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
;
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
102 static unsigned int d_hash_mask __read_mostly
;
103 static unsigned int d_hash_shift __read_mostly
;
105 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
107 static inline struct hlist_bl_head
*d_hash(unsigned int hash
)
109 return dentry_hashtable
+ (hash
>> (32 - d_hash_shift
));
112 #define IN_LOOKUP_SHIFT 10
113 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
115 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
118 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
119 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
123 /* Statistics gathering. */
124 struct dentry_stat_t dentry_stat
= {
128 static DEFINE_PER_CPU(long, nr_dentry
);
129 static DEFINE_PER_CPU(long, nr_dentry_unused
);
131 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
134 * Here we resort to our own counters instead of using generic per-cpu counters
135 * for consistency with what the vfs inode code does. We are expected to harvest
136 * better code and performance by having our own specialized counters.
138 * Please note that the loop is done over all possible CPUs, not over all online
139 * CPUs. The reason for this is that we don't want to play games with CPUs going
140 * on and off. If one of them goes off, we will just keep their counters.
142 * glommer: See cffbc8a for details, and if you ever intend to change this,
143 * please update all vfs counters to match.
145 static long get_nr_dentry(void)
149 for_each_possible_cpu(i
)
150 sum
+= per_cpu(nr_dentry
, i
);
151 return sum
< 0 ? 0 : sum
;
154 static long get_nr_dentry_unused(void)
158 for_each_possible_cpu(i
)
159 sum
+= per_cpu(nr_dentry_unused
, i
);
160 return sum
< 0 ? 0 : sum
;
163 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
164 size_t *lenp
, loff_t
*ppos
)
166 dentry_stat
.nr_dentry
= get_nr_dentry();
167 dentry_stat
.nr_unused
= get_nr_dentry_unused();
168 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
173 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
174 * The strings are both count bytes long, and count is non-zero.
176 #ifdef CONFIG_DCACHE_WORD_ACCESS
178 #include <asm/word-at-a-time.h>
180 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
181 * aligned allocation for this particular component. We don't
182 * strictly need the load_unaligned_zeropad() safety, but it
183 * doesn't hurt either.
185 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
186 * need the careful unaligned handling.
188 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
190 unsigned long a
,b
,mask
;
193 a
= *(unsigned long *)cs
;
194 b
= load_unaligned_zeropad(ct
);
195 if (tcount
< sizeof(unsigned long))
197 if (unlikely(a
!= b
))
199 cs
+= sizeof(unsigned long);
200 ct
+= sizeof(unsigned long);
201 tcount
-= sizeof(unsigned long);
205 mask
= bytemask_from_count(tcount
);
206 return unlikely(!!((a
^ b
) & mask
));
211 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
225 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
228 * Be careful about RCU walk racing with rename:
229 * use 'lockless_dereference' to fetch the name pointer.
231 * NOTE! Even if a rename will mean that the length
232 * was not loaded atomically, we don't care. The
233 * RCU walk will check the sequence count eventually,
234 * and catch it. And we won't overrun the buffer,
235 * because we're reading the name pointer atomically,
236 * and a dentry name is guaranteed to be properly
237 * terminated with a NUL byte.
239 * End result: even if 'len' is wrong, we'll exit
240 * early because the data cannot match (there can
241 * be no NUL in the ct/tcount data)
243 const unsigned char *cs
= lockless_dereference(dentry
->d_name
.name
);
245 return dentry_string_cmp(cs
, ct
, tcount
);
248 struct external_name
{
251 struct rcu_head head
;
253 unsigned char name
[];
256 static inline struct external_name
*external_name(struct dentry
*dentry
)
258 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
261 static void __d_free(struct rcu_head
*head
)
263 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
265 kmem_cache_free(dentry_cache
, dentry
);
268 static void __d_free_external(struct rcu_head
*head
)
270 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
271 kfree(external_name(dentry
));
272 kmem_cache_free(dentry_cache
, dentry
);
275 static inline int dname_external(const struct dentry
*dentry
)
277 return dentry
->d_name
.name
!= dentry
->d_iname
;
280 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
286 dentry
->d_inode
= inode
;
287 flags
= READ_ONCE(dentry
->d_flags
);
288 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
290 WRITE_ONCE(dentry
->d_flags
, flags
);
293 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
295 unsigned flags
= READ_ONCE(dentry
->d_flags
);
297 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
298 WRITE_ONCE(dentry
->d_flags
, flags
);
299 dentry
->d_inode
= NULL
;
302 static void dentry_free(struct dentry
*dentry
)
304 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
305 if (unlikely(dname_external(dentry
))) {
306 struct external_name
*p
= external_name(dentry
);
307 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
308 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
312 /* if dentry was never visible to RCU, immediate free is OK */
313 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
314 __d_free(&dentry
->d_u
.d_rcu
);
316 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
320 * Release the dentry's inode, using the filesystem
321 * d_iput() operation if defined.
323 static void dentry_unlink_inode(struct dentry
* dentry
)
324 __releases(dentry
->d_lock
)
325 __releases(dentry
->d_inode
->i_lock
)
327 struct inode
*inode
= dentry
->d_inode
;
328 bool hashed
= !d_unhashed(dentry
);
331 raw_write_seqcount_begin(&dentry
->d_seq
);
332 __d_clear_type_and_inode(dentry
);
333 hlist_del_init(&dentry
->d_u
.d_alias
);
335 raw_write_seqcount_end(&dentry
->d_seq
);
336 spin_unlock(&dentry
->d_lock
);
337 spin_unlock(&inode
->i_lock
);
339 fsnotify_inoderemove(inode
);
340 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
341 dentry
->d_op
->d_iput(dentry
, inode
);
347 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
348 * is in use - which includes both the "real" per-superblock
349 * LRU list _and_ the DCACHE_SHRINK_LIST use.
351 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
352 * on the shrink list (ie not on the superblock LRU list).
354 * The per-cpu "nr_dentry_unused" counters are updated with
355 * the DCACHE_LRU_LIST bit.
357 * These helper functions make sure we always follow the
358 * rules. d_lock must be held by the caller.
360 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
361 static void d_lru_add(struct dentry
*dentry
)
363 D_FLAG_VERIFY(dentry
, 0);
364 dentry
->d_flags
|= DCACHE_LRU_LIST
;
365 this_cpu_inc(nr_dentry_unused
);
366 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
369 static void d_lru_del(struct dentry
*dentry
)
371 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
372 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
373 this_cpu_dec(nr_dentry_unused
);
374 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
377 static void d_shrink_del(struct dentry
*dentry
)
379 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
380 list_del_init(&dentry
->d_lru
);
381 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
382 this_cpu_dec(nr_dentry_unused
);
385 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
387 D_FLAG_VERIFY(dentry
, 0);
388 list_add(&dentry
->d_lru
, list
);
389 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
390 this_cpu_inc(nr_dentry_unused
);
394 * These can only be called under the global LRU lock, ie during the
395 * callback for freeing the LRU list. "isolate" removes it from the
396 * LRU lists entirely, while shrink_move moves it to the indicated
399 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
401 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
402 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
403 this_cpu_dec(nr_dentry_unused
);
404 list_lru_isolate(lru
, &dentry
->d_lru
);
407 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
408 struct list_head
*list
)
410 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
411 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
412 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
416 * dentry_lru_(add|del)_list) must be called with d_lock held.
418 static void dentry_lru_add(struct dentry
*dentry
)
420 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
425 * d_drop - drop a dentry
426 * @dentry: dentry to drop
428 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
429 * be found through a VFS lookup any more. Note that this is different from
430 * deleting the dentry - d_delete will try to mark the dentry negative if
431 * possible, giving a successful _negative_ lookup, while d_drop will
432 * just make the cache lookup fail.
434 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
435 * reason (NFS timeouts or autofs deletes).
437 * __d_drop requires dentry->d_lock.
439 void __d_drop(struct dentry
*dentry
)
441 if (!d_unhashed(dentry
)) {
442 struct hlist_bl_head
*b
;
444 * Hashed dentries are normally on the dentry hashtable,
445 * with the exception of those newly allocated by
446 * d_obtain_alias, which are always IS_ROOT:
448 if (unlikely(IS_ROOT(dentry
)))
449 b
= &dentry
->d_sb
->s_anon
;
451 b
= d_hash(dentry
->d_name
.hash
);
454 __hlist_bl_del(&dentry
->d_hash
);
455 dentry
->d_hash
.pprev
= NULL
;
457 /* After this call, in-progress rcu-walk path lookup will fail. */
458 write_seqcount_invalidate(&dentry
->d_seq
);
461 EXPORT_SYMBOL(__d_drop
);
463 void d_drop(struct dentry
*dentry
)
465 spin_lock(&dentry
->d_lock
);
467 spin_unlock(&dentry
->d_lock
);
469 EXPORT_SYMBOL(d_drop
);
471 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
475 * Inform d_walk() and shrink_dentry_list() that we are no longer
476 * attached to the dentry tree
478 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
479 if (unlikely(list_empty(&dentry
->d_child
)))
481 __list_del_entry(&dentry
->d_child
);
483 * Cursors can move around the list of children. While we'd been
484 * a normal list member, it didn't matter - ->d_child.next would've
485 * been updated. However, from now on it won't be and for the
486 * things like d_walk() it might end up with a nasty surprise.
487 * Normally d_walk() doesn't care about cursors moving around -
488 * ->d_lock on parent prevents that and since a cursor has no children
489 * of its own, we get through it without ever unlocking the parent.
490 * There is one exception, though - if we ascend from a child that
491 * gets killed as soon as we unlock it, the next sibling is found
492 * using the value left in its ->d_child.next. And if _that_
493 * pointed to a cursor, and cursor got moved (e.g. by lseek())
494 * before d_walk() regains parent->d_lock, we'll end up skipping
495 * everything the cursor had been moved past.
497 * Solution: make sure that the pointer left behind in ->d_child.next
498 * points to something that won't be moving around. I.e. skip the
501 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
502 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
503 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
505 dentry
->d_child
.next
= next
->d_child
.next
;
509 static void __dentry_kill(struct dentry
*dentry
)
511 struct dentry
*parent
= NULL
;
512 bool can_free
= true;
513 if (!IS_ROOT(dentry
))
514 parent
= dentry
->d_parent
;
517 * The dentry is now unrecoverably dead to the world.
519 lockref_mark_dead(&dentry
->d_lockref
);
522 * inform the fs via d_prune that this dentry is about to be
523 * unhashed and destroyed.
525 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
526 dentry
->d_op
->d_prune(dentry
);
528 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
529 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
532 /* if it was on the hash then remove it */
534 dentry_unlist(dentry
, parent
);
536 spin_unlock(&parent
->d_lock
);
538 dentry_unlink_inode(dentry
);
540 spin_unlock(&dentry
->d_lock
);
541 this_cpu_dec(nr_dentry
);
542 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
543 dentry
->d_op
->d_release(dentry
);
545 spin_lock(&dentry
->d_lock
);
546 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
547 dentry
->d_flags
|= DCACHE_MAY_FREE
;
550 spin_unlock(&dentry
->d_lock
);
551 if (likely(can_free
))
556 * Finish off a dentry we've decided to kill.
557 * dentry->d_lock must be held, returns with it unlocked.
558 * If ref is non-zero, then decrement the refcount too.
559 * Returns dentry requiring refcount drop, or NULL if we're done.
561 static struct dentry
*dentry_kill(struct dentry
*dentry
)
562 __releases(dentry
->d_lock
)
564 struct inode
*inode
= dentry
->d_inode
;
565 struct dentry
*parent
= NULL
;
567 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
570 if (!IS_ROOT(dentry
)) {
571 parent
= dentry
->d_parent
;
572 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
574 spin_unlock(&inode
->i_lock
);
579 __dentry_kill(dentry
);
583 spin_unlock(&dentry
->d_lock
);
584 return dentry
; /* try again with same dentry */
587 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
589 struct dentry
*parent
= dentry
->d_parent
;
592 if (unlikely(dentry
->d_lockref
.count
< 0))
594 if (likely(spin_trylock(&parent
->d_lock
)))
597 spin_unlock(&dentry
->d_lock
);
599 parent
= ACCESS_ONCE(dentry
->d_parent
);
600 spin_lock(&parent
->d_lock
);
602 * We can't blindly lock dentry until we are sure
603 * that we won't violate the locking order.
604 * Any changes of dentry->d_parent must have
605 * been done with parent->d_lock held, so
606 * spin_lock() above is enough of a barrier
607 * for checking if it's still our child.
609 if (unlikely(parent
!= dentry
->d_parent
)) {
610 spin_unlock(&parent
->d_lock
);
614 if (parent
!= dentry
)
615 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
622 * Try to do a lockless dput(), and return whether that was successful.
624 * If unsuccessful, we return false, having already taken the dentry lock.
626 * The caller needs to hold the RCU read lock, so that the dentry is
627 * guaranteed to stay around even if the refcount goes down to zero!
629 static inline bool fast_dput(struct dentry
*dentry
)
632 unsigned int d_flags
;
635 * If we have a d_op->d_delete() operation, we sould not
636 * let the dentry count go to zero, so use "put_or_lock".
638 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
639 return lockref_put_or_lock(&dentry
->d_lockref
);
642 * .. otherwise, we can try to just decrement the
643 * lockref optimistically.
645 ret
= lockref_put_return(&dentry
->d_lockref
);
648 * If the lockref_put_return() failed due to the lock being held
649 * by somebody else, the fast path has failed. We will need to
650 * get the lock, and then check the count again.
652 if (unlikely(ret
< 0)) {
653 spin_lock(&dentry
->d_lock
);
654 if (dentry
->d_lockref
.count
> 1) {
655 dentry
->d_lockref
.count
--;
656 spin_unlock(&dentry
->d_lock
);
663 * If we weren't the last ref, we're done.
669 * Careful, careful. The reference count went down
670 * to zero, but we don't hold the dentry lock, so
671 * somebody else could get it again, and do another
672 * dput(), and we need to not race with that.
674 * However, there is a very special and common case
675 * where we don't care, because there is nothing to
676 * do: the dentry is still hashed, it does not have
677 * a 'delete' op, and it's referenced and already on
680 * NOTE! Since we aren't locked, these values are
681 * not "stable". However, it is sufficient that at
682 * some point after we dropped the reference the
683 * dentry was hashed and the flags had the proper
684 * value. Other dentry users may have re-gotten
685 * a reference to the dentry and change that, but
686 * our work is done - we can leave the dentry
687 * around with a zero refcount.
690 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
691 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
693 /* Nothing to do? Dropping the reference was all we needed? */
694 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
698 * Not the fast normal case? Get the lock. We've already decremented
699 * the refcount, but we'll need to re-check the situation after
702 spin_lock(&dentry
->d_lock
);
705 * Did somebody else grab a reference to it in the meantime, and
706 * we're no longer the last user after all? Alternatively, somebody
707 * else could have killed it and marked it dead. Either way, we
708 * don't need to do anything else.
710 if (dentry
->d_lockref
.count
) {
711 spin_unlock(&dentry
->d_lock
);
716 * Re-get the reference we optimistically dropped. We hold the
717 * lock, and we just tested that it was zero, so we can just
720 dentry
->d_lockref
.count
= 1;
728 * This is complicated by the fact that we do not want to put
729 * dentries that are no longer on any hash chain on the unused
730 * list: we'd much rather just get rid of them immediately.
732 * However, that implies that we have to traverse the dentry
733 * tree upwards to the parents which might _also_ now be
734 * scheduled for deletion (it may have been only waiting for
735 * its last child to go away).
737 * This tail recursion is done by hand as we don't want to depend
738 * on the compiler to always get this right (gcc generally doesn't).
739 * Real recursion would eat up our stack space.
743 * dput - release a dentry
744 * @dentry: dentry to release
746 * Release a dentry. This will drop the usage count and if appropriate
747 * call the dentry unlink method as well as removing it from the queues and
748 * releasing its resources. If the parent dentries were scheduled for release
749 * they too may now get deleted.
751 void dput(struct dentry
*dentry
)
753 if (unlikely(!dentry
))
760 if (likely(fast_dput(dentry
))) {
765 /* Slow case: now with the dentry lock held */
768 WARN_ON(d_in_lookup(dentry
));
770 /* Unreachable? Get rid of it */
771 if (unlikely(d_unhashed(dentry
)))
774 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
777 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
778 if (dentry
->d_op
->d_delete(dentry
))
782 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
783 dentry
->d_flags
|= DCACHE_REFERENCED
;
784 dentry_lru_add(dentry
);
786 dentry
->d_lockref
.count
--;
787 spin_unlock(&dentry
->d_lock
);
791 dentry
= dentry_kill(dentry
);
800 /* This must be called with d_lock held */
801 static inline void __dget_dlock(struct dentry
*dentry
)
803 dentry
->d_lockref
.count
++;
806 static inline void __dget(struct dentry
*dentry
)
808 lockref_get(&dentry
->d_lockref
);
811 struct dentry
*dget_parent(struct dentry
*dentry
)
817 * Do optimistic parent lookup without any
821 ret
= ACCESS_ONCE(dentry
->d_parent
);
822 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
824 if (likely(gotref
)) {
825 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
832 * Don't need rcu_dereference because we re-check it was correct under
836 ret
= dentry
->d_parent
;
837 spin_lock(&ret
->d_lock
);
838 if (unlikely(ret
!= dentry
->d_parent
)) {
839 spin_unlock(&ret
->d_lock
);
844 BUG_ON(!ret
->d_lockref
.count
);
845 ret
->d_lockref
.count
++;
846 spin_unlock(&ret
->d_lock
);
849 EXPORT_SYMBOL(dget_parent
);
852 * d_find_alias - grab a hashed alias of inode
853 * @inode: inode in question
855 * If inode has a hashed alias, or is a directory and has any alias,
856 * acquire the reference to alias and return it. Otherwise return NULL.
857 * Notice that if inode is a directory there can be only one alias and
858 * it can be unhashed only if it has no children, or if it is the root
859 * of a filesystem, or if the directory was renamed and d_revalidate
860 * was the first vfs operation to notice.
862 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
863 * any other hashed alias over that one.
865 static struct dentry
*__d_find_alias(struct inode
*inode
)
867 struct dentry
*alias
, *discon_alias
;
871 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
872 spin_lock(&alias
->d_lock
);
873 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
874 if (IS_ROOT(alias
) &&
875 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
876 discon_alias
= alias
;
879 spin_unlock(&alias
->d_lock
);
883 spin_unlock(&alias
->d_lock
);
886 alias
= discon_alias
;
887 spin_lock(&alias
->d_lock
);
888 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
890 spin_unlock(&alias
->d_lock
);
893 spin_unlock(&alias
->d_lock
);
899 struct dentry
*d_find_alias(struct inode
*inode
)
901 struct dentry
*de
= NULL
;
903 if (!hlist_empty(&inode
->i_dentry
)) {
904 spin_lock(&inode
->i_lock
);
905 de
= __d_find_alias(inode
);
906 spin_unlock(&inode
->i_lock
);
910 EXPORT_SYMBOL(d_find_alias
);
913 * Try to kill dentries associated with this inode.
914 * WARNING: you must own a reference to inode.
916 void d_prune_aliases(struct inode
*inode
)
918 struct dentry
*dentry
;
920 spin_lock(&inode
->i_lock
);
921 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
922 spin_lock(&dentry
->d_lock
);
923 if (!dentry
->d_lockref
.count
) {
924 struct dentry
*parent
= lock_parent(dentry
);
925 if (likely(!dentry
->d_lockref
.count
)) {
926 __dentry_kill(dentry
);
931 spin_unlock(&parent
->d_lock
);
933 spin_unlock(&dentry
->d_lock
);
935 spin_unlock(&inode
->i_lock
);
937 EXPORT_SYMBOL(d_prune_aliases
);
939 static void shrink_dentry_list(struct list_head
*list
)
941 struct dentry
*dentry
, *parent
;
943 while (!list_empty(list
)) {
945 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
946 spin_lock(&dentry
->d_lock
);
947 parent
= lock_parent(dentry
);
950 * The dispose list is isolated and dentries are not accounted
951 * to the LRU here, so we can simply remove it from the list
952 * here regardless of whether it is referenced or not.
954 d_shrink_del(dentry
);
957 * We found an inuse dentry which was not removed from
958 * the LRU because of laziness during lookup. Do not free it.
960 if (dentry
->d_lockref
.count
> 0) {
961 spin_unlock(&dentry
->d_lock
);
963 spin_unlock(&parent
->d_lock
);
968 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
969 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
970 spin_unlock(&dentry
->d_lock
);
972 spin_unlock(&parent
->d_lock
);
978 inode
= dentry
->d_inode
;
979 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
980 d_shrink_add(dentry
, list
);
981 spin_unlock(&dentry
->d_lock
);
983 spin_unlock(&parent
->d_lock
);
987 __dentry_kill(dentry
);
990 * We need to prune ancestors too. This is necessary to prevent
991 * quadratic behavior of shrink_dcache_parent(), but is also
992 * expected to be beneficial in reducing dentry cache
996 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
997 parent
= lock_parent(dentry
);
998 if (dentry
->d_lockref
.count
!= 1) {
999 dentry
->d_lockref
.count
--;
1000 spin_unlock(&dentry
->d_lock
);
1002 spin_unlock(&parent
->d_lock
);
1005 inode
= dentry
->d_inode
; /* can't be NULL */
1006 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1007 spin_unlock(&dentry
->d_lock
);
1009 spin_unlock(&parent
->d_lock
);
1013 __dentry_kill(dentry
);
1019 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1020 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1022 struct list_head
*freeable
= arg
;
1023 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1027 * we are inverting the lru lock/dentry->d_lock here,
1028 * so use a trylock. If we fail to get the lock, just skip
1031 if (!spin_trylock(&dentry
->d_lock
))
1035 * Referenced dentries are still in use. If they have active
1036 * counts, just remove them from the LRU. Otherwise give them
1037 * another pass through the LRU.
1039 if (dentry
->d_lockref
.count
) {
1040 d_lru_isolate(lru
, dentry
);
1041 spin_unlock(&dentry
->d_lock
);
1045 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1046 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1047 spin_unlock(&dentry
->d_lock
);
1050 * The list move itself will be made by the common LRU code. At
1051 * this point, we've dropped the dentry->d_lock but keep the
1052 * lru lock. This is safe to do, since every list movement is
1053 * protected by the lru lock even if both locks are held.
1055 * This is guaranteed by the fact that all LRU management
1056 * functions are intermediated by the LRU API calls like
1057 * list_lru_add and list_lru_del. List movement in this file
1058 * only ever occur through this functions or through callbacks
1059 * like this one, that are called from the LRU API.
1061 * The only exceptions to this are functions like
1062 * shrink_dentry_list, and code that first checks for the
1063 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1064 * operating only with stack provided lists after they are
1065 * properly isolated from the main list. It is thus, always a
1071 d_lru_shrink_move(lru
, dentry
, freeable
);
1072 spin_unlock(&dentry
->d_lock
);
1078 * prune_dcache_sb - shrink the dcache
1080 * @sc: shrink control, passed to list_lru_shrink_walk()
1082 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1083 * is done when we need more memory and called from the superblock shrinker
1086 * This function may fail to free any resources if all the dentries are in
1089 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1094 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1095 dentry_lru_isolate
, &dispose
);
1096 shrink_dentry_list(&dispose
);
1100 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1101 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1103 struct list_head
*freeable
= arg
;
1104 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1107 * we are inverting the lru lock/dentry->d_lock here,
1108 * so use a trylock. If we fail to get the lock, just skip
1111 if (!spin_trylock(&dentry
->d_lock
))
1114 d_lru_shrink_move(lru
, dentry
, freeable
);
1115 spin_unlock(&dentry
->d_lock
);
1122 * shrink_dcache_sb - shrink dcache for a superblock
1125 * Shrink the dcache for the specified super block. This is used to free
1126 * the dcache before unmounting a file system.
1128 void shrink_dcache_sb(struct super_block
*sb
)
1135 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1136 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1138 this_cpu_sub(nr_dentry_unused
, freed
);
1139 shrink_dentry_list(&dispose
);
1140 } while (freed
> 0);
1142 EXPORT_SYMBOL(shrink_dcache_sb
);
1145 * enum d_walk_ret - action to talke during tree walk
1146 * @D_WALK_CONTINUE: contrinue walk
1147 * @D_WALK_QUIT: quit walk
1148 * @D_WALK_NORETRY: quit when retry is needed
1149 * @D_WALK_SKIP: skip this dentry and its children
1159 * d_walk - walk the dentry tree
1160 * @parent: start of walk
1161 * @data: data passed to @enter() and @finish()
1162 * @enter: callback when first entering the dentry
1163 * @finish: callback when successfully finished the walk
1165 * The @enter() and @finish() callbacks are called with d_lock held.
1167 void d_walk(struct dentry
*parent
, void *data
,
1168 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1169 void (*finish
)(void *))
1171 struct dentry
*this_parent
;
1172 struct list_head
*next
;
1174 enum d_walk_ret ret
;
1178 read_seqbegin_or_lock(&rename_lock
, &seq
);
1179 this_parent
= parent
;
1180 spin_lock(&this_parent
->d_lock
);
1182 ret
= enter(data
, this_parent
);
1184 case D_WALK_CONTINUE
:
1189 case D_WALK_NORETRY
:
1194 next
= this_parent
->d_subdirs
.next
;
1196 while (next
!= &this_parent
->d_subdirs
) {
1197 struct list_head
*tmp
= next
;
1198 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1201 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1204 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1206 ret
= enter(data
, dentry
);
1208 case D_WALK_CONTINUE
:
1211 spin_unlock(&dentry
->d_lock
);
1213 case D_WALK_NORETRY
:
1217 spin_unlock(&dentry
->d_lock
);
1221 if (!list_empty(&dentry
->d_subdirs
)) {
1222 spin_unlock(&this_parent
->d_lock
);
1223 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1224 this_parent
= dentry
;
1225 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1228 spin_unlock(&dentry
->d_lock
);
1231 * All done at this level ... ascend and resume the search.
1235 if (this_parent
!= parent
) {
1236 struct dentry
*child
= this_parent
;
1237 this_parent
= child
->d_parent
;
1239 spin_unlock(&child
->d_lock
);
1240 spin_lock(&this_parent
->d_lock
);
1242 /* might go back up the wrong parent if we have had a rename. */
1243 if (need_seqretry(&rename_lock
, seq
))
1245 /* go into the first sibling still alive */
1247 next
= child
->d_child
.next
;
1248 if (next
== &this_parent
->d_subdirs
)
1250 child
= list_entry(next
, struct dentry
, d_child
);
1251 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1255 if (need_seqretry(&rename_lock
, seq
))
1262 spin_unlock(&this_parent
->d_lock
);
1263 done_seqretry(&rename_lock
, seq
);
1267 spin_unlock(&this_parent
->d_lock
);
1275 EXPORT_SYMBOL_GPL(d_walk
);
1277 struct check_mount
{
1278 struct vfsmount
*mnt
;
1279 unsigned int mounted
;
1282 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1284 struct check_mount
*info
= data
;
1285 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1287 if (likely(!d_mountpoint(dentry
)))
1288 return D_WALK_CONTINUE
;
1289 if (__path_is_mountpoint(&path
)) {
1293 return D_WALK_CONTINUE
;
1297 * path_has_submounts - check for mounts over a dentry in the
1298 * current namespace.
1299 * @parent: path to check.
1301 * Return true if the parent or its subdirectories contain
1302 * a mount point in the current namespace.
1304 int path_has_submounts(const struct path
*parent
)
1306 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1308 read_seqlock_excl(&mount_lock
);
1309 d_walk(parent
->dentry
, &data
, path_check_mount
, NULL
);
1310 read_sequnlock_excl(&mount_lock
);
1312 return data
.mounted
;
1314 EXPORT_SYMBOL(path_has_submounts
);
1317 * Called by mount code to set a mountpoint and check if the mountpoint is
1318 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1319 * subtree can become unreachable).
1321 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1322 * this reason take rename_lock and d_lock on dentry and ancestors.
1324 int d_set_mounted(struct dentry
*dentry
)
1328 write_seqlock(&rename_lock
);
1329 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1330 /* Need exclusion wrt. d_invalidate() */
1331 spin_lock(&p
->d_lock
);
1332 if (unlikely(d_unhashed(p
))) {
1333 spin_unlock(&p
->d_lock
);
1336 spin_unlock(&p
->d_lock
);
1338 spin_lock(&dentry
->d_lock
);
1339 if (!d_unlinked(dentry
)) {
1341 if (!d_mountpoint(dentry
)) {
1342 dentry
->d_flags
|= DCACHE_MOUNTED
;
1346 spin_unlock(&dentry
->d_lock
);
1348 write_sequnlock(&rename_lock
);
1353 * Search the dentry child list of the specified parent,
1354 * and move any unused dentries to the end of the unused
1355 * list for prune_dcache(). We descend to the next level
1356 * whenever the d_subdirs list is non-empty and continue
1359 * It returns zero iff there are no unused children,
1360 * otherwise it returns the number of children moved to
1361 * the end of the unused list. This may not be the total
1362 * number of unused children, because select_parent can
1363 * drop the lock and return early due to latency
1367 struct select_data
{
1368 struct dentry
*start
;
1369 struct list_head dispose
;
1373 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1375 struct select_data
*data
= _data
;
1376 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1378 if (data
->start
== dentry
)
1381 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1384 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1386 if (!dentry
->d_lockref
.count
) {
1387 d_shrink_add(dentry
, &data
->dispose
);
1392 * We can return to the caller if we have found some (this
1393 * ensures forward progress). We'll be coming back to find
1396 if (!list_empty(&data
->dispose
))
1397 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1403 * shrink_dcache_parent - prune dcache
1404 * @parent: parent of entries to prune
1406 * Prune the dcache to remove unused children of the parent dentry.
1408 void shrink_dcache_parent(struct dentry
*parent
)
1411 struct select_data data
;
1413 INIT_LIST_HEAD(&data
.dispose
);
1414 data
.start
= parent
;
1417 d_walk(parent
, &data
, select_collect
, NULL
);
1421 shrink_dentry_list(&data
.dispose
);
1425 EXPORT_SYMBOL(shrink_dcache_parent
);
1427 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1429 /* it has busy descendents; complain about those instead */
1430 if (!list_empty(&dentry
->d_subdirs
))
1431 return D_WALK_CONTINUE
;
1433 /* root with refcount 1 is fine */
1434 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1435 return D_WALK_CONTINUE
;
1437 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1438 " still in use (%d) [unmount of %s %s]\n",
1441 dentry
->d_inode
->i_ino
: 0UL,
1443 dentry
->d_lockref
.count
,
1444 dentry
->d_sb
->s_type
->name
,
1445 dentry
->d_sb
->s_id
);
1447 return D_WALK_CONTINUE
;
1450 static void do_one_tree(struct dentry
*dentry
)
1452 shrink_dcache_parent(dentry
);
1453 d_walk(dentry
, dentry
, umount_check
, NULL
);
1459 * destroy the dentries attached to a superblock on unmounting
1461 void shrink_dcache_for_umount(struct super_block
*sb
)
1463 struct dentry
*dentry
;
1465 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1467 dentry
= sb
->s_root
;
1469 do_one_tree(dentry
);
1471 while (!hlist_bl_empty(&sb
->s_anon
)) {
1472 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1473 do_one_tree(dentry
);
1477 struct detach_data
{
1478 struct select_data select
;
1479 struct dentry
*mountpoint
;
1481 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1483 struct detach_data
*data
= _data
;
1485 if (d_mountpoint(dentry
)) {
1486 __dget_dlock(dentry
);
1487 data
->mountpoint
= dentry
;
1491 return select_collect(&data
->select
, dentry
);
1494 static void check_and_drop(void *_data
)
1496 struct detach_data
*data
= _data
;
1498 if (!data
->mountpoint
&& !data
->select
.found
)
1499 __d_drop(data
->select
.start
);
1503 * d_invalidate - detach submounts, prune dcache, and drop
1504 * @dentry: dentry to invalidate (aka detach, prune and drop)
1508 * The final d_drop is done as an atomic operation relative to
1509 * rename_lock ensuring there are no races with d_set_mounted. This
1510 * ensures there are no unhashed dentries on the path to a mountpoint.
1512 void d_invalidate(struct dentry
*dentry
)
1515 * If it's already been dropped, return OK.
1517 spin_lock(&dentry
->d_lock
);
1518 if (d_unhashed(dentry
)) {
1519 spin_unlock(&dentry
->d_lock
);
1522 spin_unlock(&dentry
->d_lock
);
1524 /* Negative dentries can be dropped without further checks */
1525 if (!dentry
->d_inode
) {
1531 struct detach_data data
;
1533 data
.mountpoint
= NULL
;
1534 INIT_LIST_HEAD(&data
.select
.dispose
);
1535 data
.select
.start
= dentry
;
1536 data
.select
.found
= 0;
1538 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1540 if (data
.select
.found
)
1541 shrink_dentry_list(&data
.select
.dispose
);
1543 if (data
.mountpoint
) {
1544 detach_mounts(data
.mountpoint
);
1545 dput(data
.mountpoint
);
1548 if (!data
.mountpoint
&& !data
.select
.found
)
1554 EXPORT_SYMBOL(d_invalidate
);
1557 * __d_alloc - allocate a dcache entry
1558 * @sb: filesystem it will belong to
1559 * @name: qstr of the name
1561 * Allocates a dentry. It returns %NULL if there is insufficient memory
1562 * available. On a success the dentry is returned. The name passed in is
1563 * copied and the copy passed in may be reused after this call.
1566 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1568 struct dentry
*dentry
;
1572 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1577 * We guarantee that the inline name is always NUL-terminated.
1578 * This way the memcpy() done by the name switching in rename
1579 * will still always have a NUL at the end, even if we might
1580 * be overwriting an internal NUL character
1582 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1583 if (unlikely(!name
)) {
1584 static const struct qstr anon
= QSTR_INIT("/", 1);
1586 dname
= dentry
->d_iname
;
1587 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1588 size_t size
= offsetof(struct external_name
, name
[1]);
1589 struct external_name
*p
= kmalloc(size
+ name
->len
,
1590 GFP_KERNEL_ACCOUNT
);
1592 kmem_cache_free(dentry_cache
, dentry
);
1595 atomic_set(&p
->u
.count
, 1);
1597 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1598 kasan_unpoison_shadow(dname
,
1599 round_up(name
->len
+ 1, sizeof(unsigned long)));
1601 dname
= dentry
->d_iname
;
1604 dentry
->d_name
.len
= name
->len
;
1605 dentry
->d_name
.hash
= name
->hash
;
1606 memcpy(dname
, name
->name
, name
->len
);
1607 dname
[name
->len
] = 0;
1609 /* Make sure we always see the terminating NUL character */
1611 dentry
->d_name
.name
= dname
;
1613 dentry
->d_lockref
.count
= 1;
1614 dentry
->d_flags
= 0;
1615 spin_lock_init(&dentry
->d_lock
);
1616 seqcount_init(&dentry
->d_seq
);
1617 dentry
->d_inode
= NULL
;
1618 dentry
->d_parent
= dentry
;
1620 dentry
->d_op
= NULL
;
1621 dentry
->d_fsdata
= NULL
;
1622 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1623 INIT_LIST_HEAD(&dentry
->d_lru
);
1624 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1625 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1626 INIT_LIST_HEAD(&dentry
->d_child
);
1627 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1629 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1630 err
= dentry
->d_op
->d_init(dentry
);
1632 if (dname_external(dentry
))
1633 kfree(external_name(dentry
));
1634 kmem_cache_free(dentry_cache
, dentry
);
1639 this_cpu_inc(nr_dentry
);
1645 * d_alloc - allocate a dcache entry
1646 * @parent: parent of entry to allocate
1647 * @name: qstr of the name
1649 * Allocates a dentry. It returns %NULL if there is insufficient memory
1650 * available. On a success the dentry is returned. The name passed in is
1651 * copied and the copy passed in may be reused after this call.
1653 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1655 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1658 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1659 spin_lock(&parent
->d_lock
);
1661 * don't need child lock because it is not subject
1662 * to concurrency here
1664 __dget_dlock(parent
);
1665 dentry
->d_parent
= parent
;
1666 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1667 spin_unlock(&parent
->d_lock
);
1671 EXPORT_SYMBOL(d_alloc
);
1673 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1675 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1677 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1678 dentry
->d_parent
= dget(parent
);
1684 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1685 * @sb: the superblock
1686 * @name: qstr of the name
1688 * For a filesystem that just pins its dentries in memory and never
1689 * performs lookups at all, return an unhashed IS_ROOT dentry.
1691 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1693 return __d_alloc(sb
, name
);
1695 EXPORT_SYMBOL(d_alloc_pseudo
);
1697 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1702 q
.hash_len
= hashlen_string(parent
, name
);
1703 return d_alloc(parent
, &q
);
1705 EXPORT_SYMBOL(d_alloc_name
);
1707 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1709 WARN_ON_ONCE(dentry
->d_op
);
1710 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1712 DCACHE_OP_REVALIDATE
|
1713 DCACHE_OP_WEAK_REVALIDATE
|
1720 dentry
->d_flags
|= DCACHE_OP_HASH
;
1722 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1723 if (op
->d_revalidate
)
1724 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1725 if (op
->d_weak_revalidate
)
1726 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1728 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1730 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1732 dentry
->d_flags
|= DCACHE_OP_REAL
;
1735 EXPORT_SYMBOL(d_set_d_op
);
1739 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1740 * @dentry - The dentry to mark
1742 * Mark a dentry as falling through to the lower layer (as set with
1743 * d_pin_lower()). This flag may be recorded on the medium.
1745 void d_set_fallthru(struct dentry
*dentry
)
1747 spin_lock(&dentry
->d_lock
);
1748 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1749 spin_unlock(&dentry
->d_lock
);
1751 EXPORT_SYMBOL(d_set_fallthru
);
1753 static unsigned d_flags_for_inode(struct inode
*inode
)
1755 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1758 return DCACHE_MISS_TYPE
;
1760 if (S_ISDIR(inode
->i_mode
)) {
1761 add_flags
= DCACHE_DIRECTORY_TYPE
;
1762 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1763 if (unlikely(!inode
->i_op
->lookup
))
1764 add_flags
= DCACHE_AUTODIR_TYPE
;
1766 inode
->i_opflags
|= IOP_LOOKUP
;
1768 goto type_determined
;
1771 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1772 if (unlikely(inode
->i_op
->get_link
)) {
1773 add_flags
= DCACHE_SYMLINK_TYPE
;
1774 goto type_determined
;
1776 inode
->i_opflags
|= IOP_NOFOLLOW
;
1779 if (unlikely(!S_ISREG(inode
->i_mode
)))
1780 add_flags
= DCACHE_SPECIAL_TYPE
;
1783 if (unlikely(IS_AUTOMOUNT(inode
)))
1784 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1788 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1790 unsigned add_flags
= d_flags_for_inode(inode
);
1791 WARN_ON(d_in_lookup(dentry
));
1793 spin_lock(&dentry
->d_lock
);
1794 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1795 raw_write_seqcount_begin(&dentry
->d_seq
);
1796 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1797 raw_write_seqcount_end(&dentry
->d_seq
);
1798 fsnotify_update_flags(dentry
);
1799 spin_unlock(&dentry
->d_lock
);
1803 * d_instantiate - fill in inode information for a dentry
1804 * @entry: dentry to complete
1805 * @inode: inode to attach to this dentry
1807 * Fill in inode information in the entry.
1809 * This turns negative dentries into productive full members
1812 * NOTE! This assumes that the inode count has been incremented
1813 * (or otherwise set) by the caller to indicate that it is now
1814 * in use by the dcache.
1817 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1819 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1821 security_d_instantiate(entry
, inode
);
1822 spin_lock(&inode
->i_lock
);
1823 __d_instantiate(entry
, inode
);
1824 spin_unlock(&inode
->i_lock
);
1827 EXPORT_SYMBOL(d_instantiate
);
1830 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1831 * @entry: dentry to complete
1832 * @inode: inode to attach to this dentry
1834 * Fill in inode information in the entry. If a directory alias is found, then
1835 * return an error (and drop inode). Together with d_materialise_unique() this
1836 * guarantees that a directory inode may never have more than one alias.
1838 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1840 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1842 security_d_instantiate(entry
, inode
);
1843 spin_lock(&inode
->i_lock
);
1844 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1845 spin_unlock(&inode
->i_lock
);
1849 __d_instantiate(entry
, inode
);
1850 spin_unlock(&inode
->i_lock
);
1854 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1856 struct dentry
*d_make_root(struct inode
*root_inode
)
1858 struct dentry
*res
= NULL
;
1861 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1863 d_instantiate(res
, root_inode
);
1869 EXPORT_SYMBOL(d_make_root
);
1871 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1873 struct dentry
*alias
;
1875 if (hlist_empty(&inode
->i_dentry
))
1877 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1883 * d_find_any_alias - find any alias for a given inode
1884 * @inode: inode to find an alias for
1886 * If any aliases exist for the given inode, take and return a
1887 * reference for one of them. If no aliases exist, return %NULL.
1889 struct dentry
*d_find_any_alias(struct inode
*inode
)
1893 spin_lock(&inode
->i_lock
);
1894 de
= __d_find_any_alias(inode
);
1895 spin_unlock(&inode
->i_lock
);
1898 EXPORT_SYMBOL(d_find_any_alias
);
1900 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1907 return ERR_PTR(-ESTALE
);
1909 return ERR_CAST(inode
);
1911 res
= d_find_any_alias(inode
);
1915 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1917 res
= ERR_PTR(-ENOMEM
);
1921 security_d_instantiate(tmp
, inode
);
1922 spin_lock(&inode
->i_lock
);
1923 res
= __d_find_any_alias(inode
);
1925 spin_unlock(&inode
->i_lock
);
1930 /* attach a disconnected dentry */
1931 add_flags
= d_flags_for_inode(inode
);
1934 add_flags
|= DCACHE_DISCONNECTED
;
1936 spin_lock(&tmp
->d_lock
);
1937 __d_set_inode_and_type(tmp
, inode
, add_flags
);
1938 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1939 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1940 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1941 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1942 spin_unlock(&tmp
->d_lock
);
1943 spin_unlock(&inode
->i_lock
);
1953 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1954 * @inode: inode to allocate the dentry for
1956 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1957 * similar open by handle operations. The returned dentry may be anonymous,
1958 * or may have a full name (if the inode was already in the cache).
1960 * When called on a directory inode, we must ensure that the inode only ever
1961 * has one dentry. If a dentry is found, that is returned instead of
1962 * allocating a new one.
1964 * On successful return, the reference to the inode has been transferred
1965 * to the dentry. In case of an error the reference on the inode is released.
1966 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1967 * be passed in and the error will be propagated to the return value,
1968 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1970 struct dentry
*d_obtain_alias(struct inode
*inode
)
1972 return __d_obtain_alias(inode
, 1);
1974 EXPORT_SYMBOL(d_obtain_alias
);
1977 * d_obtain_root - find or allocate a dentry for a given inode
1978 * @inode: inode to allocate the dentry for
1980 * Obtain an IS_ROOT dentry for the root of a filesystem.
1982 * We must ensure that directory inodes only ever have one dentry. If a
1983 * dentry is found, that is returned instead of allocating a new one.
1985 * On successful return, the reference to the inode has been transferred
1986 * to the dentry. In case of an error the reference on the inode is
1987 * released. A %NULL or IS_ERR inode may be passed in and will be the
1988 * error will be propagate to the return value, with a %NULL @inode
1989 * replaced by ERR_PTR(-ESTALE).
1991 struct dentry
*d_obtain_root(struct inode
*inode
)
1993 return __d_obtain_alias(inode
, 0);
1995 EXPORT_SYMBOL(d_obtain_root
);
1998 * d_add_ci - lookup or allocate new dentry with case-exact name
1999 * @inode: the inode case-insensitive lookup has found
2000 * @dentry: the negative dentry that was passed to the parent's lookup func
2001 * @name: the case-exact name to be associated with the returned dentry
2003 * This is to avoid filling the dcache with case-insensitive names to the
2004 * same inode, only the actual correct case is stored in the dcache for
2005 * case-insensitive filesystems.
2007 * For a case-insensitive lookup match and if the the case-exact dentry
2008 * already exists in in the dcache, use it and return it.
2010 * If no entry exists with the exact case name, allocate new dentry with
2011 * the exact case, and return the spliced entry.
2013 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2016 struct dentry
*found
, *res
;
2019 * First check if a dentry matching the name already exists,
2020 * if not go ahead and create it now.
2022 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2027 if (d_in_lookup(dentry
)) {
2028 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2030 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2035 found
= d_alloc(dentry
->d_parent
, name
);
2038 return ERR_PTR(-ENOMEM
);
2041 res
= d_splice_alias(inode
, found
);
2048 EXPORT_SYMBOL(d_add_ci
);
2051 static inline bool d_same_name(const struct dentry
*dentry
,
2052 const struct dentry
*parent
,
2053 const struct qstr
*name
)
2055 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2056 if (dentry
->d_name
.len
!= name
->len
)
2058 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2060 return parent
->d_op
->d_compare(dentry
,
2061 dentry
->d_name
.len
, dentry
->d_name
.name
,
2066 * __d_lookup_rcu - search for a dentry (racy, store-free)
2067 * @parent: parent dentry
2068 * @name: qstr of name we wish to find
2069 * @seqp: returns d_seq value at the point where the dentry was found
2070 * Returns: dentry, or NULL
2072 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2073 * resolution (store-free path walking) design described in
2074 * Documentation/filesystems/path-lookup.txt.
2076 * This is not to be used outside core vfs.
2078 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2079 * held, and rcu_read_lock held. The returned dentry must not be stored into
2080 * without taking d_lock and checking d_seq sequence count against @seq
2083 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2086 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2087 * the returned dentry, so long as its parent's seqlock is checked after the
2088 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2089 * is formed, giving integrity down the path walk.
2091 * NOTE! The caller *has* to check the resulting dentry against the sequence
2092 * number we've returned before using any of the resulting dentry state!
2094 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2095 const struct qstr
*name
,
2098 u64 hashlen
= name
->hash_len
;
2099 const unsigned char *str
= name
->name
;
2100 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2101 struct hlist_bl_node
*node
;
2102 struct dentry
*dentry
;
2105 * Note: There is significant duplication with __d_lookup_rcu which is
2106 * required to prevent single threaded performance regressions
2107 * especially on architectures where smp_rmb (in seqcounts) are costly.
2108 * Keep the two functions in sync.
2112 * The hash list is protected using RCU.
2114 * Carefully use d_seq when comparing a candidate dentry, to avoid
2115 * races with d_move().
2117 * It is possible that concurrent renames can mess up our list
2118 * walk here and result in missing our dentry, resulting in the
2119 * false-negative result. d_lookup() protects against concurrent
2120 * renames using rename_lock seqlock.
2122 * See Documentation/filesystems/path-lookup.txt for more details.
2124 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2129 * The dentry sequence count protects us from concurrent
2130 * renames, and thus protects parent and name fields.
2132 * The caller must perform a seqcount check in order
2133 * to do anything useful with the returned dentry.
2135 * NOTE! We do a "raw" seqcount_begin here. That means that
2136 * we don't wait for the sequence count to stabilize if it
2137 * is in the middle of a sequence change. If we do the slow
2138 * dentry compare, we will do seqretries until it is stable,
2139 * and if we end up with a successful lookup, we actually
2140 * want to exit RCU lookup anyway.
2142 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2143 * we are still guaranteed NUL-termination of ->d_name.name.
2145 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2146 if (dentry
->d_parent
!= parent
)
2148 if (d_unhashed(dentry
))
2151 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2154 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2156 tlen
= dentry
->d_name
.len
;
2157 tname
= dentry
->d_name
.name
;
2158 /* we want a consistent (name,len) pair */
2159 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2163 if (parent
->d_op
->d_compare(dentry
,
2164 tlen
, tname
, name
) != 0)
2167 if (dentry
->d_name
.hash_len
!= hashlen
)
2169 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2179 * d_lookup - search for a dentry
2180 * @parent: parent dentry
2181 * @name: qstr of name we wish to find
2182 * Returns: dentry, or NULL
2184 * d_lookup searches the children of the parent dentry for the name in
2185 * question. If the dentry is found its reference count is incremented and the
2186 * dentry is returned. The caller must use dput to free the entry when it has
2187 * finished using it. %NULL is returned if the dentry does not exist.
2189 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2191 struct dentry
*dentry
;
2195 seq
= read_seqbegin(&rename_lock
);
2196 dentry
= __d_lookup(parent
, name
);
2199 } while (read_seqretry(&rename_lock
, seq
));
2202 EXPORT_SYMBOL(d_lookup
);
2205 * __d_lookup - search for a dentry (racy)
2206 * @parent: parent dentry
2207 * @name: qstr of name we wish to find
2208 * Returns: dentry, or NULL
2210 * __d_lookup is like d_lookup, however it may (rarely) return a
2211 * false-negative result due to unrelated rename activity.
2213 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2214 * however it must be used carefully, eg. with a following d_lookup in
2215 * the case of failure.
2217 * __d_lookup callers must be commented.
2219 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2221 unsigned int hash
= name
->hash
;
2222 struct hlist_bl_head
*b
= d_hash(hash
);
2223 struct hlist_bl_node
*node
;
2224 struct dentry
*found
= NULL
;
2225 struct dentry
*dentry
;
2228 * Note: There is significant duplication with __d_lookup_rcu which is
2229 * required to prevent single threaded performance regressions
2230 * especially on architectures where smp_rmb (in seqcounts) are costly.
2231 * Keep the two functions in sync.
2235 * The hash list is protected using RCU.
2237 * Take d_lock when comparing a candidate dentry, to avoid races
2240 * It is possible that concurrent renames can mess up our list
2241 * walk here and result in missing our dentry, resulting in the
2242 * false-negative result. d_lookup() protects against concurrent
2243 * renames using rename_lock seqlock.
2245 * See Documentation/filesystems/path-lookup.txt for more details.
2249 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2251 if (dentry
->d_name
.hash
!= hash
)
2254 spin_lock(&dentry
->d_lock
);
2255 if (dentry
->d_parent
!= parent
)
2257 if (d_unhashed(dentry
))
2260 if (!d_same_name(dentry
, parent
, name
))
2263 dentry
->d_lockref
.count
++;
2265 spin_unlock(&dentry
->d_lock
);
2268 spin_unlock(&dentry
->d_lock
);
2276 * d_hash_and_lookup - hash the qstr then search for a dentry
2277 * @dir: Directory to search in
2278 * @name: qstr of name we wish to find
2280 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2282 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2285 * Check for a fs-specific hash function. Note that we must
2286 * calculate the standard hash first, as the d_op->d_hash()
2287 * routine may choose to leave the hash value unchanged.
2289 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2290 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2291 int err
= dir
->d_op
->d_hash(dir
, name
);
2292 if (unlikely(err
< 0))
2293 return ERR_PTR(err
);
2295 return d_lookup(dir
, name
);
2297 EXPORT_SYMBOL(d_hash_and_lookup
);
2300 * When a file is deleted, we have two options:
2301 * - turn this dentry into a negative dentry
2302 * - unhash this dentry and free it.
2304 * Usually, we want to just turn this into
2305 * a negative dentry, but if anybody else is
2306 * currently using the dentry or the inode
2307 * we can't do that and we fall back on removing
2308 * it from the hash queues and waiting for
2309 * it to be deleted later when it has no users
2313 * d_delete - delete a dentry
2314 * @dentry: The dentry to delete
2316 * Turn the dentry into a negative dentry if possible, otherwise
2317 * remove it from the hash queues so it can be deleted later
2320 void d_delete(struct dentry
* dentry
)
2322 struct inode
*inode
;
2325 * Are we the only user?
2328 spin_lock(&dentry
->d_lock
);
2329 inode
= dentry
->d_inode
;
2330 isdir
= S_ISDIR(inode
->i_mode
);
2331 if (dentry
->d_lockref
.count
== 1) {
2332 if (!spin_trylock(&inode
->i_lock
)) {
2333 spin_unlock(&dentry
->d_lock
);
2337 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2338 dentry_unlink_inode(dentry
);
2339 fsnotify_nameremove(dentry
, isdir
);
2343 if (!d_unhashed(dentry
))
2346 spin_unlock(&dentry
->d_lock
);
2348 fsnotify_nameremove(dentry
, isdir
);
2350 EXPORT_SYMBOL(d_delete
);
2352 static void __d_rehash(struct dentry
*entry
)
2354 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2355 BUG_ON(!d_unhashed(entry
));
2357 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2362 * d_rehash - add an entry back to the hash
2363 * @entry: dentry to add to the hash
2365 * Adds a dentry to the hash according to its name.
2368 void d_rehash(struct dentry
* entry
)
2370 spin_lock(&entry
->d_lock
);
2372 spin_unlock(&entry
->d_lock
);
2374 EXPORT_SYMBOL(d_rehash
);
2376 static inline unsigned start_dir_add(struct inode
*dir
)
2380 unsigned n
= dir
->i_dir_seq
;
2381 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2387 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2389 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2392 static void d_wait_lookup(struct dentry
*dentry
)
2394 if (d_in_lookup(dentry
)) {
2395 DECLARE_WAITQUEUE(wait
, current
);
2396 add_wait_queue(dentry
->d_wait
, &wait
);
2398 set_current_state(TASK_UNINTERRUPTIBLE
);
2399 spin_unlock(&dentry
->d_lock
);
2401 spin_lock(&dentry
->d_lock
);
2402 } while (d_in_lookup(dentry
));
2406 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2407 const struct qstr
*name
,
2408 wait_queue_head_t
*wq
)
2410 unsigned int hash
= name
->hash
;
2411 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2412 struct hlist_bl_node
*node
;
2413 struct dentry
*new = d_alloc(parent
, name
);
2414 struct dentry
*dentry
;
2415 unsigned seq
, r_seq
, d_seq
;
2418 return ERR_PTR(-ENOMEM
);
2422 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
) & ~1;
2423 r_seq
= read_seqbegin(&rename_lock
);
2424 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2425 if (unlikely(dentry
)) {
2426 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2430 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2439 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2444 if (unlikely(parent
->d_inode
->i_dir_seq
!= seq
)) {
2450 * No changes for the parent since the beginning of d_lookup().
2451 * Since all removals from the chain happen with hlist_bl_lock(),
2452 * any potential in-lookup matches are going to stay here until
2453 * we unlock the chain. All fields are stable in everything
2456 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2457 if (dentry
->d_name
.hash
!= hash
)
2459 if (dentry
->d_parent
!= parent
)
2461 if (!d_same_name(dentry
, parent
, name
))
2464 /* now we can try to grab a reference */
2465 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2472 * somebody is likely to be still doing lookup for it;
2473 * wait for them to finish
2475 spin_lock(&dentry
->d_lock
);
2476 d_wait_lookup(dentry
);
2478 * it's not in-lookup anymore; in principle we should repeat
2479 * everything from dcache lookup, but it's likely to be what
2480 * d_lookup() would've found anyway. If it is, just return it;
2481 * otherwise we really have to repeat the whole thing.
2483 if (unlikely(dentry
->d_name
.hash
!= hash
))
2485 if (unlikely(dentry
->d_parent
!= parent
))
2487 if (unlikely(d_unhashed(dentry
)))
2489 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2491 /* OK, it *is* a hashed match; return it */
2492 spin_unlock(&dentry
->d_lock
);
2497 /* we can't take ->d_lock here; it's OK, though. */
2498 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2500 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2504 spin_unlock(&dentry
->d_lock
);
2508 EXPORT_SYMBOL(d_alloc_parallel
);
2510 void __d_lookup_done(struct dentry
*dentry
)
2512 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2513 dentry
->d_name
.hash
);
2515 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2516 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2517 wake_up_all(dentry
->d_wait
);
2518 dentry
->d_wait
= NULL
;
2520 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2521 INIT_LIST_HEAD(&dentry
->d_lru
);
2523 EXPORT_SYMBOL(__d_lookup_done
);
2525 /* inode->i_lock held if inode is non-NULL */
2527 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2529 struct inode
*dir
= NULL
;
2531 spin_lock(&dentry
->d_lock
);
2532 if (unlikely(d_in_lookup(dentry
))) {
2533 dir
= dentry
->d_parent
->d_inode
;
2534 n
= start_dir_add(dir
);
2535 __d_lookup_done(dentry
);
2538 unsigned add_flags
= d_flags_for_inode(inode
);
2539 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2540 raw_write_seqcount_begin(&dentry
->d_seq
);
2541 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2542 raw_write_seqcount_end(&dentry
->d_seq
);
2543 fsnotify_update_flags(dentry
);
2547 end_dir_add(dir
, n
);
2548 spin_unlock(&dentry
->d_lock
);
2550 spin_unlock(&inode
->i_lock
);
2554 * d_add - add dentry to hash queues
2555 * @entry: dentry to add
2556 * @inode: The inode to attach to this dentry
2558 * This adds the entry to the hash queues and initializes @inode.
2559 * The entry was actually filled in earlier during d_alloc().
2562 void d_add(struct dentry
*entry
, struct inode
*inode
)
2565 security_d_instantiate(entry
, inode
);
2566 spin_lock(&inode
->i_lock
);
2568 __d_add(entry
, inode
);
2570 EXPORT_SYMBOL(d_add
);
2573 * d_exact_alias - find and hash an exact unhashed alias
2574 * @entry: dentry to add
2575 * @inode: The inode to go with this dentry
2577 * If an unhashed dentry with the same name/parent and desired
2578 * inode already exists, hash and return it. Otherwise, return
2581 * Parent directory should be locked.
2583 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2585 struct dentry
*alias
;
2586 unsigned int hash
= entry
->d_name
.hash
;
2588 spin_lock(&inode
->i_lock
);
2589 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2591 * Don't need alias->d_lock here, because aliases with
2592 * d_parent == entry->d_parent are not subject to name or
2593 * parent changes, because the parent inode i_mutex is held.
2595 if (alias
->d_name
.hash
!= hash
)
2597 if (alias
->d_parent
!= entry
->d_parent
)
2599 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2601 spin_lock(&alias
->d_lock
);
2602 if (!d_unhashed(alias
)) {
2603 spin_unlock(&alias
->d_lock
);
2606 __dget_dlock(alias
);
2608 spin_unlock(&alias
->d_lock
);
2610 spin_unlock(&inode
->i_lock
);
2613 spin_unlock(&inode
->i_lock
);
2616 EXPORT_SYMBOL(d_exact_alias
);
2619 * dentry_update_name_case - update case insensitive dentry with a new name
2620 * @dentry: dentry to be updated
2623 * Update a case insensitive dentry with new case of name.
2625 * dentry must have been returned by d_lookup with name @name. Old and new
2626 * name lengths must match (ie. no d_compare which allows mismatched name
2629 * Parent inode i_mutex must be held over d_lookup and into this call (to
2630 * keep renames and concurrent inserts, and readdir(2) away).
2632 void dentry_update_name_case(struct dentry
*dentry
, const struct qstr
*name
)
2634 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2635 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2637 spin_lock(&dentry
->d_lock
);
2638 write_seqcount_begin(&dentry
->d_seq
);
2639 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2640 write_seqcount_end(&dentry
->d_seq
);
2641 spin_unlock(&dentry
->d_lock
);
2643 EXPORT_SYMBOL(dentry_update_name_case
);
2645 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2647 if (unlikely(dname_external(target
))) {
2648 if (unlikely(dname_external(dentry
))) {
2650 * Both external: swap the pointers
2652 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2655 * dentry:internal, target:external. Steal target's
2656 * storage and make target internal.
2658 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2659 dentry
->d_name
.len
+ 1);
2660 dentry
->d_name
.name
= target
->d_name
.name
;
2661 target
->d_name
.name
= target
->d_iname
;
2664 if (unlikely(dname_external(dentry
))) {
2666 * dentry:external, target:internal. Give dentry's
2667 * storage to target and make dentry internal
2669 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2670 target
->d_name
.len
+ 1);
2671 target
->d_name
.name
= dentry
->d_name
.name
;
2672 dentry
->d_name
.name
= dentry
->d_iname
;
2675 * Both are internal.
2678 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2679 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2680 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2681 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2682 swap(((long *) &dentry
->d_iname
)[i
],
2683 ((long *) &target
->d_iname
)[i
]);
2687 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2690 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2692 struct external_name
*old_name
= NULL
;
2693 if (unlikely(dname_external(dentry
)))
2694 old_name
= external_name(dentry
);
2695 if (unlikely(dname_external(target
))) {
2696 atomic_inc(&external_name(target
)->u
.count
);
2697 dentry
->d_name
= target
->d_name
;
2699 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2700 target
->d_name
.len
+ 1);
2701 dentry
->d_name
.name
= dentry
->d_iname
;
2702 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2704 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2705 kfree_rcu(old_name
, u
.head
);
2708 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2711 * XXXX: do we really need to take target->d_lock?
2713 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2714 spin_lock(&target
->d_parent
->d_lock
);
2716 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2717 spin_lock(&dentry
->d_parent
->d_lock
);
2718 spin_lock_nested(&target
->d_parent
->d_lock
,
2719 DENTRY_D_LOCK_NESTED
);
2721 spin_lock(&target
->d_parent
->d_lock
);
2722 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2723 DENTRY_D_LOCK_NESTED
);
2726 if (target
< dentry
) {
2727 spin_lock_nested(&target
->d_lock
, 2);
2728 spin_lock_nested(&dentry
->d_lock
, 3);
2730 spin_lock_nested(&dentry
->d_lock
, 2);
2731 spin_lock_nested(&target
->d_lock
, 3);
2735 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2737 if (target
->d_parent
!= dentry
->d_parent
)
2738 spin_unlock(&dentry
->d_parent
->d_lock
);
2739 if (target
->d_parent
!= target
)
2740 spin_unlock(&target
->d_parent
->d_lock
);
2741 spin_unlock(&target
->d_lock
);
2742 spin_unlock(&dentry
->d_lock
);
2746 * When switching names, the actual string doesn't strictly have to
2747 * be preserved in the target - because we're dropping the target
2748 * anyway. As such, we can just do a simple memcpy() to copy over
2749 * the new name before we switch, unless we are going to rehash
2750 * it. Note that if we *do* unhash the target, we are not allowed
2751 * to rehash it without giving it a new name/hash key - whether
2752 * we swap or overwrite the names here, resulting name won't match
2753 * the reality in filesystem; it's only there for d_path() purposes.
2754 * Note that all of this is happening under rename_lock, so the
2755 * any hash lookup seeing it in the middle of manipulations will
2756 * be discarded anyway. So we do not care what happens to the hash
2760 * __d_move - move a dentry
2761 * @dentry: entry to move
2762 * @target: new dentry
2763 * @exchange: exchange the two dentries
2765 * Update the dcache to reflect the move of a file name. Negative
2766 * dcache entries should not be moved in this way. Caller must hold
2767 * rename_lock, the i_mutex of the source and target directories,
2768 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2770 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2773 struct inode
*dir
= NULL
;
2775 if (!dentry
->d_inode
)
2776 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2778 BUG_ON(d_ancestor(dentry
, target
));
2779 BUG_ON(d_ancestor(target
, dentry
));
2781 dentry_lock_for_move(dentry
, target
);
2782 if (unlikely(d_in_lookup(target
))) {
2783 dir
= target
->d_parent
->d_inode
;
2784 n
= start_dir_add(dir
);
2785 __d_lookup_done(target
);
2788 write_seqcount_begin(&dentry
->d_seq
);
2789 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2792 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2796 /* Switch the names.. */
2798 swap_names(dentry
, target
);
2800 copy_name(dentry
, target
);
2802 /* rehash in new place(s) */
2807 /* ... and switch them in the tree */
2808 if (IS_ROOT(dentry
)) {
2809 /* splicing a tree */
2810 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2811 dentry
->d_parent
= target
->d_parent
;
2812 target
->d_parent
= target
;
2813 list_del_init(&target
->d_child
);
2814 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2816 /* swapping two dentries */
2817 swap(dentry
->d_parent
, target
->d_parent
);
2818 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2819 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2821 fsnotify_update_flags(target
);
2822 fsnotify_update_flags(dentry
);
2825 write_seqcount_end(&target
->d_seq
);
2826 write_seqcount_end(&dentry
->d_seq
);
2829 end_dir_add(dir
, n
);
2830 dentry_unlock_for_move(dentry
, target
);
2834 * d_move - move a dentry
2835 * @dentry: entry to move
2836 * @target: new dentry
2838 * Update the dcache to reflect the move of a file name. Negative
2839 * dcache entries should not be moved in this way. See the locking
2840 * requirements for __d_move.
2842 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2844 write_seqlock(&rename_lock
);
2845 __d_move(dentry
, target
, false);
2846 write_sequnlock(&rename_lock
);
2848 EXPORT_SYMBOL(d_move
);
2851 * d_exchange - exchange two dentries
2852 * @dentry1: first dentry
2853 * @dentry2: second dentry
2855 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2857 write_seqlock(&rename_lock
);
2859 WARN_ON(!dentry1
->d_inode
);
2860 WARN_ON(!dentry2
->d_inode
);
2861 WARN_ON(IS_ROOT(dentry1
));
2862 WARN_ON(IS_ROOT(dentry2
));
2864 __d_move(dentry1
, dentry2
, true);
2866 write_sequnlock(&rename_lock
);
2868 EXPORT_SYMBOL_GPL(d_exchange
);
2871 * d_ancestor - search for an ancestor
2872 * @p1: ancestor dentry
2875 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2876 * an ancestor of p2, else NULL.
2878 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2882 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2883 if (p
->d_parent
== p1
)
2890 * This helper attempts to cope with remotely renamed directories
2892 * It assumes that the caller is already holding
2893 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2895 * Note: If ever the locking in lock_rename() changes, then please
2896 * remember to update this too...
2898 static int __d_unalias(struct inode
*inode
,
2899 struct dentry
*dentry
, struct dentry
*alias
)
2901 struct mutex
*m1
= NULL
;
2902 struct rw_semaphore
*m2
= NULL
;
2905 /* If alias and dentry share a parent, then no extra locks required */
2906 if (alias
->d_parent
== dentry
->d_parent
)
2909 /* See lock_rename() */
2910 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2912 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2913 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2915 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2917 __d_move(alias
, dentry
, false);
2928 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2929 * @inode: the inode which may have a disconnected dentry
2930 * @dentry: a negative dentry which we want to point to the inode.
2932 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2933 * place of the given dentry and return it, else simply d_add the inode
2934 * to the dentry and return NULL.
2936 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2937 * we should error out: directories can't have multiple aliases.
2939 * This is needed in the lookup routine of any filesystem that is exportable
2940 * (via knfsd) so that we can build dcache paths to directories effectively.
2942 * If a dentry was found and moved, then it is returned. Otherwise NULL
2943 * is returned. This matches the expected return value of ->lookup.
2945 * Cluster filesystems may call this function with a negative, hashed dentry.
2946 * In that case, we know that the inode will be a regular file, and also this
2947 * will only occur during atomic_open. So we need to check for the dentry
2948 * being already hashed only in the final case.
2950 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2953 return ERR_CAST(inode
);
2955 BUG_ON(!d_unhashed(dentry
));
2960 security_d_instantiate(dentry
, inode
);
2961 spin_lock(&inode
->i_lock
);
2962 if (S_ISDIR(inode
->i_mode
)) {
2963 struct dentry
*new = __d_find_any_alias(inode
);
2964 if (unlikely(new)) {
2965 /* The reference to new ensures it remains an alias */
2966 spin_unlock(&inode
->i_lock
);
2967 write_seqlock(&rename_lock
);
2968 if (unlikely(d_ancestor(new, dentry
))) {
2969 write_sequnlock(&rename_lock
);
2971 new = ERR_PTR(-ELOOP
);
2972 pr_warn_ratelimited(
2973 "VFS: Lookup of '%s' in %s %s"
2974 " would have caused loop\n",
2975 dentry
->d_name
.name
,
2976 inode
->i_sb
->s_type
->name
,
2978 } else if (!IS_ROOT(new)) {
2979 int err
= __d_unalias(inode
, dentry
, new);
2980 write_sequnlock(&rename_lock
);
2986 __d_move(new, dentry
, false);
2987 write_sequnlock(&rename_lock
);
2994 __d_add(dentry
, inode
);
2997 EXPORT_SYMBOL(d_splice_alias
);
2999 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3003 return -ENAMETOOLONG
;
3005 memcpy(*buffer
, str
, namelen
);
3010 * prepend_name - prepend a pathname in front of current buffer pointer
3011 * @buffer: buffer pointer
3012 * @buflen: allocated length of the buffer
3013 * @name: name string and length qstr structure
3015 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3016 * make sure that either the old or the new name pointer and length are
3017 * fetched. However, there may be mismatch between length and pointer.
3018 * The length cannot be trusted, we need to copy it byte-by-byte until
3019 * the length is reached or a null byte is found. It also prepends "/" at
3020 * the beginning of the name. The sequence number check at the caller will
3021 * retry it again when a d_move() does happen. So any garbage in the buffer
3022 * due to mismatched pointer and length will be discarded.
3024 * Data dependency barrier is needed to make sure that we see that terminating
3025 * NUL. Alpha strikes again, film at 11...
3027 static int prepend_name(char **buffer
, int *buflen
, const struct qstr
*name
)
3029 const char *dname
= ACCESS_ONCE(name
->name
);
3030 u32 dlen
= ACCESS_ONCE(name
->len
);
3033 smp_read_barrier_depends();
3035 *buflen
-= dlen
+ 1;
3037 return -ENAMETOOLONG
;
3038 p
= *buffer
-= dlen
+ 1;
3050 * prepend_path - Prepend path string to a buffer
3051 * @path: the dentry/vfsmount to report
3052 * @root: root vfsmnt/dentry
3053 * @buffer: pointer to the end of the buffer
3054 * @buflen: pointer to buffer length
3056 * The function will first try to write out the pathname without taking any
3057 * lock other than the RCU read lock to make sure that dentries won't go away.
3058 * It only checks the sequence number of the global rename_lock as any change
3059 * in the dentry's d_seq will be preceded by changes in the rename_lock
3060 * sequence number. If the sequence number had been changed, it will restart
3061 * the whole pathname back-tracing sequence again by taking the rename_lock.
3062 * In this case, there is no need to take the RCU read lock as the recursive
3063 * parent pointer references will keep the dentry chain alive as long as no
3064 * rename operation is performed.
3066 static int prepend_path(const struct path
*path
,
3067 const struct path
*root
,
3068 char **buffer
, int *buflen
)
3070 struct dentry
*dentry
;
3071 struct vfsmount
*vfsmnt
;
3074 unsigned seq
, m_seq
= 0;
3080 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3087 dentry
= path
->dentry
;
3089 mnt
= real_mount(vfsmnt
);
3090 read_seqbegin_or_lock(&rename_lock
, &seq
);
3091 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3092 struct dentry
* parent
;
3094 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3095 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
3097 if (dentry
!= vfsmnt
->mnt_root
) {
3104 if (mnt
!= parent
) {
3105 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3111 error
= is_mounted(vfsmnt
) ? 1 : 2;
3114 parent
= dentry
->d_parent
;
3116 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3124 if (need_seqretry(&rename_lock
, seq
)) {
3128 done_seqretry(&rename_lock
, seq
);
3132 if (need_seqretry(&mount_lock
, m_seq
)) {
3136 done_seqretry(&mount_lock
, m_seq
);
3138 if (error
>= 0 && bptr
== *buffer
) {
3140 error
= -ENAMETOOLONG
;
3150 * __d_path - return the path of a dentry
3151 * @path: the dentry/vfsmount to report
3152 * @root: root vfsmnt/dentry
3153 * @buf: buffer to return value in
3154 * @buflen: buffer length
3156 * Convert a dentry into an ASCII path name.
3158 * Returns a pointer into the buffer or an error code if the
3159 * path was too long.
3161 * "buflen" should be positive.
3163 * If the path is not reachable from the supplied root, return %NULL.
3165 char *__d_path(const struct path
*path
,
3166 const struct path
*root
,
3167 char *buf
, int buflen
)
3169 char *res
= buf
+ buflen
;
3172 prepend(&res
, &buflen
, "\0", 1);
3173 error
= prepend_path(path
, root
, &res
, &buflen
);
3176 return ERR_PTR(error
);
3182 char *d_absolute_path(const struct path
*path
,
3183 char *buf
, int buflen
)
3185 struct path root
= {};
3186 char *res
= buf
+ buflen
;
3189 prepend(&res
, &buflen
, "\0", 1);
3190 error
= prepend_path(path
, &root
, &res
, &buflen
);
3195 return ERR_PTR(error
);
3200 * same as __d_path but appends "(deleted)" for unlinked files.
3202 static int path_with_deleted(const struct path
*path
,
3203 const struct path
*root
,
3204 char **buf
, int *buflen
)
3206 prepend(buf
, buflen
, "\0", 1);
3207 if (d_unlinked(path
->dentry
)) {
3208 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3213 return prepend_path(path
, root
, buf
, buflen
);
3216 static int prepend_unreachable(char **buffer
, int *buflen
)
3218 return prepend(buffer
, buflen
, "(unreachable)", 13);
3221 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3226 seq
= read_seqcount_begin(&fs
->seq
);
3228 } while (read_seqcount_retry(&fs
->seq
, seq
));
3232 * d_path - return the path of a dentry
3233 * @path: path to report
3234 * @buf: buffer to return value in
3235 * @buflen: buffer length
3237 * Convert a dentry into an ASCII path name. If the entry has been deleted
3238 * the string " (deleted)" is appended. Note that this is ambiguous.
3240 * Returns a pointer into the buffer or an error code if the path was
3241 * too long. Note: Callers should use the returned pointer, not the passed
3242 * in buffer, to use the name! The implementation often starts at an offset
3243 * into the buffer, and may leave 0 bytes at the start.
3245 * "buflen" should be positive.
3247 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3249 char *res
= buf
+ buflen
;
3254 * We have various synthetic filesystems that never get mounted. On
3255 * these filesystems dentries are never used for lookup purposes, and
3256 * thus don't need to be hashed. They also don't need a name until a
3257 * user wants to identify the object in /proc/pid/fd/. The little hack
3258 * below allows us to generate a name for these objects on demand:
3260 * Some pseudo inodes are mountable. When they are mounted
3261 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3262 * and instead have d_path return the mounted path.
3264 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3265 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3266 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3269 get_fs_root_rcu(current
->fs
, &root
);
3270 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3274 res
= ERR_PTR(error
);
3277 EXPORT_SYMBOL(d_path
);
3280 * Helper function for dentry_operations.d_dname() members
3282 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3283 const char *fmt
, ...)
3289 va_start(args
, fmt
);
3290 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3293 if (sz
> sizeof(temp
) || sz
> buflen
)
3294 return ERR_PTR(-ENAMETOOLONG
);
3296 buffer
+= buflen
- sz
;
3297 return memcpy(buffer
, temp
, sz
);
3300 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3302 char *end
= buffer
+ buflen
;
3303 /* these dentries are never renamed, so d_lock is not needed */
3304 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3305 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3306 prepend(&end
, &buflen
, "/", 1))
3307 end
= ERR_PTR(-ENAMETOOLONG
);
3310 EXPORT_SYMBOL(simple_dname
);
3313 * Write full pathname from the root of the filesystem into the buffer.
3315 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3317 struct dentry
*dentry
;
3330 prepend(&end
, &len
, "\0", 1);
3334 read_seqbegin_or_lock(&rename_lock
, &seq
);
3335 while (!IS_ROOT(dentry
)) {
3336 struct dentry
*parent
= dentry
->d_parent
;
3339 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3348 if (need_seqretry(&rename_lock
, seq
)) {
3352 done_seqretry(&rename_lock
, seq
);
3357 return ERR_PTR(-ENAMETOOLONG
);
3360 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3362 return __dentry_path(dentry
, buf
, buflen
);
3364 EXPORT_SYMBOL(dentry_path_raw
);
3366 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3371 if (d_unlinked(dentry
)) {
3373 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3377 retval
= __dentry_path(dentry
, buf
, buflen
);
3378 if (!IS_ERR(retval
) && p
)
3379 *p
= '/'; /* restore '/' overriden with '\0' */
3382 return ERR_PTR(-ENAMETOOLONG
);
3385 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3391 seq
= read_seqcount_begin(&fs
->seq
);
3394 } while (read_seqcount_retry(&fs
->seq
, seq
));
3398 * NOTE! The user-level library version returns a
3399 * character pointer. The kernel system call just
3400 * returns the length of the buffer filled (which
3401 * includes the ending '\0' character), or a negative
3402 * error value. So libc would do something like
3404 * char *getcwd(char * buf, size_t size)
3408 * retval = sys_getcwd(buf, size);
3415 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3418 struct path pwd
, root
;
3419 char *page
= __getname();
3425 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3428 if (!d_unlinked(pwd
.dentry
)) {
3430 char *cwd
= page
+ PATH_MAX
;
3431 int buflen
= PATH_MAX
;
3433 prepend(&cwd
, &buflen
, "\0", 1);
3434 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3440 /* Unreachable from current root */
3442 error
= prepend_unreachable(&cwd
, &buflen
);
3448 len
= PATH_MAX
+ page
- cwd
;
3451 if (copy_to_user(buf
, cwd
, len
))
3464 * Test whether new_dentry is a subdirectory of old_dentry.
3466 * Trivially implemented using the dcache structure
3470 * is_subdir - is new dentry a subdirectory of old_dentry
3471 * @new_dentry: new dentry
3472 * @old_dentry: old dentry
3474 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3475 * Returns false otherwise.
3476 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3479 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3484 if (new_dentry
== old_dentry
)
3488 /* for restarting inner loop in case of seq retry */
3489 seq
= read_seqbegin(&rename_lock
);
3491 * Need rcu_readlock to protect against the d_parent trashing
3495 if (d_ancestor(old_dentry
, new_dentry
))
3500 } while (read_seqretry(&rename_lock
, seq
));
3505 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3507 struct dentry
*root
= data
;
3508 if (dentry
!= root
) {
3509 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3512 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3513 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3514 dentry
->d_lockref
.count
--;
3517 return D_WALK_CONTINUE
;
3520 void d_genocide(struct dentry
*parent
)
3522 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3525 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3527 inode_dec_link_count(inode
);
3528 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3529 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3530 !d_unlinked(dentry
));
3531 spin_lock(&dentry
->d_parent
->d_lock
);
3532 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3533 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3534 (unsigned long long)inode
->i_ino
);
3535 spin_unlock(&dentry
->d_lock
);
3536 spin_unlock(&dentry
->d_parent
->d_lock
);
3537 d_instantiate(dentry
, inode
);
3539 EXPORT_SYMBOL(d_tmpfile
);
3541 static __initdata
unsigned long dhash_entries
;
3542 static int __init
set_dhash_entries(char *str
)
3546 dhash_entries
= simple_strtoul(str
, &str
, 0);
3549 __setup("dhash_entries=", set_dhash_entries
);
3551 static void __init
dcache_init_early(void)
3555 /* If hashes are distributed across NUMA nodes, defer
3556 * hash allocation until vmalloc space is available.
3562 alloc_large_system_hash("Dentry cache",
3563 sizeof(struct hlist_bl_head
),
3572 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3573 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3576 static void __init
dcache_init(void)
3581 * A constructor could be added for stable state like the lists,
3582 * but it is probably not worth it because of the cache nature
3585 dentry_cache
= KMEM_CACHE(dentry
,
3586 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3588 /* Hash may have been set up in dcache_init_early */
3593 alloc_large_system_hash("Dentry cache",
3594 sizeof(struct hlist_bl_head
),
3603 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3604 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3607 /* SLAB cache for __getname() consumers */
3608 struct kmem_cache
*names_cachep __read_mostly
;
3609 EXPORT_SYMBOL(names_cachep
);
3611 EXPORT_SYMBOL(d_genocide
);
3613 void __init
vfs_caches_init_early(void)
3615 dcache_init_early();
3619 void __init
vfs_caches_init(void)
3621 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3622 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3627 files_maxfiles_init();