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 <asm/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(const struct dentry
*parent
,
110 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
111 return dentry_hashtable
+ hash_32(hash
, d_hash_shift
);
114 #define IN_LOOKUP_SHIFT 10
115 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
117 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
120 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
121 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
125 /* Statistics gathering. */
126 struct dentry_stat_t dentry_stat
= {
130 static DEFINE_PER_CPU(long, nr_dentry
);
131 static DEFINE_PER_CPU(long, nr_dentry_unused
);
133 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
136 * Here we resort to our own counters instead of using generic per-cpu counters
137 * for consistency with what the vfs inode code does. We are expected to harvest
138 * better code and performance by having our own specialized counters.
140 * Please note that the loop is done over all possible CPUs, not over all online
141 * CPUs. The reason for this is that we don't want to play games with CPUs going
142 * on and off. If one of them goes off, we will just keep their counters.
144 * glommer: See cffbc8a for details, and if you ever intend to change this,
145 * please update all vfs counters to match.
147 static long get_nr_dentry(void)
151 for_each_possible_cpu(i
)
152 sum
+= per_cpu(nr_dentry
, i
);
153 return sum
< 0 ? 0 : sum
;
156 static long get_nr_dentry_unused(void)
160 for_each_possible_cpu(i
)
161 sum
+= per_cpu(nr_dentry_unused
, i
);
162 return sum
< 0 ? 0 : sum
;
165 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
166 size_t *lenp
, loff_t
*ppos
)
168 dentry_stat
.nr_dentry
= get_nr_dentry();
169 dentry_stat
.nr_unused
= get_nr_dentry_unused();
170 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
175 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
176 * The strings are both count bytes long, and count is non-zero.
178 #ifdef CONFIG_DCACHE_WORD_ACCESS
180 #include <asm/word-at-a-time.h>
182 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
183 * aligned allocation for this particular component. We don't
184 * strictly need the load_unaligned_zeropad() safety, but it
185 * doesn't hurt either.
187 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
188 * need the careful unaligned handling.
190 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
192 unsigned long a
,b
,mask
;
195 a
= *(unsigned long *)cs
;
196 b
= load_unaligned_zeropad(ct
);
197 if (tcount
< sizeof(unsigned long))
199 if (unlikely(a
!= b
))
201 cs
+= sizeof(unsigned long);
202 ct
+= sizeof(unsigned long);
203 tcount
-= sizeof(unsigned long);
207 mask
= bytemask_from_count(tcount
);
208 return unlikely(!!((a
^ b
) & mask
));
213 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
227 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
230 * Be careful about RCU walk racing with rename:
231 * use 'lockless_dereference' to fetch the name pointer.
233 * NOTE! Even if a rename will mean that the length
234 * was not loaded atomically, we don't care. The
235 * RCU walk will check the sequence count eventually,
236 * and catch it. And we won't overrun the buffer,
237 * because we're reading the name pointer atomically,
238 * and a dentry name is guaranteed to be properly
239 * terminated with a NUL byte.
241 * End result: even if 'len' is wrong, we'll exit
242 * early because the data cannot match (there can
243 * be no NUL in the ct/tcount data)
245 const unsigned char *cs
= lockless_dereference(dentry
->d_name
.name
);
247 return dentry_string_cmp(cs
, ct
, tcount
);
250 struct external_name
{
253 struct rcu_head head
;
255 unsigned char name
[];
258 static inline struct external_name
*external_name(struct dentry
*dentry
)
260 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
263 static void __d_free(struct rcu_head
*head
)
265 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
267 kmem_cache_free(dentry_cache
, dentry
);
270 static void __d_free_external(struct rcu_head
*head
)
272 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
273 kfree(external_name(dentry
));
274 kmem_cache_free(dentry_cache
, dentry
);
277 static inline int dname_external(const struct dentry
*dentry
)
279 return dentry
->d_name
.name
!= dentry
->d_iname
;
282 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
288 dentry
->d_inode
= inode
;
289 flags
= READ_ONCE(dentry
->d_flags
);
290 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
292 WRITE_ONCE(dentry
->d_flags
, flags
);
295 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
297 unsigned flags
= READ_ONCE(dentry
->d_flags
);
299 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
300 WRITE_ONCE(dentry
->d_flags
, flags
);
301 dentry
->d_inode
= NULL
;
304 static void dentry_free(struct dentry
*dentry
)
306 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
307 if (unlikely(dname_external(dentry
))) {
308 struct external_name
*p
= external_name(dentry
);
309 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
310 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
314 /* if dentry was never visible to RCU, immediate free is OK */
315 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
316 __d_free(&dentry
->d_u
.d_rcu
);
318 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
322 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
323 * @dentry: the target dentry
324 * After this call, in-progress rcu-walk path lookup will fail. This
325 * should be called after unhashing, and after changing d_inode (if
326 * the dentry has not already been unhashed).
328 static inline void dentry_rcuwalk_invalidate(struct dentry
*dentry
)
330 lockdep_assert_held(&dentry
->d_lock
);
331 /* Go through am invalidation barrier */
332 write_seqcount_invalidate(&dentry
->d_seq
);
336 * Release the dentry's inode, using the filesystem
337 * d_iput() operation if defined.
339 static void dentry_unlink_inode(struct dentry
* dentry
)
340 __releases(dentry
->d_lock
)
341 __releases(dentry
->d_inode
->i_lock
)
343 struct inode
*inode
= dentry
->d_inode
;
344 bool hashed
= !d_unhashed(dentry
);
347 raw_write_seqcount_begin(&dentry
->d_seq
);
348 __d_clear_type_and_inode(dentry
);
349 hlist_del_init(&dentry
->d_u
.d_alias
);
351 raw_write_seqcount_end(&dentry
->d_seq
);
352 spin_unlock(&dentry
->d_lock
);
353 spin_unlock(&inode
->i_lock
);
355 fsnotify_inoderemove(inode
);
356 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
357 dentry
->d_op
->d_iput(dentry
, inode
);
363 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
364 * is in use - which includes both the "real" per-superblock
365 * LRU list _and_ the DCACHE_SHRINK_LIST use.
367 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
368 * on the shrink list (ie not on the superblock LRU list).
370 * The per-cpu "nr_dentry_unused" counters are updated with
371 * the DCACHE_LRU_LIST bit.
373 * These helper functions make sure we always follow the
374 * rules. d_lock must be held by the caller.
376 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
377 static void d_lru_add(struct dentry
*dentry
)
379 D_FLAG_VERIFY(dentry
, 0);
380 dentry
->d_flags
|= DCACHE_LRU_LIST
;
381 this_cpu_inc(nr_dentry_unused
);
382 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
385 static void d_lru_del(struct dentry
*dentry
)
387 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
388 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
389 this_cpu_dec(nr_dentry_unused
);
390 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
393 static void d_shrink_del(struct dentry
*dentry
)
395 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
396 list_del_init(&dentry
->d_lru
);
397 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
398 this_cpu_dec(nr_dentry_unused
);
401 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
403 D_FLAG_VERIFY(dentry
, 0);
404 list_add(&dentry
->d_lru
, list
);
405 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
406 this_cpu_inc(nr_dentry_unused
);
410 * These can only be called under the global LRU lock, ie during the
411 * callback for freeing the LRU list. "isolate" removes it from the
412 * LRU lists entirely, while shrink_move moves it to the indicated
415 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
417 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
418 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
419 this_cpu_dec(nr_dentry_unused
);
420 list_lru_isolate(lru
, &dentry
->d_lru
);
423 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
424 struct list_head
*list
)
426 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
427 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
428 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
432 * dentry_lru_(add|del)_list) must be called with d_lock held.
434 static void dentry_lru_add(struct dentry
*dentry
)
436 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
441 * d_drop - drop a dentry
442 * @dentry: dentry to drop
444 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
445 * be found through a VFS lookup any more. Note that this is different from
446 * deleting the dentry - d_delete will try to mark the dentry negative if
447 * possible, giving a successful _negative_ lookup, while d_drop will
448 * just make the cache lookup fail.
450 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
451 * reason (NFS timeouts or autofs deletes).
453 * __d_drop requires dentry->d_lock.
455 void __d_drop(struct dentry
*dentry
)
457 if (!d_unhashed(dentry
)) {
458 struct hlist_bl_head
*b
;
460 * Hashed dentries are normally on the dentry hashtable,
461 * with the exception of those newly allocated by
462 * d_obtain_alias, which are always IS_ROOT:
464 if (unlikely(IS_ROOT(dentry
)))
465 b
= &dentry
->d_sb
->s_anon
;
467 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
470 __hlist_bl_del(&dentry
->d_hash
);
471 dentry
->d_hash
.pprev
= NULL
;
473 dentry_rcuwalk_invalidate(dentry
);
476 EXPORT_SYMBOL(__d_drop
);
478 void d_drop(struct dentry
*dentry
)
480 spin_lock(&dentry
->d_lock
);
482 spin_unlock(&dentry
->d_lock
);
484 EXPORT_SYMBOL(d_drop
);
486 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
490 * Inform d_walk() and shrink_dentry_list() that we are no longer
491 * attached to the dentry tree
493 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
494 if (unlikely(list_empty(&dentry
->d_child
)))
496 __list_del_entry(&dentry
->d_child
);
498 * Cursors can move around the list of children. While we'd been
499 * a normal list member, it didn't matter - ->d_child.next would've
500 * been updated. However, from now on it won't be and for the
501 * things like d_walk() it might end up with a nasty surprise.
502 * Normally d_walk() doesn't care about cursors moving around -
503 * ->d_lock on parent prevents that and since a cursor has no children
504 * of its own, we get through it without ever unlocking the parent.
505 * There is one exception, though - if we ascend from a child that
506 * gets killed as soon as we unlock it, the next sibling is found
507 * using the value left in its ->d_child.next. And if _that_
508 * pointed to a cursor, and cursor got moved (e.g. by lseek())
509 * before d_walk() regains parent->d_lock, we'll end up skipping
510 * everything the cursor had been moved past.
512 * Solution: make sure that the pointer left behind in ->d_child.next
513 * points to something that won't be moving around. I.e. skip the
516 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
517 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
518 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
520 dentry
->d_child
.next
= next
->d_child
.next
;
524 static void __dentry_kill(struct dentry
*dentry
)
526 struct dentry
*parent
= NULL
;
527 bool can_free
= true;
528 if (!IS_ROOT(dentry
))
529 parent
= dentry
->d_parent
;
532 * The dentry is now unrecoverably dead to the world.
534 lockref_mark_dead(&dentry
->d_lockref
);
537 * inform the fs via d_prune that this dentry is about to be
538 * unhashed and destroyed.
540 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
541 dentry
->d_op
->d_prune(dentry
);
543 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
544 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
547 /* if it was on the hash then remove it */
549 dentry_unlist(dentry
, parent
);
551 spin_unlock(&parent
->d_lock
);
553 dentry_unlink_inode(dentry
);
555 spin_unlock(&dentry
->d_lock
);
556 this_cpu_dec(nr_dentry
);
557 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
558 dentry
->d_op
->d_release(dentry
);
560 spin_lock(&dentry
->d_lock
);
561 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
562 dentry
->d_flags
|= DCACHE_MAY_FREE
;
565 spin_unlock(&dentry
->d_lock
);
566 if (likely(can_free
))
571 * Finish off a dentry we've decided to kill.
572 * dentry->d_lock must be held, returns with it unlocked.
573 * If ref is non-zero, then decrement the refcount too.
574 * Returns dentry requiring refcount drop, or NULL if we're done.
576 static struct dentry
*dentry_kill(struct dentry
*dentry
)
577 __releases(dentry
->d_lock
)
579 struct inode
*inode
= dentry
->d_inode
;
580 struct dentry
*parent
= NULL
;
582 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
585 if (!IS_ROOT(dentry
)) {
586 parent
= dentry
->d_parent
;
587 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
589 spin_unlock(&inode
->i_lock
);
594 __dentry_kill(dentry
);
598 spin_unlock(&dentry
->d_lock
);
600 return dentry
; /* try again with same dentry */
603 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
605 struct dentry
*parent
= dentry
->d_parent
;
608 if (unlikely(dentry
->d_lockref
.count
< 0))
610 if (likely(spin_trylock(&parent
->d_lock
)))
613 spin_unlock(&dentry
->d_lock
);
615 parent
= ACCESS_ONCE(dentry
->d_parent
);
616 spin_lock(&parent
->d_lock
);
618 * We can't blindly lock dentry until we are sure
619 * that we won't violate the locking order.
620 * Any changes of dentry->d_parent must have
621 * been done with parent->d_lock held, so
622 * spin_lock() above is enough of a barrier
623 * for checking if it's still our child.
625 if (unlikely(parent
!= dentry
->d_parent
)) {
626 spin_unlock(&parent
->d_lock
);
630 if (parent
!= dentry
)
631 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
638 * Try to do a lockless dput(), and return whether that was successful.
640 * If unsuccessful, we return false, having already taken the dentry lock.
642 * The caller needs to hold the RCU read lock, so that the dentry is
643 * guaranteed to stay around even if the refcount goes down to zero!
645 static inline bool fast_dput(struct dentry
*dentry
)
648 unsigned int d_flags
;
651 * If we have a d_op->d_delete() operation, we sould not
652 * let the dentry count go to zero, so use "put_or_lock".
654 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
655 return lockref_put_or_lock(&dentry
->d_lockref
);
658 * .. otherwise, we can try to just decrement the
659 * lockref optimistically.
661 ret
= lockref_put_return(&dentry
->d_lockref
);
664 * If the lockref_put_return() failed due to the lock being held
665 * by somebody else, the fast path has failed. We will need to
666 * get the lock, and then check the count again.
668 if (unlikely(ret
< 0)) {
669 spin_lock(&dentry
->d_lock
);
670 if (dentry
->d_lockref
.count
> 1) {
671 dentry
->d_lockref
.count
--;
672 spin_unlock(&dentry
->d_lock
);
679 * If we weren't the last ref, we're done.
685 * Careful, careful. The reference count went down
686 * to zero, but we don't hold the dentry lock, so
687 * somebody else could get it again, and do another
688 * dput(), and we need to not race with that.
690 * However, there is a very special and common case
691 * where we don't care, because there is nothing to
692 * do: the dentry is still hashed, it does not have
693 * a 'delete' op, and it's referenced and already on
696 * NOTE! Since we aren't locked, these values are
697 * not "stable". However, it is sufficient that at
698 * some point after we dropped the reference the
699 * dentry was hashed and the flags had the proper
700 * value. Other dentry users may have re-gotten
701 * a reference to the dentry and change that, but
702 * our work is done - we can leave the dentry
703 * around with a zero refcount.
706 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
707 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
709 /* Nothing to do? Dropping the reference was all we needed? */
710 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
714 * Not the fast normal case? Get the lock. We've already decremented
715 * the refcount, but we'll need to re-check the situation after
718 spin_lock(&dentry
->d_lock
);
721 * Did somebody else grab a reference to it in the meantime, and
722 * we're no longer the last user after all? Alternatively, somebody
723 * else could have killed it and marked it dead. Either way, we
724 * don't need to do anything else.
726 if (dentry
->d_lockref
.count
) {
727 spin_unlock(&dentry
->d_lock
);
732 * Re-get the reference we optimistically dropped. We hold the
733 * lock, and we just tested that it was zero, so we can just
736 dentry
->d_lockref
.count
= 1;
744 * This is complicated by the fact that we do not want to put
745 * dentries that are no longer on any hash chain on the unused
746 * list: we'd much rather just get rid of them immediately.
748 * However, that implies that we have to traverse the dentry
749 * tree upwards to the parents which might _also_ now be
750 * scheduled for deletion (it may have been only waiting for
751 * its last child to go away).
753 * This tail recursion is done by hand as we don't want to depend
754 * on the compiler to always get this right (gcc generally doesn't).
755 * Real recursion would eat up our stack space.
759 * dput - release a dentry
760 * @dentry: dentry to release
762 * Release a dentry. This will drop the usage count and if appropriate
763 * call the dentry unlink method as well as removing it from the queues and
764 * releasing its resources. If the parent dentries were scheduled for release
765 * they too may now get deleted.
767 void dput(struct dentry
*dentry
)
769 if (unlikely(!dentry
))
774 if (likely(fast_dput(dentry
))) {
779 /* Slow case: now with the dentry lock held */
782 WARN_ON(d_in_lookup(dentry
));
784 /* Unreachable? Get rid of it */
785 if (unlikely(d_unhashed(dentry
)))
788 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
791 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
792 if (dentry
->d_op
->d_delete(dentry
))
796 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
797 dentry
->d_flags
|= DCACHE_REFERENCED
;
798 dentry_lru_add(dentry
);
800 dentry
->d_lockref
.count
--;
801 spin_unlock(&dentry
->d_lock
);
805 dentry
= dentry_kill(dentry
);
812 /* This must be called with d_lock held */
813 static inline void __dget_dlock(struct dentry
*dentry
)
815 dentry
->d_lockref
.count
++;
818 static inline void __dget(struct dentry
*dentry
)
820 lockref_get(&dentry
->d_lockref
);
823 struct dentry
*dget_parent(struct dentry
*dentry
)
829 * Do optimistic parent lookup without any
833 ret
= ACCESS_ONCE(dentry
->d_parent
);
834 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
836 if (likely(gotref
)) {
837 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
844 * Don't need rcu_dereference because we re-check it was correct under
848 ret
= dentry
->d_parent
;
849 spin_lock(&ret
->d_lock
);
850 if (unlikely(ret
!= dentry
->d_parent
)) {
851 spin_unlock(&ret
->d_lock
);
856 BUG_ON(!ret
->d_lockref
.count
);
857 ret
->d_lockref
.count
++;
858 spin_unlock(&ret
->d_lock
);
861 EXPORT_SYMBOL(dget_parent
);
864 * d_find_alias - grab a hashed alias of inode
865 * @inode: inode in question
867 * If inode has a hashed alias, or is a directory and has any alias,
868 * acquire the reference to alias and return it. Otherwise return NULL.
869 * Notice that if inode is a directory there can be only one alias and
870 * it can be unhashed only if it has no children, or if it is the root
871 * of a filesystem, or if the directory was renamed and d_revalidate
872 * was the first vfs operation to notice.
874 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
875 * any other hashed alias over that one.
877 static struct dentry
*__d_find_alias(struct inode
*inode
)
879 struct dentry
*alias
, *discon_alias
;
883 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
884 spin_lock(&alias
->d_lock
);
885 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
886 if (IS_ROOT(alias
) &&
887 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
888 discon_alias
= alias
;
891 spin_unlock(&alias
->d_lock
);
895 spin_unlock(&alias
->d_lock
);
898 alias
= discon_alias
;
899 spin_lock(&alias
->d_lock
);
900 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
902 spin_unlock(&alias
->d_lock
);
905 spin_unlock(&alias
->d_lock
);
911 struct dentry
*d_find_alias(struct inode
*inode
)
913 struct dentry
*de
= NULL
;
915 if (!hlist_empty(&inode
->i_dentry
)) {
916 spin_lock(&inode
->i_lock
);
917 de
= __d_find_alias(inode
);
918 spin_unlock(&inode
->i_lock
);
922 EXPORT_SYMBOL(d_find_alias
);
925 * Try to kill dentries associated with this inode.
926 * WARNING: you must own a reference to inode.
928 void d_prune_aliases(struct inode
*inode
)
930 struct dentry
*dentry
;
932 spin_lock(&inode
->i_lock
);
933 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
934 spin_lock(&dentry
->d_lock
);
935 if (!dentry
->d_lockref
.count
) {
936 struct dentry
*parent
= lock_parent(dentry
);
937 if (likely(!dentry
->d_lockref
.count
)) {
938 __dentry_kill(dentry
);
943 spin_unlock(&parent
->d_lock
);
945 spin_unlock(&dentry
->d_lock
);
947 spin_unlock(&inode
->i_lock
);
949 EXPORT_SYMBOL(d_prune_aliases
);
951 static void shrink_dentry_list(struct list_head
*list
)
953 struct dentry
*dentry
, *parent
;
955 while (!list_empty(list
)) {
957 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
958 spin_lock(&dentry
->d_lock
);
959 parent
= lock_parent(dentry
);
962 * The dispose list is isolated and dentries are not accounted
963 * to the LRU here, so we can simply remove it from the list
964 * here regardless of whether it is referenced or not.
966 d_shrink_del(dentry
);
969 * We found an inuse dentry which was not removed from
970 * the LRU because of laziness during lookup. Do not free it.
972 if (dentry
->d_lockref
.count
> 0) {
973 spin_unlock(&dentry
->d_lock
);
975 spin_unlock(&parent
->d_lock
);
980 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
981 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
982 spin_unlock(&dentry
->d_lock
);
984 spin_unlock(&parent
->d_lock
);
990 inode
= dentry
->d_inode
;
991 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
992 d_shrink_add(dentry
, list
);
993 spin_unlock(&dentry
->d_lock
);
995 spin_unlock(&parent
->d_lock
);
999 __dentry_kill(dentry
);
1002 * We need to prune ancestors too. This is necessary to prevent
1003 * quadratic behavior of shrink_dcache_parent(), but is also
1004 * expected to be beneficial in reducing dentry cache
1008 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1009 parent
= lock_parent(dentry
);
1010 if (dentry
->d_lockref
.count
!= 1) {
1011 dentry
->d_lockref
.count
--;
1012 spin_unlock(&dentry
->d_lock
);
1014 spin_unlock(&parent
->d_lock
);
1017 inode
= dentry
->d_inode
; /* can't be NULL */
1018 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1019 spin_unlock(&dentry
->d_lock
);
1021 spin_unlock(&parent
->d_lock
);
1025 __dentry_kill(dentry
);
1031 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1032 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1034 struct list_head
*freeable
= arg
;
1035 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1039 * we are inverting the lru lock/dentry->d_lock here,
1040 * so use a trylock. If we fail to get the lock, just skip
1043 if (!spin_trylock(&dentry
->d_lock
))
1047 * Referenced dentries are still in use. If they have active
1048 * counts, just remove them from the LRU. Otherwise give them
1049 * another pass through the LRU.
1051 if (dentry
->d_lockref
.count
) {
1052 d_lru_isolate(lru
, dentry
);
1053 spin_unlock(&dentry
->d_lock
);
1057 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1058 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1059 spin_unlock(&dentry
->d_lock
);
1062 * The list move itself will be made by the common LRU code. At
1063 * this point, we've dropped the dentry->d_lock but keep the
1064 * lru lock. This is safe to do, since every list movement is
1065 * protected by the lru lock even if both locks are held.
1067 * This is guaranteed by the fact that all LRU management
1068 * functions are intermediated by the LRU API calls like
1069 * list_lru_add and list_lru_del. List movement in this file
1070 * only ever occur through this functions or through callbacks
1071 * like this one, that are called from the LRU API.
1073 * The only exceptions to this are functions like
1074 * shrink_dentry_list, and code that first checks for the
1075 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1076 * operating only with stack provided lists after they are
1077 * properly isolated from the main list. It is thus, always a
1083 d_lru_shrink_move(lru
, dentry
, freeable
);
1084 spin_unlock(&dentry
->d_lock
);
1090 * prune_dcache_sb - shrink the dcache
1092 * @sc: shrink control, passed to list_lru_shrink_walk()
1094 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1095 * is done when we need more memory and called from the superblock shrinker
1098 * This function may fail to free any resources if all the dentries are in
1101 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1106 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1107 dentry_lru_isolate
, &dispose
);
1108 shrink_dentry_list(&dispose
);
1112 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1113 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1115 struct list_head
*freeable
= arg
;
1116 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1119 * we are inverting the lru lock/dentry->d_lock here,
1120 * so use a trylock. If we fail to get the lock, just skip
1123 if (!spin_trylock(&dentry
->d_lock
))
1126 d_lru_shrink_move(lru
, dentry
, freeable
);
1127 spin_unlock(&dentry
->d_lock
);
1134 * shrink_dcache_sb - shrink dcache for a superblock
1137 * Shrink the dcache for the specified super block. This is used to free
1138 * the dcache before unmounting a file system.
1140 void shrink_dcache_sb(struct super_block
*sb
)
1147 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1148 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1150 this_cpu_sub(nr_dentry_unused
, freed
);
1151 shrink_dentry_list(&dispose
);
1152 } while (freed
> 0);
1154 EXPORT_SYMBOL(shrink_dcache_sb
);
1157 * enum d_walk_ret - action to talke during tree walk
1158 * @D_WALK_CONTINUE: contrinue walk
1159 * @D_WALK_QUIT: quit walk
1160 * @D_WALK_NORETRY: quit when retry is needed
1161 * @D_WALK_SKIP: skip this dentry and its children
1171 * d_walk - walk the dentry tree
1172 * @parent: start of walk
1173 * @data: data passed to @enter() and @finish()
1174 * @enter: callback when first entering the dentry
1175 * @finish: callback when successfully finished the walk
1177 * The @enter() and @finish() callbacks are called with d_lock held.
1179 static void d_walk(struct dentry
*parent
, void *data
,
1180 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1181 void (*finish
)(void *))
1183 struct dentry
*this_parent
;
1184 struct list_head
*next
;
1186 enum d_walk_ret ret
;
1190 read_seqbegin_or_lock(&rename_lock
, &seq
);
1191 this_parent
= parent
;
1192 spin_lock(&this_parent
->d_lock
);
1194 ret
= enter(data
, this_parent
);
1196 case D_WALK_CONTINUE
:
1201 case D_WALK_NORETRY
:
1206 next
= this_parent
->d_subdirs
.next
;
1208 while (next
!= &this_parent
->d_subdirs
) {
1209 struct list_head
*tmp
= next
;
1210 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1213 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1216 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1218 ret
= enter(data
, dentry
);
1220 case D_WALK_CONTINUE
:
1223 spin_unlock(&dentry
->d_lock
);
1225 case D_WALK_NORETRY
:
1229 spin_unlock(&dentry
->d_lock
);
1233 if (!list_empty(&dentry
->d_subdirs
)) {
1234 spin_unlock(&this_parent
->d_lock
);
1235 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1236 this_parent
= dentry
;
1237 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1240 spin_unlock(&dentry
->d_lock
);
1243 * All done at this level ... ascend and resume the search.
1247 if (this_parent
!= parent
) {
1248 struct dentry
*child
= this_parent
;
1249 this_parent
= child
->d_parent
;
1251 spin_unlock(&child
->d_lock
);
1252 spin_lock(&this_parent
->d_lock
);
1254 /* might go back up the wrong parent if we have had a rename. */
1255 if (need_seqretry(&rename_lock
, seq
))
1257 /* go into the first sibling still alive */
1259 next
= child
->d_child
.next
;
1260 if (next
== &this_parent
->d_subdirs
)
1262 child
= list_entry(next
, struct dentry
, d_child
);
1263 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1267 if (need_seqretry(&rename_lock
, seq
))
1274 spin_unlock(&this_parent
->d_lock
);
1275 done_seqretry(&rename_lock
, seq
);
1279 spin_unlock(&this_parent
->d_lock
);
1289 * Search for at least 1 mount point in the dentry's subdirs.
1290 * We descend to the next level whenever the d_subdirs
1291 * list is non-empty and continue searching.
1294 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1297 if (d_mountpoint(dentry
)) {
1301 return D_WALK_CONTINUE
;
1305 * have_submounts - check for mounts over a dentry
1306 * @parent: dentry to check.
1308 * Return true if the parent or its subdirectories contain
1311 int have_submounts(struct dentry
*parent
)
1315 d_walk(parent
, &ret
, check_mount
, NULL
);
1319 EXPORT_SYMBOL(have_submounts
);
1322 * Called by mount code to set a mountpoint and check if the mountpoint is
1323 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1324 * subtree can become unreachable).
1326 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1327 * this reason take rename_lock and d_lock on dentry and ancestors.
1329 int d_set_mounted(struct dentry
*dentry
)
1333 write_seqlock(&rename_lock
);
1334 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1335 /* Need exclusion wrt. d_invalidate() */
1336 spin_lock(&p
->d_lock
);
1337 if (unlikely(d_unhashed(p
))) {
1338 spin_unlock(&p
->d_lock
);
1341 spin_unlock(&p
->d_lock
);
1343 spin_lock(&dentry
->d_lock
);
1344 if (!d_unlinked(dentry
)) {
1345 dentry
->d_flags
|= DCACHE_MOUNTED
;
1348 spin_unlock(&dentry
->d_lock
);
1350 write_sequnlock(&rename_lock
);
1355 * Search the dentry child list of the specified parent,
1356 * and move any unused dentries to the end of the unused
1357 * list for prune_dcache(). We descend to the next level
1358 * whenever the d_subdirs list is non-empty and continue
1361 * It returns zero iff there are no unused children,
1362 * otherwise it returns the number of children moved to
1363 * the end of the unused list. This may not be the total
1364 * number of unused children, because select_parent can
1365 * drop the lock and return early due to latency
1369 struct select_data
{
1370 struct dentry
*start
;
1371 struct list_head dispose
;
1375 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1377 struct select_data
*data
= _data
;
1378 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1380 if (data
->start
== dentry
)
1383 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1386 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1388 if (!dentry
->d_lockref
.count
) {
1389 d_shrink_add(dentry
, &data
->dispose
);
1394 * We can return to the caller if we have found some (this
1395 * ensures forward progress). We'll be coming back to find
1398 if (!list_empty(&data
->dispose
))
1399 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1405 * shrink_dcache_parent - prune dcache
1406 * @parent: parent of entries to prune
1408 * Prune the dcache to remove unused children of the parent dentry.
1410 void shrink_dcache_parent(struct dentry
*parent
)
1413 struct select_data data
;
1415 INIT_LIST_HEAD(&data
.dispose
);
1416 data
.start
= parent
;
1419 d_walk(parent
, &data
, select_collect
, NULL
);
1423 shrink_dentry_list(&data
.dispose
);
1427 EXPORT_SYMBOL(shrink_dcache_parent
);
1429 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1431 /* it has busy descendents; complain about those instead */
1432 if (!list_empty(&dentry
->d_subdirs
))
1433 return D_WALK_CONTINUE
;
1435 /* root with refcount 1 is fine */
1436 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1437 return D_WALK_CONTINUE
;
1439 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1440 " still in use (%d) [unmount of %s %s]\n",
1443 dentry
->d_inode
->i_ino
: 0UL,
1445 dentry
->d_lockref
.count
,
1446 dentry
->d_sb
->s_type
->name
,
1447 dentry
->d_sb
->s_id
);
1449 return D_WALK_CONTINUE
;
1452 static void do_one_tree(struct dentry
*dentry
)
1454 shrink_dcache_parent(dentry
);
1455 d_walk(dentry
, dentry
, umount_check
, NULL
);
1461 * destroy the dentries attached to a superblock on unmounting
1463 void shrink_dcache_for_umount(struct super_block
*sb
)
1465 struct dentry
*dentry
;
1467 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1469 dentry
= sb
->s_root
;
1471 do_one_tree(dentry
);
1473 while (!hlist_bl_empty(&sb
->s_anon
)) {
1474 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1475 do_one_tree(dentry
);
1479 struct detach_data
{
1480 struct select_data select
;
1481 struct dentry
*mountpoint
;
1483 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1485 struct detach_data
*data
= _data
;
1487 if (d_mountpoint(dentry
)) {
1488 __dget_dlock(dentry
);
1489 data
->mountpoint
= dentry
;
1493 return select_collect(&data
->select
, dentry
);
1496 static void check_and_drop(void *_data
)
1498 struct detach_data
*data
= _data
;
1500 if (!data
->mountpoint
&& !data
->select
.found
)
1501 __d_drop(data
->select
.start
);
1505 * d_invalidate - detach submounts, prune dcache, and drop
1506 * @dentry: dentry to invalidate (aka detach, prune and drop)
1510 * The final d_drop is done as an atomic operation relative to
1511 * rename_lock ensuring there are no races with d_set_mounted. This
1512 * ensures there are no unhashed dentries on the path to a mountpoint.
1514 void d_invalidate(struct dentry
*dentry
)
1517 * If it's already been dropped, return OK.
1519 spin_lock(&dentry
->d_lock
);
1520 if (d_unhashed(dentry
)) {
1521 spin_unlock(&dentry
->d_lock
);
1524 spin_unlock(&dentry
->d_lock
);
1526 /* Negative dentries can be dropped without further checks */
1527 if (!dentry
->d_inode
) {
1533 struct detach_data data
;
1535 data
.mountpoint
= NULL
;
1536 INIT_LIST_HEAD(&data
.select
.dispose
);
1537 data
.select
.start
= dentry
;
1538 data
.select
.found
= 0;
1540 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1542 if (data
.select
.found
)
1543 shrink_dentry_list(&data
.select
.dispose
);
1545 if (data
.mountpoint
) {
1546 detach_mounts(data
.mountpoint
);
1547 dput(data
.mountpoint
);
1550 if (!data
.mountpoint
&& !data
.select
.found
)
1556 EXPORT_SYMBOL(d_invalidate
);
1559 * __d_alloc - allocate a dcache entry
1560 * @sb: filesystem it will belong to
1561 * @name: qstr of the name
1563 * Allocates a dentry. It returns %NULL if there is insufficient memory
1564 * available. On a success the dentry is returned. The name passed in is
1565 * copied and the copy passed in may be reused after this call.
1568 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1570 struct dentry
*dentry
;
1573 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1578 * We guarantee that the inline name is always NUL-terminated.
1579 * This way the memcpy() done by the name switching in rename
1580 * will still always have a NUL at the end, even if we might
1581 * be overwriting an internal NUL character
1583 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1584 if (unlikely(!name
)) {
1585 static const struct qstr anon
= QSTR_INIT("/", 1);
1587 dname
= dentry
->d_iname
;
1588 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1589 size_t size
= offsetof(struct external_name
, name
[1]);
1590 struct external_name
*p
= kmalloc(size
+ name
->len
,
1591 GFP_KERNEL_ACCOUNT
);
1593 kmem_cache_free(dentry_cache
, dentry
);
1596 atomic_set(&p
->u
.count
, 1);
1598 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1599 kasan_unpoison_shadow(dname
,
1600 round_up(name
->len
+ 1, sizeof(unsigned long)));
1602 dname
= dentry
->d_iname
;
1605 dentry
->d_name
.len
= name
->len
;
1606 dentry
->d_name
.hash
= name
->hash
;
1607 memcpy(dname
, name
->name
, name
->len
);
1608 dname
[name
->len
] = 0;
1610 /* Make sure we always see the terminating NUL character */
1612 dentry
->d_name
.name
= dname
;
1614 dentry
->d_lockref
.count
= 1;
1615 dentry
->d_flags
= 0;
1616 spin_lock_init(&dentry
->d_lock
);
1617 seqcount_init(&dentry
->d_seq
);
1618 dentry
->d_inode
= NULL
;
1619 dentry
->d_parent
= dentry
;
1621 dentry
->d_op
= NULL
;
1622 dentry
->d_fsdata
= NULL
;
1623 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1624 INIT_LIST_HEAD(&dentry
->d_lru
);
1625 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1626 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1627 INIT_LIST_HEAD(&dentry
->d_child
);
1628 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1630 this_cpu_inc(nr_dentry
);
1636 * d_alloc - allocate a dcache entry
1637 * @parent: parent of entry to allocate
1638 * @name: qstr of the name
1640 * Allocates a dentry. It returns %NULL if there is insufficient memory
1641 * available. On a success the dentry is returned. The name passed in is
1642 * copied and the copy passed in may be reused after this call.
1644 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1646 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1649 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1650 spin_lock(&parent
->d_lock
);
1652 * don't need child lock because it is not subject
1653 * to concurrency here
1655 __dget_dlock(parent
);
1656 dentry
->d_parent
= parent
;
1657 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1658 spin_unlock(&parent
->d_lock
);
1662 EXPORT_SYMBOL(d_alloc
);
1664 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1666 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1668 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1669 dentry
->d_parent
= dget(parent
);
1675 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1676 * @sb: the superblock
1677 * @name: qstr of the name
1679 * For a filesystem that just pins its dentries in memory and never
1680 * performs lookups at all, return an unhashed IS_ROOT dentry.
1682 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1684 return __d_alloc(sb
, name
);
1686 EXPORT_SYMBOL(d_alloc_pseudo
);
1688 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1693 q
.hash_len
= hashlen_string(name
);
1694 return d_alloc(parent
, &q
);
1696 EXPORT_SYMBOL(d_alloc_name
);
1698 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1700 WARN_ON_ONCE(dentry
->d_op
);
1701 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1703 DCACHE_OP_REVALIDATE
|
1704 DCACHE_OP_WEAK_REVALIDATE
|
1711 dentry
->d_flags
|= DCACHE_OP_HASH
;
1713 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1714 if (op
->d_revalidate
)
1715 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1716 if (op
->d_weak_revalidate
)
1717 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1719 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1721 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1723 dentry
->d_flags
|= DCACHE_OP_REAL
;
1726 EXPORT_SYMBOL(d_set_d_op
);
1730 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1731 * @dentry - The dentry to mark
1733 * Mark a dentry as falling through to the lower layer (as set with
1734 * d_pin_lower()). This flag may be recorded on the medium.
1736 void d_set_fallthru(struct dentry
*dentry
)
1738 spin_lock(&dentry
->d_lock
);
1739 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1740 spin_unlock(&dentry
->d_lock
);
1742 EXPORT_SYMBOL(d_set_fallthru
);
1744 static unsigned d_flags_for_inode(struct inode
*inode
)
1746 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1749 return DCACHE_MISS_TYPE
;
1751 if (S_ISDIR(inode
->i_mode
)) {
1752 add_flags
= DCACHE_DIRECTORY_TYPE
;
1753 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1754 if (unlikely(!inode
->i_op
->lookup
))
1755 add_flags
= DCACHE_AUTODIR_TYPE
;
1757 inode
->i_opflags
|= IOP_LOOKUP
;
1759 goto type_determined
;
1762 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1763 if (unlikely(inode
->i_op
->get_link
)) {
1764 add_flags
= DCACHE_SYMLINK_TYPE
;
1765 goto type_determined
;
1767 inode
->i_opflags
|= IOP_NOFOLLOW
;
1770 if (unlikely(!S_ISREG(inode
->i_mode
)))
1771 add_flags
= DCACHE_SPECIAL_TYPE
;
1774 if (unlikely(IS_AUTOMOUNT(inode
)))
1775 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1779 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1781 unsigned add_flags
= d_flags_for_inode(inode
);
1782 WARN_ON(d_in_lookup(dentry
));
1784 spin_lock(&dentry
->d_lock
);
1785 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1786 raw_write_seqcount_begin(&dentry
->d_seq
);
1787 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1788 raw_write_seqcount_end(&dentry
->d_seq
);
1789 fsnotify_update_flags(dentry
);
1790 spin_unlock(&dentry
->d_lock
);
1794 * d_instantiate - fill in inode information for a dentry
1795 * @entry: dentry to complete
1796 * @inode: inode to attach to this dentry
1798 * Fill in inode information in the entry.
1800 * This turns negative dentries into productive full members
1803 * NOTE! This assumes that the inode count has been incremented
1804 * (or otherwise set) by the caller to indicate that it is now
1805 * in use by the dcache.
1808 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1810 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1812 security_d_instantiate(entry
, inode
);
1813 spin_lock(&inode
->i_lock
);
1814 __d_instantiate(entry
, inode
);
1815 spin_unlock(&inode
->i_lock
);
1818 EXPORT_SYMBOL(d_instantiate
);
1821 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1822 * @entry: dentry to complete
1823 * @inode: inode to attach to this dentry
1825 * Fill in inode information in the entry. If a directory alias is found, then
1826 * return an error (and drop inode). Together with d_materialise_unique() this
1827 * guarantees that a directory inode may never have more than one alias.
1829 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1831 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1833 security_d_instantiate(entry
, inode
);
1834 spin_lock(&inode
->i_lock
);
1835 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1836 spin_unlock(&inode
->i_lock
);
1840 __d_instantiate(entry
, inode
);
1841 spin_unlock(&inode
->i_lock
);
1845 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1847 struct dentry
*d_make_root(struct inode
*root_inode
)
1849 struct dentry
*res
= NULL
;
1852 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1854 d_instantiate(res
, root_inode
);
1860 EXPORT_SYMBOL(d_make_root
);
1862 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1864 struct dentry
*alias
;
1866 if (hlist_empty(&inode
->i_dentry
))
1868 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1874 * d_find_any_alias - find any alias for a given inode
1875 * @inode: inode to find an alias for
1877 * If any aliases exist for the given inode, take and return a
1878 * reference for one of them. If no aliases exist, return %NULL.
1880 struct dentry
*d_find_any_alias(struct inode
*inode
)
1884 spin_lock(&inode
->i_lock
);
1885 de
= __d_find_any_alias(inode
);
1886 spin_unlock(&inode
->i_lock
);
1889 EXPORT_SYMBOL(d_find_any_alias
);
1891 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1898 return ERR_PTR(-ESTALE
);
1900 return ERR_CAST(inode
);
1902 res
= d_find_any_alias(inode
);
1906 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1908 res
= ERR_PTR(-ENOMEM
);
1912 security_d_instantiate(tmp
, inode
);
1913 spin_lock(&inode
->i_lock
);
1914 res
= __d_find_any_alias(inode
);
1916 spin_unlock(&inode
->i_lock
);
1921 /* attach a disconnected dentry */
1922 add_flags
= d_flags_for_inode(inode
);
1925 add_flags
|= DCACHE_DISCONNECTED
;
1927 spin_lock(&tmp
->d_lock
);
1928 __d_set_inode_and_type(tmp
, inode
, add_flags
);
1929 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1930 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1931 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1932 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1933 spin_unlock(&tmp
->d_lock
);
1934 spin_unlock(&inode
->i_lock
);
1944 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1945 * @inode: inode to allocate the dentry for
1947 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1948 * similar open by handle operations. The returned dentry may be anonymous,
1949 * or may have a full name (if the inode was already in the cache).
1951 * When called on a directory inode, we must ensure that the inode only ever
1952 * has one dentry. If a dentry is found, that is returned instead of
1953 * allocating a new one.
1955 * On successful return, the reference to the inode has been transferred
1956 * to the dentry. In case of an error the reference on the inode is released.
1957 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1958 * be passed in and the error will be propagated to the return value,
1959 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1961 struct dentry
*d_obtain_alias(struct inode
*inode
)
1963 return __d_obtain_alias(inode
, 1);
1965 EXPORT_SYMBOL(d_obtain_alias
);
1968 * d_obtain_root - find or allocate a dentry for a given inode
1969 * @inode: inode to allocate the dentry for
1971 * Obtain an IS_ROOT dentry for the root of a filesystem.
1973 * We must ensure that directory inodes only ever have one dentry. If a
1974 * dentry is found, that is returned instead of allocating a new one.
1976 * On successful return, the reference to the inode has been transferred
1977 * to the dentry. In case of an error the reference on the inode is
1978 * released. A %NULL or IS_ERR inode may be passed in and will be the
1979 * error will be propagate to the return value, with a %NULL @inode
1980 * replaced by ERR_PTR(-ESTALE).
1982 struct dentry
*d_obtain_root(struct inode
*inode
)
1984 return __d_obtain_alias(inode
, 0);
1986 EXPORT_SYMBOL(d_obtain_root
);
1989 * d_add_ci - lookup or allocate new dentry with case-exact name
1990 * @inode: the inode case-insensitive lookup has found
1991 * @dentry: the negative dentry that was passed to the parent's lookup func
1992 * @name: the case-exact name to be associated with the returned dentry
1994 * This is to avoid filling the dcache with case-insensitive names to the
1995 * same inode, only the actual correct case is stored in the dcache for
1996 * case-insensitive filesystems.
1998 * For a case-insensitive lookup match and if the the case-exact dentry
1999 * already exists in in the dcache, use it and return it.
2001 * If no entry exists with the exact case name, allocate new dentry with
2002 * the exact case, and return the spliced entry.
2004 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2007 struct dentry
*found
, *res
;
2010 * First check if a dentry matching the name already exists,
2011 * if not go ahead and create it now.
2013 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2018 if (d_in_lookup(dentry
)) {
2019 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2021 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2026 found
= d_alloc(dentry
->d_parent
, name
);
2029 return ERR_PTR(-ENOMEM
);
2032 res
= d_splice_alias(inode
, found
);
2039 EXPORT_SYMBOL(d_add_ci
);
2042 static inline bool d_same_name(const struct dentry
*dentry
,
2043 const struct dentry
*parent
,
2044 const struct qstr
*name
)
2046 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2047 if (dentry
->d_name
.len
!= name
->len
)
2049 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2051 return parent
->d_op
->d_compare(parent
, dentry
,
2052 dentry
->d_name
.len
, dentry
->d_name
.name
,
2057 * __d_lookup_rcu - search for a dentry (racy, store-free)
2058 * @parent: parent dentry
2059 * @name: qstr of name we wish to find
2060 * @seqp: returns d_seq value at the point where the dentry was found
2061 * Returns: dentry, or NULL
2063 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2064 * resolution (store-free path walking) design described in
2065 * Documentation/filesystems/path-lookup.txt.
2067 * This is not to be used outside core vfs.
2069 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2070 * held, and rcu_read_lock held. The returned dentry must not be stored into
2071 * without taking d_lock and checking d_seq sequence count against @seq
2074 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2077 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2078 * the returned dentry, so long as its parent's seqlock is checked after the
2079 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2080 * is formed, giving integrity down the path walk.
2082 * NOTE! The caller *has* to check the resulting dentry against the sequence
2083 * number we've returned before using any of the resulting dentry state!
2085 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2086 const struct qstr
*name
,
2089 u64 hashlen
= name
->hash_len
;
2090 const unsigned char *str
= name
->name
;
2091 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2092 struct hlist_bl_node
*node
;
2093 struct dentry
*dentry
;
2096 * Note: There is significant duplication with __d_lookup_rcu which is
2097 * required to prevent single threaded performance regressions
2098 * especially on architectures where smp_rmb (in seqcounts) are costly.
2099 * Keep the two functions in sync.
2103 * The hash list is protected using RCU.
2105 * Carefully use d_seq when comparing a candidate dentry, to avoid
2106 * races with d_move().
2108 * It is possible that concurrent renames can mess up our list
2109 * walk here and result in missing our dentry, resulting in the
2110 * false-negative result. d_lookup() protects against concurrent
2111 * renames using rename_lock seqlock.
2113 * See Documentation/filesystems/path-lookup.txt for more details.
2115 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2120 * The dentry sequence count protects us from concurrent
2121 * renames, and thus protects parent and name fields.
2123 * The caller must perform a seqcount check in order
2124 * to do anything useful with the returned dentry.
2126 * NOTE! We do a "raw" seqcount_begin here. That means that
2127 * we don't wait for the sequence count to stabilize if it
2128 * is in the middle of a sequence change. If we do the slow
2129 * dentry compare, we will do seqretries until it is stable,
2130 * and if we end up with a successful lookup, we actually
2131 * want to exit RCU lookup anyway.
2133 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2134 * we are still guaranteed NUL-termination of ->d_name.name.
2136 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2137 if (dentry
->d_parent
!= parent
)
2139 if (d_unhashed(dentry
))
2142 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2145 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2147 tlen
= dentry
->d_name
.len
;
2148 tname
= dentry
->d_name
.name
;
2149 /* we want a consistent (name,len) pair */
2150 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2154 if (parent
->d_op
->d_compare(parent
, dentry
,
2155 tlen
, tname
, name
) != 0)
2158 if (dentry
->d_name
.hash_len
!= hashlen
)
2160 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2170 * d_lookup - search for a dentry
2171 * @parent: parent dentry
2172 * @name: qstr of name we wish to find
2173 * Returns: dentry, or NULL
2175 * d_lookup searches the children of the parent dentry for the name in
2176 * question. If the dentry is found its reference count is incremented and the
2177 * dentry is returned. The caller must use dput to free the entry when it has
2178 * finished using it. %NULL is returned if the dentry does not exist.
2180 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2182 struct dentry
*dentry
;
2186 seq
= read_seqbegin(&rename_lock
);
2187 dentry
= __d_lookup(parent
, name
);
2190 } while (read_seqretry(&rename_lock
, seq
));
2193 EXPORT_SYMBOL(d_lookup
);
2196 * __d_lookup - search for a dentry (racy)
2197 * @parent: parent dentry
2198 * @name: qstr of name we wish to find
2199 * Returns: dentry, or NULL
2201 * __d_lookup is like d_lookup, however it may (rarely) return a
2202 * false-negative result due to unrelated rename activity.
2204 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2205 * however it must be used carefully, eg. with a following d_lookup in
2206 * the case of failure.
2208 * __d_lookup callers must be commented.
2210 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2212 unsigned int hash
= name
->hash
;
2213 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2214 struct hlist_bl_node
*node
;
2215 struct dentry
*found
= NULL
;
2216 struct dentry
*dentry
;
2219 * Note: There is significant duplication with __d_lookup_rcu which is
2220 * required to prevent single threaded performance regressions
2221 * especially on architectures where smp_rmb (in seqcounts) are costly.
2222 * Keep the two functions in sync.
2226 * The hash list is protected using RCU.
2228 * Take d_lock when comparing a candidate dentry, to avoid races
2231 * It is possible that concurrent renames can mess up our list
2232 * walk here and result in missing our dentry, resulting in the
2233 * false-negative result. d_lookup() protects against concurrent
2234 * renames using rename_lock seqlock.
2236 * See Documentation/filesystems/path-lookup.txt for more details.
2240 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2242 if (dentry
->d_name
.hash
!= hash
)
2245 spin_lock(&dentry
->d_lock
);
2246 if (dentry
->d_parent
!= parent
)
2248 if (d_unhashed(dentry
))
2251 if (!d_same_name(dentry
, parent
, name
))
2254 dentry
->d_lockref
.count
++;
2256 spin_unlock(&dentry
->d_lock
);
2259 spin_unlock(&dentry
->d_lock
);
2267 * d_hash_and_lookup - hash the qstr then search for a dentry
2268 * @dir: Directory to search in
2269 * @name: qstr of name we wish to find
2271 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2273 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2276 * Check for a fs-specific hash function. Note that we must
2277 * calculate the standard hash first, as the d_op->d_hash()
2278 * routine may choose to leave the hash value unchanged.
2280 name
->hash
= full_name_hash(name
->name
, name
->len
);
2281 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2282 int err
= dir
->d_op
->d_hash(dir
, name
);
2283 if (unlikely(err
< 0))
2284 return ERR_PTR(err
);
2286 return d_lookup(dir
, name
);
2288 EXPORT_SYMBOL(d_hash_and_lookup
);
2291 * When a file is deleted, we have two options:
2292 * - turn this dentry into a negative dentry
2293 * - unhash this dentry and free it.
2295 * Usually, we want to just turn this into
2296 * a negative dentry, but if anybody else is
2297 * currently using the dentry or the inode
2298 * we can't do that and we fall back on removing
2299 * it from the hash queues and waiting for
2300 * it to be deleted later when it has no users
2304 * d_delete - delete a dentry
2305 * @dentry: The dentry to delete
2307 * Turn the dentry into a negative dentry if possible, otherwise
2308 * remove it from the hash queues so it can be deleted later
2311 void d_delete(struct dentry
* dentry
)
2313 struct inode
*inode
;
2316 * Are we the only user?
2319 spin_lock(&dentry
->d_lock
);
2320 inode
= dentry
->d_inode
;
2321 isdir
= S_ISDIR(inode
->i_mode
);
2322 if (dentry
->d_lockref
.count
== 1) {
2323 if (!spin_trylock(&inode
->i_lock
)) {
2324 spin_unlock(&dentry
->d_lock
);
2328 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2329 dentry_unlink_inode(dentry
);
2330 fsnotify_nameremove(dentry
, isdir
);
2334 if (!d_unhashed(dentry
))
2337 spin_unlock(&dentry
->d_lock
);
2339 fsnotify_nameremove(dentry
, isdir
);
2341 EXPORT_SYMBOL(d_delete
);
2343 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2345 BUG_ON(!d_unhashed(entry
));
2347 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2351 static void _d_rehash(struct dentry
* entry
)
2353 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2357 * d_rehash - add an entry back to the hash
2358 * @entry: dentry to add to the hash
2360 * Adds a dentry to the hash according to its name.
2363 void d_rehash(struct dentry
* entry
)
2365 spin_lock(&entry
->d_lock
);
2367 spin_unlock(&entry
->d_lock
);
2369 EXPORT_SYMBOL(d_rehash
);
2371 static inline unsigned start_dir_add(struct inode
*dir
)
2375 unsigned n
= dir
->i_dir_seq
;
2376 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2382 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2384 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2387 static void d_wait_lookup(struct dentry
*dentry
)
2389 if (d_in_lookup(dentry
)) {
2390 DECLARE_WAITQUEUE(wait
, current
);
2391 add_wait_queue(dentry
->d_wait
, &wait
);
2393 set_current_state(TASK_UNINTERRUPTIBLE
);
2394 spin_unlock(&dentry
->d_lock
);
2396 spin_lock(&dentry
->d_lock
);
2397 } while (d_in_lookup(dentry
));
2401 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2402 const struct qstr
*name
,
2403 wait_queue_head_t
*wq
)
2405 unsigned int hash
= name
->hash
;
2406 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2407 struct hlist_bl_node
*node
;
2408 struct dentry
*new = d_alloc(parent
, name
);
2409 struct dentry
*dentry
;
2410 unsigned seq
, r_seq
, d_seq
;
2413 return ERR_PTR(-ENOMEM
);
2417 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
) & ~1;
2418 r_seq
= read_seqbegin(&rename_lock
);
2419 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2420 if (unlikely(dentry
)) {
2421 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2425 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2434 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2439 if (unlikely(parent
->d_inode
->i_dir_seq
!= seq
)) {
2445 * No changes for the parent since the beginning of d_lookup().
2446 * Since all removals from the chain happen with hlist_bl_lock(),
2447 * any potential in-lookup matches are going to stay here until
2448 * we unlock the chain. All fields are stable in everything
2451 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2452 if (dentry
->d_name
.hash
!= hash
)
2454 if (dentry
->d_parent
!= parent
)
2456 if (!d_same_name(dentry
, parent
, name
))
2459 /* now we can try to grab a reference */
2460 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2467 * somebody is likely to be still doing lookup for it;
2468 * wait for them to finish
2470 spin_lock(&dentry
->d_lock
);
2471 d_wait_lookup(dentry
);
2473 * it's not in-lookup anymore; in principle we should repeat
2474 * everything from dcache lookup, but it's likely to be what
2475 * d_lookup() would've found anyway. If it is, just return it;
2476 * otherwise we really have to repeat the whole thing.
2478 if (unlikely(dentry
->d_name
.hash
!= hash
))
2480 if (unlikely(dentry
->d_parent
!= parent
))
2482 if (unlikely(d_unhashed(dentry
)))
2484 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2486 /* OK, it *is* a hashed match; return it */
2487 spin_unlock(&dentry
->d_lock
);
2492 /* we can't take ->d_lock here; it's OK, though. */
2493 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2495 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2499 spin_unlock(&dentry
->d_lock
);
2503 EXPORT_SYMBOL(d_alloc_parallel
);
2505 void __d_lookup_done(struct dentry
*dentry
)
2507 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2508 dentry
->d_name
.hash
);
2510 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2511 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2512 wake_up_all(dentry
->d_wait
);
2513 dentry
->d_wait
= NULL
;
2515 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2516 INIT_LIST_HEAD(&dentry
->d_lru
);
2518 EXPORT_SYMBOL(__d_lookup_done
);
2520 /* inode->i_lock held if inode is non-NULL */
2522 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2524 struct inode
*dir
= NULL
;
2526 spin_lock(&dentry
->d_lock
);
2527 if (unlikely(d_in_lookup(dentry
))) {
2528 dir
= dentry
->d_parent
->d_inode
;
2529 n
= start_dir_add(dir
);
2530 __d_lookup_done(dentry
);
2533 unsigned add_flags
= d_flags_for_inode(inode
);
2534 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2535 raw_write_seqcount_begin(&dentry
->d_seq
);
2536 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2537 raw_write_seqcount_end(&dentry
->d_seq
);
2538 fsnotify_update_flags(dentry
);
2542 end_dir_add(dir
, n
);
2543 spin_unlock(&dentry
->d_lock
);
2545 spin_unlock(&inode
->i_lock
);
2549 * d_add - add dentry to hash queues
2550 * @entry: dentry to add
2551 * @inode: The inode to attach to this dentry
2553 * This adds the entry to the hash queues and initializes @inode.
2554 * The entry was actually filled in earlier during d_alloc().
2557 void d_add(struct dentry
*entry
, struct inode
*inode
)
2560 security_d_instantiate(entry
, inode
);
2561 spin_lock(&inode
->i_lock
);
2563 __d_add(entry
, inode
);
2565 EXPORT_SYMBOL(d_add
);
2568 * d_exact_alias - find and hash an exact unhashed alias
2569 * @entry: dentry to add
2570 * @inode: The inode to go with this dentry
2572 * If an unhashed dentry with the same name/parent and desired
2573 * inode already exists, hash and return it. Otherwise, return
2576 * Parent directory should be locked.
2578 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2580 struct dentry
*alias
;
2581 unsigned int hash
= entry
->d_name
.hash
;
2583 spin_lock(&inode
->i_lock
);
2584 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2586 * Don't need alias->d_lock here, because aliases with
2587 * d_parent == entry->d_parent are not subject to name or
2588 * parent changes, because the parent inode i_mutex is held.
2590 if (alias
->d_name
.hash
!= hash
)
2592 if (alias
->d_parent
!= entry
->d_parent
)
2594 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2596 spin_lock(&alias
->d_lock
);
2597 if (!d_unhashed(alias
)) {
2598 spin_unlock(&alias
->d_lock
);
2601 __dget_dlock(alias
);
2603 spin_unlock(&alias
->d_lock
);
2605 spin_unlock(&inode
->i_lock
);
2608 spin_unlock(&inode
->i_lock
);
2611 EXPORT_SYMBOL(d_exact_alias
);
2614 * dentry_update_name_case - update case insensitive dentry with a new name
2615 * @dentry: dentry to be updated
2618 * Update a case insensitive dentry with new case of name.
2620 * dentry must have been returned by d_lookup with name @name. Old and new
2621 * name lengths must match (ie. no d_compare which allows mismatched name
2624 * Parent inode i_mutex must be held over d_lookup and into this call (to
2625 * keep renames and concurrent inserts, and readdir(2) away).
2627 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2629 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2630 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2632 spin_lock(&dentry
->d_lock
);
2633 write_seqcount_begin(&dentry
->d_seq
);
2634 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2635 write_seqcount_end(&dentry
->d_seq
);
2636 spin_unlock(&dentry
->d_lock
);
2638 EXPORT_SYMBOL(dentry_update_name_case
);
2640 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2642 if (unlikely(dname_external(target
))) {
2643 if (unlikely(dname_external(dentry
))) {
2645 * Both external: swap the pointers
2647 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2650 * dentry:internal, target:external. Steal target's
2651 * storage and make target internal.
2653 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2654 dentry
->d_name
.len
+ 1);
2655 dentry
->d_name
.name
= target
->d_name
.name
;
2656 target
->d_name
.name
= target
->d_iname
;
2659 if (unlikely(dname_external(dentry
))) {
2661 * dentry:external, target:internal. Give dentry's
2662 * storage to target and make dentry internal
2664 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2665 target
->d_name
.len
+ 1);
2666 target
->d_name
.name
= dentry
->d_name
.name
;
2667 dentry
->d_name
.name
= dentry
->d_iname
;
2670 * Both are internal.
2673 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2674 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2675 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2676 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2677 swap(((long *) &dentry
->d_iname
)[i
],
2678 ((long *) &target
->d_iname
)[i
]);
2682 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2685 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2687 struct external_name
*old_name
= NULL
;
2688 if (unlikely(dname_external(dentry
)))
2689 old_name
= external_name(dentry
);
2690 if (unlikely(dname_external(target
))) {
2691 atomic_inc(&external_name(target
)->u
.count
);
2692 dentry
->d_name
= target
->d_name
;
2694 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2695 target
->d_name
.len
+ 1);
2696 dentry
->d_name
.name
= dentry
->d_iname
;
2697 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2699 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2700 kfree_rcu(old_name
, u
.head
);
2703 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2706 * XXXX: do we really need to take target->d_lock?
2708 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2709 spin_lock(&target
->d_parent
->d_lock
);
2711 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2712 spin_lock(&dentry
->d_parent
->d_lock
);
2713 spin_lock_nested(&target
->d_parent
->d_lock
,
2714 DENTRY_D_LOCK_NESTED
);
2716 spin_lock(&target
->d_parent
->d_lock
);
2717 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2718 DENTRY_D_LOCK_NESTED
);
2721 if (target
< dentry
) {
2722 spin_lock_nested(&target
->d_lock
, 2);
2723 spin_lock_nested(&dentry
->d_lock
, 3);
2725 spin_lock_nested(&dentry
->d_lock
, 2);
2726 spin_lock_nested(&target
->d_lock
, 3);
2730 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2732 if (target
->d_parent
!= dentry
->d_parent
)
2733 spin_unlock(&dentry
->d_parent
->d_lock
);
2734 if (target
->d_parent
!= target
)
2735 spin_unlock(&target
->d_parent
->d_lock
);
2736 spin_unlock(&target
->d_lock
);
2737 spin_unlock(&dentry
->d_lock
);
2741 * When switching names, the actual string doesn't strictly have to
2742 * be preserved in the target - because we're dropping the target
2743 * anyway. As such, we can just do a simple memcpy() to copy over
2744 * the new name before we switch, unless we are going to rehash
2745 * it. Note that if we *do* unhash the target, we are not allowed
2746 * to rehash it without giving it a new name/hash key - whether
2747 * we swap or overwrite the names here, resulting name won't match
2748 * the reality in filesystem; it's only there for d_path() purposes.
2749 * Note that all of this is happening under rename_lock, so the
2750 * any hash lookup seeing it in the middle of manipulations will
2751 * be discarded anyway. So we do not care what happens to the hash
2755 * __d_move - move a dentry
2756 * @dentry: entry to move
2757 * @target: new dentry
2758 * @exchange: exchange the two dentries
2760 * Update the dcache to reflect the move of a file name. Negative
2761 * dcache entries should not be moved in this way. Caller must hold
2762 * rename_lock, the i_mutex of the source and target directories,
2763 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2765 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2768 struct inode
*dir
= NULL
;
2770 if (!dentry
->d_inode
)
2771 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2773 BUG_ON(d_ancestor(dentry
, target
));
2774 BUG_ON(d_ancestor(target
, dentry
));
2776 dentry_lock_for_move(dentry
, target
);
2777 if (unlikely(d_in_lookup(target
))) {
2778 dir
= target
->d_parent
->d_inode
;
2779 n
= start_dir_add(dir
);
2780 __d_lookup_done(target
);
2783 write_seqcount_begin(&dentry
->d_seq
);
2784 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2786 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2789 * Move the dentry to the target hash queue. Don't bother checking
2790 * for the same hash queue because of how unlikely it is.
2793 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2796 * Unhash the target (d_delete() is not usable here). If exchanging
2797 * the two dentries, then rehash onto the other's hash queue.
2802 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2805 /* Switch the names.. */
2807 swap_names(dentry
, target
);
2809 copy_name(dentry
, target
);
2811 /* ... and switch them in the tree */
2812 if (IS_ROOT(dentry
)) {
2813 /* splicing a tree */
2814 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2815 dentry
->d_parent
= target
->d_parent
;
2816 target
->d_parent
= target
;
2817 list_del_init(&target
->d_child
);
2818 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2820 /* swapping two dentries */
2821 swap(dentry
->d_parent
, target
->d_parent
);
2822 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2823 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2825 fsnotify_update_flags(target
);
2826 fsnotify_update_flags(dentry
);
2829 write_seqcount_end(&target
->d_seq
);
2830 write_seqcount_end(&dentry
->d_seq
);
2833 end_dir_add(dir
, n
);
2834 dentry_unlock_for_move(dentry
, target
);
2838 * d_move - move a dentry
2839 * @dentry: entry to move
2840 * @target: new dentry
2842 * Update the dcache to reflect the move of a file name. Negative
2843 * dcache entries should not be moved in this way. See the locking
2844 * requirements for __d_move.
2846 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2848 write_seqlock(&rename_lock
);
2849 __d_move(dentry
, target
, false);
2850 write_sequnlock(&rename_lock
);
2852 EXPORT_SYMBOL(d_move
);
2855 * d_exchange - exchange two dentries
2856 * @dentry1: first dentry
2857 * @dentry2: second dentry
2859 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2861 write_seqlock(&rename_lock
);
2863 WARN_ON(!dentry1
->d_inode
);
2864 WARN_ON(!dentry2
->d_inode
);
2865 WARN_ON(IS_ROOT(dentry1
));
2866 WARN_ON(IS_ROOT(dentry2
));
2868 __d_move(dentry1
, dentry2
, true);
2870 write_sequnlock(&rename_lock
);
2874 * d_ancestor - search for an ancestor
2875 * @p1: ancestor dentry
2878 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2879 * an ancestor of p2, else NULL.
2881 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2885 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2886 if (p
->d_parent
== p1
)
2893 * This helper attempts to cope with remotely renamed directories
2895 * It assumes that the caller is already holding
2896 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2898 * Note: If ever the locking in lock_rename() changes, then please
2899 * remember to update this too...
2901 static int __d_unalias(struct inode
*inode
,
2902 struct dentry
*dentry
, struct dentry
*alias
)
2904 struct mutex
*m1
= NULL
;
2905 struct rw_semaphore
*m2
= NULL
;
2908 /* If alias and dentry share a parent, then no extra locks required */
2909 if (alias
->d_parent
== dentry
->d_parent
)
2912 /* See lock_rename() */
2913 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2915 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2916 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2918 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2920 __d_move(alias
, dentry
, false);
2931 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2932 * @inode: the inode which may have a disconnected dentry
2933 * @dentry: a negative dentry which we want to point to the inode.
2935 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2936 * place of the given dentry and return it, else simply d_add the inode
2937 * to the dentry and return NULL.
2939 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2940 * we should error out: directories can't have multiple aliases.
2942 * This is needed in the lookup routine of any filesystem that is exportable
2943 * (via knfsd) so that we can build dcache paths to directories effectively.
2945 * If a dentry was found and moved, then it is returned. Otherwise NULL
2946 * is returned. This matches the expected return value of ->lookup.
2948 * Cluster filesystems may call this function with a negative, hashed dentry.
2949 * In that case, we know that the inode will be a regular file, and also this
2950 * will only occur during atomic_open. So we need to check for the dentry
2951 * being already hashed only in the final case.
2953 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2956 return ERR_CAST(inode
);
2958 BUG_ON(!d_unhashed(dentry
));
2963 security_d_instantiate(dentry
, inode
);
2964 spin_lock(&inode
->i_lock
);
2965 if (S_ISDIR(inode
->i_mode
)) {
2966 struct dentry
*new = __d_find_any_alias(inode
);
2967 if (unlikely(new)) {
2968 /* The reference to new ensures it remains an alias */
2969 spin_unlock(&inode
->i_lock
);
2970 write_seqlock(&rename_lock
);
2971 if (unlikely(d_ancestor(new, dentry
))) {
2972 write_sequnlock(&rename_lock
);
2974 new = ERR_PTR(-ELOOP
);
2975 pr_warn_ratelimited(
2976 "VFS: Lookup of '%s' in %s %s"
2977 " would have caused loop\n",
2978 dentry
->d_name
.name
,
2979 inode
->i_sb
->s_type
->name
,
2981 } else if (!IS_ROOT(new)) {
2982 int err
= __d_unalias(inode
, dentry
, new);
2983 write_sequnlock(&rename_lock
);
2989 __d_move(new, dentry
, false);
2990 write_sequnlock(&rename_lock
);
2997 __d_add(dentry
, inode
);
3000 EXPORT_SYMBOL(d_splice_alias
);
3002 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3006 return -ENAMETOOLONG
;
3008 memcpy(*buffer
, str
, namelen
);
3013 * prepend_name - prepend a pathname in front of current buffer pointer
3014 * @buffer: buffer pointer
3015 * @buflen: allocated length of the buffer
3016 * @name: name string and length qstr structure
3018 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3019 * make sure that either the old or the new name pointer and length are
3020 * fetched. However, there may be mismatch between length and pointer.
3021 * The length cannot be trusted, we need to copy it byte-by-byte until
3022 * the length is reached or a null byte is found. It also prepends "/" at
3023 * the beginning of the name. The sequence number check at the caller will
3024 * retry it again when a d_move() does happen. So any garbage in the buffer
3025 * due to mismatched pointer and length will be discarded.
3027 * Data dependency barrier is needed to make sure that we see that terminating
3028 * NUL. Alpha strikes again, film at 11...
3030 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
3032 const char *dname
= ACCESS_ONCE(name
->name
);
3033 u32 dlen
= ACCESS_ONCE(name
->len
);
3036 smp_read_barrier_depends();
3038 *buflen
-= dlen
+ 1;
3040 return -ENAMETOOLONG
;
3041 p
= *buffer
-= dlen
+ 1;
3053 * prepend_path - Prepend path string to a buffer
3054 * @path: the dentry/vfsmount to report
3055 * @root: root vfsmnt/dentry
3056 * @buffer: pointer to the end of the buffer
3057 * @buflen: pointer to buffer length
3059 * The function will first try to write out the pathname without taking any
3060 * lock other than the RCU read lock to make sure that dentries won't go away.
3061 * It only checks the sequence number of the global rename_lock as any change
3062 * in the dentry's d_seq will be preceded by changes in the rename_lock
3063 * sequence number. If the sequence number had been changed, it will restart
3064 * the whole pathname back-tracing sequence again by taking the rename_lock.
3065 * In this case, there is no need to take the RCU read lock as the recursive
3066 * parent pointer references will keep the dentry chain alive as long as no
3067 * rename operation is performed.
3069 static int prepend_path(const struct path
*path
,
3070 const struct path
*root
,
3071 char **buffer
, int *buflen
)
3073 struct dentry
*dentry
;
3074 struct vfsmount
*vfsmnt
;
3077 unsigned seq
, m_seq
= 0;
3083 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3090 dentry
= path
->dentry
;
3092 mnt
= real_mount(vfsmnt
);
3093 read_seqbegin_or_lock(&rename_lock
, &seq
);
3094 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3095 struct dentry
* parent
;
3097 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3098 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
3100 if (dentry
!= vfsmnt
->mnt_root
) {
3107 if (mnt
!= parent
) {
3108 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3114 error
= is_mounted(vfsmnt
) ? 1 : 2;
3117 parent
= dentry
->d_parent
;
3119 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3127 if (need_seqretry(&rename_lock
, seq
)) {
3131 done_seqretry(&rename_lock
, seq
);
3135 if (need_seqretry(&mount_lock
, m_seq
)) {
3139 done_seqretry(&mount_lock
, m_seq
);
3141 if (error
>= 0 && bptr
== *buffer
) {
3143 error
= -ENAMETOOLONG
;
3153 * __d_path - return the path of a dentry
3154 * @path: the dentry/vfsmount to report
3155 * @root: root vfsmnt/dentry
3156 * @buf: buffer to return value in
3157 * @buflen: buffer length
3159 * Convert a dentry into an ASCII path name.
3161 * Returns a pointer into the buffer or an error code if the
3162 * path was too long.
3164 * "buflen" should be positive.
3166 * If the path is not reachable from the supplied root, return %NULL.
3168 char *__d_path(const struct path
*path
,
3169 const struct path
*root
,
3170 char *buf
, int buflen
)
3172 char *res
= buf
+ buflen
;
3175 prepend(&res
, &buflen
, "\0", 1);
3176 error
= prepend_path(path
, root
, &res
, &buflen
);
3179 return ERR_PTR(error
);
3185 char *d_absolute_path(const struct path
*path
,
3186 char *buf
, int buflen
)
3188 struct path root
= {};
3189 char *res
= buf
+ buflen
;
3192 prepend(&res
, &buflen
, "\0", 1);
3193 error
= prepend_path(path
, &root
, &res
, &buflen
);
3198 return ERR_PTR(error
);
3203 * same as __d_path but appends "(deleted)" for unlinked files.
3205 static int path_with_deleted(const struct path
*path
,
3206 const struct path
*root
,
3207 char **buf
, int *buflen
)
3209 prepend(buf
, buflen
, "\0", 1);
3210 if (d_unlinked(path
->dentry
)) {
3211 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3216 return prepend_path(path
, root
, buf
, buflen
);
3219 static int prepend_unreachable(char **buffer
, int *buflen
)
3221 return prepend(buffer
, buflen
, "(unreachable)", 13);
3224 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3229 seq
= read_seqcount_begin(&fs
->seq
);
3231 } while (read_seqcount_retry(&fs
->seq
, seq
));
3235 * d_path - return the path of a dentry
3236 * @path: path to report
3237 * @buf: buffer to return value in
3238 * @buflen: buffer length
3240 * Convert a dentry into an ASCII path name. If the entry has been deleted
3241 * the string " (deleted)" is appended. Note that this is ambiguous.
3243 * Returns a pointer into the buffer or an error code if the path was
3244 * too long. Note: Callers should use the returned pointer, not the passed
3245 * in buffer, to use the name! The implementation often starts at an offset
3246 * into the buffer, and may leave 0 bytes at the start.
3248 * "buflen" should be positive.
3250 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3252 char *res
= buf
+ buflen
;
3257 * We have various synthetic filesystems that never get mounted. On
3258 * these filesystems dentries are never used for lookup purposes, and
3259 * thus don't need to be hashed. They also don't need a name until a
3260 * user wants to identify the object in /proc/pid/fd/. The little hack
3261 * below allows us to generate a name for these objects on demand:
3263 * Some pseudo inodes are mountable. When they are mounted
3264 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3265 * and instead have d_path return the mounted path.
3267 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3268 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3269 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3272 get_fs_root_rcu(current
->fs
, &root
);
3273 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3277 res
= ERR_PTR(error
);
3280 EXPORT_SYMBOL(d_path
);
3283 * Helper function for dentry_operations.d_dname() members
3285 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3286 const char *fmt
, ...)
3292 va_start(args
, fmt
);
3293 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3296 if (sz
> sizeof(temp
) || sz
> buflen
)
3297 return ERR_PTR(-ENAMETOOLONG
);
3299 buffer
+= buflen
- sz
;
3300 return memcpy(buffer
, temp
, sz
);
3303 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3305 char *end
= buffer
+ buflen
;
3306 /* these dentries are never renamed, so d_lock is not needed */
3307 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3308 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3309 prepend(&end
, &buflen
, "/", 1))
3310 end
= ERR_PTR(-ENAMETOOLONG
);
3313 EXPORT_SYMBOL(simple_dname
);
3316 * Write full pathname from the root of the filesystem into the buffer.
3318 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3320 struct dentry
*dentry
;
3333 prepend(&end
, &len
, "\0", 1);
3337 read_seqbegin_or_lock(&rename_lock
, &seq
);
3338 while (!IS_ROOT(dentry
)) {
3339 struct dentry
*parent
= dentry
->d_parent
;
3342 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3351 if (need_seqretry(&rename_lock
, seq
)) {
3355 done_seqretry(&rename_lock
, seq
);
3360 return ERR_PTR(-ENAMETOOLONG
);
3363 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3365 return __dentry_path(dentry
, buf
, buflen
);
3367 EXPORT_SYMBOL(dentry_path_raw
);
3369 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3374 if (d_unlinked(dentry
)) {
3376 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3380 retval
= __dentry_path(dentry
, buf
, buflen
);
3381 if (!IS_ERR(retval
) && p
)
3382 *p
= '/'; /* restore '/' overriden with '\0' */
3385 return ERR_PTR(-ENAMETOOLONG
);
3388 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3394 seq
= read_seqcount_begin(&fs
->seq
);
3397 } while (read_seqcount_retry(&fs
->seq
, seq
));
3401 * NOTE! The user-level library version returns a
3402 * character pointer. The kernel system call just
3403 * returns the length of the buffer filled (which
3404 * includes the ending '\0' character), or a negative
3405 * error value. So libc would do something like
3407 * char *getcwd(char * buf, size_t size)
3411 * retval = sys_getcwd(buf, size);
3418 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3421 struct path pwd
, root
;
3422 char *page
= __getname();
3428 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3431 if (!d_unlinked(pwd
.dentry
)) {
3433 char *cwd
= page
+ PATH_MAX
;
3434 int buflen
= PATH_MAX
;
3436 prepend(&cwd
, &buflen
, "\0", 1);
3437 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3443 /* Unreachable from current root */
3445 error
= prepend_unreachable(&cwd
, &buflen
);
3451 len
= PATH_MAX
+ page
- cwd
;
3454 if (copy_to_user(buf
, cwd
, len
))
3467 * Test whether new_dentry is a subdirectory of old_dentry.
3469 * Trivially implemented using the dcache structure
3473 * is_subdir - is new dentry a subdirectory of old_dentry
3474 * @new_dentry: new dentry
3475 * @old_dentry: old dentry
3477 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3478 * Returns false otherwise.
3479 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3482 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3487 if (new_dentry
== old_dentry
)
3491 /* for restarting inner loop in case of seq retry */
3492 seq
= read_seqbegin(&rename_lock
);
3494 * Need rcu_readlock to protect against the d_parent trashing
3498 if (d_ancestor(old_dentry
, new_dentry
))
3503 } while (read_seqretry(&rename_lock
, seq
));
3508 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3510 struct dentry
*root
= data
;
3511 if (dentry
!= root
) {
3512 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3515 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3516 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3517 dentry
->d_lockref
.count
--;
3520 return D_WALK_CONTINUE
;
3523 void d_genocide(struct dentry
*parent
)
3525 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3528 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3530 inode_dec_link_count(inode
);
3531 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3532 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3533 !d_unlinked(dentry
));
3534 spin_lock(&dentry
->d_parent
->d_lock
);
3535 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3536 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3537 (unsigned long long)inode
->i_ino
);
3538 spin_unlock(&dentry
->d_lock
);
3539 spin_unlock(&dentry
->d_parent
->d_lock
);
3540 d_instantiate(dentry
, inode
);
3542 EXPORT_SYMBOL(d_tmpfile
);
3544 static __initdata
unsigned long dhash_entries
;
3545 static int __init
set_dhash_entries(char *str
)
3549 dhash_entries
= simple_strtoul(str
, &str
, 0);
3552 __setup("dhash_entries=", set_dhash_entries
);
3554 static void __init
dcache_init_early(void)
3558 /* If hashes are distributed across NUMA nodes, defer
3559 * hash allocation until vmalloc space is available.
3565 alloc_large_system_hash("Dentry cache",
3566 sizeof(struct hlist_bl_head
),
3575 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3576 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3579 static void __init
dcache_init(void)
3584 * A constructor could be added for stable state like the lists,
3585 * but it is probably not worth it because of the cache nature
3588 dentry_cache
= KMEM_CACHE(dentry
,
3589 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3591 /* Hash may have been set up in dcache_init_early */
3596 alloc_large_system_hash("Dentry cache",
3597 sizeof(struct hlist_bl_head
),
3606 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3607 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3610 /* SLAB cache for __getname() consumers */
3611 struct kmem_cache
*names_cachep __read_mostly
;
3612 EXPORT_SYMBOL(names_cachep
);
3614 EXPORT_SYMBOL(d_genocide
);
3616 void __init
vfs_caches_init_early(void)
3618 dcache_init_early();
3622 void __init
vfs_caches_init(void)
3624 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3625 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3630 files_maxfiles_init();