1 // SPDX-License-Identifier: GPL-2.0-only
5 * Complete reimplementation
6 * (C) 1997 Thomas Schoebel-Theuer,
7 * with heavy changes by Linus Torvalds
11 * Notes on the allocation strategy:
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.
18 #include <linux/ratelimit.h>
19 #include <linux/string.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>
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
54 * - d_parent and d_subdirs
55 * - childrens' d_child and d_parent
56 * - d_u.d_alias, d_inode
59 * dentry->d_inode->i_lock
61 * dentry->d_sb->s_dentry_lru_lock
62 * dcache_hash_bucket lock
65 * If there is an ancestor relationship:
66 * dentry->d_parent->...->d_parent->d_lock
68 * dentry->d_parent->d_lock
71 * If no ancestor relationship:
72 * arbitrary, since it's serialized on rename_lock
74 int sysctl_vfs_cache_pressure __read_mostly
= 100;
75 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
77 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
79 EXPORT_SYMBOL(rename_lock
);
81 static struct kmem_cache
*dentry_cache __read_mostly
;
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
);
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
97 static unsigned int d_hash_shift __read_mostly
;
99 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
101 static inline struct hlist_bl_head
*d_hash(unsigned int hash
)
103 return dentry_hashtable
+ (hash
>> d_hash_shift
);
106 #define IN_LOOKUP_SHIFT 10
107 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
109 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
112 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
113 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
117 /* Statistics gathering. */
118 struct dentry_stat_t dentry_stat
= {
122 static DEFINE_PER_CPU(long, nr_dentry
);
123 static DEFINE_PER_CPU(long, nr_dentry_unused
);
124 static DEFINE_PER_CPU(long, nr_dentry_negative
);
126 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
129 * Here we resort to our own counters instead of using generic per-cpu counters
130 * for consistency with what the vfs inode code does. We are expected to harvest
131 * better code and performance by having our own specialized counters.
133 * Please note that the loop is done over all possible CPUs, not over all online
134 * CPUs. The reason for this is that we don't want to play games with CPUs going
135 * on and off. If one of them goes off, we will just keep their counters.
137 * glommer: See cffbc8a for details, and if you ever intend to change this,
138 * please update all vfs counters to match.
140 static long get_nr_dentry(void)
144 for_each_possible_cpu(i
)
145 sum
+= per_cpu(nr_dentry
, i
);
146 return sum
< 0 ? 0 : sum
;
149 static long get_nr_dentry_unused(void)
153 for_each_possible_cpu(i
)
154 sum
+= per_cpu(nr_dentry_unused
, i
);
155 return sum
< 0 ? 0 : sum
;
158 static long get_nr_dentry_negative(void)
163 for_each_possible_cpu(i
)
164 sum
+= per_cpu(nr_dentry_negative
, i
);
165 return sum
< 0 ? 0 : sum
;
168 int proc_nr_dentry(struct ctl_table
*table
, int write
, void *buffer
,
169 size_t *lenp
, loff_t
*ppos
)
171 dentry_stat
.nr_dentry
= get_nr_dentry();
172 dentry_stat
.nr_unused
= get_nr_dentry_unused();
173 dentry_stat
.nr_negative
= get_nr_dentry_negative();
174 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
179 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
180 * The strings are both count bytes long, and count is non-zero.
182 #ifdef CONFIG_DCACHE_WORD_ACCESS
184 #include <asm/word-at-a-time.h>
186 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
187 * aligned allocation for this particular component. We don't
188 * strictly need the load_unaligned_zeropad() safety, but it
189 * doesn't hurt either.
191 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
192 * need the careful unaligned handling.
194 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
196 unsigned long a
,b
,mask
;
199 a
= read_word_at_a_time(cs
);
200 b
= load_unaligned_zeropad(ct
);
201 if (tcount
< sizeof(unsigned long))
203 if (unlikely(a
!= b
))
205 cs
+= sizeof(unsigned long);
206 ct
+= sizeof(unsigned long);
207 tcount
-= sizeof(unsigned long);
211 mask
= bytemask_from_count(tcount
);
212 return unlikely(!!((a
^ b
) & mask
));
217 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
231 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
234 * Be careful about RCU walk racing with rename:
235 * use 'READ_ONCE' to fetch the name pointer.
237 * NOTE! Even if a rename will mean that the length
238 * was not loaded atomically, we don't care. The
239 * RCU walk will check the sequence count eventually,
240 * and catch it. And we won't overrun the buffer,
241 * because we're reading the name pointer atomically,
242 * and a dentry name is guaranteed to be properly
243 * terminated with a NUL byte.
245 * End result: even if 'len' is wrong, we'll exit
246 * early because the data cannot match (there can
247 * be no NUL in the ct/tcount data)
249 const unsigned char *cs
= READ_ONCE(dentry
->d_name
.name
);
251 return dentry_string_cmp(cs
, ct
, tcount
);
254 struct external_name
{
257 struct rcu_head head
;
259 unsigned char name
[];
262 static inline struct external_name
*external_name(struct dentry
*dentry
)
264 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
267 static void __d_free(struct rcu_head
*head
)
269 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
271 kmem_cache_free(dentry_cache
, dentry
);
274 static void __d_free_external(struct rcu_head
*head
)
276 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
277 kfree(external_name(dentry
));
278 kmem_cache_free(dentry_cache
, dentry
);
281 static inline int dname_external(const struct dentry
*dentry
)
283 return dentry
->d_name
.name
!= dentry
->d_iname
;
286 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
288 spin_lock(&dentry
->d_lock
);
289 name
->name
= dentry
->d_name
;
290 if (unlikely(dname_external(dentry
))) {
291 atomic_inc(&external_name(dentry
)->u
.count
);
293 memcpy(name
->inline_name
, dentry
->d_iname
,
294 dentry
->d_name
.len
+ 1);
295 name
->name
.name
= name
->inline_name
;
297 spin_unlock(&dentry
->d_lock
);
299 EXPORT_SYMBOL(take_dentry_name_snapshot
);
301 void release_dentry_name_snapshot(struct name_snapshot
*name
)
303 if (unlikely(name
->name
.name
!= name
->inline_name
)) {
304 struct external_name
*p
;
305 p
= container_of(name
->name
.name
, struct external_name
, name
[0]);
306 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
307 kfree_rcu(p
, u
.head
);
310 EXPORT_SYMBOL(release_dentry_name_snapshot
);
312 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
318 dentry
->d_inode
= inode
;
319 flags
= READ_ONCE(dentry
->d_flags
);
320 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
322 smp_store_release(&dentry
->d_flags
, flags
);
325 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
327 unsigned flags
= READ_ONCE(dentry
->d_flags
);
329 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
330 WRITE_ONCE(dentry
->d_flags
, flags
);
331 dentry
->d_inode
= NULL
;
332 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
333 this_cpu_inc(nr_dentry_negative
);
336 static void dentry_free(struct dentry
*dentry
)
338 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
339 if (unlikely(dname_external(dentry
))) {
340 struct external_name
*p
= external_name(dentry
);
341 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
342 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
346 /* if dentry was never visible to RCU, immediate free is OK */
347 if (dentry
->d_flags
& DCACHE_NORCU
)
348 __d_free(&dentry
->d_u
.d_rcu
);
350 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
354 * Release the dentry's inode, using the filesystem
355 * d_iput() operation if defined.
357 static void dentry_unlink_inode(struct dentry
* dentry
)
358 __releases(dentry
->d_lock
)
359 __releases(dentry
->d_inode
->i_lock
)
361 struct inode
*inode
= dentry
->d_inode
;
363 raw_write_seqcount_begin(&dentry
->d_seq
);
364 __d_clear_type_and_inode(dentry
);
365 hlist_del_init(&dentry
->d_u
.d_alias
);
366 raw_write_seqcount_end(&dentry
->d_seq
);
367 spin_unlock(&dentry
->d_lock
);
368 spin_unlock(&inode
->i_lock
);
370 fsnotify_inoderemove(inode
);
371 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
372 dentry
->d_op
->d_iput(dentry
, inode
);
378 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
379 * is in use - which includes both the "real" per-superblock
380 * LRU list _and_ the DCACHE_SHRINK_LIST use.
382 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
383 * on the shrink list (ie not on the superblock LRU list).
385 * The per-cpu "nr_dentry_unused" counters are updated with
386 * the DCACHE_LRU_LIST bit.
388 * The per-cpu "nr_dentry_negative" counters are only updated
389 * when deleted from or added to the per-superblock LRU list, not
390 * from/to the shrink list. That is to avoid an unneeded dec/inc
391 * pair when moving from LRU to shrink list in select_collect().
393 * These helper functions make sure we always follow the
394 * rules. d_lock must be held by the caller.
396 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
397 static void d_lru_add(struct dentry
*dentry
)
399 D_FLAG_VERIFY(dentry
, 0);
400 dentry
->d_flags
|= DCACHE_LRU_LIST
;
401 this_cpu_inc(nr_dentry_unused
);
402 if (d_is_negative(dentry
))
403 this_cpu_inc(nr_dentry_negative
);
404 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
407 static void d_lru_del(struct dentry
*dentry
)
409 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
410 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
411 this_cpu_dec(nr_dentry_unused
);
412 if (d_is_negative(dentry
))
413 this_cpu_dec(nr_dentry_negative
);
414 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
417 static void d_shrink_del(struct dentry
*dentry
)
419 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
420 list_del_init(&dentry
->d_lru
);
421 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
422 this_cpu_dec(nr_dentry_unused
);
425 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
427 D_FLAG_VERIFY(dentry
, 0);
428 list_add(&dentry
->d_lru
, list
);
429 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
430 this_cpu_inc(nr_dentry_unused
);
434 * These can only be called under the global LRU lock, ie during the
435 * callback for freeing the LRU list. "isolate" removes it from the
436 * LRU lists entirely, while shrink_move moves it to the indicated
439 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
441 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
442 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
443 this_cpu_dec(nr_dentry_unused
);
444 if (d_is_negative(dentry
))
445 this_cpu_dec(nr_dentry_negative
);
446 list_lru_isolate(lru
, &dentry
->d_lru
);
449 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
450 struct list_head
*list
)
452 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
453 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
454 if (d_is_negative(dentry
))
455 this_cpu_dec(nr_dentry_negative
);
456 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
459 static void ___d_drop(struct dentry
*dentry
)
461 struct hlist_bl_head
*b
;
463 * Hashed dentries are normally on the dentry hashtable,
464 * with the exception of those newly allocated by
465 * d_obtain_root, which are always IS_ROOT:
467 if (unlikely(IS_ROOT(dentry
)))
468 b
= &dentry
->d_sb
->s_roots
;
470 b
= d_hash(dentry
->d_name
.hash
);
473 __hlist_bl_del(&dentry
->d_hash
);
477 void __d_drop(struct dentry
*dentry
)
479 if (!d_unhashed(dentry
)) {
481 dentry
->d_hash
.pprev
= NULL
;
482 write_seqcount_invalidate(&dentry
->d_seq
);
485 EXPORT_SYMBOL(__d_drop
);
488 * d_drop - drop a dentry
489 * @dentry: dentry to drop
491 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
492 * be found through a VFS lookup any more. Note that this is different from
493 * deleting the dentry - d_delete will try to mark the dentry negative if
494 * possible, giving a successful _negative_ lookup, while d_drop will
495 * just make the cache lookup fail.
497 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
498 * reason (NFS timeouts or autofs deletes).
500 * __d_drop requires dentry->d_lock
502 * ___d_drop doesn't mark dentry as "unhashed"
503 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
505 void d_drop(struct dentry
*dentry
)
507 spin_lock(&dentry
->d_lock
);
509 spin_unlock(&dentry
->d_lock
);
511 EXPORT_SYMBOL(d_drop
);
513 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
517 * Inform d_walk() and shrink_dentry_list() that we are no longer
518 * attached to the dentry tree
520 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
521 if (unlikely(list_empty(&dentry
->d_child
)))
523 __list_del_entry(&dentry
->d_child
);
525 * Cursors can move around the list of children. While we'd been
526 * a normal list member, it didn't matter - ->d_child.next would've
527 * been updated. However, from now on it won't be and for the
528 * things like d_walk() it might end up with a nasty surprise.
529 * Normally d_walk() doesn't care about cursors moving around -
530 * ->d_lock on parent prevents that and since a cursor has no children
531 * of its own, we get through it without ever unlocking the parent.
532 * There is one exception, though - if we ascend from a child that
533 * gets killed as soon as we unlock it, the next sibling is found
534 * using the value left in its ->d_child.next. And if _that_
535 * pointed to a cursor, and cursor got moved (e.g. by lseek())
536 * before d_walk() regains parent->d_lock, we'll end up skipping
537 * everything the cursor had been moved past.
539 * Solution: make sure that the pointer left behind in ->d_child.next
540 * points to something that won't be moving around. I.e. skip the
543 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
544 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
545 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
547 dentry
->d_child
.next
= next
->d_child
.next
;
551 static void __dentry_kill(struct dentry
*dentry
)
553 struct dentry
*parent
= NULL
;
554 bool can_free
= true;
555 if (!IS_ROOT(dentry
))
556 parent
= dentry
->d_parent
;
559 * The dentry is now unrecoverably dead to the world.
561 lockref_mark_dead(&dentry
->d_lockref
);
564 * inform the fs via d_prune that this dentry is about to be
565 * unhashed and destroyed.
567 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
568 dentry
->d_op
->d_prune(dentry
);
570 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
571 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
574 /* if it was on the hash then remove it */
576 dentry_unlist(dentry
, parent
);
578 spin_unlock(&parent
->d_lock
);
580 dentry_unlink_inode(dentry
);
582 spin_unlock(&dentry
->d_lock
);
583 this_cpu_dec(nr_dentry
);
584 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
585 dentry
->d_op
->d_release(dentry
);
587 spin_lock(&dentry
->d_lock
);
588 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
589 dentry
->d_flags
|= DCACHE_MAY_FREE
;
592 spin_unlock(&dentry
->d_lock
);
593 if (likely(can_free
))
598 static struct dentry
*__lock_parent(struct dentry
*dentry
)
600 struct dentry
*parent
;
602 spin_unlock(&dentry
->d_lock
);
604 parent
= READ_ONCE(dentry
->d_parent
);
605 spin_lock(&parent
->d_lock
);
607 * We can't blindly lock dentry until we are sure
608 * that we won't violate the locking order.
609 * Any changes of dentry->d_parent must have
610 * been done with parent->d_lock held, so
611 * spin_lock() above is enough of a barrier
612 * for checking if it's still our child.
614 if (unlikely(parent
!= dentry
->d_parent
)) {
615 spin_unlock(&parent
->d_lock
);
619 if (parent
!= dentry
)
620 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
626 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
628 struct dentry
*parent
= dentry
->d_parent
;
631 if (likely(spin_trylock(&parent
->d_lock
)))
633 return __lock_parent(dentry
);
636 static inline bool retain_dentry(struct dentry
*dentry
)
638 WARN_ON(d_in_lookup(dentry
));
640 /* Unreachable? Get rid of it */
641 if (unlikely(d_unhashed(dentry
)))
644 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
647 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
648 if (dentry
->d_op
->d_delete(dentry
))
652 if (unlikely(dentry
->d_flags
& DCACHE_DONTCACHE
))
655 /* retain; LRU fodder */
656 dentry
->d_lockref
.count
--;
657 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
659 else if (unlikely(!(dentry
->d_flags
& DCACHE_REFERENCED
)))
660 dentry
->d_flags
|= DCACHE_REFERENCED
;
664 void d_mark_dontcache(struct inode
*inode
)
668 spin_lock(&inode
->i_lock
);
669 hlist_for_each_entry(de
, &inode
->i_dentry
, d_u
.d_alias
) {
670 spin_lock(&de
->d_lock
);
671 de
->d_flags
|= DCACHE_DONTCACHE
;
672 spin_unlock(&de
->d_lock
);
674 inode
->i_state
|= I_DONTCACHE
;
675 spin_unlock(&inode
->i_lock
);
677 EXPORT_SYMBOL(d_mark_dontcache
);
680 * Finish off a dentry we've decided to kill.
681 * dentry->d_lock must be held, returns with it unlocked.
682 * Returns dentry requiring refcount drop, or NULL if we're done.
684 static struct dentry
*dentry_kill(struct dentry
*dentry
)
685 __releases(dentry
->d_lock
)
687 struct inode
*inode
= dentry
->d_inode
;
688 struct dentry
*parent
= NULL
;
690 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
693 if (!IS_ROOT(dentry
)) {
694 parent
= dentry
->d_parent
;
695 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
696 parent
= __lock_parent(dentry
);
697 if (likely(inode
|| !dentry
->d_inode
))
699 /* negative that became positive */
701 spin_unlock(&parent
->d_lock
);
702 inode
= dentry
->d_inode
;
706 __dentry_kill(dentry
);
710 spin_unlock(&dentry
->d_lock
);
711 spin_lock(&inode
->i_lock
);
712 spin_lock(&dentry
->d_lock
);
713 parent
= lock_parent(dentry
);
715 if (unlikely(dentry
->d_lockref
.count
!= 1)) {
716 dentry
->d_lockref
.count
--;
717 } else if (likely(!retain_dentry(dentry
))) {
718 __dentry_kill(dentry
);
721 /* we are keeping it, after all */
723 spin_unlock(&inode
->i_lock
);
725 spin_unlock(&parent
->d_lock
);
726 spin_unlock(&dentry
->d_lock
);
731 * Try to do a lockless dput(), and return whether that was successful.
733 * If unsuccessful, we return false, having already taken the dentry lock.
735 * The caller needs to hold the RCU read lock, so that the dentry is
736 * guaranteed to stay around even if the refcount goes down to zero!
738 static inline bool fast_dput(struct dentry
*dentry
)
741 unsigned int d_flags
;
744 * If we have a d_op->d_delete() operation, we sould not
745 * let the dentry count go to zero, so use "put_or_lock".
747 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
748 return lockref_put_or_lock(&dentry
->d_lockref
);
751 * .. otherwise, we can try to just decrement the
752 * lockref optimistically.
754 ret
= lockref_put_return(&dentry
->d_lockref
);
757 * If the lockref_put_return() failed due to the lock being held
758 * by somebody else, the fast path has failed. We will need to
759 * get the lock, and then check the count again.
761 if (unlikely(ret
< 0)) {
762 spin_lock(&dentry
->d_lock
);
763 if (dentry
->d_lockref
.count
> 1) {
764 dentry
->d_lockref
.count
--;
765 spin_unlock(&dentry
->d_lock
);
772 * If we weren't the last ref, we're done.
778 * Careful, careful. The reference count went down
779 * to zero, but we don't hold the dentry lock, so
780 * somebody else could get it again, and do another
781 * dput(), and we need to not race with that.
783 * However, there is a very special and common case
784 * where we don't care, because there is nothing to
785 * do: the dentry is still hashed, it does not have
786 * a 'delete' op, and it's referenced and already on
789 * NOTE! Since we aren't locked, these values are
790 * not "stable". However, it is sufficient that at
791 * some point after we dropped the reference the
792 * dentry was hashed and the flags had the proper
793 * value. Other dentry users may have re-gotten
794 * a reference to the dentry and change that, but
795 * our work is done - we can leave the dentry
796 * around with a zero refcount.
798 * Nevertheless, there are two cases that we should kill
800 * 1. free disconnected dentries as soon as their refcount
802 * 2. free dentries if they should not be cached.
805 d_flags
= READ_ONCE(dentry
->d_flags
);
806 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
|
807 DCACHE_DISCONNECTED
| DCACHE_DONTCACHE
;
809 /* Nothing to do? Dropping the reference was all we needed? */
810 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
814 * Not the fast normal case? Get the lock. We've already decremented
815 * the refcount, but we'll need to re-check the situation after
818 spin_lock(&dentry
->d_lock
);
821 * Did somebody else grab a reference to it in the meantime, and
822 * we're no longer the last user after all? Alternatively, somebody
823 * else could have killed it and marked it dead. Either way, we
824 * don't need to do anything else.
826 if (dentry
->d_lockref
.count
) {
827 spin_unlock(&dentry
->d_lock
);
832 * Re-get the reference we optimistically dropped. We hold the
833 * lock, and we just tested that it was zero, so we can just
836 dentry
->d_lockref
.count
= 1;
844 * This is complicated by the fact that we do not want to put
845 * dentries that are no longer on any hash chain on the unused
846 * list: we'd much rather just get rid of them immediately.
848 * However, that implies that we have to traverse the dentry
849 * tree upwards to the parents which might _also_ now be
850 * scheduled for deletion (it may have been only waiting for
851 * its last child to go away).
853 * This tail recursion is done by hand as we don't want to depend
854 * on the compiler to always get this right (gcc generally doesn't).
855 * Real recursion would eat up our stack space.
859 * dput - release a dentry
860 * @dentry: dentry to release
862 * Release a dentry. This will drop the usage count and if appropriate
863 * call the dentry unlink method as well as removing it from the queues and
864 * releasing its resources. If the parent dentries were scheduled for release
865 * they too may now get deleted.
867 void dput(struct dentry
*dentry
)
873 if (likely(fast_dput(dentry
))) {
878 /* Slow case: now with the dentry lock held */
881 if (likely(retain_dentry(dentry
))) {
882 spin_unlock(&dentry
->d_lock
);
886 dentry
= dentry_kill(dentry
);
891 static void __dput_to_list(struct dentry
*dentry
, struct list_head
*list
)
892 __must_hold(&dentry
->d_lock
)
894 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
895 /* let the owner of the list it's on deal with it */
896 --dentry
->d_lockref
.count
;
898 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
900 if (!--dentry
->d_lockref
.count
)
901 d_shrink_add(dentry
, list
);
905 void dput_to_list(struct dentry
*dentry
, struct list_head
*list
)
908 if (likely(fast_dput(dentry
))) {
913 if (!retain_dentry(dentry
))
914 __dput_to_list(dentry
, list
);
915 spin_unlock(&dentry
->d_lock
);
918 /* This must be called with d_lock held */
919 static inline void __dget_dlock(struct dentry
*dentry
)
921 dentry
->d_lockref
.count
++;
924 static inline void __dget(struct dentry
*dentry
)
926 lockref_get(&dentry
->d_lockref
);
929 struct dentry
*dget_parent(struct dentry
*dentry
)
936 * Do optimistic parent lookup without any
940 seq
= raw_seqcount_begin(&dentry
->d_seq
);
941 ret
= READ_ONCE(dentry
->d_parent
);
942 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
944 if (likely(gotref
)) {
945 if (!read_seqcount_retry(&dentry
->d_seq
, seq
))
952 * Don't need rcu_dereference because we re-check it was correct under
956 ret
= dentry
->d_parent
;
957 spin_lock(&ret
->d_lock
);
958 if (unlikely(ret
!= dentry
->d_parent
)) {
959 spin_unlock(&ret
->d_lock
);
964 BUG_ON(!ret
->d_lockref
.count
);
965 ret
->d_lockref
.count
++;
966 spin_unlock(&ret
->d_lock
);
969 EXPORT_SYMBOL(dget_parent
);
971 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
973 struct dentry
*alias
;
975 if (hlist_empty(&inode
->i_dentry
))
977 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
983 * d_find_any_alias - find any alias for a given inode
984 * @inode: inode to find an alias for
986 * If any aliases exist for the given inode, take and return a
987 * reference for one of them. If no aliases exist, return %NULL.
989 struct dentry
*d_find_any_alias(struct inode
*inode
)
993 spin_lock(&inode
->i_lock
);
994 de
= __d_find_any_alias(inode
);
995 spin_unlock(&inode
->i_lock
);
998 EXPORT_SYMBOL(d_find_any_alias
);
1000 static struct dentry
*__d_find_alias(struct inode
*inode
)
1002 struct dentry
*alias
;
1004 if (S_ISDIR(inode
->i_mode
))
1005 return __d_find_any_alias(inode
);
1007 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1008 spin_lock(&alias
->d_lock
);
1009 if (!d_unhashed(alias
)) {
1010 __dget_dlock(alias
);
1011 spin_unlock(&alias
->d_lock
);
1014 spin_unlock(&alias
->d_lock
);
1020 * d_find_alias - grab a hashed alias of inode
1021 * @inode: inode in question
1023 * If inode has a hashed alias, or is a directory and has any alias,
1024 * acquire the reference to alias and return it. Otherwise return NULL.
1025 * Notice that if inode is a directory there can be only one alias and
1026 * it can be unhashed only if it has no children, or if it is the root
1027 * of a filesystem, or if the directory was renamed and d_revalidate
1028 * was the first vfs operation to notice.
1030 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1031 * any other hashed alias over that one.
1033 struct dentry
*d_find_alias(struct inode
*inode
)
1035 struct dentry
*de
= NULL
;
1037 if (!hlist_empty(&inode
->i_dentry
)) {
1038 spin_lock(&inode
->i_lock
);
1039 de
= __d_find_alias(inode
);
1040 spin_unlock(&inode
->i_lock
);
1044 EXPORT_SYMBOL(d_find_alias
);
1047 * Caller MUST be holding rcu_read_lock() and be guaranteed
1048 * that inode won't get freed until rcu_read_unlock().
1050 struct dentry
*d_find_alias_rcu(struct inode
*inode
)
1052 struct hlist_head
*l
= &inode
->i_dentry
;
1053 struct dentry
*de
= NULL
;
1055 spin_lock(&inode
->i_lock
);
1056 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1057 // used without having I_FREEING set, which means no aliases left
1058 if (likely(!(inode
->i_state
& I_FREEING
) && !hlist_empty(l
))) {
1059 if (S_ISDIR(inode
->i_mode
)) {
1060 de
= hlist_entry(l
->first
, struct dentry
, d_u
.d_alias
);
1062 hlist_for_each_entry(de
, l
, d_u
.d_alias
)
1063 if (!d_unhashed(de
))
1067 spin_unlock(&inode
->i_lock
);
1072 * Try to kill dentries associated with this inode.
1073 * WARNING: you must own a reference to inode.
1075 void d_prune_aliases(struct inode
*inode
)
1077 struct dentry
*dentry
;
1079 spin_lock(&inode
->i_lock
);
1080 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
1081 spin_lock(&dentry
->d_lock
);
1082 if (!dentry
->d_lockref
.count
) {
1083 struct dentry
*parent
= lock_parent(dentry
);
1084 if (likely(!dentry
->d_lockref
.count
)) {
1085 __dentry_kill(dentry
);
1090 spin_unlock(&parent
->d_lock
);
1092 spin_unlock(&dentry
->d_lock
);
1094 spin_unlock(&inode
->i_lock
);
1096 EXPORT_SYMBOL(d_prune_aliases
);
1099 * Lock a dentry from shrink list.
1100 * Called under rcu_read_lock() and dentry->d_lock; the former
1101 * guarantees that nothing we access will be freed under us.
1102 * Note that dentry is *not* protected from concurrent dentry_kill(),
1105 * Return false if dentry has been disrupted or grabbed, leaving
1106 * the caller to kick it off-list. Otherwise, return true and have
1107 * that dentry's inode and parent both locked.
1109 static bool shrink_lock_dentry(struct dentry
*dentry
)
1111 struct inode
*inode
;
1112 struct dentry
*parent
;
1114 if (dentry
->d_lockref
.count
)
1117 inode
= dentry
->d_inode
;
1118 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1119 spin_unlock(&dentry
->d_lock
);
1120 spin_lock(&inode
->i_lock
);
1121 spin_lock(&dentry
->d_lock
);
1122 if (unlikely(dentry
->d_lockref
.count
))
1124 /* changed inode means that somebody had grabbed it */
1125 if (unlikely(inode
!= dentry
->d_inode
))
1129 parent
= dentry
->d_parent
;
1130 if (IS_ROOT(dentry
) || likely(spin_trylock(&parent
->d_lock
)))
1133 spin_unlock(&dentry
->d_lock
);
1134 spin_lock(&parent
->d_lock
);
1135 if (unlikely(parent
!= dentry
->d_parent
)) {
1136 spin_unlock(&parent
->d_lock
);
1137 spin_lock(&dentry
->d_lock
);
1140 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1141 if (likely(!dentry
->d_lockref
.count
))
1143 spin_unlock(&parent
->d_lock
);
1146 spin_unlock(&inode
->i_lock
);
1150 void shrink_dentry_list(struct list_head
*list
)
1152 while (!list_empty(list
)) {
1153 struct dentry
*dentry
, *parent
;
1155 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
1156 spin_lock(&dentry
->d_lock
);
1158 if (!shrink_lock_dentry(dentry
)) {
1159 bool can_free
= false;
1161 d_shrink_del(dentry
);
1162 if (dentry
->d_lockref
.count
< 0)
1163 can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
1164 spin_unlock(&dentry
->d_lock
);
1166 dentry_free(dentry
);
1170 d_shrink_del(dentry
);
1171 parent
= dentry
->d_parent
;
1172 if (parent
!= dentry
)
1173 __dput_to_list(parent
, list
);
1174 __dentry_kill(dentry
);
1178 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1179 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1181 struct list_head
*freeable
= arg
;
1182 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1186 * we are inverting the lru lock/dentry->d_lock here,
1187 * so use a trylock. If we fail to get the lock, just skip
1190 if (!spin_trylock(&dentry
->d_lock
))
1194 * Referenced dentries are still in use. If they have active
1195 * counts, just remove them from the LRU. Otherwise give them
1196 * another pass through the LRU.
1198 if (dentry
->d_lockref
.count
) {
1199 d_lru_isolate(lru
, dentry
);
1200 spin_unlock(&dentry
->d_lock
);
1204 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1205 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1206 spin_unlock(&dentry
->d_lock
);
1209 * The list move itself will be made by the common LRU code. At
1210 * this point, we've dropped the dentry->d_lock but keep the
1211 * lru lock. This is safe to do, since every list movement is
1212 * protected by the lru lock even if both locks are held.
1214 * This is guaranteed by the fact that all LRU management
1215 * functions are intermediated by the LRU API calls like
1216 * list_lru_add and list_lru_del. List movement in this file
1217 * only ever occur through this functions or through callbacks
1218 * like this one, that are called from the LRU API.
1220 * The only exceptions to this are functions like
1221 * shrink_dentry_list, and code that first checks for the
1222 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1223 * operating only with stack provided lists after they are
1224 * properly isolated from the main list. It is thus, always a
1230 d_lru_shrink_move(lru
, dentry
, freeable
);
1231 spin_unlock(&dentry
->d_lock
);
1237 * prune_dcache_sb - shrink the dcache
1239 * @sc: shrink control, passed to list_lru_shrink_walk()
1241 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1242 * is done when we need more memory and called from the superblock shrinker
1245 * This function may fail to free any resources if all the dentries are in
1248 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1253 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1254 dentry_lru_isolate
, &dispose
);
1255 shrink_dentry_list(&dispose
);
1259 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1260 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1262 struct list_head
*freeable
= arg
;
1263 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1266 * we are inverting the lru lock/dentry->d_lock here,
1267 * so use a trylock. If we fail to get the lock, just skip
1270 if (!spin_trylock(&dentry
->d_lock
))
1273 d_lru_shrink_move(lru
, dentry
, freeable
);
1274 spin_unlock(&dentry
->d_lock
);
1281 * shrink_dcache_sb - shrink dcache for a superblock
1284 * Shrink the dcache for the specified super block. This is used to free
1285 * the dcache before unmounting a file system.
1287 void shrink_dcache_sb(struct super_block
*sb
)
1292 list_lru_walk(&sb
->s_dentry_lru
,
1293 dentry_lru_isolate_shrink
, &dispose
, 1024);
1294 shrink_dentry_list(&dispose
);
1295 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1297 EXPORT_SYMBOL(shrink_dcache_sb
);
1300 * enum d_walk_ret - action to talke during tree walk
1301 * @D_WALK_CONTINUE: contrinue walk
1302 * @D_WALK_QUIT: quit walk
1303 * @D_WALK_NORETRY: quit when retry is needed
1304 * @D_WALK_SKIP: skip this dentry and its children
1314 * d_walk - walk the dentry tree
1315 * @parent: start of walk
1316 * @data: data passed to @enter() and @finish()
1317 * @enter: callback when first entering the dentry
1319 * The @enter() callbacks are called with d_lock held.
1321 static void d_walk(struct dentry
*parent
, void *data
,
1322 enum d_walk_ret (*enter
)(void *, struct dentry
*))
1324 struct dentry
*this_parent
;
1325 struct list_head
*next
;
1327 enum d_walk_ret ret
;
1331 read_seqbegin_or_lock(&rename_lock
, &seq
);
1332 this_parent
= parent
;
1333 spin_lock(&this_parent
->d_lock
);
1335 ret
= enter(data
, this_parent
);
1337 case D_WALK_CONTINUE
:
1342 case D_WALK_NORETRY
:
1347 next
= this_parent
->d_subdirs
.next
;
1349 while (next
!= &this_parent
->d_subdirs
) {
1350 struct list_head
*tmp
= next
;
1351 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1354 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1357 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1359 ret
= enter(data
, dentry
);
1361 case D_WALK_CONTINUE
:
1364 spin_unlock(&dentry
->d_lock
);
1366 case D_WALK_NORETRY
:
1370 spin_unlock(&dentry
->d_lock
);
1374 if (!list_empty(&dentry
->d_subdirs
)) {
1375 spin_unlock(&this_parent
->d_lock
);
1376 spin_release(&dentry
->d_lock
.dep_map
, _RET_IP_
);
1377 this_parent
= dentry
;
1378 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1381 spin_unlock(&dentry
->d_lock
);
1384 * All done at this level ... ascend and resume the search.
1388 if (this_parent
!= parent
) {
1389 struct dentry
*child
= this_parent
;
1390 this_parent
= child
->d_parent
;
1392 spin_unlock(&child
->d_lock
);
1393 spin_lock(&this_parent
->d_lock
);
1395 /* might go back up the wrong parent if we have had a rename. */
1396 if (need_seqretry(&rename_lock
, seq
))
1398 /* go into the first sibling still alive */
1400 next
= child
->d_child
.next
;
1401 if (next
== &this_parent
->d_subdirs
)
1403 child
= list_entry(next
, struct dentry
, d_child
);
1404 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1408 if (need_seqretry(&rename_lock
, seq
))
1413 spin_unlock(&this_parent
->d_lock
);
1414 done_seqretry(&rename_lock
, seq
);
1418 spin_unlock(&this_parent
->d_lock
);
1427 struct check_mount
{
1428 struct vfsmount
*mnt
;
1429 unsigned int mounted
;
1432 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1434 struct check_mount
*info
= data
;
1435 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1437 if (likely(!d_mountpoint(dentry
)))
1438 return D_WALK_CONTINUE
;
1439 if (__path_is_mountpoint(&path
)) {
1443 return D_WALK_CONTINUE
;
1447 * path_has_submounts - check for mounts over a dentry in the
1448 * current namespace.
1449 * @parent: path to check.
1451 * Return true if the parent or its subdirectories contain
1452 * a mount point in the current namespace.
1454 int path_has_submounts(const struct path
*parent
)
1456 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1458 read_seqlock_excl(&mount_lock
);
1459 d_walk(parent
->dentry
, &data
, path_check_mount
);
1460 read_sequnlock_excl(&mount_lock
);
1462 return data
.mounted
;
1464 EXPORT_SYMBOL(path_has_submounts
);
1467 * Called by mount code to set a mountpoint and check if the mountpoint is
1468 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1469 * subtree can become unreachable).
1471 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1472 * this reason take rename_lock and d_lock on dentry and ancestors.
1474 int d_set_mounted(struct dentry
*dentry
)
1478 write_seqlock(&rename_lock
);
1479 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1480 /* Need exclusion wrt. d_invalidate() */
1481 spin_lock(&p
->d_lock
);
1482 if (unlikely(d_unhashed(p
))) {
1483 spin_unlock(&p
->d_lock
);
1486 spin_unlock(&p
->d_lock
);
1488 spin_lock(&dentry
->d_lock
);
1489 if (!d_unlinked(dentry
)) {
1491 if (!d_mountpoint(dentry
)) {
1492 dentry
->d_flags
|= DCACHE_MOUNTED
;
1496 spin_unlock(&dentry
->d_lock
);
1498 write_sequnlock(&rename_lock
);
1503 * Search the dentry child list of the specified parent,
1504 * and move any unused dentries to the end of the unused
1505 * list for prune_dcache(). We descend to the next level
1506 * whenever the d_subdirs list is non-empty and continue
1509 * It returns zero iff there are no unused children,
1510 * otherwise it returns the number of children moved to
1511 * the end of the unused list. This may not be the total
1512 * number of unused children, because select_parent can
1513 * drop the lock and return early due to latency
1517 struct select_data
{
1518 struct dentry
*start
;
1521 struct dentry
*victim
;
1523 struct list_head dispose
;
1526 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1528 struct select_data
*data
= _data
;
1529 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1531 if (data
->start
== dentry
)
1534 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1537 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1539 if (!dentry
->d_lockref
.count
) {
1540 d_shrink_add(dentry
, &data
->dispose
);
1545 * We can return to the caller if we have found some (this
1546 * ensures forward progress). We'll be coming back to find
1549 if (!list_empty(&data
->dispose
))
1550 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1555 static enum d_walk_ret
select_collect2(void *_data
, struct dentry
*dentry
)
1557 struct select_data
*data
= _data
;
1558 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1560 if (data
->start
== dentry
)
1563 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1564 if (!dentry
->d_lockref
.count
) {
1566 data
->victim
= dentry
;
1570 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1572 if (!dentry
->d_lockref
.count
)
1573 d_shrink_add(dentry
, &data
->dispose
);
1576 * We can return to the caller if we have found some (this
1577 * ensures forward progress). We'll be coming back to find
1580 if (!list_empty(&data
->dispose
))
1581 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1587 * shrink_dcache_parent - prune dcache
1588 * @parent: parent of entries to prune
1590 * Prune the dcache to remove unused children of the parent dentry.
1592 void shrink_dcache_parent(struct dentry
*parent
)
1595 struct select_data data
= {.start
= parent
};
1597 INIT_LIST_HEAD(&data
.dispose
);
1598 d_walk(parent
, &data
, select_collect
);
1600 if (!list_empty(&data
.dispose
)) {
1601 shrink_dentry_list(&data
.dispose
);
1609 d_walk(parent
, &data
, select_collect2
);
1611 struct dentry
*parent
;
1612 spin_lock(&data
.victim
->d_lock
);
1613 if (!shrink_lock_dentry(data
.victim
)) {
1614 spin_unlock(&data
.victim
->d_lock
);
1618 parent
= data
.victim
->d_parent
;
1619 if (parent
!= data
.victim
)
1620 __dput_to_list(parent
, &data
.dispose
);
1621 __dentry_kill(data
.victim
);
1624 if (!list_empty(&data
.dispose
))
1625 shrink_dentry_list(&data
.dispose
);
1628 EXPORT_SYMBOL(shrink_dcache_parent
);
1630 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1632 /* it has busy descendents; complain about those instead */
1633 if (!list_empty(&dentry
->d_subdirs
))
1634 return D_WALK_CONTINUE
;
1636 /* root with refcount 1 is fine */
1637 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1638 return D_WALK_CONTINUE
;
1640 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1641 " still in use (%d) [unmount of %s %s]\n",
1644 dentry
->d_inode
->i_ino
: 0UL,
1646 dentry
->d_lockref
.count
,
1647 dentry
->d_sb
->s_type
->name
,
1648 dentry
->d_sb
->s_id
);
1650 return D_WALK_CONTINUE
;
1653 static void do_one_tree(struct dentry
*dentry
)
1655 shrink_dcache_parent(dentry
);
1656 d_walk(dentry
, dentry
, umount_check
);
1662 * destroy the dentries attached to a superblock on unmounting
1664 void shrink_dcache_for_umount(struct super_block
*sb
)
1666 struct dentry
*dentry
;
1668 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1670 dentry
= sb
->s_root
;
1672 do_one_tree(dentry
);
1674 while (!hlist_bl_empty(&sb
->s_roots
)) {
1675 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_roots
), struct dentry
, d_hash
));
1676 do_one_tree(dentry
);
1680 static enum d_walk_ret
find_submount(void *_data
, struct dentry
*dentry
)
1682 struct dentry
**victim
= _data
;
1683 if (d_mountpoint(dentry
)) {
1684 __dget_dlock(dentry
);
1688 return D_WALK_CONTINUE
;
1692 * d_invalidate - detach submounts, prune dcache, and drop
1693 * @dentry: dentry to invalidate (aka detach, prune and drop)
1695 void d_invalidate(struct dentry
*dentry
)
1697 bool had_submounts
= false;
1698 spin_lock(&dentry
->d_lock
);
1699 if (d_unhashed(dentry
)) {
1700 spin_unlock(&dentry
->d_lock
);
1704 spin_unlock(&dentry
->d_lock
);
1706 /* Negative dentries can be dropped without further checks */
1707 if (!dentry
->d_inode
)
1710 shrink_dcache_parent(dentry
);
1712 struct dentry
*victim
= NULL
;
1713 d_walk(dentry
, &victim
, find_submount
);
1716 shrink_dcache_parent(dentry
);
1719 had_submounts
= true;
1720 detach_mounts(victim
);
1724 EXPORT_SYMBOL(d_invalidate
);
1727 * __d_alloc - allocate a dcache entry
1728 * @sb: filesystem it will belong to
1729 * @name: qstr of the name
1731 * Allocates a dentry. It returns %NULL if there is insufficient memory
1732 * available. On a success the dentry is returned. The name passed in is
1733 * copied and the copy passed in may be reused after this call.
1736 static struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1738 struct dentry
*dentry
;
1742 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1747 * We guarantee that the inline name is always NUL-terminated.
1748 * This way the memcpy() done by the name switching in rename
1749 * will still always have a NUL at the end, even if we might
1750 * be overwriting an internal NUL character
1752 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1753 if (unlikely(!name
)) {
1755 dname
= dentry
->d_iname
;
1756 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1757 size_t size
= offsetof(struct external_name
, name
[1]);
1758 struct external_name
*p
= kmalloc(size
+ name
->len
,
1759 GFP_KERNEL_ACCOUNT
|
1762 kmem_cache_free(dentry_cache
, dentry
);
1765 atomic_set(&p
->u
.count
, 1);
1768 dname
= dentry
->d_iname
;
1771 dentry
->d_name
.len
= name
->len
;
1772 dentry
->d_name
.hash
= name
->hash
;
1773 memcpy(dname
, name
->name
, name
->len
);
1774 dname
[name
->len
] = 0;
1776 /* Make sure we always see the terminating NUL character */
1777 smp_store_release(&dentry
->d_name
.name
, dname
); /* ^^^ */
1779 dentry
->d_lockref
.count
= 1;
1780 dentry
->d_flags
= 0;
1781 spin_lock_init(&dentry
->d_lock
);
1782 seqcount_spinlock_init(&dentry
->d_seq
, &dentry
->d_lock
);
1783 dentry
->d_inode
= NULL
;
1784 dentry
->d_parent
= dentry
;
1786 dentry
->d_op
= NULL
;
1787 dentry
->d_fsdata
= NULL
;
1788 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1789 INIT_LIST_HEAD(&dentry
->d_lru
);
1790 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1791 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1792 INIT_LIST_HEAD(&dentry
->d_child
);
1793 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1795 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1796 err
= dentry
->d_op
->d_init(dentry
);
1798 if (dname_external(dentry
))
1799 kfree(external_name(dentry
));
1800 kmem_cache_free(dentry_cache
, dentry
);
1805 this_cpu_inc(nr_dentry
);
1811 * d_alloc - allocate a dcache entry
1812 * @parent: parent of entry to allocate
1813 * @name: qstr of the name
1815 * Allocates a dentry. It returns %NULL if there is insufficient memory
1816 * available. On a success the dentry is returned. The name passed in is
1817 * copied and the copy passed in may be reused after this call.
1819 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1821 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1824 spin_lock(&parent
->d_lock
);
1826 * don't need child lock because it is not subject
1827 * to concurrency here
1829 __dget_dlock(parent
);
1830 dentry
->d_parent
= parent
;
1831 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1832 spin_unlock(&parent
->d_lock
);
1836 EXPORT_SYMBOL(d_alloc
);
1838 struct dentry
*d_alloc_anon(struct super_block
*sb
)
1840 return __d_alloc(sb
, NULL
);
1842 EXPORT_SYMBOL(d_alloc_anon
);
1844 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1846 struct dentry
*dentry
= d_alloc_anon(parent
->d_sb
);
1848 dentry
->d_flags
|= DCACHE_DENTRY_CURSOR
;
1849 dentry
->d_parent
= dget(parent
);
1855 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1856 * @sb: the superblock
1857 * @name: qstr of the name
1859 * For a filesystem that just pins its dentries in memory and never
1860 * performs lookups at all, return an unhashed IS_ROOT dentry.
1861 * This is used for pipes, sockets et.al. - the stuff that should
1862 * never be anyone's children or parents. Unlike all other
1863 * dentries, these will not have RCU delay between dropping the
1864 * last reference and freeing them.
1866 * The only user is alloc_file_pseudo() and that's what should
1867 * be considered a public interface. Don't use directly.
1869 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1871 struct dentry
*dentry
= __d_alloc(sb
, name
);
1873 dentry
->d_flags
|= DCACHE_NORCU
;
1877 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1882 q
.hash_len
= hashlen_string(parent
, name
);
1883 return d_alloc(parent
, &q
);
1885 EXPORT_SYMBOL(d_alloc_name
);
1887 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1889 WARN_ON_ONCE(dentry
->d_op
);
1890 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1892 DCACHE_OP_REVALIDATE
|
1893 DCACHE_OP_WEAK_REVALIDATE
|
1900 dentry
->d_flags
|= DCACHE_OP_HASH
;
1902 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1903 if (op
->d_revalidate
)
1904 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1905 if (op
->d_weak_revalidate
)
1906 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1908 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1910 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1912 dentry
->d_flags
|= DCACHE_OP_REAL
;
1915 EXPORT_SYMBOL(d_set_d_op
);
1919 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1920 * @dentry - The dentry to mark
1922 * Mark a dentry as falling through to the lower layer (as set with
1923 * d_pin_lower()). This flag may be recorded on the medium.
1925 void d_set_fallthru(struct dentry
*dentry
)
1927 spin_lock(&dentry
->d_lock
);
1928 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1929 spin_unlock(&dentry
->d_lock
);
1931 EXPORT_SYMBOL(d_set_fallthru
);
1933 static unsigned d_flags_for_inode(struct inode
*inode
)
1935 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1938 return DCACHE_MISS_TYPE
;
1940 if (S_ISDIR(inode
->i_mode
)) {
1941 add_flags
= DCACHE_DIRECTORY_TYPE
;
1942 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1943 if (unlikely(!inode
->i_op
->lookup
))
1944 add_flags
= DCACHE_AUTODIR_TYPE
;
1946 inode
->i_opflags
|= IOP_LOOKUP
;
1948 goto type_determined
;
1951 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1952 if (unlikely(inode
->i_op
->get_link
)) {
1953 add_flags
= DCACHE_SYMLINK_TYPE
;
1954 goto type_determined
;
1956 inode
->i_opflags
|= IOP_NOFOLLOW
;
1959 if (unlikely(!S_ISREG(inode
->i_mode
)))
1960 add_flags
= DCACHE_SPECIAL_TYPE
;
1963 if (unlikely(IS_AUTOMOUNT(inode
)))
1964 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1968 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1970 unsigned add_flags
= d_flags_for_inode(inode
);
1971 WARN_ON(d_in_lookup(dentry
));
1973 spin_lock(&dentry
->d_lock
);
1975 * Decrement negative dentry count if it was in the LRU list.
1977 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1978 this_cpu_dec(nr_dentry_negative
);
1979 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1980 raw_write_seqcount_begin(&dentry
->d_seq
);
1981 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1982 raw_write_seqcount_end(&dentry
->d_seq
);
1983 fsnotify_update_flags(dentry
);
1984 spin_unlock(&dentry
->d_lock
);
1988 * d_instantiate - fill in inode information for a dentry
1989 * @entry: dentry to complete
1990 * @inode: inode to attach to this dentry
1992 * Fill in inode information in the entry.
1994 * This turns negative dentries into productive full members
1997 * NOTE! This assumes that the inode count has been incremented
1998 * (or otherwise set) by the caller to indicate that it is now
1999 * in use by the dcache.
2002 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
2004 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
2006 security_d_instantiate(entry
, inode
);
2007 spin_lock(&inode
->i_lock
);
2008 __d_instantiate(entry
, inode
);
2009 spin_unlock(&inode
->i_lock
);
2012 EXPORT_SYMBOL(d_instantiate
);
2015 * This should be equivalent to d_instantiate() + unlock_new_inode(),
2016 * with lockdep-related part of unlock_new_inode() done before
2017 * anything else. Use that instead of open-coding d_instantiate()/
2018 * unlock_new_inode() combinations.
2020 void d_instantiate_new(struct dentry
*entry
, struct inode
*inode
)
2022 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
2024 lockdep_annotate_inode_mutex_key(inode
);
2025 security_d_instantiate(entry
, inode
);
2026 spin_lock(&inode
->i_lock
);
2027 __d_instantiate(entry
, inode
);
2028 WARN_ON(!(inode
->i_state
& I_NEW
));
2029 inode
->i_state
&= ~I_NEW
& ~I_CREATING
;
2031 wake_up_bit(&inode
->i_state
, __I_NEW
);
2032 spin_unlock(&inode
->i_lock
);
2034 EXPORT_SYMBOL(d_instantiate_new
);
2036 struct dentry
*d_make_root(struct inode
*root_inode
)
2038 struct dentry
*res
= NULL
;
2041 res
= d_alloc_anon(root_inode
->i_sb
);
2043 d_instantiate(res
, root_inode
);
2049 EXPORT_SYMBOL(d_make_root
);
2051 static struct dentry
*__d_instantiate_anon(struct dentry
*dentry
,
2052 struct inode
*inode
,
2058 security_d_instantiate(dentry
, inode
);
2059 spin_lock(&inode
->i_lock
);
2060 res
= __d_find_any_alias(inode
);
2062 spin_unlock(&inode
->i_lock
);
2067 /* attach a disconnected dentry */
2068 add_flags
= d_flags_for_inode(inode
);
2071 add_flags
|= DCACHE_DISCONNECTED
;
2073 spin_lock(&dentry
->d_lock
);
2074 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2075 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2076 if (!disconnected
) {
2077 hlist_bl_lock(&dentry
->d_sb
->s_roots
);
2078 hlist_bl_add_head(&dentry
->d_hash
, &dentry
->d_sb
->s_roots
);
2079 hlist_bl_unlock(&dentry
->d_sb
->s_roots
);
2081 spin_unlock(&dentry
->d_lock
);
2082 spin_unlock(&inode
->i_lock
);
2091 struct dentry
*d_instantiate_anon(struct dentry
*dentry
, struct inode
*inode
)
2093 return __d_instantiate_anon(dentry
, inode
, true);
2095 EXPORT_SYMBOL(d_instantiate_anon
);
2097 static struct dentry
*__d_obtain_alias(struct inode
*inode
, bool disconnected
)
2103 return ERR_PTR(-ESTALE
);
2105 return ERR_CAST(inode
);
2107 res
= d_find_any_alias(inode
);
2111 tmp
= d_alloc_anon(inode
->i_sb
);
2113 res
= ERR_PTR(-ENOMEM
);
2117 return __d_instantiate_anon(tmp
, inode
, disconnected
);
2125 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2126 * @inode: inode to allocate the dentry for
2128 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2129 * similar open by handle operations. The returned dentry may be anonymous,
2130 * or may have a full name (if the inode was already in the cache).
2132 * When called on a directory inode, we must ensure that the inode only ever
2133 * has one dentry. If a dentry is found, that is returned instead of
2134 * allocating a new one.
2136 * On successful return, the reference to the inode has been transferred
2137 * to the dentry. In case of an error the reference on the inode is released.
2138 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2139 * be passed in and the error will be propagated to the return value,
2140 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2142 struct dentry
*d_obtain_alias(struct inode
*inode
)
2144 return __d_obtain_alias(inode
, true);
2146 EXPORT_SYMBOL(d_obtain_alias
);
2149 * d_obtain_root - find or allocate a dentry for a given inode
2150 * @inode: inode to allocate the dentry for
2152 * Obtain an IS_ROOT dentry for the root of a filesystem.
2154 * We must ensure that directory inodes only ever have one dentry. If a
2155 * dentry is found, that is returned instead of allocating a new one.
2157 * On successful return, the reference to the inode has been transferred
2158 * to the dentry. In case of an error the reference on the inode is
2159 * released. A %NULL or IS_ERR inode may be passed in and will be the
2160 * error will be propagate to the return value, with a %NULL @inode
2161 * replaced by ERR_PTR(-ESTALE).
2163 struct dentry
*d_obtain_root(struct inode
*inode
)
2165 return __d_obtain_alias(inode
, false);
2167 EXPORT_SYMBOL(d_obtain_root
);
2170 * d_add_ci - lookup or allocate new dentry with case-exact name
2171 * @inode: the inode case-insensitive lookup has found
2172 * @dentry: the negative dentry that was passed to the parent's lookup func
2173 * @name: the case-exact name to be associated with the returned dentry
2175 * This is to avoid filling the dcache with case-insensitive names to the
2176 * same inode, only the actual correct case is stored in the dcache for
2177 * case-insensitive filesystems.
2179 * For a case-insensitive lookup match and if the the case-exact dentry
2180 * already exists in in the dcache, use it and return it.
2182 * If no entry exists with the exact case name, allocate new dentry with
2183 * the exact case, and return the spliced entry.
2185 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2188 struct dentry
*found
, *res
;
2191 * First check if a dentry matching the name already exists,
2192 * if not go ahead and create it now.
2194 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2199 if (d_in_lookup(dentry
)) {
2200 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2202 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2207 found
= d_alloc(dentry
->d_parent
, name
);
2210 return ERR_PTR(-ENOMEM
);
2213 res
= d_splice_alias(inode
, found
);
2220 EXPORT_SYMBOL(d_add_ci
);
2223 static inline bool d_same_name(const struct dentry
*dentry
,
2224 const struct dentry
*parent
,
2225 const struct qstr
*name
)
2227 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2228 if (dentry
->d_name
.len
!= name
->len
)
2230 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2232 return parent
->d_op
->d_compare(dentry
,
2233 dentry
->d_name
.len
, dentry
->d_name
.name
,
2238 * __d_lookup_rcu - search for a dentry (racy, store-free)
2239 * @parent: parent dentry
2240 * @name: qstr of name we wish to find
2241 * @seqp: returns d_seq value at the point where the dentry was found
2242 * Returns: dentry, or NULL
2244 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2245 * resolution (store-free path walking) design described in
2246 * Documentation/filesystems/path-lookup.txt.
2248 * This is not to be used outside core vfs.
2250 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2251 * held, and rcu_read_lock held. The returned dentry must not be stored into
2252 * without taking d_lock and checking d_seq sequence count against @seq
2255 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2258 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2259 * the returned dentry, so long as its parent's seqlock is checked after the
2260 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2261 * is formed, giving integrity down the path walk.
2263 * NOTE! The caller *has* to check the resulting dentry against the sequence
2264 * number we've returned before using any of the resulting dentry state!
2266 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2267 const struct qstr
*name
,
2270 u64 hashlen
= name
->hash_len
;
2271 const unsigned char *str
= name
->name
;
2272 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2273 struct hlist_bl_node
*node
;
2274 struct dentry
*dentry
;
2277 * Note: There is significant duplication with __d_lookup_rcu which is
2278 * required to prevent single threaded performance regressions
2279 * especially on architectures where smp_rmb (in seqcounts) are costly.
2280 * Keep the two functions in sync.
2284 * The hash list is protected using RCU.
2286 * Carefully use d_seq when comparing a candidate dentry, to avoid
2287 * races with d_move().
2289 * It is possible that concurrent renames can mess up our list
2290 * walk here and result in missing our dentry, resulting in the
2291 * false-negative result. d_lookup() protects against concurrent
2292 * renames using rename_lock seqlock.
2294 * See Documentation/filesystems/path-lookup.txt for more details.
2296 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2301 * The dentry sequence count protects us from concurrent
2302 * renames, and thus protects parent and name fields.
2304 * The caller must perform a seqcount check in order
2305 * to do anything useful with the returned dentry.
2307 * NOTE! We do a "raw" seqcount_begin here. That means that
2308 * we don't wait for the sequence count to stabilize if it
2309 * is in the middle of a sequence change. If we do the slow
2310 * dentry compare, we will do seqretries until it is stable,
2311 * and if we end up with a successful lookup, we actually
2312 * want to exit RCU lookup anyway.
2314 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2315 * we are still guaranteed NUL-termination of ->d_name.name.
2317 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2318 if (dentry
->d_parent
!= parent
)
2320 if (d_unhashed(dentry
))
2323 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2326 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2328 tlen
= dentry
->d_name
.len
;
2329 tname
= dentry
->d_name
.name
;
2330 /* we want a consistent (name,len) pair */
2331 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2335 if (parent
->d_op
->d_compare(dentry
,
2336 tlen
, tname
, name
) != 0)
2339 if (dentry
->d_name
.hash_len
!= hashlen
)
2341 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2351 * d_lookup - search for a dentry
2352 * @parent: parent dentry
2353 * @name: qstr of name we wish to find
2354 * Returns: dentry, or NULL
2356 * d_lookup searches the children of the parent dentry for the name in
2357 * question. If the dentry is found its reference count is incremented and the
2358 * dentry is returned. The caller must use dput to free the entry when it has
2359 * finished using it. %NULL is returned if the dentry does not exist.
2361 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2363 struct dentry
*dentry
;
2367 seq
= read_seqbegin(&rename_lock
);
2368 dentry
= __d_lookup(parent
, name
);
2371 } while (read_seqretry(&rename_lock
, seq
));
2374 EXPORT_SYMBOL(d_lookup
);
2377 * __d_lookup - search for a dentry (racy)
2378 * @parent: parent dentry
2379 * @name: qstr of name we wish to find
2380 * Returns: dentry, or NULL
2382 * __d_lookup is like d_lookup, however it may (rarely) return a
2383 * false-negative result due to unrelated rename activity.
2385 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2386 * however it must be used carefully, eg. with a following d_lookup in
2387 * the case of failure.
2389 * __d_lookup callers must be commented.
2391 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2393 unsigned int hash
= name
->hash
;
2394 struct hlist_bl_head
*b
= d_hash(hash
);
2395 struct hlist_bl_node
*node
;
2396 struct dentry
*found
= NULL
;
2397 struct dentry
*dentry
;
2400 * Note: There is significant duplication with __d_lookup_rcu which is
2401 * required to prevent single threaded performance regressions
2402 * especially on architectures where smp_rmb (in seqcounts) are costly.
2403 * Keep the two functions in sync.
2407 * The hash list is protected using RCU.
2409 * Take d_lock when comparing a candidate dentry, to avoid races
2412 * It is possible that concurrent renames can mess up our list
2413 * walk here and result in missing our dentry, resulting in the
2414 * false-negative result. d_lookup() protects against concurrent
2415 * renames using rename_lock seqlock.
2417 * See Documentation/filesystems/path-lookup.txt for more details.
2421 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2423 if (dentry
->d_name
.hash
!= hash
)
2426 spin_lock(&dentry
->d_lock
);
2427 if (dentry
->d_parent
!= parent
)
2429 if (d_unhashed(dentry
))
2432 if (!d_same_name(dentry
, parent
, name
))
2435 dentry
->d_lockref
.count
++;
2437 spin_unlock(&dentry
->d_lock
);
2440 spin_unlock(&dentry
->d_lock
);
2448 * d_hash_and_lookup - hash the qstr then search for a dentry
2449 * @dir: Directory to search in
2450 * @name: qstr of name we wish to find
2452 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2454 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2457 * Check for a fs-specific hash function. Note that we must
2458 * calculate the standard hash first, as the d_op->d_hash()
2459 * routine may choose to leave the hash value unchanged.
2461 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2462 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2463 int err
= dir
->d_op
->d_hash(dir
, name
);
2464 if (unlikely(err
< 0))
2465 return ERR_PTR(err
);
2467 return d_lookup(dir
, name
);
2469 EXPORT_SYMBOL(d_hash_and_lookup
);
2472 * When a file is deleted, we have two options:
2473 * - turn this dentry into a negative dentry
2474 * - unhash this dentry and free it.
2476 * Usually, we want to just turn this into
2477 * a negative dentry, but if anybody else is
2478 * currently using the dentry or the inode
2479 * we can't do that and we fall back on removing
2480 * it from the hash queues and waiting for
2481 * it to be deleted later when it has no users
2485 * d_delete - delete a dentry
2486 * @dentry: The dentry to delete
2488 * Turn the dentry into a negative dentry if possible, otherwise
2489 * remove it from the hash queues so it can be deleted later
2492 void d_delete(struct dentry
* dentry
)
2494 struct inode
*inode
= dentry
->d_inode
;
2496 spin_lock(&inode
->i_lock
);
2497 spin_lock(&dentry
->d_lock
);
2499 * Are we the only user?
2501 if (dentry
->d_lockref
.count
== 1) {
2502 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2503 dentry_unlink_inode(dentry
);
2506 spin_unlock(&dentry
->d_lock
);
2507 spin_unlock(&inode
->i_lock
);
2510 EXPORT_SYMBOL(d_delete
);
2512 static void __d_rehash(struct dentry
*entry
)
2514 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2517 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2522 * d_rehash - add an entry back to the hash
2523 * @entry: dentry to add to the hash
2525 * Adds a dentry to the hash according to its name.
2528 void d_rehash(struct dentry
* entry
)
2530 spin_lock(&entry
->d_lock
);
2532 spin_unlock(&entry
->d_lock
);
2534 EXPORT_SYMBOL(d_rehash
);
2536 static inline unsigned start_dir_add(struct inode
*dir
)
2540 unsigned n
= dir
->i_dir_seq
;
2541 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2547 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2549 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2552 static void d_wait_lookup(struct dentry
*dentry
)
2554 if (d_in_lookup(dentry
)) {
2555 DECLARE_WAITQUEUE(wait
, current
);
2556 add_wait_queue(dentry
->d_wait
, &wait
);
2558 set_current_state(TASK_UNINTERRUPTIBLE
);
2559 spin_unlock(&dentry
->d_lock
);
2561 spin_lock(&dentry
->d_lock
);
2562 } while (d_in_lookup(dentry
));
2566 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2567 const struct qstr
*name
,
2568 wait_queue_head_t
*wq
)
2570 unsigned int hash
= name
->hash
;
2571 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2572 struct hlist_bl_node
*node
;
2573 struct dentry
*new = d_alloc(parent
, name
);
2574 struct dentry
*dentry
;
2575 unsigned seq
, r_seq
, d_seq
;
2578 return ERR_PTR(-ENOMEM
);
2582 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
);
2583 r_seq
= read_seqbegin(&rename_lock
);
2584 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2585 if (unlikely(dentry
)) {
2586 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2590 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2599 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2604 if (unlikely(seq
& 1)) {
2610 if (unlikely(READ_ONCE(parent
->d_inode
->i_dir_seq
) != seq
)) {
2616 * No changes for the parent since the beginning of d_lookup().
2617 * Since all removals from the chain happen with hlist_bl_lock(),
2618 * any potential in-lookup matches are going to stay here until
2619 * we unlock the chain. All fields are stable in everything
2622 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2623 if (dentry
->d_name
.hash
!= hash
)
2625 if (dentry
->d_parent
!= parent
)
2627 if (!d_same_name(dentry
, parent
, name
))
2630 /* now we can try to grab a reference */
2631 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2638 * somebody is likely to be still doing lookup for it;
2639 * wait for them to finish
2641 spin_lock(&dentry
->d_lock
);
2642 d_wait_lookup(dentry
);
2644 * it's not in-lookup anymore; in principle we should repeat
2645 * everything from dcache lookup, but it's likely to be what
2646 * d_lookup() would've found anyway. If it is, just return it;
2647 * otherwise we really have to repeat the whole thing.
2649 if (unlikely(dentry
->d_name
.hash
!= hash
))
2651 if (unlikely(dentry
->d_parent
!= parent
))
2653 if (unlikely(d_unhashed(dentry
)))
2655 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2657 /* OK, it *is* a hashed match; return it */
2658 spin_unlock(&dentry
->d_lock
);
2663 /* we can't take ->d_lock here; it's OK, though. */
2664 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2666 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2670 spin_unlock(&dentry
->d_lock
);
2674 EXPORT_SYMBOL(d_alloc_parallel
);
2676 void __d_lookup_done(struct dentry
*dentry
)
2678 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2679 dentry
->d_name
.hash
);
2681 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2682 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2683 wake_up_all(dentry
->d_wait
);
2684 dentry
->d_wait
= NULL
;
2686 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2687 INIT_LIST_HEAD(&dentry
->d_lru
);
2689 EXPORT_SYMBOL(__d_lookup_done
);
2691 /* inode->i_lock held if inode is non-NULL */
2693 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2695 struct inode
*dir
= NULL
;
2697 spin_lock(&dentry
->d_lock
);
2698 if (unlikely(d_in_lookup(dentry
))) {
2699 dir
= dentry
->d_parent
->d_inode
;
2700 n
= start_dir_add(dir
);
2701 __d_lookup_done(dentry
);
2704 unsigned add_flags
= d_flags_for_inode(inode
);
2705 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2706 raw_write_seqcount_begin(&dentry
->d_seq
);
2707 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2708 raw_write_seqcount_end(&dentry
->d_seq
);
2709 fsnotify_update_flags(dentry
);
2713 end_dir_add(dir
, n
);
2714 spin_unlock(&dentry
->d_lock
);
2716 spin_unlock(&inode
->i_lock
);
2720 * d_add - add dentry to hash queues
2721 * @entry: dentry to add
2722 * @inode: The inode to attach to this dentry
2724 * This adds the entry to the hash queues and initializes @inode.
2725 * The entry was actually filled in earlier during d_alloc().
2728 void d_add(struct dentry
*entry
, struct inode
*inode
)
2731 security_d_instantiate(entry
, inode
);
2732 spin_lock(&inode
->i_lock
);
2734 __d_add(entry
, inode
);
2736 EXPORT_SYMBOL(d_add
);
2739 * d_exact_alias - find and hash an exact unhashed alias
2740 * @entry: dentry to add
2741 * @inode: The inode to go with this dentry
2743 * If an unhashed dentry with the same name/parent and desired
2744 * inode already exists, hash and return it. Otherwise, return
2747 * Parent directory should be locked.
2749 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2751 struct dentry
*alias
;
2752 unsigned int hash
= entry
->d_name
.hash
;
2754 spin_lock(&inode
->i_lock
);
2755 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2757 * Don't need alias->d_lock here, because aliases with
2758 * d_parent == entry->d_parent are not subject to name or
2759 * parent changes, because the parent inode i_mutex is held.
2761 if (alias
->d_name
.hash
!= hash
)
2763 if (alias
->d_parent
!= entry
->d_parent
)
2765 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2767 spin_lock(&alias
->d_lock
);
2768 if (!d_unhashed(alias
)) {
2769 spin_unlock(&alias
->d_lock
);
2772 __dget_dlock(alias
);
2774 spin_unlock(&alias
->d_lock
);
2776 spin_unlock(&inode
->i_lock
);
2779 spin_unlock(&inode
->i_lock
);
2782 EXPORT_SYMBOL(d_exact_alias
);
2784 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2786 if (unlikely(dname_external(target
))) {
2787 if (unlikely(dname_external(dentry
))) {
2789 * Both external: swap the pointers
2791 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2794 * dentry:internal, target:external. Steal target's
2795 * storage and make target internal.
2797 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2798 dentry
->d_name
.len
+ 1);
2799 dentry
->d_name
.name
= target
->d_name
.name
;
2800 target
->d_name
.name
= target
->d_iname
;
2803 if (unlikely(dname_external(dentry
))) {
2805 * dentry:external, target:internal. Give dentry's
2806 * storage to target and make dentry internal
2808 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2809 target
->d_name
.len
+ 1);
2810 target
->d_name
.name
= dentry
->d_name
.name
;
2811 dentry
->d_name
.name
= dentry
->d_iname
;
2814 * Both are internal.
2817 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2818 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2819 swap(((long *) &dentry
->d_iname
)[i
],
2820 ((long *) &target
->d_iname
)[i
]);
2824 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2827 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2829 struct external_name
*old_name
= NULL
;
2830 if (unlikely(dname_external(dentry
)))
2831 old_name
= external_name(dentry
);
2832 if (unlikely(dname_external(target
))) {
2833 atomic_inc(&external_name(target
)->u
.count
);
2834 dentry
->d_name
= target
->d_name
;
2836 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2837 target
->d_name
.len
+ 1);
2838 dentry
->d_name
.name
= dentry
->d_iname
;
2839 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2841 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2842 kfree_rcu(old_name
, u
.head
);
2846 * __d_move - move a dentry
2847 * @dentry: entry to move
2848 * @target: new dentry
2849 * @exchange: exchange the two dentries
2851 * Update the dcache to reflect the move of a file name. Negative
2852 * dcache entries should not be moved in this way. Caller must hold
2853 * rename_lock, the i_mutex of the source and target directories,
2854 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2856 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2859 struct dentry
*old_parent
, *p
;
2860 struct inode
*dir
= NULL
;
2863 WARN_ON(!dentry
->d_inode
);
2864 if (WARN_ON(dentry
== target
))
2867 BUG_ON(d_ancestor(target
, dentry
));
2868 old_parent
= dentry
->d_parent
;
2869 p
= d_ancestor(old_parent
, target
);
2870 if (IS_ROOT(dentry
)) {
2872 spin_lock(&target
->d_parent
->d_lock
);
2874 /* target is not a descendent of dentry->d_parent */
2875 spin_lock(&target
->d_parent
->d_lock
);
2876 spin_lock_nested(&old_parent
->d_lock
, DENTRY_D_LOCK_NESTED
);
2878 BUG_ON(p
== dentry
);
2879 spin_lock(&old_parent
->d_lock
);
2881 spin_lock_nested(&target
->d_parent
->d_lock
,
2882 DENTRY_D_LOCK_NESTED
);
2884 spin_lock_nested(&dentry
->d_lock
, 2);
2885 spin_lock_nested(&target
->d_lock
, 3);
2887 if (unlikely(d_in_lookup(target
))) {
2888 dir
= target
->d_parent
->d_inode
;
2889 n
= start_dir_add(dir
);
2890 __d_lookup_done(target
);
2893 write_seqcount_begin(&dentry
->d_seq
);
2894 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2897 if (!d_unhashed(dentry
))
2899 if (!d_unhashed(target
))
2902 /* ... and switch them in the tree */
2903 dentry
->d_parent
= target
->d_parent
;
2905 copy_name(dentry
, target
);
2906 target
->d_hash
.pprev
= NULL
;
2907 dentry
->d_parent
->d_lockref
.count
++;
2908 if (dentry
!= old_parent
) /* wasn't IS_ROOT */
2909 WARN_ON(!--old_parent
->d_lockref
.count
);
2911 target
->d_parent
= old_parent
;
2912 swap_names(dentry
, target
);
2913 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2915 fsnotify_update_flags(target
);
2917 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2919 fsnotify_update_flags(dentry
);
2920 fscrypt_handle_d_move(dentry
);
2922 write_seqcount_end(&target
->d_seq
);
2923 write_seqcount_end(&dentry
->d_seq
);
2926 end_dir_add(dir
, n
);
2928 if (dentry
->d_parent
!= old_parent
)
2929 spin_unlock(&dentry
->d_parent
->d_lock
);
2930 if (dentry
!= old_parent
)
2931 spin_unlock(&old_parent
->d_lock
);
2932 spin_unlock(&target
->d_lock
);
2933 spin_unlock(&dentry
->d_lock
);
2937 * d_move - move a dentry
2938 * @dentry: entry to move
2939 * @target: new dentry
2941 * Update the dcache to reflect the move of a file name. Negative
2942 * dcache entries should not be moved in this way. See the locking
2943 * requirements for __d_move.
2945 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2947 write_seqlock(&rename_lock
);
2948 __d_move(dentry
, target
, false);
2949 write_sequnlock(&rename_lock
);
2951 EXPORT_SYMBOL(d_move
);
2954 * d_exchange - exchange two dentries
2955 * @dentry1: first dentry
2956 * @dentry2: second dentry
2958 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2960 write_seqlock(&rename_lock
);
2962 WARN_ON(!dentry1
->d_inode
);
2963 WARN_ON(!dentry2
->d_inode
);
2964 WARN_ON(IS_ROOT(dentry1
));
2965 WARN_ON(IS_ROOT(dentry2
));
2967 __d_move(dentry1
, dentry2
, true);
2969 write_sequnlock(&rename_lock
);
2973 * d_ancestor - search for an ancestor
2974 * @p1: ancestor dentry
2977 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2978 * an ancestor of p2, else NULL.
2980 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2984 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2985 if (p
->d_parent
== p1
)
2992 * This helper attempts to cope with remotely renamed directories
2994 * It assumes that the caller is already holding
2995 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2997 * Note: If ever the locking in lock_rename() changes, then please
2998 * remember to update this too...
3000 static int __d_unalias(struct inode
*inode
,
3001 struct dentry
*dentry
, struct dentry
*alias
)
3003 struct mutex
*m1
= NULL
;
3004 struct rw_semaphore
*m2
= NULL
;
3007 /* If alias and dentry share a parent, then no extra locks required */
3008 if (alias
->d_parent
== dentry
->d_parent
)
3011 /* See lock_rename() */
3012 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
3014 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
3015 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
3017 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
3019 __d_move(alias
, dentry
, false);
3030 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3031 * @inode: the inode which may have a disconnected dentry
3032 * @dentry: a negative dentry which we want to point to the inode.
3034 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3035 * place of the given dentry and return it, else simply d_add the inode
3036 * to the dentry and return NULL.
3038 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3039 * we should error out: directories can't have multiple aliases.
3041 * This is needed in the lookup routine of any filesystem that is exportable
3042 * (via knfsd) so that we can build dcache paths to directories effectively.
3044 * If a dentry was found and moved, then it is returned. Otherwise NULL
3045 * is returned. This matches the expected return value of ->lookup.
3047 * Cluster filesystems may call this function with a negative, hashed dentry.
3048 * In that case, we know that the inode will be a regular file, and also this
3049 * will only occur during atomic_open. So we need to check for the dentry
3050 * being already hashed only in the final case.
3052 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
3055 return ERR_CAST(inode
);
3057 BUG_ON(!d_unhashed(dentry
));
3062 security_d_instantiate(dentry
, inode
);
3063 spin_lock(&inode
->i_lock
);
3064 if (S_ISDIR(inode
->i_mode
)) {
3065 struct dentry
*new = __d_find_any_alias(inode
);
3066 if (unlikely(new)) {
3067 /* The reference to new ensures it remains an alias */
3068 spin_unlock(&inode
->i_lock
);
3069 write_seqlock(&rename_lock
);
3070 if (unlikely(d_ancestor(new, dentry
))) {
3071 write_sequnlock(&rename_lock
);
3073 new = ERR_PTR(-ELOOP
);
3074 pr_warn_ratelimited(
3075 "VFS: Lookup of '%s' in %s %s"
3076 " would have caused loop\n",
3077 dentry
->d_name
.name
,
3078 inode
->i_sb
->s_type
->name
,
3080 } else if (!IS_ROOT(new)) {
3081 struct dentry
*old_parent
= dget(new->d_parent
);
3082 int err
= __d_unalias(inode
, dentry
, new);
3083 write_sequnlock(&rename_lock
);
3090 __d_move(new, dentry
, false);
3091 write_sequnlock(&rename_lock
);
3098 __d_add(dentry
, inode
);
3101 EXPORT_SYMBOL(d_splice_alias
);
3104 * Test whether new_dentry is a subdirectory of old_dentry.
3106 * Trivially implemented using the dcache structure
3110 * is_subdir - is new dentry a subdirectory of old_dentry
3111 * @new_dentry: new dentry
3112 * @old_dentry: old dentry
3114 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3115 * Returns false otherwise.
3116 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3119 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3124 if (new_dentry
== old_dentry
)
3128 /* for restarting inner loop in case of seq retry */
3129 seq
= read_seqbegin(&rename_lock
);
3131 * Need rcu_readlock to protect against the d_parent trashing
3135 if (d_ancestor(old_dentry
, new_dentry
))
3140 } while (read_seqretry(&rename_lock
, seq
));
3144 EXPORT_SYMBOL(is_subdir
);
3146 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3148 struct dentry
*root
= data
;
3149 if (dentry
!= root
) {
3150 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3153 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3154 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3155 dentry
->d_lockref
.count
--;
3158 return D_WALK_CONTINUE
;
3161 void d_genocide(struct dentry
*parent
)
3163 d_walk(parent
, parent
, d_genocide_kill
);
3166 EXPORT_SYMBOL(d_genocide
);
3168 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3170 inode_dec_link_count(inode
);
3171 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3172 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3173 !d_unlinked(dentry
));
3174 spin_lock(&dentry
->d_parent
->d_lock
);
3175 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3176 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3177 (unsigned long long)inode
->i_ino
);
3178 spin_unlock(&dentry
->d_lock
);
3179 spin_unlock(&dentry
->d_parent
->d_lock
);
3180 d_instantiate(dentry
, inode
);
3182 EXPORT_SYMBOL(d_tmpfile
);
3184 static __initdata
unsigned long dhash_entries
;
3185 static int __init
set_dhash_entries(char *str
)
3189 dhash_entries
= simple_strtoul(str
, &str
, 0);
3192 __setup("dhash_entries=", set_dhash_entries
);
3194 static void __init
dcache_init_early(void)
3196 /* If hashes are distributed across NUMA nodes, defer
3197 * hash allocation until vmalloc space is available.
3203 alloc_large_system_hash("Dentry cache",
3204 sizeof(struct hlist_bl_head
),
3207 HASH_EARLY
| HASH_ZERO
,
3212 d_hash_shift
= 32 - d_hash_shift
;
3215 static void __init
dcache_init(void)
3218 * A constructor could be added for stable state like the lists,
3219 * but it is probably not worth it because of the cache nature
3222 dentry_cache
= KMEM_CACHE_USERCOPY(dentry
,
3223 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
,
3226 /* Hash may have been set up in dcache_init_early */
3231 alloc_large_system_hash("Dentry cache",
3232 sizeof(struct hlist_bl_head
),
3240 d_hash_shift
= 32 - d_hash_shift
;
3243 /* SLAB cache for __getname() consumers */
3244 struct kmem_cache
*names_cachep __read_mostly
;
3245 EXPORT_SYMBOL(names_cachep
);
3247 void __init
vfs_caches_init_early(void)
3251 for (i
= 0; i
< ARRAY_SIZE(in_lookup_hashtable
); i
++)
3252 INIT_HLIST_BL_HEAD(&in_lookup_hashtable
[i
]);
3254 dcache_init_early();
3258 void __init
vfs_caches_init(void)
3260 names_cachep
= kmem_cache_create_usercopy("names_cache", PATH_MAX
, 0,
3261 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, 0, PATH_MAX
, NULL
);
3266 files_maxfiles_init();