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 static void dentry_free(struct dentry
*dentry
)
274 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
275 if (unlikely(dname_external(dentry
))) {
276 struct external_name
*p
= external_name(dentry
);
277 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
278 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
282 /* if dentry was never visible to RCU, immediate free is OK */
283 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
284 __d_free(&dentry
->d_u
.d_rcu
);
286 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
290 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
291 * @dentry: the target dentry
292 * After this call, in-progress rcu-walk path lookup will fail. This
293 * should be called after unhashing, and after changing d_inode (if
294 * the dentry has not already been unhashed).
296 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
298 assert_spin_locked(&dentry
->d_lock
);
299 /* Go through a barrier */
300 write_seqcount_barrier(&dentry
->d_seq
);
304 * Release the dentry's inode, using the filesystem
305 * d_iput() operation if defined. Dentry has no refcount
308 static void dentry_iput(struct dentry
* dentry
)
309 __releases(dentry
->d_lock
)
310 __releases(dentry
->d_inode
->i_lock
)
312 struct inode
*inode
= dentry
->d_inode
;
314 dentry
->d_inode
= NULL
;
315 hlist_del_init(&dentry
->d_u
.d_alias
);
316 spin_unlock(&dentry
->d_lock
);
317 spin_unlock(&inode
->i_lock
);
319 fsnotify_inoderemove(inode
);
320 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
321 dentry
->d_op
->d_iput(dentry
, inode
);
325 spin_unlock(&dentry
->d_lock
);
330 * Release the dentry's inode, using the filesystem
331 * d_iput() operation if defined. dentry remains in-use.
333 static void dentry_unlink_inode(struct dentry
* dentry
)
334 __releases(dentry
->d_lock
)
335 __releases(dentry
->d_inode
->i_lock
)
337 struct inode
*inode
= dentry
->d_inode
;
338 __d_clear_type(dentry
);
339 dentry
->d_inode
= NULL
;
340 hlist_del_init(&dentry
->d_u
.d_alias
);
341 dentry_rcuwalk_barrier(dentry
);
342 spin_unlock(&dentry
->d_lock
);
343 spin_unlock(&inode
->i_lock
);
345 fsnotify_inoderemove(inode
);
346 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
347 dentry
->d_op
->d_iput(dentry
, inode
);
353 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
354 * is in use - which includes both the "real" per-superblock
355 * LRU list _and_ the DCACHE_SHRINK_LIST use.
357 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
358 * on the shrink list (ie not on the superblock LRU list).
360 * The per-cpu "nr_dentry_unused" counters are updated with
361 * the DCACHE_LRU_LIST bit.
363 * These helper functions make sure we always follow the
364 * rules. d_lock must be held by the caller.
366 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
367 static void d_lru_add(struct dentry
*dentry
)
369 D_FLAG_VERIFY(dentry
, 0);
370 dentry
->d_flags
|= DCACHE_LRU_LIST
;
371 this_cpu_inc(nr_dentry_unused
);
372 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
375 static void d_lru_del(struct dentry
*dentry
)
377 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
378 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
379 this_cpu_dec(nr_dentry_unused
);
380 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
383 static void d_shrink_del(struct dentry
*dentry
)
385 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
386 list_del_init(&dentry
->d_lru
);
387 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
388 this_cpu_dec(nr_dentry_unused
);
391 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
393 D_FLAG_VERIFY(dentry
, 0);
394 list_add(&dentry
->d_lru
, list
);
395 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
396 this_cpu_inc(nr_dentry_unused
);
400 * These can only be called under the global LRU lock, ie during the
401 * callback for freeing the LRU list. "isolate" removes it from the
402 * LRU lists entirely, while shrink_move moves it to the indicated
405 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
407 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
408 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
409 this_cpu_dec(nr_dentry_unused
);
410 list_lru_isolate(lru
, &dentry
->d_lru
);
413 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
414 struct list_head
*list
)
416 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
417 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
418 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
422 * dentry_lru_(add|del)_list) must be called with d_lock held.
424 static void dentry_lru_add(struct dentry
*dentry
)
426 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
431 * d_drop - drop a dentry
432 * @dentry: dentry to drop
434 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
435 * be found through a VFS lookup any more. Note that this is different from
436 * deleting the dentry - d_delete will try to mark the dentry negative if
437 * possible, giving a successful _negative_ lookup, while d_drop will
438 * just make the cache lookup fail.
440 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
441 * reason (NFS timeouts or autofs deletes).
443 * __d_drop requires dentry->d_lock.
445 void __d_drop(struct dentry
*dentry
)
447 if (!d_unhashed(dentry
)) {
448 struct hlist_bl_head
*b
;
450 * Hashed dentries are normally on the dentry hashtable,
451 * with the exception of those newly allocated by
452 * d_obtain_alias, which are always IS_ROOT:
454 if (unlikely(IS_ROOT(dentry
)))
455 b
= &dentry
->d_sb
->s_anon
;
457 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
460 __hlist_bl_del(&dentry
->d_hash
);
461 dentry
->d_hash
.pprev
= NULL
;
463 dentry_rcuwalk_barrier(dentry
);
466 EXPORT_SYMBOL(__d_drop
);
468 void d_drop(struct dentry
*dentry
)
470 spin_lock(&dentry
->d_lock
);
472 spin_unlock(&dentry
->d_lock
);
474 EXPORT_SYMBOL(d_drop
);
476 static void __dentry_kill(struct dentry
*dentry
)
478 struct dentry
*parent
= NULL
;
479 bool can_free
= true;
480 if (!IS_ROOT(dentry
))
481 parent
= dentry
->d_parent
;
484 * The dentry is now unrecoverably dead to the world.
486 lockref_mark_dead(&dentry
->d_lockref
);
489 * inform the fs via d_prune that this dentry is about to be
490 * unhashed and destroyed.
492 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
493 dentry
->d_op
->d_prune(dentry
);
495 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
496 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
499 /* if it was on the hash then remove it */
501 __list_del_entry(&dentry
->d_child
);
503 * Inform d_walk() that we are no longer attached to the
506 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
508 spin_unlock(&parent
->d_lock
);
511 * dentry_iput drops the locks, at which point nobody (except
512 * transient RCU lookups) can reach this dentry.
514 BUG_ON(dentry
->d_lockref
.count
> 0);
515 this_cpu_dec(nr_dentry
);
516 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
517 dentry
->d_op
->d_release(dentry
);
519 spin_lock(&dentry
->d_lock
);
520 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
521 dentry
->d_flags
|= DCACHE_MAY_FREE
;
524 spin_unlock(&dentry
->d_lock
);
525 if (likely(can_free
))
530 * Finish off a dentry we've decided to kill.
531 * dentry->d_lock must be held, returns with it unlocked.
532 * If ref is non-zero, then decrement the refcount too.
533 * Returns dentry requiring refcount drop, or NULL if we're done.
535 static struct dentry
*dentry_kill(struct dentry
*dentry
)
536 __releases(dentry
->d_lock
)
538 struct inode
*inode
= dentry
->d_inode
;
539 struct dentry
*parent
= NULL
;
541 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
544 if (!IS_ROOT(dentry
)) {
545 parent
= dentry
->d_parent
;
546 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
548 spin_unlock(&inode
->i_lock
);
553 __dentry_kill(dentry
);
557 spin_unlock(&dentry
->d_lock
);
559 return dentry
; /* try again with same dentry */
562 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
564 struct dentry
*parent
= dentry
->d_parent
;
567 if (unlikely(dentry
->d_lockref
.count
< 0))
569 if (likely(spin_trylock(&parent
->d_lock
)))
572 spin_unlock(&dentry
->d_lock
);
574 parent
= ACCESS_ONCE(dentry
->d_parent
);
575 spin_lock(&parent
->d_lock
);
577 * We can't blindly lock dentry until we are sure
578 * that we won't violate the locking order.
579 * Any changes of dentry->d_parent must have
580 * been done with parent->d_lock held, so
581 * spin_lock() above is enough of a barrier
582 * for checking if it's still our child.
584 if (unlikely(parent
!= dentry
->d_parent
)) {
585 spin_unlock(&parent
->d_lock
);
589 if (parent
!= dentry
)
590 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
597 * Try to do a lockless dput(), and return whether that was successful.
599 * If unsuccessful, we return false, having already taken the dentry lock.
601 * The caller needs to hold the RCU read lock, so that the dentry is
602 * guaranteed to stay around even if the refcount goes down to zero!
604 static inline bool fast_dput(struct dentry
*dentry
)
607 unsigned int d_flags
;
610 * If we have a d_op->d_delete() operation, we sould not
611 * let the dentry count go to zero, so use "put__or_lock".
613 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
614 return lockref_put_or_lock(&dentry
->d_lockref
);
617 * .. otherwise, we can try to just decrement the
618 * lockref optimistically.
620 ret
= lockref_put_return(&dentry
->d_lockref
);
623 * If the lockref_put_return() failed due to the lock being held
624 * by somebody else, the fast path has failed. We will need to
625 * get the lock, and then check the count again.
627 if (unlikely(ret
< 0)) {
628 spin_lock(&dentry
->d_lock
);
629 if (dentry
->d_lockref
.count
> 1) {
630 dentry
->d_lockref
.count
--;
631 spin_unlock(&dentry
->d_lock
);
638 * If we weren't the last ref, we're done.
644 * Careful, careful. The reference count went down
645 * to zero, but we don't hold the dentry lock, so
646 * somebody else could get it again, and do another
647 * dput(), and we need to not race with that.
649 * However, there is a very special and common case
650 * where we don't care, because there is nothing to
651 * do: the dentry is still hashed, it does not have
652 * a 'delete' op, and it's referenced and already on
655 * NOTE! Since we aren't locked, these values are
656 * not "stable". However, it is sufficient that at
657 * some point after we dropped the reference the
658 * dentry was hashed and the flags had the proper
659 * value. Other dentry users may have re-gotten
660 * a reference to the dentry and change that, but
661 * our work is done - we can leave the dentry
662 * around with a zero refcount.
665 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
666 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
;
668 /* Nothing to do? Dropping the reference was all we needed? */
669 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
673 * Not the fast normal case? Get the lock. We've already decremented
674 * the refcount, but we'll need to re-check the situation after
677 spin_lock(&dentry
->d_lock
);
680 * Did somebody else grab a reference to it in the meantime, and
681 * we're no longer the last user after all? Alternatively, somebody
682 * else could have killed it and marked it dead. Either way, we
683 * don't need to do anything else.
685 if (dentry
->d_lockref
.count
) {
686 spin_unlock(&dentry
->d_lock
);
691 * Re-get the reference we optimistically dropped. We hold the
692 * lock, and we just tested that it was zero, so we can just
695 dentry
->d_lockref
.count
= 1;
703 * This is complicated by the fact that we do not want to put
704 * dentries that are no longer on any hash chain on the unused
705 * list: we'd much rather just get rid of them immediately.
707 * However, that implies that we have to traverse the dentry
708 * tree upwards to the parents which might _also_ now be
709 * scheduled for deletion (it may have been only waiting for
710 * its last child to go away).
712 * This tail recursion is done by hand as we don't want to depend
713 * on the compiler to always get this right (gcc generally doesn't).
714 * Real recursion would eat up our stack space.
718 * dput - release a dentry
719 * @dentry: dentry to release
721 * Release a dentry. This will drop the usage count and if appropriate
722 * call the dentry unlink method as well as removing it from the queues and
723 * releasing its resources. If the parent dentries were scheduled for release
724 * they too may now get deleted.
726 void dput(struct dentry
*dentry
)
728 if (unlikely(!dentry
))
733 if (likely(fast_dput(dentry
))) {
738 /* Slow case: now with the dentry lock held */
741 /* Unreachable? Get rid of it */
742 if (unlikely(d_unhashed(dentry
)))
745 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
746 if (dentry
->d_op
->d_delete(dentry
))
750 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
751 dentry
->d_flags
|= DCACHE_REFERENCED
;
752 dentry_lru_add(dentry
);
754 dentry
->d_lockref
.count
--;
755 spin_unlock(&dentry
->d_lock
);
759 dentry
= dentry_kill(dentry
);
766 /* This must be called with d_lock held */
767 static inline void __dget_dlock(struct dentry
*dentry
)
769 dentry
->d_lockref
.count
++;
772 static inline void __dget(struct dentry
*dentry
)
774 lockref_get(&dentry
->d_lockref
);
777 struct dentry
*dget_parent(struct dentry
*dentry
)
783 * Do optimistic parent lookup without any
787 ret
= ACCESS_ONCE(dentry
->d_parent
);
788 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
790 if (likely(gotref
)) {
791 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
798 * Don't need rcu_dereference because we re-check it was correct under
802 ret
= dentry
->d_parent
;
803 spin_lock(&ret
->d_lock
);
804 if (unlikely(ret
!= dentry
->d_parent
)) {
805 spin_unlock(&ret
->d_lock
);
810 BUG_ON(!ret
->d_lockref
.count
);
811 ret
->d_lockref
.count
++;
812 spin_unlock(&ret
->d_lock
);
815 EXPORT_SYMBOL(dget_parent
);
818 * d_find_alias - grab a hashed alias of inode
819 * @inode: inode in question
821 * If inode has a hashed alias, or is a directory and has any alias,
822 * acquire the reference to alias and return it. Otherwise return NULL.
823 * Notice that if inode is a directory there can be only one alias and
824 * it can be unhashed only if it has no children, or if it is the root
825 * of a filesystem, or if the directory was renamed and d_revalidate
826 * was the first vfs operation to notice.
828 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
829 * any other hashed alias over that one.
831 static struct dentry
*__d_find_alias(struct inode
*inode
)
833 struct dentry
*alias
, *discon_alias
;
837 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
838 spin_lock(&alias
->d_lock
);
839 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
840 if (IS_ROOT(alias
) &&
841 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
842 discon_alias
= alias
;
845 spin_unlock(&alias
->d_lock
);
849 spin_unlock(&alias
->d_lock
);
852 alias
= discon_alias
;
853 spin_lock(&alias
->d_lock
);
854 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
856 spin_unlock(&alias
->d_lock
);
859 spin_unlock(&alias
->d_lock
);
865 struct dentry
*d_find_alias(struct inode
*inode
)
867 struct dentry
*de
= NULL
;
869 if (!hlist_empty(&inode
->i_dentry
)) {
870 spin_lock(&inode
->i_lock
);
871 de
= __d_find_alias(inode
);
872 spin_unlock(&inode
->i_lock
);
876 EXPORT_SYMBOL(d_find_alias
);
879 * Try to kill dentries associated with this inode.
880 * WARNING: you must own a reference to inode.
882 void d_prune_aliases(struct inode
*inode
)
884 struct dentry
*dentry
;
886 spin_lock(&inode
->i_lock
);
887 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
888 spin_lock(&dentry
->d_lock
);
889 if (!dentry
->d_lockref
.count
) {
890 struct dentry
*parent
= lock_parent(dentry
);
891 if (likely(!dentry
->d_lockref
.count
)) {
892 __dentry_kill(dentry
);
897 spin_unlock(&parent
->d_lock
);
899 spin_unlock(&dentry
->d_lock
);
901 spin_unlock(&inode
->i_lock
);
903 EXPORT_SYMBOL(d_prune_aliases
);
905 static void shrink_dentry_list(struct list_head
*list
)
907 struct dentry
*dentry
, *parent
;
909 while (!list_empty(list
)) {
911 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
912 spin_lock(&dentry
->d_lock
);
913 parent
= lock_parent(dentry
);
916 * The dispose list is isolated and dentries are not accounted
917 * to the LRU here, so we can simply remove it from the list
918 * here regardless of whether it is referenced or not.
920 d_shrink_del(dentry
);
923 * We found an inuse dentry which was not removed from
924 * the LRU because of laziness during lookup. Do not free it.
926 if (dentry
->d_lockref
.count
> 0) {
927 spin_unlock(&dentry
->d_lock
);
929 spin_unlock(&parent
->d_lock
);
934 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
935 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
936 spin_unlock(&dentry
->d_lock
);
938 spin_unlock(&parent
->d_lock
);
944 inode
= dentry
->d_inode
;
945 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
946 d_shrink_add(dentry
, list
);
947 spin_unlock(&dentry
->d_lock
);
949 spin_unlock(&parent
->d_lock
);
953 __dentry_kill(dentry
);
956 * We need to prune ancestors too. This is necessary to prevent
957 * quadratic behavior of shrink_dcache_parent(), but is also
958 * expected to be beneficial in reducing dentry cache
962 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
963 parent
= lock_parent(dentry
);
964 if (dentry
->d_lockref
.count
!= 1) {
965 dentry
->d_lockref
.count
--;
966 spin_unlock(&dentry
->d_lock
);
968 spin_unlock(&parent
->d_lock
);
971 inode
= dentry
->d_inode
; /* can't be NULL */
972 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
973 spin_unlock(&dentry
->d_lock
);
975 spin_unlock(&parent
->d_lock
);
979 __dentry_kill(dentry
);
985 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
986 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
988 struct list_head
*freeable
= arg
;
989 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
993 * we are inverting the lru lock/dentry->d_lock here,
994 * so use a trylock. If we fail to get the lock, just skip
997 if (!spin_trylock(&dentry
->d_lock
))
1001 * Referenced dentries are still in use. If they have active
1002 * counts, just remove them from the LRU. Otherwise give them
1003 * another pass through the LRU.
1005 if (dentry
->d_lockref
.count
) {
1006 d_lru_isolate(lru
, dentry
);
1007 spin_unlock(&dentry
->d_lock
);
1011 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1012 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1013 spin_unlock(&dentry
->d_lock
);
1016 * The list move itself will be made by the common LRU code. At
1017 * this point, we've dropped the dentry->d_lock but keep the
1018 * lru lock. This is safe to do, since every list movement is
1019 * protected by the lru lock even if both locks are held.
1021 * This is guaranteed by the fact that all LRU management
1022 * functions are intermediated by the LRU API calls like
1023 * list_lru_add and list_lru_del. List movement in this file
1024 * only ever occur through this functions or through callbacks
1025 * like this one, that are called from the LRU API.
1027 * The only exceptions to this are functions like
1028 * shrink_dentry_list, and code that first checks for the
1029 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1030 * operating only with stack provided lists after they are
1031 * properly isolated from the main list. It is thus, always a
1037 d_lru_shrink_move(lru
, dentry
, freeable
);
1038 spin_unlock(&dentry
->d_lock
);
1044 * prune_dcache_sb - shrink the dcache
1046 * @sc: shrink control, passed to list_lru_shrink_walk()
1048 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1049 * is done when we need more memory and called from the superblock shrinker
1052 * This function may fail to free any resources if all the dentries are in
1055 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1060 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1061 dentry_lru_isolate
, &dispose
);
1062 shrink_dentry_list(&dispose
);
1066 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1067 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1069 struct list_head
*freeable
= arg
;
1070 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1073 * we are inverting the lru lock/dentry->d_lock here,
1074 * so use a trylock. If we fail to get the lock, just skip
1077 if (!spin_trylock(&dentry
->d_lock
))
1080 d_lru_shrink_move(lru
, dentry
, freeable
);
1081 spin_unlock(&dentry
->d_lock
);
1088 * shrink_dcache_sb - shrink dcache for a superblock
1091 * Shrink the dcache for the specified super block. This is used to free
1092 * the dcache before unmounting a file system.
1094 void shrink_dcache_sb(struct super_block
*sb
)
1101 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1102 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1104 this_cpu_sub(nr_dentry_unused
, freed
);
1105 shrink_dentry_list(&dispose
);
1106 } while (freed
> 0);
1108 EXPORT_SYMBOL(shrink_dcache_sb
);
1111 * enum d_walk_ret - action to talke during tree walk
1112 * @D_WALK_CONTINUE: contrinue walk
1113 * @D_WALK_QUIT: quit walk
1114 * @D_WALK_NORETRY: quit when retry is needed
1115 * @D_WALK_SKIP: skip this dentry and its children
1125 * d_walk - walk the dentry tree
1126 * @parent: start of walk
1127 * @data: data passed to @enter() and @finish()
1128 * @enter: callback when first entering the dentry
1129 * @finish: callback when successfully finished the walk
1131 * The @enter() and @finish() callbacks are called with d_lock held.
1133 static void d_walk(struct dentry
*parent
, void *data
,
1134 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1135 void (*finish
)(void *))
1137 struct dentry
*this_parent
;
1138 struct list_head
*next
;
1140 enum d_walk_ret ret
;
1144 read_seqbegin_or_lock(&rename_lock
, &seq
);
1145 this_parent
= parent
;
1146 spin_lock(&this_parent
->d_lock
);
1148 ret
= enter(data
, this_parent
);
1150 case D_WALK_CONTINUE
:
1155 case D_WALK_NORETRY
:
1160 next
= this_parent
->d_subdirs
.next
;
1162 while (next
!= &this_parent
->d_subdirs
) {
1163 struct list_head
*tmp
= next
;
1164 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1167 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1169 ret
= enter(data
, dentry
);
1171 case D_WALK_CONTINUE
:
1174 spin_unlock(&dentry
->d_lock
);
1176 case D_WALK_NORETRY
:
1180 spin_unlock(&dentry
->d_lock
);
1184 if (!list_empty(&dentry
->d_subdirs
)) {
1185 spin_unlock(&this_parent
->d_lock
);
1186 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1187 this_parent
= dentry
;
1188 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1191 spin_unlock(&dentry
->d_lock
);
1194 * All done at this level ... ascend and resume the search.
1198 if (this_parent
!= parent
) {
1199 struct dentry
*child
= this_parent
;
1200 this_parent
= child
->d_parent
;
1202 spin_unlock(&child
->d_lock
);
1203 spin_lock(&this_parent
->d_lock
);
1205 /* might go back up the wrong parent if we have had a rename. */
1206 if (need_seqretry(&rename_lock
, seq
))
1208 next
= child
->d_child
.next
;
1209 while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
)) {
1210 if (next
== &this_parent
->d_subdirs
)
1212 child
= list_entry(next
, struct dentry
, d_child
);
1218 if (need_seqretry(&rename_lock
, seq
))
1225 spin_unlock(&this_parent
->d_lock
);
1226 done_seqretry(&rename_lock
, seq
);
1230 spin_unlock(&this_parent
->d_lock
);
1240 * Search for at least 1 mount point in the dentry's subdirs.
1241 * We descend to the next level whenever the d_subdirs
1242 * list is non-empty and continue searching.
1245 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1248 if (d_mountpoint(dentry
)) {
1252 return D_WALK_CONTINUE
;
1256 * have_submounts - check for mounts over a dentry
1257 * @parent: dentry to check.
1259 * Return true if the parent or its subdirectories contain
1262 int have_submounts(struct dentry
*parent
)
1266 d_walk(parent
, &ret
, check_mount
, NULL
);
1270 EXPORT_SYMBOL(have_submounts
);
1273 * Called by mount code to set a mountpoint and check if the mountpoint is
1274 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1275 * subtree can become unreachable).
1277 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1278 * this reason take rename_lock and d_lock on dentry and ancestors.
1280 int d_set_mounted(struct dentry
*dentry
)
1284 write_seqlock(&rename_lock
);
1285 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1286 /* Need exclusion wrt. d_invalidate() */
1287 spin_lock(&p
->d_lock
);
1288 if (unlikely(d_unhashed(p
))) {
1289 spin_unlock(&p
->d_lock
);
1292 spin_unlock(&p
->d_lock
);
1294 spin_lock(&dentry
->d_lock
);
1295 if (!d_unlinked(dentry
)) {
1296 dentry
->d_flags
|= DCACHE_MOUNTED
;
1299 spin_unlock(&dentry
->d_lock
);
1301 write_sequnlock(&rename_lock
);
1306 * Search the dentry child list of the specified parent,
1307 * and move any unused dentries to the end of the unused
1308 * list for prune_dcache(). We descend to the next level
1309 * whenever the d_subdirs list is non-empty and continue
1312 * It returns zero iff there are no unused children,
1313 * otherwise it returns the number of children moved to
1314 * the end of the unused list. This may not be the total
1315 * number of unused children, because select_parent can
1316 * drop the lock and return early due to latency
1320 struct select_data
{
1321 struct dentry
*start
;
1322 struct list_head dispose
;
1326 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1328 struct select_data
*data
= _data
;
1329 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1331 if (data
->start
== dentry
)
1334 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1337 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1339 if (!dentry
->d_lockref
.count
) {
1340 d_shrink_add(dentry
, &data
->dispose
);
1345 * We can return to the caller if we have found some (this
1346 * ensures forward progress). We'll be coming back to find
1349 if (!list_empty(&data
->dispose
))
1350 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1356 * shrink_dcache_parent - prune dcache
1357 * @parent: parent of entries to prune
1359 * Prune the dcache to remove unused children of the parent dentry.
1361 void shrink_dcache_parent(struct dentry
*parent
)
1364 struct select_data data
;
1366 INIT_LIST_HEAD(&data
.dispose
);
1367 data
.start
= parent
;
1370 d_walk(parent
, &data
, select_collect
, NULL
);
1374 shrink_dentry_list(&data
.dispose
);
1378 EXPORT_SYMBOL(shrink_dcache_parent
);
1380 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1382 /* it has busy descendents; complain about those instead */
1383 if (!list_empty(&dentry
->d_subdirs
))
1384 return D_WALK_CONTINUE
;
1386 /* root with refcount 1 is fine */
1387 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1388 return D_WALK_CONTINUE
;
1390 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1391 " still in use (%d) [unmount of %s %s]\n",
1394 dentry
->d_inode
->i_ino
: 0UL,
1396 dentry
->d_lockref
.count
,
1397 dentry
->d_sb
->s_type
->name
,
1398 dentry
->d_sb
->s_id
);
1400 return D_WALK_CONTINUE
;
1403 static void do_one_tree(struct dentry
*dentry
)
1405 shrink_dcache_parent(dentry
);
1406 d_walk(dentry
, dentry
, umount_check
, NULL
);
1412 * destroy the dentries attached to a superblock on unmounting
1414 void shrink_dcache_for_umount(struct super_block
*sb
)
1416 struct dentry
*dentry
;
1418 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1420 dentry
= sb
->s_root
;
1422 do_one_tree(dentry
);
1424 while (!hlist_bl_empty(&sb
->s_anon
)) {
1425 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1426 do_one_tree(dentry
);
1430 struct detach_data
{
1431 struct select_data select
;
1432 struct dentry
*mountpoint
;
1434 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1436 struct detach_data
*data
= _data
;
1438 if (d_mountpoint(dentry
)) {
1439 __dget_dlock(dentry
);
1440 data
->mountpoint
= dentry
;
1444 return select_collect(&data
->select
, dentry
);
1447 static void check_and_drop(void *_data
)
1449 struct detach_data
*data
= _data
;
1451 if (!data
->mountpoint
&& !data
->select
.found
)
1452 __d_drop(data
->select
.start
);
1456 * d_invalidate - detach submounts, prune dcache, and drop
1457 * @dentry: dentry to invalidate (aka detach, prune and drop)
1461 * The final d_drop is done as an atomic operation relative to
1462 * rename_lock ensuring there are no races with d_set_mounted. This
1463 * ensures there are no unhashed dentries on the path to a mountpoint.
1465 void d_invalidate(struct dentry
*dentry
)
1468 * If it's already been dropped, return OK.
1470 spin_lock(&dentry
->d_lock
);
1471 if (d_unhashed(dentry
)) {
1472 spin_unlock(&dentry
->d_lock
);
1475 spin_unlock(&dentry
->d_lock
);
1477 /* Negative dentries can be dropped without further checks */
1478 if (!dentry
->d_inode
) {
1484 struct detach_data data
;
1486 data
.mountpoint
= NULL
;
1487 INIT_LIST_HEAD(&data
.select
.dispose
);
1488 data
.select
.start
= dentry
;
1489 data
.select
.found
= 0;
1491 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1493 if (data
.select
.found
)
1494 shrink_dentry_list(&data
.select
.dispose
);
1496 if (data
.mountpoint
) {
1497 detach_mounts(data
.mountpoint
);
1498 dput(data
.mountpoint
);
1501 if (!data
.mountpoint
&& !data
.select
.found
)
1507 EXPORT_SYMBOL(d_invalidate
);
1510 * __d_alloc - allocate a dcache entry
1511 * @sb: filesystem it will belong to
1512 * @name: qstr of the name
1514 * Allocates a dentry. It returns %NULL if there is insufficient memory
1515 * available. On a success the dentry is returned. The name passed in is
1516 * copied and the copy passed in may be reused after this call.
1519 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1521 struct dentry
*dentry
;
1524 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1529 * We guarantee that the inline name is always NUL-terminated.
1530 * This way the memcpy() done by the name switching in rename
1531 * will still always have a NUL at the end, even if we might
1532 * be overwriting an internal NUL character
1534 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1535 if (name
->len
> DNAME_INLINE_LEN
-1) {
1536 size_t size
= offsetof(struct external_name
, name
[1]);
1537 struct external_name
*p
= kmalloc(size
+ name
->len
, GFP_KERNEL
);
1539 kmem_cache_free(dentry_cache
, dentry
);
1542 atomic_set(&p
->u
.count
, 1);
1544 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1545 kasan_unpoison_shadow(dname
,
1546 round_up(name
->len
+ 1, sizeof(unsigned long)));
1548 dname
= dentry
->d_iname
;
1551 dentry
->d_name
.len
= name
->len
;
1552 dentry
->d_name
.hash
= name
->hash
;
1553 memcpy(dname
, name
->name
, name
->len
);
1554 dname
[name
->len
] = 0;
1556 /* Make sure we always see the terminating NUL character */
1558 dentry
->d_name
.name
= dname
;
1560 dentry
->d_lockref
.count
= 1;
1561 dentry
->d_flags
= 0;
1562 spin_lock_init(&dentry
->d_lock
);
1563 seqcount_init(&dentry
->d_seq
);
1564 dentry
->d_inode
= NULL
;
1565 dentry
->d_parent
= dentry
;
1567 dentry
->d_op
= NULL
;
1568 dentry
->d_fsdata
= NULL
;
1569 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1570 INIT_LIST_HEAD(&dentry
->d_lru
);
1571 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1572 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1573 INIT_LIST_HEAD(&dentry
->d_child
);
1574 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1576 this_cpu_inc(nr_dentry
);
1582 * d_alloc - allocate a dcache entry
1583 * @parent: parent of entry to allocate
1584 * @name: qstr of the name
1586 * Allocates a dentry. It returns %NULL if there is insufficient memory
1587 * available. On a success the dentry is returned. The name passed in is
1588 * copied and the copy passed in may be reused after this call.
1590 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1592 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1596 spin_lock(&parent
->d_lock
);
1598 * don't need child lock because it is not subject
1599 * to concurrency here
1601 __dget_dlock(parent
);
1602 dentry
->d_parent
= parent
;
1603 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1604 spin_unlock(&parent
->d_lock
);
1608 EXPORT_SYMBOL(d_alloc
);
1611 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1612 * @sb: the superblock
1613 * @name: qstr of the name
1615 * For a filesystem that just pins its dentries in memory and never
1616 * performs lookups at all, return an unhashed IS_ROOT dentry.
1618 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1620 return __d_alloc(sb
, name
);
1622 EXPORT_SYMBOL(d_alloc_pseudo
);
1624 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1629 q
.len
= strlen(name
);
1630 q
.hash
= full_name_hash(q
.name
, q
.len
);
1631 return d_alloc(parent
, &q
);
1633 EXPORT_SYMBOL(d_alloc_name
);
1635 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1637 WARN_ON_ONCE(dentry
->d_op
);
1638 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1640 DCACHE_OP_REVALIDATE
|
1641 DCACHE_OP_WEAK_REVALIDATE
|
1642 DCACHE_OP_DELETE
));
1647 dentry
->d_flags
|= DCACHE_OP_HASH
;
1649 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1650 if (op
->d_revalidate
)
1651 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1652 if (op
->d_weak_revalidate
)
1653 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1655 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1657 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1660 EXPORT_SYMBOL(d_set_d_op
);
1664 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1665 * @dentry - The dentry to mark
1667 * Mark a dentry as falling through to the lower layer (as set with
1668 * d_pin_lower()). This flag may be recorded on the medium.
1670 void d_set_fallthru(struct dentry
*dentry
)
1672 spin_lock(&dentry
->d_lock
);
1673 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1674 spin_unlock(&dentry
->d_lock
);
1676 EXPORT_SYMBOL(d_set_fallthru
);
1678 static unsigned d_flags_for_inode(struct inode
*inode
)
1680 unsigned add_flags
= DCACHE_FILE_TYPE
;
1683 return DCACHE_MISS_TYPE
;
1685 if (S_ISDIR(inode
->i_mode
)) {
1686 add_flags
= DCACHE_DIRECTORY_TYPE
;
1687 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1688 if (unlikely(!inode
->i_op
->lookup
))
1689 add_flags
= DCACHE_AUTODIR_TYPE
;
1691 inode
->i_opflags
|= IOP_LOOKUP
;
1693 } else if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1694 if (unlikely(inode
->i_op
->follow_link
))
1695 add_flags
= DCACHE_SYMLINK_TYPE
;
1697 inode
->i_opflags
|= IOP_NOFOLLOW
;
1700 if (unlikely(IS_AUTOMOUNT(inode
)))
1701 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1705 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1707 unsigned add_flags
= d_flags_for_inode(inode
);
1709 spin_lock(&dentry
->d_lock
);
1710 dentry
->d_flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
1711 dentry
->d_flags
|= add_flags
;
1713 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1714 dentry
->d_inode
= inode
;
1715 dentry_rcuwalk_barrier(dentry
);
1716 spin_unlock(&dentry
->d_lock
);
1717 fsnotify_d_instantiate(dentry
, inode
);
1721 * d_instantiate - fill in inode information for a dentry
1722 * @entry: dentry to complete
1723 * @inode: inode to attach to this dentry
1725 * Fill in inode information in the entry.
1727 * This turns negative dentries into productive full members
1730 * NOTE! This assumes that the inode count has been incremented
1731 * (or otherwise set) by the caller to indicate that it is now
1732 * in use by the dcache.
1735 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1737 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1739 spin_lock(&inode
->i_lock
);
1740 __d_instantiate(entry
, inode
);
1742 spin_unlock(&inode
->i_lock
);
1743 security_d_instantiate(entry
, inode
);
1745 EXPORT_SYMBOL(d_instantiate
);
1748 * d_instantiate_unique - instantiate a non-aliased dentry
1749 * @entry: dentry to instantiate
1750 * @inode: inode to attach to this dentry
1752 * Fill in inode information in the entry. On success, it returns NULL.
1753 * If an unhashed alias of "entry" already exists, then we return the
1754 * aliased dentry instead and drop one reference to inode.
1756 * Note that in order to avoid conflicts with rename() etc, the caller
1757 * had better be holding the parent directory semaphore.
1759 * This also assumes that the inode count has been incremented
1760 * (or otherwise set) by the caller to indicate that it is now
1761 * in use by the dcache.
1763 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1764 struct inode
*inode
)
1766 struct dentry
*alias
;
1767 int len
= entry
->d_name
.len
;
1768 const char *name
= entry
->d_name
.name
;
1769 unsigned int hash
= entry
->d_name
.hash
;
1772 __d_instantiate(entry
, NULL
);
1776 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1778 * Don't need alias->d_lock here, because aliases with
1779 * d_parent == entry->d_parent are not subject to name or
1780 * parent changes, because the parent inode i_mutex is held.
1782 if (alias
->d_name
.hash
!= hash
)
1784 if (alias
->d_parent
!= entry
->d_parent
)
1786 if (alias
->d_name
.len
!= len
)
1788 if (dentry_cmp(alias
, name
, len
))
1794 __d_instantiate(entry
, inode
);
1798 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1800 struct dentry
*result
;
1802 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1805 spin_lock(&inode
->i_lock
);
1806 result
= __d_instantiate_unique(entry
, inode
);
1808 spin_unlock(&inode
->i_lock
);
1811 security_d_instantiate(entry
, inode
);
1815 BUG_ON(!d_unhashed(result
));
1820 EXPORT_SYMBOL(d_instantiate_unique
);
1823 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1824 * @entry: dentry to complete
1825 * @inode: inode to attach to this dentry
1827 * Fill in inode information in the entry. If a directory alias is found, then
1828 * return an error (and drop inode). Together with d_materialise_unique() this
1829 * guarantees that a directory inode may never have more than one alias.
1831 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1833 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1835 spin_lock(&inode
->i_lock
);
1836 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1837 spin_unlock(&inode
->i_lock
);
1841 __d_instantiate(entry
, inode
);
1842 spin_unlock(&inode
->i_lock
);
1843 security_d_instantiate(entry
, inode
);
1847 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1849 struct dentry
*d_make_root(struct inode
*root_inode
)
1851 struct dentry
*res
= NULL
;
1854 static const struct qstr name
= QSTR_INIT("/", 1);
1856 res
= __d_alloc(root_inode
->i_sb
, &name
);
1858 d_instantiate(res
, root_inode
);
1864 EXPORT_SYMBOL(d_make_root
);
1866 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1868 struct dentry
*alias
;
1870 if (hlist_empty(&inode
->i_dentry
))
1872 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1878 * d_find_any_alias - find any alias for a given inode
1879 * @inode: inode to find an alias for
1881 * If any aliases exist for the given inode, take and return a
1882 * reference for one of them. If no aliases exist, return %NULL.
1884 struct dentry
*d_find_any_alias(struct inode
*inode
)
1888 spin_lock(&inode
->i_lock
);
1889 de
= __d_find_any_alias(inode
);
1890 spin_unlock(&inode
->i_lock
);
1893 EXPORT_SYMBOL(d_find_any_alias
);
1895 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1897 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1903 return ERR_PTR(-ESTALE
);
1905 return ERR_CAST(inode
);
1907 res
= d_find_any_alias(inode
);
1911 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1913 res
= ERR_PTR(-ENOMEM
);
1917 spin_lock(&inode
->i_lock
);
1918 res
= __d_find_any_alias(inode
);
1920 spin_unlock(&inode
->i_lock
);
1925 /* attach a disconnected dentry */
1926 add_flags
= d_flags_for_inode(inode
);
1929 add_flags
|= DCACHE_DISCONNECTED
;
1931 spin_lock(&tmp
->d_lock
);
1932 tmp
->d_inode
= inode
;
1933 tmp
->d_flags
|= add_flags
;
1934 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1935 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1936 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1937 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1938 spin_unlock(&tmp
->d_lock
);
1939 spin_unlock(&inode
->i_lock
);
1940 security_d_instantiate(tmp
, inode
);
1945 if (res
&& !IS_ERR(res
))
1946 security_d_instantiate(res
, inode
);
1952 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1953 * @inode: inode to allocate the dentry for
1955 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1956 * similar open by handle operations. The returned dentry may be anonymous,
1957 * or may have a full name (if the inode was already in the cache).
1959 * When called on a directory inode, we must ensure that the inode only ever
1960 * has one dentry. If a dentry is found, that is returned instead of
1961 * allocating a new one.
1963 * On successful return, the reference to the inode has been transferred
1964 * to the dentry. In case of an error the reference on the inode is released.
1965 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1966 * be passed in and the error will be propagated to the return value,
1967 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1969 struct dentry
*d_obtain_alias(struct inode
*inode
)
1971 return __d_obtain_alias(inode
, 1);
1973 EXPORT_SYMBOL(d_obtain_alias
);
1976 * d_obtain_root - find or allocate a dentry for a given inode
1977 * @inode: inode to allocate the dentry for
1979 * Obtain an IS_ROOT dentry for the root of a filesystem.
1981 * We must ensure that directory inodes only ever have one dentry. If a
1982 * dentry is found, that is returned instead of allocating a new one.
1984 * On successful return, the reference to the inode has been transferred
1985 * to the dentry. In case of an error the reference on the inode is
1986 * released. A %NULL or IS_ERR inode may be passed in and will be the
1987 * error will be propagate to the return value, with a %NULL @inode
1988 * replaced by ERR_PTR(-ESTALE).
1990 struct dentry
*d_obtain_root(struct inode
*inode
)
1992 return __d_obtain_alias(inode
, 0);
1994 EXPORT_SYMBOL(d_obtain_root
);
1997 * d_add_ci - lookup or allocate new dentry with case-exact name
1998 * @inode: the inode case-insensitive lookup has found
1999 * @dentry: the negative dentry that was passed to the parent's lookup func
2000 * @name: the case-exact name to be associated with the returned dentry
2002 * This is to avoid filling the dcache with case-insensitive names to the
2003 * same inode, only the actual correct case is stored in the dcache for
2004 * case-insensitive filesystems.
2006 * For a case-insensitive lookup match and if the the case-exact dentry
2007 * already exists in in the dcache, use it and return it.
2009 * If no entry exists with the exact case name, allocate new dentry with
2010 * the exact case, and return the spliced entry.
2012 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2015 struct dentry
*found
;
2019 * First check if a dentry matching the name already exists,
2020 * if not go ahead and create it now.
2022 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2024 new = d_alloc(dentry
->d_parent
, name
);
2026 found
= ERR_PTR(-ENOMEM
);
2028 found
= d_splice_alias(inode
, new);
2039 EXPORT_SYMBOL(d_add_ci
);
2042 * Do the slow-case of the dentry name compare.
2044 * Unlike the dentry_cmp() function, we need to atomically
2045 * load the name and length information, so that the
2046 * filesystem can rely on them, and can use the 'name' and
2047 * 'len' information without worrying about walking off the
2048 * end of memory etc.
2050 * Thus the read_seqcount_retry() and the "duplicate" info
2051 * in arguments (the low-level filesystem should not look
2052 * at the dentry inode or name contents directly, since
2053 * rename can change them while we're in RCU mode).
2055 enum slow_d_compare
{
2061 static noinline
enum slow_d_compare
slow_dentry_cmp(
2062 const struct dentry
*parent
,
2063 struct dentry
*dentry
,
2065 const struct qstr
*name
)
2067 int tlen
= dentry
->d_name
.len
;
2068 const char *tname
= dentry
->d_name
.name
;
2070 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2072 return D_COMP_SEQRETRY
;
2074 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2075 return D_COMP_NOMATCH
;
2080 * __d_lookup_rcu - search for a dentry (racy, store-free)
2081 * @parent: parent dentry
2082 * @name: qstr of name we wish to find
2083 * @seqp: returns d_seq value at the point where the dentry was found
2084 * Returns: dentry, or NULL
2086 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2087 * resolution (store-free path walking) design described in
2088 * Documentation/filesystems/path-lookup.txt.
2090 * This is not to be used outside core vfs.
2092 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2093 * held, and rcu_read_lock held. The returned dentry must not be stored into
2094 * without taking d_lock and checking d_seq sequence count against @seq
2097 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2100 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2101 * the returned dentry, so long as its parent's seqlock is checked after the
2102 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2103 * is formed, giving integrity down the path walk.
2105 * NOTE! The caller *has* to check the resulting dentry against the sequence
2106 * number we've returned before using any of the resulting dentry state!
2108 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2109 const struct qstr
*name
,
2112 u64 hashlen
= name
->hash_len
;
2113 const unsigned char *str
= name
->name
;
2114 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2115 struct hlist_bl_node
*node
;
2116 struct dentry
*dentry
;
2119 * Note: There is significant duplication with __d_lookup_rcu which is
2120 * required to prevent single threaded performance regressions
2121 * especially on architectures where smp_rmb (in seqcounts) are costly.
2122 * Keep the two functions in sync.
2126 * The hash list is protected using RCU.
2128 * Carefully use d_seq when comparing a candidate dentry, to avoid
2129 * races with d_move().
2131 * It is possible that concurrent renames can mess up our list
2132 * walk here and result in missing our dentry, resulting in the
2133 * false-negative result. d_lookup() protects against concurrent
2134 * renames using rename_lock seqlock.
2136 * See Documentation/filesystems/path-lookup.txt for more details.
2138 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2143 * The dentry sequence count protects us from concurrent
2144 * renames, and thus protects parent and name fields.
2146 * The caller must perform a seqcount check in order
2147 * to do anything useful with the returned dentry.
2149 * NOTE! We do a "raw" seqcount_begin here. That means that
2150 * we don't wait for the sequence count to stabilize if it
2151 * is in the middle of a sequence change. If we do the slow
2152 * dentry compare, we will do seqretries until it is stable,
2153 * and if we end up with a successful lookup, we actually
2154 * want to exit RCU lookup anyway.
2156 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2157 if (dentry
->d_parent
!= parent
)
2159 if (d_unhashed(dentry
))
2162 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2163 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2166 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2169 case D_COMP_NOMATCH
:
2176 if (dentry
->d_name
.hash_len
!= hashlen
)
2179 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2186 * d_lookup - search for a dentry
2187 * @parent: parent dentry
2188 * @name: qstr of name we wish to find
2189 * Returns: dentry, or NULL
2191 * d_lookup searches the children of the parent dentry for the name in
2192 * question. If the dentry is found its reference count is incremented and the
2193 * dentry is returned. The caller must use dput to free the entry when it has
2194 * finished using it. %NULL is returned if the dentry does not exist.
2196 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2198 struct dentry
*dentry
;
2202 seq
= read_seqbegin(&rename_lock
);
2203 dentry
= __d_lookup(parent
, name
);
2206 } while (read_seqretry(&rename_lock
, seq
));
2209 EXPORT_SYMBOL(d_lookup
);
2212 * __d_lookup - search for a dentry (racy)
2213 * @parent: parent dentry
2214 * @name: qstr of name we wish to find
2215 * Returns: dentry, or NULL
2217 * __d_lookup is like d_lookup, however it may (rarely) return a
2218 * false-negative result due to unrelated rename activity.
2220 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2221 * however it must be used carefully, eg. with a following d_lookup in
2222 * the case of failure.
2224 * __d_lookup callers must be commented.
2226 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2228 unsigned int len
= name
->len
;
2229 unsigned int hash
= name
->hash
;
2230 const unsigned char *str
= name
->name
;
2231 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2232 struct hlist_bl_node
*node
;
2233 struct dentry
*found
= NULL
;
2234 struct dentry
*dentry
;
2237 * Note: There is significant duplication with __d_lookup_rcu which is
2238 * required to prevent single threaded performance regressions
2239 * especially on architectures where smp_rmb (in seqcounts) are costly.
2240 * Keep the two functions in sync.
2244 * The hash list is protected using RCU.
2246 * Take d_lock when comparing a candidate dentry, to avoid races
2249 * It is possible that concurrent renames can mess up our list
2250 * walk here and result in missing our dentry, resulting in the
2251 * false-negative result. d_lookup() protects against concurrent
2252 * renames using rename_lock seqlock.
2254 * See Documentation/filesystems/path-lookup.txt for more details.
2258 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2260 if (dentry
->d_name
.hash
!= hash
)
2263 spin_lock(&dentry
->d_lock
);
2264 if (dentry
->d_parent
!= parent
)
2266 if (d_unhashed(dentry
))
2270 * It is safe to compare names since d_move() cannot
2271 * change the qstr (protected by d_lock).
2273 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2274 int tlen
= dentry
->d_name
.len
;
2275 const char *tname
= dentry
->d_name
.name
;
2276 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2279 if (dentry
->d_name
.len
!= len
)
2281 if (dentry_cmp(dentry
, str
, len
))
2285 dentry
->d_lockref
.count
++;
2287 spin_unlock(&dentry
->d_lock
);
2290 spin_unlock(&dentry
->d_lock
);
2298 * d_hash_and_lookup - hash the qstr then search for a dentry
2299 * @dir: Directory to search in
2300 * @name: qstr of name we wish to find
2302 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2304 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2307 * Check for a fs-specific hash function. Note that we must
2308 * calculate the standard hash first, as the d_op->d_hash()
2309 * routine may choose to leave the hash value unchanged.
2311 name
->hash
= full_name_hash(name
->name
, name
->len
);
2312 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2313 int err
= dir
->d_op
->d_hash(dir
, name
);
2314 if (unlikely(err
< 0))
2315 return ERR_PTR(err
);
2317 return d_lookup(dir
, name
);
2319 EXPORT_SYMBOL(d_hash_and_lookup
);
2322 * When a file is deleted, we have two options:
2323 * - turn this dentry into a negative dentry
2324 * - unhash this dentry and free it.
2326 * Usually, we want to just turn this into
2327 * a negative dentry, but if anybody else is
2328 * currently using the dentry or the inode
2329 * we can't do that and we fall back on removing
2330 * it from the hash queues and waiting for
2331 * it to be deleted later when it has no users
2335 * d_delete - delete a dentry
2336 * @dentry: The dentry to delete
2338 * Turn the dentry into a negative dentry if possible, otherwise
2339 * remove it from the hash queues so it can be deleted later
2342 void d_delete(struct dentry
* dentry
)
2344 struct inode
*inode
;
2347 * Are we the only user?
2350 spin_lock(&dentry
->d_lock
);
2351 inode
= dentry
->d_inode
;
2352 isdir
= S_ISDIR(inode
->i_mode
);
2353 if (dentry
->d_lockref
.count
== 1) {
2354 if (!spin_trylock(&inode
->i_lock
)) {
2355 spin_unlock(&dentry
->d_lock
);
2359 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2360 dentry_unlink_inode(dentry
);
2361 fsnotify_nameremove(dentry
, isdir
);
2365 if (!d_unhashed(dentry
))
2368 spin_unlock(&dentry
->d_lock
);
2370 fsnotify_nameremove(dentry
, isdir
);
2372 EXPORT_SYMBOL(d_delete
);
2374 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2376 BUG_ON(!d_unhashed(entry
));
2378 entry
->d_flags
|= DCACHE_RCUACCESS
;
2379 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2383 static void _d_rehash(struct dentry
* entry
)
2385 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2389 * d_rehash - add an entry back to the hash
2390 * @entry: dentry to add to the hash
2392 * Adds a dentry to the hash according to its name.
2395 void d_rehash(struct dentry
* entry
)
2397 spin_lock(&entry
->d_lock
);
2399 spin_unlock(&entry
->d_lock
);
2401 EXPORT_SYMBOL(d_rehash
);
2404 * dentry_update_name_case - update case insensitive dentry with a new name
2405 * @dentry: dentry to be updated
2408 * Update a case insensitive dentry with new case of name.
2410 * dentry must have been returned by d_lookup with name @name. Old and new
2411 * name lengths must match (ie. no d_compare which allows mismatched name
2414 * Parent inode i_mutex must be held over d_lookup and into this call (to
2415 * keep renames and concurrent inserts, and readdir(2) away).
2417 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2419 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2420 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2422 spin_lock(&dentry
->d_lock
);
2423 write_seqcount_begin(&dentry
->d_seq
);
2424 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2425 write_seqcount_end(&dentry
->d_seq
);
2426 spin_unlock(&dentry
->d_lock
);
2428 EXPORT_SYMBOL(dentry_update_name_case
);
2430 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2432 if (unlikely(dname_external(target
))) {
2433 if (unlikely(dname_external(dentry
))) {
2435 * Both external: swap the pointers
2437 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2440 * dentry:internal, target:external. Steal target's
2441 * storage and make target internal.
2443 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2444 dentry
->d_name
.len
+ 1);
2445 dentry
->d_name
.name
= target
->d_name
.name
;
2446 target
->d_name
.name
= target
->d_iname
;
2449 if (unlikely(dname_external(dentry
))) {
2451 * dentry:external, target:internal. Give dentry's
2452 * storage to target and make dentry internal
2454 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2455 target
->d_name
.len
+ 1);
2456 target
->d_name
.name
= dentry
->d_name
.name
;
2457 dentry
->d_name
.name
= dentry
->d_iname
;
2460 * Both are internal.
2463 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2464 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2465 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2466 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2467 swap(((long *) &dentry
->d_iname
)[i
],
2468 ((long *) &target
->d_iname
)[i
]);
2472 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2475 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2477 struct external_name
*old_name
= NULL
;
2478 if (unlikely(dname_external(dentry
)))
2479 old_name
= external_name(dentry
);
2480 if (unlikely(dname_external(target
))) {
2481 atomic_inc(&external_name(target
)->u
.count
);
2482 dentry
->d_name
= target
->d_name
;
2484 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2485 target
->d_name
.len
+ 1);
2486 dentry
->d_name
.name
= dentry
->d_iname
;
2487 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2489 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2490 kfree_rcu(old_name
, u
.head
);
2493 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2496 * XXXX: do we really need to take target->d_lock?
2498 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2499 spin_lock(&target
->d_parent
->d_lock
);
2501 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2502 spin_lock(&dentry
->d_parent
->d_lock
);
2503 spin_lock_nested(&target
->d_parent
->d_lock
,
2504 DENTRY_D_LOCK_NESTED
);
2506 spin_lock(&target
->d_parent
->d_lock
);
2507 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2508 DENTRY_D_LOCK_NESTED
);
2511 if (target
< dentry
) {
2512 spin_lock_nested(&target
->d_lock
, 2);
2513 spin_lock_nested(&dentry
->d_lock
, 3);
2515 spin_lock_nested(&dentry
->d_lock
, 2);
2516 spin_lock_nested(&target
->d_lock
, 3);
2520 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2522 if (target
->d_parent
!= dentry
->d_parent
)
2523 spin_unlock(&dentry
->d_parent
->d_lock
);
2524 if (target
->d_parent
!= target
)
2525 spin_unlock(&target
->d_parent
->d_lock
);
2526 spin_unlock(&target
->d_lock
);
2527 spin_unlock(&dentry
->d_lock
);
2531 * When switching names, the actual string doesn't strictly have to
2532 * be preserved in the target - because we're dropping the target
2533 * anyway. As such, we can just do a simple memcpy() to copy over
2534 * the new name before we switch, unless we are going to rehash
2535 * it. Note that if we *do* unhash the target, we are not allowed
2536 * to rehash it without giving it a new name/hash key - whether
2537 * we swap or overwrite the names here, resulting name won't match
2538 * the reality in filesystem; it's only there for d_path() purposes.
2539 * Note that all of this is happening under rename_lock, so the
2540 * any hash lookup seeing it in the middle of manipulations will
2541 * be discarded anyway. So we do not care what happens to the hash
2545 * __d_move - move a dentry
2546 * @dentry: entry to move
2547 * @target: new dentry
2548 * @exchange: exchange the two dentries
2550 * Update the dcache to reflect the move of a file name. Negative
2551 * dcache entries should not be moved in this way. Caller must hold
2552 * rename_lock, the i_mutex of the source and target directories,
2553 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2555 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2558 if (!dentry
->d_inode
)
2559 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2561 BUG_ON(d_ancestor(dentry
, target
));
2562 BUG_ON(d_ancestor(target
, dentry
));
2564 dentry_lock_for_move(dentry
, target
);
2566 write_seqcount_begin(&dentry
->d_seq
);
2567 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2569 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2572 * Move the dentry to the target hash queue. Don't bother checking
2573 * for the same hash queue because of how unlikely it is.
2576 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2579 * Unhash the target (d_delete() is not usable here). If exchanging
2580 * the two dentries, then rehash onto the other's hash queue.
2585 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2588 /* Switch the names.. */
2590 swap_names(dentry
, target
);
2592 copy_name(dentry
, target
);
2594 /* ... and switch them in the tree */
2595 if (IS_ROOT(dentry
)) {
2596 /* splicing a tree */
2597 dentry
->d_parent
= target
->d_parent
;
2598 target
->d_parent
= target
;
2599 list_del_init(&target
->d_child
);
2600 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2602 /* swapping two dentries */
2603 swap(dentry
->d_parent
, target
->d_parent
);
2604 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2605 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2607 fsnotify_d_move(target
);
2608 fsnotify_d_move(dentry
);
2611 write_seqcount_end(&target
->d_seq
);
2612 write_seqcount_end(&dentry
->d_seq
);
2614 dentry_unlock_for_move(dentry
, target
);
2618 * d_move - move a dentry
2619 * @dentry: entry to move
2620 * @target: new dentry
2622 * Update the dcache to reflect the move of a file name. Negative
2623 * dcache entries should not be moved in this way. See the locking
2624 * requirements for __d_move.
2626 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2628 write_seqlock(&rename_lock
);
2629 __d_move(dentry
, target
, false);
2630 write_sequnlock(&rename_lock
);
2632 EXPORT_SYMBOL(d_move
);
2635 * d_exchange - exchange two dentries
2636 * @dentry1: first dentry
2637 * @dentry2: second dentry
2639 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2641 write_seqlock(&rename_lock
);
2643 WARN_ON(!dentry1
->d_inode
);
2644 WARN_ON(!dentry2
->d_inode
);
2645 WARN_ON(IS_ROOT(dentry1
));
2646 WARN_ON(IS_ROOT(dentry2
));
2648 __d_move(dentry1
, dentry2
, true);
2650 write_sequnlock(&rename_lock
);
2654 * d_ancestor - search for an ancestor
2655 * @p1: ancestor dentry
2658 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2659 * an ancestor of p2, else NULL.
2661 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2665 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2666 if (p
->d_parent
== p1
)
2673 * This helper attempts to cope with remotely renamed directories
2675 * It assumes that the caller is already holding
2676 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2678 * Note: If ever the locking in lock_rename() changes, then please
2679 * remember to update this too...
2681 static int __d_unalias(struct inode
*inode
,
2682 struct dentry
*dentry
, struct dentry
*alias
)
2684 struct mutex
*m1
= NULL
, *m2
= NULL
;
2687 /* If alias and dentry share a parent, then no extra locks required */
2688 if (alias
->d_parent
== dentry
->d_parent
)
2691 /* See lock_rename() */
2692 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2694 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2695 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2697 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2699 __d_move(alias
, dentry
, false);
2702 spin_unlock(&inode
->i_lock
);
2711 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2712 * @inode: the inode which may have a disconnected dentry
2713 * @dentry: a negative dentry which we want to point to the inode.
2715 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2716 * place of the given dentry and return it, else simply d_add the inode
2717 * to the dentry and return NULL.
2719 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2720 * we should error out: directories can't have multiple aliases.
2722 * This is needed in the lookup routine of any filesystem that is exportable
2723 * (via knfsd) so that we can build dcache paths to directories effectively.
2725 * If a dentry was found and moved, then it is returned. Otherwise NULL
2726 * is returned. This matches the expected return value of ->lookup.
2728 * Cluster filesystems may call this function with a negative, hashed dentry.
2729 * In that case, we know that the inode will be a regular file, and also this
2730 * will only occur during atomic_open. So we need to check for the dentry
2731 * being already hashed only in the final case.
2733 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2736 return ERR_CAST(inode
);
2738 BUG_ON(!d_unhashed(dentry
));
2741 __d_instantiate(dentry
, NULL
);
2744 spin_lock(&inode
->i_lock
);
2745 if (S_ISDIR(inode
->i_mode
)) {
2746 struct dentry
*new = __d_find_any_alias(inode
);
2747 if (unlikely(new)) {
2748 write_seqlock(&rename_lock
);
2749 if (unlikely(d_ancestor(new, dentry
))) {
2750 write_sequnlock(&rename_lock
);
2751 spin_unlock(&inode
->i_lock
);
2753 new = ERR_PTR(-ELOOP
);
2754 pr_warn_ratelimited(
2755 "VFS: Lookup of '%s' in %s %s"
2756 " would have caused loop\n",
2757 dentry
->d_name
.name
,
2758 inode
->i_sb
->s_type
->name
,
2760 } else if (!IS_ROOT(new)) {
2761 int err
= __d_unalias(inode
, dentry
, new);
2762 write_sequnlock(&rename_lock
);
2768 __d_move(new, dentry
, false);
2769 write_sequnlock(&rename_lock
);
2770 spin_unlock(&inode
->i_lock
);
2771 security_d_instantiate(new, inode
);
2777 /* already taking inode->i_lock, so d_add() by hand */
2778 __d_instantiate(dentry
, inode
);
2779 spin_unlock(&inode
->i_lock
);
2781 security_d_instantiate(dentry
, inode
);
2785 EXPORT_SYMBOL(d_splice_alias
);
2787 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2791 return -ENAMETOOLONG
;
2793 memcpy(*buffer
, str
, namelen
);
2798 * prepend_name - prepend a pathname in front of current buffer pointer
2799 * @buffer: buffer pointer
2800 * @buflen: allocated length of the buffer
2801 * @name: name string and length qstr structure
2803 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2804 * make sure that either the old or the new name pointer and length are
2805 * fetched. However, there may be mismatch between length and pointer.
2806 * The length cannot be trusted, we need to copy it byte-by-byte until
2807 * the length is reached or a null byte is found. It also prepends "/" at
2808 * the beginning of the name. The sequence number check at the caller will
2809 * retry it again when a d_move() does happen. So any garbage in the buffer
2810 * due to mismatched pointer and length will be discarded.
2812 * Data dependency barrier is needed to make sure that we see that terminating
2813 * NUL. Alpha strikes again, film at 11...
2815 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2817 const char *dname
= ACCESS_ONCE(name
->name
);
2818 u32 dlen
= ACCESS_ONCE(name
->len
);
2821 smp_read_barrier_depends();
2823 *buflen
-= dlen
+ 1;
2825 return -ENAMETOOLONG
;
2826 p
= *buffer
-= dlen
+ 1;
2838 * prepend_path - Prepend path string to a buffer
2839 * @path: the dentry/vfsmount to report
2840 * @root: root vfsmnt/dentry
2841 * @buffer: pointer to the end of the buffer
2842 * @buflen: pointer to buffer length
2844 * The function will first try to write out the pathname without taking any
2845 * lock other than the RCU read lock to make sure that dentries won't go away.
2846 * It only checks the sequence number of the global rename_lock as any change
2847 * in the dentry's d_seq will be preceded by changes in the rename_lock
2848 * sequence number. If the sequence number had been changed, it will restart
2849 * the whole pathname back-tracing sequence again by taking the rename_lock.
2850 * In this case, there is no need to take the RCU read lock as the recursive
2851 * parent pointer references will keep the dentry chain alive as long as no
2852 * rename operation is performed.
2854 static int prepend_path(const struct path
*path
,
2855 const struct path
*root
,
2856 char **buffer
, int *buflen
)
2858 struct dentry
*dentry
;
2859 struct vfsmount
*vfsmnt
;
2862 unsigned seq
, m_seq
= 0;
2868 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
2875 dentry
= path
->dentry
;
2877 mnt
= real_mount(vfsmnt
);
2878 read_seqbegin_or_lock(&rename_lock
, &seq
);
2879 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2880 struct dentry
* parent
;
2882 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2883 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
2885 if (mnt
!= parent
) {
2886 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
2892 * Filesystems needing to implement special "root names"
2893 * should do so with ->d_dname()
2895 if (IS_ROOT(dentry
) &&
2896 (dentry
->d_name
.len
!= 1 ||
2897 dentry
->d_name
.name
[0] != '/')) {
2898 WARN(1, "Root dentry has weird name <%.*s>\n",
2899 (int) dentry
->d_name
.len
,
2900 dentry
->d_name
.name
);
2903 error
= is_mounted(vfsmnt
) ? 1 : 2;
2906 parent
= dentry
->d_parent
;
2908 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
2916 if (need_seqretry(&rename_lock
, seq
)) {
2920 done_seqretry(&rename_lock
, seq
);
2924 if (need_seqretry(&mount_lock
, m_seq
)) {
2928 done_seqretry(&mount_lock
, m_seq
);
2930 if (error
>= 0 && bptr
== *buffer
) {
2932 error
= -ENAMETOOLONG
;
2942 * __d_path - return the path of a dentry
2943 * @path: the dentry/vfsmount to report
2944 * @root: root vfsmnt/dentry
2945 * @buf: buffer to return value in
2946 * @buflen: buffer length
2948 * Convert a dentry into an ASCII path name.
2950 * Returns a pointer into the buffer or an error code if the
2951 * path was too long.
2953 * "buflen" should be positive.
2955 * If the path is not reachable from the supplied root, return %NULL.
2957 char *__d_path(const struct path
*path
,
2958 const struct path
*root
,
2959 char *buf
, int buflen
)
2961 char *res
= buf
+ buflen
;
2964 prepend(&res
, &buflen
, "\0", 1);
2965 error
= prepend_path(path
, root
, &res
, &buflen
);
2968 return ERR_PTR(error
);
2974 char *d_absolute_path(const struct path
*path
,
2975 char *buf
, int buflen
)
2977 struct path root
= {};
2978 char *res
= buf
+ buflen
;
2981 prepend(&res
, &buflen
, "\0", 1);
2982 error
= prepend_path(path
, &root
, &res
, &buflen
);
2987 return ERR_PTR(error
);
2992 * same as __d_path but appends "(deleted)" for unlinked files.
2994 static int path_with_deleted(const struct path
*path
,
2995 const struct path
*root
,
2996 char **buf
, int *buflen
)
2998 prepend(buf
, buflen
, "\0", 1);
2999 if (d_unlinked(path
->dentry
)) {
3000 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3005 return prepend_path(path
, root
, buf
, buflen
);
3008 static int prepend_unreachable(char **buffer
, int *buflen
)
3010 return prepend(buffer
, buflen
, "(unreachable)", 13);
3013 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3018 seq
= read_seqcount_begin(&fs
->seq
);
3020 } while (read_seqcount_retry(&fs
->seq
, seq
));
3024 * d_path - return the path of a dentry
3025 * @path: path to report
3026 * @buf: buffer to return value in
3027 * @buflen: buffer length
3029 * Convert a dentry into an ASCII path name. If the entry has been deleted
3030 * the string " (deleted)" is appended. Note that this is ambiguous.
3032 * Returns a pointer into the buffer or an error code if the path was
3033 * too long. Note: Callers should use the returned pointer, not the passed
3034 * in buffer, to use the name! The implementation often starts at an offset
3035 * into the buffer, and may leave 0 bytes at the start.
3037 * "buflen" should be positive.
3039 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3041 char *res
= buf
+ buflen
;
3046 * We have various synthetic filesystems that never get mounted. On
3047 * these filesystems dentries are never used for lookup purposes, and
3048 * thus don't need to be hashed. They also don't need a name until a
3049 * user wants to identify the object in /proc/pid/fd/. The little hack
3050 * below allows us to generate a name for these objects on demand:
3052 * Some pseudo inodes are mountable. When they are mounted
3053 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3054 * and instead have d_path return the mounted path.
3056 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3057 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3058 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3061 get_fs_root_rcu(current
->fs
, &root
);
3062 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3066 res
= ERR_PTR(error
);
3069 EXPORT_SYMBOL(d_path
);
3072 * Helper function for dentry_operations.d_dname() members
3074 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3075 const char *fmt
, ...)
3081 va_start(args
, fmt
);
3082 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3085 if (sz
> sizeof(temp
) || sz
> buflen
)
3086 return ERR_PTR(-ENAMETOOLONG
);
3088 buffer
+= buflen
- sz
;
3089 return memcpy(buffer
, temp
, sz
);
3092 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3094 char *end
= buffer
+ buflen
;
3095 /* these dentries are never renamed, so d_lock is not needed */
3096 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3097 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3098 prepend(&end
, &buflen
, "/", 1))
3099 end
= ERR_PTR(-ENAMETOOLONG
);
3102 EXPORT_SYMBOL(simple_dname
);
3105 * Write full pathname from the root of the filesystem into the buffer.
3107 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3109 struct dentry
*dentry
;
3122 prepend(&end
, &len
, "\0", 1);
3126 read_seqbegin_or_lock(&rename_lock
, &seq
);
3127 while (!IS_ROOT(dentry
)) {
3128 struct dentry
*parent
= dentry
->d_parent
;
3131 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3140 if (need_seqretry(&rename_lock
, seq
)) {
3144 done_seqretry(&rename_lock
, seq
);
3149 return ERR_PTR(-ENAMETOOLONG
);
3152 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3154 return __dentry_path(dentry
, buf
, buflen
);
3156 EXPORT_SYMBOL(dentry_path_raw
);
3158 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3163 if (d_unlinked(dentry
)) {
3165 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3169 retval
= __dentry_path(dentry
, buf
, buflen
);
3170 if (!IS_ERR(retval
) && p
)
3171 *p
= '/'; /* restore '/' overriden with '\0' */
3174 return ERR_PTR(-ENAMETOOLONG
);
3177 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3183 seq
= read_seqcount_begin(&fs
->seq
);
3186 } while (read_seqcount_retry(&fs
->seq
, seq
));
3190 * NOTE! The user-level library version returns a
3191 * character pointer. The kernel system call just
3192 * returns the length of the buffer filled (which
3193 * includes the ending '\0' character), or a negative
3194 * error value. So libc would do something like
3196 * char *getcwd(char * buf, size_t size)
3200 * retval = sys_getcwd(buf, size);
3207 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3210 struct path pwd
, root
;
3211 char *page
= __getname();
3217 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3220 if (!d_unlinked(pwd
.dentry
)) {
3222 char *cwd
= page
+ PATH_MAX
;
3223 int buflen
= PATH_MAX
;
3225 prepend(&cwd
, &buflen
, "\0", 1);
3226 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3232 /* Unreachable from current root */
3234 error
= prepend_unreachable(&cwd
, &buflen
);
3240 len
= PATH_MAX
+ page
- cwd
;
3243 if (copy_to_user(buf
, cwd
, len
))
3256 * Test whether new_dentry is a subdirectory of old_dentry.
3258 * Trivially implemented using the dcache structure
3262 * is_subdir - is new dentry a subdirectory of old_dentry
3263 * @new_dentry: new dentry
3264 * @old_dentry: old dentry
3266 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3267 * Returns 0 otherwise.
3268 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3271 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3276 if (new_dentry
== old_dentry
)
3280 /* for restarting inner loop in case of seq retry */
3281 seq
= read_seqbegin(&rename_lock
);
3283 * Need rcu_readlock to protect against the d_parent trashing
3287 if (d_ancestor(old_dentry
, new_dentry
))
3292 } while (read_seqretry(&rename_lock
, seq
));
3297 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3299 struct dentry
*root
= data
;
3300 if (dentry
!= root
) {
3301 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3304 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3305 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3306 dentry
->d_lockref
.count
--;
3309 return D_WALK_CONTINUE
;
3312 void d_genocide(struct dentry
*parent
)
3314 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3317 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3319 inode_dec_link_count(inode
);
3320 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3321 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3322 !d_unlinked(dentry
));
3323 spin_lock(&dentry
->d_parent
->d_lock
);
3324 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3325 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3326 (unsigned long long)inode
->i_ino
);
3327 spin_unlock(&dentry
->d_lock
);
3328 spin_unlock(&dentry
->d_parent
->d_lock
);
3329 d_instantiate(dentry
, inode
);
3331 EXPORT_SYMBOL(d_tmpfile
);
3333 static __initdata
unsigned long dhash_entries
;
3334 static int __init
set_dhash_entries(char *str
)
3338 dhash_entries
= simple_strtoul(str
, &str
, 0);
3341 __setup("dhash_entries=", set_dhash_entries
);
3343 static void __init
dcache_init_early(void)
3347 /* If hashes are distributed across NUMA nodes, defer
3348 * hash allocation until vmalloc space is available.
3354 alloc_large_system_hash("Dentry cache",
3355 sizeof(struct hlist_bl_head
),
3364 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3365 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3368 static void __init
dcache_init(void)
3373 * A constructor could be added for stable state like the lists,
3374 * but it is probably not worth it because of the cache nature
3377 dentry_cache
= KMEM_CACHE(dentry
,
3378 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3380 /* Hash may have been set up in dcache_init_early */
3385 alloc_large_system_hash("Dentry cache",
3386 sizeof(struct hlist_bl_head
),
3395 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3396 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3399 /* SLAB cache for __getname() consumers */
3400 struct kmem_cache
*names_cachep __read_mostly
;
3401 EXPORT_SYMBOL(names_cachep
);
3403 EXPORT_SYMBOL(d_genocide
);
3405 void __init
vfs_caches_init_early(void)
3407 dcache_init_early();
3411 void __init
vfs_caches_init(unsigned long mempages
)
3413 unsigned long reserve
;
3415 /* Base hash sizes on available memory, with a reserve equal to
3416 150% of current kernel size */
3418 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3419 mempages
-= reserve
;
3421 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3422 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3426 files_init(mempages
);