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>
46 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_u.d_alias, d_inode of aliases
48 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
52 * dentry->d_sb->s_dentry_lru_lock protects:
53 * - the dcache lru lists and counters
60 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
62 * - d_u.d_alias, d_inode
65 * dentry->d_inode->i_lock
67 * dentry->d_sb->s_dentry_lru_lock
68 * dcache_hash_bucket lock
71 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
74 * dentry->d_parent->d_lock
77 * If no ancestor relationship:
78 * if (dentry1 < dentry2)
82 int sysctl_vfs_cache_pressure __read_mostly
= 100;
83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
85 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
87 EXPORT_SYMBOL(rename_lock
);
89 static struct kmem_cache
*dentry_cache __read_mostly
;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
100 static unsigned int d_hash_mask __read_mostly
;
101 static unsigned int d_hash_shift __read_mostly
;
103 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
105 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
108 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
109 return dentry_hashtable
+ hash_32(hash
, d_hash_shift
);
112 /* Statistics gathering. */
113 struct dentry_stat_t dentry_stat
= {
117 static DEFINE_PER_CPU(long, nr_dentry
);
118 static DEFINE_PER_CPU(long, nr_dentry_unused
);
120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 * Here we resort to our own counters instead of using generic per-cpu counters
124 * for consistency with what the vfs inode code does. We are expected to harvest
125 * better code and performance by having our own specialized counters.
127 * Please note that the loop is done over all possible CPUs, not over all online
128 * CPUs. The reason for this is that we don't want to play games with CPUs going
129 * on and off. If one of them goes off, we will just keep their counters.
131 * glommer: See cffbc8a for details, and if you ever intend to change this,
132 * please update all vfs counters to match.
134 static long get_nr_dentry(void)
138 for_each_possible_cpu(i
)
139 sum
+= per_cpu(nr_dentry
, i
);
140 return sum
< 0 ? 0 : sum
;
143 static long get_nr_dentry_unused(void)
147 for_each_possible_cpu(i
)
148 sum
+= per_cpu(nr_dentry_unused
, i
);
149 return sum
< 0 ? 0 : sum
;
152 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
153 size_t *lenp
, loff_t
*ppos
)
155 dentry_stat
.nr_dentry
= get_nr_dentry();
156 dentry_stat
.nr_unused
= get_nr_dentry_unused();
157 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
162 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
163 * The strings are both count bytes long, and count is non-zero.
165 #ifdef CONFIG_DCACHE_WORD_ACCESS
167 #include <asm/word-at-a-time.h>
169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
170 * aligned allocation for this particular component. We don't
171 * strictly need the load_unaligned_zeropad() safety, but it
172 * doesn't hurt either.
174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
175 * need the careful unaligned handling.
177 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
179 unsigned long a
,b
,mask
;
182 a
= *(unsigned long *)cs
;
183 b
= load_unaligned_zeropad(ct
);
184 if (tcount
< sizeof(unsigned long))
186 if (unlikely(a
!= b
))
188 cs
+= sizeof(unsigned long);
189 ct
+= sizeof(unsigned long);
190 tcount
-= sizeof(unsigned long);
194 mask
= bytemask_from_count(tcount
);
195 return unlikely(!!((a
^ b
) & mask
));
200 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
214 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
216 const unsigned char *cs
;
218 * Be careful about RCU walk racing with rename:
219 * use ACCESS_ONCE to fetch the name pointer.
221 * NOTE! Even if a rename will mean that the length
222 * was not loaded atomically, we don't care. The
223 * RCU walk will check the sequence count eventually,
224 * and catch it. And we won't overrun the buffer,
225 * because we're reading the name pointer atomically,
226 * and a dentry name is guaranteed to be properly
227 * terminated with a NUL byte.
229 * End result: even if 'len' is wrong, we'll exit
230 * early because the data cannot match (there can
231 * be no NUL in the ct/tcount data)
233 cs
= ACCESS_ONCE(dentry
->d_name
.name
);
234 smp_read_barrier_depends();
235 return dentry_string_cmp(cs
, ct
, tcount
);
238 struct external_name
{
241 struct rcu_head head
;
243 unsigned char name
[];
246 static inline struct external_name
*external_name(struct dentry
*dentry
)
248 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
251 static void __d_free(struct rcu_head
*head
)
253 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
255 kmem_cache_free(dentry_cache
, dentry
);
258 static void __d_free_external(struct rcu_head
*head
)
260 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
261 kfree(external_name(dentry
));
262 kmem_cache_free(dentry_cache
, dentry
);
265 static inline int dname_external(const struct dentry
*dentry
)
267 return dentry
->d_name
.name
!= dentry
->d_iname
;
270 static void dentry_free(struct dentry
*dentry
)
272 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
273 if (unlikely(dname_external(dentry
))) {
274 struct external_name
*p
= external_name(dentry
);
275 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
276 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
280 /* if dentry was never visible to RCU, immediate free is OK */
281 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
282 __d_free(&dentry
->d_u
.d_rcu
);
284 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
288 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
289 * @dentry: the target dentry
290 * After this call, in-progress rcu-walk path lookup will fail. This
291 * should be called after unhashing, and after changing d_inode (if
292 * the dentry has not already been unhashed).
294 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
296 assert_spin_locked(&dentry
->d_lock
);
297 /* Go through a barrier */
298 write_seqcount_barrier(&dentry
->d_seq
);
302 * Release the dentry's inode, using the filesystem
303 * d_iput() operation if defined. Dentry has no refcount
306 static void dentry_iput(struct dentry
* dentry
)
307 __releases(dentry
->d_lock
)
308 __releases(dentry
->d_inode
->i_lock
)
310 struct inode
*inode
= dentry
->d_inode
;
312 dentry
->d_inode
= NULL
;
313 hlist_del_init(&dentry
->d_u
.d_alias
);
314 spin_unlock(&dentry
->d_lock
);
315 spin_unlock(&inode
->i_lock
);
317 fsnotify_inoderemove(inode
);
318 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
319 dentry
->d_op
->d_iput(dentry
, inode
);
323 spin_unlock(&dentry
->d_lock
);
328 * Release the dentry's inode, using the filesystem
329 * d_iput() operation if defined. dentry remains in-use.
331 static void dentry_unlink_inode(struct dentry
* dentry
)
332 __releases(dentry
->d_lock
)
333 __releases(dentry
->d_inode
->i_lock
)
335 struct inode
*inode
= dentry
->d_inode
;
336 __d_clear_type(dentry
);
337 dentry
->d_inode
= NULL
;
338 hlist_del_init(&dentry
->d_u
.d_alias
);
339 dentry_rcuwalk_barrier(dentry
);
340 spin_unlock(&dentry
->d_lock
);
341 spin_unlock(&inode
->i_lock
);
343 fsnotify_inoderemove(inode
);
344 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
345 dentry
->d_op
->d_iput(dentry
, inode
);
351 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
352 * is in use - which includes both the "real" per-superblock
353 * LRU list _and_ the DCACHE_SHRINK_LIST use.
355 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
356 * on the shrink list (ie not on the superblock LRU list).
358 * The per-cpu "nr_dentry_unused" counters are updated with
359 * the DCACHE_LRU_LIST bit.
361 * These helper functions make sure we always follow the
362 * rules. d_lock must be held by the caller.
364 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
365 static void d_lru_add(struct dentry
*dentry
)
367 D_FLAG_VERIFY(dentry
, 0);
368 dentry
->d_flags
|= DCACHE_LRU_LIST
;
369 this_cpu_inc(nr_dentry_unused
);
370 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
373 static void d_lru_del(struct dentry
*dentry
)
375 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
376 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
377 this_cpu_dec(nr_dentry_unused
);
378 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
381 static void d_shrink_del(struct dentry
*dentry
)
383 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
384 list_del_init(&dentry
->d_lru
);
385 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
386 this_cpu_dec(nr_dentry_unused
);
389 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
391 D_FLAG_VERIFY(dentry
, 0);
392 list_add(&dentry
->d_lru
, list
);
393 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
394 this_cpu_inc(nr_dentry_unused
);
398 * These can only be called under the global LRU lock, ie during the
399 * callback for freeing the LRU list. "isolate" removes it from the
400 * LRU lists entirely, while shrink_move moves it to the indicated
403 static void d_lru_isolate(struct dentry
*dentry
)
405 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
406 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
407 this_cpu_dec(nr_dentry_unused
);
408 list_del_init(&dentry
->d_lru
);
411 static void d_lru_shrink_move(struct dentry
*dentry
, struct list_head
*list
)
413 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
414 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
415 list_move_tail(&dentry
->d_lru
, list
);
419 * dentry_lru_(add|del)_list) must be called with d_lock held.
421 static void dentry_lru_add(struct dentry
*dentry
)
423 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
428 * d_drop - drop a dentry
429 * @dentry: dentry to drop
431 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
432 * be found through a VFS lookup any more. Note that this is different from
433 * deleting the dentry - d_delete will try to mark the dentry negative if
434 * possible, giving a successful _negative_ lookup, while d_drop will
435 * just make the cache lookup fail.
437 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
438 * reason (NFS timeouts or autofs deletes).
440 * __d_drop requires dentry->d_lock.
442 void __d_drop(struct dentry
*dentry
)
444 if (!d_unhashed(dentry
)) {
445 struct hlist_bl_head
*b
;
447 * Hashed dentries are normally on the dentry hashtable,
448 * with the exception of those newly allocated by
449 * d_obtain_alias, which are always IS_ROOT:
451 if (unlikely(IS_ROOT(dentry
)))
452 b
= &dentry
->d_sb
->s_anon
;
454 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
457 __hlist_bl_del(&dentry
->d_hash
);
458 dentry
->d_hash
.pprev
= NULL
;
460 dentry_rcuwalk_barrier(dentry
);
463 EXPORT_SYMBOL(__d_drop
);
465 void d_drop(struct dentry
*dentry
)
467 spin_lock(&dentry
->d_lock
);
469 spin_unlock(&dentry
->d_lock
);
471 EXPORT_SYMBOL(d_drop
);
473 static void __dentry_kill(struct dentry
*dentry
)
475 struct dentry
*parent
= NULL
;
476 bool can_free
= true;
477 if (!IS_ROOT(dentry
))
478 parent
= dentry
->d_parent
;
481 * The dentry is now unrecoverably dead to the world.
483 lockref_mark_dead(&dentry
->d_lockref
);
486 * inform the fs via d_prune that this dentry is about to be
487 * unhashed and destroyed.
489 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
490 dentry
->d_op
->d_prune(dentry
);
492 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
493 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
496 /* if it was on the hash then remove it */
498 __list_del_entry(&dentry
->d_child
);
500 * Inform d_walk() that we are no longer attached to the
503 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
505 spin_unlock(&parent
->d_lock
);
508 * dentry_iput drops the locks, at which point nobody (except
509 * transient RCU lookups) can reach this dentry.
511 BUG_ON(dentry
->d_lockref
.count
> 0);
512 this_cpu_dec(nr_dentry
);
513 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
514 dentry
->d_op
->d_release(dentry
);
516 spin_lock(&dentry
->d_lock
);
517 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
518 dentry
->d_flags
|= DCACHE_MAY_FREE
;
521 spin_unlock(&dentry
->d_lock
);
522 if (likely(can_free
))
527 * Finish off a dentry we've decided to kill.
528 * dentry->d_lock must be held, returns with it unlocked.
529 * If ref is non-zero, then decrement the refcount too.
530 * Returns dentry requiring refcount drop, or NULL if we're done.
532 static struct dentry
*dentry_kill(struct dentry
*dentry
)
533 __releases(dentry
->d_lock
)
535 struct inode
*inode
= dentry
->d_inode
;
536 struct dentry
*parent
= NULL
;
538 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
541 if (!IS_ROOT(dentry
)) {
542 parent
= dentry
->d_parent
;
543 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
545 spin_unlock(&inode
->i_lock
);
550 __dentry_kill(dentry
);
554 spin_unlock(&dentry
->d_lock
);
556 return dentry
; /* try again with same dentry */
559 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
561 struct dentry
*parent
= dentry
->d_parent
;
564 if (unlikely(dentry
->d_lockref
.count
< 0))
566 if (likely(spin_trylock(&parent
->d_lock
)))
569 spin_unlock(&dentry
->d_lock
);
571 parent
= ACCESS_ONCE(dentry
->d_parent
);
572 spin_lock(&parent
->d_lock
);
574 * We can't blindly lock dentry until we are sure
575 * that we won't violate the locking order.
576 * Any changes of dentry->d_parent must have
577 * been done with parent->d_lock held, so
578 * spin_lock() above is enough of a barrier
579 * for checking if it's still our child.
581 if (unlikely(parent
!= dentry
->d_parent
)) {
582 spin_unlock(&parent
->d_lock
);
586 if (parent
!= dentry
)
587 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
594 * Try to do a lockless dput(), and return whether that was successful.
596 * If unsuccessful, we return false, having already taken the dentry lock.
598 * The caller needs to hold the RCU read lock, so that the dentry is
599 * guaranteed to stay around even if the refcount goes down to zero!
601 static inline bool fast_dput(struct dentry
*dentry
)
604 unsigned int d_flags
;
607 * If we have a d_op->d_delete() operation, we sould not
608 * let the dentry count go to zero, so use "put__or_lock".
610 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
611 return lockref_put_or_lock(&dentry
->d_lockref
);
614 * .. otherwise, we can try to just decrement the
615 * lockref optimistically.
617 ret
= lockref_put_return(&dentry
->d_lockref
);
620 * If the lockref_put_return() failed due to the lock being held
621 * by somebody else, the fast path has failed. We will need to
622 * get the lock, and then check the count again.
624 if (unlikely(ret
< 0)) {
625 spin_lock(&dentry
->d_lock
);
626 if (dentry
->d_lockref
.count
> 1) {
627 dentry
->d_lockref
.count
--;
628 spin_unlock(&dentry
->d_lock
);
635 * If we weren't the last ref, we're done.
641 * Careful, careful. The reference count went down
642 * to zero, but we don't hold the dentry lock, so
643 * somebody else could get it again, and do another
644 * dput(), and we need to not race with that.
646 * However, there is a very special and common case
647 * where we don't care, because there is nothing to
648 * do: the dentry is still hashed, it does not have
649 * a 'delete' op, and it's referenced and already on
652 * NOTE! Since we aren't locked, these values are
653 * not "stable". However, it is sufficient that at
654 * some point after we dropped the reference the
655 * dentry was hashed and the flags had the proper
656 * value. Other dentry users may have re-gotten
657 * a reference to the dentry and change that, but
658 * our work is done - we can leave the dentry
659 * around with a zero refcount.
662 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
663 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
;
665 /* Nothing to do? Dropping the reference was all we needed? */
666 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
670 * Not the fast normal case? Get the lock. We've already decremented
671 * the refcount, but we'll need to re-check the situation after
674 spin_lock(&dentry
->d_lock
);
677 * Did somebody else grab a reference to it in the meantime, and
678 * we're no longer the last user after all? Alternatively, somebody
679 * else could have killed it and marked it dead. Either way, we
680 * don't need to do anything else.
682 if (dentry
->d_lockref
.count
) {
683 spin_unlock(&dentry
->d_lock
);
688 * Re-get the reference we optimistically dropped. We hold the
689 * lock, and we just tested that it was zero, so we can just
692 dentry
->d_lockref
.count
= 1;
700 * This is complicated by the fact that we do not want to put
701 * dentries that are no longer on any hash chain on the unused
702 * list: we'd much rather just get rid of them immediately.
704 * However, that implies that we have to traverse the dentry
705 * tree upwards to the parents which might _also_ now be
706 * scheduled for deletion (it may have been only waiting for
707 * its last child to go away).
709 * This tail recursion is done by hand as we don't want to depend
710 * on the compiler to always get this right (gcc generally doesn't).
711 * Real recursion would eat up our stack space.
715 * dput - release a dentry
716 * @dentry: dentry to release
718 * Release a dentry. This will drop the usage count and if appropriate
719 * call the dentry unlink method as well as removing it from the queues and
720 * releasing its resources. If the parent dentries were scheduled for release
721 * they too may now get deleted.
723 void dput(struct dentry
*dentry
)
725 if (unlikely(!dentry
))
730 if (likely(fast_dput(dentry
))) {
735 /* Slow case: now with the dentry lock held */
738 /* Unreachable? Get rid of it */
739 if (unlikely(d_unhashed(dentry
)))
742 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
743 if (dentry
->d_op
->d_delete(dentry
))
747 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
748 dentry
->d_flags
|= DCACHE_REFERENCED
;
749 dentry_lru_add(dentry
);
751 dentry
->d_lockref
.count
--;
752 spin_unlock(&dentry
->d_lock
);
756 dentry
= dentry_kill(dentry
);
763 /* This must be called with d_lock held */
764 static inline void __dget_dlock(struct dentry
*dentry
)
766 dentry
->d_lockref
.count
++;
769 static inline void __dget(struct dentry
*dentry
)
771 lockref_get(&dentry
->d_lockref
);
774 struct dentry
*dget_parent(struct dentry
*dentry
)
780 * Do optimistic parent lookup without any
784 ret
= ACCESS_ONCE(dentry
->d_parent
);
785 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
787 if (likely(gotref
)) {
788 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
795 * Don't need rcu_dereference because we re-check it was correct under
799 ret
= dentry
->d_parent
;
800 spin_lock(&ret
->d_lock
);
801 if (unlikely(ret
!= dentry
->d_parent
)) {
802 spin_unlock(&ret
->d_lock
);
807 BUG_ON(!ret
->d_lockref
.count
);
808 ret
->d_lockref
.count
++;
809 spin_unlock(&ret
->d_lock
);
812 EXPORT_SYMBOL(dget_parent
);
815 * d_find_alias - grab a hashed alias of inode
816 * @inode: inode in question
818 * If inode has a hashed alias, or is a directory and has any alias,
819 * acquire the reference to alias and return it. Otherwise return NULL.
820 * Notice that if inode is a directory there can be only one alias and
821 * it can be unhashed only if it has no children, or if it is the root
822 * of a filesystem, or if the directory was renamed and d_revalidate
823 * was the first vfs operation to notice.
825 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
826 * any other hashed alias over that one.
828 static struct dentry
*__d_find_alias(struct inode
*inode
)
830 struct dentry
*alias
, *discon_alias
;
834 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
835 spin_lock(&alias
->d_lock
);
836 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
837 if (IS_ROOT(alias
) &&
838 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
839 discon_alias
= alias
;
842 spin_unlock(&alias
->d_lock
);
846 spin_unlock(&alias
->d_lock
);
849 alias
= discon_alias
;
850 spin_lock(&alias
->d_lock
);
851 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
853 spin_unlock(&alias
->d_lock
);
856 spin_unlock(&alias
->d_lock
);
862 struct dentry
*d_find_alias(struct inode
*inode
)
864 struct dentry
*de
= NULL
;
866 if (!hlist_empty(&inode
->i_dentry
)) {
867 spin_lock(&inode
->i_lock
);
868 de
= __d_find_alias(inode
);
869 spin_unlock(&inode
->i_lock
);
873 EXPORT_SYMBOL(d_find_alias
);
876 * Try to kill dentries associated with this inode.
877 * WARNING: you must own a reference to inode.
879 void d_prune_aliases(struct inode
*inode
)
881 struct dentry
*dentry
;
883 spin_lock(&inode
->i_lock
);
884 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
885 spin_lock(&dentry
->d_lock
);
886 if (!dentry
->d_lockref
.count
) {
887 struct dentry
*parent
= lock_parent(dentry
);
888 if (likely(!dentry
->d_lockref
.count
)) {
889 __dentry_kill(dentry
);
894 spin_unlock(&parent
->d_lock
);
896 spin_unlock(&dentry
->d_lock
);
898 spin_unlock(&inode
->i_lock
);
900 EXPORT_SYMBOL(d_prune_aliases
);
902 static void shrink_dentry_list(struct list_head
*list
)
904 struct dentry
*dentry
, *parent
;
906 while (!list_empty(list
)) {
908 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
909 spin_lock(&dentry
->d_lock
);
910 parent
= lock_parent(dentry
);
913 * The dispose list is isolated and dentries are not accounted
914 * to the LRU here, so we can simply remove it from the list
915 * here regardless of whether it is referenced or not.
917 d_shrink_del(dentry
);
920 * We found an inuse dentry which was not removed from
921 * the LRU because of laziness during lookup. Do not free it.
923 if (dentry
->d_lockref
.count
> 0) {
924 spin_unlock(&dentry
->d_lock
);
926 spin_unlock(&parent
->d_lock
);
931 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
932 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
933 spin_unlock(&dentry
->d_lock
);
935 spin_unlock(&parent
->d_lock
);
941 inode
= dentry
->d_inode
;
942 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
943 d_shrink_add(dentry
, list
);
944 spin_unlock(&dentry
->d_lock
);
946 spin_unlock(&parent
->d_lock
);
950 __dentry_kill(dentry
);
953 * We need to prune ancestors too. This is necessary to prevent
954 * quadratic behavior of shrink_dcache_parent(), but is also
955 * expected to be beneficial in reducing dentry cache
959 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
960 parent
= lock_parent(dentry
);
961 if (dentry
->d_lockref
.count
!= 1) {
962 dentry
->d_lockref
.count
--;
963 spin_unlock(&dentry
->d_lock
);
965 spin_unlock(&parent
->d_lock
);
968 inode
= dentry
->d_inode
; /* can't be NULL */
969 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
970 spin_unlock(&dentry
->d_lock
);
972 spin_unlock(&parent
->d_lock
);
976 __dentry_kill(dentry
);
982 static enum lru_status
983 dentry_lru_isolate(struct list_head
*item
, spinlock_t
*lru_lock
, void *arg
)
985 struct list_head
*freeable
= arg
;
986 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
990 * we are inverting the lru lock/dentry->d_lock here,
991 * so use a trylock. If we fail to get the lock, just skip
994 if (!spin_trylock(&dentry
->d_lock
))
998 * Referenced dentries are still in use. If they have active
999 * counts, just remove them from the LRU. Otherwise give them
1000 * another pass through the LRU.
1002 if (dentry
->d_lockref
.count
) {
1003 d_lru_isolate(dentry
);
1004 spin_unlock(&dentry
->d_lock
);
1008 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1009 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1010 spin_unlock(&dentry
->d_lock
);
1013 * The list move itself will be made by the common LRU code. At
1014 * this point, we've dropped the dentry->d_lock but keep the
1015 * lru lock. This is safe to do, since every list movement is
1016 * protected by the lru lock even if both locks are held.
1018 * This is guaranteed by the fact that all LRU management
1019 * functions are intermediated by the LRU API calls like
1020 * list_lru_add and list_lru_del. List movement in this file
1021 * only ever occur through this functions or through callbacks
1022 * like this one, that are called from the LRU API.
1024 * The only exceptions to this are functions like
1025 * shrink_dentry_list, and code that first checks for the
1026 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1027 * operating only with stack provided lists after they are
1028 * properly isolated from the main list. It is thus, always a
1034 d_lru_shrink_move(dentry
, freeable
);
1035 spin_unlock(&dentry
->d_lock
);
1041 * prune_dcache_sb - shrink the dcache
1043 * @nr_to_scan : number of entries to try to free
1044 * @nid: which node to scan for freeable entities
1046 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
1047 * done when we need more memory an called from the superblock shrinker
1050 * This function may fail to free any resources if all the dentries are in
1053 long prune_dcache_sb(struct super_block
*sb
, unsigned long nr_to_scan
,
1059 freed
= list_lru_walk_node(&sb
->s_dentry_lru
, nid
, dentry_lru_isolate
,
1060 &dispose
, &nr_to_scan
);
1061 shrink_dentry_list(&dispose
);
1065 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1066 spinlock_t
*lru_lock
, void *arg
)
1068 struct list_head
*freeable
= arg
;
1069 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1072 * we are inverting the lru lock/dentry->d_lock here,
1073 * so use a trylock. If we fail to get the lock, just skip
1076 if (!spin_trylock(&dentry
->d_lock
))
1079 d_lru_shrink_move(dentry
, freeable
);
1080 spin_unlock(&dentry
->d_lock
);
1087 * shrink_dcache_sb - shrink dcache for a superblock
1090 * Shrink the dcache for the specified super block. This is used to free
1091 * the dcache before unmounting a file system.
1093 void shrink_dcache_sb(struct super_block
*sb
)
1100 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1101 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1103 this_cpu_sub(nr_dentry_unused
, freed
);
1104 shrink_dentry_list(&dispose
);
1105 } while (freed
> 0);
1107 EXPORT_SYMBOL(shrink_dcache_sb
);
1110 * enum d_walk_ret - action to talke during tree walk
1111 * @D_WALK_CONTINUE: contrinue walk
1112 * @D_WALK_QUIT: quit walk
1113 * @D_WALK_NORETRY: quit when retry is needed
1114 * @D_WALK_SKIP: skip this dentry and its children
1124 * d_walk - walk the dentry tree
1125 * @parent: start of walk
1126 * @data: data passed to @enter() and @finish()
1127 * @enter: callback when first entering the dentry
1128 * @finish: callback when successfully finished the walk
1130 * The @enter() and @finish() callbacks are called with d_lock held.
1132 static void d_walk(struct dentry
*parent
, void *data
,
1133 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1134 void (*finish
)(void *))
1136 struct dentry
*this_parent
;
1137 struct list_head
*next
;
1139 enum d_walk_ret ret
;
1143 read_seqbegin_or_lock(&rename_lock
, &seq
);
1144 this_parent
= parent
;
1145 spin_lock(&this_parent
->d_lock
);
1147 ret
= enter(data
, this_parent
);
1149 case D_WALK_CONTINUE
:
1154 case D_WALK_NORETRY
:
1159 next
= this_parent
->d_subdirs
.next
;
1161 while (next
!= &this_parent
->d_subdirs
) {
1162 struct list_head
*tmp
= next
;
1163 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1166 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1168 ret
= enter(data
, dentry
);
1170 case D_WALK_CONTINUE
:
1173 spin_unlock(&dentry
->d_lock
);
1175 case D_WALK_NORETRY
:
1179 spin_unlock(&dentry
->d_lock
);
1183 if (!list_empty(&dentry
->d_subdirs
)) {
1184 spin_unlock(&this_parent
->d_lock
);
1185 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1186 this_parent
= dentry
;
1187 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1190 spin_unlock(&dentry
->d_lock
);
1193 * All done at this level ... ascend and resume the search.
1197 if (this_parent
!= parent
) {
1198 struct dentry
*child
= this_parent
;
1199 this_parent
= child
->d_parent
;
1201 spin_unlock(&child
->d_lock
);
1202 spin_lock(&this_parent
->d_lock
);
1204 /* might go back up the wrong parent if we have had a rename. */
1205 if (need_seqretry(&rename_lock
, seq
))
1207 next
= child
->d_child
.next
;
1208 while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
)) {
1209 if (next
== &this_parent
->d_subdirs
)
1211 child
= list_entry(next
, struct dentry
, d_child
);
1217 if (need_seqretry(&rename_lock
, seq
))
1224 spin_unlock(&this_parent
->d_lock
);
1225 done_seqretry(&rename_lock
, seq
);
1229 spin_unlock(&this_parent
->d_lock
);
1239 * Search for at least 1 mount point in the dentry's subdirs.
1240 * We descend to the next level whenever the d_subdirs
1241 * list is non-empty and continue searching.
1244 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1247 if (d_mountpoint(dentry
)) {
1251 return D_WALK_CONTINUE
;
1255 * have_submounts - check for mounts over a dentry
1256 * @parent: dentry to check.
1258 * Return true if the parent or its subdirectories contain
1261 int have_submounts(struct dentry
*parent
)
1265 d_walk(parent
, &ret
, check_mount
, NULL
);
1269 EXPORT_SYMBOL(have_submounts
);
1272 * Called by mount code to set a mountpoint and check if the mountpoint is
1273 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1274 * subtree can become unreachable).
1276 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1277 * this reason take rename_lock and d_lock on dentry and ancestors.
1279 int d_set_mounted(struct dentry
*dentry
)
1283 write_seqlock(&rename_lock
);
1284 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1285 /* Need exclusion wrt. d_invalidate() */
1286 spin_lock(&p
->d_lock
);
1287 if (unlikely(d_unhashed(p
))) {
1288 spin_unlock(&p
->d_lock
);
1291 spin_unlock(&p
->d_lock
);
1293 spin_lock(&dentry
->d_lock
);
1294 if (!d_unlinked(dentry
)) {
1295 dentry
->d_flags
|= DCACHE_MOUNTED
;
1298 spin_unlock(&dentry
->d_lock
);
1300 write_sequnlock(&rename_lock
);
1305 * Search the dentry child list of the specified parent,
1306 * and move any unused dentries to the end of the unused
1307 * list for prune_dcache(). We descend to the next level
1308 * whenever the d_subdirs list is non-empty and continue
1311 * It returns zero iff there are no unused children,
1312 * otherwise it returns the number of children moved to
1313 * the end of the unused list. This may not be the total
1314 * number of unused children, because select_parent can
1315 * drop the lock and return early due to latency
1319 struct select_data
{
1320 struct dentry
*start
;
1321 struct list_head dispose
;
1325 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1327 struct select_data
*data
= _data
;
1328 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1330 if (data
->start
== dentry
)
1333 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1336 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1338 if (!dentry
->d_lockref
.count
) {
1339 d_shrink_add(dentry
, &data
->dispose
);
1344 * We can return to the caller if we have found some (this
1345 * ensures forward progress). We'll be coming back to find
1348 if (!list_empty(&data
->dispose
))
1349 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1355 * shrink_dcache_parent - prune dcache
1356 * @parent: parent of entries to prune
1358 * Prune the dcache to remove unused children of the parent dentry.
1360 void shrink_dcache_parent(struct dentry
*parent
)
1363 struct select_data data
;
1365 INIT_LIST_HEAD(&data
.dispose
);
1366 data
.start
= parent
;
1369 d_walk(parent
, &data
, select_collect
, NULL
);
1373 shrink_dentry_list(&data
.dispose
);
1377 EXPORT_SYMBOL(shrink_dcache_parent
);
1379 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1381 /* it has busy descendents; complain about those instead */
1382 if (!list_empty(&dentry
->d_subdirs
))
1383 return D_WALK_CONTINUE
;
1385 /* root with refcount 1 is fine */
1386 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1387 return D_WALK_CONTINUE
;
1389 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1390 " still in use (%d) [unmount of %s %s]\n",
1393 dentry
->d_inode
->i_ino
: 0UL,
1395 dentry
->d_lockref
.count
,
1396 dentry
->d_sb
->s_type
->name
,
1397 dentry
->d_sb
->s_id
);
1399 return D_WALK_CONTINUE
;
1402 static void do_one_tree(struct dentry
*dentry
)
1404 shrink_dcache_parent(dentry
);
1405 d_walk(dentry
, dentry
, umount_check
, NULL
);
1411 * destroy the dentries attached to a superblock on unmounting
1413 void shrink_dcache_for_umount(struct super_block
*sb
)
1415 struct dentry
*dentry
;
1417 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1419 dentry
= sb
->s_root
;
1421 do_one_tree(dentry
);
1423 while (!hlist_bl_empty(&sb
->s_anon
)) {
1424 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1425 do_one_tree(dentry
);
1429 struct detach_data
{
1430 struct select_data select
;
1431 struct dentry
*mountpoint
;
1433 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1435 struct detach_data
*data
= _data
;
1437 if (d_mountpoint(dentry
)) {
1438 __dget_dlock(dentry
);
1439 data
->mountpoint
= dentry
;
1443 return select_collect(&data
->select
, dentry
);
1446 static void check_and_drop(void *_data
)
1448 struct detach_data
*data
= _data
;
1450 if (!data
->mountpoint
&& !data
->select
.found
)
1451 __d_drop(data
->select
.start
);
1455 * d_invalidate - detach submounts, prune dcache, and drop
1456 * @dentry: dentry to invalidate (aka detach, prune and drop)
1460 * The final d_drop is done as an atomic operation relative to
1461 * rename_lock ensuring there are no races with d_set_mounted. This
1462 * ensures there are no unhashed dentries on the path to a mountpoint.
1464 void d_invalidate(struct dentry
*dentry
)
1467 * If it's already been dropped, return OK.
1469 spin_lock(&dentry
->d_lock
);
1470 if (d_unhashed(dentry
)) {
1471 spin_unlock(&dentry
->d_lock
);
1474 spin_unlock(&dentry
->d_lock
);
1476 /* Negative dentries can be dropped without further checks */
1477 if (!dentry
->d_inode
) {
1483 struct detach_data data
;
1485 data
.mountpoint
= NULL
;
1486 INIT_LIST_HEAD(&data
.select
.dispose
);
1487 data
.select
.start
= dentry
;
1488 data
.select
.found
= 0;
1490 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1492 if (data
.select
.found
)
1493 shrink_dentry_list(&data
.select
.dispose
);
1495 if (data
.mountpoint
) {
1496 detach_mounts(data
.mountpoint
);
1497 dput(data
.mountpoint
);
1500 if (!data
.mountpoint
&& !data
.select
.found
)
1506 EXPORT_SYMBOL(d_invalidate
);
1509 * __d_alloc - allocate a dcache entry
1510 * @sb: filesystem it will belong to
1511 * @name: qstr of the name
1513 * Allocates a dentry. It returns %NULL if there is insufficient memory
1514 * available. On a success the dentry is returned. The name passed in is
1515 * copied and the copy passed in may be reused after this call.
1518 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1520 struct dentry
*dentry
;
1523 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1528 * We guarantee that the inline name is always NUL-terminated.
1529 * This way the memcpy() done by the name switching in rename
1530 * will still always have a NUL at the end, even if we might
1531 * be overwriting an internal NUL character
1533 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1534 if (name
->len
> DNAME_INLINE_LEN
-1) {
1535 size_t size
= offsetof(struct external_name
, name
[1]);
1536 struct external_name
*p
= kmalloc(size
+ name
->len
, GFP_KERNEL
);
1538 kmem_cache_free(dentry_cache
, dentry
);
1541 atomic_set(&p
->u
.count
, 1);
1544 dname
= dentry
->d_iname
;
1547 dentry
->d_name
.len
= name
->len
;
1548 dentry
->d_name
.hash
= name
->hash
;
1549 memcpy(dname
, name
->name
, name
->len
);
1550 dname
[name
->len
] = 0;
1552 /* Make sure we always see the terminating NUL character */
1554 dentry
->d_name
.name
= dname
;
1556 dentry
->d_lockref
.count
= 1;
1557 dentry
->d_flags
= 0;
1558 spin_lock_init(&dentry
->d_lock
);
1559 seqcount_init(&dentry
->d_seq
);
1560 dentry
->d_inode
= NULL
;
1561 dentry
->d_parent
= dentry
;
1563 dentry
->d_op
= NULL
;
1564 dentry
->d_fsdata
= NULL
;
1565 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1566 INIT_LIST_HEAD(&dentry
->d_lru
);
1567 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1568 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1569 INIT_LIST_HEAD(&dentry
->d_child
);
1570 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1572 this_cpu_inc(nr_dentry
);
1578 * d_alloc - allocate a dcache entry
1579 * @parent: parent of entry to allocate
1580 * @name: qstr of the name
1582 * Allocates a dentry. It returns %NULL if there is insufficient memory
1583 * available. On a success the dentry is returned. The name passed in is
1584 * copied and the copy passed in may be reused after this call.
1586 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1588 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1592 spin_lock(&parent
->d_lock
);
1594 * don't need child lock because it is not subject
1595 * to concurrency here
1597 __dget_dlock(parent
);
1598 dentry
->d_parent
= parent
;
1599 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1600 spin_unlock(&parent
->d_lock
);
1604 EXPORT_SYMBOL(d_alloc
);
1607 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1608 * @sb: the superblock
1609 * @name: qstr of the name
1611 * For a filesystem that just pins its dentries in memory and never
1612 * performs lookups at all, return an unhashed IS_ROOT dentry.
1614 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1616 return __d_alloc(sb
, name
);
1618 EXPORT_SYMBOL(d_alloc_pseudo
);
1620 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1625 q
.len
= strlen(name
);
1626 q
.hash
= full_name_hash(q
.name
, q
.len
);
1627 return d_alloc(parent
, &q
);
1629 EXPORT_SYMBOL(d_alloc_name
);
1631 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1633 WARN_ON_ONCE(dentry
->d_op
);
1634 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1636 DCACHE_OP_REVALIDATE
|
1637 DCACHE_OP_WEAK_REVALIDATE
|
1638 DCACHE_OP_DELETE
));
1643 dentry
->d_flags
|= DCACHE_OP_HASH
;
1645 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1646 if (op
->d_revalidate
)
1647 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1648 if (op
->d_weak_revalidate
)
1649 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1651 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1653 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1656 EXPORT_SYMBOL(d_set_d_op
);
1658 static unsigned d_flags_for_inode(struct inode
*inode
)
1660 unsigned add_flags
= DCACHE_FILE_TYPE
;
1663 return DCACHE_MISS_TYPE
;
1665 if (S_ISDIR(inode
->i_mode
)) {
1666 add_flags
= DCACHE_DIRECTORY_TYPE
;
1667 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1668 if (unlikely(!inode
->i_op
->lookup
))
1669 add_flags
= DCACHE_AUTODIR_TYPE
;
1671 inode
->i_opflags
|= IOP_LOOKUP
;
1673 } else if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1674 if (unlikely(inode
->i_op
->follow_link
))
1675 add_flags
= DCACHE_SYMLINK_TYPE
;
1677 inode
->i_opflags
|= IOP_NOFOLLOW
;
1680 if (unlikely(IS_AUTOMOUNT(inode
)))
1681 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1685 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1687 unsigned add_flags
= d_flags_for_inode(inode
);
1689 spin_lock(&dentry
->d_lock
);
1690 __d_set_type(dentry
, add_flags
);
1692 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1693 dentry
->d_inode
= inode
;
1694 dentry_rcuwalk_barrier(dentry
);
1695 spin_unlock(&dentry
->d_lock
);
1696 fsnotify_d_instantiate(dentry
, inode
);
1700 * d_instantiate - fill in inode information for a dentry
1701 * @entry: dentry to complete
1702 * @inode: inode to attach to this dentry
1704 * Fill in inode information in the entry.
1706 * This turns negative dentries into productive full members
1709 * NOTE! This assumes that the inode count has been incremented
1710 * (or otherwise set) by the caller to indicate that it is now
1711 * in use by the dcache.
1714 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1716 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1718 spin_lock(&inode
->i_lock
);
1719 __d_instantiate(entry
, inode
);
1721 spin_unlock(&inode
->i_lock
);
1722 security_d_instantiate(entry
, inode
);
1724 EXPORT_SYMBOL(d_instantiate
);
1727 * d_instantiate_unique - instantiate a non-aliased dentry
1728 * @entry: dentry to instantiate
1729 * @inode: inode to attach to this dentry
1731 * Fill in inode information in the entry. On success, it returns NULL.
1732 * If an unhashed alias of "entry" already exists, then we return the
1733 * aliased dentry instead and drop one reference to inode.
1735 * Note that in order to avoid conflicts with rename() etc, the caller
1736 * had better be holding the parent directory semaphore.
1738 * This also assumes that the inode count has been incremented
1739 * (or otherwise set) by the caller to indicate that it is now
1740 * in use by the dcache.
1742 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1743 struct inode
*inode
)
1745 struct dentry
*alias
;
1746 int len
= entry
->d_name
.len
;
1747 const char *name
= entry
->d_name
.name
;
1748 unsigned int hash
= entry
->d_name
.hash
;
1751 __d_instantiate(entry
, NULL
);
1755 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1757 * Don't need alias->d_lock here, because aliases with
1758 * d_parent == entry->d_parent are not subject to name or
1759 * parent changes, because the parent inode i_mutex is held.
1761 if (alias
->d_name
.hash
!= hash
)
1763 if (alias
->d_parent
!= entry
->d_parent
)
1765 if (alias
->d_name
.len
!= len
)
1767 if (dentry_cmp(alias
, name
, len
))
1773 __d_instantiate(entry
, inode
);
1777 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1779 struct dentry
*result
;
1781 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1784 spin_lock(&inode
->i_lock
);
1785 result
= __d_instantiate_unique(entry
, inode
);
1787 spin_unlock(&inode
->i_lock
);
1790 security_d_instantiate(entry
, inode
);
1794 BUG_ON(!d_unhashed(result
));
1799 EXPORT_SYMBOL(d_instantiate_unique
);
1802 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1803 * @entry: dentry to complete
1804 * @inode: inode to attach to this dentry
1806 * Fill in inode information in the entry. If a directory alias is found, then
1807 * return an error (and drop inode). Together with d_materialise_unique() this
1808 * guarantees that a directory inode may never have more than one alias.
1810 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1812 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1814 spin_lock(&inode
->i_lock
);
1815 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1816 spin_unlock(&inode
->i_lock
);
1820 __d_instantiate(entry
, inode
);
1821 spin_unlock(&inode
->i_lock
);
1822 security_d_instantiate(entry
, inode
);
1826 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1828 struct dentry
*d_make_root(struct inode
*root_inode
)
1830 struct dentry
*res
= NULL
;
1833 static const struct qstr name
= QSTR_INIT("/", 1);
1835 res
= __d_alloc(root_inode
->i_sb
, &name
);
1837 d_instantiate(res
, root_inode
);
1843 EXPORT_SYMBOL(d_make_root
);
1845 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1847 struct dentry
*alias
;
1849 if (hlist_empty(&inode
->i_dentry
))
1851 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1857 * d_find_any_alias - find any alias for a given inode
1858 * @inode: inode to find an alias for
1860 * If any aliases exist for the given inode, take and return a
1861 * reference for one of them. If no aliases exist, return %NULL.
1863 struct dentry
*d_find_any_alias(struct inode
*inode
)
1867 spin_lock(&inode
->i_lock
);
1868 de
= __d_find_any_alias(inode
);
1869 spin_unlock(&inode
->i_lock
);
1872 EXPORT_SYMBOL(d_find_any_alias
);
1874 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1876 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1882 return ERR_PTR(-ESTALE
);
1884 return ERR_CAST(inode
);
1886 res
= d_find_any_alias(inode
);
1890 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1892 res
= ERR_PTR(-ENOMEM
);
1896 spin_lock(&inode
->i_lock
);
1897 res
= __d_find_any_alias(inode
);
1899 spin_unlock(&inode
->i_lock
);
1904 /* attach a disconnected dentry */
1905 add_flags
= d_flags_for_inode(inode
);
1908 add_flags
|= DCACHE_DISCONNECTED
;
1910 spin_lock(&tmp
->d_lock
);
1911 tmp
->d_inode
= inode
;
1912 tmp
->d_flags
|= add_flags
;
1913 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1914 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1915 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1916 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1917 spin_unlock(&tmp
->d_lock
);
1918 spin_unlock(&inode
->i_lock
);
1919 security_d_instantiate(tmp
, inode
);
1924 if (res
&& !IS_ERR(res
))
1925 security_d_instantiate(res
, inode
);
1931 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1932 * @inode: inode to allocate the dentry for
1934 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1935 * similar open by handle operations. The returned dentry may be anonymous,
1936 * or may have a full name (if the inode was already in the cache).
1938 * When called on a directory inode, we must ensure that the inode only ever
1939 * has one dentry. If a dentry is found, that is returned instead of
1940 * allocating a new one.
1942 * On successful return, the reference to the inode has been transferred
1943 * to the dentry. In case of an error the reference on the inode is released.
1944 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1945 * be passed in and the error will be propagated to the return value,
1946 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1948 struct dentry
*d_obtain_alias(struct inode
*inode
)
1950 return __d_obtain_alias(inode
, 1);
1952 EXPORT_SYMBOL(d_obtain_alias
);
1955 * d_obtain_root - find or allocate a dentry for a given inode
1956 * @inode: inode to allocate the dentry for
1958 * Obtain an IS_ROOT dentry for the root of a filesystem.
1960 * We must ensure that directory inodes only ever have one dentry. If a
1961 * dentry is found, that is returned instead of 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
1965 * released. A %NULL or IS_ERR inode may be passed in and will be the
1966 * error will be propagate to the return value, with a %NULL @inode
1967 * replaced by ERR_PTR(-ESTALE).
1969 struct dentry
*d_obtain_root(struct inode
*inode
)
1971 return __d_obtain_alias(inode
, 0);
1973 EXPORT_SYMBOL(d_obtain_root
);
1976 * d_add_ci - lookup or allocate new dentry with case-exact name
1977 * @inode: the inode case-insensitive lookup has found
1978 * @dentry: the negative dentry that was passed to the parent's lookup func
1979 * @name: the case-exact name to be associated with the returned dentry
1981 * This is to avoid filling the dcache with case-insensitive names to the
1982 * same inode, only the actual correct case is stored in the dcache for
1983 * case-insensitive filesystems.
1985 * For a case-insensitive lookup match and if the the case-exact dentry
1986 * already exists in in the dcache, use it and return it.
1988 * If no entry exists with the exact case name, allocate new dentry with
1989 * the exact case, and return the spliced entry.
1991 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1994 struct dentry
*found
;
1998 * First check if a dentry matching the name already exists,
1999 * if not go ahead and create it now.
2001 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2003 new = d_alloc(dentry
->d_parent
, name
);
2005 found
= ERR_PTR(-ENOMEM
);
2007 found
= d_splice_alias(inode
, new);
2018 EXPORT_SYMBOL(d_add_ci
);
2021 * Do the slow-case of the dentry name compare.
2023 * Unlike the dentry_cmp() function, we need to atomically
2024 * load the name and length information, so that the
2025 * filesystem can rely on them, and can use the 'name' and
2026 * 'len' information without worrying about walking off the
2027 * end of memory etc.
2029 * Thus the read_seqcount_retry() and the "duplicate" info
2030 * in arguments (the low-level filesystem should not look
2031 * at the dentry inode or name contents directly, since
2032 * rename can change them while we're in RCU mode).
2034 enum slow_d_compare
{
2040 static noinline
enum slow_d_compare
slow_dentry_cmp(
2041 const struct dentry
*parent
,
2042 struct dentry
*dentry
,
2044 const struct qstr
*name
)
2046 int tlen
= dentry
->d_name
.len
;
2047 const char *tname
= dentry
->d_name
.name
;
2049 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2051 return D_COMP_SEQRETRY
;
2053 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2054 return D_COMP_NOMATCH
;
2059 * __d_lookup_rcu - search for a dentry (racy, store-free)
2060 * @parent: parent dentry
2061 * @name: qstr of name we wish to find
2062 * @seqp: returns d_seq value at the point where the dentry was found
2063 * Returns: dentry, or NULL
2065 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2066 * resolution (store-free path walking) design described in
2067 * Documentation/filesystems/path-lookup.txt.
2069 * This is not to be used outside core vfs.
2071 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2072 * held, and rcu_read_lock held. The returned dentry must not be stored into
2073 * without taking d_lock and checking d_seq sequence count against @seq
2076 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2079 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2080 * the returned dentry, so long as its parent's seqlock is checked after the
2081 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2082 * is formed, giving integrity down the path walk.
2084 * NOTE! The caller *has* to check the resulting dentry against the sequence
2085 * number we've returned before using any of the resulting dentry state!
2087 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2088 const struct qstr
*name
,
2091 u64 hashlen
= name
->hash_len
;
2092 const unsigned char *str
= name
->name
;
2093 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2094 struct hlist_bl_node
*node
;
2095 struct dentry
*dentry
;
2098 * Note: There is significant duplication with __d_lookup_rcu which is
2099 * required to prevent single threaded performance regressions
2100 * especially on architectures where smp_rmb (in seqcounts) are costly.
2101 * Keep the two functions in sync.
2105 * The hash list is protected using RCU.
2107 * Carefully use d_seq when comparing a candidate dentry, to avoid
2108 * races with d_move().
2110 * It is possible that concurrent renames can mess up our list
2111 * walk here and result in missing our dentry, resulting in the
2112 * false-negative result. d_lookup() protects against concurrent
2113 * renames using rename_lock seqlock.
2115 * See Documentation/filesystems/path-lookup.txt for more details.
2117 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2122 * The dentry sequence count protects us from concurrent
2123 * renames, and thus protects parent and name fields.
2125 * The caller must perform a seqcount check in order
2126 * to do anything useful with the returned dentry.
2128 * NOTE! We do a "raw" seqcount_begin here. That means that
2129 * we don't wait for the sequence count to stabilize if it
2130 * is in the middle of a sequence change. If we do the slow
2131 * dentry compare, we will do seqretries until it is stable,
2132 * and if we end up with a successful lookup, we actually
2133 * want to exit RCU lookup anyway.
2135 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2136 if (dentry
->d_parent
!= parent
)
2138 if (d_unhashed(dentry
))
2141 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2142 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2145 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2148 case D_COMP_NOMATCH
:
2155 if (dentry
->d_name
.hash_len
!= hashlen
)
2158 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2165 * d_lookup - search for a dentry
2166 * @parent: parent dentry
2167 * @name: qstr of name we wish to find
2168 * Returns: dentry, or NULL
2170 * d_lookup searches the children of the parent dentry for the name in
2171 * question. If the dentry is found its reference count is incremented and the
2172 * dentry is returned. The caller must use dput to free the entry when it has
2173 * finished using it. %NULL is returned if the dentry does not exist.
2175 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2177 struct dentry
*dentry
;
2181 seq
= read_seqbegin(&rename_lock
);
2182 dentry
= __d_lookup(parent
, name
);
2185 } while (read_seqretry(&rename_lock
, seq
));
2188 EXPORT_SYMBOL(d_lookup
);
2191 * __d_lookup - search for a dentry (racy)
2192 * @parent: parent dentry
2193 * @name: qstr of name we wish to find
2194 * Returns: dentry, or NULL
2196 * __d_lookup is like d_lookup, however it may (rarely) return a
2197 * false-negative result due to unrelated rename activity.
2199 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2200 * however it must be used carefully, eg. with a following d_lookup in
2201 * the case of failure.
2203 * __d_lookup callers must be commented.
2205 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2207 unsigned int len
= name
->len
;
2208 unsigned int hash
= name
->hash
;
2209 const unsigned char *str
= name
->name
;
2210 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2211 struct hlist_bl_node
*node
;
2212 struct dentry
*found
= NULL
;
2213 struct dentry
*dentry
;
2216 * Note: There is significant duplication with __d_lookup_rcu which is
2217 * required to prevent single threaded performance regressions
2218 * especially on architectures where smp_rmb (in seqcounts) are costly.
2219 * Keep the two functions in sync.
2223 * The hash list is protected using RCU.
2225 * Take d_lock when comparing a candidate dentry, to avoid races
2228 * It is possible that concurrent renames can mess up our list
2229 * walk here and result in missing our dentry, resulting in the
2230 * false-negative result. d_lookup() protects against concurrent
2231 * renames using rename_lock seqlock.
2233 * See Documentation/filesystems/path-lookup.txt for more details.
2237 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2239 if (dentry
->d_name
.hash
!= hash
)
2242 spin_lock(&dentry
->d_lock
);
2243 if (dentry
->d_parent
!= parent
)
2245 if (d_unhashed(dentry
))
2249 * It is safe to compare names since d_move() cannot
2250 * change the qstr (protected by d_lock).
2252 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2253 int tlen
= dentry
->d_name
.len
;
2254 const char *tname
= dentry
->d_name
.name
;
2255 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2258 if (dentry
->d_name
.len
!= len
)
2260 if (dentry_cmp(dentry
, str
, len
))
2264 dentry
->d_lockref
.count
++;
2266 spin_unlock(&dentry
->d_lock
);
2269 spin_unlock(&dentry
->d_lock
);
2277 * d_hash_and_lookup - hash the qstr then search for a dentry
2278 * @dir: Directory to search in
2279 * @name: qstr of name we wish to find
2281 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2283 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2286 * Check for a fs-specific hash function. Note that we must
2287 * calculate the standard hash first, as the d_op->d_hash()
2288 * routine may choose to leave the hash value unchanged.
2290 name
->hash
= full_name_hash(name
->name
, name
->len
);
2291 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2292 int err
= dir
->d_op
->d_hash(dir
, name
);
2293 if (unlikely(err
< 0))
2294 return ERR_PTR(err
);
2296 return d_lookup(dir
, name
);
2298 EXPORT_SYMBOL(d_hash_and_lookup
);
2301 * When a file is deleted, we have two options:
2302 * - turn this dentry into a negative dentry
2303 * - unhash this dentry and free it.
2305 * Usually, we want to just turn this into
2306 * a negative dentry, but if anybody else is
2307 * currently using the dentry or the inode
2308 * we can't do that and we fall back on removing
2309 * it from the hash queues and waiting for
2310 * it to be deleted later when it has no users
2314 * d_delete - delete a dentry
2315 * @dentry: The dentry to delete
2317 * Turn the dentry into a negative dentry if possible, otherwise
2318 * remove it from the hash queues so it can be deleted later
2321 void d_delete(struct dentry
* dentry
)
2323 struct inode
*inode
;
2326 * Are we the only user?
2329 spin_lock(&dentry
->d_lock
);
2330 inode
= dentry
->d_inode
;
2331 isdir
= S_ISDIR(inode
->i_mode
);
2332 if (dentry
->d_lockref
.count
== 1) {
2333 if (!spin_trylock(&inode
->i_lock
)) {
2334 spin_unlock(&dentry
->d_lock
);
2338 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2339 dentry_unlink_inode(dentry
);
2340 fsnotify_nameremove(dentry
, isdir
);
2344 if (!d_unhashed(dentry
))
2347 spin_unlock(&dentry
->d_lock
);
2349 fsnotify_nameremove(dentry
, isdir
);
2351 EXPORT_SYMBOL(d_delete
);
2353 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2355 BUG_ON(!d_unhashed(entry
));
2357 entry
->d_flags
|= DCACHE_RCUACCESS
;
2358 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2362 static void _d_rehash(struct dentry
* entry
)
2364 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2368 * d_rehash - add an entry back to the hash
2369 * @entry: dentry to add to the hash
2371 * Adds a dentry to the hash according to its name.
2374 void d_rehash(struct dentry
* entry
)
2376 spin_lock(&entry
->d_lock
);
2378 spin_unlock(&entry
->d_lock
);
2380 EXPORT_SYMBOL(d_rehash
);
2383 * dentry_update_name_case - update case insensitive dentry with a new name
2384 * @dentry: dentry to be updated
2387 * Update a case insensitive dentry with new case of name.
2389 * dentry must have been returned by d_lookup with name @name. Old and new
2390 * name lengths must match (ie. no d_compare which allows mismatched name
2393 * Parent inode i_mutex must be held over d_lookup and into this call (to
2394 * keep renames and concurrent inserts, and readdir(2) away).
2396 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2398 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2399 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2401 spin_lock(&dentry
->d_lock
);
2402 write_seqcount_begin(&dentry
->d_seq
);
2403 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2404 write_seqcount_end(&dentry
->d_seq
);
2405 spin_unlock(&dentry
->d_lock
);
2407 EXPORT_SYMBOL(dentry_update_name_case
);
2409 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2411 if (unlikely(dname_external(target
))) {
2412 if (unlikely(dname_external(dentry
))) {
2414 * Both external: swap the pointers
2416 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2419 * dentry:internal, target:external. Steal target's
2420 * storage and make target internal.
2422 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2423 dentry
->d_name
.len
+ 1);
2424 dentry
->d_name
.name
= target
->d_name
.name
;
2425 target
->d_name
.name
= target
->d_iname
;
2428 if (unlikely(dname_external(dentry
))) {
2430 * dentry:external, target:internal. Give dentry's
2431 * storage to target and make dentry internal
2433 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2434 target
->d_name
.len
+ 1);
2435 target
->d_name
.name
= dentry
->d_name
.name
;
2436 dentry
->d_name
.name
= dentry
->d_iname
;
2439 * Both are internal.
2442 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2443 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2444 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2445 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2446 swap(((long *) &dentry
->d_iname
)[i
],
2447 ((long *) &target
->d_iname
)[i
]);
2451 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2454 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2456 struct external_name
*old_name
= NULL
;
2457 if (unlikely(dname_external(dentry
)))
2458 old_name
= external_name(dentry
);
2459 if (unlikely(dname_external(target
))) {
2460 atomic_inc(&external_name(target
)->u
.count
);
2461 dentry
->d_name
= target
->d_name
;
2463 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2464 target
->d_name
.len
+ 1);
2465 dentry
->d_name
.name
= dentry
->d_iname
;
2466 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2468 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2469 kfree_rcu(old_name
, u
.head
);
2472 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2475 * XXXX: do we really need to take target->d_lock?
2477 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2478 spin_lock(&target
->d_parent
->d_lock
);
2480 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2481 spin_lock(&dentry
->d_parent
->d_lock
);
2482 spin_lock_nested(&target
->d_parent
->d_lock
,
2483 DENTRY_D_LOCK_NESTED
);
2485 spin_lock(&target
->d_parent
->d_lock
);
2486 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2487 DENTRY_D_LOCK_NESTED
);
2490 if (target
< dentry
) {
2491 spin_lock_nested(&target
->d_lock
, 2);
2492 spin_lock_nested(&dentry
->d_lock
, 3);
2494 spin_lock_nested(&dentry
->d_lock
, 2);
2495 spin_lock_nested(&target
->d_lock
, 3);
2499 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2501 if (target
->d_parent
!= dentry
->d_parent
)
2502 spin_unlock(&dentry
->d_parent
->d_lock
);
2503 if (target
->d_parent
!= target
)
2504 spin_unlock(&target
->d_parent
->d_lock
);
2505 spin_unlock(&target
->d_lock
);
2506 spin_unlock(&dentry
->d_lock
);
2510 * When switching names, the actual string doesn't strictly have to
2511 * be preserved in the target - because we're dropping the target
2512 * anyway. As such, we can just do a simple memcpy() to copy over
2513 * the new name before we switch, unless we are going to rehash
2514 * it. Note that if we *do* unhash the target, we are not allowed
2515 * to rehash it without giving it a new name/hash key - whether
2516 * we swap or overwrite the names here, resulting name won't match
2517 * the reality in filesystem; it's only there for d_path() purposes.
2518 * Note that all of this is happening under rename_lock, so the
2519 * any hash lookup seeing it in the middle of manipulations will
2520 * be discarded anyway. So we do not care what happens to the hash
2524 * __d_move - move a dentry
2525 * @dentry: entry to move
2526 * @target: new dentry
2527 * @exchange: exchange the two dentries
2529 * Update the dcache to reflect the move of a file name. Negative
2530 * dcache entries should not be moved in this way. Caller must hold
2531 * rename_lock, the i_mutex of the source and target directories,
2532 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2534 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2537 if (!dentry
->d_inode
)
2538 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2540 BUG_ON(d_ancestor(dentry
, target
));
2541 BUG_ON(d_ancestor(target
, dentry
));
2543 dentry_lock_for_move(dentry
, target
);
2545 write_seqcount_begin(&dentry
->d_seq
);
2546 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2548 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2551 * Move the dentry to the target hash queue. Don't bother checking
2552 * for the same hash queue because of how unlikely it is.
2555 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2558 * Unhash the target (d_delete() is not usable here). If exchanging
2559 * the two dentries, then rehash onto the other's hash queue.
2564 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2567 /* Switch the names.. */
2569 swap_names(dentry
, target
);
2571 copy_name(dentry
, target
);
2573 /* ... and switch them in the tree */
2574 if (IS_ROOT(dentry
)) {
2575 /* splicing a tree */
2576 dentry
->d_parent
= target
->d_parent
;
2577 target
->d_parent
= target
;
2578 list_del_init(&target
->d_child
);
2579 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2581 /* swapping two dentries */
2582 swap(dentry
->d_parent
, target
->d_parent
);
2583 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2584 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2586 fsnotify_d_move(target
);
2587 fsnotify_d_move(dentry
);
2590 write_seqcount_end(&target
->d_seq
);
2591 write_seqcount_end(&dentry
->d_seq
);
2593 dentry_unlock_for_move(dentry
, target
);
2597 * d_move - move a dentry
2598 * @dentry: entry to move
2599 * @target: new dentry
2601 * Update the dcache to reflect the move of a file name. Negative
2602 * dcache entries should not be moved in this way. See the locking
2603 * requirements for __d_move.
2605 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2607 write_seqlock(&rename_lock
);
2608 __d_move(dentry
, target
, false);
2609 write_sequnlock(&rename_lock
);
2611 EXPORT_SYMBOL(d_move
);
2614 * d_exchange - exchange two dentries
2615 * @dentry1: first dentry
2616 * @dentry2: second dentry
2618 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2620 write_seqlock(&rename_lock
);
2622 WARN_ON(!dentry1
->d_inode
);
2623 WARN_ON(!dentry2
->d_inode
);
2624 WARN_ON(IS_ROOT(dentry1
));
2625 WARN_ON(IS_ROOT(dentry2
));
2627 __d_move(dentry1
, dentry2
, true);
2629 write_sequnlock(&rename_lock
);
2633 * d_ancestor - search for an ancestor
2634 * @p1: ancestor dentry
2637 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2638 * an ancestor of p2, else NULL.
2640 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2644 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2645 if (p
->d_parent
== p1
)
2652 * This helper attempts to cope with remotely renamed directories
2654 * It assumes that the caller is already holding
2655 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2657 * Note: If ever the locking in lock_rename() changes, then please
2658 * remember to update this too...
2660 static int __d_unalias(struct inode
*inode
,
2661 struct dentry
*dentry
, struct dentry
*alias
)
2663 struct mutex
*m1
= NULL
, *m2
= NULL
;
2666 /* If alias and dentry share a parent, then no extra locks required */
2667 if (alias
->d_parent
== dentry
->d_parent
)
2670 /* See lock_rename() */
2671 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2673 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2674 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2676 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2678 __d_move(alias
, dentry
, false);
2681 spin_unlock(&inode
->i_lock
);
2690 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2691 * @inode: the inode which may have a disconnected dentry
2692 * @dentry: a negative dentry which we want to point to the inode.
2694 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2695 * place of the given dentry and return it, else simply d_add the inode
2696 * to the dentry and return NULL.
2698 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2699 * we should error out: directories can't have multiple aliases.
2701 * This is needed in the lookup routine of any filesystem that is exportable
2702 * (via knfsd) so that we can build dcache paths to directories effectively.
2704 * If a dentry was found and moved, then it is returned. Otherwise NULL
2705 * is returned. This matches the expected return value of ->lookup.
2707 * Cluster filesystems may call this function with a negative, hashed dentry.
2708 * In that case, we know that the inode will be a regular file, and also this
2709 * will only occur during atomic_open. So we need to check for the dentry
2710 * being already hashed only in the final case.
2712 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2715 return ERR_CAST(inode
);
2717 BUG_ON(!d_unhashed(dentry
));
2720 __d_instantiate(dentry
, NULL
);
2723 spin_lock(&inode
->i_lock
);
2724 if (S_ISDIR(inode
->i_mode
)) {
2725 struct dentry
*new = __d_find_any_alias(inode
);
2726 if (unlikely(new)) {
2727 write_seqlock(&rename_lock
);
2728 if (unlikely(d_ancestor(new, dentry
))) {
2729 write_sequnlock(&rename_lock
);
2730 spin_unlock(&inode
->i_lock
);
2732 new = ERR_PTR(-ELOOP
);
2733 pr_warn_ratelimited(
2734 "VFS: Lookup of '%s' in %s %s"
2735 " would have caused loop\n",
2736 dentry
->d_name
.name
,
2737 inode
->i_sb
->s_type
->name
,
2739 } else if (!IS_ROOT(new)) {
2740 int err
= __d_unalias(inode
, dentry
, new);
2741 write_sequnlock(&rename_lock
);
2747 __d_move(new, dentry
, false);
2748 write_sequnlock(&rename_lock
);
2749 spin_unlock(&inode
->i_lock
);
2750 security_d_instantiate(new, inode
);
2756 /* already taking inode->i_lock, so d_add() by hand */
2757 __d_instantiate(dentry
, inode
);
2758 spin_unlock(&inode
->i_lock
);
2760 security_d_instantiate(dentry
, inode
);
2764 EXPORT_SYMBOL(d_splice_alias
);
2766 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2770 return -ENAMETOOLONG
;
2772 memcpy(*buffer
, str
, namelen
);
2777 * prepend_name - prepend a pathname in front of current buffer pointer
2778 * @buffer: buffer pointer
2779 * @buflen: allocated length of the buffer
2780 * @name: name string and length qstr structure
2782 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2783 * make sure that either the old or the new name pointer and length are
2784 * fetched. However, there may be mismatch between length and pointer.
2785 * The length cannot be trusted, we need to copy it byte-by-byte until
2786 * the length is reached or a null byte is found. It also prepends "/" at
2787 * the beginning of the name. The sequence number check at the caller will
2788 * retry it again when a d_move() does happen. So any garbage in the buffer
2789 * due to mismatched pointer and length will be discarded.
2791 * Data dependency barrier is needed to make sure that we see that terminating
2792 * NUL. Alpha strikes again, film at 11...
2794 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2796 const char *dname
= ACCESS_ONCE(name
->name
);
2797 u32 dlen
= ACCESS_ONCE(name
->len
);
2800 smp_read_barrier_depends();
2802 *buflen
-= dlen
+ 1;
2804 return -ENAMETOOLONG
;
2805 p
= *buffer
-= dlen
+ 1;
2817 * prepend_path - Prepend path string to a buffer
2818 * @path: the dentry/vfsmount to report
2819 * @root: root vfsmnt/dentry
2820 * @buffer: pointer to the end of the buffer
2821 * @buflen: pointer to buffer length
2823 * The function will first try to write out the pathname without taking any
2824 * lock other than the RCU read lock to make sure that dentries won't go away.
2825 * It only checks the sequence number of the global rename_lock as any change
2826 * in the dentry's d_seq will be preceded by changes in the rename_lock
2827 * sequence number. If the sequence number had been changed, it will restart
2828 * the whole pathname back-tracing sequence again by taking the rename_lock.
2829 * In this case, there is no need to take the RCU read lock as the recursive
2830 * parent pointer references will keep the dentry chain alive as long as no
2831 * rename operation is performed.
2833 static int prepend_path(const struct path
*path
,
2834 const struct path
*root
,
2835 char **buffer
, int *buflen
)
2837 struct dentry
*dentry
;
2838 struct vfsmount
*vfsmnt
;
2841 unsigned seq
, m_seq
= 0;
2847 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
2854 dentry
= path
->dentry
;
2856 mnt
= real_mount(vfsmnt
);
2857 read_seqbegin_or_lock(&rename_lock
, &seq
);
2858 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2859 struct dentry
* parent
;
2861 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2862 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
2864 if (mnt
!= parent
) {
2865 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
2871 * Filesystems needing to implement special "root names"
2872 * should do so with ->d_dname()
2874 if (IS_ROOT(dentry
) &&
2875 (dentry
->d_name
.len
!= 1 ||
2876 dentry
->d_name
.name
[0] != '/')) {
2877 WARN(1, "Root dentry has weird name <%.*s>\n",
2878 (int) dentry
->d_name
.len
,
2879 dentry
->d_name
.name
);
2882 error
= is_mounted(vfsmnt
) ? 1 : 2;
2885 parent
= dentry
->d_parent
;
2887 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
2895 if (need_seqretry(&rename_lock
, seq
)) {
2899 done_seqretry(&rename_lock
, seq
);
2903 if (need_seqretry(&mount_lock
, m_seq
)) {
2907 done_seqretry(&mount_lock
, m_seq
);
2909 if (error
>= 0 && bptr
== *buffer
) {
2911 error
= -ENAMETOOLONG
;
2921 * __d_path - return the path of a dentry
2922 * @path: the dentry/vfsmount to report
2923 * @root: root vfsmnt/dentry
2924 * @buf: buffer to return value in
2925 * @buflen: buffer length
2927 * Convert a dentry into an ASCII path name.
2929 * Returns a pointer into the buffer or an error code if the
2930 * path was too long.
2932 * "buflen" should be positive.
2934 * If the path is not reachable from the supplied root, return %NULL.
2936 char *__d_path(const struct path
*path
,
2937 const struct path
*root
,
2938 char *buf
, int buflen
)
2940 char *res
= buf
+ buflen
;
2943 prepend(&res
, &buflen
, "\0", 1);
2944 error
= prepend_path(path
, root
, &res
, &buflen
);
2947 return ERR_PTR(error
);
2953 char *d_absolute_path(const struct path
*path
,
2954 char *buf
, int buflen
)
2956 struct path root
= {};
2957 char *res
= buf
+ buflen
;
2960 prepend(&res
, &buflen
, "\0", 1);
2961 error
= prepend_path(path
, &root
, &res
, &buflen
);
2966 return ERR_PTR(error
);
2971 * same as __d_path but appends "(deleted)" for unlinked files.
2973 static int path_with_deleted(const struct path
*path
,
2974 const struct path
*root
,
2975 char **buf
, int *buflen
)
2977 prepend(buf
, buflen
, "\0", 1);
2978 if (d_unlinked(path
->dentry
)) {
2979 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2984 return prepend_path(path
, root
, buf
, buflen
);
2987 static int prepend_unreachable(char **buffer
, int *buflen
)
2989 return prepend(buffer
, buflen
, "(unreachable)", 13);
2992 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
2997 seq
= read_seqcount_begin(&fs
->seq
);
2999 } while (read_seqcount_retry(&fs
->seq
, seq
));
3003 * d_path - return the path of a dentry
3004 * @path: path to report
3005 * @buf: buffer to return value in
3006 * @buflen: buffer length
3008 * Convert a dentry into an ASCII path name. If the entry has been deleted
3009 * the string " (deleted)" is appended. Note that this is ambiguous.
3011 * Returns a pointer into the buffer or an error code if the path was
3012 * too long. Note: Callers should use the returned pointer, not the passed
3013 * in buffer, to use the name! The implementation often starts at an offset
3014 * into the buffer, and may leave 0 bytes at the start.
3016 * "buflen" should be positive.
3018 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3020 char *res
= buf
+ buflen
;
3025 * We have various synthetic filesystems that never get mounted. On
3026 * these filesystems dentries are never used for lookup purposes, and
3027 * thus don't need to be hashed. They also don't need a name until a
3028 * user wants to identify the object in /proc/pid/fd/. The little hack
3029 * below allows us to generate a name for these objects on demand:
3031 * Some pseudo inodes are mountable. When they are mounted
3032 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3033 * and instead have d_path return the mounted path.
3035 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3036 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3037 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3040 get_fs_root_rcu(current
->fs
, &root
);
3041 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3045 res
= ERR_PTR(error
);
3048 EXPORT_SYMBOL(d_path
);
3051 * Helper function for dentry_operations.d_dname() members
3053 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3054 const char *fmt
, ...)
3060 va_start(args
, fmt
);
3061 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3064 if (sz
> sizeof(temp
) || sz
> buflen
)
3065 return ERR_PTR(-ENAMETOOLONG
);
3067 buffer
+= buflen
- sz
;
3068 return memcpy(buffer
, temp
, sz
);
3071 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3073 char *end
= buffer
+ buflen
;
3074 /* these dentries are never renamed, so d_lock is not needed */
3075 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3076 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3077 prepend(&end
, &buflen
, "/", 1))
3078 end
= ERR_PTR(-ENAMETOOLONG
);
3081 EXPORT_SYMBOL(simple_dname
);
3084 * Write full pathname from the root of the filesystem into the buffer.
3086 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3088 struct dentry
*dentry
;
3101 prepend(&end
, &len
, "\0", 1);
3105 read_seqbegin_or_lock(&rename_lock
, &seq
);
3106 while (!IS_ROOT(dentry
)) {
3107 struct dentry
*parent
= dentry
->d_parent
;
3110 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3119 if (need_seqretry(&rename_lock
, seq
)) {
3123 done_seqretry(&rename_lock
, seq
);
3128 return ERR_PTR(-ENAMETOOLONG
);
3131 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3133 return __dentry_path(dentry
, buf
, buflen
);
3135 EXPORT_SYMBOL(dentry_path_raw
);
3137 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3142 if (d_unlinked(dentry
)) {
3144 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3148 retval
= __dentry_path(dentry
, buf
, buflen
);
3149 if (!IS_ERR(retval
) && p
)
3150 *p
= '/'; /* restore '/' overriden with '\0' */
3153 return ERR_PTR(-ENAMETOOLONG
);
3156 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3162 seq
= read_seqcount_begin(&fs
->seq
);
3165 } while (read_seqcount_retry(&fs
->seq
, seq
));
3169 * NOTE! The user-level library version returns a
3170 * character pointer. The kernel system call just
3171 * returns the length of the buffer filled (which
3172 * includes the ending '\0' character), or a negative
3173 * error value. So libc would do something like
3175 * char *getcwd(char * buf, size_t size)
3179 * retval = sys_getcwd(buf, size);
3186 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3189 struct path pwd
, root
;
3190 char *page
= __getname();
3196 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3199 if (!d_unlinked(pwd
.dentry
)) {
3201 char *cwd
= page
+ PATH_MAX
;
3202 int buflen
= PATH_MAX
;
3204 prepend(&cwd
, &buflen
, "\0", 1);
3205 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3211 /* Unreachable from current root */
3213 error
= prepend_unreachable(&cwd
, &buflen
);
3219 len
= PATH_MAX
+ page
- cwd
;
3222 if (copy_to_user(buf
, cwd
, len
))
3235 * Test whether new_dentry is a subdirectory of old_dentry.
3237 * Trivially implemented using the dcache structure
3241 * is_subdir - is new dentry a subdirectory of old_dentry
3242 * @new_dentry: new dentry
3243 * @old_dentry: old dentry
3245 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3246 * Returns 0 otherwise.
3247 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3250 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3255 if (new_dentry
== old_dentry
)
3259 /* for restarting inner loop in case of seq retry */
3260 seq
= read_seqbegin(&rename_lock
);
3262 * Need rcu_readlock to protect against the d_parent trashing
3266 if (d_ancestor(old_dentry
, new_dentry
))
3271 } while (read_seqretry(&rename_lock
, seq
));
3276 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3278 struct dentry
*root
= data
;
3279 if (dentry
!= root
) {
3280 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3283 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3284 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3285 dentry
->d_lockref
.count
--;
3288 return D_WALK_CONTINUE
;
3291 void d_genocide(struct dentry
*parent
)
3293 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3296 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3298 inode_dec_link_count(inode
);
3299 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3300 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3301 !d_unlinked(dentry
));
3302 spin_lock(&dentry
->d_parent
->d_lock
);
3303 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3304 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3305 (unsigned long long)inode
->i_ino
);
3306 spin_unlock(&dentry
->d_lock
);
3307 spin_unlock(&dentry
->d_parent
->d_lock
);
3308 d_instantiate(dentry
, inode
);
3310 EXPORT_SYMBOL(d_tmpfile
);
3312 static __initdata
unsigned long dhash_entries
;
3313 static int __init
set_dhash_entries(char *str
)
3317 dhash_entries
= simple_strtoul(str
, &str
, 0);
3320 __setup("dhash_entries=", set_dhash_entries
);
3322 static void __init
dcache_init_early(void)
3326 /* If hashes are distributed across NUMA nodes, defer
3327 * hash allocation until vmalloc space is available.
3333 alloc_large_system_hash("Dentry cache",
3334 sizeof(struct hlist_bl_head
),
3343 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3344 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3347 static void __init
dcache_init(void)
3352 * A constructor could be added for stable state like the lists,
3353 * but it is probably not worth it because of the cache nature
3356 dentry_cache
= KMEM_CACHE(dentry
,
3357 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3359 /* Hash may have been set up in dcache_init_early */
3364 alloc_large_system_hash("Dentry cache",
3365 sizeof(struct hlist_bl_head
),
3374 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3375 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3378 /* SLAB cache for __getname() consumers */
3379 struct kmem_cache
*names_cachep __read_mostly
;
3380 EXPORT_SYMBOL(names_cachep
);
3382 EXPORT_SYMBOL(d_genocide
);
3384 void __init
vfs_caches_init_early(void)
3386 dcache_init_early();
3390 void __init
vfs_caches_init(unsigned long mempages
)
3392 unsigned long reserve
;
3394 /* Base hash sizes on available memory, with a reserve equal to
3395 150% of current kernel size */
3397 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3398 mempages
-= reserve
;
3400 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3401 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3405 files_init(mempages
);