : * To make it easier to use in export operations a %NULL or IS_ERR inode may
2167 : * be passed in and the error will be propagated to the return value,
2168 : * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2169 : */
2170 345204669 : struct dentry *d_obtain_alias(struct inode *inode)
2171 : {
2172 345204669 : return __d_obtain_alias(inode, true);
2173 : }
2174 : EXPORT_SYMBOL(d_obtain_alias);
2175 :
2176 : /**
2177 : * d_obtain_root - find or allocate a dentry for a given inode
2178 : * @inode: inode to allocate the dentry for
2179 : *
2180 : * Obtain an IS_ROOT dentry for the root of a filesystem.
2181 : *
2182 : * We must ensure that directory inodes only ever have one dentry. If a
2183 : * dentry is found, that is returned instead of allocating a new one.
2184 : *
2185 : * On successful return, the reference to the inode has been transferred
2186 : * to the dentry. In case of an error the reference on the inode is
2187 : * released. A %NULL or IS_ERR inode may be passed in and will be the
2188 : * error will be propagate to the return value, with a %NULL @inode
2189 : * replaced by ERR_PTR(-ESTALE).
2190 : */
2191 0 : struct dentry *d_obtain_root(struct inode *inode)
2192 : {
2193 0 : return __d_obtain_alias(inode, false);
2194 : }
2195 : EXPORT_SYMBOL(d_obtain_root);
2196 :
2197 : /**
2198 : * d_add_ci - lookup or allocate new dentry with case-exact name
2199 : * @inode: the inode case-insensitive lookup has found
2200 : * @dentry: the negative dentry that was passed to the parent's lookup func
2201 : * @name: the case-exact name to be associated with the returned dentry
2202 : *
2203 : * This is to avoid filling the dcache with case-insensitive names to the
2204 : * same inode, only the actual correct case is stored in the dcache for
2205 : * case-insensitive filesystems.
2206 : *
2207 : * For a case-insensitive lookup match and if the case-exact dentry
2208 : * already exists in the dcache, use it and return it.
2209 : *
2210 : * If no entry exists with the exact case name, allocate new dentry with
2211 : * the exact case, and return the spliced entry.
2212 : */
2213 484528 : struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2214 : struct qstr *name)
2215 : {
2216 484528 : struct dentry *found, *res;
2217 :
2218 : /*
2219 : * First check if a dentry matching the name already exists,
2220 : * if not go ahead and create it now.
2221 : */
2222 484528 : found = d_hash_and_lookup(dentry->d_parent, name);
2223 484528 : if (found) {
2224 357324 : iput(inode);
2225 357324 : return found;
2226 : }
2227 127204 : if (d_in_lookup(dentry)) {
2228 126951 : found = d_alloc_parallel(dentry->d_parent, name,
2229 : dentry->d_wait);
2230 126951 : if (IS_ERR(found) || !d_in_lookup(found)) {
2231 0 : iput(inode);
2232 0 : return found;
2233 : }
2234 : } else {
2235 253 : found = d_alloc(dentry->d_parent, name);
2236 253 : if (!found) {
2237 0 : iput(inode);
2238 0 : return ERR_PTR(-ENOMEM);
2239 : }
2240 : }
2241 127204 : res = d_splice_alias(inode, found);
2242 127204 : if (res) {
2243 0 : d_lookup_done(found);
2244 0 : dput(found);
2245 0 : return res;
2246 : }
2247 : return found;
2248 : }
2249 : EXPORT_SYMBOL(d_add_ci);
2250 :
2251 : /**
2252 : * d_same_name - compare dentry name with case-exact name
2253 : * @parent: parent dentry
2254 : * @dentry: the negative dentry that was passed to the parent's lookup func
2255 : * @name: the case-exact name to be associated with the returned dentry
2256 : *
2257 : * Return: true if names are same, or false
2258 : */
2259 1428200375 : bool d_same_name(const struct dentry *dentry, const struct dentry *parent,
2260 : const struct qstr *name)
2261 : {
2262 1428200375 : if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2263 1427637402 : if (dentry->d_name.len != name->len)
2264 : return false;
2265 1427456643 : return dentry_cmp(dentry, name->name, name->len) == 0;
2266 : }
2267 562973 : return parent->d_op->d_compare(dentry,
2268 562973 : dentry->d_name.len, dentry->d_name.name,
2269 562973 : name) == 0;
2270 : }
2271 : EXPORT_SYMBOL_GPL(d_same_name);
2272 :
2273 : /*
2274 : * This is __d_lookup_rcu() when the parent dentry has
2275 : * DCACHE_OP_COMPARE, which makes things much nastier.
2276 : */
2277 3159 : static noinline struct dentry *__d_lookup_rcu_op_compare(
2278 : const struct dentry *parent,
2279 : const struct qstr *name,
2280 : unsigned *seqp)
2281 : {
2282 3159 : u64 hashlen = name->hash_len;
2283 3159 : struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2284 3155 : struct hlist_bl_node *node;
2285 3155 : struct dentry *dentry;
2286 :
2287 3196 : hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2288 41 : int tlen;
2289 41 : const char *tname;
2290 41 : unsigned seq;
2291 :
2292 41 : seqretry:
2293 41 : seq = raw_seqcount_begin(&dentry->d_seq);
2294 41 : if (dentry->d_parent != parent)
2295 41 : continue;
2296 0 : if (d_unhashed(dentry))
2297 0 : continue;
2298 0 : if (dentry->d_name.hash != hashlen_hash(hashlen))
2299 0 : continue;
2300 0 : tlen = dentry->d_name.len;
2301 0 : tname = dentry->d_name.name;
2302 : /* we want a consistent (name,len) pair */
2303 0 : if (read_seqcount_retry(&dentry->d_seq, seq)) {
2304 0 : cpu_relax();
2305 0 : goto seqretry;
2306 : }
2307 0 : if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0)
2308 0 : continue;
2309 0 : *seqp = seq;
2310 0 : return dentry;
2311 : }
2312 : return NULL;
2313 : }
2314 :
2315 : /**
2316 : * __d_lookup_rcu - search for a dentry (racy, store-free)
2317 : * @parent: parent dentry
2318 : * @name: qstr of name we wish to find
2319 : * @seqp: returns d_seq value at the point where the dentry was found
2320 : * Returns: dentry, or NULL
2321 : *
2322 : * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2323 : * resolution (store-free path walking) design described in
2324 : * Documentation/filesystems/path-lookup.txt.
2325 : *
2326 : * This is not to be used outside core vfs.
2327 : *
2328 : * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2329 : * held, and rcu_read_lock held. The returned dentry must not be stored into
2330 : * without taking d_lock and checking d_seq sequence count against @seq
2331 : * returned here.
2332 : *
2333 : * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2334 : * function.
2335 : *
2336 : * Alternatively, __d_lookup_rcu may be called again to look up the child of
2337 : * the returned dentry, so long as its parent's seqlock is checked after the
2338 : * child is looked up. Thus, an interlocking stepping of sequence lock checks
2339 : * is formed, giving integrity down the path walk.
2340 : *
2341 : * NOTE! The caller *has* to check the resulting dentry against the sequence
2342 : * number we've returned before using any of the resulting dentry state!
2343 : */
2344 72911040817 : struct dentry *__d_lookup_rcu(const struct dentry *parent,
2345 : const struct qstr *name,
2346 : unsigned *seqp)
2347 : {
2348 72911040817 : u64 hashlen = name->hash_len;
2349 72911040817 : const unsigned char *str = name->name;
2350 72911040817 : struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2351 72911642324 : struct hlist_bl_node *node;
2352 72911642324 : struct dentry *dentry;
2353 :
2354 : /*
2355 : * Note: There is significant duplication with __d_lookup_rcu which is
2356 : * required to prevent single threaded performance regressions
2357 : * especially on architectures where smp_rmb (in seqcounts) are costly.
2358 : * Keep the two functions in sync.
2359 : */
2360 :
2361 72911642324 : if (unlikely(parent->d_flags & DCACHE_OP_COMPARE))
2362 3162 : return __d_lookup_rcu_op_compare(parent, name, seqp);
2363 :
2364 : /*
2365 : * The hash list is protected using RCU.
2366 : *
2367 : * Carefully use d_seq when comparing a candidate dentry, to avoid
2368 : * races with d_move().
2369 : *
2370 : * It is possible that concurrent renames can mess up our list
2371 : * walk here and result in missing our dentry, resulting in the
2372 : * false-negative result. d_lookup() protects against concurrent
2373 : * renames using rename_lock seqlock.
2374 : *
2375 : * See Documentation/filesystems/path-lookup.txt for more details.
2376 : */
2377 82143487976 : hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2378 79958320883 : unsigned seq;
2379 :
2380 : /*
2381 : * The dentry sequence count protects us from concurrent
2382 : * renames, and thus protects parent and name fields.
2383 : *
2384 : * The caller must perform a seqcount check in order
2385 : * to do anything useful with the returned dentry.
2386 : *
2387 : * NOTE! We do a "raw" seqcount_begin here. That means that
2388 : * we don't wait for the sequence count to stabilize if it
2389 : * is in the middle of a sequence change. If we do the slow
2390 : * dentry compare, we will do seqretries until it is stable,
2391 : * and if we end up with a successful lookup, we actually
2392 : * want to exit RCU lookup anyway.
2393 : *
2394 : * Note that raw_seqcount_begin still *does* smp_rmb(), so
2395 : * we are still guaranteed NUL-termination of ->d_name.name.
2396 : */
2397 79958320883 : seq = raw_seqcount_begin(&dentry->d_seq);
2398 79960663559 : if (dentry->d_parent != parent)
2399 9196813628 : continue;
2400 70763849931 : if (d_unhashed(dentry))
2401 44606 : continue;
2402 70763805325 : if (dentry->d_name.hash_len != hashlen)
2403 34960613 : continue;
2404 70728844712 : if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2405 29967 : continue;
2406 70743123121 : *seqp = seq;
2407 70743123121 : return dentry;
2408 : }
2409 : return NULL;
2410 : }
2411 :
2412 : /**
2413 : * d_lookup - search for a dentry
2414 : * @parent: parent dentry
2415 : * @name: qstr of name we wish to find
2416 : * Returns: dentry, or NULL
2417 : *
2418 : * d_lookup searches the children of the parent dentry for the name in
2419 : * question. If the dentry is found its reference count is incremented and the
2420 : * dentry is returned. The caller must use dput to free the entry when it has
2421 : * finished using it. %NULL is returned if the dentry does not exist.
2422 : */
2423 1018536047 : struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2424 : {
2425 1018551057 : struct dentry *dentry;
2426 1018551057 : unsigned seq;
2427 :
2428 1018551057 : do {
2429 1018551057 : seq = read_seqbegin(&rename_lock);
2430 1018357967 : dentry = __d_lookup(parent, name);
2431 1019445566 : if (dentry)
2432 : break;
2433 224528816 : } while (read_seqretry(&rename_lock, seq));
2434 1019442099 : return dentry;
2435 : }
2436 : EXPORT_SYMBOL(d_lookup);
2437 :
2438 : /**
2439 : * __d_lookup - search for a dentry (racy)
2440 : * @parent: parent dentry
2441 : * @name: qstr of name we wish to find
2442 : * Returns: dentry, or NULL
2443 : *
2444 : * __d_lookup is like d_lookup, however it may (rarely) return a
2445 : * false-negative result due to unrelated rename activity.
2446 : *
2447 : * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2448 : * however it must be used carefully, eg. with a following d_lookup in
2449 : * the case of failure.
2450 : *
2451 : * __d_lookup callers must be commented.
2452 : */
2453 1666599412 : struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2454 : {
2455 1666599412 : unsigned int hash = name->hash;
2456 1666599412 : struct hlist_bl_head *b = d_hash(hash);
2457 1666381256 : struct hlist_bl_node *node;
2458 1666381256 : struct dentry *found = NULL;
2459 1666381256 : struct dentry *dentry;
2460 :
2461 : /*
2462 : * Note: There is significant duplication with __d_lookup_rcu which is
2463 : * required to prevent single threaded performance regressions
2464 : * especially on architectures where smp_rmb (in seqcounts) are costly.
2465 : * Keep the two functions in sync.
2466 : */
2467 :
2468 : /*
2469 : * The hash list is protected using RCU.
2470 : *
2471 : * Take d_lock when comparing a candidate dentry, to avoid races
2472 : * with d_move().
2473 : *
2474 : * It is possible that concurrent renames can mess up our list
2475 : * walk here and result in missing our dentry, resulting in the
2476 : * false-negative result. d_lookup() protects against concurrent
2477 : * renames using rename_lock seqlock.
2478 : *
2479 : * See Documentation/filesystems/path-lookup.txt for more details.
2480 : */
2481 1666381256 : rcu_read_lock();
2482 :
2483 1760017889 : hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2484 :
2485 1512634935 : if (dentry->d_name.hash != hash)
2486 92643409 : continue;
2487 :
2488 1419991526 : spin_lock(&dentry->d_lock);
2489 1420946977 : if (dentry->d_parent != parent)
2490 32030 : goto next;
2491 1420914947 : if (d_unhashed(dentry))
2492 393 : goto next;
2493 :
2494 1420914554 : if (!d_same_name(dentry, parent, name))
2495 18581 : goto next;
2496 :
2497 1420618866 : dentry->d_lockref.count++;
2498 1420618866 : found = dentry;
2499 1420618866 : spin_unlock(&dentry->d_lock);
2500 : break;
2501 51004 : next:
2502 51004 : spin_unlock(&dentry->d_lock);
2503 : }
2504 1668256915 : rcu_read_unlock();
2505 :
2506 1668165283 : return found;
2507 : }
2508 :
2509 : /**
2510 : * d_hash_and_lookup - hash the qstr then search for a dentry
2511 : * @dir: Directory to search in
2512 : * @name: qstr of name we wish to find
2513 : *
2514 : * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2515 : */
2516 22624745 : struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2517 : {
2518 : /*
2519 : * Check for a fs-specific hash function. Note that we must
2520 : * calculate the standard hash first, as the d_op->d_hash()
2521 : * routine may choose to leave the hash value unchanged.
2522 : */
2523 22624745 : name->hash = full_name_hash(dir, name->name, name->len);
2524 22624745 : if (dir->d_flags & DCACHE_OP_HASH) {
2525 0 : int err = dir->d_op->d_hash(dir, name);
2526 0 : if (unlikely(err < 0))
2527 0 : return ERR_PTR(err);
2528 : }
2529 22624745 : return d_lookup(dir, name);
2530 : }
2531 : EXPORT_SYMBOL(d_hash_and_lookup);
2532 :
2533 : /*
2534 : * When a file is deleted, we have two options:
2535 : * - turn this dentry into a negative dentry
2536 : * - unhash this dentry and free it.
2537 : *
2538 : * Usually, we want to just turn this into
2539 : * a negative dentry, but if anybody else is
2540 : * currently using the dentry or the inode
2541 : * we can't do that and we fall back on removing
2542 : * it from the hash queues and waiting for
2543 : * it to be deleted later when it has no users
2544 : */
2545 :
2546 : /**
2547 : * d_delete - delete a dentry
2548 : * @dentry: The dentry to delete
2549 : *
2550 : * Turn the dentry into a negative dentry if possible, otherwise
2551 : * remove it from the hash queues so it can be deleted later
2552 : */
2553 :
2554 71946051 : void d_delete(struct dentry * dentry)
2555 : {
2556 71946051 : struct inode *inode = dentry->d_inode;
2557 :
2558 71946051 : spin_lock(&inode->i_lock);
2559 71947675 : spin_lock(&dentry->d_lock);
2560 : /*
2561 : * Are we the only user?
2562 : */
2563 71946857 : if (dentry->d_lockref.count == 1) {
2564 71630880 : dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2565 71630880 : dentry_unlink_inode(dentry);
2566 : } else {
2567 315977 : __d_drop(dentry);
2568 315974 : spin_unlock(&dentry->d_lock);
2569 315978 : spin_unlock(&inode->i_lock);
2570 : }
2571 71946825 : }
2572 : EXPORT_SYMBOL(d_delete);
2573 :
2574 1312698411 : static void __d_rehash(struct dentry *entry)
2575 : {
2576 1312698411 : struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2577 :
2578 1312282733 : hlist_bl_lock(b);
2579 1312981060 : hlist_bl_add_head_rcu(&entry->d_hash, b);
2580 1312979413 : hlist_bl_unlock(b);
2581 1312890962 : }
2582 :
2583 : /**
2584 : * d_rehash - add an entry back to the hash
2585 : * @entry: dentry to add to the hash
2586 : *
2587 : * Adds a dentry to the hash according to its name.
2588 : */
2589 :
2590 0 : void d_rehash(struct dentry * entry)
2591 : {
2592 0 : spin_lock(&entry->d_lock);
2593 0 : __d_rehash(entry);
2594 0 : spin_unlock(&entry->d_lock);
2595 0 : }
2596 : EXPORT_SYMBOL(d_rehash);
2597 :
2598 1111008111 : static inline unsigned start_dir_add(struct inode *dir)
2599 : {
2600 1111008111 : preempt_disable_nested();
2601 1118251333 : for (;;) {
2602 1118251333 : unsigned n = dir->i_dir_seq;
2603 1118251333 : if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2604 1111518528 : return n;
2605 6997099 : cpu_relax();
2606 : }
2607 : }
2608 :
2609 1111467210 : static inline void end_dir_add(struct inode *dir, unsigned int n,
2610 : wait_queue_head_t *d_wait)
2611 : {
2612 1111467210 : smp_store_release(&dir->i_dir_seq, n + 2);
2613 1111458416 : preempt_enable_nested();
2614 1111458416 : wake_up_all(d_wait);
2615 1110915121 : }
2616 :
2617 3705171 : static void d_wait_lookup(struct dentry *dentry)
2618 : {
2619 3705171 : if (d_in_lookup(dentry)) {
2620 115043 : DECLARE_WAITQUEUE(wait, current);
2621 115043 : add_wait_queue(dentry->d_wait, &wait);
2622 115043 : do {
2623 115043 : set_current_state(TASK_UNINTERRUPTIBLE);
2624 115043 : spin_unlock(&dentry->d_lock);
2625 115043 : schedule();
2626 115031 : spin_lock(&dentry->d_lock);
2627 115043 : } while (d_in_lookup(dentry));
2628 : }
2629 3705171 : }
2630 :
2631 1121852517 : struct dentry *d_alloc_parallel(struct dentry *parent,
2632 : const struct qstr *name,
2633 : wait_queue_head_t *wq)
2634 : {
2635 1121852517 : unsigned int hash = name->hash;
2636 1121852517 : struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2637 1121852517 : struct hlist_bl_node *node;
2638 1121852517 : struct dentry *new = d_alloc(parent, name);
2639 1122430717 : struct dentry *dentry;
2640 1122430717 : unsigned seq, r_seq, d_seq;
2641 :
2642 1122430717 : if (unlikely(!new))
2643 : return ERR_PTR(-ENOMEM);
2644 :
2645 1122430717 : retry:
2646 1153956882 : rcu_read_lock();
2647 1153888680 : seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2648 1153888680 : r_seq = read_seqbegin(&rename_lock);
2649 1153915931 : dentry = __d_lookup_rcu(parent, name, &d_seq);
2650 1153802910 : if (unlikely(dentry)) {
2651 7852177 : if (!lockref_get_not_dead(&dentry->d_lockref)) {
2652 1827533 : rcu_read_unlock();
2653 1827531 : goto retry;
2654 : }
2655 6037313 : if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2656 0 : rcu_read_unlock();
2657 0 : dput(dentry);
2658 0 : goto retry;
2659 : }
2660 6037285 : rcu_read_unlock();
2661 6036404 : dput(new);
2662 6036404 : return dentry;
2663 : }
2664 1145950733 : if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2665 2506 : rcu_read_unlock();
2666 2506 : goto retry;
2667 : }
2668 :
2669 1145887240 : if (unlikely(seq & 1)) {
2670 27851939 : rcu_read_unlock();
2671 27874998 : goto retry;
2672 : }
2673 :
2674 1118035301 : hlist_bl_lock(b);
2675 1118337963 : if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2676 1818790 : hlist_bl_unlock(b);
2677 1819710 : rcu_read_unlock();
2678 1819886 : goto retry;
2679 : }
2680 : /*
2681 : * No changes for the parent since the beginning of d_lookup().
2682 : * Since all removals from the chain happen with hlist_bl_lock(),
2683 : * any potential in-lookup matches are going to stay here until
2684 : * we unlock the chain. All fields are stable in everything
2685 : * we encounter.
2686 : */
2687 1116530754 : hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2688 3731079 : if (dentry->d_name.hash != hash)
2689 11560 : continue;
2690 3719519 : if (dentry->d_parent != parent)
2691 0 : continue;
2692 3719519 : if (!d_same_name(dentry, parent, name))
2693 21 : continue;
2694 3705618 : hlist_bl_unlock(b);
2695 : /* now we can try to grab a reference */
2696 3705160 : if (!lockref_get_not_dead(&dentry->d_lockref)) {
2697 447 : rcu_read_unlock();
2698 447 : goto retry;
2699 : }
2700 :
2701 3705167 : rcu_read_unlock();
2702 : /*
2703 : * somebody is likely to be still doing lookup for it;
2704 : * wait for them to finish
2705 : */
2706 3705166 : spin_lock(&dentry->d_lock);
2707 3705171 : d_wait_lookup(dentry);
2708 : /*
2709 : * it's not in-lookup anymore; in principle we should repeat
2710 : * everything from dcache lookup, but it's likely to be what
2711 : * d_lookup() would've found anyway. If it is, just return it;
2712 : * otherwise we really have to repeat the whole thing.
2713 : */
2714 3705170 : if (unlikely(dentry->d_name.hash != hash))
2715 0 : goto mismatch;
2716 3705170 : if (unlikely(dentry->d_parent != parent))
2717 0 : goto mismatch;
2718 3705170 : if (unlikely(d_unhashed(dentry)))
2719 797 : goto mismatch;
2720 3704373 : if (unlikely(!d_same_name(dentry, parent, name)))
2721 0 : goto mismatch;
2722 : /* OK, it *is* a hashed match; return it */
2723 3704373 : spin_unlock(&dentry->d_lock);
2724 3704374 : dput(new);
2725 3704374 : return dentry;
2726 : }
2727 1112799675 : rcu_read_unlock();
2728 : /* we can't take ->d_lock here; it's OK, though. */
2729 1112781492 : new->d_flags |= DCACHE_PAR_LOOKUP;
2730 1112781492 : new->d_wait = wq;
2731 1112781492 : hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2732 1112570262 : hlist_bl_unlock(b);
2733 1112570262 : return new;
2734 797 : mismatch:
2735 797 : spin_unlock(&dentry->d_lock);
2736 797 : dput(dentry);
2737 797 : goto retry;
2738 : }
2739 : EXPORT_SYMBOL(d_alloc_parallel);
2740 :
2741 : /*
2742 : * - Unhash the dentry
2743 : * - Retrieve and clear the waitqueue head in dentry
2744 : * - Return the waitqueue head
2745 : */
2746 1112775493 : static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry)
2747 : {
2748 1112775493 : wait_queue_head_t *d_wait;
2749 1112775493 : struct hlist_bl_head *b;
2750 :
2751 1112775493 : lockdep_assert_held(&dentry->d_lock);
2752 :
2753 1112775493 : b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash);
2754 1112775493 : hlist_bl_lock(b);
2755 1112890298 : dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2756 1112890298 : __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2757 1112788027 : d_wait = dentry->d_wait;
2758 1112788027 : dentry->d_wait = NULL;
2759 1112788027 : hlist_bl_unlock(b);
2760 1112759363 : INIT_HLIST_NODE(&dentry->d_u.d_alias);
2761 1112759363 : INIT_LIST_HEAD(&dentry->d_lru);
2762 1112759363 : return d_wait;
2763 : }
2764 :
2765 1287267 : void __d_lookup_unhash_wake(struct dentry *dentry)
2766 : {
2767 1287267 : spin_lock(&dentry->d_lock);
2768 1287262 : wake_up_all(__d_lookup_unhash(dentry));
2769 1287245 : spin_unlock(&dentry->d_lock);
2770 1287247 : }
2771 : EXPORT_SYMBOL(__d_lookup_unhash_wake);
2772 :
2773 : /* inode->i_lock held if inode is non-NULL */
2774 :
2775 1241169145 : static inline void __d_add(struct dentry *dentry, struct inode *inode)
2776 : {
2777 1241169145 : wait_queue_head_t *d_wait;
2778 1241169145 : struct inode *dir = NULL;
2779 1241169145 : unsigned n;
2780 1241169145 : spin_lock(&dentry->d_lock);
2781 1241493333 : if (unlikely(d_in_lookup(dentry))) {
2782 1108545555 : dir = dentry->d_parent->d_inode;
2783 1108545555 : n = start_dir_add(dir);
2784 1108883064 : d_wait = __d_lookup_unhash(dentry);
2785 : }
2786 1241813684 : if (inode) {
2787 81827870 : unsigned add_flags = d_flags_for_inode(inode);
2788 81817378 : hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2789 81817378 : raw_write_seqcount_begin(&dentry->d_seq);
2790 81817448 : __d_set_inode_and_type(dentry, inode, add_flags);
2791 81817437 : raw_write_seqcount_end(&dentry->d_seq);
2792 81817210 : fsnotify_update_flags(dentry);
2793 : }
2794 1241800804 : __d_rehash(dentry);
2795 1241857101 : if (dir)
2796 1108861376 : end_dir_add(dir, n, d_wait);
2797 1241224984 : spin_unlock(&dentry->d_lock);
2798 1241446147 : if (inode)
2799 81824715 : spin_unlock(&inode->i_lock);
2800 1241433626 : }
2801 :
2802 : /**
2803 : * d_add - add dentry to hash queues
2804 : * @entry: dentry to add
2805 : * @inode: The inode to attach to this dentry
2806 : *
2807 : * This adds the entry to the hash queues and initializes @inode.
2808 : * The entry was actually filled in earlier during d_alloc().
2809 : */
2810 :
2811 991267851 : void d_add(struct dentry *entry, struct inode *inode)
2812 : {
2813 991267851 : if (inode) {
2814 165 : security_d_instantiate(entry, inode);
2815 165 : spin_lock(&inode->i_lock);
2816 : }
2817 991267851 : __d_add(entry, inode);
2818 991416003 : }
2819 : EXPORT_SYMBOL(d_add);
2820 :
2821 : /**
2822 : * d_exact_alias - find and hash an exact unhashed alias
2823 : * @entry: dentry to add
2824 : * @inode: The inode to go with this dentry
2825 : *
2826 : * If an unhashed dentry with the same name/parent and desired
2827 : * inode already exists, hash and return it. Otherwise, return
2828 : * NULL.
2829 : *
2830 : * Parent directory should be locked.
2831 : */
2832 38290 : struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2833 : {
2834 38290 : struct dentry *alias;
2835 38290 : unsigned int hash = entry->d_name.hash;
2836 :
2837 38290 : spin_lock(&inode->i_lock);
2838 76578 : hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2839 : /*
2840 : * Don't need alias->d_lock here, because aliases with
2841 : * d_parent == entry->d_parent are not subject to name or
2842 : * parent changes, because the parent inode i_mutex is held.
2843 : */
2844 1 : if (alias->d_name.hash != hash)
2845 0 : continue;
2846 1 : if (alias->d_parent != entry->d_parent)
2847 0 : continue;
2848 1 : if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2849 0 : continue;
2850 0 : spin_lock(&alias->d_lock);
2851 0 : if (!d_unhashed(alias)) {
2852 0 : spin_unlock(&alias->d_lock);
2853 0 : alias = NULL;
2854 : } else {
2855 0 : __dget_dlock(alias);
2856 0 : __d_rehash(alias);
2857 0 : spin_unlock(&alias->d_lock);
2858 : }
2859 0 : spin_unlock(&inode->i_lock);
2860 0 : return alias;
2861 : }
2862 38288 : spin_unlock(&inode->i_lock);
2863 38288 : return NULL;
2864 : }
2865 : EXPORT_SYMBOL(d_exact_alias);
2866 :
2867 13889662 : static void swap_names(struct dentry *dentry, struct dentry *target)
2868 : {
2869 13889662 : if (unlikely(dname_external(target))) {
2870 1670 : if (unlikely(dname_external(dentry))) {
2871 : /*
2872 : * Both external: swap the pointers
2873 : */
2874 485 : swap(target->d_name.name, dentry->d_name.name);
2875 : } else {
2876 : /*
2877 : * dentry:internal, target:external. Steal target's
2878 : * storage and make target internal.
2879 : */
2880 2370 : memcpy(target->d_iname, dentry->d_name.name,
2881 : dentry->d_name.len + 1);
2882 1185 : dentry->d_name.name = target->d_name.name;
2883 1185 : target->d_name.name = target->d_iname;
2884 : }
2885 : } else {
2886 13887992 : if (unlikely(dname_external(dentry))) {
2887 : /*
2888 : * dentry:external, target:internal. Give dentry's
2889 : * storage to target and make dentry internal
2890 : */
2891 2444 : memcpy(dentry->d_iname, target->d_name.name,
2892 : target->d_name.len + 1);
2893 1222 : target->d_name.name = dentry->d_name.name;
2894 1222 : dentry->d_name.name = dentry->d_iname;
2895 : } else {
2896 : /*
2897 : * Both are internal.
2898 : */
2899 : unsigned int i;
2900 : BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2901 69433850 : for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2902 55547080 : swap(((long *) &dentry->d_iname)[i],
2903 : ((long *) &target->d_iname)[i]);
2904 : }
2905 : }
2906 : }
2907 13889662 : swap(dentry->d_name.hash_len, target->d_name.hash_len);
2908 13889662 : }
2909 :
2910 43274832 : static void copy_name(struct dentry *dentry, struct dentry *target)
2911 : {
2912 43274832 : struct external_name *old_name = NULL;
2913 43274832 : if (unlikely(dname_external(dentry)))
2914 1291219 : old_name = external_name(dentry);
2915 43274832 : if (unlikely(dname_external(target))) {
2916 1212755 : atomic_inc(&external_name(target)->u.count);
2917 1212755 : dentry->d_name = target->d_name;
2918 : } else {
2919 84124154 : memcpy(dentry->d_iname, target->d_name.name,
2920 : target->d_name.len + 1);
2921 42062077 : dentry->d_name.name = dentry->d_iname;
2922 42062077 : dentry->d_name.hash_len = target->d_name.hash_len;
2923 : }
2924 43274832 : if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2925 0 : kfree_rcu(old_name, u.head);
2926 43274832 : }
2927 :
2928 : /*
2929 : * __d_move - move a dentry
2930 : * @dentry: entry to move
2931 : * @target: new dentry
2932 : * @exchange: exchange the two dentries
2933 : *
2934 : * Update the dcache to reflect the move of a file name. Negative
2935 : * dcache entries should not be moved in this way. Caller must hold
2936 : * rename_lock, the i_mutex of the source and target directories,
2937 : * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2938 : */
2939 57164494 : static void __d_move(struct dentry *dentry, struct dentry *target,
2940 : bool exchange)
2941 : {
2942 57164494 : struct dentry *old_parent, *p;
2943 57164494 : wait_queue_head_t *d_wait;
2944 57164494 : struct inode *dir = NULL;
2945 57164494 : unsigned n;
2946 :
2947 57164494 : WARN_ON(!dentry->d_inode);
2948 57164494 : if (WARN_ON(dentry == target))
2949 : return;
2950 :
2951 57164494 : BUG_ON(d_ancestor(target, dentry));
2952 57164494 : old_parent = dentry->d_parent;
2953 57164494 : p = d_ancestor(old_parent, target);
2954 57164494 : if (IS_ROOT(dentry)) {
2955 2619817 : BUG_ON(p);
2956 2619817 : spin_lock(&target->d_parent->d_lock);
2957 54544677 : } else if (!p) {
2958 : /* target is not a descendent of dentry->d_parent */
2959 45493720 : spin_lock(&target->d_parent->d_lock);
2960 45493720 : spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2961 : } else {
2962 9050957 : BUG_ON(p == dentry);
2963 9050957 : spin_lock(&old_parent->d_lock);
2964 9050957 : if (p != target)
2965 2056206 : spin_lock_nested(&target->d_parent->d_lock,
2966 : DENTRY_D_LOCK_NESTED);
2967 : }
2968 57164494 : spin_lock_nested(&dentry->d_lock, 2);
2969 57164494 : spin_lock_nested(&target->d_lock, 3);
2970 :
2971 57164494 : if (unlikely(d_in_lookup(target))) {
2972 2619817 : dir = target->d_parent->d_inode;
2973 2619817 : n = start_dir_add(dir);
2974 2619817 : d_wait = __d_lookup_unhash(target);
2975 : }
2976 :
2977 57164494 : write_seqcount_begin(&dentry->d_seq);
2978 57164494 : write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2979 :
2980 : /* unhash both */
2981 57164494 : if (!d_unhashed(dentry))
2982 54544677 : ___d_drop(dentry);
2983 57164494 : if (!d_unhashed(target))
2984 54544677 : ___d_drop(target);
2985 :
2986 : /* ... and switch them in the tree */
2987 57164494 : dentry->d_parent = target->d_parent;
2988 57164494 : if (!exchange) {
2989 43274832 : copy_name(dentry, target);
2990 43274832 : target->d_hash.pprev = NULL;
2991 43274832 : dentry->d_parent->d_lockref.count++;
2992 43274832 : if (dentry != old_parent) /* wasn't IS_ROOT */
2993 40655015 : WARN_ON(!--old_parent->d_lockref.count);
2994 : } else {
2995 13889662 : target->d_parent = old_parent;
2996 13889662 : swap_names(dentry, target);
2997 13889662 : list_move(&target->d_child, &target->d_parent->d_subdirs);
2998 13889662 : __d_rehash(target);
2999 13889662 : fsnotify_update_flags(target);
3000 : }
3001 57164494 : list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
3002 57164494 : __d_rehash(dentry);
3003 57164494 : fsnotify_update_flags(dentry);
3004 57164494 : fscrypt_handle_d_move(dentry);
3005 :
3006 57164494 : write_seqcount_end(&target->d_seq);
3007 57164494 : write_seqcount_end(&dentry->d_seq);
3008 :
3009 57164494 : if (dir)
3010 2619817 : end_dir_add(dir, n, d_wait);
3011 :
3012 57164494 : if (dentry->d_parent != old_parent)
3013 50169743 : spin_unlock(&dentry->d_parent->d_lock);
3014 57164494 : if (dentry != old_parent)
3015 54544677 : spin_unlock(&old_parent->d_lock);
3016 57164494 : spin_unlock(&target->d_lock);
3017 57164494 : spin_unlock(&dentry->d_lock);
3018 : }
3019 :
3020 : /*
3021 : * d_move - move a dentry
3022 : * @dentry: entry to move
3023 : * @target: new dentry
3024 : *
3025 : * Update the dcache to reflect the move of a file name. Negative
3026 : * dcache entries should not be moved in this way. See the locking
3027 : * requirements for __d_move.
3028 : */
3029 40654380 : void d_move(struct dentry *dentry, struct dentry *target)
3030 : {
3031 40654380 : write_seqlock(&rename_lock);
3032 40655015 : __d_move(dentry, target, false);
3033 40655015 : write_sequnlock(&rename_lock);
3034 40655015 : }
3035 : EXPORT_SYMBOL(d_move);
3036 :
3037 : /*
3038 : * d_exchange - exchange two dentries
3039 : * @dentry1: first dentry
3040 : * @dentry2: second dentry
3041 : */
3042 13889625 : void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
3043 : {
3044 13889625 : write_seqlock(&rename_lock);
3045 :
3046 13889662 : WARN_ON(!dentry1->d_inode);
3047 13889662 : WARN_ON(!dentry2->d_inode);
3048 13889662 : WARN_ON(IS_ROOT(dentry1));
3049 13889662 : WARN_ON(IS_ROOT(dentry2));
3050 :
3051 13889662 : __d_move(dentry1, dentry2, true);
3052 :
3053 13889662 : write_sequnlock(&rename_lock);
3054 13889662 : }
3055 :
3056 : /**
3057 : * d_ancestor - search for an ancestor
3058 : * @p1: ancestor dentry
3059 : * @p2: child dentry
3060 : *
3061 : * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
3062 : * an ancestor of p2, else NULL.
3063 : */
3064 96875442 : struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
3065 : {
3066 154039936 : struct dentry *p;
3067 :
3068 10996480064 : for (p = p2; !IS_ROOT(p); p = p->d_parent) {
3069 10795935135 : if (p->d_parent == p1)
3070 5242163 : return p;
3071 : }
3072 : return NULL;
3073 : }
3074 :
3075 : /*
3076 : * This helper attempts to cope with remotely renamed directories
3077 : *
3078 : * It assumes that the caller is already holding
3079 : * dentry->d_parent->d_inode->i_mutex, and rename_lock
3080 : *
3081 : * Note: If ever the locking in lock_rename() changes, then please
3082 : * remember to update this too...
3083 : */
3084 0 : static int __d_unalias(struct inode *inode,
3085 : struct dentry *dentry, struct dentry *alias)
3086 : {
3087 0 : struct mutex *m1 = NULL;
3088 0 : struct rw_semaphore *m2 = NULL;
3089 0 : int ret = -ESTALE;
3090 :
3091 : /* If alias and dentry share a parent, then no extra locks required */
3092 0 : if (alias->d_parent == dentry->d_parent)
3093 0 : goto out_unalias;
3094 :
3095 : /* See lock_rename() */
3096 0 : if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
3097 0 : goto out_err;
3098 0 : m1 = &dentry->d_sb->s_vfs_rename_mutex;
3099 0 : if (!inode_trylock_shared(alias->d_parent->d_inode))
3100 0 : goto out_err;
3101 0 : m2 = &alias->d_parent->d_inode->i_rwsem;
3102 0 : out_unalias:
3103 0 : __d_move(alias, dentry, false);
3104 0 : ret = 0;
3105 0 : out_err:
3106 0 : if (m2)
3107 0 : up_read(m2);
3108 0 : if (m1)
3109 0 : mutex_unlock(m1);
3110 0 : return ret;
3111 : }
3112 :
3113 : /**
3114 : * d_splice_alias - splice a disconnected dentry into the tree if one exists
3115 : * @inode: the inode which may have a disconnected dentry
3116 : * @dentry: a negative dentry which we want to point to the inode.
3117 : *
3118 : * If inode is a directory and has an IS_ROOT alias, then d_move that in
3119 : * place of the given dentry and return it, else simply d_add the inode
3120 : * to the dentry and return NULL.
3121 : *
3122 : * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3123 : * we should error out: directories can't have multiple aliases.
3124 : *
3125 : * This is needed in the lookup routine of any filesystem that is exportable
3126 : * (via knfsd) so that we can build dcache paths to directories effectively.
3127 : *
3128 : * If a dentry was found and moved, then it is returned. Otherwise NULL
3129 : * is returned. This matches the expected return value of ->lookup.
3130 : *
3131 : * Cluster filesystems may call this function with a negative, hashed dentry.
3132 : * In that case, we know that the inode will be a regular file, and also this
3133 : * will only occur during atomic_open. So we need to check for the dentry
3134 : * being already hashed only in the final case.
3135 : */
3136 252969694 : struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3137 : {
3138 252969694 : if (IS_ERR(inode))
3139 : return ERR_CAST(inode);
3140 :
3141 252893814 : BUG_ON(!d_unhashed(dentry));
3142 :
3143 252893814 : if (!inode)
3144 168471913 : goto out;
3145 :
3146 84421901 : security_d_instantiate(dentry, inode);
3147 84421901 : spin_lock(&inode->i_lock);
3148 84432606 : if (S_ISDIR(inode->i_mode)) {
3149 26151061 : struct dentry *new = __d_find_any_alias(inode);
3150 26150546 : if (unlikely(new)) {
3151 : /* The reference to new ensures it remains an alias */
3152 2617773 : spin_unlock(&inode->i_lock);
3153 2617509 : write_seqlock(&rename_lock);
3154 2619817 : if (unlikely(d_ancestor(new, dentry))) {
3155 0 : write_sequnlock(&rename_lock);
3156 0 : dput(new);
3157 0 : new = ERR_PTR(-ELOOP);
3158 0 : pr_warn_ratelimited(
3159 : "VFS: Lookup of '%s' in %s %s"
3160 : " would have caused loop\n",
3161 : dentry->d_name.name,
3162 : inode->i_sb->s_type->name,
3163 : inode->i_sb->s_id);
3164 2619817 : } else if (!IS_ROOT(new)) {
3165 0 : struct dentry *old_parent = dget(new->d_parent);
3166 0 : int err = __d_unalias(inode, dentry, new);
3167 0 : write_sequnlock(&rename_lock);
3168 0 : if (err) {
3169 0 : dput(new);
3170 0 : new = ERR_PTR(err);
3171 : }
3172 0 : dput(old_parent);
3173 : } else {
3174 2619817 : __d_move(new, dentry, false);
3175 2619817 : write_sequnlock(&rename_lock);
3176 : }
3177 2619817 : iput(inode);
3178 2619817 : return new;
3179 : }
3180 : }
3181 81814318 : out:
3182 250286231 : __d_add(dentry, inode);
3183 250286231 : return NULL;
3184 : }
3185 : EXPORT_SYMBOL(d_splice_alias);
3186 :
3187 : /*
3188 : * Test whether new_dentry is a subdirectory of old_dentry.
3189 : *
3190 : * Trivially implemented using the dcache structure
3191 : */
3192 :
3193 : /**
3194 : * is_subdir - is new dentry a subdirectory of old_dentry
3195 : * @new_dentry: new dentry
3196 : * @old_dentry: old dentry
3197 : *
3198 : * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3199 : * Returns false otherwise.
3200 : * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3201 : */
3202 :
3203 16632517 : bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3204 : {
3205 16632517 : bool result;
3206 16632517 : unsigned seq;
3207 :
3208 16632517 : if (new_dentry == old_dentry)
3209 : return true;
3210 :
3211 16620342 : do {
3212 : /* for restarting inner loop in case of seq retry */
3213 16620342 : seq = read_seqbegin(&rename_lock);
3214 : /*
3215 : * Need rcu_readlock to protect against the d_parent trashing
3216 : * due to d_move
3217 : */
3218 16621012 : rcu_read_lock();
3219 16620652 : if (d_ancestor(old_dentry, new_dentry))
3220 : result = true;
3221 : else
3222 1013807 : result = false;
3223 16620652 : rcu_read_unlock();
3224 16620969 : } while (read_seqretry(&rename_lock, seq));
3225 :
3226 : return result;
3227 : }
3228 : EXPORT_SYMBOL(is_subdir);
3229 :
3230 1429 : static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3231 : {
3232 1429 : struct dentry *root = data;
3233 1429 : if (dentry != root) {
3234 513 : if (d_unhashed(dentry) || !dentry->d_inode)
3235 : return D_WALK_SKIP;
3236 :
3237 513 : if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3238 513 : dentry->d_flags |= DCACHE_GENOCIDE;
3239 513 : dentry->d_lockref.count--;
3240 : }
3241 : }
3242 : return D_WALK_CONTINUE;
3243 : }
3244 :
3245 916 : void d_genocide(struct dentry *parent)
3246 : {
3247 916 : d_walk(parent, parent, d_genocide_kill);
3248 916 : }
3249 :
3250 : EXPORT_SYMBOL(d_genocide);
3251 :
3252 8559422 : void d_tmpfile(struct file *file, struct inode *inode)
3253 : {
3254 8559422 : struct dentry *dentry = file->f_path.dentry;
3255 :
3256 8559422 : inode_dec_link_count(inode);
3257 17962218 : BUG_ON(dentry->d_name.name != dentry->d_iname ||
3258 : !hlist_unhashed(&dentry->d_u.d_alias) ||
3259 : !d_unlinked(dentry));
3260 8981109 : spin_lock(&dentry->d_parent->d_lock);
3261 8994464 : spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3262 8984499 : dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3263 8984499 : (unsigned long long)inode->i_ino);
3264 8984499 : spin_unlock(&dentry->d_lock);
3265 8261494 : spin_unlock(&dentry->d_parent->d_lock);
3266 8601048 : d_instantiate(dentry, inode);
3267 8862337 : }
3268 : EXPORT_SYMBOL(d_tmpfile);
3269 :
3270 : static __initdata unsigned long dhash_entries;
3271 0 : static int __init set_dhash_entries(char *str)
3272 : {
3273 0 : if (!str)
3274 : return 0;
3275 0 : dhash_entries = simple_strtoul(str, &str, 0);
3276 0 : return 1;
3277 : }
3278 : __setup("dhash_entries=", set_dhash_entries);
3279 :
3280 0 : static void __init dcache_init_early(void)
3281 : {
3282 : /* If hashes are distributed across NUMA nodes, defer
3283 : * hash allocation until vmalloc space is available.
3284 : */
3285 0 : if (hashdist)
3286 : return;
3287 :
3288 0 : dentry_hashtable =
3289 0 : alloc_large_system_hash("Dentry cache",
3290 : sizeof(struct hlist_bl_head),
3291 : dhash_entries,
3292 : 13,
3293 : HASH_EARLY | HASH_ZERO,
3294 : &d_hash_shift,
3295 : NULL,
3296 : 0,
3297 : 0);
3298 0 : d_hash_shift = 32 - d_hash_shift;
3299 : }
3300 :
3301 0 : static void __init dcache_init(void)
3302 : {
3303 : /*
3304 : * A constructor could be added for stable state like the lists,
3305 : * but it is probably not worth it because of the cache nature
3306 : * of the dcache.
3307 : */
3308 0 : dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3309 : SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3310 : d_iname);
3311 :
3312 : /* Hash may have been set up in dcache_init_early */
3313 0 : if (!hashdist)
3314 : return;
3315 :
3316 0 : dentry_hashtable =
3317 0 : alloc_large_system_hash("Dentry cache",
3318 : sizeof(struct hlist_bl_head),
3319 : dhash_entries,
3320 : 13,
3321 : HASH_ZERO,
3322 : &d_hash_shift,
3323 : NULL,
3324 : 0,
3325 : 0);
3326 0 : d_hash_shift = 32 - d_hash_shift;
3327 : }
3328 :
3329 : /* SLAB cache for __getname() consumers */
3330 : struct kmem_cache *names_cachep __read_mostly;
3331 : EXPORT_SYMBOL(names_cachep);
3332 :
3333 0 : void __init vfs_caches_init_early(void)
3334 : {
3335 0 : int i;
3336 :
3337 0 : for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3338 0 : INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3339 :
3340 0 : dcache_init_early();
3341 0 : inode_init_early();
3342 0 : }
3343 :
3344 0 : void __init vfs_caches_init(void)
3345 : {
3346 0 : names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3347 : SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3348 :
3349 0 : dcache_init();
3350 0 : inode_init();
3351 0 : files_init();
3352 0 : files_maxfiles_init();
3353 0 : mnt_init();
3354 0 : bdev_cache_init();
3355 0 : chrdev_init();
3356 0 : }
|
