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 __ro_after_init
;
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
);
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.
95 * This hash-function tries to avoid losing too many bits of hash
96 * information, yet avoid using a prime hash-size or similar.
99 static unsigned int d_hash_shift __ro_after_init
;
101 static struct hlist_bl_head
*dentry_hashtable __ro_after_init
;
103 static inline struct hlist_bl_head
*d_hash(unsigned int hash
)
105 return dentry_hashtable
+ (hash
>> d_hash_shift
);
108 #define IN_LOOKUP_SHIFT 10
109 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
111 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
114 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
115 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
118 struct dentry_stat_t
{
121 long age_limit
; /* age in seconds */
122 long want_pages
; /* pages requested by system */
123 long nr_negative
; /* # of unused negative dentries */
124 long dummy
; /* Reserved for future use */
127 static DEFINE_PER_CPU(long, nr_dentry
);
128 static DEFINE_PER_CPU(long, nr_dentry_unused
);
129 static DEFINE_PER_CPU(long, nr_dentry_negative
);
131 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
132 /* Statistics gathering. */
133 static struct dentry_stat_t dentry_stat
= {
138 * Here we resort to our own counters instead of using generic per-cpu counters
139 * for consistency with what the vfs inode code does. We are expected to harvest
140 * better code and performance by having our own specialized counters.
142 * Please note that the loop is done over all possible CPUs, not over all online
143 * CPUs. The reason for this is that we don't want to play games with CPUs going
144 * on and off. If one of them goes off, we will just keep their counters.
146 * glommer: See cffbc8a for details, and if you ever intend to change this,
147 * please update all vfs counters to match.
149 static long get_nr_dentry(void)
153 for_each_possible_cpu(i
)
154 sum
+= per_cpu(nr_dentry
, i
);
155 return sum
< 0 ? 0 : sum
;
158 static long get_nr_dentry_unused(void)
162 for_each_possible_cpu(i
)
163 sum
+= per_cpu(nr_dentry_unused
, i
);
164 return sum
< 0 ? 0 : sum
;
167 static long get_nr_dentry_negative(void)
172 for_each_possible_cpu(i
)
173 sum
+= per_cpu(nr_dentry_negative
, i
);
174 return sum
< 0 ? 0 : sum
;
177 static int proc_nr_dentry(struct ctl_table
*table
, int write
, void *buffer
,
178 size_t *lenp
, loff_t
*ppos
)
180 dentry_stat
.nr_dentry
= get_nr_dentry();
181 dentry_stat
.nr_unused
= get_nr_dentry_unused();
182 dentry_stat
.nr_negative
= get_nr_dentry_negative();
183 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
186 static struct ctl_table fs_dcache_sysctls
[] = {
188 .procname
= "dentry-state",
189 .data
= &dentry_stat
,
190 .maxlen
= 6*sizeof(long),
192 .proc_handler
= proc_nr_dentry
,
197 static int __init
init_fs_dcache_sysctls(void)
199 register_sysctl_init("fs", fs_dcache_sysctls
);
202 fs_initcall(init_fs_dcache_sysctls
);
206 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
207 * The strings are both count bytes long, and count is non-zero.
209 #ifdef CONFIG_DCACHE_WORD_ACCESS
211 #include <asm/word-at-a-time.h>
213 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
214 * aligned allocation for this particular component. We don't
215 * strictly need the load_unaligned_zeropad() safety, but it
216 * doesn't hurt either.
218 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
219 * need the careful unaligned handling.
221 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
223 unsigned long a
,b
,mask
;
226 a
= read_word_at_a_time(cs
);
227 b
= load_unaligned_zeropad(ct
);
228 if (tcount
< sizeof(unsigned long))
230 if (unlikely(a
!= b
))
232 cs
+= sizeof(unsigned long);
233 ct
+= sizeof(unsigned long);
234 tcount
-= sizeof(unsigned long);
238 mask
= bytemask_from_count(tcount
);
239 return unlikely(!!((a
^ b
) & mask
));
244 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
258 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
261 * Be careful about RCU walk racing with rename:
262 * use 'READ_ONCE' to fetch the name pointer.
264 * NOTE! Even if a rename will mean that the length
265 * was not loaded atomically, we don't care. The
266 * RCU walk will check the sequence count eventually,
267 * and catch it. And we won't overrun the buffer,
268 * because we're reading the name pointer atomically,
269 * and a dentry name is guaranteed to be properly
270 * terminated with a NUL byte.
272 * End result: even if 'len' is wrong, we'll exit
273 * early because the data cannot match (there can
274 * be no NUL in the ct/tcount data)
276 const unsigned char *cs
= READ_ONCE(dentry
->d_name
.name
);
278 return dentry_string_cmp(cs
, ct
, tcount
);
281 struct external_name
{
284 struct rcu_head head
;
286 unsigned char name
[];
289 static inline struct external_name
*external_name(struct dentry
*dentry
)
291 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
294 static void __d_free(struct rcu_head
*head
)
296 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
298 kmem_cache_free(dentry_cache
, dentry
);
301 static void __d_free_external(struct rcu_head
*head
)
303 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
304 kfree(external_name(dentry
));
305 kmem_cache_free(dentry_cache
, dentry
);
308 static inline int dname_external(const struct dentry
*dentry
)
310 return dentry
->d_name
.name
!= dentry
->d_iname
;
313 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
315 spin_lock(&dentry
->d_lock
);
316 name
->name
= dentry
->d_name
;
317 if (unlikely(dname_external(dentry
))) {
318 atomic_inc(&external_name(dentry
)->u
.count
);
320 memcpy(name
->inline_name
, dentry
->d_iname
,
321 dentry
->d_name
.len
+ 1);
322 name
->name
.name
= name
->inline_name
;
324 spin_unlock(&dentry
->d_lock
);
326 EXPORT_SYMBOL(take_dentry_name_snapshot
);
328 void release_dentry_name_snapshot(struct name_snapshot
*name
)
330 if (unlikely(name
->name
.name
!= name
->inline_name
)) {
331 struct external_name
*p
;
332 p
= container_of(name
->name
.name
, struct external_name
, name
[0]);
333 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
334 kfree_rcu(p
, u
.head
);
337 EXPORT_SYMBOL(release_dentry_name_snapshot
);
339 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
345 dentry
->d_inode
= inode
;
346 flags
= READ_ONCE(dentry
->d_flags
);
347 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
349 smp_store_release(&dentry
->d_flags
, flags
);
352 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
354 unsigned flags
= READ_ONCE(dentry
->d_flags
);
356 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
357 WRITE_ONCE(dentry
->d_flags
, flags
);
358 dentry
->d_inode
= NULL
;
359 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
360 this_cpu_inc(nr_dentry_negative
);
363 static void dentry_free(struct dentry
*dentry
)
365 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
366 if (unlikely(dname_external(dentry
))) {
367 struct external_name
*p
= external_name(dentry
);
368 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
369 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
373 /* if dentry was never visible to RCU, immediate free is OK */
374 if (dentry
->d_flags
& DCACHE_NORCU
)
375 __d_free(&dentry
->d_u
.d_rcu
);
377 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
381 * Release the dentry's inode, using the filesystem
382 * d_iput() operation if defined.
384 static void dentry_unlink_inode(struct dentry
* dentry
)
385 __releases(dentry
->d_lock
)
386 __releases(dentry
->d_inode
->i_lock
)
388 struct inode
*inode
= dentry
->d_inode
;
390 raw_write_seqcount_begin(&dentry
->d_seq
);
391 __d_clear_type_and_inode(dentry
);
392 hlist_del_init(&dentry
->d_u
.d_alias
);
393 raw_write_seqcount_end(&dentry
->d_seq
);
394 spin_unlock(&dentry
->d_lock
);
395 spin_unlock(&inode
->i_lock
);
397 fsnotify_inoderemove(inode
);
398 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
399 dentry
->d_op
->d_iput(dentry
, inode
);
405 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
406 * is in use - which includes both the "real" per-superblock
407 * LRU list _and_ the DCACHE_SHRINK_LIST use.
409 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
410 * on the shrink list (ie not on the superblock LRU list).
412 * The per-cpu "nr_dentry_unused" counters are updated with
413 * the DCACHE_LRU_LIST bit.
415 * The per-cpu "nr_dentry_negative" counters are only updated
416 * when deleted from or added to the per-superblock LRU list, not
417 * from/to the shrink list. That is to avoid an unneeded dec/inc
418 * pair when moving from LRU to shrink list in select_collect().
420 * These helper functions make sure we always follow the
421 * rules. d_lock must be held by the caller.
423 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
424 static void d_lru_add(struct dentry
*dentry
)
426 D_FLAG_VERIFY(dentry
, 0);
427 dentry
->d_flags
|= DCACHE_LRU_LIST
;
428 this_cpu_inc(nr_dentry_unused
);
429 if (d_is_negative(dentry
))
430 this_cpu_inc(nr_dentry_negative
);
431 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
434 static void d_lru_del(struct dentry
*dentry
)
436 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
437 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
438 this_cpu_dec(nr_dentry_unused
);
439 if (d_is_negative(dentry
))
440 this_cpu_dec(nr_dentry_negative
);
441 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
444 static void d_shrink_del(struct dentry
*dentry
)
446 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
447 list_del_init(&dentry
->d_lru
);
448 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
449 this_cpu_dec(nr_dentry_unused
);
452 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
454 D_FLAG_VERIFY(dentry
, 0);
455 list_add(&dentry
->d_lru
, list
);
456 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
457 this_cpu_inc(nr_dentry_unused
);
461 * These can only be called under the global LRU lock, ie during the
462 * callback for freeing the LRU list. "isolate" removes it from the
463 * LRU lists entirely, while shrink_move moves it to the indicated
466 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
468 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
469 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
470 this_cpu_dec(nr_dentry_unused
);
471 if (d_is_negative(dentry
))
472 this_cpu_dec(nr_dentry_negative
);
473 list_lru_isolate(lru
, &dentry
->d_lru
);
476 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
477 struct list_head
*list
)
479 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
480 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
481 if (d_is_negative(dentry
))
482 this_cpu_dec(nr_dentry_negative
);
483 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
486 static void ___d_drop(struct dentry
*dentry
)
488 struct hlist_bl_head
*b
;
490 * Hashed dentries are normally on the dentry hashtable,
491 * with the exception of those newly allocated by
492 * d_obtain_root, which are always IS_ROOT:
494 if (unlikely(IS_ROOT(dentry
)))
495 b
= &dentry
->d_sb
->s_roots
;
497 b
= d_hash(dentry
->d_name
.hash
);
500 __hlist_bl_del(&dentry
->d_hash
);
504 void __d_drop(struct dentry
*dentry
)
506 if (!d_unhashed(dentry
)) {
508 dentry
->d_hash
.pprev
= NULL
;
509 write_seqcount_invalidate(&dentry
->d_seq
);
512 EXPORT_SYMBOL(__d_drop
);
515 * d_drop - drop a dentry
516 * @dentry: dentry to drop
518 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
519 * be found through a VFS lookup any more. Note that this is different from
520 * deleting the dentry - d_delete will try to mark the dentry negative if
521 * possible, giving a successful _negative_ lookup, while d_drop will
522 * just make the cache lookup fail.
524 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
525 * reason (NFS timeouts or autofs deletes).
527 * __d_drop requires dentry->d_lock
529 * ___d_drop doesn't mark dentry as "unhashed"
530 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
532 void d_drop(struct dentry
*dentry
)
534 spin_lock(&dentry
->d_lock
);
536 spin_unlock(&dentry
->d_lock
);
538 EXPORT_SYMBOL(d_drop
);
540 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
544 * Inform d_walk() and shrink_dentry_list() that we are no longer
545 * attached to the dentry tree
547 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
548 if (unlikely(list_empty(&dentry
->d_child
)))
550 __list_del_entry(&dentry
->d_child
);
552 * Cursors can move around the list of children. While we'd been
553 * a normal list member, it didn't matter - ->d_child.next would've
554 * been updated. However, from now on it won't be and for the
555 * things like d_walk() it might end up with a nasty surprise.
556 * Normally d_walk() doesn't care about cursors moving around -
557 * ->d_lock on parent prevents that and since a cursor has no children
558 * of its own, we get through it without ever unlocking the parent.
559 * There is one exception, though - if we ascend from a child that
560 * gets killed as soon as we unlock it, the next sibling is found
561 * using the value left in its ->d_child.next. And if _that_
562 * pointed to a cursor, and cursor got moved (e.g. by lseek())
563 * before d_walk() regains parent->d_lock, we'll end up skipping
564 * everything the cursor had been moved past.
566 * Solution: make sure that the pointer left behind in ->d_child.next
567 * points to something that won't be moving around. I.e. skip the
570 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
571 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
572 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
574 dentry
->d_child
.next
= next
->d_child
.next
;
578 static void __dentry_kill(struct dentry
*dentry
)
580 struct dentry
*parent
= NULL
;
581 bool can_free
= true;
582 if (!IS_ROOT(dentry
))
583 parent
= dentry
->d_parent
;
586 * The dentry is now unrecoverably dead to the world.
588 lockref_mark_dead(&dentry
->d_lockref
);
591 * inform the fs via d_prune that this dentry is about to be
592 * unhashed and destroyed.
594 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
595 dentry
->d_op
->d_prune(dentry
);
597 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
598 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
601 /* if it was on the hash then remove it */
603 dentry_unlist(dentry
, parent
);
605 spin_unlock(&parent
->d_lock
);
607 dentry_unlink_inode(dentry
);
609 spin_unlock(&dentry
->d_lock
);
610 this_cpu_dec(nr_dentry
);
611 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
612 dentry
->d_op
->d_release(dentry
);
614 spin_lock(&dentry
->d_lock
);
615 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
616 dentry
->d_flags
|= DCACHE_MAY_FREE
;
619 spin_unlock(&dentry
->d_lock
);
620 if (likely(can_free
))
625 static struct dentry
*__lock_parent(struct dentry
*dentry
)
627 struct dentry
*parent
;
629 spin_unlock(&dentry
->d_lock
);
631 parent
= READ_ONCE(dentry
->d_parent
);
632 spin_lock(&parent
->d_lock
);
634 * We can't blindly lock dentry until we are sure
635 * that we won't violate the locking order.
636 * Any changes of dentry->d_parent must have
637 * been done with parent->d_lock held, so
638 * spin_lock() above is enough of a barrier
639 * for checking if it's still our child.
641 if (unlikely(parent
!= dentry
->d_parent
)) {
642 spin_unlock(&parent
->d_lock
);
646 if (parent
!= dentry
)
647 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
653 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
655 struct dentry
*parent
= dentry
->d_parent
;
658 if (likely(spin_trylock(&parent
->d_lock
)))
660 return __lock_parent(dentry
);
663 static inline bool retain_dentry(struct dentry
*dentry
)
665 WARN_ON(d_in_lookup(dentry
));
667 /* Unreachable? Get rid of it */
668 if (unlikely(d_unhashed(dentry
)))
671 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
674 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
675 if (dentry
->d_op
->d_delete(dentry
))
679 if (unlikely(dentry
->d_flags
& DCACHE_DONTCACHE
))
682 /* retain; LRU fodder */
683 dentry
->d_lockref
.count
--;
684 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
686 else if (unlikely(!(dentry
->d_flags
& DCACHE_REFERENCED
)))
687 dentry
->d_flags
|= DCACHE_REFERENCED
;
691 void d_mark_dontcache(struct inode
*inode
)
695 spin_lock(&inode
->i_lock
);
696 hlist_for_each_entry(de
, &inode
->i_dentry
, d_u
.d_alias
) {
697 spin_lock(&de
->d_lock
);
698 de
->d_flags
|= DCACHE_DONTCACHE
;
699 spin_unlock(&de
->d_lock
);
701 inode
->i_state
|= I_DONTCACHE
;
702 spin_unlock(&inode
->i_lock
);
704 EXPORT_SYMBOL(d_mark_dontcache
);
707 * Finish off a dentry we've decided to kill.
708 * dentry->d_lock must be held, returns with it unlocked.
709 * Returns dentry requiring refcount drop, or NULL if we're done.
711 static struct dentry
*dentry_kill(struct dentry
*dentry
)
712 __releases(dentry
->d_lock
)
714 struct inode
*inode
= dentry
->d_inode
;
715 struct dentry
*parent
= NULL
;
717 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
720 if (!IS_ROOT(dentry
)) {
721 parent
= dentry
->d_parent
;
722 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
723 parent
= __lock_parent(dentry
);
724 if (likely(inode
|| !dentry
->d_inode
))
726 /* negative that became positive */
728 spin_unlock(&parent
->d_lock
);
729 inode
= dentry
->d_inode
;
733 __dentry_kill(dentry
);
737 spin_unlock(&dentry
->d_lock
);
738 spin_lock(&inode
->i_lock
);
739 spin_lock(&dentry
->d_lock
);
740 parent
= lock_parent(dentry
);
742 if (unlikely(dentry
->d_lockref
.count
!= 1)) {
743 dentry
->d_lockref
.count
--;
744 } else if (likely(!retain_dentry(dentry
))) {
745 __dentry_kill(dentry
);
748 /* we are keeping it, after all */
750 spin_unlock(&inode
->i_lock
);
752 spin_unlock(&parent
->d_lock
);
753 spin_unlock(&dentry
->d_lock
);
758 * Try to do a lockless dput(), and return whether that was successful.
760 * If unsuccessful, we return false, having already taken the dentry lock.
762 * The caller needs to hold the RCU read lock, so that the dentry is
763 * guaranteed to stay around even if the refcount goes down to zero!
765 static inline bool fast_dput(struct dentry
*dentry
)
768 unsigned int d_flags
;
771 * If we have a d_op->d_delete() operation, we sould not
772 * let the dentry count go to zero, so use "put_or_lock".
774 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
775 return lockref_put_or_lock(&dentry
->d_lockref
);
778 * .. otherwise, we can try to just decrement the
779 * lockref optimistically.
781 ret
= lockref_put_return(&dentry
->d_lockref
);
784 * If the lockref_put_return() failed due to the lock being held
785 * by somebody else, the fast path has failed. We will need to
786 * get the lock, and then check the count again.
788 if (unlikely(ret
< 0)) {
789 spin_lock(&dentry
->d_lock
);
790 if (dentry
->d_lockref
.count
> 1) {
791 dentry
->d_lockref
.count
--;
792 spin_unlock(&dentry
->d_lock
);
799 * If we weren't the last ref, we're done.
805 * Careful, careful. The reference count went down
806 * to zero, but we don't hold the dentry lock, so
807 * somebody else could get it again, and do another
808 * dput(), and we need to not race with that.
810 * However, there is a very special and common case
811 * where we don't care, because there is nothing to
812 * do: the dentry is still hashed, it does not have
813 * a 'delete' op, and it's referenced and already on
816 * NOTE! Since we aren't locked, these values are
817 * not "stable". However, it is sufficient that at
818 * some point after we dropped the reference the
819 * dentry was hashed and the flags had the proper
820 * value. Other dentry users may have re-gotten
821 * a reference to the dentry and change that, but
822 * our work is done - we can leave the dentry
823 * around with a zero refcount.
825 * Nevertheless, there are two cases that we should kill
827 * 1. free disconnected dentries as soon as their refcount
829 * 2. free dentries if they should not be cached.
832 d_flags
= READ_ONCE(dentry
->d_flags
);
833 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
|
834 DCACHE_DISCONNECTED
| DCACHE_DONTCACHE
;
836 /* Nothing to do? Dropping the reference was all we needed? */
837 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
841 * Not the fast normal case? Get the lock. We've already decremented
842 * the refcount, but we'll need to re-check the situation after
845 spin_lock(&dentry
->d_lock
);
848 * Did somebody else grab a reference to it in the meantime, and
849 * we're no longer the last user after all? Alternatively, somebody
850 * else could have killed it and marked it dead. Either way, we
851 * don't need to do anything else.
853 if (dentry
->d_lockref
.count
) {
854 spin_unlock(&dentry
->d_lock
);
859 * Re-get the reference we optimistically dropped. We hold the
860 * lock, and we just tested that it was zero, so we can just
863 dentry
->d_lockref
.count
= 1;
871 * This is complicated by the fact that we do not want to put
872 * dentries that are no longer on any hash chain on the unused
873 * list: we'd much rather just get rid of them immediately.
875 * However, that implies that we have to traverse the dentry
876 * tree upwards to the parents which might _also_ now be
877 * scheduled for deletion (it may have been only waiting for
878 * its last child to go away).
880 * This tail recursion is done by hand as we don't want to depend
881 * on the compiler to always get this right (gcc generally doesn't).
882 * Real recursion would eat up our stack space.
886 * dput - release a dentry
887 * @dentry: dentry to release
889 * Release a dentry. This will drop the usage count and if appropriate
890 * call the dentry unlink method as well as removing it from the queues and
891 * releasing its resources. If the parent dentries were scheduled for release
892 * they too may now get deleted.
894 void dput(struct dentry
*dentry
)
900 if (likely(fast_dput(dentry
))) {
905 /* Slow case: now with the dentry lock held */
908 if (likely(retain_dentry(dentry
))) {
909 spin_unlock(&dentry
->d_lock
);
913 dentry
= dentry_kill(dentry
);
918 static void __dput_to_list(struct dentry
*dentry
, struct list_head
*list
)
919 __must_hold(&dentry
->d_lock
)
921 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
922 /* let the owner of the list it's on deal with it */
923 --dentry
->d_lockref
.count
;
925 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
927 if (!--dentry
->d_lockref
.count
)
928 d_shrink_add(dentry
, list
);
932 void dput_to_list(struct dentry
*dentry
, struct list_head
*list
)
935 if (likely(fast_dput(dentry
))) {
940 if (!retain_dentry(dentry
))
941 __dput_to_list(dentry
, list
);
942 spin_unlock(&dentry
->d_lock
);
945 /* This must be called with d_lock held */
946 static inline void __dget_dlock(struct dentry
*dentry
)
948 dentry
->d_lockref
.count
++;
951 static inline void __dget(struct dentry
*dentry
)
953 lockref_get(&dentry
->d_lockref
);
956 struct dentry
*dget_parent(struct dentry
*dentry
)
963 * Do optimistic parent lookup without any
967 seq
= raw_seqcount_begin(&dentry
->d_seq
);
968 ret
= READ_ONCE(dentry
->d_parent
);
969 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
971 if (likely(gotref
)) {
972 if (!read_seqcount_retry(&dentry
->d_seq
, seq
))
979 * Don't need rcu_dereference because we re-check it was correct under
983 ret
= dentry
->d_parent
;
984 spin_lock(&ret
->d_lock
);
985 if (unlikely(ret
!= dentry
->d_parent
)) {
986 spin_unlock(&ret
->d_lock
);
991 BUG_ON(!ret
->d_lockref
.count
);
992 ret
->d_lockref
.count
++;
993 spin_unlock(&ret
->d_lock
);
996 EXPORT_SYMBOL(dget_parent
);
998 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1000 struct dentry
*alias
;
1002 if (hlist_empty(&inode
->i_dentry
))
1004 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1010 * d_find_any_alias - find any alias for a given inode
1011 * @inode: inode to find an alias for
1013 * If any aliases exist for the given inode, take and return a
1014 * reference for one of them. If no aliases exist, return %NULL.
1016 struct dentry
*d_find_any_alias(struct inode
*inode
)
1020 spin_lock(&inode
->i_lock
);
1021 de
= __d_find_any_alias(inode
);
1022 spin_unlock(&inode
->i_lock
);
1025 EXPORT_SYMBOL(d_find_any_alias
);
1027 static struct dentry
*__d_find_alias(struct inode
*inode
)
1029 struct dentry
*alias
;
1031 if (S_ISDIR(inode
->i_mode
))
1032 return __d_find_any_alias(inode
);
1034 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1035 spin_lock(&alias
->d_lock
);
1036 if (!d_unhashed(alias
)) {
1037 __dget_dlock(alias
);
1038 spin_unlock(&alias
->d_lock
);
1041 spin_unlock(&alias
->d_lock
);
1047 * d_find_alias - grab a hashed alias of inode
1048 * @inode: inode in question
1050 * If inode has a hashed alias, or is a directory and has any alias,
1051 * acquire the reference to alias and return it. Otherwise return NULL.
1052 * Notice that if inode is a directory there can be only one alias and
1053 * it can be unhashed only if it has no children, or if it is the root
1054 * of a filesystem, or if the directory was renamed and d_revalidate
1055 * was the first vfs operation to notice.
1057 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1058 * any other hashed alias over that one.
1060 struct dentry
*d_find_alias(struct inode
*inode
)
1062 struct dentry
*de
= NULL
;
1064 if (!hlist_empty(&inode
->i_dentry
)) {
1065 spin_lock(&inode
->i_lock
);
1066 de
= __d_find_alias(inode
);
1067 spin_unlock(&inode
->i_lock
);
1071 EXPORT_SYMBOL(d_find_alias
);
1074 * Caller MUST be holding rcu_read_lock() and be guaranteed
1075 * that inode won't get freed until rcu_read_unlock().
1077 struct dentry
*d_find_alias_rcu(struct inode
*inode
)
1079 struct hlist_head
*l
= &inode
->i_dentry
;
1080 struct dentry
*de
= NULL
;
1082 spin_lock(&inode
->i_lock
);
1083 // ->i_dentry and ->i_rcu are colocated, but the latter won't be
1084 // used without having I_FREEING set, which means no aliases left
1085 if (likely(!(inode
->i_state
& I_FREEING
) && !hlist_empty(l
))) {
1086 if (S_ISDIR(inode
->i_mode
)) {
1087 de
= hlist_entry(l
->first
, struct dentry
, d_u
.d_alias
);
1089 hlist_for_each_entry(de
, l
, d_u
.d_alias
)
1090 if (!d_unhashed(de
))
1094 spin_unlock(&inode
->i_lock
);
1099 * Try to kill dentries associated with this inode.
1100 * WARNING: you must own a reference to inode.
1102 void d_prune_aliases(struct inode
*inode
)
1104 struct dentry
*dentry
;
1106 spin_lock(&inode
->i_lock
);
1107 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
1108 spin_lock(&dentry
->d_lock
);
1109 if (!dentry
->d_lockref
.count
) {
1110 struct dentry
*parent
= lock_parent(dentry
);
1111 if (likely(!dentry
->d_lockref
.count
)) {
1112 __dentry_kill(dentry
);
1117 spin_unlock(&parent
->d_lock
);
1119 spin_unlock(&dentry
->d_lock
);
1121 spin_unlock(&inode
->i_lock
);
1123 EXPORT_SYMBOL(d_prune_aliases
);
1126 * Lock a dentry from shrink list.
1127 * Called under rcu_read_lock() and dentry->d_lock; the former
1128 * guarantees that nothing we access will be freed under us.
1129 * Note that dentry is *not* protected from concurrent dentry_kill(),
1132 * Return false if dentry has been disrupted or grabbed, leaving
1133 * the caller to kick it off-list. Otherwise, return true and have
1134 * that dentry's inode and parent both locked.
1136 static bool shrink_lock_dentry(struct dentry
*dentry
)
1138 struct inode
*inode
;
1139 struct dentry
*parent
;
1141 if (dentry
->d_lockref
.count
)
1144 inode
= dentry
->d_inode
;
1145 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1146 spin_unlock(&dentry
->d_lock
);
1147 spin_lock(&inode
->i_lock
);
1148 spin_lock(&dentry
->d_lock
);
1149 if (unlikely(dentry
->d_lockref
.count
))
1151 /* changed inode means that somebody had grabbed it */
1152 if (unlikely(inode
!= dentry
->d_inode
))
1156 parent
= dentry
->d_parent
;
1157 if (IS_ROOT(dentry
) || likely(spin_trylock(&parent
->d_lock
)))
1160 spin_unlock(&dentry
->d_lock
);
1161 spin_lock(&parent
->d_lock
);
1162 if (unlikely(parent
!= dentry
->d_parent
)) {
1163 spin_unlock(&parent
->d_lock
);
1164 spin_lock(&dentry
->d_lock
);
1167 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1168 if (likely(!dentry
->d_lockref
.count
))
1170 spin_unlock(&parent
->d_lock
);
1173 spin_unlock(&inode
->i_lock
);
1177 void shrink_dentry_list(struct list_head
*list
)
1179 while (!list_empty(list
)) {
1180 struct dentry
*dentry
, *parent
;
1182 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
1183 spin_lock(&dentry
->d_lock
);
1185 if (!shrink_lock_dentry(dentry
)) {
1186 bool can_free
= false;
1188 d_shrink_del(dentry
);
1189 if (dentry
->d_lockref
.count
< 0)
1190 can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
1191 spin_unlock(&dentry
->d_lock
);
1193 dentry_free(dentry
);
1197 d_shrink_del(dentry
);
1198 parent
= dentry
->d_parent
;
1199 if (parent
!= dentry
)
1200 __dput_to_list(parent
, list
);
1201 __dentry_kill(dentry
);
1205 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1206 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1208 struct list_head
*freeable
= arg
;
1209 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1213 * we are inverting the lru lock/dentry->d_lock here,
1214 * so use a trylock. If we fail to get the lock, just skip
1217 if (!spin_trylock(&dentry
->d_lock
))
1221 * Referenced dentries are still in use. If they have active
1222 * counts, just remove them from the LRU. Otherwise give them
1223 * another pass through the LRU.
1225 if (dentry
->d_lockref
.count
) {
1226 d_lru_isolate(lru
, dentry
);
1227 spin_unlock(&dentry
->d_lock
);
1231 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1232 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1233 spin_unlock(&dentry
->d_lock
);
1236 * The list move itself will be made by the common LRU code. At
1237 * this point, we've dropped the dentry->d_lock but keep the
1238 * lru lock. This is safe to do, since every list movement is
1239 * protected by the lru lock even if both locks are held.
1241 * This is guaranteed by the fact that all LRU management
1242 * functions are intermediated by the LRU API calls like
1243 * list_lru_add and list_lru_del. List movement in this file
1244 * only ever occur through this functions or through callbacks
1245 * like this one, that are called from the LRU API.
1247 * The only exceptions to this are functions like
1248 * shrink_dentry_list, and code that first checks for the
1249 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1250 * operating only with stack provided lists after they are
1251 * properly isolated from the main list. It is thus, always a
1257 d_lru_shrink_move(lru
, dentry
, freeable
);
1258 spin_unlock(&dentry
->d_lock
);
1264 * prune_dcache_sb - shrink the dcache
1266 * @sc: shrink control, passed to list_lru_shrink_walk()
1268 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1269 * is done when we need more memory and called from the superblock shrinker
1272 * This function may fail to free any resources if all the dentries are in
1275 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1280 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1281 dentry_lru_isolate
, &dispose
);
1282 shrink_dentry_list(&dispose
);
1286 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1287 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1289 struct list_head
*freeable
= arg
;
1290 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1293 * we are inverting the lru lock/dentry->d_lock here,
1294 * so use a trylock. If we fail to get the lock, just skip
1297 if (!spin_trylock(&dentry
->d_lock
))
1300 d_lru_shrink_move(lru
, dentry
, freeable
);
1301 spin_unlock(&dentry
->d_lock
);
1308 * shrink_dcache_sb - shrink dcache for a superblock
1311 * Shrink the dcache for the specified super block. This is used to free
1312 * the dcache before unmounting a file system.
1314 void shrink_dcache_sb(struct super_block
*sb
)
1319 list_lru_walk(&sb
->s_dentry_lru
,
1320 dentry_lru_isolate_shrink
, &dispose
, 1024);
1321 shrink_dentry_list(&dispose
);
1322 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1324 EXPORT_SYMBOL(shrink_dcache_sb
);
1327 * enum d_walk_ret - action to talke during tree walk
1328 * @D_WALK_CONTINUE: contrinue walk
1329 * @D_WALK_QUIT: quit walk
1330 * @D_WALK_NORETRY: quit when retry is needed
1331 * @D_WALK_SKIP: skip this dentry and its children
1341 * d_walk - walk the dentry tree
1342 * @parent: start of walk
1343 * @data: data passed to @enter() and @finish()
1344 * @enter: callback when first entering the dentry
1346 * The @enter() callbacks are called with d_lock held.
1348 static void d_walk(struct dentry
*parent
, void *data
,
1349 enum d_walk_ret (*enter
)(void *, struct dentry
*))
1351 struct dentry
*this_parent
;
1352 struct list_head
*next
;
1354 enum d_walk_ret ret
;
1358 read_seqbegin_or_lock(&rename_lock
, &seq
);
1359 this_parent
= parent
;
1360 spin_lock(&this_parent
->d_lock
);
1362 ret
= enter(data
, this_parent
);
1364 case D_WALK_CONTINUE
:
1369 case D_WALK_NORETRY
:
1374 next
= this_parent
->d_subdirs
.next
;
1376 while (next
!= &this_parent
->d_subdirs
) {
1377 struct list_head
*tmp
= next
;
1378 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1381 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1384 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1386 ret
= enter(data
, dentry
);
1388 case D_WALK_CONTINUE
:
1391 spin_unlock(&dentry
->d_lock
);
1393 case D_WALK_NORETRY
:
1397 spin_unlock(&dentry
->d_lock
);
1401 if (!list_empty(&dentry
->d_subdirs
)) {
1402 spin_unlock(&this_parent
->d_lock
);
1403 spin_release(&dentry
->d_lock
.dep_map
, _RET_IP_
);
1404 this_parent
= dentry
;
1405 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1408 spin_unlock(&dentry
->d_lock
);
1411 * All done at this level ... ascend and resume the search.
1415 if (this_parent
!= parent
) {
1416 struct dentry
*child
= this_parent
;
1417 this_parent
= child
->d_parent
;
1419 spin_unlock(&child
->d_lock
);
1420 spin_lock(&this_parent
->d_lock
);
1422 /* might go back up the wrong parent if we have had a rename. */
1423 if (need_seqretry(&rename_lock
, seq
))
1425 /* go into the first sibling still alive */
1427 next
= child
->d_child
.next
;
1428 if (next
== &this_parent
->d_subdirs
)
1430 child
= list_entry(next
, struct dentry
, d_child
);
1431 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1435 if (need_seqretry(&rename_lock
, seq
))
1440 spin_unlock(&this_parent
->d_lock
);
1441 done_seqretry(&rename_lock
, seq
);
1445 spin_unlock(&this_parent
->d_lock
);
1454 struct check_mount
{
1455 struct vfsmount
*mnt
;
1456 unsigned int mounted
;
1459 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1461 struct check_mount
*info
= data
;
1462 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1464 if (likely(!d_mountpoint(dentry
)))
1465 return D_WALK_CONTINUE
;
1466 if (__path_is_mountpoint(&path
)) {
1470 return D_WALK_CONTINUE
;
1474 * path_has_submounts - check for mounts over a dentry in the
1475 * current namespace.
1476 * @parent: path to check.
1478 * Return true if the parent or its subdirectories contain
1479 * a mount point in the current namespace.
1481 int path_has_submounts(const struct path
*parent
)
1483 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1485 read_seqlock_excl(&mount_lock
);
1486 d_walk(parent
->dentry
, &data
, path_check_mount
);
1487 read_sequnlock_excl(&mount_lock
);
1489 return data
.mounted
;
1491 EXPORT_SYMBOL(path_has_submounts
);
1494 * Called by mount code to set a mountpoint and check if the mountpoint is
1495 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1496 * subtree can become unreachable).
1498 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1499 * this reason take rename_lock and d_lock on dentry and ancestors.
1501 int d_set_mounted(struct dentry
*dentry
)
1505 write_seqlock(&rename_lock
);
1506 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1507 /* Need exclusion wrt. d_invalidate() */
1508 spin_lock(&p
->d_lock
);
1509 if (unlikely(d_unhashed(p
))) {
1510 spin_unlock(&p
->d_lock
);
1513 spin_unlock(&p
->d_lock
);
1515 spin_lock(&dentry
->d_lock
);
1516 if (!d_unlinked(dentry
)) {
1518 if (!d_mountpoint(dentry
)) {
1519 dentry
->d_flags
|= DCACHE_MOUNTED
;
1523 spin_unlock(&dentry
->d_lock
);
1525 write_sequnlock(&rename_lock
);
1530 * Search the dentry child list of the specified parent,
1531 * and move any unused dentries to the end of the unused
1532 * list for prune_dcache(). We descend to the next level
1533 * whenever the d_subdirs list is non-empty and continue
1536 * It returns zero iff there are no unused children,
1537 * otherwise it returns the number of children moved to
1538 * the end of the unused list. This may not be the total
1539 * number of unused children, because select_parent can
1540 * drop the lock and return early due to latency
1544 struct select_data
{
1545 struct dentry
*start
;
1548 struct dentry
*victim
;
1550 struct list_head dispose
;
1553 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1555 struct select_data
*data
= _data
;
1556 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1558 if (data
->start
== dentry
)
1561 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1564 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1566 if (!dentry
->d_lockref
.count
) {
1567 d_shrink_add(dentry
, &data
->dispose
);
1572 * We can return to the caller if we have found some (this
1573 * ensures forward progress). We'll be coming back to find
1576 if (!list_empty(&data
->dispose
))
1577 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1582 static enum d_walk_ret
select_collect2(void *_data
, struct dentry
*dentry
)
1584 struct select_data
*data
= _data
;
1585 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1587 if (data
->start
== dentry
)
1590 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1591 if (!dentry
->d_lockref
.count
) {
1593 data
->victim
= dentry
;
1597 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1599 if (!dentry
->d_lockref
.count
)
1600 d_shrink_add(dentry
, &data
->dispose
);
1603 * We can return to the caller if we have found some (this
1604 * ensures forward progress). We'll be coming back to find
1607 if (!list_empty(&data
->dispose
))
1608 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1614 * shrink_dcache_parent - prune dcache
1615 * @parent: parent of entries to prune
1617 * Prune the dcache to remove unused children of the parent dentry.
1619 void shrink_dcache_parent(struct dentry
*parent
)
1622 struct select_data data
= {.start
= parent
};
1624 INIT_LIST_HEAD(&data
.dispose
);
1625 d_walk(parent
, &data
, select_collect
);
1627 if (!list_empty(&data
.dispose
)) {
1628 shrink_dentry_list(&data
.dispose
);
1636 d_walk(parent
, &data
, select_collect2
);
1638 struct dentry
*parent
;
1639 spin_lock(&data
.victim
->d_lock
);
1640 if (!shrink_lock_dentry(data
.victim
)) {
1641 spin_unlock(&data
.victim
->d_lock
);
1645 parent
= data
.victim
->d_parent
;
1646 if (parent
!= data
.victim
)
1647 __dput_to_list(parent
, &data
.dispose
);
1648 __dentry_kill(data
.victim
);
1651 if (!list_empty(&data
.dispose
))
1652 shrink_dentry_list(&data
.dispose
);
1655 EXPORT_SYMBOL(shrink_dcache_parent
);
1657 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1659 /* it has busy descendents; complain about those instead */
1660 if (!list_empty(&dentry
->d_subdirs
))
1661 return D_WALK_CONTINUE
;
1663 /* root with refcount 1 is fine */
1664 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1665 return D_WALK_CONTINUE
;
1667 WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} "
1668 " still in use (%d) [unmount of %s %s]\n",
1671 dentry
->d_inode
->i_ino
: 0UL,
1673 dentry
->d_lockref
.count
,
1674 dentry
->d_sb
->s_type
->name
,
1675 dentry
->d_sb
->s_id
);
1676 return D_WALK_CONTINUE
;
1679 static void do_one_tree(struct dentry
*dentry
)
1681 shrink_dcache_parent(dentry
);
1682 d_walk(dentry
, dentry
, umount_check
);
1688 * destroy the dentries attached to a superblock on unmounting
1690 void shrink_dcache_for_umount(struct super_block
*sb
)
1692 struct dentry
*dentry
;
1694 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1696 dentry
= sb
->s_root
;
1698 do_one_tree(dentry
);
1700 while (!hlist_bl_empty(&sb
->s_roots
)) {
1701 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_roots
), struct dentry
, d_hash
));
1702 do_one_tree(dentry
);
1706 static enum d_walk_ret
find_submount(void *_data
, struct dentry
*dentry
)
1708 struct dentry
**victim
= _data
;
1709 if (d_mountpoint(dentry
)) {
1710 __dget_dlock(dentry
);
1714 return D_WALK_CONTINUE
;
1718 * d_invalidate - detach submounts, prune dcache, and drop
1719 * @dentry: dentry to invalidate (aka detach, prune and drop)
1721 void d_invalidate(struct dentry
*dentry
)
1723 bool had_submounts
= false;
1724 spin_lock(&dentry
->d_lock
);
1725 if (d_unhashed(dentry
)) {
1726 spin_unlock(&dentry
->d_lock
);
1730 spin_unlock(&dentry
->d_lock
);
1732 /* Negative dentries can be dropped without further checks */
1733 if (!dentry
->d_inode
)
1736 shrink_dcache_parent(dentry
);
1738 struct dentry
*victim
= NULL
;
1739 d_walk(dentry
, &victim
, find_submount
);
1742 shrink_dcache_parent(dentry
);
1745 had_submounts
= true;
1746 detach_mounts(victim
);
1750 EXPORT_SYMBOL(d_invalidate
);
1753 * __d_alloc - allocate a dcache entry
1754 * @sb: filesystem it will belong to
1755 * @name: qstr of the name
1757 * Allocates a dentry. It returns %NULL if there is insufficient memory
1758 * available. On a success the dentry is returned. The name passed in is
1759 * copied and the copy passed in may be reused after this call.
1762 static struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1764 struct dentry
*dentry
;
1768 dentry
= kmem_cache_alloc_lru(dentry_cache
, &sb
->s_dentry_lru
,
1774 * We guarantee that the inline name is always NUL-terminated.
1775 * This way the memcpy() done by the name switching in rename
1776 * will still always have a NUL at the end, even if we might
1777 * be overwriting an internal NUL character
1779 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1780 if (unlikely(!name
)) {
1782 dname
= dentry
->d_iname
;
1783 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1784 size_t size
= offsetof(struct external_name
, name
[1]);
1785 struct external_name
*p
= kmalloc(size
+ name
->len
,
1786 GFP_KERNEL_ACCOUNT
|
1789 kmem_cache_free(dentry_cache
, dentry
);
1792 atomic_set(&p
->u
.count
, 1);
1795 dname
= dentry
->d_iname
;
1798 dentry
->d_name
.len
= name
->len
;
1799 dentry
->d_name
.hash
= name
->hash
;
1800 memcpy(dname
, name
->name
, name
->len
);
1801 dname
[name
->len
] = 0;
1803 /* Make sure we always see the terminating NUL character */
1804 smp_store_release(&dentry
->d_name
.name
, dname
); /* ^^^ */
1806 dentry
->d_lockref
.count
= 1;
1807 dentry
->d_flags
= 0;
1808 spin_lock_init(&dentry
->d_lock
);
1809 seqcount_spinlock_init(&dentry
->d_seq
, &dentry
->d_lock
);
1810 dentry
->d_inode
= NULL
;
1811 dentry
->d_parent
= dentry
;
1813 dentry
->d_op
= NULL
;
1814 dentry
->d_fsdata
= NULL
;
1815 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1816 INIT_LIST_HEAD(&dentry
->d_lru
);
1817 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1818 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1819 INIT_LIST_HEAD(&dentry
->d_child
);
1820 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1822 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1823 err
= dentry
->d_op
->d_init(dentry
);
1825 if (dname_external(dentry
))
1826 kfree(external_name(dentry
));
1827 kmem_cache_free(dentry_cache
, dentry
);
1832 this_cpu_inc(nr_dentry
);
1838 * d_alloc - allocate a dcache entry
1839 * @parent: parent of entry to allocate
1840 * @name: qstr of the name
1842 * Allocates a dentry. It returns %NULL if there is insufficient memory
1843 * available. On a success the dentry is returned. The name passed in is
1844 * copied and the copy passed in may be reused after this call.
1846 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1848 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1851 spin_lock(&parent
->d_lock
);
1853 * don't need child lock because it is not subject
1854 * to concurrency here
1856 __dget_dlock(parent
);
1857 dentry
->d_parent
= parent
;
1858 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1859 spin_unlock(&parent
->d_lock
);
1863 EXPORT_SYMBOL(d_alloc
);
1865 struct dentry
*d_alloc_anon(struct super_block
*sb
)
1867 return __d_alloc(sb
, NULL
);
1869 EXPORT_SYMBOL(d_alloc_anon
);
1871 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1873 struct dentry
*dentry
= d_alloc_anon(parent
->d_sb
);
1875 dentry
->d_flags
|= DCACHE_DENTRY_CURSOR
;
1876 dentry
->d_parent
= dget(parent
);
1882 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1883 * @sb: the superblock
1884 * @name: qstr of the name
1886 * For a filesystem that just pins its dentries in memory and never
1887 * performs lookups at all, return an unhashed IS_ROOT dentry.
1888 * This is used for pipes, sockets et.al. - the stuff that should
1889 * never be anyone's children or parents. Unlike all other
1890 * dentries, these will not have RCU delay between dropping the
1891 * last reference and freeing them.
1893 * The only user is alloc_file_pseudo() and that's what should
1894 * be considered a public interface. Don't use directly.
1896 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1898 struct dentry
*dentry
= __d_alloc(sb
, name
);
1900 dentry
->d_flags
|= DCACHE_NORCU
;
1904 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1909 q
.hash_len
= hashlen_string(parent
, name
);
1910 return d_alloc(parent
, &q
);
1912 EXPORT_SYMBOL(d_alloc_name
);
1914 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1916 WARN_ON_ONCE(dentry
->d_op
);
1917 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1919 DCACHE_OP_REVALIDATE
|
1920 DCACHE_OP_WEAK_REVALIDATE
|
1927 dentry
->d_flags
|= DCACHE_OP_HASH
;
1929 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1930 if (op
->d_revalidate
)
1931 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1932 if (op
->d_weak_revalidate
)
1933 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1935 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1937 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1939 dentry
->d_flags
|= DCACHE_OP_REAL
;
1942 EXPORT_SYMBOL(d_set_d_op
);
1946 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1947 * @dentry - The dentry to mark
1949 * Mark a dentry as falling through to the lower layer (as set with
1950 * d_pin_lower()). This flag may be recorded on the medium.
1952 void d_set_fallthru(struct dentry
*dentry
)
1954 spin_lock(&dentry
->d_lock
);
1955 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1956 spin_unlock(&dentry
->d_lock
);
1958 EXPORT_SYMBOL(d_set_fallthru
);
1960 static unsigned d_flags_for_inode(struct inode
*inode
)
1962 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1965 return DCACHE_MISS_TYPE
;
1967 if (S_ISDIR(inode
->i_mode
)) {
1968 add_flags
= DCACHE_DIRECTORY_TYPE
;
1969 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1970 if (unlikely(!inode
->i_op
->lookup
))
1971 add_flags
= DCACHE_AUTODIR_TYPE
;
1973 inode
->i_opflags
|= IOP_LOOKUP
;
1975 goto type_determined
;
1978 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1979 if (unlikely(inode
->i_op
->get_link
)) {
1980 add_flags
= DCACHE_SYMLINK_TYPE
;
1981 goto type_determined
;
1983 inode
->i_opflags
|= IOP_NOFOLLOW
;
1986 if (unlikely(!S_ISREG(inode
->i_mode
)))
1987 add_flags
= DCACHE_SPECIAL_TYPE
;
1990 if (unlikely(IS_AUTOMOUNT(inode
)))
1991 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1995 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1997 unsigned add_flags
= d_flags_for_inode(inode
);
1998 WARN_ON(d_in_lookup(dentry
));
2000 spin_lock(&dentry
->d_lock
);
2002 * Decrement negative dentry count if it was in the LRU list.
2004 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
2005 this_cpu_dec(nr_dentry_negative
);
2006 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2007 raw_write_seqcount_begin(&dentry
->d_seq
);
2008 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2009 raw_write_seqcount_end(&dentry
->d_seq
);
2010 fsnotify_update_flags(dentry
);
2011 spin_unlock(&dentry
->d_lock
);
2015 * d_instantiate - fill in inode information for a dentry
2016 * @entry: dentry to complete
2017 * @inode: inode to attach to this dentry
2019 * Fill in inode information in the entry.
2021 * This turns negative dentries into productive full members
2024 * NOTE! This assumes that the inode count has been incremented
2025 * (or otherwise set) by the caller to indicate that it is now
2026 * in use by the dcache.
2029 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
2031 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
2033 security_d_instantiate(entry
, inode
);
2034 spin_lock(&inode
->i_lock
);
2035 __d_instantiate(entry
, inode
);
2036 spin_unlock(&inode
->i_lock
);
2039 EXPORT_SYMBOL(d_instantiate
);
2042 * This should be equivalent to d_instantiate() + unlock_new_inode(),
2043 * with lockdep-related part of unlock_new_inode() done before
2044 * anything else. Use that instead of open-coding d_instantiate()/
2045 * unlock_new_inode() combinations.
2047 void d_instantiate_new(struct dentry
*entry
, struct inode
*inode
)
2049 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
2051 lockdep_annotate_inode_mutex_key(inode
);
2052 security_d_instantiate(entry
, inode
);
2053 spin_lock(&inode
->i_lock
);
2054 __d_instantiate(entry
, inode
);
2055 WARN_ON(!(inode
->i_state
& I_NEW
));
2056 inode
->i_state
&= ~I_NEW
& ~I_CREATING
;
2058 wake_up_bit(&inode
->i_state
, __I_NEW
);
2059 spin_unlock(&inode
->i_lock
);
2061 EXPORT_SYMBOL(d_instantiate_new
);
2063 struct dentry
*d_make_root(struct inode
*root_inode
)
2065 struct dentry
*res
= NULL
;
2068 res
= d_alloc_anon(root_inode
->i_sb
);
2070 d_instantiate(res
, root_inode
);
2076 EXPORT_SYMBOL(d_make_root
);
2078 static struct dentry
*__d_instantiate_anon(struct dentry
*dentry
,
2079 struct inode
*inode
,
2085 security_d_instantiate(dentry
, inode
);
2086 spin_lock(&inode
->i_lock
);
2087 res
= __d_find_any_alias(inode
);
2089 spin_unlock(&inode
->i_lock
);
2094 /* attach a disconnected dentry */
2095 add_flags
= d_flags_for_inode(inode
);
2098 add_flags
|= DCACHE_DISCONNECTED
;
2100 spin_lock(&dentry
->d_lock
);
2101 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2102 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2103 if (!disconnected
) {
2104 hlist_bl_lock(&dentry
->d_sb
->s_roots
);
2105 hlist_bl_add_head(&dentry
->d_hash
, &dentry
->d_sb
->s_roots
);
2106 hlist_bl_unlock(&dentry
->d_sb
->s_roots
);
2108 spin_unlock(&dentry
->d_lock
);
2109 spin_unlock(&inode
->i_lock
);
2118 struct dentry
*d_instantiate_anon(struct dentry
*dentry
, struct inode
*inode
)
2120 return __d_instantiate_anon(dentry
, inode
, true);
2122 EXPORT_SYMBOL(d_instantiate_anon
);
2124 static struct dentry
*__d_obtain_alias(struct inode
*inode
, bool disconnected
)
2130 return ERR_PTR(-ESTALE
);
2132 return ERR_CAST(inode
);
2134 res
= d_find_any_alias(inode
);
2138 tmp
= d_alloc_anon(inode
->i_sb
);
2140 res
= ERR_PTR(-ENOMEM
);
2144 return __d_instantiate_anon(tmp
, inode
, disconnected
);
2152 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2153 * @inode: inode to allocate the dentry for
2155 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2156 * similar open by handle operations. The returned dentry may be anonymous,
2157 * or may have a full name (if the inode was already in the cache).
2159 * When called on a directory inode, we must ensure that the inode only ever
2160 * has one dentry. If a dentry is found, that is returned instead of
2161 * allocating a new one.
2163 * On successful return, the reference to the inode has been transferred
2164 * to the dentry. In case of an error the reference on the inode is released.
2165 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2166 * be passed in and the error will be propagated to the return value,
2167 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2169 struct dentry
*d_obtain_alias(struct inode
*inode
)
2171 return __d_obtain_alias(inode
, true);
2173 EXPORT_SYMBOL(d_obtain_alias
);
2176 * d_obtain_root - find or allocate a dentry for a given inode
2177 * @inode: inode to allocate the dentry for
2179 * Obtain an IS_ROOT dentry for the root of a filesystem.
2181 * We must ensure that directory inodes only ever have one dentry. If a
2182 * dentry is found, that is returned instead of allocating a new one.
2184 * On successful return, the reference to the inode has been transferred
2185 * to the dentry. In case of an error the reference on the inode is
2186 * released. A %NULL or IS_ERR inode may be passed in and will be the
2187 * error will be propagate to the return value, with a %NULL @inode
2188 * replaced by ERR_PTR(-ESTALE).
2190 struct dentry
*d_obtain_root(struct inode
*inode
)
2192 return __d_obtain_alias(inode
, false);
2194 EXPORT_SYMBOL(d_obtain_root
);
2197 * d_add_ci - lookup or allocate new dentry with case-exact name
2198 * @inode: the inode case-insensitive lookup has found
2199 * @dentry: the negative dentry that was passed to the parent's lookup func
2200 * @name: the case-exact name to be associated with the returned dentry
2202 * This is to avoid filling the dcache with case-insensitive names to the
2203 * same inode, only the actual correct case is stored in the dcache for
2204 * case-insensitive filesystems.
2206 * For a case-insensitive lookup match and if the case-exact dentry
2207 * already exists in the dcache, use it and return it.
2209 * If no entry exists with the exact case name, allocate new dentry with
2210 * the exact case, and return the spliced entry.
2212 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2215 struct dentry
*found
, *res
;
2218 * First check if a dentry matching the name already exists,
2219 * if not go ahead and create it now.
2221 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2226 if (d_in_lookup(dentry
)) {
2227 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2229 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2234 found
= d_alloc(dentry
->d_parent
, name
);
2237 return ERR_PTR(-ENOMEM
);
2240 res
= d_splice_alias(inode
, found
);
2242 d_lookup_done(found
);
2248 EXPORT_SYMBOL(d_add_ci
);
2251 * d_same_name - compare dentry name with case-exact name
2252 * @parent: parent dentry
2253 * @dentry: the negative dentry that was passed to the parent's lookup func
2254 * @name: the case-exact name to be associated with the returned dentry
2256 * Return: true if names are same, or false
2258 bool d_same_name(const struct dentry
*dentry
, const struct dentry
*parent
,
2259 const struct qstr
*name
)
2261 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2262 if (dentry
->d_name
.len
!= name
->len
)
2264 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2266 return parent
->d_op
->d_compare(dentry
,
2267 dentry
->d_name
.len
, dentry
->d_name
.name
,
2270 EXPORT_SYMBOL_GPL(d_same_name
);
2273 * This is __d_lookup_rcu() when the parent dentry has
2274 * DCACHE_OP_COMPARE, which makes things much nastier.
2276 static noinline
struct dentry
*__d_lookup_rcu_op_compare(
2277 const struct dentry
*parent
,
2278 const struct qstr
*name
,
2281 u64 hashlen
= name
->hash_len
;
2282 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2283 struct hlist_bl_node
*node
;
2284 struct dentry
*dentry
;
2286 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2292 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2293 if (dentry
->d_parent
!= parent
)
2295 if (d_unhashed(dentry
))
2297 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2299 tlen
= dentry
->d_name
.len
;
2300 tname
= dentry
->d_name
.name
;
2301 /* we want a consistent (name,len) pair */
2302 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2306 if (parent
->d_op
->d_compare(dentry
, tlen
, tname
, name
) != 0)
2315 * __d_lookup_rcu - search for a dentry (racy, store-free)
2316 * @parent: parent dentry
2317 * @name: qstr of name we wish to find
2318 * @seqp: returns d_seq value at the point where the dentry was found
2319 * Returns: dentry, or NULL
2321 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2322 * resolution (store-free path walking) design described in
2323 * Documentation/filesystems/path-lookup.txt.
2325 * This is not to be used outside core vfs.
2327 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2328 * held, and rcu_read_lock held. The returned dentry must not be stored into
2329 * without taking d_lock and checking d_seq sequence count against @seq
2332 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2335 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2336 * the returned dentry, so long as its parent's seqlock is checked after the
2337 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2338 * is formed, giving integrity down the path walk.
2340 * NOTE! The caller *has* to check the resulting dentry against the sequence
2341 * number we've returned before using any of the resulting dentry state!
2343 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2344 const struct qstr
*name
,
2347 u64 hashlen
= name
->hash_len
;
2348 const unsigned char *str
= name
->name
;
2349 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2350 struct hlist_bl_node
*node
;
2351 struct dentry
*dentry
;
2354 * Note: There is significant duplication with __d_lookup_rcu which is
2355 * required to prevent single threaded performance regressions
2356 * especially on architectures where smp_rmb (in seqcounts) are costly.
2357 * Keep the two functions in sync.
2360 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
))
2361 return __d_lookup_rcu_op_compare(parent
, name
, seqp
);
2364 * The hash list is protected using RCU.
2366 * Carefully use d_seq when comparing a candidate dentry, to avoid
2367 * races with d_move().
2369 * It is possible that concurrent renames can mess up our list
2370 * walk here and result in missing our dentry, resulting in the
2371 * false-negative result. d_lookup() protects against concurrent
2372 * renames using rename_lock seqlock.
2374 * See Documentation/filesystems/path-lookup.txt for more details.
2376 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2380 * The dentry sequence count protects us from concurrent
2381 * renames, and thus protects parent and name fields.
2383 * The caller must perform a seqcount check in order
2384 * to do anything useful with the returned dentry.
2386 * NOTE! We do a "raw" seqcount_begin here. That means that
2387 * we don't wait for the sequence count to stabilize if it
2388 * is in the middle of a sequence change. If we do the slow
2389 * dentry compare, we will do seqretries until it is stable,
2390 * and if we end up with a successful lookup, we actually
2391 * want to exit RCU lookup anyway.
2393 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2394 * we are still guaranteed NUL-termination of ->d_name.name.
2396 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2397 if (dentry
->d_parent
!= parent
)
2399 if (d_unhashed(dentry
))
2401 if (dentry
->d_name
.hash_len
!= hashlen
)
2403 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2412 * d_lookup - search for a dentry
2413 * @parent: parent dentry
2414 * @name: qstr of name we wish to find
2415 * Returns: dentry, or NULL
2417 * d_lookup searches the children of the parent dentry for the name in
2418 * question. If the dentry is found its reference count is incremented and the
2419 * dentry is returned. The caller must use dput to free the entry when it has
2420 * finished using it. %NULL is returned if the dentry does not exist.
2422 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2424 struct dentry
*dentry
;
2428 seq
= read_seqbegin(&rename_lock
);
2429 dentry
= __d_lookup(parent
, name
);
2432 } while (read_seqretry(&rename_lock
, seq
));
2435 EXPORT_SYMBOL(d_lookup
);
2438 * __d_lookup - search for a dentry (racy)
2439 * @parent: parent dentry
2440 * @name: qstr of name we wish to find
2441 * Returns: dentry, or NULL
2443 * __d_lookup is like d_lookup, however it may (rarely) return a
2444 * false-negative result due to unrelated rename activity.
2446 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2447 * however it must be used carefully, eg. with a following d_lookup in
2448 * the case of failure.
2450 * __d_lookup callers must be commented.
2452 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2454 unsigned int hash
= name
->hash
;
2455 struct hlist_bl_head
*b
= d_hash(hash
);
2456 struct hlist_bl_node
*node
;
2457 struct dentry
*found
= NULL
;
2458 struct dentry
*dentry
;
2461 * Note: There is significant duplication with __d_lookup_rcu which is
2462 * required to prevent single threaded performance regressions
2463 * especially on architectures where smp_rmb (in seqcounts) are costly.
2464 * Keep the two functions in sync.
2468 * The hash list is protected using RCU.
2470 * Take d_lock when comparing a candidate dentry, to avoid races
2473 * It is possible that concurrent renames can mess up our list
2474 * walk here and result in missing our dentry, resulting in the
2475 * false-negative result. d_lookup() protects against concurrent
2476 * renames using rename_lock seqlock.
2478 * See Documentation/filesystems/path-lookup.txt for more details.
2482 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2484 if (dentry
->d_name
.hash
!= hash
)
2487 spin_lock(&dentry
->d_lock
);
2488 if (dentry
->d_parent
!= parent
)
2490 if (d_unhashed(dentry
))
2493 if (!d_same_name(dentry
, parent
, name
))
2496 dentry
->d_lockref
.count
++;
2498 spin_unlock(&dentry
->d_lock
);
2501 spin_unlock(&dentry
->d_lock
);
2509 * d_hash_and_lookup - hash the qstr then search for a dentry
2510 * @dir: Directory to search in
2511 * @name: qstr of name we wish to find
2513 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2515 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2518 * Check for a fs-specific hash function. Note that we must
2519 * calculate the standard hash first, as the d_op->d_hash()
2520 * routine may choose to leave the hash value unchanged.
2522 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2523 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2524 int err
= dir
->d_op
->d_hash(dir
, name
);
2525 if (unlikely(err
< 0))
2526 return ERR_PTR(err
);
2528 return d_lookup(dir
, name
);
2530 EXPORT_SYMBOL(d_hash_and_lookup
);
2533 * When a file is deleted, we have two options:
2534 * - turn this dentry into a negative dentry
2535 * - unhash this dentry and free it.
2537 * Usually, we want to just turn this into
2538 * a negative dentry, but if anybody else is
2539 * currently using the dentry or the inode
2540 * we can't do that and we fall back on removing
2541 * it from the hash queues and waiting for
2542 * it to be deleted later when it has no users
2546 * d_delete - delete a dentry
2547 * @dentry: The dentry to delete
2549 * Turn the dentry into a negative dentry if possible, otherwise
2550 * remove it from the hash queues so it can be deleted later
2553 void d_delete(struct dentry
* dentry
)
2555 struct inode
*inode
= dentry
->d_inode
;
2557 spin_lock(&inode
->i_lock
);
2558 spin_lock(&dentry
->d_lock
);
2560 * Are we the only user?
2562 if (dentry
->d_lockref
.count
== 1) {
2563 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2564 dentry_unlink_inode(dentry
);
2567 spin_unlock(&dentry
->d_lock
);
2568 spin_unlock(&inode
->i_lock
);
2571 EXPORT_SYMBOL(d_delete
);
2573 static void __d_rehash(struct dentry
*entry
)
2575 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2578 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2583 * d_rehash - add an entry back to the hash
2584 * @entry: dentry to add to the hash
2586 * Adds a dentry to the hash according to its name.
2589 void d_rehash(struct dentry
* entry
)
2591 spin_lock(&entry
->d_lock
);
2593 spin_unlock(&entry
->d_lock
);
2595 EXPORT_SYMBOL(d_rehash
);
2597 static inline unsigned start_dir_add(struct inode
*dir
)
2599 preempt_disable_nested();
2601 unsigned n
= dir
->i_dir_seq
;
2602 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2608 static inline void end_dir_add(struct inode
*dir
, unsigned int n
,
2609 wait_queue_head_t
*d_wait
)
2611 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2612 preempt_enable_nested();
2613 wake_up_all(d_wait
);
2616 static void d_wait_lookup(struct dentry
*dentry
)
2618 if (d_in_lookup(dentry
)) {
2619 DECLARE_WAITQUEUE(wait
, current
);
2620 add_wait_queue(dentry
->d_wait
, &wait
);
2622 set_current_state(TASK_UNINTERRUPTIBLE
);
2623 spin_unlock(&dentry
->d_lock
);
2625 spin_lock(&dentry
->d_lock
);
2626 } while (d_in_lookup(dentry
));
2630 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2631 const struct qstr
*name
,
2632 wait_queue_head_t
*wq
)
2634 unsigned int hash
= name
->hash
;
2635 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2636 struct hlist_bl_node
*node
;
2637 struct dentry
*new = d_alloc(parent
, name
);
2638 struct dentry
*dentry
;
2639 unsigned seq
, r_seq
, d_seq
;
2642 return ERR_PTR(-ENOMEM
);
2646 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
);
2647 r_seq
= read_seqbegin(&rename_lock
);
2648 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2649 if (unlikely(dentry
)) {
2650 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2654 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2663 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2668 if (unlikely(seq
& 1)) {
2674 if (unlikely(READ_ONCE(parent
->d_inode
->i_dir_seq
) != seq
)) {
2680 * No changes for the parent since the beginning of d_lookup().
2681 * Since all removals from the chain happen with hlist_bl_lock(),
2682 * any potential in-lookup matches are going to stay here until
2683 * we unlock the chain. All fields are stable in everything
2686 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2687 if (dentry
->d_name
.hash
!= hash
)
2689 if (dentry
->d_parent
!= parent
)
2691 if (!d_same_name(dentry
, parent
, name
))
2694 /* now we can try to grab a reference */
2695 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2702 * somebody is likely to be still doing lookup for it;
2703 * wait for them to finish
2705 spin_lock(&dentry
->d_lock
);
2706 d_wait_lookup(dentry
);
2708 * it's not in-lookup anymore; in principle we should repeat
2709 * everything from dcache lookup, but it's likely to be what
2710 * d_lookup() would've found anyway. If it is, just return it;
2711 * otherwise we really have to repeat the whole thing.
2713 if (unlikely(dentry
->d_name
.hash
!= hash
))
2715 if (unlikely(dentry
->d_parent
!= parent
))
2717 if (unlikely(d_unhashed(dentry
)))
2719 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2721 /* OK, it *is* a hashed match; return it */
2722 spin_unlock(&dentry
->d_lock
);
2727 /* we can't take ->d_lock here; it's OK, though. */
2728 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2730 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2734 spin_unlock(&dentry
->d_lock
);
2738 EXPORT_SYMBOL(d_alloc_parallel
);
2741 * - Unhash the dentry
2742 * - Retrieve and clear the waitqueue head in dentry
2743 * - Return the waitqueue head
2745 static wait_queue_head_t
*__d_lookup_unhash(struct dentry
*dentry
)
2747 wait_queue_head_t
*d_wait
;
2748 struct hlist_bl_head
*b
;
2750 lockdep_assert_held(&dentry
->d_lock
);
2752 b
= in_lookup_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
2754 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2755 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2756 d_wait
= dentry
->d_wait
;
2757 dentry
->d_wait
= NULL
;
2759 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2760 INIT_LIST_HEAD(&dentry
->d_lru
);
2764 void __d_lookup_unhash_wake(struct dentry
*dentry
)
2766 spin_lock(&dentry
->d_lock
);
2767 wake_up_all(__d_lookup_unhash(dentry
));
2768 spin_unlock(&dentry
->d_lock
);
2770 EXPORT_SYMBOL(__d_lookup_unhash_wake
);
2772 /* inode->i_lock held if inode is non-NULL */
2774 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2776 wait_queue_head_t
*d_wait
;
2777 struct inode
*dir
= NULL
;
2779 spin_lock(&dentry
->d_lock
);
2780 if (unlikely(d_in_lookup(dentry
))) {
2781 dir
= dentry
->d_parent
->d_inode
;
2782 n
= start_dir_add(dir
);
2783 d_wait
= __d_lookup_unhash(dentry
);
2786 unsigned add_flags
= d_flags_for_inode(inode
);
2787 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2788 raw_write_seqcount_begin(&dentry
->d_seq
);
2789 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2790 raw_write_seqcount_end(&dentry
->d_seq
);
2791 fsnotify_update_flags(dentry
);
2795 end_dir_add(dir
, n
, d_wait
);
2796 spin_unlock(&dentry
->d_lock
);
2798 spin_unlock(&inode
->i_lock
);
2802 * d_add - add dentry to hash queues
2803 * @entry: dentry to add
2804 * @inode: The inode to attach to this dentry
2806 * This adds the entry to the hash queues and initializes @inode.
2807 * The entry was actually filled in earlier during d_alloc().
2810 void d_add(struct dentry
*entry
, struct inode
*inode
)
2813 security_d_instantiate(entry
, inode
);
2814 spin_lock(&inode
->i_lock
);
2816 __d_add(entry
, inode
);
2818 EXPORT_SYMBOL(d_add
);
2821 * d_exact_alias - find and hash an exact unhashed alias
2822 * @entry: dentry to add
2823 * @inode: The inode to go with this dentry
2825 * If an unhashed dentry with the same name/parent and desired
2826 * inode already exists, hash and return it. Otherwise, return
2829 * Parent directory should be locked.
2831 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2833 struct dentry
*alias
;
2834 unsigned int hash
= entry
->d_name
.hash
;
2836 spin_lock(&inode
->i_lock
);
2837 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2839 * Don't need alias->d_lock here, because aliases with
2840 * d_parent == entry->d_parent are not subject to name or
2841 * parent changes, because the parent inode i_mutex is held.
2843 if (alias
->d_name
.hash
!= hash
)
2845 if (alias
->d_parent
!= entry
->d_parent
)
2847 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2849 spin_lock(&alias
->d_lock
);
2850 if (!d_unhashed(alias
)) {
2851 spin_unlock(&alias
->d_lock
);
2854 __dget_dlock(alias
);
2856 spin_unlock(&alias
->d_lock
);
2858 spin_unlock(&inode
->i_lock
);
2861 spin_unlock(&inode
->i_lock
);
2864 EXPORT_SYMBOL(d_exact_alias
);
2866 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2868 if (unlikely(dname_external(target
))) {
2869 if (unlikely(dname_external(dentry
))) {
2871 * Both external: swap the pointers
2873 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2876 * dentry:internal, target:external. Steal target's
2877 * storage and make target internal.
2879 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2880 dentry
->d_name
.len
+ 1);
2881 dentry
->d_name
.name
= target
->d_name
.name
;
2882 target
->d_name
.name
= target
->d_iname
;
2885 if (unlikely(dname_external(dentry
))) {
2887 * dentry:external, target:internal. Give dentry's
2888 * storage to target and make dentry internal
2890 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2891 target
->d_name
.len
+ 1);
2892 target
->d_name
.name
= dentry
->d_name
.name
;
2893 dentry
->d_name
.name
= dentry
->d_iname
;
2896 * Both are internal.
2899 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2900 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2901 swap(((long *) &dentry
->d_iname
)[i
],
2902 ((long *) &target
->d_iname
)[i
]);
2906 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2909 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2911 struct external_name
*old_name
= NULL
;
2912 if (unlikely(dname_external(dentry
)))
2913 old_name
= external_name(dentry
);
2914 if (unlikely(dname_external(target
))) {
2915 atomic_inc(&external_name(target
)->u
.count
);
2916 dentry
->d_name
= target
->d_name
;
2918 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2919 target
->d_name
.len
+ 1);
2920 dentry
->d_name
.name
= dentry
->d_iname
;
2921 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2923 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2924 kfree_rcu(old_name
, u
.head
);
2928 * __d_move - move a dentry
2929 * @dentry: entry to move
2930 * @target: new dentry
2931 * @exchange: exchange the two dentries
2933 * Update the dcache to reflect the move of a file name. Negative
2934 * dcache entries should not be moved in this way. Caller must hold
2935 * rename_lock, the i_mutex of the source and target directories,
2936 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2938 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2941 struct dentry
*old_parent
, *p
;
2942 wait_queue_head_t
*d_wait
;
2943 struct inode
*dir
= NULL
;
2946 WARN_ON(!dentry
->d_inode
);
2947 if (WARN_ON(dentry
== target
))
2950 BUG_ON(d_ancestor(target
, dentry
));
2951 old_parent
= dentry
->d_parent
;
2952 p
= d_ancestor(old_parent
, target
);
2953 if (IS_ROOT(dentry
)) {
2955 spin_lock(&target
->d_parent
->d_lock
);
2957 /* target is not a descendent of dentry->d_parent */
2958 spin_lock(&target
->d_parent
->d_lock
);
2959 spin_lock_nested(&old_parent
->d_lock
, DENTRY_D_LOCK_NESTED
);
2961 BUG_ON(p
== dentry
);
2962 spin_lock(&old_parent
->d_lock
);
2964 spin_lock_nested(&target
->d_parent
->d_lock
,
2965 DENTRY_D_LOCK_NESTED
);
2967 spin_lock_nested(&dentry
->d_lock
, 2);
2968 spin_lock_nested(&target
->d_lock
, 3);
2970 if (unlikely(d_in_lookup(target
))) {
2971 dir
= target
->d_parent
->d_inode
;
2972 n
= start_dir_add(dir
);
2973 d_wait
= __d_lookup_unhash(target
);
2976 write_seqcount_begin(&dentry
->d_seq
);
2977 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2980 if (!d_unhashed(dentry
))
2982 if (!d_unhashed(target
))
2985 /* ... and switch them in the tree */
2986 dentry
->d_parent
= target
->d_parent
;
2988 copy_name(dentry
, target
);
2989 target
->d_hash
.pprev
= NULL
;
2990 dentry
->d_parent
->d_lockref
.count
++;
2991 if (dentry
!= old_parent
) /* wasn't IS_ROOT */
2992 WARN_ON(!--old_parent
->d_lockref
.count
);
2994 target
->d_parent
= old_parent
;
2995 swap_names(dentry
, target
);
2996 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2998 fsnotify_update_flags(target
);
3000 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
3002 fsnotify_update_flags(dentry
);
3003 fscrypt_handle_d_move(dentry
);
3005 write_seqcount_end(&target
->d_seq
);
3006 write_seqcount_end(&dentry
->d_seq
);
3009 end_dir_add(dir
, n
, d_wait
);
3011 if (dentry
->d_parent
!= old_parent
)
3012 spin_unlock(&dentry
->d_parent
->d_lock
);
3013 if (dentry
!= old_parent
)
3014 spin_unlock(&old_parent
->d_lock
);
3015 spin_unlock(&target
->d_lock
);
3016 spin_unlock(&dentry
->d_lock
);
3020 * d_move - move a dentry
3021 * @dentry: entry to move
3022 * @target: new dentry
3024 * Update the dcache to reflect the move of a file name. Negative
3025 * dcache entries should not be moved in this way. See the locking
3026 * requirements for __d_move.
3028 void d_move(struct dentry
*dentry
, struct dentry
*target
)
3030 write_seqlock(&rename_lock
);
3031 __d_move(dentry
, target
, false);
3032 write_sequnlock(&rename_lock
);
3034 EXPORT_SYMBOL(d_move
);
3037 * d_exchange - exchange two dentries
3038 * @dentry1: first dentry
3039 * @dentry2: second dentry
3041 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
3043 write_seqlock(&rename_lock
);
3045 WARN_ON(!dentry1
->d_inode
);
3046 WARN_ON(!dentry2
->d_inode
);
3047 WARN_ON(IS_ROOT(dentry1
));
3048 WARN_ON(IS_ROOT(dentry2
));
3050 __d_move(dentry1
, dentry2
, true);
3052 write_sequnlock(&rename_lock
);
3056 * d_ancestor - search for an ancestor
3057 * @p1: ancestor dentry
3060 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
3061 * an ancestor of p2, else NULL.
3063 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
3067 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
3068 if (p
->d_parent
== p1
)
3075 * This helper attempts to cope with remotely renamed directories
3077 * It assumes that the caller is already holding
3078 * dentry->d_parent->d_inode->i_mutex, and rename_lock
3080 * Note: If ever the locking in lock_rename() changes, then please
3081 * remember to update this too...
3083 static int __d_unalias(struct inode
*inode
,
3084 struct dentry
*dentry
, struct dentry
*alias
)
3086 struct mutex
*m1
= NULL
;
3087 struct rw_semaphore
*m2
= NULL
;
3090 /* If alias and dentry share a parent, then no extra locks required */
3091 if (alias
->d_parent
== dentry
->d_parent
)
3094 /* See lock_rename() */
3095 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
3097 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
3098 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
3100 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
3102 __d_move(alias
, dentry
, false);
3113 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3114 * @inode: the inode which may have a disconnected dentry
3115 * @dentry: a negative dentry which we want to point to the inode.
3117 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3118 * place of the given dentry and return it, else simply d_add the inode
3119 * to the dentry and return NULL.
3121 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3122 * we should error out: directories can't have multiple aliases.
3124 * This is needed in the lookup routine of any filesystem that is exportable
3125 * (via knfsd) so that we can build dcache paths to directories effectively.
3127 * If a dentry was found and moved, then it is returned. Otherwise NULL
3128 * is returned. This matches the expected return value of ->lookup.
3130 * Cluster filesystems may call this function with a negative, hashed dentry.
3131 * In that case, we know that the inode will be a regular file, and also this
3132 * will only occur during atomic_open. So we need to check for the dentry
3133 * being already hashed only in the final case.
3135 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
3138 return ERR_CAST(inode
);
3140 BUG_ON(!d_unhashed(dentry
));
3145 security_d_instantiate(dentry
, inode
);
3146 spin_lock(&inode
->i_lock
);
3147 if (S_ISDIR(inode
->i_mode
)) {
3148 struct dentry
*new = __d_find_any_alias(inode
);
3149 if (unlikely(new)) {
3150 /* The reference to new ensures it remains an alias */
3151 spin_unlock(&inode
->i_lock
);
3152 write_seqlock(&rename_lock
);
3153 if (unlikely(d_ancestor(new, dentry
))) {
3154 write_sequnlock(&rename_lock
);
3156 new = ERR_PTR(-ELOOP
);
3157 pr_warn_ratelimited(
3158 "VFS: Lookup of '%s' in %s %s"
3159 " would have caused loop\n",
3160 dentry
->d_name
.name
,
3161 inode
->i_sb
->s_type
->name
,
3163 } else if (!IS_ROOT(new)) {
3164 struct dentry
*old_parent
= dget(new->d_parent
);
3165 int err
= __d_unalias(inode
, dentry
, new);
3166 write_sequnlock(&rename_lock
);
3173 __d_move(new, dentry
, false);
3174 write_sequnlock(&rename_lock
);
3181 __d_add(dentry
, inode
);
3184 EXPORT_SYMBOL(d_splice_alias
);
3187 * Test whether new_dentry is a subdirectory of old_dentry.
3189 * Trivially implemented using the dcache structure
3193 * is_subdir - is new dentry a subdirectory of old_dentry
3194 * @new_dentry: new dentry
3195 * @old_dentry: old dentry
3197 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3198 * Returns false otherwise.
3199 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3202 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3207 if (new_dentry
== old_dentry
)
3211 /* for restarting inner loop in case of seq retry */
3212 seq
= read_seqbegin(&rename_lock
);
3214 * Need rcu_readlock to protect against the d_parent trashing
3218 if (d_ancestor(old_dentry
, new_dentry
))
3223 } while (read_seqretry(&rename_lock
, seq
));
3227 EXPORT_SYMBOL(is_subdir
);
3229 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3231 struct dentry
*root
= data
;
3232 if (dentry
!= root
) {
3233 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3236 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3237 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3238 dentry
->d_lockref
.count
--;
3241 return D_WALK_CONTINUE
;
3244 void d_genocide(struct dentry
*parent
)
3246 d_walk(parent
, parent
, d_genocide_kill
);
3249 void d_mark_tmpfile(struct file
*file
, struct inode
*inode
)
3251 struct dentry
*dentry
= file
->f_path
.dentry
;
3253 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3254 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3255 !d_unlinked(dentry
));
3256 spin_lock(&dentry
->d_parent
->d_lock
);
3257 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3258 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3259 (unsigned long long)inode
->i_ino
);
3260 spin_unlock(&dentry
->d_lock
);
3261 spin_unlock(&dentry
->d_parent
->d_lock
);
3263 EXPORT_SYMBOL(d_mark_tmpfile
);
3265 void d_tmpfile(struct file
*file
, struct inode
*inode
)
3267 struct dentry
*dentry
= file
->f_path
.dentry
;
3269 inode_dec_link_count(inode
);
3270 d_mark_tmpfile(file
, inode
);
3271 d_instantiate(dentry
, inode
);
3273 EXPORT_SYMBOL(d_tmpfile
);
3275 static __initdata
unsigned long dhash_entries
;
3276 static int __init
set_dhash_entries(char *str
)
3280 dhash_entries
= simple_strtoul(str
, &str
, 0);
3283 __setup("dhash_entries=", set_dhash_entries
);
3285 static void __init
dcache_init_early(void)
3287 /* If hashes are distributed across NUMA nodes, defer
3288 * hash allocation until vmalloc space is available.
3294 alloc_large_system_hash("Dentry cache",
3295 sizeof(struct hlist_bl_head
),
3298 HASH_EARLY
| HASH_ZERO
,
3303 d_hash_shift
= 32 - d_hash_shift
;
3306 static void __init
dcache_init(void)
3309 * A constructor could be added for stable state like the lists,
3310 * but it is probably not worth it because of the cache nature
3313 dentry_cache
= KMEM_CACHE_USERCOPY(dentry
,
3314 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
,
3317 /* Hash may have been set up in dcache_init_early */
3322 alloc_large_system_hash("Dentry cache",
3323 sizeof(struct hlist_bl_head
),
3331 d_hash_shift
= 32 - d_hash_shift
;
3334 /* SLAB cache for __getname() consumers */
3335 struct kmem_cache
*names_cachep __ro_after_init
;
3336 EXPORT_SYMBOL(names_cachep
);
3338 void __init
vfs_caches_init_early(void)
3342 for (i
= 0; i
< ARRAY_SIZE(in_lookup_hashtable
); i
++)
3343 INIT_HLIST_BL_HEAD(&in_lookup_hashtable
[i
]);
3345 dcache_init_early();
3349 void __init
vfs_caches_init(void)
3351 names_cachep
= kmem_cache_create_usercopy("names_cache", PATH_MAX
, 0,
3352 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, 0, PATH_MAX
, NULL
);
3357 files_maxfiles_init();