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