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>
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
64 * dentry->d_inode->i_lock
67 * dcache_hash_bucket lock
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
73 * dentry->d_parent->d_lock
76 * If no ancestor relationship:
77 * if (dentry1 < dentry2)
81 int sysctl_vfs_cache_pressure __read_mostly
= 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
84 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
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.
99 #define D_HASHBITS d_hash_shift
100 #define D_HASHMASK d_hash_mask
102 static unsigned int d_hash_mask __read_mostly
;
103 static unsigned int d_hash_shift __read_mostly
;
105 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
107 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
110 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
111 hash
= hash
+ (hash
>> D_HASHBITS
);
112 return dentry_hashtable
+ (hash
& D_HASHMASK
);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat
= {
120 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 static int get_nr_dentry(void)
127 for_each_possible_cpu(i
)
128 sum
+= per_cpu(nr_dentry
, i
);
129 return sum
< 0 ? 0 : sum
;
132 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
133 size_t *lenp
, loff_t
*ppos
)
135 dentry_stat
.nr_dentry
= get_nr_dentry();
136 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
144 #ifdef CONFIG_DCACHE_WORD_ACCESS
146 #include <asm/word-at-a-time.h>
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
156 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
158 unsigned long a
,b
,mask
;
161 a
= *(unsigned long *)cs
;
162 b
= load_unaligned_zeropad(ct
);
163 if (tcount
< sizeof(unsigned long))
165 if (unlikely(a
!= b
))
167 cs
+= sizeof(unsigned long);
168 ct
+= sizeof(unsigned long);
169 tcount
-= sizeof(unsigned long);
173 mask
= ~(~0ul << tcount
*8);
174 return unlikely(!!((a
^ b
) & mask
));
179 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
193 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
195 const unsigned char *cs
;
197 * Be careful about RCU walk racing with rename:
198 * use ACCESS_ONCE to fetch the name pointer.
200 * NOTE! Even if a rename will mean that the length
201 * was not loaded atomically, we don't care. The
202 * RCU walk will check the sequence count eventually,
203 * and catch it. And we won't overrun the buffer,
204 * because we're reading the name pointer atomically,
205 * and a dentry name is guaranteed to be properly
206 * terminated with a NUL byte.
208 * End result: even if 'len' is wrong, we'll exit
209 * early because the data cannot match (there can
210 * be no NUL in the ct/tcount data)
212 cs
= ACCESS_ONCE(dentry
->d_name
.name
);
213 smp_read_barrier_depends();
214 return dentry_string_cmp(cs
, ct
, tcount
);
217 static void __d_free(struct rcu_head
*head
)
219 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
221 WARN_ON(!hlist_unhashed(&dentry
->d_alias
));
222 if (dname_external(dentry
))
223 kfree(dentry
->d_name
.name
);
224 kmem_cache_free(dentry_cache
, dentry
);
230 static void d_free(struct dentry
*dentry
)
232 BUG_ON(dentry
->d_lockref
.count
);
233 this_cpu_dec(nr_dentry
);
234 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
235 dentry
->d_op
->d_release(dentry
);
237 /* if dentry was never visible to RCU, immediate free is OK */
238 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
239 __d_free(&dentry
->d_u
.d_rcu
);
241 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
245 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
246 * @dentry: the target dentry
247 * After this call, in-progress rcu-walk path lookup will fail. This
248 * should be called after unhashing, and after changing d_inode (if
249 * the dentry has not already been unhashed).
251 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
253 assert_spin_locked(&dentry
->d_lock
);
254 /* Go through a barrier */
255 write_seqcount_barrier(&dentry
->d_seq
);
259 * Release the dentry's inode, using the filesystem
260 * d_iput() operation if defined. Dentry has no refcount
263 static void dentry_iput(struct dentry
* dentry
)
264 __releases(dentry
->d_lock
)
265 __releases(dentry
->d_inode
->i_lock
)
267 struct inode
*inode
= dentry
->d_inode
;
269 dentry
->d_inode
= NULL
;
270 hlist_del_init(&dentry
->d_alias
);
271 spin_unlock(&dentry
->d_lock
);
272 spin_unlock(&inode
->i_lock
);
274 fsnotify_inoderemove(inode
);
275 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
276 dentry
->d_op
->d_iput(dentry
, inode
);
280 spin_unlock(&dentry
->d_lock
);
285 * Release the dentry's inode, using the filesystem
286 * d_iput() operation if defined. dentry remains in-use.
288 static void dentry_unlink_inode(struct dentry
* dentry
)
289 __releases(dentry
->d_lock
)
290 __releases(dentry
->d_inode
->i_lock
)
292 struct inode
*inode
= dentry
->d_inode
;
293 dentry
->d_inode
= NULL
;
294 hlist_del_init(&dentry
->d_alias
);
295 dentry_rcuwalk_barrier(dentry
);
296 spin_unlock(&dentry
->d_lock
);
297 spin_unlock(&inode
->i_lock
);
299 fsnotify_inoderemove(inode
);
300 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
301 dentry
->d_op
->d_iput(dentry
, inode
);
307 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
309 static void dentry_lru_add(struct dentry
*dentry
)
311 if (list_empty(&dentry
->d_lru
)) {
312 spin_lock(&dcache_lru_lock
);
313 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
314 dentry
->d_sb
->s_nr_dentry_unused
++;
315 dentry_stat
.nr_unused
++;
316 spin_unlock(&dcache_lru_lock
);
320 static void __dentry_lru_del(struct dentry
*dentry
)
322 list_del_init(&dentry
->d_lru
);
323 dentry
->d_flags
&= ~DCACHE_SHRINK_LIST
;
324 dentry
->d_sb
->s_nr_dentry_unused
--;
325 dentry_stat
.nr_unused
--;
329 * Remove a dentry with references from the LRU.
331 static void dentry_lru_del(struct dentry
*dentry
)
333 if (!list_empty(&dentry
->d_lru
)) {
334 spin_lock(&dcache_lru_lock
);
335 __dentry_lru_del(dentry
);
336 spin_unlock(&dcache_lru_lock
);
340 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
342 spin_lock(&dcache_lru_lock
);
343 if (list_empty(&dentry
->d_lru
)) {
344 list_add_tail(&dentry
->d_lru
, list
);
345 dentry
->d_sb
->s_nr_dentry_unused
++;
346 dentry_stat
.nr_unused
++;
348 list_move_tail(&dentry
->d_lru
, list
);
350 spin_unlock(&dcache_lru_lock
);
354 * d_kill - kill dentry and return parent
355 * @dentry: dentry to kill
356 * @parent: parent dentry
358 * The dentry must already be unhashed and removed from the LRU.
360 * If this is the root of the dentry tree, return NULL.
362 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
365 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
366 __releases(dentry
->d_lock
)
367 __releases(parent
->d_lock
)
368 __releases(dentry
->d_inode
->i_lock
)
370 list_del(&dentry
->d_u
.d_child
);
372 * Inform try_to_ascend() that we are no longer attached to the
375 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
377 spin_unlock(&parent
->d_lock
);
380 * dentry_iput drops the locks, at which point nobody (except
381 * transient RCU lookups) can reach this dentry.
388 * Unhash a dentry without inserting an RCU walk barrier or checking that
389 * dentry->d_lock is locked. The caller must take care of that, if
392 static void __d_shrink(struct dentry
*dentry
)
394 if (!d_unhashed(dentry
)) {
395 struct hlist_bl_head
*b
;
396 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
397 b
= &dentry
->d_sb
->s_anon
;
399 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
402 __hlist_bl_del(&dentry
->d_hash
);
403 dentry
->d_hash
.pprev
= NULL
;
409 * d_drop - drop a dentry
410 * @dentry: dentry to drop
412 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
413 * be found through a VFS lookup any more. Note that this is different from
414 * deleting the dentry - d_delete will try to mark the dentry negative if
415 * possible, giving a successful _negative_ lookup, while d_drop will
416 * just make the cache lookup fail.
418 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
419 * reason (NFS timeouts or autofs deletes).
421 * __d_drop requires dentry->d_lock.
423 void __d_drop(struct dentry
*dentry
)
425 if (!d_unhashed(dentry
)) {
427 dentry_rcuwalk_barrier(dentry
);
430 EXPORT_SYMBOL(__d_drop
);
432 void d_drop(struct dentry
*dentry
)
434 spin_lock(&dentry
->d_lock
);
436 spin_unlock(&dentry
->d_lock
);
438 EXPORT_SYMBOL(d_drop
);
441 * Finish off a dentry we've decided to kill.
442 * dentry->d_lock must be held, returns with it unlocked.
443 * If ref is non-zero, then decrement the refcount too.
444 * Returns dentry requiring refcount drop, or NULL if we're done.
446 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
447 __releases(dentry
->d_lock
)
450 struct dentry
*parent
;
452 inode
= dentry
->d_inode
;
453 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
455 spin_unlock(&dentry
->d_lock
);
457 return dentry
; /* try again with same dentry */
462 parent
= dentry
->d_parent
;
463 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
465 spin_unlock(&inode
->i_lock
);
470 dentry
->d_lockref
.count
--;
472 * inform the fs via d_prune that this dentry is about to be
473 * unhashed and destroyed.
475 if ((dentry
->d_flags
& DCACHE_OP_PRUNE
) && !d_unhashed(dentry
))
476 dentry
->d_op
->d_prune(dentry
);
478 dentry_lru_del(dentry
);
479 /* if it was on the hash then remove it */
481 return d_kill(dentry
, parent
);
487 * This is complicated by the fact that we do not want to put
488 * dentries that are no longer on any hash chain on the unused
489 * list: we'd much rather just get rid of them immediately.
491 * However, that implies that we have to traverse the dentry
492 * tree upwards to the parents which might _also_ now be
493 * scheduled for deletion (it may have been only waiting for
494 * its last child to go away).
496 * This tail recursion is done by hand as we don't want to depend
497 * on the compiler to always get this right (gcc generally doesn't).
498 * Real recursion would eat up our stack space.
502 * dput - release a dentry
503 * @dentry: dentry to release
505 * Release a dentry. This will drop the usage count and if appropriate
506 * call the dentry unlink method as well as removing it from the queues and
507 * releasing its resources. If the parent dentries were scheduled for release
508 * they too may now get deleted.
510 void dput(struct dentry
*dentry
)
516 if (dentry
->d_lockref
.count
== 1)
518 if (lockref_put_or_lock(&dentry
->d_lockref
))
521 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
522 if (dentry
->d_op
->d_delete(dentry
))
526 /* Unreachable? Get rid of it */
527 if (d_unhashed(dentry
))
530 dentry
->d_flags
|= DCACHE_REFERENCED
;
531 dentry_lru_add(dentry
);
533 dentry
->d_lockref
.count
--;
534 spin_unlock(&dentry
->d_lock
);
538 dentry
= dentry_kill(dentry
, 1);
545 * d_invalidate - invalidate a dentry
546 * @dentry: dentry to invalidate
548 * Try to invalidate the dentry if it turns out to be
549 * possible. If there are other dentries that can be
550 * reached through this one we can't delete it and we
551 * return -EBUSY. On success we return 0.
556 int d_invalidate(struct dentry
* dentry
)
559 * If it's already been dropped, return OK.
561 spin_lock(&dentry
->d_lock
);
562 if (d_unhashed(dentry
)) {
563 spin_unlock(&dentry
->d_lock
);
567 * Check whether to do a partial shrink_dcache
568 * to get rid of unused child entries.
570 if (!list_empty(&dentry
->d_subdirs
)) {
571 spin_unlock(&dentry
->d_lock
);
572 shrink_dcache_parent(dentry
);
573 spin_lock(&dentry
->d_lock
);
577 * Somebody else still using it?
579 * If it's a directory, we can't drop it
580 * for fear of somebody re-populating it
581 * with children (even though dropping it
582 * would make it unreachable from the root,
583 * we might still populate it if it was a
584 * working directory or similar).
585 * We also need to leave mountpoints alone,
588 if (dentry
->d_lockref
.count
> 1 && dentry
->d_inode
) {
589 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
590 spin_unlock(&dentry
->d_lock
);
596 spin_unlock(&dentry
->d_lock
);
599 EXPORT_SYMBOL(d_invalidate
);
601 /* This must be called with d_lock held */
602 static inline void __dget_dlock(struct dentry
*dentry
)
604 dentry
->d_lockref
.count
++;
607 static inline void __dget(struct dentry
*dentry
)
609 lockref_get(&dentry
->d_lockref
);
612 struct dentry
*dget_parent(struct dentry
*dentry
)
618 * Do optimistic parent lookup without any
622 ret
= ACCESS_ONCE(dentry
->d_parent
);
623 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
625 if (likely(gotref
)) {
626 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
633 * Don't need rcu_dereference because we re-check it was correct under
637 ret
= dentry
->d_parent
;
638 spin_lock(&ret
->d_lock
);
639 if (unlikely(ret
!= dentry
->d_parent
)) {
640 spin_unlock(&ret
->d_lock
);
645 BUG_ON(!ret
->d_lockref
.count
);
646 ret
->d_lockref
.count
++;
647 spin_unlock(&ret
->d_lock
);
650 EXPORT_SYMBOL(dget_parent
);
653 * d_find_alias - grab a hashed alias of inode
654 * @inode: inode in question
655 * @want_discon: flag, used by d_splice_alias, to request
656 * that only a DISCONNECTED alias be returned.
658 * If inode has a hashed alias, or is a directory and has any alias,
659 * acquire the reference to alias and return it. Otherwise return NULL.
660 * Notice that if inode is a directory there can be only one alias and
661 * it can be unhashed only if it has no children, or if it is the root
664 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
665 * any other hashed alias over that one unless @want_discon is set,
666 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
668 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
670 struct dentry
*alias
, *discon_alias
;
674 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
675 spin_lock(&alias
->d_lock
);
676 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
677 if (IS_ROOT(alias
) &&
678 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
679 discon_alias
= alias
;
680 } else if (!want_discon
) {
682 spin_unlock(&alias
->d_lock
);
686 spin_unlock(&alias
->d_lock
);
689 alias
= discon_alias
;
690 spin_lock(&alias
->d_lock
);
691 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
692 if (IS_ROOT(alias
) &&
693 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
695 spin_unlock(&alias
->d_lock
);
699 spin_unlock(&alias
->d_lock
);
705 struct dentry
*d_find_alias(struct inode
*inode
)
707 struct dentry
*de
= NULL
;
709 if (!hlist_empty(&inode
->i_dentry
)) {
710 spin_lock(&inode
->i_lock
);
711 de
= __d_find_alias(inode
, 0);
712 spin_unlock(&inode
->i_lock
);
716 EXPORT_SYMBOL(d_find_alias
);
719 * Try to kill dentries associated with this inode.
720 * WARNING: you must own a reference to inode.
722 void d_prune_aliases(struct inode
*inode
)
724 struct dentry
*dentry
;
726 spin_lock(&inode
->i_lock
);
727 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
728 spin_lock(&dentry
->d_lock
);
729 if (!dentry
->d_lockref
.count
) {
731 * inform the fs via d_prune that this dentry
732 * is about to be unhashed and destroyed.
734 if ((dentry
->d_flags
& DCACHE_OP_PRUNE
) &&
736 dentry
->d_op
->d_prune(dentry
);
738 __dget_dlock(dentry
);
740 spin_unlock(&dentry
->d_lock
);
741 spin_unlock(&inode
->i_lock
);
745 spin_unlock(&dentry
->d_lock
);
747 spin_unlock(&inode
->i_lock
);
749 EXPORT_SYMBOL(d_prune_aliases
);
752 * Try to throw away a dentry - free the inode, dput the parent.
753 * Requires dentry->d_lock is held, and dentry->d_count == 0.
754 * Releases dentry->d_lock.
756 * This may fail if locks cannot be acquired no problem, just try again.
758 static void try_prune_one_dentry(struct dentry
*dentry
)
759 __releases(dentry
->d_lock
)
761 struct dentry
*parent
;
763 parent
= dentry_kill(dentry
, 0);
765 * If dentry_kill returns NULL, we have nothing more to do.
766 * if it returns the same dentry, trylocks failed. In either
767 * case, just loop again.
769 * Otherwise, we need to prune ancestors too. This is necessary
770 * to prevent quadratic behavior of shrink_dcache_parent(), but
771 * is also expected to be beneficial in reducing dentry cache
776 if (parent
== dentry
)
779 /* Prune ancestors. */
782 if (lockref_put_or_lock(&dentry
->d_lockref
))
784 dentry
= dentry_kill(dentry
, 1);
788 static void shrink_dentry_list(struct list_head
*list
)
790 struct dentry
*dentry
;
794 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
795 if (&dentry
->d_lru
== list
)
797 spin_lock(&dentry
->d_lock
);
798 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
799 spin_unlock(&dentry
->d_lock
);
804 * We found an inuse dentry which was not removed from
805 * the LRU because of laziness during lookup. Do not free
806 * it - just keep it off the LRU list.
808 if (dentry
->d_lockref
.count
) {
809 dentry_lru_del(dentry
);
810 spin_unlock(&dentry
->d_lock
);
816 try_prune_one_dentry(dentry
);
824 * prune_dcache_sb - shrink the dcache
826 * @count: number of entries to try to free
828 * Attempt to shrink the superblock dcache LRU by @count entries. This is
829 * done when we need more memory an called from the superblock shrinker
832 * This function may fail to free any resources if all the dentries are in
835 void prune_dcache_sb(struct super_block
*sb
, int count
)
837 struct dentry
*dentry
;
838 LIST_HEAD(referenced
);
842 spin_lock(&dcache_lru_lock
);
843 while (!list_empty(&sb
->s_dentry_lru
)) {
844 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
845 struct dentry
, d_lru
);
846 BUG_ON(dentry
->d_sb
!= sb
);
848 if (!spin_trylock(&dentry
->d_lock
)) {
849 spin_unlock(&dcache_lru_lock
);
854 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
855 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
856 list_move(&dentry
->d_lru
, &referenced
);
857 spin_unlock(&dentry
->d_lock
);
859 list_move_tail(&dentry
->d_lru
, &tmp
);
860 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
861 spin_unlock(&dentry
->d_lock
);
865 cond_resched_lock(&dcache_lru_lock
);
867 if (!list_empty(&referenced
))
868 list_splice(&referenced
, &sb
->s_dentry_lru
);
869 spin_unlock(&dcache_lru_lock
);
871 shrink_dentry_list(&tmp
);
875 * shrink_dcache_sb - shrink dcache for a superblock
878 * Shrink the dcache for the specified super block. This is used to free
879 * the dcache before unmounting a file system.
881 void shrink_dcache_sb(struct super_block
*sb
)
885 spin_lock(&dcache_lru_lock
);
886 while (!list_empty(&sb
->s_dentry_lru
)) {
887 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
888 spin_unlock(&dcache_lru_lock
);
889 shrink_dentry_list(&tmp
);
890 spin_lock(&dcache_lru_lock
);
892 spin_unlock(&dcache_lru_lock
);
894 EXPORT_SYMBOL(shrink_dcache_sb
);
897 * destroy a single subtree of dentries for unmount
898 * - see the comments on shrink_dcache_for_umount() for a description of the
901 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
903 struct dentry
*parent
;
905 BUG_ON(!IS_ROOT(dentry
));
908 /* descend to the first leaf in the current subtree */
909 while (!list_empty(&dentry
->d_subdirs
))
910 dentry
= list_entry(dentry
->d_subdirs
.next
,
911 struct dentry
, d_u
.d_child
);
913 /* consume the dentries from this leaf up through its parents
914 * until we find one with children or run out altogether */
919 * inform the fs that this dentry is about to be
920 * unhashed and destroyed.
922 if ((dentry
->d_flags
& DCACHE_OP_PRUNE
) &&
924 dentry
->d_op
->d_prune(dentry
);
926 dentry_lru_del(dentry
);
929 if (dentry
->d_lockref
.count
!= 0) {
931 "BUG: Dentry %p{i=%lx,n=%s}"
933 " [unmount of %s %s]\n",
936 dentry
->d_inode
->i_ino
: 0UL,
938 dentry
->d_lockref
.count
,
939 dentry
->d_sb
->s_type
->name
,
944 if (IS_ROOT(dentry
)) {
946 list_del(&dentry
->d_u
.d_child
);
948 parent
= dentry
->d_parent
;
949 parent
->d_lockref
.count
--;
950 list_del(&dentry
->d_u
.d_child
);
953 inode
= dentry
->d_inode
;
955 dentry
->d_inode
= NULL
;
956 hlist_del_init(&dentry
->d_alias
);
957 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
958 dentry
->d_op
->d_iput(dentry
, inode
);
965 /* finished when we fall off the top of the tree,
966 * otherwise we ascend to the parent and move to the
967 * next sibling if there is one */
971 } while (list_empty(&dentry
->d_subdirs
));
973 dentry
= list_entry(dentry
->d_subdirs
.next
,
974 struct dentry
, d_u
.d_child
);
979 * destroy the dentries attached to a superblock on unmounting
980 * - we don't need to use dentry->d_lock because:
981 * - the superblock is detached from all mountings and open files, so the
982 * dentry trees will not be rearranged by the VFS
983 * - s_umount is write-locked, so the memory pressure shrinker will ignore
984 * any dentries belonging to this superblock that it comes across
985 * - the filesystem itself is no longer permitted to rearrange the dentries
988 void shrink_dcache_for_umount(struct super_block
*sb
)
990 struct dentry
*dentry
;
992 if (down_read_trylock(&sb
->s_umount
))
997 dentry
->d_lockref
.count
--;
998 shrink_dcache_for_umount_subtree(dentry
);
1000 while (!hlist_bl_empty(&sb
->s_anon
)) {
1001 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
1002 shrink_dcache_for_umount_subtree(dentry
);
1007 * This tries to ascend one level of parenthood, but
1008 * we can race with renaming, so we need to re-check
1009 * the parenthood after dropping the lock and check
1010 * that the sequence number still matches.
1012 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
1014 struct dentry
*new = old
->d_parent
;
1017 spin_unlock(&old
->d_lock
);
1018 spin_lock(&new->d_lock
);
1021 * might go back up the wrong parent if we have had a rename
1024 if (new != old
->d_parent
||
1025 (old
->d_flags
& DCACHE_DENTRY_KILLED
) ||
1026 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1027 spin_unlock(&new->d_lock
);
1035 * enum d_walk_ret - action to talke during tree walk
1036 * @D_WALK_CONTINUE: contrinue walk
1037 * @D_WALK_QUIT: quit walk
1038 * @D_WALK_NORETRY: quit when retry is needed
1039 * @D_WALK_SKIP: skip this dentry and its children
1049 * d_walk - walk the dentry tree
1050 * @parent: start of walk
1051 * @data: data passed to @enter() and @finish()
1052 * @enter: callback when first entering the dentry
1053 * @finish: callback when successfully finished the walk
1055 * The @enter() and @finish() callbacks are called with d_lock held.
1057 static void d_walk(struct dentry
*parent
, void *data
,
1058 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1059 void (*finish
)(void *))
1061 struct dentry
*this_parent
;
1062 struct list_head
*next
;
1065 enum d_walk_ret ret
;
1068 seq
= read_seqbegin(&rename_lock
);
1070 this_parent
= parent
;
1071 spin_lock(&this_parent
->d_lock
);
1073 ret
= enter(data
, this_parent
);
1075 case D_WALK_CONTINUE
:
1080 case D_WALK_NORETRY
:
1085 next
= this_parent
->d_subdirs
.next
;
1087 while (next
!= &this_parent
->d_subdirs
) {
1088 struct list_head
*tmp
= next
;
1089 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1092 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1094 ret
= enter(data
, dentry
);
1096 case D_WALK_CONTINUE
:
1099 spin_unlock(&dentry
->d_lock
);
1101 case D_WALK_NORETRY
:
1105 spin_unlock(&dentry
->d_lock
);
1109 if (!list_empty(&dentry
->d_subdirs
)) {
1110 spin_unlock(&this_parent
->d_lock
);
1111 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1112 this_parent
= dentry
;
1113 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1116 spin_unlock(&dentry
->d_lock
);
1119 * All done at this level ... ascend and resume the search.
1121 if (this_parent
!= parent
) {
1122 struct dentry
*child
= this_parent
;
1123 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1126 next
= child
->d_u
.d_child
.next
;
1129 if (!locked
&& read_seqretry(&rename_lock
, seq
)) {
1130 spin_unlock(&this_parent
->d_lock
);
1137 spin_unlock(&this_parent
->d_lock
);
1139 write_sequnlock(&rename_lock
);
1148 write_seqlock(&rename_lock
);
1153 * Search for at least 1 mount point in the dentry's subdirs.
1154 * We descend to the next level whenever the d_subdirs
1155 * list is non-empty and continue searching.
1159 * have_submounts - check for mounts over a dentry
1160 * @parent: dentry to check.
1162 * Return true if the parent or its subdirectories contain
1166 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1169 if (d_mountpoint(dentry
)) {
1173 return D_WALK_CONTINUE
;
1176 int have_submounts(struct dentry
*parent
)
1180 d_walk(parent
, &ret
, check_mount
, NULL
);
1184 EXPORT_SYMBOL(have_submounts
);
1187 * Called by mount code to set a mountpoint and check if the mountpoint is
1188 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1189 * subtree can become unreachable).
1191 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1192 * this reason take rename_lock and d_lock on dentry and ancestors.
1194 int d_set_mounted(struct dentry
*dentry
)
1198 write_seqlock(&rename_lock
);
1199 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1200 /* Need exclusion wrt. check_submounts_and_drop() */
1201 spin_lock(&p
->d_lock
);
1202 if (unlikely(d_unhashed(p
))) {
1203 spin_unlock(&p
->d_lock
);
1206 spin_unlock(&p
->d_lock
);
1208 spin_lock(&dentry
->d_lock
);
1209 if (!d_unlinked(dentry
)) {
1210 dentry
->d_flags
|= DCACHE_MOUNTED
;
1213 spin_unlock(&dentry
->d_lock
);
1215 write_sequnlock(&rename_lock
);
1220 * Search the dentry child list of the specified parent,
1221 * and move any unused dentries to the end of the unused
1222 * list for prune_dcache(). We descend to the next level
1223 * whenever the d_subdirs list is non-empty and continue
1226 * It returns zero iff there are no unused children,
1227 * otherwise it returns the number of children moved to
1228 * the end of the unused list. This may not be the total
1229 * number of unused children, because select_parent can
1230 * drop the lock and return early due to latency
1234 struct select_data
{
1235 struct dentry
*start
;
1236 struct list_head dispose
;
1240 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1242 struct select_data
*data
= _data
;
1243 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1245 if (data
->start
== dentry
)
1249 * move only zero ref count dentries to the dispose list.
1251 * Those which are presently on the shrink list, being processed
1252 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1253 * loop in shrink_dcache_parent() might not make any progress
1256 if (dentry
->d_lockref
.count
) {
1257 dentry_lru_del(dentry
);
1258 } else if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
)) {
1259 dentry_lru_move_list(dentry
, &data
->dispose
);
1260 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
1262 ret
= D_WALK_NORETRY
;
1265 * We can return to the caller if we have found some (this
1266 * ensures forward progress). We'll be coming back to find
1269 if (data
->found
&& need_resched())
1276 * shrink_dcache_parent - prune dcache
1277 * @parent: parent of entries to prune
1279 * Prune the dcache to remove unused children of the parent dentry.
1281 void shrink_dcache_parent(struct dentry
*parent
)
1284 struct select_data data
;
1286 INIT_LIST_HEAD(&data
.dispose
);
1287 data
.start
= parent
;
1290 d_walk(parent
, &data
, select_collect
, NULL
);
1294 shrink_dentry_list(&data
.dispose
);
1298 EXPORT_SYMBOL(shrink_dcache_parent
);
1300 static enum d_walk_ret
check_and_collect(void *_data
, struct dentry
*dentry
)
1302 struct select_data
*data
= _data
;
1304 if (d_mountpoint(dentry
)) {
1305 data
->found
= -EBUSY
;
1309 return select_collect(_data
, dentry
);
1312 static void check_and_drop(void *_data
)
1314 struct select_data
*data
= _data
;
1316 if (d_mountpoint(data
->start
))
1317 data
->found
= -EBUSY
;
1319 __d_drop(data
->start
);
1323 * check_submounts_and_drop - prune dcache, check for submounts and drop
1325 * All done as a single atomic operation relative to has_unlinked_ancestor().
1326 * Returns 0 if successfully unhashed @parent. If there were submounts then
1329 * @dentry: dentry to prune and drop
1331 int check_submounts_and_drop(struct dentry
*dentry
)
1335 /* Negative dentries can be dropped without further checks */
1336 if (!dentry
->d_inode
) {
1342 struct select_data data
;
1344 INIT_LIST_HEAD(&data
.dispose
);
1345 data
.start
= dentry
;
1348 d_walk(dentry
, &data
, check_and_collect
, check_and_drop
);
1351 if (!list_empty(&data
.dispose
))
1352 shrink_dentry_list(&data
.dispose
);
1363 EXPORT_SYMBOL(check_submounts_and_drop
);
1366 * __d_alloc - allocate a dcache entry
1367 * @sb: filesystem it will belong to
1368 * @name: qstr of the name
1370 * Allocates a dentry. It returns %NULL if there is insufficient memory
1371 * available. On a success the dentry is returned. The name passed in is
1372 * copied and the copy passed in may be reused after this call.
1375 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1377 struct dentry
*dentry
;
1380 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1385 * We guarantee that the inline name is always NUL-terminated.
1386 * This way the memcpy() done by the name switching in rename
1387 * will still always have a NUL at the end, even if we might
1388 * be overwriting an internal NUL character
1390 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1391 if (name
->len
> DNAME_INLINE_LEN
-1) {
1392 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1394 kmem_cache_free(dentry_cache
, dentry
);
1398 dname
= dentry
->d_iname
;
1401 dentry
->d_name
.len
= name
->len
;
1402 dentry
->d_name
.hash
= name
->hash
;
1403 memcpy(dname
, name
->name
, name
->len
);
1404 dname
[name
->len
] = 0;
1406 /* Make sure we always see the terminating NUL character */
1408 dentry
->d_name
.name
= dname
;
1410 dentry
->d_lockref
.count
= 1;
1411 dentry
->d_flags
= 0;
1412 spin_lock_init(&dentry
->d_lock
);
1413 seqcount_init(&dentry
->d_seq
);
1414 dentry
->d_inode
= NULL
;
1415 dentry
->d_parent
= dentry
;
1417 dentry
->d_op
= NULL
;
1418 dentry
->d_fsdata
= NULL
;
1419 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1420 INIT_LIST_HEAD(&dentry
->d_lru
);
1421 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1422 INIT_HLIST_NODE(&dentry
->d_alias
);
1423 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1424 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1426 this_cpu_inc(nr_dentry
);
1432 * d_alloc - allocate a dcache entry
1433 * @parent: parent of entry to allocate
1434 * @name: qstr of the name
1436 * Allocates a dentry. It returns %NULL if there is insufficient memory
1437 * available. On a success the dentry is returned. The name passed in is
1438 * copied and the copy passed in may be reused after this call.
1440 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1442 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1446 spin_lock(&parent
->d_lock
);
1448 * don't need child lock because it is not subject
1449 * to concurrency here
1451 __dget_dlock(parent
);
1452 dentry
->d_parent
= parent
;
1453 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1454 spin_unlock(&parent
->d_lock
);
1458 EXPORT_SYMBOL(d_alloc
);
1460 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1462 struct dentry
*dentry
= __d_alloc(sb
, name
);
1464 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1467 EXPORT_SYMBOL(d_alloc_pseudo
);
1469 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1474 q
.len
= strlen(name
);
1475 q
.hash
= full_name_hash(q
.name
, q
.len
);
1476 return d_alloc(parent
, &q
);
1478 EXPORT_SYMBOL(d_alloc_name
);
1480 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1482 WARN_ON_ONCE(dentry
->d_op
);
1483 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1485 DCACHE_OP_REVALIDATE
|
1486 DCACHE_OP_WEAK_REVALIDATE
|
1487 DCACHE_OP_DELETE
));
1492 dentry
->d_flags
|= DCACHE_OP_HASH
;
1494 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1495 if (op
->d_revalidate
)
1496 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1497 if (op
->d_weak_revalidate
)
1498 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1500 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1502 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1505 EXPORT_SYMBOL(d_set_d_op
);
1507 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1509 spin_lock(&dentry
->d_lock
);
1511 if (unlikely(IS_AUTOMOUNT(inode
)))
1512 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1513 hlist_add_head(&dentry
->d_alias
, &inode
->i_dentry
);
1515 dentry
->d_inode
= inode
;
1516 dentry_rcuwalk_barrier(dentry
);
1517 spin_unlock(&dentry
->d_lock
);
1518 fsnotify_d_instantiate(dentry
, inode
);
1522 * d_instantiate - fill in inode information for a dentry
1523 * @entry: dentry to complete
1524 * @inode: inode to attach to this dentry
1526 * Fill in inode information in the entry.
1528 * This turns negative dentries into productive full members
1531 * NOTE! This assumes that the inode count has been incremented
1532 * (or otherwise set) by the caller to indicate that it is now
1533 * in use by the dcache.
1536 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1538 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1540 spin_lock(&inode
->i_lock
);
1541 __d_instantiate(entry
, inode
);
1543 spin_unlock(&inode
->i_lock
);
1544 security_d_instantiate(entry
, inode
);
1546 EXPORT_SYMBOL(d_instantiate
);
1549 * d_instantiate_unique - instantiate a non-aliased dentry
1550 * @entry: dentry to instantiate
1551 * @inode: inode to attach to this dentry
1553 * Fill in inode information in the entry. On success, it returns NULL.
1554 * If an unhashed alias of "entry" already exists, then we return the
1555 * aliased dentry instead and drop one reference to inode.
1557 * Note that in order to avoid conflicts with rename() etc, the caller
1558 * had better be holding the parent directory semaphore.
1560 * This also assumes that the inode count has been incremented
1561 * (or otherwise set) by the caller to indicate that it is now
1562 * in use by the dcache.
1564 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1565 struct inode
*inode
)
1567 struct dentry
*alias
;
1568 int len
= entry
->d_name
.len
;
1569 const char *name
= entry
->d_name
.name
;
1570 unsigned int hash
= entry
->d_name
.hash
;
1573 __d_instantiate(entry
, NULL
);
1577 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1579 * Don't need alias->d_lock here, because aliases with
1580 * d_parent == entry->d_parent are not subject to name or
1581 * parent changes, because the parent inode i_mutex is held.
1583 if (alias
->d_name
.hash
!= hash
)
1585 if (alias
->d_parent
!= entry
->d_parent
)
1587 if (alias
->d_name
.len
!= len
)
1589 if (dentry_cmp(alias
, name
, len
))
1595 __d_instantiate(entry
, inode
);
1599 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1601 struct dentry
*result
;
1603 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1606 spin_lock(&inode
->i_lock
);
1607 result
= __d_instantiate_unique(entry
, inode
);
1609 spin_unlock(&inode
->i_lock
);
1612 security_d_instantiate(entry
, inode
);
1616 BUG_ON(!d_unhashed(result
));
1621 EXPORT_SYMBOL(d_instantiate_unique
);
1623 struct dentry
*d_make_root(struct inode
*root_inode
)
1625 struct dentry
*res
= NULL
;
1628 static const struct qstr name
= QSTR_INIT("/", 1);
1630 res
= __d_alloc(root_inode
->i_sb
, &name
);
1632 d_instantiate(res
, root_inode
);
1638 EXPORT_SYMBOL(d_make_root
);
1640 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1642 struct dentry
*alias
;
1644 if (hlist_empty(&inode
->i_dentry
))
1646 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_alias
);
1652 * d_find_any_alias - find any alias for a given inode
1653 * @inode: inode to find an alias for
1655 * If any aliases exist for the given inode, take and return a
1656 * reference for one of them. If no aliases exist, return %NULL.
1658 struct dentry
*d_find_any_alias(struct inode
*inode
)
1662 spin_lock(&inode
->i_lock
);
1663 de
= __d_find_any_alias(inode
);
1664 spin_unlock(&inode
->i_lock
);
1667 EXPORT_SYMBOL(d_find_any_alias
);
1670 * d_obtain_alias - find or allocate a dentry for a given inode
1671 * @inode: inode to allocate the dentry for
1673 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1674 * similar open by handle operations. The returned dentry may be anonymous,
1675 * or may have a full name (if the inode was already in the cache).
1677 * When called on a directory inode, we must ensure that the inode only ever
1678 * has one dentry. If a dentry is found, that is returned instead of
1679 * allocating a new one.
1681 * On successful return, the reference to the inode has been transferred
1682 * to the dentry. In case of an error the reference on the inode is released.
1683 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1684 * be passed in and will be the error will be propagate to the return value,
1685 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1687 struct dentry
*d_obtain_alias(struct inode
*inode
)
1689 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1694 return ERR_PTR(-ESTALE
);
1696 return ERR_CAST(inode
);
1698 res
= d_find_any_alias(inode
);
1702 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1704 res
= ERR_PTR(-ENOMEM
);
1708 spin_lock(&inode
->i_lock
);
1709 res
= __d_find_any_alias(inode
);
1711 spin_unlock(&inode
->i_lock
);
1716 /* attach a disconnected dentry */
1717 spin_lock(&tmp
->d_lock
);
1718 tmp
->d_inode
= inode
;
1719 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1720 hlist_add_head(&tmp
->d_alias
, &inode
->i_dentry
);
1721 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1722 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1723 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1724 spin_unlock(&tmp
->d_lock
);
1725 spin_unlock(&inode
->i_lock
);
1726 security_d_instantiate(tmp
, inode
);
1731 if (res
&& !IS_ERR(res
))
1732 security_d_instantiate(res
, inode
);
1736 EXPORT_SYMBOL(d_obtain_alias
);
1739 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1740 * @inode: the inode which may have a disconnected dentry
1741 * @dentry: a negative dentry which we want to point to the inode.
1743 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1744 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1745 * and return it, else simply d_add the inode to the dentry and return NULL.
1747 * This is needed in the lookup routine of any filesystem that is exportable
1748 * (via knfsd) so that we can build dcache paths to directories effectively.
1750 * If a dentry was found and moved, then it is returned. Otherwise NULL
1751 * is returned. This matches the expected return value of ->lookup.
1753 * Cluster filesystems may call this function with a negative, hashed dentry.
1754 * In that case, we know that the inode will be a regular file, and also this
1755 * will only occur during atomic_open. So we need to check for the dentry
1756 * being already hashed only in the final case.
1758 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1760 struct dentry
*new = NULL
;
1763 return ERR_CAST(inode
);
1765 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1766 spin_lock(&inode
->i_lock
);
1767 new = __d_find_alias(inode
, 1);
1769 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1770 spin_unlock(&inode
->i_lock
);
1771 security_d_instantiate(new, inode
);
1772 d_move(new, dentry
);
1775 /* already taking inode->i_lock, so d_add() by hand */
1776 __d_instantiate(dentry
, inode
);
1777 spin_unlock(&inode
->i_lock
);
1778 security_d_instantiate(dentry
, inode
);
1782 d_instantiate(dentry
, inode
);
1783 if (d_unhashed(dentry
))
1788 EXPORT_SYMBOL(d_splice_alias
);
1791 * d_add_ci - lookup or allocate new dentry with case-exact name
1792 * @inode: the inode case-insensitive lookup has found
1793 * @dentry: the negative dentry that was passed to the parent's lookup func
1794 * @name: the case-exact name to be associated with the returned dentry
1796 * This is to avoid filling the dcache with case-insensitive names to the
1797 * same inode, only the actual correct case is stored in the dcache for
1798 * case-insensitive filesystems.
1800 * For a case-insensitive lookup match and if the the case-exact dentry
1801 * already exists in in the dcache, use it and return it.
1803 * If no entry exists with the exact case name, allocate new dentry with
1804 * the exact case, and return the spliced entry.
1806 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1809 struct dentry
*found
;
1813 * First check if a dentry matching the name already exists,
1814 * if not go ahead and create it now.
1816 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1817 if (unlikely(IS_ERR(found
)))
1820 new = d_alloc(dentry
->d_parent
, name
);
1822 found
= ERR_PTR(-ENOMEM
);
1826 found
= d_splice_alias(inode
, new);
1835 * If a matching dentry exists, and it's not negative use it.
1837 * Decrement the reference count to balance the iget() done
1840 if (found
->d_inode
) {
1841 if (unlikely(found
->d_inode
!= inode
)) {
1842 /* This can't happen because bad inodes are unhashed. */
1843 BUG_ON(!is_bad_inode(inode
));
1844 BUG_ON(!is_bad_inode(found
->d_inode
));
1851 * Negative dentry: instantiate it unless the inode is a directory and
1852 * already has a dentry.
1854 new = d_splice_alias(inode
, found
);
1865 EXPORT_SYMBOL(d_add_ci
);
1868 * Do the slow-case of the dentry name compare.
1870 * Unlike the dentry_cmp() function, we need to atomically
1871 * load the name and length information, so that the
1872 * filesystem can rely on them, and can use the 'name' and
1873 * 'len' information without worrying about walking off the
1874 * end of memory etc.
1876 * Thus the read_seqcount_retry() and the "duplicate" info
1877 * in arguments (the low-level filesystem should not look
1878 * at the dentry inode or name contents directly, since
1879 * rename can change them while we're in RCU mode).
1881 enum slow_d_compare
{
1887 static noinline
enum slow_d_compare
slow_dentry_cmp(
1888 const struct dentry
*parent
,
1889 struct dentry
*dentry
,
1891 const struct qstr
*name
)
1893 int tlen
= dentry
->d_name
.len
;
1894 const char *tname
= dentry
->d_name
.name
;
1896 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
1898 return D_COMP_SEQRETRY
;
1900 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
1901 return D_COMP_NOMATCH
;
1906 * __d_lookup_rcu - search for a dentry (racy, store-free)
1907 * @parent: parent dentry
1908 * @name: qstr of name we wish to find
1909 * @seqp: returns d_seq value at the point where the dentry was found
1910 * Returns: dentry, or NULL
1912 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1913 * resolution (store-free path walking) design described in
1914 * Documentation/filesystems/path-lookup.txt.
1916 * This is not to be used outside core vfs.
1918 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1919 * held, and rcu_read_lock held. The returned dentry must not be stored into
1920 * without taking d_lock and checking d_seq sequence count against @seq
1923 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1926 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1927 * the returned dentry, so long as its parent's seqlock is checked after the
1928 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1929 * is formed, giving integrity down the path walk.
1931 * NOTE! The caller *has* to check the resulting dentry against the sequence
1932 * number we've returned before using any of the resulting dentry state!
1934 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
1935 const struct qstr
*name
,
1938 u64 hashlen
= name
->hash_len
;
1939 const unsigned char *str
= name
->name
;
1940 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
1941 struct hlist_bl_node
*node
;
1942 struct dentry
*dentry
;
1945 * Note: There is significant duplication with __d_lookup_rcu which is
1946 * required to prevent single threaded performance regressions
1947 * especially on architectures where smp_rmb (in seqcounts) are costly.
1948 * Keep the two functions in sync.
1952 * The hash list is protected using RCU.
1954 * Carefully use d_seq when comparing a candidate dentry, to avoid
1955 * races with d_move().
1957 * It is possible that concurrent renames can mess up our list
1958 * walk here and result in missing our dentry, resulting in the
1959 * false-negative result. d_lookup() protects against concurrent
1960 * renames using rename_lock seqlock.
1962 * See Documentation/filesystems/path-lookup.txt for more details.
1964 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1969 * The dentry sequence count protects us from concurrent
1970 * renames, and thus protects parent and name fields.
1972 * The caller must perform a seqcount check in order
1973 * to do anything useful with the returned dentry.
1975 * NOTE! We do a "raw" seqcount_begin here. That means that
1976 * we don't wait for the sequence count to stabilize if it
1977 * is in the middle of a sequence change. If we do the slow
1978 * dentry compare, we will do seqretries until it is stable,
1979 * and if we end up with a successful lookup, we actually
1980 * want to exit RCU lookup anyway.
1982 seq
= raw_seqcount_begin(&dentry
->d_seq
);
1983 if (dentry
->d_parent
!= parent
)
1985 if (d_unhashed(dentry
))
1988 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1989 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
1992 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
1995 case D_COMP_NOMATCH
:
2002 if (dentry
->d_name
.hash_len
!= hashlen
)
2005 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2012 * d_lookup - search for a dentry
2013 * @parent: parent dentry
2014 * @name: qstr of name we wish to find
2015 * Returns: dentry, or NULL
2017 * d_lookup searches the children of the parent dentry for the name in
2018 * question. If the dentry is found its reference count is incremented and the
2019 * dentry is returned. The caller must use dput to free the entry when it has
2020 * finished using it. %NULL is returned if the dentry does not exist.
2022 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2024 struct dentry
*dentry
;
2028 seq
= read_seqbegin(&rename_lock
);
2029 dentry
= __d_lookup(parent
, name
);
2032 } while (read_seqretry(&rename_lock
, seq
));
2035 EXPORT_SYMBOL(d_lookup
);
2038 * __d_lookup - search for a dentry (racy)
2039 * @parent: parent dentry
2040 * @name: qstr of name we wish to find
2041 * Returns: dentry, or NULL
2043 * __d_lookup is like d_lookup, however it may (rarely) return a
2044 * false-negative result due to unrelated rename activity.
2046 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2047 * however it must be used carefully, eg. with a following d_lookup in
2048 * the case of failure.
2050 * __d_lookup callers must be commented.
2052 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2054 unsigned int len
= name
->len
;
2055 unsigned int hash
= name
->hash
;
2056 const unsigned char *str
= name
->name
;
2057 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2058 struct hlist_bl_node
*node
;
2059 struct dentry
*found
= NULL
;
2060 struct dentry
*dentry
;
2063 * Note: There is significant duplication with __d_lookup_rcu which is
2064 * required to prevent single threaded performance regressions
2065 * especially on architectures where smp_rmb (in seqcounts) are costly.
2066 * Keep the two functions in sync.
2070 * The hash list is protected using RCU.
2072 * Take d_lock when comparing a candidate dentry, to avoid races
2075 * It is possible that concurrent renames can mess up our list
2076 * walk here and result in missing our dentry, resulting in the
2077 * false-negative result. d_lookup() protects against concurrent
2078 * renames using rename_lock seqlock.
2080 * See Documentation/filesystems/path-lookup.txt for more details.
2084 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2086 if (dentry
->d_name
.hash
!= hash
)
2089 spin_lock(&dentry
->d_lock
);
2090 if (dentry
->d_parent
!= parent
)
2092 if (d_unhashed(dentry
))
2096 * It is safe to compare names since d_move() cannot
2097 * change the qstr (protected by d_lock).
2099 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2100 int tlen
= dentry
->d_name
.len
;
2101 const char *tname
= dentry
->d_name
.name
;
2102 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2105 if (dentry
->d_name
.len
!= len
)
2107 if (dentry_cmp(dentry
, str
, len
))
2111 dentry
->d_lockref
.count
++;
2113 spin_unlock(&dentry
->d_lock
);
2116 spin_unlock(&dentry
->d_lock
);
2124 * d_hash_and_lookup - hash the qstr then search for a dentry
2125 * @dir: Directory to search in
2126 * @name: qstr of name we wish to find
2128 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2130 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2133 * Check for a fs-specific hash function. Note that we must
2134 * calculate the standard hash first, as the d_op->d_hash()
2135 * routine may choose to leave the hash value unchanged.
2137 name
->hash
= full_name_hash(name
->name
, name
->len
);
2138 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2139 int err
= dir
->d_op
->d_hash(dir
, name
);
2140 if (unlikely(err
< 0))
2141 return ERR_PTR(err
);
2143 return d_lookup(dir
, name
);
2145 EXPORT_SYMBOL(d_hash_and_lookup
);
2148 * d_validate - verify dentry provided from insecure source (deprecated)
2149 * @dentry: The dentry alleged to be valid child of @dparent
2150 * @dparent: The parent dentry (known to be valid)
2152 * An insecure source has sent us a dentry, here we verify it and dget() it.
2153 * This is used by ncpfs in its readdir implementation.
2154 * Zero is returned in the dentry is invalid.
2156 * This function is slow for big directories, and deprecated, do not use it.
2158 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
2160 struct dentry
*child
;
2162 spin_lock(&dparent
->d_lock
);
2163 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
2164 if (dentry
== child
) {
2165 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2166 __dget_dlock(dentry
);
2167 spin_unlock(&dentry
->d_lock
);
2168 spin_unlock(&dparent
->d_lock
);
2172 spin_unlock(&dparent
->d_lock
);
2176 EXPORT_SYMBOL(d_validate
);
2179 * When a file is deleted, we have two options:
2180 * - turn this dentry into a negative dentry
2181 * - unhash this dentry and free it.
2183 * Usually, we want to just turn this into
2184 * a negative dentry, but if anybody else is
2185 * currently using the dentry or the inode
2186 * we can't do that and we fall back on removing
2187 * it from the hash queues and waiting for
2188 * it to be deleted later when it has no users
2192 * d_delete - delete a dentry
2193 * @dentry: The dentry to delete
2195 * Turn the dentry into a negative dentry if possible, otherwise
2196 * remove it from the hash queues so it can be deleted later
2199 void d_delete(struct dentry
* dentry
)
2201 struct inode
*inode
;
2204 * Are we the only user?
2207 spin_lock(&dentry
->d_lock
);
2208 inode
= dentry
->d_inode
;
2209 isdir
= S_ISDIR(inode
->i_mode
);
2210 if (dentry
->d_lockref
.count
== 1) {
2211 if (!spin_trylock(&inode
->i_lock
)) {
2212 spin_unlock(&dentry
->d_lock
);
2216 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2217 dentry_unlink_inode(dentry
);
2218 fsnotify_nameremove(dentry
, isdir
);
2222 if (!d_unhashed(dentry
))
2225 spin_unlock(&dentry
->d_lock
);
2227 fsnotify_nameremove(dentry
, isdir
);
2229 EXPORT_SYMBOL(d_delete
);
2231 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2233 BUG_ON(!d_unhashed(entry
));
2235 entry
->d_flags
|= DCACHE_RCUACCESS
;
2236 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2240 static void _d_rehash(struct dentry
* entry
)
2242 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2246 * d_rehash - add an entry back to the hash
2247 * @entry: dentry to add to the hash
2249 * Adds a dentry to the hash according to its name.
2252 void d_rehash(struct dentry
* entry
)
2254 spin_lock(&entry
->d_lock
);
2256 spin_unlock(&entry
->d_lock
);
2258 EXPORT_SYMBOL(d_rehash
);
2261 * dentry_update_name_case - update case insensitive dentry with a new name
2262 * @dentry: dentry to be updated
2265 * Update a case insensitive dentry with new case of name.
2267 * dentry must have been returned by d_lookup with name @name. Old and new
2268 * name lengths must match (ie. no d_compare which allows mismatched name
2271 * Parent inode i_mutex must be held over d_lookup and into this call (to
2272 * keep renames and concurrent inserts, and readdir(2) away).
2274 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2276 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2277 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2279 spin_lock(&dentry
->d_lock
);
2280 write_seqcount_begin(&dentry
->d_seq
);
2281 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2282 write_seqcount_end(&dentry
->d_seq
);
2283 spin_unlock(&dentry
->d_lock
);
2285 EXPORT_SYMBOL(dentry_update_name_case
);
2287 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2289 if (dname_external(target
)) {
2290 if (dname_external(dentry
)) {
2292 * Both external: swap the pointers
2294 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2297 * dentry:internal, target:external. Steal target's
2298 * storage and make target internal.
2300 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2301 dentry
->d_name
.len
+ 1);
2302 dentry
->d_name
.name
= target
->d_name
.name
;
2303 target
->d_name
.name
= target
->d_iname
;
2306 if (dname_external(dentry
)) {
2308 * dentry:external, target:internal. Give dentry's
2309 * storage to target and make dentry internal
2311 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2312 target
->d_name
.len
+ 1);
2313 target
->d_name
.name
= dentry
->d_name
.name
;
2314 dentry
->d_name
.name
= dentry
->d_iname
;
2317 * Both are internal. Just copy target to dentry
2319 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2320 target
->d_name
.len
+ 1);
2321 dentry
->d_name
.len
= target
->d_name
.len
;
2325 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2328 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2331 * XXXX: do we really need to take target->d_lock?
2333 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2334 spin_lock(&target
->d_parent
->d_lock
);
2336 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2337 spin_lock(&dentry
->d_parent
->d_lock
);
2338 spin_lock_nested(&target
->d_parent
->d_lock
,
2339 DENTRY_D_LOCK_NESTED
);
2341 spin_lock(&target
->d_parent
->d_lock
);
2342 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2343 DENTRY_D_LOCK_NESTED
);
2346 if (target
< dentry
) {
2347 spin_lock_nested(&target
->d_lock
, 2);
2348 spin_lock_nested(&dentry
->d_lock
, 3);
2350 spin_lock_nested(&dentry
->d_lock
, 2);
2351 spin_lock_nested(&target
->d_lock
, 3);
2355 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2356 struct dentry
*target
)
2358 if (target
->d_parent
!= dentry
->d_parent
)
2359 spin_unlock(&dentry
->d_parent
->d_lock
);
2360 if (target
->d_parent
!= target
)
2361 spin_unlock(&target
->d_parent
->d_lock
);
2365 * When switching names, the actual string doesn't strictly have to
2366 * be preserved in the target - because we're dropping the target
2367 * anyway. As such, we can just do a simple memcpy() to copy over
2368 * the new name before we switch.
2370 * Note that we have to be a lot more careful about getting the hash
2371 * switched - we have to switch the hash value properly even if it
2372 * then no longer matches the actual (corrupted) string of the target.
2373 * The hash value has to match the hash queue that the dentry is on..
2376 * __d_move - move a dentry
2377 * @dentry: entry to move
2378 * @target: new dentry
2380 * Update the dcache to reflect the move of a file name. Negative
2381 * dcache entries should not be moved in this way. Caller must hold
2382 * rename_lock, the i_mutex of the source and target directories,
2383 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2385 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2387 if (!dentry
->d_inode
)
2388 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2390 BUG_ON(d_ancestor(dentry
, target
));
2391 BUG_ON(d_ancestor(target
, dentry
));
2393 dentry_lock_for_move(dentry
, target
);
2395 write_seqcount_begin(&dentry
->d_seq
);
2396 write_seqcount_begin(&target
->d_seq
);
2398 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2401 * Move the dentry to the target hash queue. Don't bother checking
2402 * for the same hash queue because of how unlikely it is.
2405 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2407 /* Unhash the target: dput() will then get rid of it */
2410 list_del(&dentry
->d_u
.d_child
);
2411 list_del(&target
->d_u
.d_child
);
2413 /* Switch the names.. */
2414 switch_names(dentry
, target
);
2415 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2417 /* ... and switch the parents */
2418 if (IS_ROOT(dentry
)) {
2419 dentry
->d_parent
= target
->d_parent
;
2420 target
->d_parent
= target
;
2421 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2423 swap(dentry
->d_parent
, target
->d_parent
);
2425 /* And add them back to the (new) parent lists */
2426 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2429 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2431 write_seqcount_end(&target
->d_seq
);
2432 write_seqcount_end(&dentry
->d_seq
);
2434 dentry_unlock_parents_for_move(dentry
, target
);
2435 spin_unlock(&target
->d_lock
);
2436 fsnotify_d_move(dentry
);
2437 spin_unlock(&dentry
->d_lock
);
2441 * d_move - move a dentry
2442 * @dentry: entry to move
2443 * @target: new dentry
2445 * Update the dcache to reflect the move of a file name. Negative
2446 * dcache entries should not be moved in this way. See the locking
2447 * requirements for __d_move.
2449 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2451 write_seqlock(&rename_lock
);
2452 __d_move(dentry
, target
);
2453 write_sequnlock(&rename_lock
);
2455 EXPORT_SYMBOL(d_move
);
2458 * d_ancestor - search for an ancestor
2459 * @p1: ancestor dentry
2462 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2463 * an ancestor of p2, else NULL.
2465 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2469 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2470 if (p
->d_parent
== p1
)
2477 * This helper attempts to cope with remotely renamed directories
2479 * It assumes that the caller is already holding
2480 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2482 * Note: If ever the locking in lock_rename() changes, then please
2483 * remember to update this too...
2485 static struct dentry
*__d_unalias(struct inode
*inode
,
2486 struct dentry
*dentry
, struct dentry
*alias
)
2488 struct mutex
*m1
= NULL
, *m2
= NULL
;
2489 struct dentry
*ret
= ERR_PTR(-EBUSY
);
2491 /* If alias and dentry share a parent, then no extra locks required */
2492 if (alias
->d_parent
== dentry
->d_parent
)
2495 /* See lock_rename() */
2496 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2498 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2499 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2501 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2503 if (likely(!d_mountpoint(alias
))) {
2504 __d_move(alias
, dentry
);
2508 spin_unlock(&inode
->i_lock
);
2517 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2518 * named dentry in place of the dentry to be replaced.
2519 * returns with anon->d_lock held!
2521 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2523 struct dentry
*dparent
;
2525 dentry_lock_for_move(anon
, dentry
);
2527 write_seqcount_begin(&dentry
->d_seq
);
2528 write_seqcount_begin(&anon
->d_seq
);
2530 dparent
= dentry
->d_parent
;
2532 switch_names(dentry
, anon
);
2533 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2535 dentry
->d_parent
= dentry
;
2536 list_del_init(&dentry
->d_u
.d_child
);
2537 anon
->d_parent
= dparent
;
2538 list_move(&anon
->d_u
.d_child
, &dparent
->d_subdirs
);
2540 write_seqcount_end(&dentry
->d_seq
);
2541 write_seqcount_end(&anon
->d_seq
);
2543 dentry_unlock_parents_for_move(anon
, dentry
);
2544 spin_unlock(&dentry
->d_lock
);
2546 /* anon->d_lock still locked, returns locked */
2547 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2551 * d_materialise_unique - introduce an inode into the tree
2552 * @dentry: candidate dentry
2553 * @inode: inode to bind to the dentry, to which aliases may be attached
2555 * Introduces an dentry into the tree, substituting an extant disconnected
2556 * root directory alias in its place if there is one. Caller must hold the
2557 * i_mutex of the parent directory.
2559 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2561 struct dentry
*actual
;
2563 BUG_ON(!d_unhashed(dentry
));
2567 __d_instantiate(dentry
, NULL
);
2572 spin_lock(&inode
->i_lock
);
2574 if (S_ISDIR(inode
->i_mode
)) {
2575 struct dentry
*alias
;
2577 /* Does an aliased dentry already exist? */
2578 alias
= __d_find_alias(inode
, 0);
2581 write_seqlock(&rename_lock
);
2583 if (d_ancestor(alias
, dentry
)) {
2584 /* Check for loops */
2585 actual
= ERR_PTR(-ELOOP
);
2586 spin_unlock(&inode
->i_lock
);
2587 } else if (IS_ROOT(alias
)) {
2588 /* Is this an anonymous mountpoint that we
2589 * could splice into our tree? */
2590 __d_materialise_dentry(dentry
, alias
);
2591 write_sequnlock(&rename_lock
);
2595 /* Nope, but we must(!) avoid directory
2596 * aliasing. This drops inode->i_lock */
2597 actual
= __d_unalias(inode
, dentry
, alias
);
2599 write_sequnlock(&rename_lock
);
2600 if (IS_ERR(actual
)) {
2601 if (PTR_ERR(actual
) == -ELOOP
)
2602 pr_warn_ratelimited(
2603 "VFS: Lookup of '%s' in %s %s"
2604 " would have caused loop\n",
2605 dentry
->d_name
.name
,
2606 inode
->i_sb
->s_type
->name
,
2614 /* Add a unique reference */
2615 actual
= __d_instantiate_unique(dentry
, inode
);
2619 BUG_ON(!d_unhashed(actual
));
2621 spin_lock(&actual
->d_lock
);
2624 spin_unlock(&actual
->d_lock
);
2625 spin_unlock(&inode
->i_lock
);
2627 if (actual
== dentry
) {
2628 security_d_instantiate(dentry
, inode
);
2635 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2637 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2641 return -ENAMETOOLONG
;
2643 memcpy(*buffer
, str
, namelen
);
2647 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2649 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2653 * prepend_path - Prepend path string to a buffer
2654 * @path: the dentry/vfsmount to report
2655 * @root: root vfsmnt/dentry
2656 * @buffer: pointer to the end of the buffer
2657 * @buflen: pointer to buffer length
2659 * Caller holds the rename_lock.
2661 static int prepend_path(const struct path
*path
,
2662 const struct path
*root
,
2663 char **buffer
, int *buflen
)
2665 struct dentry
*dentry
= path
->dentry
;
2666 struct vfsmount
*vfsmnt
= path
->mnt
;
2667 struct mount
*mnt
= real_mount(vfsmnt
);
2671 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2672 struct dentry
* parent
;
2674 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2676 if (!mnt_has_parent(mnt
))
2678 dentry
= mnt
->mnt_mountpoint
;
2679 mnt
= mnt
->mnt_parent
;
2683 parent
= dentry
->d_parent
;
2685 spin_lock(&dentry
->d_lock
);
2686 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2687 spin_unlock(&dentry
->d_lock
);
2689 error
= prepend(buffer
, buflen
, "/", 1);
2697 if (!error
&& !slash
)
2698 error
= prepend(buffer
, buflen
, "/", 1);
2704 * Filesystems needing to implement special "root names"
2705 * should do so with ->d_dname()
2707 if (IS_ROOT(dentry
) &&
2708 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2709 WARN(1, "Root dentry has weird name <%.*s>\n",
2710 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2713 error
= prepend(buffer
, buflen
, "/", 1);
2715 error
= is_mounted(vfsmnt
) ? 1 : 2;
2720 * __d_path - return the path of a dentry
2721 * @path: the dentry/vfsmount to report
2722 * @root: root vfsmnt/dentry
2723 * @buf: buffer to return value in
2724 * @buflen: buffer length
2726 * Convert a dentry into an ASCII path name.
2728 * Returns a pointer into the buffer or an error code if the
2729 * path was too long.
2731 * "buflen" should be positive.
2733 * If the path is not reachable from the supplied root, return %NULL.
2735 char *__d_path(const struct path
*path
,
2736 const struct path
*root
,
2737 char *buf
, int buflen
)
2739 char *res
= buf
+ buflen
;
2742 prepend(&res
, &buflen
, "\0", 1);
2743 br_read_lock(&vfsmount_lock
);
2744 write_seqlock(&rename_lock
);
2745 error
= prepend_path(path
, root
, &res
, &buflen
);
2746 write_sequnlock(&rename_lock
);
2747 br_read_unlock(&vfsmount_lock
);
2750 return ERR_PTR(error
);
2756 char *d_absolute_path(const struct path
*path
,
2757 char *buf
, int buflen
)
2759 struct path root
= {};
2760 char *res
= buf
+ buflen
;
2763 prepend(&res
, &buflen
, "\0", 1);
2764 br_read_lock(&vfsmount_lock
);
2765 write_seqlock(&rename_lock
);
2766 error
= prepend_path(path
, &root
, &res
, &buflen
);
2767 write_sequnlock(&rename_lock
);
2768 br_read_unlock(&vfsmount_lock
);
2773 return ERR_PTR(error
);
2778 * same as __d_path but appends "(deleted)" for unlinked files.
2780 static int path_with_deleted(const struct path
*path
,
2781 const struct path
*root
,
2782 char **buf
, int *buflen
)
2784 prepend(buf
, buflen
, "\0", 1);
2785 if (d_unlinked(path
->dentry
)) {
2786 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2791 return prepend_path(path
, root
, buf
, buflen
);
2794 static int prepend_unreachable(char **buffer
, int *buflen
)
2796 return prepend(buffer
, buflen
, "(unreachable)", 13);
2800 * d_path - return the path of a dentry
2801 * @path: path to report
2802 * @buf: buffer to return value in
2803 * @buflen: buffer length
2805 * Convert a dentry into an ASCII path name. If the entry has been deleted
2806 * the string " (deleted)" is appended. Note that this is ambiguous.
2808 * Returns a pointer into the buffer or an error code if the path was
2809 * too long. Note: Callers should use the returned pointer, not the passed
2810 * in buffer, to use the name! The implementation often starts at an offset
2811 * into the buffer, and may leave 0 bytes at the start.
2813 * "buflen" should be positive.
2815 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2817 char *res
= buf
+ buflen
;
2822 * We have various synthetic filesystems that never get mounted. On
2823 * these filesystems dentries are never used for lookup purposes, and
2824 * thus don't need to be hashed. They also don't need a name until a
2825 * user wants to identify the object in /proc/pid/fd/. The little hack
2826 * below allows us to generate a name for these objects on demand:
2828 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2829 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2831 get_fs_root(current
->fs
, &root
);
2832 br_read_lock(&vfsmount_lock
);
2833 write_seqlock(&rename_lock
);
2834 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2835 write_sequnlock(&rename_lock
);
2836 br_read_unlock(&vfsmount_lock
);
2838 res
= ERR_PTR(error
);
2842 EXPORT_SYMBOL(d_path
);
2845 * Helper function for dentry_operations.d_dname() members
2847 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2848 const char *fmt
, ...)
2854 va_start(args
, fmt
);
2855 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2858 if (sz
> sizeof(temp
) || sz
> buflen
)
2859 return ERR_PTR(-ENAMETOOLONG
);
2861 buffer
+= buflen
- sz
;
2862 return memcpy(buffer
, temp
, sz
);
2865 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
2867 char *end
= buffer
+ buflen
;
2868 /* these dentries are never renamed, so d_lock is not needed */
2869 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
2870 prepend_name(&end
, &buflen
, &dentry
->d_name
) ||
2871 prepend(&end
, &buflen
, "/", 1))
2872 end
= ERR_PTR(-ENAMETOOLONG
);
2877 * Write full pathname from the root of the filesystem into the buffer.
2879 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2881 char *end
= buf
+ buflen
;
2884 prepend(&end
, &buflen
, "\0", 1);
2891 while (!IS_ROOT(dentry
)) {
2892 struct dentry
*parent
= dentry
->d_parent
;
2896 spin_lock(&dentry
->d_lock
);
2897 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2898 spin_unlock(&dentry
->d_lock
);
2899 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2907 return ERR_PTR(-ENAMETOOLONG
);
2910 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2914 write_seqlock(&rename_lock
);
2915 retval
= __dentry_path(dentry
, buf
, buflen
);
2916 write_sequnlock(&rename_lock
);
2920 EXPORT_SYMBOL(dentry_path_raw
);
2922 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2927 write_seqlock(&rename_lock
);
2928 if (d_unlinked(dentry
)) {
2930 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2934 retval
= __dentry_path(dentry
, buf
, buflen
);
2935 write_sequnlock(&rename_lock
);
2936 if (!IS_ERR(retval
) && p
)
2937 *p
= '/'; /* restore '/' overriden with '\0' */
2940 return ERR_PTR(-ENAMETOOLONG
);
2944 * NOTE! The user-level library version returns a
2945 * character pointer. The kernel system call just
2946 * returns the length of the buffer filled (which
2947 * includes the ending '\0' character), or a negative
2948 * error value. So libc would do something like
2950 * char *getcwd(char * buf, size_t size)
2954 * retval = sys_getcwd(buf, size);
2961 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2964 struct path pwd
, root
;
2965 char *page
= (char *) __get_free_page(GFP_USER
);
2970 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2973 br_read_lock(&vfsmount_lock
);
2974 write_seqlock(&rename_lock
);
2975 if (!d_unlinked(pwd
.dentry
)) {
2977 char *cwd
= page
+ PAGE_SIZE
;
2978 int buflen
= PAGE_SIZE
;
2980 prepend(&cwd
, &buflen
, "\0", 1);
2981 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2982 write_sequnlock(&rename_lock
);
2983 br_read_unlock(&vfsmount_lock
);
2988 /* Unreachable from current root */
2990 error
= prepend_unreachable(&cwd
, &buflen
);
2996 len
= PAGE_SIZE
+ page
- cwd
;
2999 if (copy_to_user(buf
, cwd
, len
))
3003 write_sequnlock(&rename_lock
);
3004 br_read_unlock(&vfsmount_lock
);
3010 free_page((unsigned long) page
);
3015 * Test whether new_dentry is a subdirectory of old_dentry.
3017 * Trivially implemented using the dcache structure
3021 * is_subdir - is new dentry a subdirectory of old_dentry
3022 * @new_dentry: new dentry
3023 * @old_dentry: old dentry
3025 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3026 * Returns 0 otherwise.
3027 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3030 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3035 if (new_dentry
== old_dentry
)
3039 /* for restarting inner loop in case of seq retry */
3040 seq
= read_seqbegin(&rename_lock
);
3042 * Need rcu_readlock to protect against the d_parent trashing
3046 if (d_ancestor(old_dentry
, new_dentry
))
3051 } while (read_seqretry(&rename_lock
, seq
));
3056 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3058 struct dentry
*root
= data
;
3059 if (dentry
!= root
) {
3060 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3063 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3064 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3065 dentry
->d_lockref
.count
--;
3068 return D_WALK_CONTINUE
;
3071 void d_genocide(struct dentry
*parent
)
3073 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3076 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3078 inode_dec_link_count(inode
);
3079 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3080 !hlist_unhashed(&dentry
->d_alias
) ||
3081 !d_unlinked(dentry
));
3082 spin_lock(&dentry
->d_parent
->d_lock
);
3083 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3084 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3085 (unsigned long long)inode
->i_ino
);
3086 spin_unlock(&dentry
->d_lock
);
3087 spin_unlock(&dentry
->d_parent
->d_lock
);
3088 d_instantiate(dentry
, inode
);
3090 EXPORT_SYMBOL(d_tmpfile
);
3092 static __initdata
unsigned long dhash_entries
;
3093 static int __init
set_dhash_entries(char *str
)
3097 dhash_entries
= simple_strtoul(str
, &str
, 0);
3100 __setup("dhash_entries=", set_dhash_entries
);
3102 static void __init
dcache_init_early(void)
3106 /* If hashes are distributed across NUMA nodes, defer
3107 * hash allocation until vmalloc space is available.
3113 alloc_large_system_hash("Dentry cache",
3114 sizeof(struct hlist_bl_head
),
3123 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3124 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3127 static void __init
dcache_init(void)
3132 * A constructor could be added for stable state like the lists,
3133 * but it is probably not worth it because of the cache nature
3136 dentry_cache
= KMEM_CACHE(dentry
,
3137 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3139 /* Hash may have been set up in dcache_init_early */
3144 alloc_large_system_hash("Dentry cache",
3145 sizeof(struct hlist_bl_head
),
3154 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3155 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3158 /* SLAB cache for __getname() consumers */
3159 struct kmem_cache
*names_cachep __read_mostly
;
3160 EXPORT_SYMBOL(names_cachep
);
3162 EXPORT_SYMBOL(d_genocide
);
3164 void __init
vfs_caches_init_early(void)
3166 dcache_init_early();
3170 void __init
vfs_caches_init(unsigned long mempages
)
3172 unsigned long reserve
;
3174 /* Base hash sizes on available memory, with a reserve equal to
3175 150% of current kernel size */
3177 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3178 mempages
-= reserve
;
3180 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3181 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3185 files_init(mempages
);