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_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
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 WARN_ON(!hlist_unhashed(&dentry
->d_alias
));
256 kmem_cache_free(dentry_cache
, dentry
);
259 static void __d_free_external(struct rcu_head
*head
)
261 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
262 WARN_ON(!hlist_unhashed(&dentry
->d_alias
));
263 kfree(external_name(dentry
));
264 kmem_cache_free(dentry_cache
, dentry
);
267 static void dentry_free(struct dentry
*dentry
)
269 if (unlikely(dname_external(dentry
))) {
270 struct external_name
*p
= external_name(dentry
);
271 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
272 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
276 /* if dentry was never visible to RCU, immediate free is OK */
277 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
278 __d_free(&dentry
->d_u
.d_rcu
);
280 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
284 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
285 * @dentry: the target dentry
286 * After this call, in-progress rcu-walk path lookup will fail. This
287 * should be called after unhashing, and after changing d_inode (if
288 * the dentry has not already been unhashed).
290 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
292 assert_spin_locked(&dentry
->d_lock
);
293 /* Go through a barrier */
294 write_seqcount_barrier(&dentry
->d_seq
);
298 * Release the dentry's inode, using the filesystem
299 * d_iput() operation if defined. Dentry has no refcount
302 static void dentry_iput(struct dentry
* dentry
)
303 __releases(dentry
->d_lock
)
304 __releases(dentry
->d_inode
->i_lock
)
306 struct inode
*inode
= dentry
->d_inode
;
308 dentry
->d_inode
= NULL
;
309 hlist_del_init(&dentry
->d_alias
);
310 spin_unlock(&dentry
->d_lock
);
311 spin_unlock(&inode
->i_lock
);
313 fsnotify_inoderemove(inode
);
314 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
315 dentry
->d_op
->d_iput(dentry
, inode
);
319 spin_unlock(&dentry
->d_lock
);
324 * Release the dentry's inode, using the filesystem
325 * d_iput() operation if defined. dentry remains in-use.
327 static void dentry_unlink_inode(struct dentry
* dentry
)
328 __releases(dentry
->d_lock
)
329 __releases(dentry
->d_inode
->i_lock
)
331 struct inode
*inode
= dentry
->d_inode
;
332 __d_clear_type(dentry
);
333 dentry
->d_inode
= NULL
;
334 hlist_del_init(&dentry
->d_alias
);
335 dentry_rcuwalk_barrier(dentry
);
336 spin_unlock(&dentry
->d_lock
);
337 spin_unlock(&inode
->i_lock
);
339 fsnotify_inoderemove(inode
);
340 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
341 dentry
->d_op
->d_iput(dentry
, inode
);
347 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
348 * is in use - which includes both the "real" per-superblock
349 * LRU list _and_ the DCACHE_SHRINK_LIST use.
351 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
352 * on the shrink list (ie not on the superblock LRU list).
354 * The per-cpu "nr_dentry_unused" counters are updated with
355 * the DCACHE_LRU_LIST bit.
357 * These helper functions make sure we always follow the
358 * rules. d_lock must be held by the caller.
360 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
361 static void d_lru_add(struct dentry
*dentry
)
363 D_FLAG_VERIFY(dentry
, 0);
364 dentry
->d_flags
|= DCACHE_LRU_LIST
;
365 this_cpu_inc(nr_dentry_unused
);
366 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
369 static void d_lru_del(struct dentry
*dentry
)
371 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
372 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
373 this_cpu_dec(nr_dentry_unused
);
374 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
377 static void d_shrink_del(struct dentry
*dentry
)
379 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
380 list_del_init(&dentry
->d_lru
);
381 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
382 this_cpu_dec(nr_dentry_unused
);
385 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
387 D_FLAG_VERIFY(dentry
, 0);
388 list_add(&dentry
->d_lru
, list
);
389 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
390 this_cpu_inc(nr_dentry_unused
);
394 * These can only be called under the global LRU lock, ie during the
395 * callback for freeing the LRU list. "isolate" removes it from the
396 * LRU lists entirely, while shrink_move moves it to the indicated
399 static void d_lru_isolate(struct dentry
*dentry
)
401 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
402 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
403 this_cpu_dec(nr_dentry_unused
);
404 list_del_init(&dentry
->d_lru
);
407 static void d_lru_shrink_move(struct dentry
*dentry
, struct list_head
*list
)
409 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
410 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
411 list_move_tail(&dentry
->d_lru
, list
);
415 * dentry_lru_(add|del)_list) must be called with d_lock held.
417 static void dentry_lru_add(struct dentry
*dentry
)
419 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
424 * d_drop - drop a dentry
425 * @dentry: dentry to drop
427 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
428 * be found through a VFS lookup any more. Note that this is different from
429 * deleting the dentry - d_delete will try to mark the dentry negative if
430 * possible, giving a successful _negative_ lookup, while d_drop will
431 * just make the cache lookup fail.
433 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
434 * reason (NFS timeouts or autofs deletes).
436 * __d_drop requires dentry->d_lock.
438 void __d_drop(struct dentry
*dentry
)
440 if (!d_unhashed(dentry
)) {
441 struct hlist_bl_head
*b
;
443 * Hashed dentries are normally on the dentry hashtable,
444 * with the exception of those newly allocated by
445 * d_obtain_alias, which are always IS_ROOT:
447 if (unlikely(IS_ROOT(dentry
)))
448 b
= &dentry
->d_sb
->s_anon
;
450 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
453 __hlist_bl_del(&dentry
->d_hash
);
454 dentry
->d_hash
.pprev
= NULL
;
456 dentry_rcuwalk_barrier(dentry
);
459 EXPORT_SYMBOL(__d_drop
);
461 void d_drop(struct dentry
*dentry
)
463 spin_lock(&dentry
->d_lock
);
465 spin_unlock(&dentry
->d_lock
);
467 EXPORT_SYMBOL(d_drop
);
469 static void __dentry_kill(struct dentry
*dentry
)
471 struct dentry
*parent
= NULL
;
472 bool can_free
= true;
473 if (!IS_ROOT(dentry
))
474 parent
= dentry
->d_parent
;
477 * The dentry is now unrecoverably dead to the world.
479 lockref_mark_dead(&dentry
->d_lockref
);
482 * inform the fs via d_prune that this dentry is about to be
483 * unhashed and destroyed.
485 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
486 dentry
->d_op
->d_prune(dentry
);
488 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
489 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
492 /* if it was on the hash then remove it */
494 list_del(&dentry
->d_u
.d_child
);
496 * Inform d_walk() that we are no longer attached to the
499 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
501 spin_unlock(&parent
->d_lock
);
504 * dentry_iput drops the locks, at which point nobody (except
505 * transient RCU lookups) can reach this dentry.
507 BUG_ON((int)dentry
->d_lockref
.count
> 0);
508 this_cpu_dec(nr_dentry
);
509 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
510 dentry
->d_op
->d_release(dentry
);
512 spin_lock(&dentry
->d_lock
);
513 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
514 dentry
->d_flags
|= DCACHE_MAY_FREE
;
517 spin_unlock(&dentry
->d_lock
);
518 if (likely(can_free
))
523 * Finish off a dentry we've decided to kill.
524 * dentry->d_lock must be held, returns with it unlocked.
525 * If ref is non-zero, then decrement the refcount too.
526 * Returns dentry requiring refcount drop, or NULL if we're done.
528 static struct dentry
*dentry_kill(struct dentry
*dentry
)
529 __releases(dentry
->d_lock
)
531 struct inode
*inode
= dentry
->d_inode
;
532 struct dentry
*parent
= NULL
;
534 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
537 if (!IS_ROOT(dentry
)) {
538 parent
= dentry
->d_parent
;
539 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
541 spin_unlock(&inode
->i_lock
);
546 __dentry_kill(dentry
);
550 spin_unlock(&dentry
->d_lock
);
552 return dentry
; /* try again with same dentry */
555 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
557 struct dentry
*parent
= dentry
->d_parent
;
560 if (unlikely((int)dentry
->d_lockref
.count
< 0))
562 if (likely(spin_trylock(&parent
->d_lock
)))
565 spin_unlock(&dentry
->d_lock
);
567 parent
= ACCESS_ONCE(dentry
->d_parent
);
568 spin_lock(&parent
->d_lock
);
570 * We can't blindly lock dentry until we are sure
571 * that we won't violate the locking order.
572 * Any changes of dentry->d_parent must have
573 * been done with parent->d_lock held, so
574 * spin_lock() above is enough of a barrier
575 * for checking if it's still our child.
577 if (unlikely(parent
!= dentry
->d_parent
)) {
578 spin_unlock(&parent
->d_lock
);
582 if (parent
!= dentry
)
583 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
592 * This is complicated by the fact that we do not want to put
593 * dentries that are no longer on any hash chain on the unused
594 * list: we'd much rather just get rid of them immediately.
596 * However, that implies that we have to traverse the dentry
597 * tree upwards to the parents which might _also_ now be
598 * scheduled for deletion (it may have been only waiting for
599 * its last child to go away).
601 * This tail recursion is done by hand as we don't want to depend
602 * on the compiler to always get this right (gcc generally doesn't).
603 * Real recursion would eat up our stack space.
607 * dput - release a dentry
608 * @dentry: dentry to release
610 * Release a dentry. This will drop the usage count and if appropriate
611 * call the dentry unlink method as well as removing it from the queues and
612 * releasing its resources. If the parent dentries were scheduled for release
613 * they too may now get deleted.
615 void dput(struct dentry
*dentry
)
617 if (unlikely(!dentry
))
621 if (lockref_put_or_lock(&dentry
->d_lockref
))
624 /* Unreachable? Get rid of it */
625 if (unlikely(d_unhashed(dentry
)))
628 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
629 if (dentry
->d_op
->d_delete(dentry
))
633 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
634 dentry
->d_flags
|= DCACHE_REFERENCED
;
635 dentry_lru_add(dentry
);
637 dentry
->d_lockref
.count
--;
638 spin_unlock(&dentry
->d_lock
);
642 dentry
= dentry_kill(dentry
);
649 /* This must be called with d_lock held */
650 static inline void __dget_dlock(struct dentry
*dentry
)
652 dentry
->d_lockref
.count
++;
655 static inline void __dget(struct dentry
*dentry
)
657 lockref_get(&dentry
->d_lockref
);
660 struct dentry
*dget_parent(struct dentry
*dentry
)
666 * Do optimistic parent lookup without any
670 ret
= ACCESS_ONCE(dentry
->d_parent
);
671 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
673 if (likely(gotref
)) {
674 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
681 * Don't need rcu_dereference because we re-check it was correct under
685 ret
= dentry
->d_parent
;
686 spin_lock(&ret
->d_lock
);
687 if (unlikely(ret
!= dentry
->d_parent
)) {
688 spin_unlock(&ret
->d_lock
);
693 BUG_ON(!ret
->d_lockref
.count
);
694 ret
->d_lockref
.count
++;
695 spin_unlock(&ret
->d_lock
);
698 EXPORT_SYMBOL(dget_parent
);
701 * d_find_alias - grab a hashed alias of inode
702 * @inode: inode in question
704 * If inode has a hashed alias, or is a directory and has any alias,
705 * acquire the reference to alias and return it. Otherwise return NULL.
706 * Notice that if inode is a directory there can be only one alias and
707 * it can be unhashed only if it has no children, or if it is the root
708 * of a filesystem, or if the directory was renamed and d_revalidate
709 * was the first vfs operation to notice.
711 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
712 * any other hashed alias over that one.
714 static struct dentry
*__d_find_alias(struct inode
*inode
)
716 struct dentry
*alias
, *discon_alias
;
720 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
721 spin_lock(&alias
->d_lock
);
722 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
723 if (IS_ROOT(alias
) &&
724 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
725 discon_alias
= alias
;
728 spin_unlock(&alias
->d_lock
);
732 spin_unlock(&alias
->d_lock
);
735 alias
= discon_alias
;
736 spin_lock(&alias
->d_lock
);
737 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
739 spin_unlock(&alias
->d_lock
);
742 spin_unlock(&alias
->d_lock
);
748 struct dentry
*d_find_alias(struct inode
*inode
)
750 struct dentry
*de
= NULL
;
752 if (!hlist_empty(&inode
->i_dentry
)) {
753 spin_lock(&inode
->i_lock
);
754 de
= __d_find_alias(inode
);
755 spin_unlock(&inode
->i_lock
);
759 EXPORT_SYMBOL(d_find_alias
);
762 * Try to kill dentries associated with this inode.
763 * WARNING: you must own a reference to inode.
765 void d_prune_aliases(struct inode
*inode
)
767 struct dentry
*dentry
;
769 spin_lock(&inode
->i_lock
);
770 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
771 spin_lock(&dentry
->d_lock
);
772 if (!dentry
->d_lockref
.count
) {
773 struct dentry
*parent
= lock_parent(dentry
);
774 if (likely(!dentry
->d_lockref
.count
)) {
775 __dentry_kill(dentry
);
779 spin_unlock(&parent
->d_lock
);
781 spin_unlock(&dentry
->d_lock
);
783 spin_unlock(&inode
->i_lock
);
785 EXPORT_SYMBOL(d_prune_aliases
);
787 static void shrink_dentry_list(struct list_head
*list
)
789 struct dentry
*dentry
, *parent
;
791 while (!list_empty(list
)) {
793 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
794 spin_lock(&dentry
->d_lock
);
795 parent
= lock_parent(dentry
);
798 * The dispose list is isolated and dentries are not accounted
799 * to the LRU here, so we can simply remove it from the list
800 * here regardless of whether it is referenced or not.
802 d_shrink_del(dentry
);
805 * We found an inuse dentry which was not removed from
806 * the LRU because of laziness during lookup. Do not free it.
808 if ((int)dentry
->d_lockref
.count
> 0) {
809 spin_unlock(&dentry
->d_lock
);
811 spin_unlock(&parent
->d_lock
);
816 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
817 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
818 spin_unlock(&dentry
->d_lock
);
820 spin_unlock(&parent
->d_lock
);
826 inode
= dentry
->d_inode
;
827 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
828 d_shrink_add(dentry
, list
);
829 spin_unlock(&dentry
->d_lock
);
831 spin_unlock(&parent
->d_lock
);
835 __dentry_kill(dentry
);
838 * We need to prune ancestors too. This is necessary to prevent
839 * quadratic behavior of shrink_dcache_parent(), but is also
840 * expected to be beneficial in reducing dentry cache
844 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
845 parent
= lock_parent(dentry
);
846 if (dentry
->d_lockref
.count
!= 1) {
847 dentry
->d_lockref
.count
--;
848 spin_unlock(&dentry
->d_lock
);
850 spin_unlock(&parent
->d_lock
);
853 inode
= dentry
->d_inode
; /* can't be NULL */
854 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
855 spin_unlock(&dentry
->d_lock
);
857 spin_unlock(&parent
->d_lock
);
861 __dentry_kill(dentry
);
867 static enum lru_status
868 dentry_lru_isolate(struct list_head
*item
, spinlock_t
*lru_lock
, void *arg
)
870 struct list_head
*freeable
= arg
;
871 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
875 * we are inverting the lru lock/dentry->d_lock here,
876 * so use a trylock. If we fail to get the lock, just skip
879 if (!spin_trylock(&dentry
->d_lock
))
883 * Referenced dentries are still in use. If they have active
884 * counts, just remove them from the LRU. Otherwise give them
885 * another pass through the LRU.
887 if (dentry
->d_lockref
.count
) {
888 d_lru_isolate(dentry
);
889 spin_unlock(&dentry
->d_lock
);
893 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
894 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
895 spin_unlock(&dentry
->d_lock
);
898 * The list move itself will be made by the common LRU code. At
899 * this point, we've dropped the dentry->d_lock but keep the
900 * lru lock. This is safe to do, since every list movement is
901 * protected by the lru lock even if both locks are held.
903 * This is guaranteed by the fact that all LRU management
904 * functions are intermediated by the LRU API calls like
905 * list_lru_add and list_lru_del. List movement in this file
906 * only ever occur through this functions or through callbacks
907 * like this one, that are called from the LRU API.
909 * The only exceptions to this are functions like
910 * shrink_dentry_list, and code that first checks for the
911 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
912 * operating only with stack provided lists after they are
913 * properly isolated from the main list. It is thus, always a
919 d_lru_shrink_move(dentry
, freeable
);
920 spin_unlock(&dentry
->d_lock
);
926 * prune_dcache_sb - shrink the dcache
928 * @nr_to_scan : number of entries to try to free
929 * @nid: which node to scan for freeable entities
931 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
932 * done when we need more memory an called from the superblock shrinker
935 * This function may fail to free any resources if all the dentries are in
938 long prune_dcache_sb(struct super_block
*sb
, unsigned long nr_to_scan
,
944 freed
= list_lru_walk_node(&sb
->s_dentry_lru
, nid
, dentry_lru_isolate
,
945 &dispose
, &nr_to_scan
);
946 shrink_dentry_list(&dispose
);
950 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
951 spinlock_t
*lru_lock
, void *arg
)
953 struct list_head
*freeable
= arg
;
954 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
957 * we are inverting the lru lock/dentry->d_lock here,
958 * so use a trylock. If we fail to get the lock, just skip
961 if (!spin_trylock(&dentry
->d_lock
))
964 d_lru_shrink_move(dentry
, freeable
);
965 spin_unlock(&dentry
->d_lock
);
972 * shrink_dcache_sb - shrink dcache for a superblock
975 * Shrink the dcache for the specified super block. This is used to free
976 * the dcache before unmounting a file system.
978 void shrink_dcache_sb(struct super_block
*sb
)
985 freed
= list_lru_walk(&sb
->s_dentry_lru
,
986 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
988 this_cpu_sub(nr_dentry_unused
, freed
);
989 shrink_dentry_list(&dispose
);
992 EXPORT_SYMBOL(shrink_dcache_sb
);
995 * enum d_walk_ret - action to talke during tree walk
996 * @D_WALK_CONTINUE: contrinue walk
997 * @D_WALK_QUIT: quit walk
998 * @D_WALK_NORETRY: quit when retry is needed
999 * @D_WALK_SKIP: skip this dentry and its children
1009 * d_walk - walk the dentry tree
1010 * @parent: start of walk
1011 * @data: data passed to @enter() and @finish()
1012 * @enter: callback when first entering the dentry
1013 * @finish: callback when successfully finished the walk
1015 * The @enter() and @finish() callbacks are called with d_lock held.
1017 static void d_walk(struct dentry
*parent
, void *data
,
1018 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1019 void (*finish
)(void *))
1021 struct dentry
*this_parent
;
1022 struct list_head
*next
;
1024 enum d_walk_ret ret
;
1028 read_seqbegin_or_lock(&rename_lock
, &seq
);
1029 this_parent
= parent
;
1030 spin_lock(&this_parent
->d_lock
);
1032 ret
= enter(data
, this_parent
);
1034 case D_WALK_CONTINUE
:
1039 case D_WALK_NORETRY
:
1044 next
= this_parent
->d_subdirs
.next
;
1046 while (next
!= &this_parent
->d_subdirs
) {
1047 struct list_head
*tmp
= next
;
1048 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1051 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1053 ret
= enter(data
, dentry
);
1055 case D_WALK_CONTINUE
:
1058 spin_unlock(&dentry
->d_lock
);
1060 case D_WALK_NORETRY
:
1064 spin_unlock(&dentry
->d_lock
);
1068 if (!list_empty(&dentry
->d_subdirs
)) {
1069 spin_unlock(&this_parent
->d_lock
);
1070 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1071 this_parent
= dentry
;
1072 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1075 spin_unlock(&dentry
->d_lock
);
1078 * All done at this level ... ascend and resume the search.
1080 if (this_parent
!= parent
) {
1081 struct dentry
*child
= this_parent
;
1082 this_parent
= child
->d_parent
;
1085 spin_unlock(&child
->d_lock
);
1086 spin_lock(&this_parent
->d_lock
);
1089 * might go back up the wrong parent if we have had a rename
1092 if (this_parent
!= child
->d_parent
||
1093 (child
->d_flags
& DCACHE_DENTRY_KILLED
) ||
1094 need_seqretry(&rename_lock
, seq
)) {
1095 spin_unlock(&this_parent
->d_lock
);
1100 next
= child
->d_u
.d_child
.next
;
1103 if (need_seqretry(&rename_lock
, seq
)) {
1104 spin_unlock(&this_parent
->d_lock
);
1111 spin_unlock(&this_parent
->d_lock
);
1112 done_seqretry(&rename_lock
, seq
);
1123 * Search for at least 1 mount point in the dentry's subdirs.
1124 * We descend to the next level whenever the d_subdirs
1125 * list is non-empty and continue searching.
1128 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1131 if (d_mountpoint(dentry
)) {
1135 return D_WALK_CONTINUE
;
1139 * have_submounts - check for mounts over a dentry
1140 * @parent: dentry to check.
1142 * Return true if the parent or its subdirectories contain
1145 int have_submounts(struct dentry
*parent
)
1149 d_walk(parent
, &ret
, check_mount
, NULL
);
1153 EXPORT_SYMBOL(have_submounts
);
1156 * Called by mount code to set a mountpoint and check if the mountpoint is
1157 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1158 * subtree can become unreachable).
1160 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1161 * this reason take rename_lock and d_lock on dentry and ancestors.
1163 int d_set_mounted(struct dentry
*dentry
)
1167 write_seqlock(&rename_lock
);
1168 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1169 /* Need exclusion wrt. d_invalidate() */
1170 spin_lock(&p
->d_lock
);
1171 if (unlikely(d_unhashed(p
))) {
1172 spin_unlock(&p
->d_lock
);
1175 spin_unlock(&p
->d_lock
);
1177 spin_lock(&dentry
->d_lock
);
1178 if (!d_unlinked(dentry
)) {
1179 dentry
->d_flags
|= DCACHE_MOUNTED
;
1182 spin_unlock(&dentry
->d_lock
);
1184 write_sequnlock(&rename_lock
);
1189 * Search the dentry child list of the specified parent,
1190 * and move any unused dentries to the end of the unused
1191 * list for prune_dcache(). We descend to the next level
1192 * whenever the d_subdirs list is non-empty and continue
1195 * It returns zero iff there are no unused children,
1196 * otherwise it returns the number of children moved to
1197 * the end of the unused list. This may not be the total
1198 * number of unused children, because select_parent can
1199 * drop the lock and return early due to latency
1203 struct select_data
{
1204 struct dentry
*start
;
1205 struct list_head dispose
;
1209 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1211 struct select_data
*data
= _data
;
1212 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1214 if (data
->start
== dentry
)
1217 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1220 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1222 if (!dentry
->d_lockref
.count
) {
1223 d_shrink_add(dentry
, &data
->dispose
);
1228 * We can return to the caller if we have found some (this
1229 * ensures forward progress). We'll be coming back to find
1232 if (!list_empty(&data
->dispose
))
1233 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1239 * shrink_dcache_parent - prune dcache
1240 * @parent: parent of entries to prune
1242 * Prune the dcache to remove unused children of the parent dentry.
1244 void shrink_dcache_parent(struct dentry
*parent
)
1247 struct select_data data
;
1249 INIT_LIST_HEAD(&data
.dispose
);
1250 data
.start
= parent
;
1253 d_walk(parent
, &data
, select_collect
, NULL
);
1257 shrink_dentry_list(&data
.dispose
);
1261 EXPORT_SYMBOL(shrink_dcache_parent
);
1263 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1265 /* it has busy descendents; complain about those instead */
1266 if (!list_empty(&dentry
->d_subdirs
))
1267 return D_WALK_CONTINUE
;
1269 /* root with refcount 1 is fine */
1270 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1271 return D_WALK_CONTINUE
;
1273 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1274 " still in use (%d) [unmount of %s %s]\n",
1277 dentry
->d_inode
->i_ino
: 0UL,
1279 dentry
->d_lockref
.count
,
1280 dentry
->d_sb
->s_type
->name
,
1281 dentry
->d_sb
->s_id
);
1283 return D_WALK_CONTINUE
;
1286 static void do_one_tree(struct dentry
*dentry
)
1288 shrink_dcache_parent(dentry
);
1289 d_walk(dentry
, dentry
, umount_check
, NULL
);
1295 * destroy the dentries attached to a superblock on unmounting
1297 void shrink_dcache_for_umount(struct super_block
*sb
)
1299 struct dentry
*dentry
;
1301 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1303 dentry
= sb
->s_root
;
1305 do_one_tree(dentry
);
1307 while (!hlist_bl_empty(&sb
->s_anon
)) {
1308 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1309 do_one_tree(dentry
);
1313 struct detach_data
{
1314 struct select_data select
;
1315 struct dentry
*mountpoint
;
1317 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1319 struct detach_data
*data
= _data
;
1321 if (d_mountpoint(dentry
)) {
1322 __dget_dlock(dentry
);
1323 data
->mountpoint
= dentry
;
1327 return select_collect(&data
->select
, dentry
);
1330 static void check_and_drop(void *_data
)
1332 struct detach_data
*data
= _data
;
1334 if (!data
->mountpoint
&& !data
->select
.found
)
1335 __d_drop(data
->select
.start
);
1339 * d_invalidate - detach submounts, prune dcache, and drop
1340 * @dentry: dentry to invalidate (aka detach, prune and drop)
1344 * The final d_drop is done as an atomic operation relative to
1345 * rename_lock ensuring there are no races with d_set_mounted. This
1346 * ensures there are no unhashed dentries on the path to a mountpoint.
1348 void d_invalidate(struct dentry
*dentry
)
1351 * If it's already been dropped, return OK.
1353 spin_lock(&dentry
->d_lock
);
1354 if (d_unhashed(dentry
)) {
1355 spin_unlock(&dentry
->d_lock
);
1358 spin_unlock(&dentry
->d_lock
);
1360 /* Negative dentries can be dropped without further checks */
1361 if (!dentry
->d_inode
) {
1367 struct detach_data data
;
1369 data
.mountpoint
= NULL
;
1370 INIT_LIST_HEAD(&data
.select
.dispose
);
1371 data
.select
.start
= dentry
;
1372 data
.select
.found
= 0;
1374 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1376 if (data
.select
.found
)
1377 shrink_dentry_list(&data
.select
.dispose
);
1379 if (data
.mountpoint
) {
1380 detach_mounts(data
.mountpoint
);
1381 dput(data
.mountpoint
);
1384 if (!data
.mountpoint
&& !data
.select
.found
)
1390 EXPORT_SYMBOL(d_invalidate
);
1393 * __d_alloc - allocate a dcache entry
1394 * @sb: filesystem it will belong to
1395 * @name: qstr of the name
1397 * Allocates a dentry. It returns %NULL if there is insufficient memory
1398 * available. On a success the dentry is returned. The name passed in is
1399 * copied and the copy passed in may be reused after this call.
1402 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1404 struct dentry
*dentry
;
1407 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1412 * We guarantee that the inline name is always NUL-terminated.
1413 * This way the memcpy() done by the name switching in rename
1414 * will still always have a NUL at the end, even if we might
1415 * be overwriting an internal NUL character
1417 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1418 if (name
->len
> DNAME_INLINE_LEN
-1) {
1419 size_t size
= offsetof(struct external_name
, name
[1]);
1420 struct external_name
*p
= kmalloc(size
+ name
->len
, GFP_KERNEL
);
1422 kmem_cache_free(dentry_cache
, dentry
);
1425 atomic_set(&p
->u
.count
, 1);
1428 dname
= dentry
->d_iname
;
1431 dentry
->d_name
.len
= name
->len
;
1432 dentry
->d_name
.hash
= name
->hash
;
1433 memcpy(dname
, name
->name
, name
->len
);
1434 dname
[name
->len
] = 0;
1436 /* Make sure we always see the terminating NUL character */
1438 dentry
->d_name
.name
= dname
;
1440 dentry
->d_lockref
.count
= 1;
1441 dentry
->d_flags
= 0;
1442 spin_lock_init(&dentry
->d_lock
);
1443 seqcount_init(&dentry
->d_seq
);
1444 dentry
->d_inode
= NULL
;
1445 dentry
->d_parent
= dentry
;
1447 dentry
->d_op
= NULL
;
1448 dentry
->d_fsdata
= NULL
;
1449 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1450 INIT_LIST_HEAD(&dentry
->d_lru
);
1451 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1452 INIT_HLIST_NODE(&dentry
->d_alias
);
1453 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1454 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1456 this_cpu_inc(nr_dentry
);
1462 * d_alloc - allocate a dcache entry
1463 * @parent: parent of entry to allocate
1464 * @name: qstr of the name
1466 * Allocates a dentry. It returns %NULL if there is insufficient memory
1467 * available. On a success the dentry is returned. The name passed in is
1468 * copied and the copy passed in may be reused after this call.
1470 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1472 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1476 spin_lock(&parent
->d_lock
);
1478 * don't need child lock because it is not subject
1479 * to concurrency here
1481 __dget_dlock(parent
);
1482 dentry
->d_parent
= parent
;
1483 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1484 spin_unlock(&parent
->d_lock
);
1488 EXPORT_SYMBOL(d_alloc
);
1491 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1492 * @sb: the superblock
1493 * @name: qstr of the name
1495 * For a filesystem that just pins its dentries in memory and never
1496 * performs lookups at all, return an unhashed IS_ROOT dentry.
1498 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1500 return __d_alloc(sb
, name
);
1502 EXPORT_SYMBOL(d_alloc_pseudo
);
1504 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1509 q
.len
= strlen(name
);
1510 q
.hash
= full_name_hash(q
.name
, q
.len
);
1511 return d_alloc(parent
, &q
);
1513 EXPORT_SYMBOL(d_alloc_name
);
1515 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1517 WARN_ON_ONCE(dentry
->d_op
);
1518 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1520 DCACHE_OP_REVALIDATE
|
1521 DCACHE_OP_WEAK_REVALIDATE
|
1522 DCACHE_OP_DELETE
));
1527 dentry
->d_flags
|= DCACHE_OP_HASH
;
1529 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1530 if (op
->d_revalidate
)
1531 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1532 if (op
->d_weak_revalidate
)
1533 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1535 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1537 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1540 EXPORT_SYMBOL(d_set_d_op
);
1542 static unsigned d_flags_for_inode(struct inode
*inode
)
1544 unsigned add_flags
= DCACHE_FILE_TYPE
;
1547 return DCACHE_MISS_TYPE
;
1549 if (S_ISDIR(inode
->i_mode
)) {
1550 add_flags
= DCACHE_DIRECTORY_TYPE
;
1551 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1552 if (unlikely(!inode
->i_op
->lookup
))
1553 add_flags
= DCACHE_AUTODIR_TYPE
;
1555 inode
->i_opflags
|= IOP_LOOKUP
;
1557 } else if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1558 if (unlikely(inode
->i_op
->follow_link
))
1559 add_flags
= DCACHE_SYMLINK_TYPE
;
1561 inode
->i_opflags
|= IOP_NOFOLLOW
;
1564 if (unlikely(IS_AUTOMOUNT(inode
)))
1565 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1569 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1571 unsigned add_flags
= d_flags_for_inode(inode
);
1573 spin_lock(&dentry
->d_lock
);
1574 __d_set_type(dentry
, add_flags
);
1576 hlist_add_head(&dentry
->d_alias
, &inode
->i_dentry
);
1577 dentry
->d_inode
= inode
;
1578 dentry_rcuwalk_barrier(dentry
);
1579 spin_unlock(&dentry
->d_lock
);
1580 fsnotify_d_instantiate(dentry
, inode
);
1584 * d_instantiate - fill in inode information for a dentry
1585 * @entry: dentry to complete
1586 * @inode: inode to attach to this dentry
1588 * Fill in inode information in the entry.
1590 * This turns negative dentries into productive full members
1593 * NOTE! This assumes that the inode count has been incremented
1594 * (or otherwise set) by the caller to indicate that it is now
1595 * in use by the dcache.
1598 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1600 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1602 spin_lock(&inode
->i_lock
);
1603 __d_instantiate(entry
, inode
);
1605 spin_unlock(&inode
->i_lock
);
1606 security_d_instantiate(entry
, inode
);
1608 EXPORT_SYMBOL(d_instantiate
);
1611 * d_instantiate_unique - instantiate a non-aliased dentry
1612 * @entry: dentry to instantiate
1613 * @inode: inode to attach to this dentry
1615 * Fill in inode information in the entry. On success, it returns NULL.
1616 * If an unhashed alias of "entry" already exists, then we return the
1617 * aliased dentry instead and drop one reference to inode.
1619 * Note that in order to avoid conflicts with rename() etc, the caller
1620 * had better be holding the parent directory semaphore.
1622 * This also assumes that the inode count has been incremented
1623 * (or otherwise set) by the caller to indicate that it is now
1624 * in use by the dcache.
1626 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1627 struct inode
*inode
)
1629 struct dentry
*alias
;
1630 int len
= entry
->d_name
.len
;
1631 const char *name
= entry
->d_name
.name
;
1632 unsigned int hash
= entry
->d_name
.hash
;
1635 __d_instantiate(entry
, NULL
);
1639 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1641 * Don't need alias->d_lock here, because aliases with
1642 * d_parent == entry->d_parent are not subject to name or
1643 * parent changes, because the parent inode i_mutex is held.
1645 if (alias
->d_name
.hash
!= hash
)
1647 if (alias
->d_parent
!= entry
->d_parent
)
1649 if (alias
->d_name
.len
!= len
)
1651 if (dentry_cmp(alias
, name
, len
))
1657 __d_instantiate(entry
, inode
);
1661 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1663 struct dentry
*result
;
1665 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1668 spin_lock(&inode
->i_lock
);
1669 result
= __d_instantiate_unique(entry
, inode
);
1671 spin_unlock(&inode
->i_lock
);
1674 security_d_instantiate(entry
, inode
);
1678 BUG_ON(!d_unhashed(result
));
1683 EXPORT_SYMBOL(d_instantiate_unique
);
1686 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1687 * @entry: dentry to complete
1688 * @inode: inode to attach to this dentry
1690 * Fill in inode information in the entry. If a directory alias is found, then
1691 * return an error (and drop inode). Together with d_materialise_unique() this
1692 * guarantees that a directory inode may never have more than one alias.
1694 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1696 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1698 spin_lock(&inode
->i_lock
);
1699 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1700 spin_unlock(&inode
->i_lock
);
1704 __d_instantiate(entry
, inode
);
1705 spin_unlock(&inode
->i_lock
);
1706 security_d_instantiate(entry
, inode
);
1710 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1712 struct dentry
*d_make_root(struct inode
*root_inode
)
1714 struct dentry
*res
= NULL
;
1717 static const struct qstr name
= QSTR_INIT("/", 1);
1719 res
= __d_alloc(root_inode
->i_sb
, &name
);
1721 d_instantiate(res
, root_inode
);
1727 EXPORT_SYMBOL(d_make_root
);
1729 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1731 struct dentry
*alias
;
1733 if (hlist_empty(&inode
->i_dentry
))
1735 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_alias
);
1741 * d_find_any_alias - find any alias for a given inode
1742 * @inode: inode to find an alias for
1744 * If any aliases exist for the given inode, take and return a
1745 * reference for one of them. If no aliases exist, return %NULL.
1747 struct dentry
*d_find_any_alias(struct inode
*inode
)
1751 spin_lock(&inode
->i_lock
);
1752 de
= __d_find_any_alias(inode
);
1753 spin_unlock(&inode
->i_lock
);
1756 EXPORT_SYMBOL(d_find_any_alias
);
1758 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1760 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1766 return ERR_PTR(-ESTALE
);
1768 return ERR_CAST(inode
);
1770 res
= d_find_any_alias(inode
);
1774 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1776 res
= ERR_PTR(-ENOMEM
);
1780 spin_lock(&inode
->i_lock
);
1781 res
= __d_find_any_alias(inode
);
1783 spin_unlock(&inode
->i_lock
);
1788 /* attach a disconnected dentry */
1789 add_flags
= d_flags_for_inode(inode
);
1792 add_flags
|= DCACHE_DISCONNECTED
;
1794 spin_lock(&tmp
->d_lock
);
1795 tmp
->d_inode
= inode
;
1796 tmp
->d_flags
|= add_flags
;
1797 hlist_add_head(&tmp
->d_alias
, &inode
->i_dentry
);
1798 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1799 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1800 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1801 spin_unlock(&tmp
->d_lock
);
1802 spin_unlock(&inode
->i_lock
);
1803 security_d_instantiate(tmp
, inode
);
1808 if (res
&& !IS_ERR(res
))
1809 security_d_instantiate(res
, inode
);
1815 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1816 * @inode: inode to allocate the dentry for
1818 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1819 * similar open by handle operations. The returned dentry may be anonymous,
1820 * or may have a full name (if the inode was already in the cache).
1822 * When called on a directory inode, we must ensure that the inode only ever
1823 * has one dentry. If a dentry is found, that is returned instead of
1824 * allocating a new one.
1826 * On successful return, the reference to the inode has been transferred
1827 * to the dentry. In case of an error the reference on the inode is released.
1828 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1829 * be passed in and the error will be propagated to the return value,
1830 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1832 struct dentry
*d_obtain_alias(struct inode
*inode
)
1834 return __d_obtain_alias(inode
, 1);
1836 EXPORT_SYMBOL(d_obtain_alias
);
1839 * d_obtain_root - find or allocate a dentry for a given inode
1840 * @inode: inode to allocate the dentry for
1842 * Obtain an IS_ROOT dentry for the root of a filesystem.
1844 * We must ensure that directory inodes only ever have one dentry. If a
1845 * dentry is found, that is returned instead of allocating a new one.
1847 * On successful return, the reference to the inode has been transferred
1848 * to the dentry. In case of an error the reference on the inode is
1849 * released. A %NULL or IS_ERR inode may be passed in and will be the
1850 * error will be propagate to the return value, with a %NULL @inode
1851 * replaced by ERR_PTR(-ESTALE).
1853 struct dentry
*d_obtain_root(struct inode
*inode
)
1855 return __d_obtain_alias(inode
, 0);
1857 EXPORT_SYMBOL(d_obtain_root
);
1860 * d_add_ci - lookup or allocate new dentry with case-exact name
1861 * @inode: the inode case-insensitive lookup has found
1862 * @dentry: the negative dentry that was passed to the parent's lookup func
1863 * @name: the case-exact name to be associated with the returned dentry
1865 * This is to avoid filling the dcache with case-insensitive names to the
1866 * same inode, only the actual correct case is stored in the dcache for
1867 * case-insensitive filesystems.
1869 * For a case-insensitive lookup match and if the the case-exact dentry
1870 * already exists in in the dcache, use it and return it.
1872 * If no entry exists with the exact case name, allocate new dentry with
1873 * the exact case, and return the spliced entry.
1875 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1878 struct dentry
*found
;
1882 * First check if a dentry matching the name already exists,
1883 * if not go ahead and create it now.
1885 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1886 if (unlikely(IS_ERR(found
)))
1889 new = d_alloc(dentry
->d_parent
, name
);
1891 found
= ERR_PTR(-ENOMEM
);
1895 found
= d_splice_alias(inode
, new);
1904 * If a matching dentry exists, and it's not negative use it.
1906 * Decrement the reference count to balance the iget() done
1909 if (found
->d_inode
) {
1910 if (unlikely(found
->d_inode
!= inode
)) {
1911 /* This can't happen because bad inodes are unhashed. */
1912 BUG_ON(!is_bad_inode(inode
));
1913 BUG_ON(!is_bad_inode(found
->d_inode
));
1920 * Negative dentry: instantiate it unless the inode is a directory and
1921 * already has a dentry.
1923 new = d_splice_alias(inode
, found
);
1934 EXPORT_SYMBOL(d_add_ci
);
1937 * Do the slow-case of the dentry name compare.
1939 * Unlike the dentry_cmp() function, we need to atomically
1940 * load the name and length information, so that the
1941 * filesystem can rely on them, and can use the 'name' and
1942 * 'len' information without worrying about walking off the
1943 * end of memory etc.
1945 * Thus the read_seqcount_retry() and the "duplicate" info
1946 * in arguments (the low-level filesystem should not look
1947 * at the dentry inode or name contents directly, since
1948 * rename can change them while we're in RCU mode).
1950 enum slow_d_compare
{
1956 static noinline
enum slow_d_compare
slow_dentry_cmp(
1957 const struct dentry
*parent
,
1958 struct dentry
*dentry
,
1960 const struct qstr
*name
)
1962 int tlen
= dentry
->d_name
.len
;
1963 const char *tname
= dentry
->d_name
.name
;
1965 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
1967 return D_COMP_SEQRETRY
;
1969 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
1970 return D_COMP_NOMATCH
;
1975 * __d_lookup_rcu - search for a dentry (racy, store-free)
1976 * @parent: parent dentry
1977 * @name: qstr of name we wish to find
1978 * @seqp: returns d_seq value at the point where the dentry was found
1979 * Returns: dentry, or NULL
1981 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1982 * resolution (store-free path walking) design described in
1983 * Documentation/filesystems/path-lookup.txt.
1985 * This is not to be used outside core vfs.
1987 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1988 * held, and rcu_read_lock held. The returned dentry must not be stored into
1989 * without taking d_lock and checking d_seq sequence count against @seq
1992 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1995 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1996 * the returned dentry, so long as its parent's seqlock is checked after the
1997 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1998 * is formed, giving integrity down the path walk.
2000 * NOTE! The caller *has* to check the resulting dentry against the sequence
2001 * number we've returned before using any of the resulting dentry state!
2003 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2004 const struct qstr
*name
,
2007 u64 hashlen
= name
->hash_len
;
2008 const unsigned char *str
= name
->name
;
2009 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2010 struct hlist_bl_node
*node
;
2011 struct dentry
*dentry
;
2014 * Note: There is significant duplication with __d_lookup_rcu which is
2015 * required to prevent single threaded performance regressions
2016 * especially on architectures where smp_rmb (in seqcounts) are costly.
2017 * Keep the two functions in sync.
2021 * The hash list is protected using RCU.
2023 * Carefully use d_seq when comparing a candidate dentry, to avoid
2024 * races with d_move().
2026 * It is possible that concurrent renames can mess up our list
2027 * walk here and result in missing our dentry, resulting in the
2028 * false-negative result. d_lookup() protects against concurrent
2029 * renames using rename_lock seqlock.
2031 * See Documentation/filesystems/path-lookup.txt for more details.
2033 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2038 * The dentry sequence count protects us from concurrent
2039 * renames, and thus protects parent and name fields.
2041 * The caller must perform a seqcount check in order
2042 * to do anything useful with the returned dentry.
2044 * NOTE! We do a "raw" seqcount_begin here. That means that
2045 * we don't wait for the sequence count to stabilize if it
2046 * is in the middle of a sequence change. If we do the slow
2047 * dentry compare, we will do seqretries until it is stable,
2048 * and if we end up with a successful lookup, we actually
2049 * want to exit RCU lookup anyway.
2051 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2052 if (dentry
->d_parent
!= parent
)
2054 if (d_unhashed(dentry
))
2057 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2058 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2061 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2064 case D_COMP_NOMATCH
:
2071 if (dentry
->d_name
.hash_len
!= hashlen
)
2074 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2081 * d_lookup - search for a dentry
2082 * @parent: parent dentry
2083 * @name: qstr of name we wish to find
2084 * Returns: dentry, or NULL
2086 * d_lookup searches the children of the parent dentry for the name in
2087 * question. If the dentry is found its reference count is incremented and the
2088 * dentry is returned. The caller must use dput to free the entry when it has
2089 * finished using it. %NULL is returned if the dentry does not exist.
2091 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2093 struct dentry
*dentry
;
2097 seq
= read_seqbegin(&rename_lock
);
2098 dentry
= __d_lookup(parent
, name
);
2101 } while (read_seqretry(&rename_lock
, seq
));
2104 EXPORT_SYMBOL(d_lookup
);
2107 * __d_lookup - search for a dentry (racy)
2108 * @parent: parent dentry
2109 * @name: qstr of name we wish to find
2110 * Returns: dentry, or NULL
2112 * __d_lookup is like d_lookup, however it may (rarely) return a
2113 * false-negative result due to unrelated rename activity.
2115 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2116 * however it must be used carefully, eg. with a following d_lookup in
2117 * the case of failure.
2119 * __d_lookup callers must be commented.
2121 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2123 unsigned int len
= name
->len
;
2124 unsigned int hash
= name
->hash
;
2125 const unsigned char *str
= name
->name
;
2126 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2127 struct hlist_bl_node
*node
;
2128 struct dentry
*found
= NULL
;
2129 struct dentry
*dentry
;
2132 * Note: There is significant duplication with __d_lookup_rcu which is
2133 * required to prevent single threaded performance regressions
2134 * especially on architectures where smp_rmb (in seqcounts) are costly.
2135 * Keep the two functions in sync.
2139 * The hash list is protected using RCU.
2141 * Take d_lock when comparing a candidate dentry, to avoid races
2144 * It is possible that concurrent renames can mess up our list
2145 * walk here and result in missing our dentry, resulting in the
2146 * false-negative result. d_lookup() protects against concurrent
2147 * renames using rename_lock seqlock.
2149 * See Documentation/filesystems/path-lookup.txt for more details.
2153 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2155 if (dentry
->d_name
.hash
!= hash
)
2158 spin_lock(&dentry
->d_lock
);
2159 if (dentry
->d_parent
!= parent
)
2161 if (d_unhashed(dentry
))
2165 * It is safe to compare names since d_move() cannot
2166 * change the qstr (protected by d_lock).
2168 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2169 int tlen
= dentry
->d_name
.len
;
2170 const char *tname
= dentry
->d_name
.name
;
2171 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2174 if (dentry
->d_name
.len
!= len
)
2176 if (dentry_cmp(dentry
, str
, len
))
2180 dentry
->d_lockref
.count
++;
2182 spin_unlock(&dentry
->d_lock
);
2185 spin_unlock(&dentry
->d_lock
);
2193 * d_hash_and_lookup - hash the qstr then search for a dentry
2194 * @dir: Directory to search in
2195 * @name: qstr of name we wish to find
2197 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2199 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2202 * Check for a fs-specific hash function. Note that we must
2203 * calculate the standard hash first, as the d_op->d_hash()
2204 * routine may choose to leave the hash value unchanged.
2206 name
->hash
= full_name_hash(name
->name
, name
->len
);
2207 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2208 int err
= dir
->d_op
->d_hash(dir
, name
);
2209 if (unlikely(err
< 0))
2210 return ERR_PTR(err
);
2212 return d_lookup(dir
, name
);
2214 EXPORT_SYMBOL(d_hash_and_lookup
);
2217 * d_validate - verify dentry provided from insecure source (deprecated)
2218 * @dentry: The dentry alleged to be valid child of @dparent
2219 * @dparent: The parent dentry (known to be valid)
2221 * An insecure source has sent us a dentry, here we verify it and dget() it.
2222 * This is used by ncpfs in its readdir implementation.
2223 * Zero is returned in the dentry is invalid.
2225 * This function is slow for big directories, and deprecated, do not use it.
2227 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
2229 struct dentry
*child
;
2231 spin_lock(&dparent
->d_lock
);
2232 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
2233 if (dentry
== child
) {
2234 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2235 __dget_dlock(dentry
);
2236 spin_unlock(&dentry
->d_lock
);
2237 spin_unlock(&dparent
->d_lock
);
2241 spin_unlock(&dparent
->d_lock
);
2245 EXPORT_SYMBOL(d_validate
);
2248 * When a file is deleted, we have two options:
2249 * - turn this dentry into a negative dentry
2250 * - unhash this dentry and free it.
2252 * Usually, we want to just turn this into
2253 * a negative dentry, but if anybody else is
2254 * currently using the dentry or the inode
2255 * we can't do that and we fall back on removing
2256 * it from the hash queues and waiting for
2257 * it to be deleted later when it has no users
2261 * d_delete - delete a dentry
2262 * @dentry: The dentry to delete
2264 * Turn the dentry into a negative dentry if possible, otherwise
2265 * remove it from the hash queues so it can be deleted later
2268 void d_delete(struct dentry
* dentry
)
2270 struct inode
*inode
;
2273 * Are we the only user?
2276 spin_lock(&dentry
->d_lock
);
2277 inode
= dentry
->d_inode
;
2278 isdir
= S_ISDIR(inode
->i_mode
);
2279 if (dentry
->d_lockref
.count
== 1) {
2280 if (!spin_trylock(&inode
->i_lock
)) {
2281 spin_unlock(&dentry
->d_lock
);
2285 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2286 dentry_unlink_inode(dentry
);
2287 fsnotify_nameremove(dentry
, isdir
);
2291 if (!d_unhashed(dentry
))
2294 spin_unlock(&dentry
->d_lock
);
2296 fsnotify_nameremove(dentry
, isdir
);
2298 EXPORT_SYMBOL(d_delete
);
2300 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2302 BUG_ON(!d_unhashed(entry
));
2304 entry
->d_flags
|= DCACHE_RCUACCESS
;
2305 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2309 static void _d_rehash(struct dentry
* entry
)
2311 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2315 * d_rehash - add an entry back to the hash
2316 * @entry: dentry to add to the hash
2318 * Adds a dentry to the hash according to its name.
2321 void d_rehash(struct dentry
* entry
)
2323 spin_lock(&entry
->d_lock
);
2325 spin_unlock(&entry
->d_lock
);
2327 EXPORT_SYMBOL(d_rehash
);
2330 * dentry_update_name_case - update case insensitive dentry with a new name
2331 * @dentry: dentry to be updated
2334 * Update a case insensitive dentry with new case of name.
2336 * dentry must have been returned by d_lookup with name @name. Old and new
2337 * name lengths must match (ie. no d_compare which allows mismatched name
2340 * Parent inode i_mutex must be held over d_lookup and into this call (to
2341 * keep renames and concurrent inserts, and readdir(2) away).
2343 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2345 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2346 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2348 spin_lock(&dentry
->d_lock
);
2349 write_seqcount_begin(&dentry
->d_seq
);
2350 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2351 write_seqcount_end(&dentry
->d_seq
);
2352 spin_unlock(&dentry
->d_lock
);
2354 EXPORT_SYMBOL(dentry_update_name_case
);
2356 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2358 if (unlikely(dname_external(target
))) {
2359 if (unlikely(dname_external(dentry
))) {
2361 * Both external: swap the pointers
2363 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2366 * dentry:internal, target:external. Steal target's
2367 * storage and make target internal.
2369 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2370 dentry
->d_name
.len
+ 1);
2371 dentry
->d_name
.name
= target
->d_name
.name
;
2372 target
->d_name
.name
= target
->d_iname
;
2375 if (unlikely(dname_external(dentry
))) {
2377 * dentry:external, target:internal. Give dentry's
2378 * storage to target and make dentry internal
2380 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2381 target
->d_name
.len
+ 1);
2382 target
->d_name
.name
= dentry
->d_name
.name
;
2383 dentry
->d_name
.name
= dentry
->d_iname
;
2386 * Both are internal.
2389 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2390 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2391 swap(((long *) &dentry
->d_iname
)[i
],
2392 ((long *) &target
->d_iname
)[i
]);
2396 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2399 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2401 struct external_name
*old_name
= NULL
;
2402 if (unlikely(dname_external(dentry
)))
2403 old_name
= external_name(dentry
);
2404 if (unlikely(dname_external(target
))) {
2405 atomic_inc(&external_name(target
)->u
.count
);
2406 dentry
->d_name
= target
->d_name
;
2408 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2409 target
->d_name
.len
+ 1);
2410 dentry
->d_name
.name
= dentry
->d_iname
;
2411 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2413 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2414 kfree_rcu(old_name
, u
.head
);
2417 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2420 * XXXX: do we really need to take target->d_lock?
2422 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2423 spin_lock(&target
->d_parent
->d_lock
);
2425 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2426 spin_lock(&dentry
->d_parent
->d_lock
);
2427 spin_lock_nested(&target
->d_parent
->d_lock
,
2428 DENTRY_D_LOCK_NESTED
);
2430 spin_lock(&target
->d_parent
->d_lock
);
2431 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2432 DENTRY_D_LOCK_NESTED
);
2435 if (target
< dentry
) {
2436 spin_lock_nested(&target
->d_lock
, 2);
2437 spin_lock_nested(&dentry
->d_lock
, 3);
2439 spin_lock_nested(&dentry
->d_lock
, 2);
2440 spin_lock_nested(&target
->d_lock
, 3);
2444 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2446 if (target
->d_parent
!= dentry
->d_parent
)
2447 spin_unlock(&dentry
->d_parent
->d_lock
);
2448 if (target
->d_parent
!= target
)
2449 spin_unlock(&target
->d_parent
->d_lock
);
2450 spin_unlock(&target
->d_lock
);
2451 spin_unlock(&dentry
->d_lock
);
2455 * When switching names, the actual string doesn't strictly have to
2456 * be preserved in the target - because we're dropping the target
2457 * anyway. As such, we can just do a simple memcpy() to copy over
2458 * the new name before we switch, unless we are going to rehash
2459 * it. Note that if we *do* unhash the target, we are not allowed
2460 * to rehash it without giving it a new name/hash key - whether
2461 * we swap or overwrite the names here, resulting name won't match
2462 * the reality in filesystem; it's only there for d_path() purposes.
2463 * Note that all of this is happening under rename_lock, so the
2464 * any hash lookup seeing it in the middle of manipulations will
2465 * be discarded anyway. So we do not care what happens to the hash
2469 * __d_move - move a dentry
2470 * @dentry: entry to move
2471 * @target: new dentry
2472 * @exchange: exchange the two dentries
2474 * Update the dcache to reflect the move of a file name. Negative
2475 * dcache entries should not be moved in this way. Caller must hold
2476 * rename_lock, the i_mutex of the source and target directories,
2477 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2479 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2482 if (!dentry
->d_inode
)
2483 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2485 BUG_ON(d_ancestor(dentry
, target
));
2486 BUG_ON(d_ancestor(target
, dentry
));
2488 dentry_lock_for_move(dentry
, target
);
2490 write_seqcount_begin(&dentry
->d_seq
);
2491 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2493 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2496 * Move the dentry to the target hash queue. Don't bother checking
2497 * for the same hash queue because of how unlikely it is.
2500 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2503 * Unhash the target (d_delete() is not usable here). If exchanging
2504 * the two dentries, then rehash onto the other's hash queue.
2509 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2512 /* Switch the names.. */
2514 swap_names(dentry
, target
);
2516 copy_name(dentry
, target
);
2518 /* ... and switch them in the tree */
2519 if (IS_ROOT(dentry
)) {
2520 /* splicing a tree */
2521 dentry
->d_parent
= target
->d_parent
;
2522 target
->d_parent
= target
;
2523 list_del_init(&target
->d_u
.d_child
);
2524 list_move(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2526 /* swapping two dentries */
2527 swap(dentry
->d_parent
, target
->d_parent
);
2528 list_move(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2529 list_move(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2531 fsnotify_d_move(target
);
2532 fsnotify_d_move(dentry
);
2535 write_seqcount_end(&target
->d_seq
);
2536 write_seqcount_end(&dentry
->d_seq
);
2538 dentry_unlock_for_move(dentry
, target
);
2542 * d_move - move a dentry
2543 * @dentry: entry to move
2544 * @target: new dentry
2546 * Update the dcache to reflect the move of a file name. Negative
2547 * dcache entries should not be moved in this way. See the locking
2548 * requirements for __d_move.
2550 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2552 write_seqlock(&rename_lock
);
2553 __d_move(dentry
, target
, false);
2554 write_sequnlock(&rename_lock
);
2556 EXPORT_SYMBOL(d_move
);
2559 * d_exchange - exchange two dentries
2560 * @dentry1: first dentry
2561 * @dentry2: second dentry
2563 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2565 write_seqlock(&rename_lock
);
2567 WARN_ON(!dentry1
->d_inode
);
2568 WARN_ON(!dentry2
->d_inode
);
2569 WARN_ON(IS_ROOT(dentry1
));
2570 WARN_ON(IS_ROOT(dentry2
));
2572 __d_move(dentry1
, dentry2
, true);
2574 write_sequnlock(&rename_lock
);
2578 * d_ancestor - search for an ancestor
2579 * @p1: ancestor dentry
2582 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2583 * an ancestor of p2, else NULL.
2585 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2589 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2590 if (p
->d_parent
== p1
)
2597 * This helper attempts to cope with remotely renamed directories
2599 * It assumes that the caller is already holding
2600 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2602 * Note: If ever the locking in lock_rename() changes, then please
2603 * remember to update this too...
2605 static struct dentry
*__d_unalias(struct inode
*inode
,
2606 struct dentry
*dentry
, struct dentry
*alias
)
2608 struct mutex
*m1
= NULL
, *m2
= NULL
;
2609 struct dentry
*ret
= ERR_PTR(-EBUSY
);
2611 /* If alias and dentry share a parent, then no extra locks required */
2612 if (alias
->d_parent
== dentry
->d_parent
)
2615 /* See lock_rename() */
2616 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2618 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2619 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2621 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2623 __d_move(alias
, dentry
, false);
2626 spin_unlock(&inode
->i_lock
);
2635 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2636 * @inode: the inode which may have a disconnected dentry
2637 * @dentry: a negative dentry which we want to point to the inode.
2639 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2640 * place of the given dentry and return it, else simply d_add the inode
2641 * to the dentry and return NULL.
2643 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2644 * we should error out: directories can't have multiple aliases.
2646 * This is needed in the lookup routine of any filesystem that is exportable
2647 * (via knfsd) so that we can build dcache paths to directories effectively.
2649 * If a dentry was found and moved, then it is returned. Otherwise NULL
2650 * is returned. This matches the expected return value of ->lookup.
2652 * Cluster filesystems may call this function with a negative, hashed dentry.
2653 * In that case, we know that the inode will be a regular file, and also this
2654 * will only occur during atomic_open. So we need to check for the dentry
2655 * being already hashed only in the final case.
2657 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2659 struct dentry
*new = NULL
;
2662 return ERR_CAST(inode
);
2664 if (inode
&& S_ISDIR(inode
->i_mode
)) {
2665 spin_lock(&inode
->i_lock
);
2666 new = __d_find_any_alias(inode
);
2668 if (!IS_ROOT(new)) {
2669 spin_unlock(&inode
->i_lock
);
2671 return ERR_PTR(-EIO
);
2673 if (d_ancestor(new, dentry
)) {
2674 spin_unlock(&inode
->i_lock
);
2676 return ERR_PTR(-EIO
);
2678 write_seqlock(&rename_lock
);
2679 __d_move(new, dentry
, false);
2680 write_sequnlock(&rename_lock
);
2681 spin_unlock(&inode
->i_lock
);
2682 security_d_instantiate(new, inode
);
2685 /* already taking inode->i_lock, so d_add() by hand */
2686 __d_instantiate(dentry
, inode
);
2687 spin_unlock(&inode
->i_lock
);
2688 security_d_instantiate(dentry
, inode
);
2692 d_instantiate(dentry
, inode
);
2693 if (d_unhashed(dentry
))
2698 EXPORT_SYMBOL(d_splice_alias
);
2701 * d_materialise_unique - introduce an inode into the tree
2702 * @dentry: candidate dentry
2703 * @inode: inode to bind to the dentry, to which aliases may be attached
2705 * Introduces an dentry into the tree, substituting an extant disconnected
2706 * root directory alias in its place if there is one. Caller must hold the
2707 * i_mutex of the parent directory.
2709 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2711 struct dentry
*actual
;
2713 BUG_ON(!d_unhashed(dentry
));
2717 __d_instantiate(dentry
, NULL
);
2722 spin_lock(&inode
->i_lock
);
2724 if (S_ISDIR(inode
->i_mode
)) {
2725 struct dentry
*alias
;
2727 /* Does an aliased dentry already exist? */
2728 alias
= __d_find_alias(inode
);
2731 write_seqlock(&rename_lock
);
2733 if (d_ancestor(alias
, dentry
)) {
2734 /* Check for loops */
2735 actual
= ERR_PTR(-ELOOP
);
2736 spin_unlock(&inode
->i_lock
);
2737 } else if (IS_ROOT(alias
)) {
2738 /* Is this an anonymous mountpoint that we
2739 * could splice into our tree? */
2740 __d_move(alias
, dentry
, false);
2741 write_sequnlock(&rename_lock
);
2744 /* Nope, but we must(!) avoid directory
2745 * aliasing. This drops inode->i_lock */
2746 actual
= __d_unalias(inode
, dentry
, alias
);
2748 write_sequnlock(&rename_lock
);
2749 if (IS_ERR(actual
)) {
2750 if (PTR_ERR(actual
) == -ELOOP
)
2751 pr_warn_ratelimited(
2752 "VFS: Lookup of '%s' in %s %s"
2753 " would have caused loop\n",
2754 dentry
->d_name
.name
,
2755 inode
->i_sb
->s_type
->name
,
2763 /* Add a unique reference */
2764 actual
= __d_instantiate_unique(dentry
, inode
);
2770 spin_unlock(&inode
->i_lock
);
2772 if (actual
== dentry
) {
2773 security_d_instantiate(dentry
, inode
);
2780 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2782 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2786 return -ENAMETOOLONG
;
2788 memcpy(*buffer
, str
, namelen
);
2793 * prepend_name - prepend a pathname in front of current buffer pointer
2794 * @buffer: buffer pointer
2795 * @buflen: allocated length of the buffer
2796 * @name: name string and length qstr structure
2798 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2799 * make sure that either the old or the new name pointer and length are
2800 * fetched. However, there may be mismatch between length and pointer.
2801 * The length cannot be trusted, we need to copy it byte-by-byte until
2802 * the length is reached or a null byte is found. It also prepends "/" at
2803 * the beginning of the name. The sequence number check at the caller will
2804 * retry it again when a d_move() does happen. So any garbage in the buffer
2805 * due to mismatched pointer and length will be discarded.
2807 * Data dependency barrier is needed to make sure that we see that terminating
2808 * NUL. Alpha strikes again, film at 11...
2810 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2812 const char *dname
= ACCESS_ONCE(name
->name
);
2813 u32 dlen
= ACCESS_ONCE(name
->len
);
2816 smp_read_barrier_depends();
2818 *buflen
-= dlen
+ 1;
2820 return -ENAMETOOLONG
;
2821 p
= *buffer
-= dlen
+ 1;
2833 * prepend_path - Prepend path string to a buffer
2834 * @path: the dentry/vfsmount to report
2835 * @root: root vfsmnt/dentry
2836 * @buffer: pointer to the end of the buffer
2837 * @buflen: pointer to buffer length
2839 * The function will first try to write out the pathname without taking any
2840 * lock other than the RCU read lock to make sure that dentries won't go away.
2841 * It only checks the sequence number of the global rename_lock as any change
2842 * in the dentry's d_seq will be preceded by changes in the rename_lock
2843 * sequence number. If the sequence number had been changed, it will restart
2844 * the whole pathname back-tracing sequence again by taking the rename_lock.
2845 * In this case, there is no need to take the RCU read lock as the recursive
2846 * parent pointer references will keep the dentry chain alive as long as no
2847 * rename operation is performed.
2849 static int prepend_path(const struct path
*path
,
2850 const struct path
*root
,
2851 char **buffer
, int *buflen
)
2853 struct dentry
*dentry
;
2854 struct vfsmount
*vfsmnt
;
2857 unsigned seq
, m_seq
= 0;
2863 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
2870 dentry
= path
->dentry
;
2872 mnt
= real_mount(vfsmnt
);
2873 read_seqbegin_or_lock(&rename_lock
, &seq
);
2874 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2875 struct dentry
* parent
;
2877 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2878 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
2880 if (mnt
!= parent
) {
2881 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
2887 * Filesystems needing to implement special "root names"
2888 * should do so with ->d_dname()
2890 if (IS_ROOT(dentry
) &&
2891 (dentry
->d_name
.len
!= 1 ||
2892 dentry
->d_name
.name
[0] != '/')) {
2893 WARN(1, "Root dentry has weird name <%.*s>\n",
2894 (int) dentry
->d_name
.len
,
2895 dentry
->d_name
.name
);
2898 error
= is_mounted(vfsmnt
) ? 1 : 2;
2901 parent
= dentry
->d_parent
;
2903 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
2911 if (need_seqretry(&rename_lock
, seq
)) {
2915 done_seqretry(&rename_lock
, seq
);
2919 if (need_seqretry(&mount_lock
, m_seq
)) {
2923 done_seqretry(&mount_lock
, m_seq
);
2925 if (error
>= 0 && bptr
== *buffer
) {
2927 error
= -ENAMETOOLONG
;
2937 * __d_path - return the path of a dentry
2938 * @path: the dentry/vfsmount to report
2939 * @root: root vfsmnt/dentry
2940 * @buf: buffer to return value in
2941 * @buflen: buffer length
2943 * Convert a dentry into an ASCII path name.
2945 * Returns a pointer into the buffer or an error code if the
2946 * path was too long.
2948 * "buflen" should be positive.
2950 * If the path is not reachable from the supplied root, return %NULL.
2952 char *__d_path(const struct path
*path
,
2953 const struct path
*root
,
2954 char *buf
, int buflen
)
2956 char *res
= buf
+ buflen
;
2959 prepend(&res
, &buflen
, "\0", 1);
2960 error
= prepend_path(path
, root
, &res
, &buflen
);
2963 return ERR_PTR(error
);
2969 char *d_absolute_path(const struct path
*path
,
2970 char *buf
, int buflen
)
2972 struct path root
= {};
2973 char *res
= buf
+ buflen
;
2976 prepend(&res
, &buflen
, "\0", 1);
2977 error
= prepend_path(path
, &root
, &res
, &buflen
);
2982 return ERR_PTR(error
);
2987 * same as __d_path but appends "(deleted)" for unlinked files.
2989 static int path_with_deleted(const struct path
*path
,
2990 const struct path
*root
,
2991 char **buf
, int *buflen
)
2993 prepend(buf
, buflen
, "\0", 1);
2994 if (d_unlinked(path
->dentry
)) {
2995 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3000 return prepend_path(path
, root
, buf
, buflen
);
3003 static int prepend_unreachable(char **buffer
, int *buflen
)
3005 return prepend(buffer
, buflen
, "(unreachable)", 13);
3008 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3013 seq
= read_seqcount_begin(&fs
->seq
);
3015 } while (read_seqcount_retry(&fs
->seq
, seq
));
3019 * d_path - return the path of a dentry
3020 * @path: path to report
3021 * @buf: buffer to return value in
3022 * @buflen: buffer length
3024 * Convert a dentry into an ASCII path name. If the entry has been deleted
3025 * the string " (deleted)" is appended. Note that this is ambiguous.
3027 * Returns a pointer into the buffer or an error code if the path was
3028 * too long. Note: Callers should use the returned pointer, not the passed
3029 * in buffer, to use the name! The implementation often starts at an offset
3030 * into the buffer, and may leave 0 bytes at the start.
3032 * "buflen" should be positive.
3034 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3036 char *res
= buf
+ buflen
;
3041 * We have various synthetic filesystems that never get mounted. On
3042 * these filesystems dentries are never used for lookup purposes, and
3043 * thus don't need to be hashed. They also don't need a name until a
3044 * user wants to identify the object in /proc/pid/fd/. The little hack
3045 * below allows us to generate a name for these objects on demand:
3047 * Some pseudo inodes are mountable. When they are mounted
3048 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3049 * and instead have d_path return the mounted path.
3051 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3052 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3053 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3056 get_fs_root_rcu(current
->fs
, &root
);
3057 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3061 res
= ERR_PTR(error
);
3064 EXPORT_SYMBOL(d_path
);
3067 * Helper function for dentry_operations.d_dname() members
3069 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3070 const char *fmt
, ...)
3076 va_start(args
, fmt
);
3077 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3080 if (sz
> sizeof(temp
) || sz
> buflen
)
3081 return ERR_PTR(-ENAMETOOLONG
);
3083 buffer
+= buflen
- sz
;
3084 return memcpy(buffer
, temp
, sz
);
3087 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3089 char *end
= buffer
+ buflen
;
3090 /* these dentries are never renamed, so d_lock is not needed */
3091 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3092 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3093 prepend(&end
, &buflen
, "/", 1))
3094 end
= ERR_PTR(-ENAMETOOLONG
);
3097 EXPORT_SYMBOL(simple_dname
);
3100 * Write full pathname from the root of the filesystem into the buffer.
3102 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3104 struct dentry
*dentry
;
3117 prepend(&end
, &len
, "\0", 1);
3121 read_seqbegin_or_lock(&rename_lock
, &seq
);
3122 while (!IS_ROOT(dentry
)) {
3123 struct dentry
*parent
= dentry
->d_parent
;
3126 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3135 if (need_seqretry(&rename_lock
, seq
)) {
3139 done_seqretry(&rename_lock
, seq
);
3144 return ERR_PTR(-ENAMETOOLONG
);
3147 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3149 return __dentry_path(dentry
, buf
, buflen
);
3151 EXPORT_SYMBOL(dentry_path_raw
);
3153 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3158 if (d_unlinked(dentry
)) {
3160 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3164 retval
= __dentry_path(dentry
, buf
, buflen
);
3165 if (!IS_ERR(retval
) && p
)
3166 *p
= '/'; /* restore '/' overriden with '\0' */
3169 return ERR_PTR(-ENAMETOOLONG
);
3172 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3178 seq
= read_seqcount_begin(&fs
->seq
);
3181 } while (read_seqcount_retry(&fs
->seq
, seq
));
3185 * NOTE! The user-level library version returns a
3186 * character pointer. The kernel system call just
3187 * returns the length of the buffer filled (which
3188 * includes the ending '\0' character), or a negative
3189 * error value. So libc would do something like
3191 * char *getcwd(char * buf, size_t size)
3195 * retval = sys_getcwd(buf, size);
3202 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3205 struct path pwd
, root
;
3206 char *page
= __getname();
3212 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3215 if (!d_unlinked(pwd
.dentry
)) {
3217 char *cwd
= page
+ PATH_MAX
;
3218 int buflen
= PATH_MAX
;
3220 prepend(&cwd
, &buflen
, "\0", 1);
3221 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3227 /* Unreachable from current root */
3229 error
= prepend_unreachable(&cwd
, &buflen
);
3235 len
= PATH_MAX
+ page
- cwd
;
3238 if (copy_to_user(buf
, cwd
, len
))
3251 * Test whether new_dentry is a subdirectory of old_dentry.
3253 * Trivially implemented using the dcache structure
3257 * is_subdir - is new dentry a subdirectory of old_dentry
3258 * @new_dentry: new dentry
3259 * @old_dentry: old dentry
3261 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3262 * Returns 0 otherwise.
3263 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3266 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3271 if (new_dentry
== old_dentry
)
3275 /* for restarting inner loop in case of seq retry */
3276 seq
= read_seqbegin(&rename_lock
);
3278 * Need rcu_readlock to protect against the d_parent trashing
3282 if (d_ancestor(old_dentry
, new_dentry
))
3287 } while (read_seqretry(&rename_lock
, seq
));
3292 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3294 struct dentry
*root
= data
;
3295 if (dentry
!= root
) {
3296 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3299 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3300 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3301 dentry
->d_lockref
.count
--;
3304 return D_WALK_CONTINUE
;
3307 void d_genocide(struct dentry
*parent
)
3309 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3312 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3314 inode_dec_link_count(inode
);
3315 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3316 !hlist_unhashed(&dentry
->d_alias
) ||
3317 !d_unlinked(dentry
));
3318 spin_lock(&dentry
->d_parent
->d_lock
);
3319 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3320 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3321 (unsigned long long)inode
->i_ino
);
3322 spin_unlock(&dentry
->d_lock
);
3323 spin_unlock(&dentry
->d_parent
->d_lock
);
3324 d_instantiate(dentry
, inode
);
3326 EXPORT_SYMBOL(d_tmpfile
);
3328 static __initdata
unsigned long dhash_entries
;
3329 static int __init
set_dhash_entries(char *str
)
3333 dhash_entries
= simple_strtoul(str
, &str
, 0);
3336 __setup("dhash_entries=", set_dhash_entries
);
3338 static void __init
dcache_init_early(void)
3342 /* If hashes are distributed across NUMA nodes, defer
3343 * hash allocation until vmalloc space is available.
3349 alloc_large_system_hash("Dentry cache",
3350 sizeof(struct hlist_bl_head
),
3359 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3360 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3363 static void __init
dcache_init(void)
3368 * A constructor could be added for stable state like the lists,
3369 * but it is probably not worth it because of the cache nature
3372 dentry_cache
= KMEM_CACHE(dentry
,
3373 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3375 /* Hash may have been set up in dcache_init_early */
3380 alloc_large_system_hash("Dentry cache",
3381 sizeof(struct hlist_bl_head
),
3390 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3391 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3394 /* SLAB cache for __getname() consumers */
3395 struct kmem_cache
*names_cachep __read_mostly
;
3396 EXPORT_SYMBOL(names_cachep
);
3398 EXPORT_SYMBOL(d_genocide
);
3400 void __init
vfs_caches_init_early(void)
3402 dcache_init_early();
3406 void __init
vfs_caches_init(unsigned long mempages
)
3408 unsigned long reserve
;
3410 /* Base hash sizes on available memory, with a reserve equal to
3411 150% of current kernel size */
3413 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3414 mempages
-= reserve
;
3416 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3417 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3421 files_init(mempages
);