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 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat
= {
119 static DEFINE_PER_CPU(long, nr_dentry
);
120 static DEFINE_PER_CPU(long, nr_dentry_unused
);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
125 * Here we resort to our own counters instead of using generic per-cpu counters
126 * for consistency with what the vfs inode code does. We are expected to harvest
127 * better code and performance by having our own specialized counters.
129 * Please note that the loop is done over all possible CPUs, not over all online
130 * CPUs. The reason for this is that we don't want to play games with CPUs going
131 * on and off. If one of them goes off, we will just keep their counters.
133 * glommer: See cffbc8a for details, and if you ever intend to change this,
134 * please update all vfs counters to match.
136 static long get_nr_dentry(void)
140 for_each_possible_cpu(i
)
141 sum
+= per_cpu(nr_dentry
, i
);
142 return sum
< 0 ? 0 : sum
;
145 static long get_nr_dentry_unused(void)
149 for_each_possible_cpu(i
)
150 sum
+= per_cpu(nr_dentry_unused
, i
);
151 return sum
< 0 ? 0 : sum
;
154 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
155 size_t *lenp
, loff_t
*ppos
)
157 dentry_stat
.nr_dentry
= get_nr_dentry();
158 dentry_stat
.nr_unused
= get_nr_dentry_unused();
159 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
164 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165 * The strings are both count bytes long, and count is non-zero.
167 #ifdef CONFIG_DCACHE_WORD_ACCESS
169 #include <asm/word-at-a-time.h>
171 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172 * aligned allocation for this particular component. We don't
173 * strictly need the load_unaligned_zeropad() safety, but it
174 * doesn't hurt either.
176 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177 * need the careful unaligned handling.
179 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
181 unsigned long a
,b
,mask
;
184 a
= *(unsigned long *)cs
;
185 b
= load_unaligned_zeropad(ct
);
186 if (tcount
< sizeof(unsigned long))
188 if (unlikely(a
!= b
))
190 cs
+= sizeof(unsigned long);
191 ct
+= sizeof(unsigned long);
192 tcount
-= sizeof(unsigned long);
196 mask
= bytemask_from_count(tcount
);
197 return unlikely(!!((a
^ b
) & mask
));
202 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
216 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
218 const unsigned char *cs
;
220 * Be careful about RCU walk racing with rename:
221 * use ACCESS_ONCE to fetch the name pointer.
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
235 cs
= ACCESS_ONCE(dentry
->d_name
.name
);
236 smp_read_barrier_depends();
237 return dentry_string_cmp(cs
, ct
, tcount
);
240 struct external_name
{
243 struct rcu_head head
;
245 unsigned char name
[];
248 static inline struct external_name
*external_name(struct dentry
*dentry
)
250 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
253 static void __d_free(struct rcu_head
*head
)
255 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
257 kmem_cache_free(dentry_cache
, dentry
);
260 static void __d_free_external(struct rcu_head
*head
)
262 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
263 kfree(external_name(dentry
));
264 kmem_cache_free(dentry_cache
, dentry
);
267 static inline int dname_external(const struct dentry
*dentry
)
269 return dentry
->d_name
.name
!= dentry
->d_iname
;
272 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
274 spin_lock(&dentry
->d_lock
);
275 if (unlikely(dname_external(dentry
))) {
276 struct external_name
*p
= external_name(dentry
);
277 atomic_inc(&p
->u
.count
);
278 spin_unlock(&dentry
->d_lock
);
279 name
->name
= p
->name
;
281 memcpy(name
->inline_name
, dentry
->d_iname
,
282 dentry
->d_name
.len
+ 1);
283 spin_unlock(&dentry
->d_lock
);
284 name
->name
= name
->inline_name
;
287 EXPORT_SYMBOL(take_dentry_name_snapshot
);
289 void release_dentry_name_snapshot(struct name_snapshot
*name
)
291 if (unlikely(name
->name
!= name
->inline_name
)) {
292 struct external_name
*p
;
293 p
= container_of(name
->name
, struct external_name
, name
[0]);
294 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
295 kfree_rcu(p
, u
.head
);
298 EXPORT_SYMBOL(release_dentry_name_snapshot
);
300 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
306 dentry
->d_inode
= inode
;
307 flags
= READ_ONCE(dentry
->d_flags
);
308 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
310 WRITE_ONCE(dentry
->d_flags
, flags
);
313 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
315 unsigned flags
= READ_ONCE(dentry
->d_flags
);
317 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
318 WRITE_ONCE(dentry
->d_flags
, flags
);
319 dentry
->d_inode
= NULL
;
322 static void dentry_free(struct dentry
*dentry
)
324 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
325 if (unlikely(dname_external(dentry
))) {
326 struct external_name
*p
= external_name(dentry
);
327 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
328 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
332 /* if dentry was never visible to RCU, immediate free is OK */
333 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
334 __d_free(&dentry
->d_u
.d_rcu
);
336 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
340 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
341 * @dentry: the target dentry
342 * After this call, in-progress rcu-walk path lookup will fail. This
343 * should be called after unhashing, and after changing d_inode (if
344 * the dentry has not already been unhashed).
346 static inline void dentry_rcuwalk_invalidate(struct dentry
*dentry
)
348 lockdep_assert_held(&dentry
->d_lock
);
349 /* Go through am invalidation barrier */
350 write_seqcount_invalidate(&dentry
->d_seq
);
354 * Release the dentry's inode, using the filesystem
355 * d_iput() operation if defined. Dentry has no refcount
358 static void dentry_iput(struct dentry
* dentry
)
359 __releases(dentry
->d_lock
)
360 __releases(dentry
->d_inode
->i_lock
)
362 struct inode
*inode
= dentry
->d_inode
;
364 __d_clear_type_and_inode(dentry
);
365 hlist_del_init(&dentry
->d_u
.d_alias
);
366 spin_unlock(&dentry
->d_lock
);
367 spin_unlock(&inode
->i_lock
);
369 fsnotify_inoderemove(inode
);
370 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
371 dentry
->d_op
->d_iput(dentry
, inode
);
375 spin_unlock(&dentry
->d_lock
);
380 * Release the dentry's inode, using the filesystem
381 * d_iput() operation if defined. dentry remains in-use.
383 static void dentry_unlink_inode(struct dentry
* dentry
)
384 __releases(dentry
->d_lock
)
385 __releases(dentry
->d_inode
->i_lock
)
387 struct inode
*inode
= dentry
->d_inode
;
389 raw_write_seqcount_begin(&dentry
->d_seq
);
390 __d_clear_type_and_inode(dentry
);
391 hlist_del_init(&dentry
->d_u
.d_alias
);
392 raw_write_seqcount_end(&dentry
->d_seq
);
393 spin_unlock(&dentry
->d_lock
);
394 spin_unlock(&inode
->i_lock
);
396 fsnotify_inoderemove(inode
);
397 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
398 dentry
->d_op
->d_iput(dentry
, inode
);
404 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
405 * is in use - which includes both the "real" per-superblock
406 * LRU list _and_ the DCACHE_SHRINK_LIST use.
408 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
409 * on the shrink list (ie not on the superblock LRU list).
411 * The per-cpu "nr_dentry_unused" counters are updated with
412 * the DCACHE_LRU_LIST bit.
414 * These helper functions make sure we always follow the
415 * rules. d_lock must be held by the caller.
417 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
418 static void d_lru_add(struct dentry
*dentry
)
420 D_FLAG_VERIFY(dentry
, 0);
421 dentry
->d_flags
|= DCACHE_LRU_LIST
;
422 this_cpu_inc(nr_dentry_unused
);
423 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
426 static void d_lru_del(struct dentry
*dentry
)
428 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
429 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
430 this_cpu_dec(nr_dentry_unused
);
431 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
434 static void d_shrink_del(struct dentry
*dentry
)
436 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
437 list_del_init(&dentry
->d_lru
);
438 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
439 this_cpu_dec(nr_dentry_unused
);
442 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
444 D_FLAG_VERIFY(dentry
, 0);
445 list_add(&dentry
->d_lru
, list
);
446 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
447 this_cpu_inc(nr_dentry_unused
);
451 * These can only be called under the global LRU lock, ie during the
452 * callback for freeing the LRU list. "isolate" removes it from the
453 * LRU lists entirely, while shrink_move moves it to the indicated
456 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
458 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
459 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
460 this_cpu_dec(nr_dentry_unused
);
461 list_lru_isolate(lru
, &dentry
->d_lru
);
464 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
465 struct list_head
*list
)
467 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
468 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
469 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
473 * dentry_lru_(add|del)_list) must be called with d_lock held.
475 static void dentry_lru_add(struct dentry
*dentry
)
477 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
482 * d_drop - drop a dentry
483 * @dentry: dentry to drop
485 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
486 * be found through a VFS lookup any more. Note that this is different from
487 * deleting the dentry - d_delete will try to mark the dentry negative if
488 * possible, giving a successful _negative_ lookup, while d_drop will
489 * just make the cache lookup fail.
491 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
492 * reason (NFS timeouts or autofs deletes).
494 * __d_drop requires dentry->d_lock.
496 void __d_drop(struct dentry
*dentry
)
498 if (!d_unhashed(dentry
)) {
499 struct hlist_bl_head
*b
;
501 * Hashed dentries are normally on the dentry hashtable,
502 * with the exception of those newly allocated by
503 * d_obtain_alias, which are always IS_ROOT:
505 if (unlikely(IS_ROOT(dentry
)))
506 b
= &dentry
->d_sb
->s_anon
;
508 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
511 __hlist_bl_del(&dentry
->d_hash
);
512 dentry
->d_hash
.pprev
= NULL
;
514 dentry_rcuwalk_invalidate(dentry
);
517 EXPORT_SYMBOL(__d_drop
);
519 void d_drop(struct dentry
*dentry
)
521 spin_lock(&dentry
->d_lock
);
523 spin_unlock(&dentry
->d_lock
);
525 EXPORT_SYMBOL(d_drop
);
527 static void __dentry_kill(struct dentry
*dentry
)
529 struct dentry
*parent
= NULL
;
530 bool can_free
= true;
531 if (!IS_ROOT(dentry
))
532 parent
= dentry
->d_parent
;
535 * The dentry is now unrecoverably dead to the world.
537 lockref_mark_dead(&dentry
->d_lockref
);
540 * inform the fs via d_prune that this dentry is about to be
541 * unhashed and destroyed.
543 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
544 dentry
->d_op
->d_prune(dentry
);
546 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
547 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
550 /* if it was on the hash then remove it */
552 __list_del_entry(&dentry
->d_child
);
554 * Inform d_walk() that we are no longer attached to the
557 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
559 spin_unlock(&parent
->d_lock
);
562 * dentry_iput drops the locks, at which point nobody (except
563 * transient RCU lookups) can reach this dentry.
565 BUG_ON(dentry
->d_lockref
.count
> 0);
566 this_cpu_dec(nr_dentry
);
567 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
568 dentry
->d_op
->d_release(dentry
);
570 spin_lock(&dentry
->d_lock
);
571 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
572 dentry
->d_flags
|= DCACHE_MAY_FREE
;
575 spin_unlock(&dentry
->d_lock
);
576 if (likely(can_free
))
581 * Finish off a dentry we've decided to kill.
582 * dentry->d_lock must be held, returns with it unlocked.
583 * If ref is non-zero, then decrement the refcount too.
584 * Returns dentry requiring refcount drop, or NULL if we're done.
586 static struct dentry
*dentry_kill(struct dentry
*dentry
)
587 __releases(dentry
->d_lock
)
589 struct inode
*inode
= dentry
->d_inode
;
590 struct dentry
*parent
= NULL
;
592 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
595 if (!IS_ROOT(dentry
)) {
596 parent
= dentry
->d_parent
;
597 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
599 spin_unlock(&inode
->i_lock
);
604 __dentry_kill(dentry
);
608 spin_unlock(&dentry
->d_lock
);
609 return dentry
; /* try again with same dentry */
612 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
614 struct dentry
*parent
= dentry
->d_parent
;
617 if (unlikely(dentry
->d_lockref
.count
< 0))
619 if (likely(spin_trylock(&parent
->d_lock
)))
622 spin_unlock(&dentry
->d_lock
);
624 parent
= ACCESS_ONCE(dentry
->d_parent
);
625 spin_lock(&parent
->d_lock
);
627 * We can't blindly lock dentry until we are sure
628 * that we won't violate the locking order.
629 * Any changes of dentry->d_parent must have
630 * been done with parent->d_lock held, so
631 * spin_lock() above is enough of a barrier
632 * for checking if it's still our child.
634 if (unlikely(parent
!= dentry
->d_parent
)) {
635 spin_unlock(&parent
->d_lock
);
638 if (parent
!= dentry
) {
639 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
640 if (unlikely(dentry
->d_lockref
.count
< 0)) {
641 spin_unlock(&parent
->d_lock
);
652 * Try to do a lockless dput(), and return whether that was successful.
654 * If unsuccessful, we return false, having already taken the dentry lock.
656 * The caller needs to hold the RCU read lock, so that the dentry is
657 * guaranteed to stay around even if the refcount goes down to zero!
659 static inline bool fast_dput(struct dentry
*dentry
)
662 unsigned int d_flags
;
665 * If we have a d_op->d_delete() operation, we sould not
666 * let the dentry count go to zero, so use "put_or_lock".
668 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
669 return lockref_put_or_lock(&dentry
->d_lockref
);
672 * .. otherwise, we can try to just decrement the
673 * lockref optimistically.
675 ret
= lockref_put_return(&dentry
->d_lockref
);
678 * If the lockref_put_return() failed due to the lock being held
679 * by somebody else, the fast path has failed. We will need to
680 * get the lock, and then check the count again.
682 if (unlikely(ret
< 0)) {
683 spin_lock(&dentry
->d_lock
);
684 if (dentry
->d_lockref
.count
> 1) {
685 dentry
->d_lockref
.count
--;
686 spin_unlock(&dentry
->d_lock
);
693 * If we weren't the last ref, we're done.
699 * Careful, careful. The reference count went down
700 * to zero, but we don't hold the dentry lock, so
701 * somebody else could get it again, and do another
702 * dput(), and we need to not race with that.
704 * However, there is a very special and common case
705 * where we don't care, because there is nothing to
706 * do: the dentry is still hashed, it does not have
707 * a 'delete' op, and it's referenced and already on
710 * NOTE! Since we aren't locked, these values are
711 * not "stable". However, it is sufficient that at
712 * some point after we dropped the reference the
713 * dentry was hashed and the flags had the proper
714 * value. Other dentry users may have re-gotten
715 * a reference to the dentry and change that, but
716 * our work is done - we can leave the dentry
717 * around with a zero refcount.
720 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
721 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
723 /* Nothing to do? Dropping the reference was all we needed? */
724 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
728 * Not the fast normal case? Get the lock. We've already decremented
729 * the refcount, but we'll need to re-check the situation after
732 spin_lock(&dentry
->d_lock
);
735 * Did somebody else grab a reference to it in the meantime, and
736 * we're no longer the last user after all? Alternatively, somebody
737 * else could have killed it and marked it dead. Either way, we
738 * don't need to do anything else.
740 if (dentry
->d_lockref
.count
) {
741 spin_unlock(&dentry
->d_lock
);
746 * Re-get the reference we optimistically dropped. We hold the
747 * lock, and we just tested that it was zero, so we can just
750 dentry
->d_lockref
.count
= 1;
758 * This is complicated by the fact that we do not want to put
759 * dentries that are no longer on any hash chain on the unused
760 * list: we'd much rather just get rid of them immediately.
762 * However, that implies that we have to traverse the dentry
763 * tree upwards to the parents which might _also_ now be
764 * scheduled for deletion (it may have been only waiting for
765 * its last child to go away).
767 * This tail recursion is done by hand as we don't want to depend
768 * on the compiler to always get this right (gcc generally doesn't).
769 * Real recursion would eat up our stack space.
773 * dput - release a dentry
774 * @dentry: dentry to release
776 * Release a dentry. This will drop the usage count and if appropriate
777 * call the dentry unlink method as well as removing it from the queues and
778 * releasing its resources. If the parent dentries were scheduled for release
779 * they too may now get deleted.
781 void dput(struct dentry
*dentry
)
783 if (unlikely(!dentry
))
790 if (likely(fast_dput(dentry
))) {
795 /* Slow case: now with the dentry lock held */
798 /* Unreachable? Get rid of it */
799 if (unlikely(d_unhashed(dentry
)))
802 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
805 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
806 if (dentry
->d_op
->d_delete(dentry
))
810 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
811 dentry
->d_flags
|= DCACHE_REFERENCED
;
812 dentry_lru_add(dentry
);
814 dentry
->d_lockref
.count
--;
815 spin_unlock(&dentry
->d_lock
);
819 dentry
= dentry_kill(dentry
);
828 /* This must be called with d_lock held */
829 static inline void __dget_dlock(struct dentry
*dentry
)
831 dentry
->d_lockref
.count
++;
834 static inline void __dget(struct dentry
*dentry
)
836 lockref_get(&dentry
->d_lockref
);
839 struct dentry
*dget_parent(struct dentry
*dentry
)
845 * Do optimistic parent lookup without any
849 ret
= ACCESS_ONCE(dentry
->d_parent
);
850 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
852 if (likely(gotref
)) {
853 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
860 * Don't need rcu_dereference because we re-check it was correct under
864 ret
= dentry
->d_parent
;
865 spin_lock(&ret
->d_lock
);
866 if (unlikely(ret
!= dentry
->d_parent
)) {
867 spin_unlock(&ret
->d_lock
);
872 BUG_ON(!ret
->d_lockref
.count
);
873 ret
->d_lockref
.count
++;
874 spin_unlock(&ret
->d_lock
);
877 EXPORT_SYMBOL(dget_parent
);
880 * d_find_alias - grab a hashed alias of inode
881 * @inode: inode in question
883 * If inode has a hashed alias, or is a directory and has any alias,
884 * acquire the reference to alias and return it. Otherwise return NULL.
885 * Notice that if inode is a directory there can be only one alias and
886 * it can be unhashed only if it has no children, or if it is the root
887 * of a filesystem, or if the directory was renamed and d_revalidate
888 * was the first vfs operation to notice.
890 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
891 * any other hashed alias over that one.
893 static struct dentry
*__d_find_alias(struct inode
*inode
)
895 struct dentry
*alias
, *discon_alias
;
899 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
900 spin_lock(&alias
->d_lock
);
901 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
902 if (IS_ROOT(alias
) &&
903 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
904 discon_alias
= alias
;
907 spin_unlock(&alias
->d_lock
);
911 spin_unlock(&alias
->d_lock
);
914 alias
= discon_alias
;
915 spin_lock(&alias
->d_lock
);
916 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
918 spin_unlock(&alias
->d_lock
);
921 spin_unlock(&alias
->d_lock
);
927 struct dentry
*d_find_alias(struct inode
*inode
)
929 struct dentry
*de
= NULL
;
931 if (!hlist_empty(&inode
->i_dentry
)) {
932 spin_lock(&inode
->i_lock
);
933 de
= __d_find_alias(inode
);
934 spin_unlock(&inode
->i_lock
);
938 EXPORT_SYMBOL(d_find_alias
);
941 * Try to kill dentries associated with this inode.
942 * WARNING: you must own a reference to inode.
944 void d_prune_aliases(struct inode
*inode
)
946 struct dentry
*dentry
;
948 spin_lock(&inode
->i_lock
);
949 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
950 spin_lock(&dentry
->d_lock
);
951 if (!dentry
->d_lockref
.count
) {
952 struct dentry
*parent
= lock_parent(dentry
);
953 if (likely(!dentry
->d_lockref
.count
)) {
954 __dentry_kill(dentry
);
959 spin_unlock(&parent
->d_lock
);
961 spin_unlock(&dentry
->d_lock
);
963 spin_unlock(&inode
->i_lock
);
965 EXPORT_SYMBOL(d_prune_aliases
);
967 static void shrink_dentry_list(struct list_head
*list
)
969 struct dentry
*dentry
, *parent
;
971 while (!list_empty(list
)) {
973 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
974 spin_lock(&dentry
->d_lock
);
975 parent
= lock_parent(dentry
);
978 * The dispose list is isolated and dentries are not accounted
979 * to the LRU here, so we can simply remove it from the list
980 * here regardless of whether it is referenced or not.
982 d_shrink_del(dentry
);
985 * We found an inuse dentry which was not removed from
986 * the LRU because of laziness during lookup. Do not free it.
988 if (dentry
->d_lockref
.count
> 0) {
989 spin_unlock(&dentry
->d_lock
);
991 spin_unlock(&parent
->d_lock
);
996 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
997 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
998 spin_unlock(&dentry
->d_lock
);
1000 spin_unlock(&parent
->d_lock
);
1002 dentry_free(dentry
);
1006 inode
= dentry
->d_inode
;
1007 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1008 d_shrink_add(dentry
, list
);
1009 spin_unlock(&dentry
->d_lock
);
1011 spin_unlock(&parent
->d_lock
);
1015 __dentry_kill(dentry
);
1018 * We need to prune ancestors too. This is necessary to prevent
1019 * quadratic behavior of shrink_dcache_parent(), but is also
1020 * expected to be beneficial in reducing dentry cache
1024 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1025 parent
= lock_parent(dentry
);
1026 if (dentry
->d_lockref
.count
!= 1) {
1027 dentry
->d_lockref
.count
--;
1028 spin_unlock(&dentry
->d_lock
);
1030 spin_unlock(&parent
->d_lock
);
1033 inode
= dentry
->d_inode
; /* can't be NULL */
1034 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1035 spin_unlock(&dentry
->d_lock
);
1037 spin_unlock(&parent
->d_lock
);
1041 __dentry_kill(dentry
);
1047 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1048 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1050 struct list_head
*freeable
= arg
;
1051 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1055 * we are inverting the lru lock/dentry->d_lock here,
1056 * so use a trylock. If we fail to get the lock, just skip
1059 if (!spin_trylock(&dentry
->d_lock
))
1063 * Referenced dentries are still in use. If they have active
1064 * counts, just remove them from the LRU. Otherwise give them
1065 * another pass through the LRU.
1067 if (dentry
->d_lockref
.count
) {
1068 d_lru_isolate(lru
, dentry
);
1069 spin_unlock(&dentry
->d_lock
);
1073 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1074 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1075 spin_unlock(&dentry
->d_lock
);
1078 * The list move itself will be made by the common LRU code. At
1079 * this point, we've dropped the dentry->d_lock but keep the
1080 * lru lock. This is safe to do, since every list movement is
1081 * protected by the lru lock even if both locks are held.
1083 * This is guaranteed by the fact that all LRU management
1084 * functions are intermediated by the LRU API calls like
1085 * list_lru_add and list_lru_del. List movement in this file
1086 * only ever occur through this functions or through callbacks
1087 * like this one, that are called from the LRU API.
1089 * The only exceptions to this are functions like
1090 * shrink_dentry_list, and code that first checks for the
1091 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1092 * operating only with stack provided lists after they are
1093 * properly isolated from the main list. It is thus, always a
1099 d_lru_shrink_move(lru
, dentry
, freeable
);
1100 spin_unlock(&dentry
->d_lock
);
1106 * prune_dcache_sb - shrink the dcache
1108 * @sc: shrink control, passed to list_lru_shrink_walk()
1110 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1111 * is done when we need more memory and called from the superblock shrinker
1114 * This function may fail to free any resources if all the dentries are in
1117 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1122 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1123 dentry_lru_isolate
, &dispose
);
1124 shrink_dentry_list(&dispose
);
1128 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1129 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1131 struct list_head
*freeable
= arg
;
1132 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1135 * we are inverting the lru lock/dentry->d_lock here,
1136 * so use a trylock. If we fail to get the lock, just skip
1139 if (!spin_trylock(&dentry
->d_lock
))
1142 d_lru_shrink_move(lru
, dentry
, freeable
);
1143 spin_unlock(&dentry
->d_lock
);
1150 * shrink_dcache_sb - shrink dcache for a superblock
1153 * Shrink the dcache for the specified super block. This is used to free
1154 * the dcache before unmounting a file system.
1156 void shrink_dcache_sb(struct super_block
*sb
)
1163 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1164 dentry_lru_isolate_shrink
, &dispose
, 1024);
1166 this_cpu_sub(nr_dentry_unused
, freed
);
1167 shrink_dentry_list(&dispose
);
1169 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1171 EXPORT_SYMBOL(shrink_dcache_sb
);
1174 * enum d_walk_ret - action to talke during tree walk
1175 * @D_WALK_CONTINUE: contrinue walk
1176 * @D_WALK_QUIT: quit walk
1177 * @D_WALK_NORETRY: quit when retry is needed
1178 * @D_WALK_SKIP: skip this dentry and its children
1188 * d_walk - walk the dentry tree
1189 * @parent: start of walk
1190 * @data: data passed to @enter() and @finish()
1191 * @enter: callback when first entering the dentry
1192 * @finish: callback when successfully finished the walk
1194 * The @enter() and @finish() callbacks are called with d_lock held.
1196 static void d_walk(struct dentry
*parent
, void *data
,
1197 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1198 void (*finish
)(void *))
1200 struct dentry
*this_parent
;
1201 struct list_head
*next
;
1203 enum d_walk_ret ret
;
1207 read_seqbegin_or_lock(&rename_lock
, &seq
);
1208 this_parent
= parent
;
1209 spin_lock(&this_parent
->d_lock
);
1211 ret
= enter(data
, this_parent
);
1213 case D_WALK_CONTINUE
:
1218 case D_WALK_NORETRY
:
1223 next
= this_parent
->d_subdirs
.next
;
1225 while (next
!= &this_parent
->d_subdirs
) {
1226 struct list_head
*tmp
= next
;
1227 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1230 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1232 ret
= enter(data
, dentry
);
1234 case D_WALK_CONTINUE
:
1237 spin_unlock(&dentry
->d_lock
);
1239 case D_WALK_NORETRY
:
1243 spin_unlock(&dentry
->d_lock
);
1247 if (!list_empty(&dentry
->d_subdirs
)) {
1248 spin_unlock(&this_parent
->d_lock
);
1249 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1250 this_parent
= dentry
;
1251 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1254 spin_unlock(&dentry
->d_lock
);
1257 * All done at this level ... ascend and resume the search.
1261 if (this_parent
!= parent
) {
1262 struct dentry
*child
= this_parent
;
1263 this_parent
= child
->d_parent
;
1265 spin_unlock(&child
->d_lock
);
1266 spin_lock(&this_parent
->d_lock
);
1268 /* might go back up the wrong parent if we have had a rename. */
1269 if (need_seqretry(&rename_lock
, seq
))
1271 /* go into the first sibling still alive */
1273 next
= child
->d_child
.next
;
1274 if (next
== &this_parent
->d_subdirs
)
1276 child
= list_entry(next
, struct dentry
, d_child
);
1277 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1281 if (need_seqretry(&rename_lock
, seq
))
1288 spin_unlock(&this_parent
->d_lock
);
1289 done_seqretry(&rename_lock
, seq
);
1293 spin_unlock(&this_parent
->d_lock
);
1303 * Search for at least 1 mount point in the dentry's subdirs.
1304 * We descend to the next level whenever the d_subdirs
1305 * list is non-empty and continue searching.
1308 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1311 if (d_mountpoint(dentry
)) {
1315 return D_WALK_CONTINUE
;
1319 * have_submounts - check for mounts over a dentry
1320 * @parent: dentry to check.
1322 * Return true if the parent or its subdirectories contain
1325 int have_submounts(struct dentry
*parent
)
1329 d_walk(parent
, &ret
, check_mount
, NULL
);
1333 EXPORT_SYMBOL(have_submounts
);
1336 * Called by mount code to set a mountpoint and check if the mountpoint is
1337 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1338 * subtree can become unreachable).
1340 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1341 * this reason take rename_lock and d_lock on dentry and ancestors.
1343 int d_set_mounted(struct dentry
*dentry
)
1347 write_seqlock(&rename_lock
);
1348 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1349 /* Need exclusion wrt. d_invalidate() */
1350 spin_lock(&p
->d_lock
);
1351 if (unlikely(d_unhashed(p
))) {
1352 spin_unlock(&p
->d_lock
);
1355 spin_unlock(&p
->d_lock
);
1357 spin_lock(&dentry
->d_lock
);
1358 if (!d_unlinked(dentry
)) {
1360 if (!d_mountpoint(dentry
)) {
1361 dentry
->d_flags
|= DCACHE_MOUNTED
;
1365 spin_unlock(&dentry
->d_lock
);
1367 write_sequnlock(&rename_lock
);
1372 * Search the dentry child list of the specified parent,
1373 * and move any unused dentries to the end of the unused
1374 * list for prune_dcache(). We descend to the next level
1375 * whenever the d_subdirs list is non-empty and continue
1378 * It returns zero iff there are no unused children,
1379 * otherwise it returns the number of children moved to
1380 * the end of the unused list. This may not be the total
1381 * number of unused children, because select_parent can
1382 * drop the lock and return early due to latency
1386 struct select_data
{
1387 struct dentry
*start
;
1388 struct list_head dispose
;
1392 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1394 struct select_data
*data
= _data
;
1395 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1397 if (data
->start
== dentry
)
1400 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1403 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1405 if (!dentry
->d_lockref
.count
) {
1406 d_shrink_add(dentry
, &data
->dispose
);
1411 * We can return to the caller if we have found some (this
1412 * ensures forward progress). We'll be coming back to find
1415 if (!list_empty(&data
->dispose
))
1416 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1422 * shrink_dcache_parent - prune dcache
1423 * @parent: parent of entries to prune
1425 * Prune the dcache to remove unused children of the parent dentry.
1427 void shrink_dcache_parent(struct dentry
*parent
)
1430 struct select_data data
;
1432 INIT_LIST_HEAD(&data
.dispose
);
1433 data
.start
= parent
;
1436 d_walk(parent
, &data
, select_collect
, NULL
);
1440 shrink_dentry_list(&data
.dispose
);
1444 EXPORT_SYMBOL(shrink_dcache_parent
);
1446 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1448 /* it has busy descendents; complain about those instead */
1449 if (!list_empty(&dentry
->d_subdirs
))
1450 return D_WALK_CONTINUE
;
1452 /* root with refcount 1 is fine */
1453 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1454 return D_WALK_CONTINUE
;
1456 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1457 " still in use (%d) [unmount of %s %s]\n",
1460 dentry
->d_inode
->i_ino
: 0UL,
1462 dentry
->d_lockref
.count
,
1463 dentry
->d_sb
->s_type
->name
,
1464 dentry
->d_sb
->s_id
);
1466 return D_WALK_CONTINUE
;
1469 static void do_one_tree(struct dentry
*dentry
)
1471 shrink_dcache_parent(dentry
);
1472 d_walk(dentry
, dentry
, umount_check
, NULL
);
1478 * destroy the dentries attached to a superblock on unmounting
1480 void shrink_dcache_for_umount(struct super_block
*sb
)
1482 struct dentry
*dentry
;
1484 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1486 dentry
= sb
->s_root
;
1488 do_one_tree(dentry
);
1490 while (!hlist_bl_empty(&sb
->s_anon
)) {
1491 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1492 do_one_tree(dentry
);
1496 struct detach_data
{
1497 struct select_data select
;
1498 struct dentry
*mountpoint
;
1500 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1502 struct detach_data
*data
= _data
;
1504 if (d_mountpoint(dentry
)) {
1505 __dget_dlock(dentry
);
1506 data
->mountpoint
= dentry
;
1510 return select_collect(&data
->select
, dentry
);
1513 static void check_and_drop(void *_data
)
1515 struct detach_data
*data
= _data
;
1517 if (!data
->mountpoint
&& !data
->select
.found
)
1518 __d_drop(data
->select
.start
);
1522 * d_invalidate - detach submounts, prune dcache, and drop
1523 * @dentry: dentry to invalidate (aka detach, prune and drop)
1527 * The final d_drop is done as an atomic operation relative to
1528 * rename_lock ensuring there are no races with d_set_mounted. This
1529 * ensures there are no unhashed dentries on the path to a mountpoint.
1531 void d_invalidate(struct dentry
*dentry
)
1534 * If it's already been dropped, return OK.
1536 spin_lock(&dentry
->d_lock
);
1537 if (d_unhashed(dentry
)) {
1538 spin_unlock(&dentry
->d_lock
);
1541 spin_unlock(&dentry
->d_lock
);
1543 /* Negative dentries can be dropped without further checks */
1544 if (!dentry
->d_inode
) {
1550 struct detach_data data
;
1552 data
.mountpoint
= NULL
;
1553 INIT_LIST_HEAD(&data
.select
.dispose
);
1554 data
.select
.start
= dentry
;
1555 data
.select
.found
= 0;
1557 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1559 if (data
.select
.found
)
1560 shrink_dentry_list(&data
.select
.dispose
);
1562 if (data
.mountpoint
) {
1563 detach_mounts(data
.mountpoint
);
1564 dput(data
.mountpoint
);
1567 if (!data
.mountpoint
&& !data
.select
.found
)
1573 EXPORT_SYMBOL(d_invalidate
);
1576 * __d_alloc - allocate a dcache entry
1577 * @sb: filesystem it will belong to
1578 * @name: qstr of the name
1580 * Allocates a dentry. It returns %NULL if there is insufficient memory
1581 * available. On a success the dentry is returned. The name passed in is
1582 * copied and the copy passed in may be reused after this call.
1585 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1587 struct dentry
*dentry
;
1590 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1595 * We guarantee that the inline name is always NUL-terminated.
1596 * This way the memcpy() done by the name switching in rename
1597 * will still always have a NUL at the end, even if we might
1598 * be overwriting an internal NUL character
1600 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1601 if (name
->len
> DNAME_INLINE_LEN
-1) {
1602 size_t size
= offsetof(struct external_name
, name
[1]);
1603 struct external_name
*p
= kmalloc(size
+ name
->len
, GFP_KERNEL
);
1605 kmem_cache_free(dentry_cache
, dentry
);
1608 atomic_set(&p
->u
.count
, 1);
1610 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1611 kasan_unpoison_shadow(dname
,
1612 round_up(name
->len
+ 1, sizeof(unsigned long)));
1614 dname
= dentry
->d_iname
;
1617 dentry
->d_name
.len
= name
->len
;
1618 dentry
->d_name
.hash
= name
->hash
;
1619 memcpy(dname
, name
->name
, name
->len
);
1620 dname
[name
->len
] = 0;
1622 /* Make sure we always see the terminating NUL character */
1624 dentry
->d_name
.name
= dname
;
1626 dentry
->d_lockref
.count
= 1;
1627 dentry
->d_flags
= 0;
1628 spin_lock_init(&dentry
->d_lock
);
1629 seqcount_init(&dentry
->d_seq
);
1630 dentry
->d_inode
= NULL
;
1631 dentry
->d_parent
= dentry
;
1633 dentry
->d_op
= NULL
;
1634 dentry
->d_fsdata
= NULL
;
1635 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1636 INIT_LIST_HEAD(&dentry
->d_lru
);
1637 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1638 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1639 INIT_LIST_HEAD(&dentry
->d_child
);
1640 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1642 this_cpu_inc(nr_dentry
);
1648 * d_alloc - allocate a dcache entry
1649 * @parent: parent of entry to allocate
1650 * @name: qstr of the name
1652 * Allocates a dentry. It returns %NULL if there is insufficient memory
1653 * available. On a success the dentry is returned. The name passed in is
1654 * copied and the copy passed in may be reused after this call.
1656 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1658 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1661 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1662 spin_lock(&parent
->d_lock
);
1664 * don't need child lock because it is not subject
1665 * to concurrency here
1667 __dget_dlock(parent
);
1668 dentry
->d_parent
= parent
;
1669 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1670 spin_unlock(&parent
->d_lock
);
1674 EXPORT_SYMBOL(d_alloc
);
1677 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1678 * @sb: the superblock
1679 * @name: qstr of the name
1681 * For a filesystem that just pins its dentries in memory and never
1682 * performs lookups at all, return an unhashed IS_ROOT dentry.
1684 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1686 return __d_alloc(sb
, name
);
1688 EXPORT_SYMBOL(d_alloc_pseudo
);
1690 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1695 q
.len
= strlen(name
);
1696 q
.hash
= full_name_hash(q
.name
, q
.len
);
1697 return d_alloc(parent
, &q
);
1699 EXPORT_SYMBOL(d_alloc_name
);
1701 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1703 WARN_ON_ONCE(dentry
->d_op
);
1704 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1706 DCACHE_OP_REVALIDATE
|
1707 DCACHE_OP_WEAK_REVALIDATE
|
1709 DCACHE_OP_SELECT_INODE
|
1715 dentry
->d_flags
|= DCACHE_OP_HASH
;
1717 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1718 if (op
->d_revalidate
)
1719 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1720 if (op
->d_weak_revalidate
)
1721 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1723 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1725 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1726 if (op
->d_select_inode
)
1727 dentry
->d_flags
|= DCACHE_OP_SELECT_INODE
;
1729 dentry
->d_flags
|= DCACHE_OP_REAL
;
1732 EXPORT_SYMBOL(d_set_d_op
);
1736 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1737 * @dentry - The dentry to mark
1739 * Mark a dentry as falling through to the lower layer (as set with
1740 * d_pin_lower()). This flag may be recorded on the medium.
1742 void d_set_fallthru(struct dentry
*dentry
)
1744 spin_lock(&dentry
->d_lock
);
1745 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1746 spin_unlock(&dentry
->d_lock
);
1748 EXPORT_SYMBOL(d_set_fallthru
);
1750 static unsigned d_flags_for_inode(struct inode
*inode
)
1752 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1755 return DCACHE_MISS_TYPE
;
1757 if (S_ISDIR(inode
->i_mode
)) {
1758 add_flags
= DCACHE_DIRECTORY_TYPE
;
1759 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1760 if (unlikely(!inode
->i_op
->lookup
))
1761 add_flags
= DCACHE_AUTODIR_TYPE
;
1763 inode
->i_opflags
|= IOP_LOOKUP
;
1765 goto type_determined
;
1768 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1769 if (unlikely(inode
->i_op
->follow_link
)) {
1770 add_flags
= DCACHE_SYMLINK_TYPE
;
1771 goto type_determined
;
1773 inode
->i_opflags
|= IOP_NOFOLLOW
;
1776 if (unlikely(!S_ISREG(inode
->i_mode
)))
1777 add_flags
= DCACHE_SPECIAL_TYPE
;
1780 if (unlikely(IS_AUTOMOUNT(inode
)))
1781 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1785 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1787 unsigned add_flags
= d_flags_for_inode(inode
);
1789 spin_lock(&dentry
->d_lock
);
1791 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1792 raw_write_seqcount_begin(&dentry
->d_seq
);
1793 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1794 raw_write_seqcount_end(&dentry
->d_seq
);
1795 spin_unlock(&dentry
->d_lock
);
1796 fsnotify_d_instantiate(dentry
, inode
);
1800 * d_instantiate - fill in inode information for a dentry
1801 * @entry: dentry to complete
1802 * @inode: inode to attach to this dentry
1804 * Fill in inode information in the entry.
1806 * This turns negative dentries into productive full members
1809 * NOTE! This assumes that the inode count has been incremented
1810 * (or otherwise set) by the caller to indicate that it is now
1811 * in use by the dcache.
1814 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1816 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1818 spin_lock(&inode
->i_lock
);
1819 __d_instantiate(entry
, inode
);
1821 spin_unlock(&inode
->i_lock
);
1822 security_d_instantiate(entry
, inode
);
1824 EXPORT_SYMBOL(d_instantiate
);
1827 * d_instantiate_unique - instantiate a non-aliased dentry
1828 * @entry: dentry to instantiate
1829 * @inode: inode to attach to this dentry
1831 * Fill in inode information in the entry. On success, it returns NULL.
1832 * If an unhashed alias of "entry" already exists, then we return the
1833 * aliased dentry instead and drop one reference to inode.
1835 * Note that in order to avoid conflicts with rename() etc, the caller
1836 * had better be holding the parent directory semaphore.
1838 * This also assumes that the inode count has been incremented
1839 * (or otherwise set) by the caller to indicate that it is now
1840 * in use by the dcache.
1842 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1843 struct inode
*inode
)
1845 struct dentry
*alias
;
1846 int len
= entry
->d_name
.len
;
1847 const char *name
= entry
->d_name
.name
;
1848 unsigned int hash
= entry
->d_name
.hash
;
1851 __d_instantiate(entry
, NULL
);
1855 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1857 * Don't need alias->d_lock here, because aliases with
1858 * d_parent == entry->d_parent are not subject to name or
1859 * parent changes, because the parent inode i_mutex is held.
1861 if (alias
->d_name
.hash
!= hash
)
1863 if (alias
->d_parent
!= entry
->d_parent
)
1865 if (alias
->d_name
.len
!= len
)
1867 if (dentry_cmp(alias
, name
, len
))
1873 __d_instantiate(entry
, inode
);
1877 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1879 struct dentry
*result
;
1881 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1884 spin_lock(&inode
->i_lock
);
1885 result
= __d_instantiate_unique(entry
, inode
);
1887 spin_unlock(&inode
->i_lock
);
1890 security_d_instantiate(entry
, inode
);
1894 BUG_ON(!d_unhashed(result
));
1899 EXPORT_SYMBOL(d_instantiate_unique
);
1902 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1903 * with lockdep-related part of unlock_new_inode() done before
1904 * anything else. Use that instead of open-coding d_instantiate()/
1905 * unlock_new_inode() combinations.
1907 void d_instantiate_new(struct dentry
*entry
, struct inode
*inode
)
1909 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1911 lockdep_annotate_inode_mutex_key(inode
);
1912 security_d_instantiate(entry
, inode
);
1913 spin_lock(&inode
->i_lock
);
1914 __d_instantiate(entry
, inode
);
1915 WARN_ON(!(inode
->i_state
& I_NEW
));
1916 inode
->i_state
&= ~I_NEW
;
1918 wake_up_bit(&inode
->i_state
, __I_NEW
);
1919 spin_unlock(&inode
->i_lock
);
1921 EXPORT_SYMBOL(d_instantiate_new
);
1924 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1925 * @entry: dentry to complete
1926 * @inode: inode to attach to this dentry
1928 * Fill in inode information in the entry. If a directory alias is found, then
1929 * return an error (and drop inode). Together with d_materialise_unique() this
1930 * guarantees that a directory inode may never have more than one alias.
1932 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1934 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1936 spin_lock(&inode
->i_lock
);
1937 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1938 spin_unlock(&inode
->i_lock
);
1942 __d_instantiate(entry
, inode
);
1943 spin_unlock(&inode
->i_lock
);
1944 security_d_instantiate(entry
, inode
);
1948 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1950 struct dentry
*d_make_root(struct inode
*root_inode
)
1952 struct dentry
*res
= NULL
;
1955 static const struct qstr name
= QSTR_INIT("/", 1);
1957 res
= __d_alloc(root_inode
->i_sb
, &name
);
1959 res
->d_flags
|= DCACHE_RCUACCESS
;
1960 d_instantiate(res
, root_inode
);
1967 EXPORT_SYMBOL(d_make_root
);
1969 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1971 struct dentry
*alias
;
1973 if (hlist_empty(&inode
->i_dentry
))
1975 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1981 * d_find_any_alias - find any alias for a given inode
1982 * @inode: inode to find an alias for
1984 * If any aliases exist for the given inode, take and return a
1985 * reference for one of them. If no aliases exist, return %NULL.
1987 struct dentry
*d_find_any_alias(struct inode
*inode
)
1991 spin_lock(&inode
->i_lock
);
1992 de
= __d_find_any_alias(inode
);
1993 spin_unlock(&inode
->i_lock
);
1996 EXPORT_SYMBOL(d_find_any_alias
);
1998 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
2000 static const struct qstr anonstring
= QSTR_INIT("/", 1);
2006 return ERR_PTR(-ESTALE
);
2008 return ERR_CAST(inode
);
2010 res
= d_find_any_alias(inode
);
2014 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
2016 res
= ERR_PTR(-ENOMEM
);
2020 spin_lock(&inode
->i_lock
);
2021 res
= __d_find_any_alias(inode
);
2023 spin_unlock(&inode
->i_lock
);
2028 /* attach a disconnected dentry */
2029 add_flags
= d_flags_for_inode(inode
);
2032 add_flags
|= DCACHE_DISCONNECTED
;
2034 spin_lock(&tmp
->d_lock
);
2035 __d_set_inode_and_type(tmp
, inode
, add_flags
);
2036 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
2037 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
2038 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
2039 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
2040 spin_unlock(&tmp
->d_lock
);
2041 spin_unlock(&inode
->i_lock
);
2042 security_d_instantiate(tmp
, inode
);
2047 if (res
&& !IS_ERR(res
))
2048 security_d_instantiate(res
, inode
);
2054 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2055 * @inode: inode to allocate the dentry for
2057 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2058 * similar open by handle operations. The returned dentry may be anonymous,
2059 * or may have a full name (if the inode was already in the cache).
2061 * When called on a directory inode, we must ensure that the inode only ever
2062 * has one dentry. If a dentry is found, that is returned instead of
2063 * allocating a new one.
2065 * On successful return, the reference to the inode has been transferred
2066 * to the dentry. In case of an error the reference on the inode is released.
2067 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2068 * be passed in and the error will be propagated to the return value,
2069 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2071 struct dentry
*d_obtain_alias(struct inode
*inode
)
2073 return __d_obtain_alias(inode
, 1);
2075 EXPORT_SYMBOL(d_obtain_alias
);
2078 * d_obtain_root - find or allocate a dentry for a given inode
2079 * @inode: inode to allocate the dentry for
2081 * Obtain an IS_ROOT dentry for the root of a filesystem.
2083 * We must ensure that directory inodes only ever have one dentry. If a
2084 * dentry is found, that is returned instead of allocating a new one.
2086 * On successful return, the reference to the inode has been transferred
2087 * to the dentry. In case of an error the reference on the inode is
2088 * released. A %NULL or IS_ERR inode may be passed in and will be the
2089 * error will be propagate to the return value, with a %NULL @inode
2090 * replaced by ERR_PTR(-ESTALE).
2092 struct dentry
*d_obtain_root(struct inode
*inode
)
2094 return __d_obtain_alias(inode
, 0);
2096 EXPORT_SYMBOL(d_obtain_root
);
2099 * d_add_ci - lookup or allocate new dentry with case-exact name
2100 * @inode: the inode case-insensitive lookup has found
2101 * @dentry: the negative dentry that was passed to the parent's lookup func
2102 * @name: the case-exact name to be associated with the returned dentry
2104 * This is to avoid filling the dcache with case-insensitive names to the
2105 * same inode, only the actual correct case is stored in the dcache for
2106 * case-insensitive filesystems.
2108 * For a case-insensitive lookup match and if the the case-exact dentry
2109 * already exists in in the dcache, use it and return it.
2111 * If no entry exists with the exact case name, allocate new dentry with
2112 * the exact case, and return the spliced entry.
2114 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2117 struct dentry
*found
;
2121 * First check if a dentry matching the name already exists,
2122 * if not go ahead and create it now.
2124 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2126 new = d_alloc(dentry
->d_parent
, name
);
2128 found
= ERR_PTR(-ENOMEM
);
2130 found
= d_splice_alias(inode
, new);
2141 EXPORT_SYMBOL(d_add_ci
);
2144 * Do the slow-case of the dentry name compare.
2146 * Unlike the dentry_cmp() function, we need to atomically
2147 * load the name and length information, so that the
2148 * filesystem can rely on them, and can use the 'name' and
2149 * 'len' information without worrying about walking off the
2150 * end of memory etc.
2152 * Thus the read_seqcount_retry() and the "duplicate" info
2153 * in arguments (the low-level filesystem should not look
2154 * at the dentry inode or name contents directly, since
2155 * rename can change them while we're in RCU mode).
2157 enum slow_d_compare
{
2163 static noinline
enum slow_d_compare
slow_dentry_cmp(
2164 const struct dentry
*parent
,
2165 struct dentry
*dentry
,
2167 const struct qstr
*name
)
2169 int tlen
= dentry
->d_name
.len
;
2170 const char *tname
= dentry
->d_name
.name
;
2172 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2174 return D_COMP_SEQRETRY
;
2176 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2177 return D_COMP_NOMATCH
;
2182 * __d_lookup_rcu - search for a dentry (racy, store-free)
2183 * @parent: parent dentry
2184 * @name: qstr of name we wish to find
2185 * @seqp: returns d_seq value at the point where the dentry was found
2186 * Returns: dentry, or NULL
2188 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2189 * resolution (store-free path walking) design described in
2190 * Documentation/filesystems/path-lookup.txt.
2192 * This is not to be used outside core vfs.
2194 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2195 * held, and rcu_read_lock held. The returned dentry must not be stored into
2196 * without taking d_lock and checking d_seq sequence count against @seq
2199 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2202 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2203 * the returned dentry, so long as its parent's seqlock is checked after the
2204 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2205 * is formed, giving integrity down the path walk.
2207 * NOTE! The caller *has* to check the resulting dentry against the sequence
2208 * number we've returned before using any of the resulting dentry state!
2210 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2211 const struct qstr
*name
,
2214 u64 hashlen
= name
->hash_len
;
2215 const unsigned char *str
= name
->name
;
2216 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2217 struct hlist_bl_node
*node
;
2218 struct dentry
*dentry
;
2221 * Note: There is significant duplication with __d_lookup_rcu which is
2222 * required to prevent single threaded performance regressions
2223 * especially on architectures where smp_rmb (in seqcounts) are costly.
2224 * Keep the two functions in sync.
2228 * The hash list is protected using RCU.
2230 * Carefully use d_seq when comparing a candidate dentry, to avoid
2231 * races with d_move().
2233 * It is possible that concurrent renames can mess up our list
2234 * walk here and result in missing our dentry, resulting in the
2235 * false-negative result. d_lookup() protects against concurrent
2236 * renames using rename_lock seqlock.
2238 * See Documentation/filesystems/path-lookup.txt for more details.
2240 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2245 * The dentry sequence count protects us from concurrent
2246 * renames, and thus protects parent and name fields.
2248 * The caller must perform a seqcount check in order
2249 * to do anything useful with the returned dentry.
2251 * NOTE! We do a "raw" seqcount_begin here. That means that
2252 * we don't wait for the sequence count to stabilize if it
2253 * is in the middle of a sequence change. If we do the slow
2254 * dentry compare, we will do seqretries until it is stable,
2255 * and if we end up with a successful lookup, we actually
2256 * want to exit RCU lookup anyway.
2258 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2259 if (dentry
->d_parent
!= parent
)
2261 if (d_unhashed(dentry
))
2264 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2265 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2268 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2271 case D_COMP_NOMATCH
:
2278 if (dentry
->d_name
.hash_len
!= hashlen
)
2281 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2288 * d_lookup - search for a dentry
2289 * @parent: parent dentry
2290 * @name: qstr of name we wish to find
2291 * Returns: dentry, or NULL
2293 * d_lookup searches the children of the parent dentry for the name in
2294 * question. If the dentry is found its reference count is incremented and the
2295 * dentry is returned. The caller must use dput to free the entry when it has
2296 * finished using it. %NULL is returned if the dentry does not exist.
2298 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2300 struct dentry
*dentry
;
2304 seq
= read_seqbegin(&rename_lock
);
2305 dentry
= __d_lookup(parent
, name
);
2308 } while (read_seqretry(&rename_lock
, seq
));
2311 EXPORT_SYMBOL(d_lookup
);
2314 * __d_lookup - search for a dentry (racy)
2315 * @parent: parent dentry
2316 * @name: qstr of name we wish to find
2317 * Returns: dentry, or NULL
2319 * __d_lookup is like d_lookup, however it may (rarely) return a
2320 * false-negative result due to unrelated rename activity.
2322 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2323 * however it must be used carefully, eg. with a following d_lookup in
2324 * the case of failure.
2326 * __d_lookup callers must be commented.
2328 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2330 unsigned int len
= name
->len
;
2331 unsigned int hash
= name
->hash
;
2332 const unsigned char *str
= name
->name
;
2333 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2334 struct hlist_bl_node
*node
;
2335 struct dentry
*found
= NULL
;
2336 struct dentry
*dentry
;
2339 * Note: There is significant duplication with __d_lookup_rcu which is
2340 * required to prevent single threaded performance regressions
2341 * especially on architectures where smp_rmb (in seqcounts) are costly.
2342 * Keep the two functions in sync.
2346 * The hash list is protected using RCU.
2348 * Take d_lock when comparing a candidate dentry, to avoid races
2351 * It is possible that concurrent renames can mess up our list
2352 * walk here and result in missing our dentry, resulting in the
2353 * false-negative result. d_lookup() protects against concurrent
2354 * renames using rename_lock seqlock.
2356 * See Documentation/filesystems/path-lookup.txt for more details.
2360 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2362 if (dentry
->d_name
.hash
!= hash
)
2365 spin_lock(&dentry
->d_lock
);
2366 if (dentry
->d_parent
!= parent
)
2368 if (d_unhashed(dentry
))
2372 * It is safe to compare names since d_move() cannot
2373 * change the qstr (protected by d_lock).
2375 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2376 int tlen
= dentry
->d_name
.len
;
2377 const char *tname
= dentry
->d_name
.name
;
2378 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2381 if (dentry
->d_name
.len
!= len
)
2383 if (dentry_cmp(dentry
, str
, len
))
2387 dentry
->d_lockref
.count
++;
2389 spin_unlock(&dentry
->d_lock
);
2392 spin_unlock(&dentry
->d_lock
);
2400 * d_hash_and_lookup - hash the qstr then search for a dentry
2401 * @dir: Directory to search in
2402 * @name: qstr of name we wish to find
2404 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2406 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2409 * Check for a fs-specific hash function. Note that we must
2410 * calculate the standard hash first, as the d_op->d_hash()
2411 * routine may choose to leave the hash value unchanged.
2413 name
->hash
= full_name_hash(name
->name
, name
->len
);
2414 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2415 int err
= dir
->d_op
->d_hash(dir
, name
);
2416 if (unlikely(err
< 0))
2417 return ERR_PTR(err
);
2419 return d_lookup(dir
, name
);
2421 EXPORT_SYMBOL(d_hash_and_lookup
);
2424 * When a file is deleted, we have two options:
2425 * - turn this dentry into a negative dentry
2426 * - unhash this dentry and free it.
2428 * Usually, we want to just turn this into
2429 * a negative dentry, but if anybody else is
2430 * currently using the dentry or the inode
2431 * we can't do that and we fall back on removing
2432 * it from the hash queues and waiting for
2433 * it to be deleted later when it has no users
2437 * d_delete - delete a dentry
2438 * @dentry: The dentry to delete
2440 * Turn the dentry into a negative dentry if possible, otherwise
2441 * remove it from the hash queues so it can be deleted later
2444 void d_delete(struct dentry
* dentry
)
2446 struct inode
*inode
;
2449 * Are we the only user?
2452 spin_lock(&dentry
->d_lock
);
2453 inode
= dentry
->d_inode
;
2454 isdir
= S_ISDIR(inode
->i_mode
);
2455 if (dentry
->d_lockref
.count
== 1) {
2456 if (!spin_trylock(&inode
->i_lock
)) {
2457 spin_unlock(&dentry
->d_lock
);
2461 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2462 dentry_unlink_inode(dentry
);
2463 fsnotify_nameremove(dentry
, isdir
);
2467 if (!d_unhashed(dentry
))
2470 spin_unlock(&dentry
->d_lock
);
2472 fsnotify_nameremove(dentry
, isdir
);
2474 EXPORT_SYMBOL(d_delete
);
2476 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2478 BUG_ON(!d_unhashed(entry
));
2480 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2484 static void _d_rehash(struct dentry
* entry
)
2486 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2490 * d_rehash - add an entry back to the hash
2491 * @entry: dentry to add to the hash
2493 * Adds a dentry to the hash according to its name.
2496 void d_rehash(struct dentry
* entry
)
2498 spin_lock(&entry
->d_lock
);
2500 spin_unlock(&entry
->d_lock
);
2502 EXPORT_SYMBOL(d_rehash
);
2505 * dentry_update_name_case - update case insensitive dentry with a new name
2506 * @dentry: dentry to be updated
2509 * Update a case insensitive dentry with new case of name.
2511 * dentry must have been returned by d_lookup with name @name. Old and new
2512 * name lengths must match (ie. no d_compare which allows mismatched name
2515 * Parent inode i_mutex must be held over d_lookup and into this call (to
2516 * keep renames and concurrent inserts, and readdir(2) away).
2518 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2520 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2521 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2523 spin_lock(&dentry
->d_lock
);
2524 write_seqcount_begin(&dentry
->d_seq
);
2525 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2526 write_seqcount_end(&dentry
->d_seq
);
2527 spin_unlock(&dentry
->d_lock
);
2529 EXPORT_SYMBOL(dentry_update_name_case
);
2531 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2533 if (unlikely(dname_external(target
))) {
2534 if (unlikely(dname_external(dentry
))) {
2536 * Both external: swap the pointers
2538 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2541 * dentry:internal, target:external. Steal target's
2542 * storage and make target internal.
2544 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2545 dentry
->d_name
.len
+ 1);
2546 dentry
->d_name
.name
= target
->d_name
.name
;
2547 target
->d_name
.name
= target
->d_iname
;
2550 if (unlikely(dname_external(dentry
))) {
2552 * dentry:external, target:internal. Give dentry's
2553 * storage to target and make dentry internal
2555 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2556 target
->d_name
.len
+ 1);
2557 target
->d_name
.name
= dentry
->d_name
.name
;
2558 dentry
->d_name
.name
= dentry
->d_iname
;
2561 * Both are internal.
2564 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2565 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2566 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2567 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2568 swap(((long *) &dentry
->d_iname
)[i
],
2569 ((long *) &target
->d_iname
)[i
]);
2573 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2576 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2578 struct external_name
*old_name
= NULL
;
2579 if (unlikely(dname_external(dentry
)))
2580 old_name
= external_name(dentry
);
2581 if (unlikely(dname_external(target
))) {
2582 atomic_inc(&external_name(target
)->u
.count
);
2583 dentry
->d_name
= target
->d_name
;
2585 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2586 target
->d_name
.len
+ 1);
2587 dentry
->d_name
.name
= dentry
->d_iname
;
2588 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2590 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2591 kfree_rcu(old_name
, u
.head
);
2594 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2597 * XXXX: do we really need to take target->d_lock?
2599 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2600 spin_lock(&target
->d_parent
->d_lock
);
2602 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2603 spin_lock(&dentry
->d_parent
->d_lock
);
2604 spin_lock_nested(&target
->d_parent
->d_lock
,
2605 DENTRY_D_LOCK_NESTED
);
2607 spin_lock(&target
->d_parent
->d_lock
);
2608 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2609 DENTRY_D_LOCK_NESTED
);
2612 if (target
< dentry
) {
2613 spin_lock_nested(&target
->d_lock
, 2);
2614 spin_lock_nested(&dentry
->d_lock
, 3);
2616 spin_lock_nested(&dentry
->d_lock
, 2);
2617 spin_lock_nested(&target
->d_lock
, 3);
2621 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2623 if (target
->d_parent
!= dentry
->d_parent
)
2624 spin_unlock(&dentry
->d_parent
->d_lock
);
2625 if (target
->d_parent
!= target
)
2626 spin_unlock(&target
->d_parent
->d_lock
);
2627 spin_unlock(&target
->d_lock
);
2628 spin_unlock(&dentry
->d_lock
);
2632 * When switching names, the actual string doesn't strictly have to
2633 * be preserved in the target - because we're dropping the target
2634 * anyway. As such, we can just do a simple memcpy() to copy over
2635 * the new name before we switch, unless we are going to rehash
2636 * it. Note that if we *do* unhash the target, we are not allowed
2637 * to rehash it without giving it a new name/hash key - whether
2638 * we swap or overwrite the names here, resulting name won't match
2639 * the reality in filesystem; it's only there for d_path() purposes.
2640 * Note that all of this is happening under rename_lock, so the
2641 * any hash lookup seeing it in the middle of manipulations will
2642 * be discarded anyway. So we do not care what happens to the hash
2646 * __d_move - move a dentry
2647 * @dentry: entry to move
2648 * @target: new dentry
2649 * @exchange: exchange the two dentries
2651 * Update the dcache to reflect the move of a file name. Negative
2652 * dcache entries should not be moved in this way. Caller must hold
2653 * rename_lock, the i_mutex of the source and target directories,
2654 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2656 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2659 if (!dentry
->d_inode
)
2660 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2662 BUG_ON(d_ancestor(dentry
, target
));
2663 BUG_ON(d_ancestor(target
, dentry
));
2665 dentry_lock_for_move(dentry
, target
);
2667 write_seqcount_begin(&dentry
->d_seq
);
2668 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2670 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2673 * Move the dentry to the target hash queue. Don't bother checking
2674 * for the same hash queue because of how unlikely it is.
2677 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2680 * Unhash the target (d_delete() is not usable here). If exchanging
2681 * the two dentries, then rehash onto the other's hash queue.
2686 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2689 /* Switch the names.. */
2691 swap_names(dentry
, target
);
2693 copy_name(dentry
, target
);
2695 /* ... and switch them in the tree */
2696 if (IS_ROOT(dentry
)) {
2697 /* splicing a tree */
2698 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2699 dentry
->d_parent
= target
->d_parent
;
2700 target
->d_parent
= target
;
2701 list_del_init(&target
->d_child
);
2702 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2704 /* swapping two dentries */
2705 swap(dentry
->d_parent
, target
->d_parent
);
2706 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2707 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2709 fsnotify_d_move(target
);
2710 fsnotify_d_move(dentry
);
2713 write_seqcount_end(&target
->d_seq
);
2714 write_seqcount_end(&dentry
->d_seq
);
2716 dentry_unlock_for_move(dentry
, target
);
2720 * d_move - move a dentry
2721 * @dentry: entry to move
2722 * @target: new dentry
2724 * Update the dcache to reflect the move of a file name. Negative
2725 * dcache entries should not be moved in this way. See the locking
2726 * requirements for __d_move.
2728 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2730 write_seqlock(&rename_lock
);
2731 __d_move(dentry
, target
, false);
2732 write_sequnlock(&rename_lock
);
2734 EXPORT_SYMBOL(d_move
);
2737 * d_exchange - exchange two dentries
2738 * @dentry1: first dentry
2739 * @dentry2: second dentry
2741 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2743 write_seqlock(&rename_lock
);
2745 WARN_ON(!dentry1
->d_inode
);
2746 WARN_ON(!dentry2
->d_inode
);
2747 WARN_ON(IS_ROOT(dentry1
));
2748 WARN_ON(IS_ROOT(dentry2
));
2750 __d_move(dentry1
, dentry2
, true);
2752 write_sequnlock(&rename_lock
);
2756 * d_ancestor - search for an ancestor
2757 * @p1: ancestor dentry
2760 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2761 * an ancestor of p2, else NULL.
2763 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2767 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2768 if (p
->d_parent
== p1
)
2775 * This helper attempts to cope with remotely renamed directories
2777 * It assumes that the caller is already holding
2778 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2780 * Note: If ever the locking in lock_rename() changes, then please
2781 * remember to update this too...
2783 static int __d_unalias(struct inode
*inode
,
2784 struct dentry
*dentry
, struct dentry
*alias
)
2786 struct mutex
*m1
= NULL
, *m2
= NULL
;
2789 /* If alias and dentry share a parent, then no extra locks required */
2790 if (alias
->d_parent
== dentry
->d_parent
)
2793 /* See lock_rename() */
2794 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2796 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2797 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2799 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2801 __d_move(alias
, dentry
, false);
2812 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2813 * @inode: the inode which may have a disconnected dentry
2814 * @dentry: a negative dentry which we want to point to the inode.
2816 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2817 * place of the given dentry and return it, else simply d_add the inode
2818 * to the dentry and return NULL.
2820 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2821 * we should error out: directories can't have multiple aliases.
2823 * This is needed in the lookup routine of any filesystem that is exportable
2824 * (via knfsd) so that we can build dcache paths to directories effectively.
2826 * If a dentry was found and moved, then it is returned. Otherwise NULL
2827 * is returned. This matches the expected return value of ->lookup.
2829 * Cluster filesystems may call this function with a negative, hashed dentry.
2830 * In that case, we know that the inode will be a regular file, and also this
2831 * will only occur during atomic_open. So we need to check for the dentry
2832 * being already hashed only in the final case.
2834 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2837 return ERR_CAST(inode
);
2839 BUG_ON(!d_unhashed(dentry
));
2842 __d_instantiate(dentry
, NULL
);
2845 spin_lock(&inode
->i_lock
);
2846 if (S_ISDIR(inode
->i_mode
)) {
2847 struct dentry
*new = __d_find_any_alias(inode
);
2848 if (unlikely(new)) {
2849 /* The reference to new ensures it remains an alias */
2850 spin_unlock(&inode
->i_lock
);
2851 write_seqlock(&rename_lock
);
2852 if (unlikely(d_ancestor(new, dentry
))) {
2853 write_sequnlock(&rename_lock
);
2855 new = ERR_PTR(-ELOOP
);
2856 pr_warn_ratelimited(
2857 "VFS: Lookup of '%s' in %s %s"
2858 " would have caused loop\n",
2859 dentry
->d_name
.name
,
2860 inode
->i_sb
->s_type
->name
,
2862 } else if (!IS_ROOT(new)) {
2863 int err
= __d_unalias(inode
, dentry
, new);
2864 write_sequnlock(&rename_lock
);
2870 __d_move(new, dentry
, false);
2871 write_sequnlock(&rename_lock
);
2872 security_d_instantiate(new, inode
);
2878 /* already taking inode->i_lock, so d_add() by hand */
2879 __d_instantiate(dentry
, inode
);
2880 spin_unlock(&inode
->i_lock
);
2882 security_d_instantiate(dentry
, inode
);
2886 EXPORT_SYMBOL(d_splice_alias
);
2888 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2892 return -ENAMETOOLONG
;
2894 memcpy(*buffer
, str
, namelen
);
2899 * prepend_name - prepend a pathname in front of current buffer pointer
2900 * @buffer: buffer pointer
2901 * @buflen: allocated length of the buffer
2902 * @name: name string and length qstr structure
2904 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2905 * make sure that either the old or the new name pointer and length are
2906 * fetched. However, there may be mismatch between length and pointer.
2907 * The length cannot be trusted, we need to copy it byte-by-byte until
2908 * the length is reached or a null byte is found. It also prepends "/" at
2909 * the beginning of the name. The sequence number check at the caller will
2910 * retry it again when a d_move() does happen. So any garbage in the buffer
2911 * due to mismatched pointer and length will be discarded.
2913 * Data dependency barrier is needed to make sure that we see that terminating
2914 * NUL. Alpha strikes again, film at 11...
2916 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2918 const char *dname
= ACCESS_ONCE(name
->name
);
2919 u32 dlen
= ACCESS_ONCE(name
->len
);
2922 smp_read_barrier_depends();
2924 *buflen
-= dlen
+ 1;
2926 return -ENAMETOOLONG
;
2927 p
= *buffer
-= dlen
+ 1;
2939 * prepend_path - Prepend path string to a buffer
2940 * @path: the dentry/vfsmount to report
2941 * @root: root vfsmnt/dentry
2942 * @buffer: pointer to the end of the buffer
2943 * @buflen: pointer to buffer length
2945 * The function will first try to write out the pathname without taking any
2946 * lock other than the RCU read lock to make sure that dentries won't go away.
2947 * It only checks the sequence number of the global rename_lock as any change
2948 * in the dentry's d_seq will be preceded by changes in the rename_lock
2949 * sequence number. If the sequence number had been changed, it will restart
2950 * the whole pathname back-tracing sequence again by taking the rename_lock.
2951 * In this case, there is no need to take the RCU read lock as the recursive
2952 * parent pointer references will keep the dentry chain alive as long as no
2953 * rename operation is performed.
2955 static int prepend_path(const struct path
*path
,
2956 const struct path
*root
,
2957 char **buffer
, int *buflen
)
2959 struct dentry
*dentry
;
2960 struct vfsmount
*vfsmnt
;
2963 unsigned seq
, m_seq
= 0;
2969 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
2976 dentry
= path
->dentry
;
2978 mnt
= real_mount(vfsmnt
);
2979 read_seqbegin_or_lock(&rename_lock
, &seq
);
2980 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2981 struct dentry
* parent
;
2983 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2984 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
2986 if (dentry
!= vfsmnt
->mnt_root
) {
2993 if (mnt
!= parent
) {
2994 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3000 error
= is_mounted(vfsmnt
) ? 1 : 2;
3003 parent
= dentry
->d_parent
;
3005 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3013 if (need_seqretry(&rename_lock
, seq
)) {
3017 done_seqretry(&rename_lock
, seq
);
3021 if (need_seqretry(&mount_lock
, m_seq
)) {
3025 done_seqretry(&mount_lock
, m_seq
);
3027 if (error
>= 0 && bptr
== *buffer
) {
3029 error
= -ENAMETOOLONG
;
3039 * __d_path - return the path of a dentry
3040 * @path: the dentry/vfsmount to report
3041 * @root: root vfsmnt/dentry
3042 * @buf: buffer to return value in
3043 * @buflen: buffer length
3045 * Convert a dentry into an ASCII path name.
3047 * Returns a pointer into the buffer or an error code if the
3048 * path was too long.
3050 * "buflen" should be positive.
3052 * If the path is not reachable from the supplied root, return %NULL.
3054 char *__d_path(const struct path
*path
,
3055 const struct path
*root
,
3056 char *buf
, int buflen
)
3058 char *res
= buf
+ buflen
;
3061 prepend(&res
, &buflen
, "\0", 1);
3062 error
= prepend_path(path
, root
, &res
, &buflen
);
3065 return ERR_PTR(error
);
3071 char *d_absolute_path(const struct path
*path
,
3072 char *buf
, int buflen
)
3074 struct path root
= {};
3075 char *res
= buf
+ buflen
;
3078 prepend(&res
, &buflen
, "\0", 1);
3079 error
= prepend_path(path
, &root
, &res
, &buflen
);
3084 return ERR_PTR(error
);
3089 * same as __d_path but appends "(deleted)" for unlinked files.
3091 static int path_with_deleted(const struct path
*path
,
3092 const struct path
*root
,
3093 char **buf
, int *buflen
)
3095 prepend(buf
, buflen
, "\0", 1);
3096 if (d_unlinked(path
->dentry
)) {
3097 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3102 return prepend_path(path
, root
, buf
, buflen
);
3105 static int prepend_unreachable(char **buffer
, int *buflen
)
3107 return prepend(buffer
, buflen
, "(unreachable)", 13);
3110 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3115 seq
= read_seqcount_begin(&fs
->seq
);
3117 } while (read_seqcount_retry(&fs
->seq
, seq
));
3121 * d_path - return the path of a dentry
3122 * @path: path to report
3123 * @buf: buffer to return value in
3124 * @buflen: buffer length
3126 * Convert a dentry into an ASCII path name. If the entry has been deleted
3127 * the string " (deleted)" is appended. Note that this is ambiguous.
3129 * Returns a pointer into the buffer or an error code if the path was
3130 * too long. Note: Callers should use the returned pointer, not the passed
3131 * in buffer, to use the name! The implementation often starts at an offset
3132 * into the buffer, and may leave 0 bytes at the start.
3134 * "buflen" should be positive.
3136 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3138 char *res
= buf
+ buflen
;
3143 * We have various synthetic filesystems that never get mounted. On
3144 * these filesystems dentries are never used for lookup purposes, and
3145 * thus don't need to be hashed. They also don't need a name until a
3146 * user wants to identify the object in /proc/pid/fd/. The little hack
3147 * below allows us to generate a name for these objects on demand:
3149 * Some pseudo inodes are mountable. When they are mounted
3150 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3151 * and instead have d_path return the mounted path.
3153 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3154 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3155 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3158 get_fs_root_rcu(current
->fs
, &root
);
3159 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3163 res
= ERR_PTR(error
);
3166 EXPORT_SYMBOL(d_path
);
3169 * Helper function for dentry_operations.d_dname() members
3171 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3172 const char *fmt
, ...)
3178 va_start(args
, fmt
);
3179 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3182 if (sz
> sizeof(temp
) || sz
> buflen
)
3183 return ERR_PTR(-ENAMETOOLONG
);
3185 buffer
+= buflen
- sz
;
3186 return memcpy(buffer
, temp
, sz
);
3189 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3191 char *end
= buffer
+ buflen
;
3192 /* these dentries are never renamed, so d_lock is not needed */
3193 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3194 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3195 prepend(&end
, &buflen
, "/", 1))
3196 end
= ERR_PTR(-ENAMETOOLONG
);
3199 EXPORT_SYMBOL(simple_dname
);
3202 * Write full pathname from the root of the filesystem into the buffer.
3204 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3206 struct dentry
*dentry
;
3219 prepend(&end
, &len
, "\0", 1);
3223 read_seqbegin_or_lock(&rename_lock
, &seq
);
3224 while (!IS_ROOT(dentry
)) {
3225 struct dentry
*parent
= dentry
->d_parent
;
3228 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3237 if (need_seqretry(&rename_lock
, seq
)) {
3241 done_seqretry(&rename_lock
, seq
);
3246 return ERR_PTR(-ENAMETOOLONG
);
3249 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3251 return __dentry_path(dentry
, buf
, buflen
);
3253 EXPORT_SYMBOL(dentry_path_raw
);
3255 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3260 if (d_unlinked(dentry
)) {
3262 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3266 retval
= __dentry_path(dentry
, buf
, buflen
);
3267 if (!IS_ERR(retval
) && p
)
3268 *p
= '/'; /* restore '/' overriden with '\0' */
3271 return ERR_PTR(-ENAMETOOLONG
);
3274 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3280 seq
= read_seqcount_begin(&fs
->seq
);
3283 } while (read_seqcount_retry(&fs
->seq
, seq
));
3287 * NOTE! The user-level library version returns a
3288 * character pointer. The kernel system call just
3289 * returns the length of the buffer filled (which
3290 * includes the ending '\0' character), or a negative
3291 * error value. So libc would do something like
3293 * char *getcwd(char * buf, size_t size)
3297 * retval = sys_getcwd(buf, size);
3304 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3307 struct path pwd
, root
;
3308 char *page
= __getname();
3314 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3317 if (!d_unlinked(pwd
.dentry
)) {
3319 char *cwd
= page
+ PATH_MAX
;
3320 int buflen
= PATH_MAX
;
3322 prepend(&cwd
, &buflen
, "\0", 1);
3323 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3329 /* Unreachable from current root */
3331 error
= prepend_unreachable(&cwd
, &buflen
);
3337 len
= PATH_MAX
+ page
- cwd
;
3340 if (copy_to_user(buf
, cwd
, len
))
3353 * Test whether new_dentry is a subdirectory of old_dentry.
3355 * Trivially implemented using the dcache structure
3359 * is_subdir - is new dentry a subdirectory of old_dentry
3360 * @new_dentry: new dentry
3361 * @old_dentry: old dentry
3363 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3364 * Returns 0 otherwise.
3365 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3368 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3373 if (new_dentry
== old_dentry
)
3377 /* for restarting inner loop in case of seq retry */
3378 seq
= read_seqbegin(&rename_lock
);
3380 * Need rcu_readlock to protect against the d_parent trashing
3384 if (d_ancestor(old_dentry
, new_dentry
))
3389 } while (read_seqretry(&rename_lock
, seq
));
3394 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3396 struct dentry
*root
= data
;
3397 if (dentry
!= root
) {
3398 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3401 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3402 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3403 dentry
->d_lockref
.count
--;
3406 return D_WALK_CONTINUE
;
3409 void d_genocide(struct dentry
*parent
)
3411 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3414 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3416 inode_dec_link_count(inode
);
3417 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3418 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3419 !d_unlinked(dentry
));
3420 spin_lock(&dentry
->d_parent
->d_lock
);
3421 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3422 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3423 (unsigned long long)inode
->i_ino
);
3424 spin_unlock(&dentry
->d_lock
);
3425 spin_unlock(&dentry
->d_parent
->d_lock
);
3426 d_instantiate(dentry
, inode
);
3428 EXPORT_SYMBOL(d_tmpfile
);
3430 static __initdata
unsigned long dhash_entries
;
3431 static int __init
set_dhash_entries(char *str
)
3435 dhash_entries
= simple_strtoul(str
, &str
, 0);
3438 __setup("dhash_entries=", set_dhash_entries
);
3440 static void __init
dcache_init_early(void)
3444 /* If hashes are distributed across NUMA nodes, defer
3445 * hash allocation until vmalloc space is available.
3451 alloc_large_system_hash("Dentry cache",
3452 sizeof(struct hlist_bl_head
),
3461 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3462 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3465 static void __init
dcache_init(void)
3470 * A constructor could be added for stable state like the lists,
3471 * but it is probably not worth it because of the cache nature
3474 dentry_cache
= KMEM_CACHE(dentry
,
3475 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3477 /* Hash may have been set up in dcache_init_early */
3482 alloc_large_system_hash("Dentry cache",
3483 sizeof(struct hlist_bl_head
),
3492 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3493 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3496 /* SLAB cache for __getname() consumers */
3497 struct kmem_cache
*names_cachep __read_mostly
;
3498 EXPORT_SYMBOL(names_cachep
);
3500 EXPORT_SYMBOL(d_genocide
);
3502 void __init
vfs_caches_init_early(void)
3504 dcache_init_early();
3508 void __init
vfs_caches_init(void)
3510 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3511 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3516 files_maxfiles_init();