4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <linux/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
41 #include <linux/kasan.h>
48 * dcache->d_inode->i_lock protects:
49 * - i_dentry, d_u.d_alias, d_inode of aliases
50 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
54 * dentry->d_sb->s_dentry_lru_lock protects:
55 * - the dcache lru lists and counters
62 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
64 * - d_u.d_alias, d_inode
67 * dentry->d_inode->i_lock
69 * dentry->d_sb->s_dentry_lru_lock
70 * dcache_hash_bucket lock
73 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
76 * dentry->d_parent->d_lock
79 * If no ancestor relationship:
80 * if (dentry1 < dentry2)
84 int sysctl_vfs_cache_pressure __read_mostly
= 100;
85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
87 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
89 EXPORT_SYMBOL(rename_lock
);
91 static struct kmem_cache
*dentry_cache __read_mostly
;
93 const struct qstr empty_name
= QSTR_INIT("", 0);
94 EXPORT_SYMBOL(empty_name
);
95 const struct qstr slash_name
= QSTR_INIT("/", 1);
96 EXPORT_SYMBOL(slash_name
);
99 * This is the single most critical data structure when it comes
100 * to the dcache: the hashtable for lookups. Somebody should try
101 * to make this good - I've just made it work.
103 * This hash-function tries to avoid losing too many bits of hash
104 * information, yet avoid using a prime hash-size or similar.
107 static unsigned int d_hash_mask __read_mostly
;
108 static unsigned int d_hash_shift __read_mostly
;
110 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
112 static inline struct hlist_bl_head
*d_hash(unsigned int hash
)
114 return dentry_hashtable
+ (hash
>> (32 - d_hash_shift
));
117 #define IN_LOOKUP_SHIFT 10
118 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
120 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
123 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
124 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
128 /* Statistics gathering. */
129 struct dentry_stat_t dentry_stat
= {
133 static DEFINE_PER_CPU(long, nr_dentry
);
134 static DEFINE_PER_CPU(long, nr_dentry_unused
);
136 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
139 * Here we resort to our own counters instead of using generic per-cpu counters
140 * for consistency with what the vfs inode code does. We are expected to harvest
141 * better code and performance by having our own specialized counters.
143 * Please note that the loop is done over all possible CPUs, not over all online
144 * CPUs. The reason for this is that we don't want to play games with CPUs going
145 * on and off. If one of them goes off, we will just keep their counters.
147 * glommer: See cffbc8a for details, and if you ever intend to change this,
148 * please update all vfs counters to match.
150 static long get_nr_dentry(void)
154 for_each_possible_cpu(i
)
155 sum
+= per_cpu(nr_dentry
, i
);
156 return sum
< 0 ? 0 : sum
;
159 static long get_nr_dentry_unused(void)
163 for_each_possible_cpu(i
)
164 sum
+= per_cpu(nr_dentry_unused
, i
);
165 return sum
< 0 ? 0 : sum
;
168 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*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 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
178 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
179 * The strings are both count bytes long, and count is non-zero.
181 #ifdef CONFIG_DCACHE_WORD_ACCESS
183 #include <asm/word-at-a-time.h>
185 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
186 * aligned allocation for this particular component. We don't
187 * strictly need the load_unaligned_zeropad() safety, but it
188 * doesn't hurt either.
190 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
191 * need the careful unaligned handling.
193 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
195 unsigned long a
,b
,mask
;
198 a
= *(unsigned long *)cs
;
199 b
= load_unaligned_zeropad(ct
);
200 if (tcount
< sizeof(unsigned long))
202 if (unlikely(a
!= b
))
204 cs
+= sizeof(unsigned long);
205 ct
+= sizeof(unsigned long);
206 tcount
-= sizeof(unsigned long);
210 mask
= bytemask_from_count(tcount
);
211 return unlikely(!!((a
^ b
) & mask
));
216 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
230 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
233 * Be careful about RCU walk racing with rename:
234 * use 'READ_ONCE' to fetch the name pointer.
236 * NOTE! Even if a rename will mean that the length
237 * was not loaded atomically, we don't care. The
238 * RCU walk will check the sequence count eventually,
239 * and catch it. And we won't overrun the buffer,
240 * because we're reading the name pointer atomically,
241 * and a dentry name is guaranteed to be properly
242 * terminated with a NUL byte.
244 * End result: even if 'len' is wrong, we'll exit
245 * early because the data cannot match (there can
246 * be no NUL in the ct/tcount data)
248 const unsigned char *cs
= READ_ONCE(dentry
->d_name
.name
);
250 return dentry_string_cmp(cs
, ct
, tcount
);
253 struct external_name
{
256 struct rcu_head head
;
258 unsigned char name
[];
261 static inline struct external_name
*external_name(struct dentry
*dentry
)
263 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
266 static void __d_free(struct rcu_head
*head
)
268 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
270 kmem_cache_free(dentry_cache
, dentry
);
273 static void __d_free_external_name(struct rcu_head
*head
)
275 struct external_name
*name
= container_of(head
, struct external_name
,
278 mod_node_page_state(page_pgdat(virt_to_page(name
)),
279 NR_INDIRECTLY_RECLAIMABLE_BYTES
,
285 static void __d_free_external(struct rcu_head
*head
)
287 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
289 __d_free_external_name(&external_name(dentry
)->u
.head
);
291 kmem_cache_free(dentry_cache
, dentry
);
294 static inline int dname_external(const struct dentry
*dentry
)
296 return dentry
->d_name
.name
!= dentry
->d_iname
;
299 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
301 spin_lock(&dentry
->d_lock
);
302 if (unlikely(dname_external(dentry
))) {
303 struct external_name
*p
= external_name(dentry
);
304 atomic_inc(&p
->u
.count
);
305 spin_unlock(&dentry
->d_lock
);
306 name
->name
= p
->name
;
308 memcpy(name
->inline_name
, dentry
->d_iname
,
309 dentry
->d_name
.len
+ 1);
310 spin_unlock(&dentry
->d_lock
);
311 name
->name
= name
->inline_name
;
314 EXPORT_SYMBOL(take_dentry_name_snapshot
);
316 void release_dentry_name_snapshot(struct name_snapshot
*name
)
318 if (unlikely(name
->name
!= name
->inline_name
)) {
319 struct external_name
*p
;
320 p
= container_of(name
->name
, struct external_name
, name
[0]);
321 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
322 call_rcu(&p
->u
.head
, __d_free_external_name
);
325 EXPORT_SYMBOL(release_dentry_name_snapshot
);
327 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
333 dentry
->d_inode
= inode
;
334 flags
= READ_ONCE(dentry
->d_flags
);
335 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
337 WRITE_ONCE(dentry
->d_flags
, flags
);
340 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
342 unsigned flags
= READ_ONCE(dentry
->d_flags
);
344 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
345 WRITE_ONCE(dentry
->d_flags
, flags
);
346 dentry
->d_inode
= NULL
;
349 static void dentry_free(struct dentry
*dentry
)
351 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
352 if (unlikely(dname_external(dentry
))) {
353 struct external_name
*p
= external_name(dentry
);
354 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
355 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
359 /* if dentry was never visible to RCU, immediate free is OK */
360 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
361 __d_free(&dentry
->d_u
.d_rcu
);
363 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
367 * Release the dentry's inode, using the filesystem
368 * d_iput() operation if defined.
370 static void dentry_unlink_inode(struct dentry
* dentry
)
371 __releases(dentry
->d_lock
)
372 __releases(dentry
->d_inode
->i_lock
)
374 struct inode
*inode
= dentry
->d_inode
;
376 raw_write_seqcount_begin(&dentry
->d_seq
);
377 __d_clear_type_and_inode(dentry
);
378 hlist_del_init(&dentry
->d_u
.d_alias
);
379 raw_write_seqcount_end(&dentry
->d_seq
);
380 spin_unlock(&dentry
->d_lock
);
381 spin_unlock(&inode
->i_lock
);
383 fsnotify_inoderemove(inode
);
384 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
385 dentry
->d_op
->d_iput(dentry
, inode
);
391 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
392 * is in use - which includes both the "real" per-superblock
393 * LRU list _and_ the DCACHE_SHRINK_LIST use.
395 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
396 * on the shrink list (ie not on the superblock LRU list).
398 * The per-cpu "nr_dentry_unused" counters are updated with
399 * the DCACHE_LRU_LIST bit.
401 * These helper functions make sure we always follow the
402 * rules. d_lock must be held by the caller.
404 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
405 static void d_lru_add(struct dentry
*dentry
)
407 D_FLAG_VERIFY(dentry
, 0);
408 dentry
->d_flags
|= DCACHE_LRU_LIST
;
409 this_cpu_inc(nr_dentry_unused
);
410 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
413 static void d_lru_del(struct dentry
*dentry
)
415 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
416 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
417 this_cpu_dec(nr_dentry_unused
);
418 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
421 static void d_shrink_del(struct dentry
*dentry
)
423 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
424 list_del_init(&dentry
->d_lru
);
425 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
426 this_cpu_dec(nr_dentry_unused
);
429 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
431 D_FLAG_VERIFY(dentry
, 0);
432 list_add(&dentry
->d_lru
, list
);
433 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
434 this_cpu_inc(nr_dentry_unused
);
438 * These can only be called under the global LRU lock, ie during the
439 * callback for freeing the LRU list. "isolate" removes it from the
440 * LRU lists entirely, while shrink_move moves it to the indicated
443 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
445 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
446 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
447 this_cpu_dec(nr_dentry_unused
);
448 list_lru_isolate(lru
, &dentry
->d_lru
);
451 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
452 struct list_head
*list
)
454 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
455 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
456 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
460 * dentry_lru_(add|del)_list) must be called with d_lock held.
462 static void dentry_lru_add(struct dentry
*dentry
)
464 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
466 else if (unlikely(!(dentry
->d_flags
& DCACHE_REFERENCED
)))
467 dentry
->d_flags
|= DCACHE_REFERENCED
;
471 * d_drop - drop a dentry
472 * @dentry: dentry to drop
474 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
475 * be found through a VFS lookup any more. Note that this is different from
476 * deleting the dentry - d_delete will try to mark the dentry negative if
477 * possible, giving a successful _negative_ lookup, while d_drop will
478 * just make the cache lookup fail.
480 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
481 * reason (NFS timeouts or autofs deletes).
483 * __d_drop requires dentry->d_lock
484 * ___d_drop doesn't mark dentry as "unhashed"
485 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
487 static void ___d_drop(struct dentry
*dentry
)
489 if (!d_unhashed(dentry
)) {
490 struct hlist_bl_head
*b
;
492 * Hashed dentries are normally on the dentry hashtable,
493 * with the exception of those newly allocated by
494 * d_obtain_alias, which are always IS_ROOT:
496 if (unlikely(IS_ROOT(dentry
)))
497 b
= &dentry
->d_sb
->s_anon
;
499 b
= d_hash(dentry
->d_name
.hash
);
502 __hlist_bl_del(&dentry
->d_hash
);
504 /* After this call, in-progress rcu-walk path lookup will fail. */
505 write_seqcount_invalidate(&dentry
->d_seq
);
509 void __d_drop(struct dentry
*dentry
)
512 dentry
->d_hash
.pprev
= NULL
;
514 EXPORT_SYMBOL(__d_drop
);
516 void d_drop(struct dentry
*dentry
)
518 spin_lock(&dentry
->d_lock
);
520 spin_unlock(&dentry
->d_lock
);
522 EXPORT_SYMBOL(d_drop
);
524 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
528 * Inform d_walk() and shrink_dentry_list() that we are no longer
529 * attached to the dentry tree
531 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
532 if (unlikely(list_empty(&dentry
->d_child
)))
534 __list_del_entry(&dentry
->d_child
);
536 * Cursors can move around the list of children. While we'd been
537 * a normal list member, it didn't matter - ->d_child.next would've
538 * been updated. However, from now on it won't be and for the
539 * things like d_walk() it might end up with a nasty surprise.
540 * Normally d_walk() doesn't care about cursors moving around -
541 * ->d_lock on parent prevents that and since a cursor has no children
542 * of its own, we get through it without ever unlocking the parent.
543 * There is one exception, though - if we ascend from a child that
544 * gets killed as soon as we unlock it, the next sibling is found
545 * using the value left in its ->d_child.next. And if _that_
546 * pointed to a cursor, and cursor got moved (e.g. by lseek())
547 * before d_walk() regains parent->d_lock, we'll end up skipping
548 * everything the cursor had been moved past.
550 * Solution: make sure that the pointer left behind in ->d_child.next
551 * points to something that won't be moving around. I.e. skip the
554 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
555 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
556 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
558 dentry
->d_child
.next
= next
->d_child
.next
;
562 static void __dentry_kill(struct dentry
*dentry
)
564 struct dentry
*parent
= NULL
;
565 bool can_free
= true;
566 if (!IS_ROOT(dentry
))
567 parent
= dentry
->d_parent
;
570 * The dentry is now unrecoverably dead to the world.
572 lockref_mark_dead(&dentry
->d_lockref
);
575 * inform the fs via d_prune that this dentry is about to be
576 * unhashed and destroyed.
578 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
579 dentry
->d_op
->d_prune(dentry
);
581 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
582 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
585 /* if it was on the hash then remove it */
587 dentry_unlist(dentry
, parent
);
589 spin_unlock(&parent
->d_lock
);
591 dentry_unlink_inode(dentry
);
593 spin_unlock(&dentry
->d_lock
);
594 this_cpu_dec(nr_dentry
);
595 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
596 dentry
->d_op
->d_release(dentry
);
598 spin_lock(&dentry
->d_lock
);
599 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
600 dentry
->d_flags
|= DCACHE_MAY_FREE
;
603 spin_unlock(&dentry
->d_lock
);
604 if (likely(can_free
))
609 * Finish off a dentry we've decided to kill.
610 * dentry->d_lock must be held, returns with it unlocked.
611 * If ref is non-zero, then decrement the refcount too.
612 * Returns dentry requiring refcount drop, or NULL if we're done.
614 static struct dentry
*dentry_kill(struct dentry
*dentry
)
615 __releases(dentry
->d_lock
)
617 struct inode
*inode
= dentry
->d_inode
;
618 struct dentry
*parent
= NULL
;
620 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
623 if (!IS_ROOT(dentry
)) {
624 parent
= dentry
->d_parent
;
625 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
627 spin_unlock(&inode
->i_lock
);
632 __dentry_kill(dentry
);
636 spin_unlock(&dentry
->d_lock
);
637 return dentry
; /* try again with same dentry */
640 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
642 struct dentry
*parent
= dentry
->d_parent
;
645 if (unlikely(dentry
->d_lockref
.count
< 0))
647 if (likely(spin_trylock(&parent
->d_lock
)))
650 spin_unlock(&dentry
->d_lock
);
652 parent
= READ_ONCE(dentry
->d_parent
);
653 spin_lock(&parent
->d_lock
);
655 * We can't blindly lock dentry until we are sure
656 * that we won't violate the locking order.
657 * Any changes of dentry->d_parent must have
658 * been done with parent->d_lock held, so
659 * spin_lock() above is enough of a barrier
660 * for checking if it's still our child.
662 if (unlikely(parent
!= dentry
->d_parent
)) {
663 spin_unlock(&parent
->d_lock
);
666 if (parent
!= dentry
) {
667 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
668 if (unlikely(dentry
->d_lockref
.count
< 0)) {
669 spin_unlock(&parent
->d_lock
);
680 * Try to do a lockless dput(), and return whether that was successful.
682 * If unsuccessful, we return false, having already taken the dentry lock.
684 * The caller needs to hold the RCU read lock, so that the dentry is
685 * guaranteed to stay around even if the refcount goes down to zero!
687 static inline bool fast_dput(struct dentry
*dentry
)
690 unsigned int d_flags
;
693 * If we have a d_op->d_delete() operation, we sould not
694 * let the dentry count go to zero, so use "put_or_lock".
696 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
697 return lockref_put_or_lock(&dentry
->d_lockref
);
700 * .. otherwise, we can try to just decrement the
701 * lockref optimistically.
703 ret
= lockref_put_return(&dentry
->d_lockref
);
706 * If the lockref_put_return() failed due to the lock being held
707 * by somebody else, the fast path has failed. We will need to
708 * get the lock, and then check the count again.
710 if (unlikely(ret
< 0)) {
711 spin_lock(&dentry
->d_lock
);
712 if (dentry
->d_lockref
.count
> 1) {
713 dentry
->d_lockref
.count
--;
714 spin_unlock(&dentry
->d_lock
);
721 * If we weren't the last ref, we're done.
727 * Careful, careful. The reference count went down
728 * to zero, but we don't hold the dentry lock, so
729 * somebody else could get it again, and do another
730 * dput(), and we need to not race with that.
732 * However, there is a very special and common case
733 * where we don't care, because there is nothing to
734 * do: the dentry is still hashed, it does not have
735 * a 'delete' op, and it's referenced and already on
738 * NOTE! Since we aren't locked, these values are
739 * not "stable". However, it is sufficient that at
740 * some point after we dropped the reference the
741 * dentry was hashed and the flags had the proper
742 * value. Other dentry users may have re-gotten
743 * a reference to the dentry and change that, but
744 * our work is done - we can leave the dentry
745 * around with a zero refcount.
748 d_flags
= READ_ONCE(dentry
->d_flags
);
749 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
751 /* Nothing to do? Dropping the reference was all we needed? */
752 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
756 * Not the fast normal case? Get the lock. We've already decremented
757 * the refcount, but we'll need to re-check the situation after
760 spin_lock(&dentry
->d_lock
);
763 * Did somebody else grab a reference to it in the meantime, and
764 * we're no longer the last user after all? Alternatively, somebody
765 * else could have killed it and marked it dead. Either way, we
766 * don't need to do anything else.
768 if (dentry
->d_lockref
.count
) {
769 spin_unlock(&dentry
->d_lock
);
774 * Re-get the reference we optimistically dropped. We hold the
775 * lock, and we just tested that it was zero, so we can just
778 dentry
->d_lockref
.count
= 1;
786 * This is complicated by the fact that we do not want to put
787 * dentries that are no longer on any hash chain on the unused
788 * list: we'd much rather just get rid of them immediately.
790 * However, that implies that we have to traverse the dentry
791 * tree upwards to the parents which might _also_ now be
792 * scheduled for deletion (it may have been only waiting for
793 * its last child to go away).
795 * This tail recursion is done by hand as we don't want to depend
796 * on the compiler to always get this right (gcc generally doesn't).
797 * Real recursion would eat up our stack space.
801 * dput - release a dentry
802 * @dentry: dentry to release
804 * Release a dentry. This will drop the usage count and if appropriate
805 * call the dentry unlink method as well as removing it from the queues and
806 * releasing its resources. If the parent dentries were scheduled for release
807 * they too may now get deleted.
809 void dput(struct dentry
*dentry
)
811 if (unlikely(!dentry
))
818 if (likely(fast_dput(dentry
))) {
823 /* Slow case: now with the dentry lock held */
826 WARN_ON(d_in_lookup(dentry
));
828 /* Unreachable? Get rid of it */
829 if (unlikely(d_unhashed(dentry
)))
832 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
835 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
836 if (dentry
->d_op
->d_delete(dentry
))
840 dentry_lru_add(dentry
);
842 dentry
->d_lockref
.count
--;
843 spin_unlock(&dentry
->d_lock
);
847 dentry
= dentry_kill(dentry
);
856 /* This must be called with d_lock held */
857 static inline void __dget_dlock(struct dentry
*dentry
)
859 dentry
->d_lockref
.count
++;
862 static inline void __dget(struct dentry
*dentry
)
864 lockref_get(&dentry
->d_lockref
);
867 struct dentry
*dget_parent(struct dentry
*dentry
)
873 * Do optimistic parent lookup without any
877 ret
= READ_ONCE(dentry
->d_parent
);
878 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
880 if (likely(gotref
)) {
881 if (likely(ret
== READ_ONCE(dentry
->d_parent
)))
888 * Don't need rcu_dereference because we re-check it was correct under
892 ret
= dentry
->d_parent
;
893 spin_lock(&ret
->d_lock
);
894 if (unlikely(ret
!= dentry
->d_parent
)) {
895 spin_unlock(&ret
->d_lock
);
900 BUG_ON(!ret
->d_lockref
.count
);
901 ret
->d_lockref
.count
++;
902 spin_unlock(&ret
->d_lock
);
905 EXPORT_SYMBOL(dget_parent
);
908 * d_find_alias - grab a hashed alias of inode
909 * @inode: inode in question
911 * If inode has a hashed alias, or is a directory and has any alias,
912 * acquire the reference to alias and return it. Otherwise return NULL.
913 * Notice that if inode is a directory there can be only one alias and
914 * it can be unhashed only if it has no children, or if it is the root
915 * of a filesystem, or if the directory was renamed and d_revalidate
916 * was the first vfs operation to notice.
918 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
919 * any other hashed alias over that one.
921 static struct dentry
*__d_find_alias(struct inode
*inode
)
923 struct dentry
*alias
, *discon_alias
;
927 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
928 spin_lock(&alias
->d_lock
);
929 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
930 if (IS_ROOT(alias
) &&
931 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
932 discon_alias
= alias
;
935 spin_unlock(&alias
->d_lock
);
939 spin_unlock(&alias
->d_lock
);
942 alias
= discon_alias
;
943 spin_lock(&alias
->d_lock
);
944 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
946 spin_unlock(&alias
->d_lock
);
949 spin_unlock(&alias
->d_lock
);
955 struct dentry
*d_find_alias(struct inode
*inode
)
957 struct dentry
*de
= NULL
;
959 if (!hlist_empty(&inode
->i_dentry
)) {
960 spin_lock(&inode
->i_lock
);
961 de
= __d_find_alias(inode
);
962 spin_unlock(&inode
->i_lock
);
966 EXPORT_SYMBOL(d_find_alias
);
969 * Try to kill dentries associated with this inode.
970 * WARNING: you must own a reference to inode.
972 void d_prune_aliases(struct inode
*inode
)
974 struct dentry
*dentry
;
976 spin_lock(&inode
->i_lock
);
977 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
978 spin_lock(&dentry
->d_lock
);
979 if (!dentry
->d_lockref
.count
) {
980 struct dentry
*parent
= lock_parent(dentry
);
981 if (likely(!dentry
->d_lockref
.count
)) {
982 __dentry_kill(dentry
);
987 spin_unlock(&parent
->d_lock
);
989 spin_unlock(&dentry
->d_lock
);
991 spin_unlock(&inode
->i_lock
);
993 EXPORT_SYMBOL(d_prune_aliases
);
995 static void shrink_dentry_list(struct list_head
*list
)
997 struct dentry
*dentry
, *parent
;
999 while (!list_empty(list
)) {
1000 struct inode
*inode
;
1001 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
1002 spin_lock(&dentry
->d_lock
);
1003 parent
= lock_parent(dentry
);
1006 * The dispose list is isolated and dentries are not accounted
1007 * to the LRU here, so we can simply remove it from the list
1008 * here regardless of whether it is referenced or not.
1010 d_shrink_del(dentry
);
1013 * We found an inuse dentry which was not removed from
1014 * the LRU because of laziness during lookup. Do not free it.
1016 if (dentry
->d_lockref
.count
> 0) {
1017 spin_unlock(&dentry
->d_lock
);
1019 spin_unlock(&parent
->d_lock
);
1024 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
1025 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
1026 spin_unlock(&dentry
->d_lock
);
1028 spin_unlock(&parent
->d_lock
);
1030 dentry_free(dentry
);
1034 inode
= dentry
->d_inode
;
1035 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1036 d_shrink_add(dentry
, list
);
1037 spin_unlock(&dentry
->d_lock
);
1039 spin_unlock(&parent
->d_lock
);
1043 __dentry_kill(dentry
);
1046 * We need to prune ancestors too. This is necessary to prevent
1047 * quadratic behavior of shrink_dcache_parent(), but is also
1048 * expected to be beneficial in reducing dentry cache
1052 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1053 parent
= lock_parent(dentry
);
1054 if (dentry
->d_lockref
.count
!= 1) {
1055 dentry
->d_lockref
.count
--;
1056 spin_unlock(&dentry
->d_lock
);
1058 spin_unlock(&parent
->d_lock
);
1061 inode
= dentry
->d_inode
; /* can't be NULL */
1062 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1063 spin_unlock(&dentry
->d_lock
);
1065 spin_unlock(&parent
->d_lock
);
1069 __dentry_kill(dentry
);
1075 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1076 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1078 struct list_head
*freeable
= arg
;
1079 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1083 * we are inverting the lru lock/dentry->d_lock here,
1084 * so use a trylock. If we fail to get the lock, just skip
1087 if (!spin_trylock(&dentry
->d_lock
))
1091 * Referenced dentries are still in use. If they have active
1092 * counts, just remove them from the LRU. Otherwise give them
1093 * another pass through the LRU.
1095 if (dentry
->d_lockref
.count
) {
1096 d_lru_isolate(lru
, dentry
);
1097 spin_unlock(&dentry
->d_lock
);
1101 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1102 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1103 spin_unlock(&dentry
->d_lock
);
1106 * The list move itself will be made by the common LRU code. At
1107 * this point, we've dropped the dentry->d_lock but keep the
1108 * lru lock. This is safe to do, since every list movement is
1109 * protected by the lru lock even if both locks are held.
1111 * This is guaranteed by the fact that all LRU management
1112 * functions are intermediated by the LRU API calls like
1113 * list_lru_add and list_lru_del. List movement in this file
1114 * only ever occur through this functions or through callbacks
1115 * like this one, that are called from the LRU API.
1117 * The only exceptions to this are functions like
1118 * shrink_dentry_list, and code that first checks for the
1119 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1120 * operating only with stack provided lists after they are
1121 * properly isolated from the main list. It is thus, always a
1127 d_lru_shrink_move(lru
, dentry
, freeable
);
1128 spin_unlock(&dentry
->d_lock
);
1134 * prune_dcache_sb - shrink the dcache
1136 * @sc: shrink control, passed to list_lru_shrink_walk()
1138 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1139 * is done when we need more memory and called from the superblock shrinker
1142 * This function may fail to free any resources if all the dentries are in
1145 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1150 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1151 dentry_lru_isolate
, &dispose
);
1152 shrink_dentry_list(&dispose
);
1156 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1157 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1159 struct list_head
*freeable
= arg
;
1160 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1163 * we are inverting the lru lock/dentry->d_lock here,
1164 * so use a trylock. If we fail to get the lock, just skip
1167 if (!spin_trylock(&dentry
->d_lock
))
1170 d_lru_shrink_move(lru
, dentry
, freeable
);
1171 spin_unlock(&dentry
->d_lock
);
1178 * shrink_dcache_sb - shrink dcache for a superblock
1181 * Shrink the dcache for the specified super block. This is used to free
1182 * the dcache before unmounting a file system.
1184 void shrink_dcache_sb(struct super_block
*sb
)
1189 list_lru_walk(&sb
->s_dentry_lru
,
1190 dentry_lru_isolate_shrink
, &dispose
, 1024);
1191 shrink_dentry_list(&dispose
);
1193 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1195 EXPORT_SYMBOL(shrink_dcache_sb
);
1198 * enum d_walk_ret - action to talke during tree walk
1199 * @D_WALK_CONTINUE: contrinue walk
1200 * @D_WALK_QUIT: quit walk
1201 * @D_WALK_NORETRY: quit when retry is needed
1202 * @D_WALK_SKIP: skip this dentry and its children
1212 * d_walk - walk the dentry tree
1213 * @parent: start of walk
1214 * @data: data passed to @enter() and @finish()
1215 * @enter: callback when first entering the dentry
1216 * @finish: callback when successfully finished the walk
1218 * The @enter() and @finish() callbacks are called with d_lock held.
1220 void d_walk(struct dentry
*parent
, void *data
,
1221 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1222 void (*finish
)(void *))
1224 struct dentry
*this_parent
;
1225 struct list_head
*next
;
1227 enum d_walk_ret ret
;
1231 read_seqbegin_or_lock(&rename_lock
, &seq
);
1232 this_parent
= parent
;
1233 spin_lock(&this_parent
->d_lock
);
1235 ret
= enter(data
, this_parent
);
1237 case D_WALK_CONTINUE
:
1242 case D_WALK_NORETRY
:
1247 next
= this_parent
->d_subdirs
.next
;
1249 while (next
!= &this_parent
->d_subdirs
) {
1250 struct list_head
*tmp
= next
;
1251 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1254 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1257 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1259 ret
= enter(data
, dentry
);
1261 case D_WALK_CONTINUE
:
1264 spin_unlock(&dentry
->d_lock
);
1266 case D_WALK_NORETRY
:
1270 spin_unlock(&dentry
->d_lock
);
1274 if (!list_empty(&dentry
->d_subdirs
)) {
1275 spin_unlock(&this_parent
->d_lock
);
1276 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1277 this_parent
= dentry
;
1278 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1281 spin_unlock(&dentry
->d_lock
);
1284 * All done at this level ... ascend and resume the search.
1288 if (this_parent
!= parent
) {
1289 struct dentry
*child
= this_parent
;
1290 this_parent
= child
->d_parent
;
1292 spin_unlock(&child
->d_lock
);
1293 spin_lock(&this_parent
->d_lock
);
1295 /* might go back up the wrong parent if we have had a rename. */
1296 if (need_seqretry(&rename_lock
, seq
))
1298 /* go into the first sibling still alive */
1300 next
= child
->d_child
.next
;
1301 if (next
== &this_parent
->d_subdirs
)
1303 child
= list_entry(next
, struct dentry
, d_child
);
1304 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1308 if (need_seqretry(&rename_lock
, seq
))
1315 spin_unlock(&this_parent
->d_lock
);
1316 done_seqretry(&rename_lock
, seq
);
1320 spin_unlock(&this_parent
->d_lock
);
1328 EXPORT_SYMBOL_GPL(d_walk
);
1330 struct check_mount
{
1331 struct vfsmount
*mnt
;
1332 unsigned int mounted
;
1335 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1337 struct check_mount
*info
= data
;
1338 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1340 if (likely(!d_mountpoint(dentry
)))
1341 return D_WALK_CONTINUE
;
1342 if (__path_is_mountpoint(&path
)) {
1346 return D_WALK_CONTINUE
;
1350 * path_has_submounts - check for mounts over a dentry in the
1351 * current namespace.
1352 * @parent: path to check.
1354 * Return true if the parent or its subdirectories contain
1355 * a mount point in the current namespace.
1357 int path_has_submounts(const struct path
*parent
)
1359 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1361 read_seqlock_excl(&mount_lock
);
1362 d_walk(parent
->dentry
, &data
, path_check_mount
, NULL
);
1363 read_sequnlock_excl(&mount_lock
);
1365 return data
.mounted
;
1367 EXPORT_SYMBOL(path_has_submounts
);
1370 * Called by mount code to set a mountpoint and check if the mountpoint is
1371 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1372 * subtree can become unreachable).
1374 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1375 * this reason take rename_lock and d_lock on dentry and ancestors.
1377 int d_set_mounted(struct dentry
*dentry
)
1381 write_seqlock(&rename_lock
);
1382 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1383 /* Need exclusion wrt. d_invalidate() */
1384 spin_lock(&p
->d_lock
);
1385 if (unlikely(d_unhashed(p
))) {
1386 spin_unlock(&p
->d_lock
);
1389 spin_unlock(&p
->d_lock
);
1391 spin_lock(&dentry
->d_lock
);
1392 if (!d_unlinked(dentry
)) {
1394 if (!d_mountpoint(dentry
)) {
1395 dentry
->d_flags
|= DCACHE_MOUNTED
;
1399 spin_unlock(&dentry
->d_lock
);
1401 write_sequnlock(&rename_lock
);
1406 * Search the dentry child list of the specified parent,
1407 * and move any unused dentries to the end of the unused
1408 * list for prune_dcache(). We descend to the next level
1409 * whenever the d_subdirs list is non-empty and continue
1412 * It returns zero iff there are no unused children,
1413 * otherwise it returns the number of children moved to
1414 * the end of the unused list. This may not be the total
1415 * number of unused children, because select_parent can
1416 * drop the lock and return early due to latency
1420 struct select_data
{
1421 struct dentry
*start
;
1422 struct list_head dispose
;
1426 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1428 struct select_data
*data
= _data
;
1429 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1431 if (data
->start
== dentry
)
1434 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1437 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1439 if (!dentry
->d_lockref
.count
) {
1440 d_shrink_add(dentry
, &data
->dispose
);
1445 * We can return to the caller if we have found some (this
1446 * ensures forward progress). We'll be coming back to find
1449 if (!list_empty(&data
->dispose
))
1450 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1456 * shrink_dcache_parent - prune dcache
1457 * @parent: parent of entries to prune
1459 * Prune the dcache to remove unused children of the parent dentry.
1461 void shrink_dcache_parent(struct dentry
*parent
)
1464 struct select_data data
;
1466 INIT_LIST_HEAD(&data
.dispose
);
1467 data
.start
= parent
;
1470 d_walk(parent
, &data
, select_collect
, NULL
);
1474 shrink_dentry_list(&data
.dispose
);
1478 EXPORT_SYMBOL(shrink_dcache_parent
);
1480 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1482 /* it has busy descendents; complain about those instead */
1483 if (!list_empty(&dentry
->d_subdirs
))
1484 return D_WALK_CONTINUE
;
1486 /* root with refcount 1 is fine */
1487 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1488 return D_WALK_CONTINUE
;
1490 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1491 " still in use (%d) [unmount of %s %s]\n",
1494 dentry
->d_inode
->i_ino
: 0UL,
1496 dentry
->d_lockref
.count
,
1497 dentry
->d_sb
->s_type
->name
,
1498 dentry
->d_sb
->s_id
);
1500 return D_WALK_CONTINUE
;
1503 static void do_one_tree(struct dentry
*dentry
)
1505 shrink_dcache_parent(dentry
);
1506 d_walk(dentry
, dentry
, umount_check
, NULL
);
1512 * destroy the dentries attached to a superblock on unmounting
1514 void shrink_dcache_for_umount(struct super_block
*sb
)
1516 struct dentry
*dentry
;
1518 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1520 dentry
= sb
->s_root
;
1522 do_one_tree(dentry
);
1524 while (!hlist_bl_empty(&sb
->s_anon
)) {
1525 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1526 do_one_tree(dentry
);
1530 struct detach_data
{
1531 struct select_data select
;
1532 struct dentry
*mountpoint
;
1534 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1536 struct detach_data
*data
= _data
;
1538 if (d_mountpoint(dentry
)) {
1539 __dget_dlock(dentry
);
1540 data
->mountpoint
= dentry
;
1544 return select_collect(&data
->select
, dentry
);
1547 static void check_and_drop(void *_data
)
1549 struct detach_data
*data
= _data
;
1551 if (!data
->mountpoint
&& list_empty(&data
->select
.dispose
))
1552 __d_drop(data
->select
.start
);
1556 * d_invalidate - detach submounts, prune dcache, and drop
1557 * @dentry: dentry to invalidate (aka detach, prune and drop)
1561 * The final d_drop is done as an atomic operation relative to
1562 * rename_lock ensuring there are no races with d_set_mounted. This
1563 * ensures there are no unhashed dentries on the path to a mountpoint.
1565 void d_invalidate(struct dentry
*dentry
)
1568 * If it's already been dropped, return OK.
1570 spin_lock(&dentry
->d_lock
);
1571 if (d_unhashed(dentry
)) {
1572 spin_unlock(&dentry
->d_lock
);
1575 spin_unlock(&dentry
->d_lock
);
1577 /* Negative dentries can be dropped without further checks */
1578 if (!dentry
->d_inode
) {
1584 struct detach_data data
;
1586 data
.mountpoint
= NULL
;
1587 INIT_LIST_HEAD(&data
.select
.dispose
);
1588 data
.select
.start
= dentry
;
1589 data
.select
.found
= 0;
1591 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1593 if (!list_empty(&data
.select
.dispose
))
1594 shrink_dentry_list(&data
.select
.dispose
);
1595 else if (!data
.mountpoint
)
1598 if (data
.mountpoint
) {
1599 detach_mounts(data
.mountpoint
);
1600 dput(data
.mountpoint
);
1605 EXPORT_SYMBOL(d_invalidate
);
1608 * __d_alloc - allocate a dcache entry
1609 * @sb: filesystem it will belong to
1610 * @name: qstr of the name
1612 * Allocates a dentry. It returns %NULL if there is insufficient memory
1613 * available. On a success the dentry is returned. The name passed in is
1614 * copied and the copy passed in may be reused after this call.
1617 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1619 struct external_name
*ext
= NULL
;
1620 struct dentry
*dentry
;
1624 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1629 * We guarantee that the inline name is always NUL-terminated.
1630 * This way the memcpy() done by the name switching in rename
1631 * will still always have a NUL at the end, even if we might
1632 * be overwriting an internal NUL character
1634 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1635 if (unlikely(!name
)) {
1637 dname
= dentry
->d_iname
;
1638 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1639 size_t size
= offsetof(struct external_name
, name
[1]);
1640 ext
= kmalloc(size
+ name
->len
, GFP_KERNEL_ACCOUNT
);
1642 kmem_cache_free(dentry_cache
, dentry
);
1645 atomic_set(&ext
->u
.count
, 1);
1647 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1648 kasan_unpoison_shadow(dname
,
1649 round_up(name
->len
+ 1, sizeof(unsigned long)));
1651 dname
= dentry
->d_iname
;
1654 dentry
->d_name
.len
= name
->len
;
1655 dentry
->d_name
.hash
= name
->hash
;
1656 memcpy(dname
, name
->name
, name
->len
);
1657 dname
[name
->len
] = 0;
1659 /* Make sure we always see the terminating NUL character */
1661 dentry
->d_name
.name
= dname
;
1663 dentry
->d_lockref
.count
= 1;
1664 dentry
->d_flags
= 0;
1665 spin_lock_init(&dentry
->d_lock
);
1666 seqcount_init(&dentry
->d_seq
);
1667 dentry
->d_inode
= NULL
;
1668 dentry
->d_parent
= dentry
;
1670 dentry
->d_op
= NULL
;
1671 dentry
->d_fsdata
= NULL
;
1672 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1673 INIT_LIST_HEAD(&dentry
->d_lru
);
1674 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1675 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1676 INIT_LIST_HEAD(&dentry
->d_child
);
1677 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1679 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1680 err
= dentry
->d_op
->d_init(dentry
);
1682 if (dname_external(dentry
))
1683 kfree(external_name(dentry
));
1684 kmem_cache_free(dentry_cache
, dentry
);
1689 if (unlikely(ext
)) {
1690 pg_data_t
*pgdat
= page_pgdat(virt_to_page(ext
));
1691 mod_node_page_state(pgdat
, NR_INDIRECTLY_RECLAIMABLE_BYTES
,
1695 this_cpu_inc(nr_dentry
);
1701 * d_alloc - allocate a dcache entry
1702 * @parent: parent of entry to allocate
1703 * @name: qstr of the name
1705 * Allocates a dentry. It returns %NULL if there is insufficient memory
1706 * available. On a success the dentry is returned. The name passed in is
1707 * copied and the copy passed in may be reused after this call.
1709 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1711 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1714 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1715 spin_lock(&parent
->d_lock
);
1717 * don't need child lock because it is not subject
1718 * to concurrency here
1720 __dget_dlock(parent
);
1721 dentry
->d_parent
= parent
;
1722 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1723 spin_unlock(&parent
->d_lock
);
1727 EXPORT_SYMBOL(d_alloc
);
1729 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1731 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1733 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1734 dentry
->d_parent
= dget(parent
);
1740 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1741 * @sb: the superblock
1742 * @name: qstr of the name
1744 * For a filesystem that just pins its dentries in memory and never
1745 * performs lookups at all, return an unhashed IS_ROOT dentry.
1747 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1749 return __d_alloc(sb
, name
);
1751 EXPORT_SYMBOL(d_alloc_pseudo
);
1753 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1758 q
.hash_len
= hashlen_string(parent
, name
);
1759 return d_alloc(parent
, &q
);
1761 EXPORT_SYMBOL(d_alloc_name
);
1763 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1765 WARN_ON_ONCE(dentry
->d_op
);
1766 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1768 DCACHE_OP_REVALIDATE
|
1769 DCACHE_OP_WEAK_REVALIDATE
|
1776 dentry
->d_flags
|= DCACHE_OP_HASH
;
1778 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1779 if (op
->d_revalidate
)
1780 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1781 if (op
->d_weak_revalidate
)
1782 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1784 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1786 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1788 dentry
->d_flags
|= DCACHE_OP_REAL
;
1791 EXPORT_SYMBOL(d_set_d_op
);
1795 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1796 * @dentry - The dentry to mark
1798 * Mark a dentry as falling through to the lower layer (as set with
1799 * d_pin_lower()). This flag may be recorded on the medium.
1801 void d_set_fallthru(struct dentry
*dentry
)
1803 spin_lock(&dentry
->d_lock
);
1804 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1805 spin_unlock(&dentry
->d_lock
);
1807 EXPORT_SYMBOL(d_set_fallthru
);
1809 static unsigned d_flags_for_inode(struct inode
*inode
)
1811 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1814 return DCACHE_MISS_TYPE
;
1816 if (S_ISDIR(inode
->i_mode
)) {
1817 add_flags
= DCACHE_DIRECTORY_TYPE
;
1818 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1819 if (unlikely(!inode
->i_op
->lookup
))
1820 add_flags
= DCACHE_AUTODIR_TYPE
;
1822 inode
->i_opflags
|= IOP_LOOKUP
;
1824 goto type_determined
;
1827 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1828 if (unlikely(inode
->i_op
->get_link
)) {
1829 add_flags
= DCACHE_SYMLINK_TYPE
;
1830 goto type_determined
;
1832 inode
->i_opflags
|= IOP_NOFOLLOW
;
1835 if (unlikely(!S_ISREG(inode
->i_mode
)))
1836 add_flags
= DCACHE_SPECIAL_TYPE
;
1839 if (unlikely(IS_AUTOMOUNT(inode
)))
1840 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1844 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1846 unsigned add_flags
= d_flags_for_inode(inode
);
1847 WARN_ON(d_in_lookup(dentry
));
1849 spin_lock(&dentry
->d_lock
);
1850 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1851 raw_write_seqcount_begin(&dentry
->d_seq
);
1852 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1853 raw_write_seqcount_end(&dentry
->d_seq
);
1854 fsnotify_update_flags(dentry
);
1855 spin_unlock(&dentry
->d_lock
);
1859 * d_instantiate - fill in inode information for a dentry
1860 * @entry: dentry to complete
1861 * @inode: inode to attach to this dentry
1863 * Fill in inode information in the entry.
1865 * This turns negative dentries into productive full members
1868 * NOTE! This assumes that the inode count has been incremented
1869 * (or otherwise set) by the caller to indicate that it is now
1870 * in use by the dcache.
1873 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1875 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1877 security_d_instantiate(entry
, inode
);
1878 spin_lock(&inode
->i_lock
);
1879 __d_instantiate(entry
, inode
);
1880 spin_unlock(&inode
->i_lock
);
1883 EXPORT_SYMBOL(d_instantiate
);
1886 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1887 * with lockdep-related part of unlock_new_inode() done before
1888 * anything else. Use that instead of open-coding d_instantiate()/
1889 * unlock_new_inode() combinations.
1891 void d_instantiate_new(struct dentry
*entry
, struct inode
*inode
)
1893 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1895 lockdep_annotate_inode_mutex_key(inode
);
1896 security_d_instantiate(entry
, inode
);
1897 spin_lock(&inode
->i_lock
);
1898 __d_instantiate(entry
, inode
);
1899 WARN_ON(!(inode
->i_state
& I_NEW
));
1900 inode
->i_state
&= ~I_NEW
;
1902 wake_up_bit(&inode
->i_state
, __I_NEW
);
1903 spin_unlock(&inode
->i_lock
);
1905 EXPORT_SYMBOL(d_instantiate_new
);
1908 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1909 * @entry: dentry to complete
1910 * @inode: inode to attach to this dentry
1912 * Fill in inode information in the entry. If a directory alias is found, then
1913 * return an error (and drop inode). Together with d_materialise_unique() this
1914 * guarantees that a directory inode may never have more than one alias.
1916 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1918 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1920 security_d_instantiate(entry
, inode
);
1921 spin_lock(&inode
->i_lock
);
1922 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1923 spin_unlock(&inode
->i_lock
);
1927 __d_instantiate(entry
, inode
);
1928 spin_unlock(&inode
->i_lock
);
1932 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1934 struct dentry
*d_make_root(struct inode
*root_inode
)
1936 struct dentry
*res
= NULL
;
1939 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1941 res
->d_flags
|= DCACHE_RCUACCESS
;
1942 d_instantiate(res
, root_inode
);
1949 EXPORT_SYMBOL(d_make_root
);
1951 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1953 struct dentry
*alias
;
1955 if (hlist_empty(&inode
->i_dentry
))
1957 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1963 * d_find_any_alias - find any alias for a given inode
1964 * @inode: inode to find an alias for
1966 * If any aliases exist for the given inode, take and return a
1967 * reference for one of them. If no aliases exist, return %NULL.
1969 struct dentry
*d_find_any_alias(struct inode
*inode
)
1973 spin_lock(&inode
->i_lock
);
1974 de
= __d_find_any_alias(inode
);
1975 spin_unlock(&inode
->i_lock
);
1978 EXPORT_SYMBOL(d_find_any_alias
);
1980 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1987 return ERR_PTR(-ESTALE
);
1989 return ERR_CAST(inode
);
1991 res
= d_find_any_alias(inode
);
1995 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1997 res
= ERR_PTR(-ENOMEM
);
2001 security_d_instantiate(tmp
, inode
);
2002 spin_lock(&inode
->i_lock
);
2003 res
= __d_find_any_alias(inode
);
2005 spin_unlock(&inode
->i_lock
);
2010 /* attach a disconnected dentry */
2011 add_flags
= d_flags_for_inode(inode
);
2014 add_flags
|= DCACHE_DISCONNECTED
;
2016 spin_lock(&tmp
->d_lock
);
2017 __d_set_inode_and_type(tmp
, inode
, add_flags
);
2018 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
2019 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
2020 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
2021 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
2022 spin_unlock(&tmp
->d_lock
);
2023 spin_unlock(&inode
->i_lock
);
2033 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2034 * @inode: inode to allocate the dentry for
2036 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2037 * similar open by handle operations. The returned dentry may be anonymous,
2038 * or may have a full name (if the inode was already in the cache).
2040 * When called on a directory inode, we must ensure that the inode only ever
2041 * has one dentry. If a dentry is found, that is returned instead of
2042 * allocating a new one.
2044 * On successful return, the reference to the inode has been transferred
2045 * to the dentry. In case of an error the reference on the inode is released.
2046 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2047 * be passed in and the error will be propagated to the return value,
2048 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2050 struct dentry
*d_obtain_alias(struct inode
*inode
)
2052 return __d_obtain_alias(inode
, 1);
2054 EXPORT_SYMBOL(d_obtain_alias
);
2057 * d_obtain_root - find or allocate a dentry for a given inode
2058 * @inode: inode to allocate the dentry for
2060 * Obtain an IS_ROOT dentry for the root of a filesystem.
2062 * We must ensure that directory inodes only ever have one dentry. If a
2063 * dentry is found, that is returned instead of allocating a new one.
2065 * On successful return, the reference to the inode has been transferred
2066 * to the dentry. In case of an error the reference on the inode is
2067 * released. A %NULL or IS_ERR inode may be passed in and will be the
2068 * error will be propagate to the return value, with a %NULL @inode
2069 * replaced by ERR_PTR(-ESTALE).
2071 struct dentry
*d_obtain_root(struct inode
*inode
)
2073 return __d_obtain_alias(inode
, 0);
2075 EXPORT_SYMBOL(d_obtain_root
);
2078 * d_add_ci - lookup or allocate new dentry with case-exact name
2079 * @inode: the inode case-insensitive lookup has found
2080 * @dentry: the negative dentry that was passed to the parent's lookup func
2081 * @name: the case-exact name to be associated with the returned dentry
2083 * This is to avoid filling the dcache with case-insensitive names to the
2084 * same inode, only the actual correct case is stored in the dcache for
2085 * case-insensitive filesystems.
2087 * For a case-insensitive lookup match and if the the case-exact dentry
2088 * already exists in in the dcache, use it and return it.
2090 * If no entry exists with the exact case name, allocate new dentry with
2091 * the exact case, and return the spliced entry.
2093 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2096 struct dentry
*found
, *res
;
2099 * First check if a dentry matching the name already exists,
2100 * if not go ahead and create it now.
2102 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2107 if (d_in_lookup(dentry
)) {
2108 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2110 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2115 found
= d_alloc(dentry
->d_parent
, name
);
2118 return ERR_PTR(-ENOMEM
);
2121 res
= d_splice_alias(inode
, found
);
2128 EXPORT_SYMBOL(d_add_ci
);
2131 static inline bool d_same_name(const struct dentry
*dentry
,
2132 const struct dentry
*parent
,
2133 const struct qstr
*name
)
2135 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2136 if (dentry
->d_name
.len
!= name
->len
)
2138 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2140 return parent
->d_op
->d_compare(dentry
,
2141 dentry
->d_name
.len
, dentry
->d_name
.name
,
2146 * __d_lookup_rcu - search for a dentry (racy, store-free)
2147 * @parent: parent dentry
2148 * @name: qstr of name we wish to find
2149 * @seqp: returns d_seq value at the point where the dentry was found
2150 * Returns: dentry, or NULL
2152 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2153 * resolution (store-free path walking) design described in
2154 * Documentation/filesystems/path-lookup.txt.
2156 * This is not to be used outside core vfs.
2158 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2159 * held, and rcu_read_lock held. The returned dentry must not be stored into
2160 * without taking d_lock and checking d_seq sequence count against @seq
2163 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2166 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2167 * the returned dentry, so long as its parent's seqlock is checked after the
2168 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2169 * is formed, giving integrity down the path walk.
2171 * NOTE! The caller *has* to check the resulting dentry against the sequence
2172 * number we've returned before using any of the resulting dentry state!
2174 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2175 const struct qstr
*name
,
2178 u64 hashlen
= name
->hash_len
;
2179 const unsigned char *str
= name
->name
;
2180 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2181 struct hlist_bl_node
*node
;
2182 struct dentry
*dentry
;
2185 * Note: There is significant duplication with __d_lookup_rcu which is
2186 * required to prevent single threaded performance regressions
2187 * especially on architectures where smp_rmb (in seqcounts) are costly.
2188 * Keep the two functions in sync.
2192 * The hash list is protected using RCU.
2194 * Carefully use d_seq when comparing a candidate dentry, to avoid
2195 * races with d_move().
2197 * It is possible that concurrent renames can mess up our list
2198 * walk here and result in missing our dentry, resulting in the
2199 * false-negative result. d_lookup() protects against concurrent
2200 * renames using rename_lock seqlock.
2202 * See Documentation/filesystems/path-lookup.txt for more details.
2204 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2209 * The dentry sequence count protects us from concurrent
2210 * renames, and thus protects parent and name fields.
2212 * The caller must perform a seqcount check in order
2213 * to do anything useful with the returned dentry.
2215 * NOTE! We do a "raw" seqcount_begin here. That means that
2216 * we don't wait for the sequence count to stabilize if it
2217 * is in the middle of a sequence change. If we do the slow
2218 * dentry compare, we will do seqretries until it is stable,
2219 * and if we end up with a successful lookup, we actually
2220 * want to exit RCU lookup anyway.
2222 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2223 * we are still guaranteed NUL-termination of ->d_name.name.
2225 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2226 if (dentry
->d_parent
!= parent
)
2228 if (d_unhashed(dentry
))
2231 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2234 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2236 tlen
= dentry
->d_name
.len
;
2237 tname
= dentry
->d_name
.name
;
2238 /* we want a consistent (name,len) pair */
2239 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2243 if (parent
->d_op
->d_compare(dentry
,
2244 tlen
, tname
, name
) != 0)
2247 if (dentry
->d_name
.hash_len
!= hashlen
)
2249 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2259 * d_lookup - search for a dentry
2260 * @parent: parent dentry
2261 * @name: qstr of name we wish to find
2262 * Returns: dentry, or NULL
2264 * d_lookup searches the children of the parent dentry for the name in
2265 * question. If the dentry is found its reference count is incremented and the
2266 * dentry is returned. The caller must use dput to free the entry when it has
2267 * finished using it. %NULL is returned if the dentry does not exist.
2269 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2271 struct dentry
*dentry
;
2275 seq
= read_seqbegin(&rename_lock
);
2276 dentry
= __d_lookup(parent
, name
);
2279 } while (read_seqretry(&rename_lock
, seq
));
2282 EXPORT_SYMBOL(d_lookup
);
2285 * __d_lookup - search for a dentry (racy)
2286 * @parent: parent dentry
2287 * @name: qstr of name we wish to find
2288 * Returns: dentry, or NULL
2290 * __d_lookup is like d_lookup, however it may (rarely) return a
2291 * false-negative result due to unrelated rename activity.
2293 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2294 * however it must be used carefully, eg. with a following d_lookup in
2295 * the case of failure.
2297 * __d_lookup callers must be commented.
2299 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2301 unsigned int hash
= name
->hash
;
2302 struct hlist_bl_head
*b
= d_hash(hash
);
2303 struct hlist_bl_node
*node
;
2304 struct dentry
*found
= NULL
;
2305 struct dentry
*dentry
;
2308 * Note: There is significant duplication with __d_lookup_rcu which is
2309 * required to prevent single threaded performance regressions
2310 * especially on architectures where smp_rmb (in seqcounts) are costly.
2311 * Keep the two functions in sync.
2315 * The hash list is protected using RCU.
2317 * Take d_lock when comparing a candidate dentry, to avoid races
2320 * It is possible that concurrent renames can mess up our list
2321 * walk here and result in missing our dentry, resulting in the
2322 * false-negative result. d_lookup() protects against concurrent
2323 * renames using rename_lock seqlock.
2325 * See Documentation/filesystems/path-lookup.txt for more details.
2329 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2331 if (dentry
->d_name
.hash
!= hash
)
2334 spin_lock(&dentry
->d_lock
);
2335 if (dentry
->d_parent
!= parent
)
2337 if (d_unhashed(dentry
))
2340 if (!d_same_name(dentry
, parent
, name
))
2343 dentry
->d_lockref
.count
++;
2345 spin_unlock(&dentry
->d_lock
);
2348 spin_unlock(&dentry
->d_lock
);
2356 * d_hash_and_lookup - hash the qstr then search for a dentry
2357 * @dir: Directory to search in
2358 * @name: qstr of name we wish to find
2360 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2362 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2365 * Check for a fs-specific hash function. Note that we must
2366 * calculate the standard hash first, as the d_op->d_hash()
2367 * routine may choose to leave the hash value unchanged.
2369 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2370 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2371 int err
= dir
->d_op
->d_hash(dir
, name
);
2372 if (unlikely(err
< 0))
2373 return ERR_PTR(err
);
2375 return d_lookup(dir
, name
);
2377 EXPORT_SYMBOL(d_hash_and_lookup
);
2380 * When a file is deleted, we have two options:
2381 * - turn this dentry into a negative dentry
2382 * - unhash this dentry and free it.
2384 * Usually, we want to just turn this into
2385 * a negative dentry, but if anybody else is
2386 * currently using the dentry or the inode
2387 * we can't do that and we fall back on removing
2388 * it from the hash queues and waiting for
2389 * it to be deleted later when it has no users
2393 * d_delete - delete a dentry
2394 * @dentry: The dentry to delete
2396 * Turn the dentry into a negative dentry if possible, otherwise
2397 * remove it from the hash queues so it can be deleted later
2400 void d_delete(struct dentry
* dentry
)
2402 struct inode
*inode
;
2405 * Are we the only user?
2408 spin_lock(&dentry
->d_lock
);
2409 inode
= dentry
->d_inode
;
2410 isdir
= S_ISDIR(inode
->i_mode
);
2411 if (dentry
->d_lockref
.count
== 1) {
2412 if (!spin_trylock(&inode
->i_lock
)) {
2413 spin_unlock(&dentry
->d_lock
);
2417 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2418 dentry_unlink_inode(dentry
);
2419 fsnotify_nameremove(dentry
, isdir
);
2423 if (!d_unhashed(dentry
))
2426 spin_unlock(&dentry
->d_lock
);
2428 fsnotify_nameremove(dentry
, isdir
);
2430 EXPORT_SYMBOL(d_delete
);
2432 static void __d_rehash(struct dentry
*entry
)
2434 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2437 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2442 * d_rehash - add an entry back to the hash
2443 * @entry: dentry to add to the hash
2445 * Adds a dentry to the hash according to its name.
2448 void d_rehash(struct dentry
* entry
)
2450 spin_lock(&entry
->d_lock
);
2452 spin_unlock(&entry
->d_lock
);
2454 EXPORT_SYMBOL(d_rehash
);
2456 static inline unsigned start_dir_add(struct inode
*dir
)
2460 unsigned n
= dir
->i_dir_seq
;
2461 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2467 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2469 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2472 static void d_wait_lookup(struct dentry
*dentry
)
2474 if (d_in_lookup(dentry
)) {
2475 DECLARE_WAITQUEUE(wait
, current
);
2476 add_wait_queue(dentry
->d_wait
, &wait
);
2478 set_current_state(TASK_UNINTERRUPTIBLE
);
2479 spin_unlock(&dentry
->d_lock
);
2481 spin_lock(&dentry
->d_lock
);
2482 } while (d_in_lookup(dentry
));
2486 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2487 const struct qstr
*name
,
2488 wait_queue_head_t
*wq
)
2490 unsigned int hash
= name
->hash
;
2491 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2492 struct hlist_bl_node
*node
;
2493 struct dentry
*new = d_alloc(parent
, name
);
2494 struct dentry
*dentry
;
2495 unsigned seq
, r_seq
, d_seq
;
2498 return ERR_PTR(-ENOMEM
);
2502 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
);
2503 r_seq
= read_seqbegin(&rename_lock
);
2504 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2505 if (unlikely(dentry
)) {
2506 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2510 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2519 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2524 if (unlikely(seq
& 1)) {
2530 if (unlikely(READ_ONCE(parent
->d_inode
->i_dir_seq
) != seq
)) {
2536 * No changes for the parent since the beginning of d_lookup().
2537 * Since all removals from the chain happen with hlist_bl_lock(),
2538 * any potential in-lookup matches are going to stay here until
2539 * we unlock the chain. All fields are stable in everything
2542 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2543 if (dentry
->d_name
.hash
!= hash
)
2545 if (dentry
->d_parent
!= parent
)
2547 if (!d_same_name(dentry
, parent
, name
))
2550 /* now we can try to grab a reference */
2551 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2558 * somebody is likely to be still doing lookup for it;
2559 * wait for them to finish
2561 spin_lock(&dentry
->d_lock
);
2562 d_wait_lookup(dentry
);
2564 * it's not in-lookup anymore; in principle we should repeat
2565 * everything from dcache lookup, but it's likely to be what
2566 * d_lookup() would've found anyway. If it is, just return it;
2567 * otherwise we really have to repeat the whole thing.
2569 if (unlikely(dentry
->d_name
.hash
!= hash
))
2571 if (unlikely(dentry
->d_parent
!= parent
))
2573 if (unlikely(d_unhashed(dentry
)))
2575 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2577 /* OK, it *is* a hashed match; return it */
2578 spin_unlock(&dentry
->d_lock
);
2583 /* we can't take ->d_lock here; it's OK, though. */
2584 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2586 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2590 spin_unlock(&dentry
->d_lock
);
2594 EXPORT_SYMBOL(d_alloc_parallel
);
2596 void __d_lookup_done(struct dentry
*dentry
)
2598 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2599 dentry
->d_name
.hash
);
2601 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2602 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2603 wake_up_all(dentry
->d_wait
);
2604 dentry
->d_wait
= NULL
;
2606 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2607 INIT_LIST_HEAD(&dentry
->d_lru
);
2609 EXPORT_SYMBOL(__d_lookup_done
);
2611 /* inode->i_lock held if inode is non-NULL */
2613 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2615 struct inode
*dir
= NULL
;
2617 spin_lock(&dentry
->d_lock
);
2618 if (unlikely(d_in_lookup(dentry
))) {
2619 dir
= dentry
->d_parent
->d_inode
;
2620 n
= start_dir_add(dir
);
2621 __d_lookup_done(dentry
);
2624 unsigned add_flags
= d_flags_for_inode(inode
);
2625 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2626 raw_write_seqcount_begin(&dentry
->d_seq
);
2627 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2628 raw_write_seqcount_end(&dentry
->d_seq
);
2629 fsnotify_update_flags(dentry
);
2633 end_dir_add(dir
, n
);
2634 spin_unlock(&dentry
->d_lock
);
2636 spin_unlock(&inode
->i_lock
);
2640 * d_add - add dentry to hash queues
2641 * @entry: dentry to add
2642 * @inode: The inode to attach to this dentry
2644 * This adds the entry to the hash queues and initializes @inode.
2645 * The entry was actually filled in earlier during d_alloc().
2648 void d_add(struct dentry
*entry
, struct inode
*inode
)
2651 security_d_instantiate(entry
, inode
);
2652 spin_lock(&inode
->i_lock
);
2654 __d_add(entry
, inode
);
2656 EXPORT_SYMBOL(d_add
);
2659 * d_exact_alias - find and hash an exact unhashed alias
2660 * @entry: dentry to add
2661 * @inode: The inode to go with this dentry
2663 * If an unhashed dentry with the same name/parent and desired
2664 * inode already exists, hash and return it. Otherwise, return
2667 * Parent directory should be locked.
2669 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2671 struct dentry
*alias
;
2672 unsigned int hash
= entry
->d_name
.hash
;
2674 spin_lock(&inode
->i_lock
);
2675 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2677 * Don't need alias->d_lock here, because aliases with
2678 * d_parent == entry->d_parent are not subject to name or
2679 * parent changes, because the parent inode i_mutex is held.
2681 if (alias
->d_name
.hash
!= hash
)
2683 if (alias
->d_parent
!= entry
->d_parent
)
2685 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2687 spin_lock(&alias
->d_lock
);
2688 if (!d_unhashed(alias
)) {
2689 spin_unlock(&alias
->d_lock
);
2692 __dget_dlock(alias
);
2694 spin_unlock(&alias
->d_lock
);
2696 spin_unlock(&inode
->i_lock
);
2699 spin_unlock(&inode
->i_lock
);
2702 EXPORT_SYMBOL(d_exact_alias
);
2705 * dentry_update_name_case - update case insensitive dentry with a new name
2706 * @dentry: dentry to be updated
2709 * Update a case insensitive dentry with new case of name.
2711 * dentry must have been returned by d_lookup with name @name. Old and new
2712 * name lengths must match (ie. no d_compare which allows mismatched name
2715 * Parent inode i_mutex must be held over d_lookup and into this call (to
2716 * keep renames and concurrent inserts, and readdir(2) away).
2718 void dentry_update_name_case(struct dentry
*dentry
, const struct qstr
*name
)
2720 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2721 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2723 spin_lock(&dentry
->d_lock
);
2724 write_seqcount_begin(&dentry
->d_seq
);
2725 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2726 write_seqcount_end(&dentry
->d_seq
);
2727 spin_unlock(&dentry
->d_lock
);
2729 EXPORT_SYMBOL(dentry_update_name_case
);
2731 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2733 if (unlikely(dname_external(target
))) {
2734 if (unlikely(dname_external(dentry
))) {
2736 * Both external: swap the pointers
2738 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2741 * dentry:internal, target:external. Steal target's
2742 * storage and make target internal.
2744 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2745 dentry
->d_name
.len
+ 1);
2746 dentry
->d_name
.name
= target
->d_name
.name
;
2747 target
->d_name
.name
= target
->d_iname
;
2750 if (unlikely(dname_external(dentry
))) {
2752 * dentry:external, target:internal. Give dentry's
2753 * storage to target and make dentry internal
2755 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2756 target
->d_name
.len
+ 1);
2757 target
->d_name
.name
= dentry
->d_name
.name
;
2758 dentry
->d_name
.name
= dentry
->d_iname
;
2761 * Both are internal.
2764 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2765 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2766 swap(((long *) &dentry
->d_iname
)[i
],
2767 ((long *) &target
->d_iname
)[i
]);
2771 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2774 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2776 struct external_name
*old_name
= NULL
;
2777 if (unlikely(dname_external(dentry
)))
2778 old_name
= external_name(dentry
);
2779 if (unlikely(dname_external(target
))) {
2780 atomic_inc(&external_name(target
)->u
.count
);
2781 dentry
->d_name
= target
->d_name
;
2783 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2784 target
->d_name
.len
+ 1);
2785 dentry
->d_name
.name
= dentry
->d_iname
;
2786 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2788 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2789 call_rcu(&old_name
->u
.head
, __d_free_external_name
);
2792 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2795 * XXXX: do we really need to take target->d_lock?
2797 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2798 spin_lock(&target
->d_parent
->d_lock
);
2800 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2801 spin_lock(&dentry
->d_parent
->d_lock
);
2802 spin_lock_nested(&target
->d_parent
->d_lock
,
2803 DENTRY_D_LOCK_NESTED
);
2805 spin_lock(&target
->d_parent
->d_lock
);
2806 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2807 DENTRY_D_LOCK_NESTED
);
2810 if (target
< dentry
) {
2811 spin_lock_nested(&target
->d_lock
, 2);
2812 spin_lock_nested(&dentry
->d_lock
, 3);
2814 spin_lock_nested(&dentry
->d_lock
, 2);
2815 spin_lock_nested(&target
->d_lock
, 3);
2819 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2821 if (target
->d_parent
!= dentry
->d_parent
)
2822 spin_unlock(&dentry
->d_parent
->d_lock
);
2823 if (target
->d_parent
!= target
)
2824 spin_unlock(&target
->d_parent
->d_lock
);
2825 spin_unlock(&target
->d_lock
);
2826 spin_unlock(&dentry
->d_lock
);
2830 * When switching names, the actual string doesn't strictly have to
2831 * be preserved in the target - because we're dropping the target
2832 * anyway. As such, we can just do a simple memcpy() to copy over
2833 * the new name before we switch, unless we are going to rehash
2834 * it. Note that if we *do* unhash the target, we are not allowed
2835 * to rehash it without giving it a new name/hash key - whether
2836 * we swap or overwrite the names here, resulting name won't match
2837 * the reality in filesystem; it's only there for d_path() purposes.
2838 * Note that all of this is happening under rename_lock, so the
2839 * any hash lookup seeing it in the middle of manipulations will
2840 * be discarded anyway. So we do not care what happens to the hash
2844 * __d_move - move a dentry
2845 * @dentry: entry to move
2846 * @target: new dentry
2847 * @exchange: exchange the two dentries
2849 * Update the dcache to reflect the move of a file name. Negative
2850 * dcache entries should not be moved in this way. Caller must hold
2851 * rename_lock, the i_mutex of the source and target directories,
2852 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2854 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2857 struct inode
*dir
= NULL
;
2859 if (!dentry
->d_inode
)
2860 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2862 BUG_ON(d_ancestor(dentry
, target
));
2863 BUG_ON(d_ancestor(target
, dentry
));
2865 dentry_lock_for_move(dentry
, target
);
2866 if (unlikely(d_in_lookup(target
))) {
2867 dir
= target
->d_parent
->d_inode
;
2868 n
= start_dir_add(dir
);
2869 __d_lookup_done(target
);
2872 write_seqcount_begin(&dentry
->d_seq
);
2873 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2876 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2880 /* Switch the names.. */
2882 swap_names(dentry
, target
);
2884 copy_name(dentry
, target
);
2886 /* rehash in new place(s) */
2891 target
->d_hash
.pprev
= NULL
;
2893 /* ... and switch them in the tree */
2894 if (IS_ROOT(dentry
)) {
2895 /* splicing a tree */
2896 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2897 dentry
->d_parent
= target
->d_parent
;
2898 target
->d_parent
= target
;
2899 list_del_init(&target
->d_child
);
2900 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2902 /* swapping two dentries */
2903 swap(dentry
->d_parent
, target
->d_parent
);
2904 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2905 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2907 fsnotify_update_flags(target
);
2908 fsnotify_update_flags(dentry
);
2911 write_seqcount_end(&target
->d_seq
);
2912 write_seqcount_end(&dentry
->d_seq
);
2915 end_dir_add(dir
, n
);
2916 dentry_unlock_for_move(dentry
, target
);
2920 * d_move - move a dentry
2921 * @dentry: entry to move
2922 * @target: new dentry
2924 * Update the dcache to reflect the move of a file name. Negative
2925 * dcache entries should not be moved in this way. See the locking
2926 * requirements for __d_move.
2928 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2930 write_seqlock(&rename_lock
);
2931 __d_move(dentry
, target
, false);
2932 write_sequnlock(&rename_lock
);
2934 EXPORT_SYMBOL(d_move
);
2937 * d_exchange - exchange two dentries
2938 * @dentry1: first dentry
2939 * @dentry2: second dentry
2941 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2943 write_seqlock(&rename_lock
);
2945 WARN_ON(!dentry1
->d_inode
);
2946 WARN_ON(!dentry2
->d_inode
);
2947 WARN_ON(IS_ROOT(dentry1
));
2948 WARN_ON(IS_ROOT(dentry2
));
2950 __d_move(dentry1
, dentry2
, true);
2952 write_sequnlock(&rename_lock
);
2954 EXPORT_SYMBOL_GPL(d_exchange
);
2957 * d_ancestor - search for an ancestor
2958 * @p1: ancestor dentry
2961 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2962 * an ancestor of p2, else NULL.
2964 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2968 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2969 if (p
->d_parent
== p1
)
2976 * This helper attempts to cope with remotely renamed directories
2978 * It assumes that the caller is already holding
2979 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2981 * Note: If ever the locking in lock_rename() changes, then please
2982 * remember to update this too...
2984 static int __d_unalias(struct inode
*inode
,
2985 struct dentry
*dentry
, struct dentry
*alias
)
2987 struct mutex
*m1
= NULL
;
2988 struct rw_semaphore
*m2
= NULL
;
2991 /* If alias and dentry share a parent, then no extra locks required */
2992 if (alias
->d_parent
== dentry
->d_parent
)
2995 /* See lock_rename() */
2996 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2998 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2999 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
3001 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
3003 __d_move(alias
, dentry
, false);
3014 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3015 * @inode: the inode which may have a disconnected dentry
3016 * @dentry: a negative dentry which we want to point to the inode.
3018 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3019 * place of the given dentry and return it, else simply d_add the inode
3020 * to the dentry and return NULL.
3022 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3023 * we should error out: directories can't have multiple aliases.
3025 * This is needed in the lookup routine of any filesystem that is exportable
3026 * (via knfsd) so that we can build dcache paths to directories effectively.
3028 * If a dentry was found and moved, then it is returned. Otherwise NULL
3029 * is returned. This matches the expected return value of ->lookup.
3031 * Cluster filesystems may call this function with a negative, hashed dentry.
3032 * In that case, we know that the inode will be a regular file, and also this
3033 * will only occur during atomic_open. So we need to check for the dentry
3034 * being already hashed only in the final case.
3036 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
3039 return ERR_CAST(inode
);
3041 BUG_ON(!d_unhashed(dentry
));
3046 security_d_instantiate(dentry
, inode
);
3047 spin_lock(&inode
->i_lock
);
3048 if (S_ISDIR(inode
->i_mode
)) {
3049 struct dentry
*new = __d_find_any_alias(inode
);
3050 if (unlikely(new)) {
3051 /* The reference to new ensures it remains an alias */
3052 spin_unlock(&inode
->i_lock
);
3053 write_seqlock(&rename_lock
);
3054 if (unlikely(d_ancestor(new, dentry
))) {
3055 write_sequnlock(&rename_lock
);
3057 new = ERR_PTR(-ELOOP
);
3058 pr_warn_ratelimited(
3059 "VFS: Lookup of '%s' in %s %s"
3060 " would have caused loop\n",
3061 dentry
->d_name
.name
,
3062 inode
->i_sb
->s_type
->name
,
3064 } else if (!IS_ROOT(new)) {
3065 int err
= __d_unalias(inode
, dentry
, new);
3066 write_sequnlock(&rename_lock
);
3072 __d_move(new, dentry
, false);
3073 write_sequnlock(&rename_lock
);
3080 __d_add(dentry
, inode
);
3083 EXPORT_SYMBOL(d_splice_alias
);
3085 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3089 return -ENAMETOOLONG
;
3091 memcpy(*buffer
, str
, namelen
);
3096 * prepend_name - prepend a pathname in front of current buffer pointer
3097 * @buffer: buffer pointer
3098 * @buflen: allocated length of the buffer
3099 * @name: name string and length qstr structure
3101 * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
3102 * make sure that either the old or the new name pointer and length are
3103 * fetched. However, there may be mismatch between length and pointer.
3104 * The length cannot be trusted, we need to copy it byte-by-byte until
3105 * the length is reached or a null byte is found. It also prepends "/" at
3106 * the beginning of the name. The sequence number check at the caller will
3107 * retry it again when a d_move() does happen. So any garbage in the buffer
3108 * due to mismatched pointer and length will be discarded.
3110 * Data dependency barrier is needed to make sure that we see that terminating
3111 * NUL. Alpha strikes again, film at 11...
3113 static int prepend_name(char **buffer
, int *buflen
, const struct qstr
*name
)
3115 const char *dname
= READ_ONCE(name
->name
);
3116 u32 dlen
= READ_ONCE(name
->len
);
3119 smp_read_barrier_depends();
3121 *buflen
-= dlen
+ 1;
3123 return -ENAMETOOLONG
;
3124 p
= *buffer
-= dlen
+ 1;
3136 * prepend_path - Prepend path string to a buffer
3137 * @path: the dentry/vfsmount to report
3138 * @root: root vfsmnt/dentry
3139 * @buffer: pointer to the end of the buffer
3140 * @buflen: pointer to buffer length
3142 * The function will first try to write out the pathname without taking any
3143 * lock other than the RCU read lock to make sure that dentries won't go away.
3144 * It only checks the sequence number of the global rename_lock as any change
3145 * in the dentry's d_seq will be preceded by changes in the rename_lock
3146 * sequence number. If the sequence number had been changed, it will restart
3147 * the whole pathname back-tracing sequence again by taking the rename_lock.
3148 * In this case, there is no need to take the RCU read lock as the recursive
3149 * parent pointer references will keep the dentry chain alive as long as no
3150 * rename operation is performed.
3152 static int prepend_path(const struct path
*path
,
3153 const struct path
*root
,
3154 char **buffer
, int *buflen
)
3156 struct dentry
*dentry
;
3157 struct vfsmount
*vfsmnt
;
3160 unsigned seq
, m_seq
= 0;
3166 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3173 dentry
= path
->dentry
;
3175 mnt
= real_mount(vfsmnt
);
3176 read_seqbegin_or_lock(&rename_lock
, &seq
);
3177 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3178 struct dentry
* parent
;
3180 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3181 struct mount
*parent
= READ_ONCE(mnt
->mnt_parent
);
3183 if (dentry
!= vfsmnt
->mnt_root
) {
3190 if (mnt
!= parent
) {
3191 dentry
= READ_ONCE(mnt
->mnt_mountpoint
);
3197 error
= is_mounted(vfsmnt
) ? 1 : 2;
3200 parent
= dentry
->d_parent
;
3202 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3210 if (need_seqretry(&rename_lock
, seq
)) {
3214 done_seqretry(&rename_lock
, seq
);
3218 if (need_seqretry(&mount_lock
, m_seq
)) {
3222 done_seqretry(&mount_lock
, m_seq
);
3224 if (error
>= 0 && bptr
== *buffer
) {
3226 error
= -ENAMETOOLONG
;
3236 * __d_path - return the path of a dentry
3237 * @path: the dentry/vfsmount to report
3238 * @root: root vfsmnt/dentry
3239 * @buf: buffer to return value in
3240 * @buflen: buffer length
3242 * Convert a dentry into an ASCII path name.
3244 * Returns a pointer into the buffer or an error code if the
3245 * path was too long.
3247 * "buflen" should be positive.
3249 * If the path is not reachable from the supplied root, return %NULL.
3251 char *__d_path(const struct path
*path
,
3252 const struct path
*root
,
3253 char *buf
, int buflen
)
3255 char *res
= buf
+ buflen
;
3258 prepend(&res
, &buflen
, "\0", 1);
3259 error
= prepend_path(path
, root
, &res
, &buflen
);
3262 return ERR_PTR(error
);
3268 char *d_absolute_path(const struct path
*path
,
3269 char *buf
, int buflen
)
3271 struct path root
= {};
3272 char *res
= buf
+ buflen
;
3275 prepend(&res
, &buflen
, "\0", 1);
3276 error
= prepend_path(path
, &root
, &res
, &buflen
);
3281 return ERR_PTR(error
);
3286 * same as __d_path but appends "(deleted)" for unlinked files.
3288 static int path_with_deleted(const struct path
*path
,
3289 const struct path
*root
,
3290 char **buf
, int *buflen
)
3292 prepend(buf
, buflen
, "\0", 1);
3293 if (d_unlinked(path
->dentry
)) {
3294 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3299 return prepend_path(path
, root
, buf
, buflen
);
3302 static int prepend_unreachable(char **buffer
, int *buflen
)
3304 return prepend(buffer
, buflen
, "(unreachable)", 13);
3307 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3312 seq
= read_seqcount_begin(&fs
->seq
);
3314 } while (read_seqcount_retry(&fs
->seq
, seq
));
3318 * d_path - return the path of a dentry
3319 * @path: path to report
3320 * @buf: buffer to return value in
3321 * @buflen: buffer length
3323 * Convert a dentry into an ASCII path name. If the entry has been deleted
3324 * the string " (deleted)" is appended. Note that this is ambiguous.
3326 * Returns a pointer into the buffer or an error code if the path was
3327 * too long. Note: Callers should use the returned pointer, not the passed
3328 * in buffer, to use the name! The implementation often starts at an offset
3329 * into the buffer, and may leave 0 bytes at the start.
3331 * "buflen" should be positive.
3333 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3335 char *res
= buf
+ buflen
;
3340 * We have various synthetic filesystems that never get mounted. On
3341 * these filesystems dentries are never used for lookup purposes, and
3342 * thus don't need to be hashed. They also don't need a name until a
3343 * user wants to identify the object in /proc/pid/fd/. The little hack
3344 * below allows us to generate a name for these objects on demand:
3346 * Some pseudo inodes are mountable. When they are mounted
3347 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3348 * and instead have d_path return the mounted path.
3350 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3351 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3352 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3355 get_fs_root_rcu(current
->fs
, &root
);
3356 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3360 res
= ERR_PTR(error
);
3363 EXPORT_SYMBOL(d_path
);
3366 * Helper function for dentry_operations.d_dname() members
3368 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3369 const char *fmt
, ...)
3375 va_start(args
, fmt
);
3376 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3379 if (sz
> sizeof(temp
) || sz
> buflen
)
3380 return ERR_PTR(-ENAMETOOLONG
);
3382 buffer
+= buflen
- sz
;
3383 return memcpy(buffer
, temp
, sz
);
3386 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3388 char *end
= buffer
+ buflen
;
3389 /* these dentries are never renamed, so d_lock is not needed */
3390 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3391 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3392 prepend(&end
, &buflen
, "/", 1))
3393 end
= ERR_PTR(-ENAMETOOLONG
);
3396 EXPORT_SYMBOL(simple_dname
);
3399 * Write full pathname from the root of the filesystem into the buffer.
3401 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3403 struct dentry
*dentry
;
3416 prepend(&end
, &len
, "\0", 1);
3420 read_seqbegin_or_lock(&rename_lock
, &seq
);
3421 while (!IS_ROOT(dentry
)) {
3422 struct dentry
*parent
= dentry
->d_parent
;
3425 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3434 if (need_seqretry(&rename_lock
, seq
)) {
3438 done_seqretry(&rename_lock
, seq
);
3443 return ERR_PTR(-ENAMETOOLONG
);
3446 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3448 return __dentry_path(dentry
, buf
, buflen
);
3450 EXPORT_SYMBOL(dentry_path_raw
);
3452 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3457 if (d_unlinked(dentry
)) {
3459 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3463 retval
= __dentry_path(dentry
, buf
, buflen
);
3464 if (!IS_ERR(retval
) && p
)
3465 *p
= '/'; /* restore '/' overriden with '\0' */
3468 return ERR_PTR(-ENAMETOOLONG
);
3471 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3477 seq
= read_seqcount_begin(&fs
->seq
);
3480 } while (read_seqcount_retry(&fs
->seq
, seq
));
3484 * NOTE! The user-level library version returns a
3485 * character pointer. The kernel system call just
3486 * returns the length of the buffer filled (which
3487 * includes the ending '\0' character), or a negative
3488 * error value. So libc would do something like
3490 * char *getcwd(char * buf, size_t size)
3494 * retval = sys_getcwd(buf, size);
3501 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3504 struct path pwd
, root
;
3505 char *page
= __getname();
3511 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3514 if (!d_unlinked(pwd
.dentry
)) {
3516 char *cwd
= page
+ PATH_MAX
;
3517 int buflen
= PATH_MAX
;
3519 prepend(&cwd
, &buflen
, "\0", 1);
3520 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3526 /* Unreachable from current root */
3528 error
= prepend_unreachable(&cwd
, &buflen
);
3534 len
= PATH_MAX
+ page
- cwd
;
3537 if (copy_to_user(buf
, cwd
, len
))
3550 * Test whether new_dentry is a subdirectory of old_dentry.
3552 * Trivially implemented using the dcache structure
3556 * is_subdir - is new dentry a subdirectory of old_dentry
3557 * @new_dentry: new dentry
3558 * @old_dentry: old dentry
3560 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3561 * Returns false otherwise.
3562 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3565 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3570 if (new_dentry
== old_dentry
)
3574 /* for restarting inner loop in case of seq retry */
3575 seq
= read_seqbegin(&rename_lock
);
3577 * Need rcu_readlock to protect against the d_parent trashing
3581 if (d_ancestor(old_dentry
, new_dentry
))
3586 } while (read_seqretry(&rename_lock
, seq
));
3591 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3593 struct dentry
*root
= data
;
3594 if (dentry
!= root
) {
3595 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3598 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3599 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3600 dentry
->d_lockref
.count
--;
3603 return D_WALK_CONTINUE
;
3606 void d_genocide(struct dentry
*parent
)
3608 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3611 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3613 inode_dec_link_count(inode
);
3614 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3615 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3616 !d_unlinked(dentry
));
3617 spin_lock(&dentry
->d_parent
->d_lock
);
3618 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3619 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3620 (unsigned long long)inode
->i_ino
);
3621 spin_unlock(&dentry
->d_lock
);
3622 spin_unlock(&dentry
->d_parent
->d_lock
);
3623 d_instantiate(dentry
, inode
);
3625 EXPORT_SYMBOL(d_tmpfile
);
3627 static __initdata
unsigned long dhash_entries
;
3628 static int __init
set_dhash_entries(char *str
)
3632 dhash_entries
= simple_strtoul(str
, &str
, 0);
3635 __setup("dhash_entries=", set_dhash_entries
);
3637 static void __init
dcache_init_early(void)
3639 /* If hashes are distributed across NUMA nodes, defer
3640 * hash allocation until vmalloc space is available.
3646 alloc_large_system_hash("Dentry cache",
3647 sizeof(struct hlist_bl_head
),
3650 HASH_EARLY
| HASH_ZERO
,
3657 static void __init
dcache_init(void)
3660 * A constructor could be added for stable state like the lists,
3661 * but it is probably not worth it because of the cache nature
3664 dentry_cache
= KMEM_CACHE(dentry
,
3665 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3667 /* Hash may have been set up in dcache_init_early */
3672 alloc_large_system_hash("Dentry cache",
3673 sizeof(struct hlist_bl_head
),
3683 /* SLAB cache for __getname() consumers */
3684 struct kmem_cache
*names_cachep __read_mostly
;
3685 EXPORT_SYMBOL(names_cachep
);
3687 EXPORT_SYMBOL(d_genocide
);
3689 void __init
vfs_caches_init_early(void)
3693 for (i
= 0; i
< ARRAY_SIZE(in_lookup_hashtable
); i
++)
3694 INIT_HLIST_BL_HEAD(&in_lookup_hashtable
[i
]);
3696 dcache_init_early();
3700 void __init
vfs_caches_init(void)
3702 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3703 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3708 files_maxfiles_init();