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_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
);
341 * Remove a dentry that is unreferenced and about to be pruned
342 * (unhashed and destroyed) from the LRU, and inform the file system.
343 * This wrapper should be called _prior_ to unhashing a victim dentry.
345 static void dentry_lru_prune(struct dentry
*dentry
)
347 if (!list_empty(&dentry
->d_lru
)) {
348 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
349 dentry
->d_op
->d_prune(dentry
);
351 spin_lock(&dcache_lru_lock
);
352 __dentry_lru_del(dentry
);
353 spin_unlock(&dcache_lru_lock
);
357 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
359 spin_lock(&dcache_lru_lock
);
360 if (list_empty(&dentry
->d_lru
)) {
361 list_add_tail(&dentry
->d_lru
, list
);
362 dentry
->d_sb
->s_nr_dentry_unused
++;
363 dentry_stat
.nr_unused
++;
365 list_move_tail(&dentry
->d_lru
, list
);
367 spin_unlock(&dcache_lru_lock
);
371 * d_kill - kill dentry and return parent
372 * @dentry: dentry to kill
373 * @parent: parent dentry
375 * The dentry must already be unhashed and removed from the LRU.
377 * If this is the root of the dentry tree, return NULL.
379 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
382 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
383 __releases(dentry
->d_lock
)
384 __releases(parent
->d_lock
)
385 __releases(dentry
->d_inode
->i_lock
)
387 list_del(&dentry
->d_u
.d_child
);
389 * Inform try_to_ascend() that we are no longer attached to the
392 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
394 spin_unlock(&parent
->d_lock
);
397 * dentry_iput drops the locks, at which point nobody (except
398 * transient RCU lookups) can reach this dentry.
405 * Unhash a dentry without inserting an RCU walk barrier or checking that
406 * dentry->d_lock is locked. The caller must take care of that, if
409 static void __d_shrink(struct dentry
*dentry
)
411 if (!d_unhashed(dentry
)) {
412 struct hlist_bl_head
*b
;
413 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
414 b
= &dentry
->d_sb
->s_anon
;
416 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
419 __hlist_bl_del(&dentry
->d_hash
);
420 dentry
->d_hash
.pprev
= NULL
;
426 * d_drop - drop a dentry
427 * @dentry: dentry to drop
429 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
430 * be found through a VFS lookup any more. Note that this is different from
431 * deleting the dentry - d_delete will try to mark the dentry negative if
432 * possible, giving a successful _negative_ lookup, while d_drop will
433 * just make the cache lookup fail.
435 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
436 * reason (NFS timeouts or autofs deletes).
438 * __d_drop requires dentry->d_lock.
440 void __d_drop(struct dentry
*dentry
)
442 if (!d_unhashed(dentry
)) {
444 dentry_rcuwalk_barrier(dentry
);
447 EXPORT_SYMBOL(__d_drop
);
449 void d_drop(struct dentry
*dentry
)
451 spin_lock(&dentry
->d_lock
);
453 spin_unlock(&dentry
->d_lock
);
455 EXPORT_SYMBOL(d_drop
);
458 * Finish off a dentry we've decided to kill.
459 * dentry->d_lock must be held, returns with it unlocked.
460 * If ref is non-zero, then decrement the refcount too.
461 * Returns dentry requiring refcount drop, or NULL if we're done.
463 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
464 __releases(dentry
->d_lock
)
467 struct dentry
*parent
;
469 inode
= dentry
->d_inode
;
470 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
472 spin_unlock(&dentry
->d_lock
);
474 return dentry
; /* try again with same dentry */
479 parent
= dentry
->d_parent
;
480 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
482 spin_unlock(&inode
->i_lock
);
489 * if dentry was on the d_lru list delete it from there.
490 * inform the fs via d_prune that this dentry is about to be
491 * unhashed and destroyed.
493 dentry_lru_prune(dentry
);
494 /* if it was on the hash then remove it */
496 return d_kill(dentry
, parent
);
502 * This is complicated by the fact that we do not want to put
503 * dentries that are no longer on any hash chain on the unused
504 * list: we'd much rather just get rid of them immediately.
506 * However, that implies that we have to traverse the dentry
507 * tree upwards to the parents which might _also_ now be
508 * scheduled for deletion (it may have been only waiting for
509 * its last child to go away).
511 * This tail recursion is done by hand as we don't want to depend
512 * on the compiler to always get this right (gcc generally doesn't).
513 * Real recursion would eat up our stack space.
517 * dput - release a dentry
518 * @dentry: dentry to release
520 * Release a dentry. This will drop the usage count and if appropriate
521 * call the dentry unlink method as well as removing it from the queues and
522 * releasing its resources. If the parent dentries were scheduled for release
523 * they too may now get deleted.
525 void dput(struct dentry
*dentry
)
531 if (dentry
->d_count
== 1)
533 spin_lock(&dentry
->d_lock
);
534 BUG_ON(!dentry
->d_count
);
535 if (dentry
->d_count
> 1) {
537 spin_unlock(&dentry
->d_lock
);
541 if (dentry
->d_flags
& DCACHE_OP_DELETE
) {
542 if (dentry
->d_op
->d_delete(dentry
))
546 /* Unreachable? Get rid of it */
547 if (d_unhashed(dentry
))
550 dentry
->d_flags
|= DCACHE_REFERENCED
;
551 dentry_lru_add(dentry
);
554 spin_unlock(&dentry
->d_lock
);
558 dentry
= dentry_kill(dentry
, 1);
565 * d_invalidate - invalidate a dentry
566 * @dentry: dentry to invalidate
568 * Try to invalidate the dentry if it turns out to be
569 * possible. If there are other dentries that can be
570 * reached through this one we can't delete it and we
571 * return -EBUSY. On success we return 0.
576 int d_invalidate(struct dentry
* dentry
)
579 * If it's already been dropped, return OK.
581 spin_lock(&dentry
->d_lock
);
582 if (d_unhashed(dentry
)) {
583 spin_unlock(&dentry
->d_lock
);
587 * Check whether to do a partial shrink_dcache
588 * to get rid of unused child entries.
590 if (!list_empty(&dentry
->d_subdirs
)) {
591 spin_unlock(&dentry
->d_lock
);
592 shrink_dcache_parent(dentry
);
593 spin_lock(&dentry
->d_lock
);
597 * Somebody else still using it?
599 * If it's a directory, we can't drop it
600 * for fear of somebody re-populating it
601 * with children (even though dropping it
602 * would make it unreachable from the root,
603 * we might still populate it if it was a
604 * working directory or similar).
605 * We also need to leave mountpoints alone,
608 if (dentry
->d_count
> 1 && dentry
->d_inode
) {
609 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
610 spin_unlock(&dentry
->d_lock
);
616 spin_unlock(&dentry
->d_lock
);
619 EXPORT_SYMBOL(d_invalidate
);
621 /* This must be called with d_lock held */
622 static inline void __dget_dlock(struct dentry
*dentry
)
627 static inline void __dget(struct dentry
*dentry
)
629 spin_lock(&dentry
->d_lock
);
630 __dget_dlock(dentry
);
631 spin_unlock(&dentry
->d_lock
);
634 struct dentry
*dget_parent(struct dentry
*dentry
)
640 * Don't need rcu_dereference because we re-check it was correct under
644 ret
= dentry
->d_parent
;
645 spin_lock(&ret
->d_lock
);
646 if (unlikely(ret
!= dentry
->d_parent
)) {
647 spin_unlock(&ret
->d_lock
);
652 BUG_ON(!ret
->d_count
);
654 spin_unlock(&ret
->d_lock
);
657 EXPORT_SYMBOL(dget_parent
);
660 * d_find_alias - grab a hashed alias of inode
661 * @inode: inode in question
662 * @want_discon: flag, used by d_splice_alias, to request
663 * that only a DISCONNECTED alias be returned.
665 * If inode has a hashed alias, or is a directory and has any alias,
666 * acquire the reference to alias and return it. Otherwise return NULL.
667 * Notice that if inode is a directory there can be only one alias and
668 * it can be unhashed only if it has no children, or if it is the root
671 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
672 * any other hashed alias over that one unless @want_discon is set,
673 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
675 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
677 struct dentry
*alias
, *discon_alias
;
678 struct hlist_node
*p
;
682 hlist_for_each_entry(alias
, p
, &inode
->i_dentry
, d_alias
) {
683 spin_lock(&alias
->d_lock
);
684 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
685 if (IS_ROOT(alias
) &&
686 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
687 discon_alias
= alias
;
688 } else if (!want_discon
) {
690 spin_unlock(&alias
->d_lock
);
694 spin_unlock(&alias
->d_lock
);
697 alias
= discon_alias
;
698 spin_lock(&alias
->d_lock
);
699 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
700 if (IS_ROOT(alias
) &&
701 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
703 spin_unlock(&alias
->d_lock
);
707 spin_unlock(&alias
->d_lock
);
713 struct dentry
*d_find_alias(struct inode
*inode
)
715 struct dentry
*de
= NULL
;
717 if (!hlist_empty(&inode
->i_dentry
)) {
718 spin_lock(&inode
->i_lock
);
719 de
= __d_find_alias(inode
, 0);
720 spin_unlock(&inode
->i_lock
);
724 EXPORT_SYMBOL(d_find_alias
);
727 * Try to kill dentries associated with this inode.
728 * WARNING: you must own a reference to inode.
730 void d_prune_aliases(struct inode
*inode
)
732 struct dentry
*dentry
;
733 struct hlist_node
*p
;
735 spin_lock(&inode
->i_lock
);
736 hlist_for_each_entry(dentry
, p
, &inode
->i_dentry
, d_alias
) {
737 spin_lock(&dentry
->d_lock
);
738 if (!dentry
->d_count
) {
739 __dget_dlock(dentry
);
741 spin_unlock(&dentry
->d_lock
);
742 spin_unlock(&inode
->i_lock
);
746 spin_unlock(&dentry
->d_lock
);
748 spin_unlock(&inode
->i_lock
);
750 EXPORT_SYMBOL(d_prune_aliases
);
753 * Try to throw away a dentry - free the inode, dput the parent.
754 * Requires dentry->d_lock is held, and dentry->d_count == 0.
755 * Releases dentry->d_lock.
757 * This may fail if locks cannot be acquired no problem, just try again.
759 static void try_prune_one_dentry(struct dentry
*dentry
)
760 __releases(dentry
->d_lock
)
762 struct dentry
*parent
;
764 parent
= dentry_kill(dentry
, 0);
766 * If dentry_kill returns NULL, we have nothing more to do.
767 * if it returns the same dentry, trylocks failed. In either
768 * case, just loop again.
770 * Otherwise, we need to prune ancestors too. This is necessary
771 * to prevent quadratic behavior of shrink_dcache_parent(), but
772 * is also expected to be beneficial in reducing dentry cache
777 if (parent
== dentry
)
780 /* Prune ancestors. */
783 spin_lock(&dentry
->d_lock
);
784 if (dentry
->d_count
> 1) {
786 spin_unlock(&dentry
->d_lock
);
789 dentry
= dentry_kill(dentry
, 1);
793 static void shrink_dentry_list(struct list_head
*list
)
795 struct dentry
*dentry
;
799 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
800 if (&dentry
->d_lru
== list
)
802 spin_lock(&dentry
->d_lock
);
803 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
804 spin_unlock(&dentry
->d_lock
);
809 * We found an inuse dentry which was not removed from
810 * the LRU because of laziness during lookup. Do not free
811 * it - just keep it off the LRU list.
813 if (dentry
->d_count
) {
814 dentry_lru_del(dentry
);
815 spin_unlock(&dentry
->d_lock
);
821 try_prune_one_dentry(dentry
);
829 * prune_dcache_sb - shrink the dcache
831 * @count: number of entries to try to free
833 * Attempt to shrink the superblock dcache LRU by @count entries. This is
834 * done when we need more memory an called from the superblock shrinker
837 * This function may fail to free any resources if all the dentries are in
840 void prune_dcache_sb(struct super_block
*sb
, int count
)
842 struct dentry
*dentry
;
843 LIST_HEAD(referenced
);
847 spin_lock(&dcache_lru_lock
);
848 while (!list_empty(&sb
->s_dentry_lru
)) {
849 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
850 struct dentry
, d_lru
);
851 BUG_ON(dentry
->d_sb
!= sb
);
853 if (!spin_trylock(&dentry
->d_lock
)) {
854 spin_unlock(&dcache_lru_lock
);
859 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
860 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
861 list_move(&dentry
->d_lru
, &referenced
);
862 spin_unlock(&dentry
->d_lock
);
864 list_move_tail(&dentry
->d_lru
, &tmp
);
865 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
866 spin_unlock(&dentry
->d_lock
);
870 cond_resched_lock(&dcache_lru_lock
);
872 if (!list_empty(&referenced
))
873 list_splice(&referenced
, &sb
->s_dentry_lru
);
874 spin_unlock(&dcache_lru_lock
);
876 shrink_dentry_list(&tmp
);
880 * shrink_dcache_sb - shrink dcache for a superblock
883 * Shrink the dcache for the specified super block. This is used to free
884 * the dcache before unmounting a file system.
886 void shrink_dcache_sb(struct super_block
*sb
)
890 spin_lock(&dcache_lru_lock
);
891 while (!list_empty(&sb
->s_dentry_lru
)) {
892 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
893 spin_unlock(&dcache_lru_lock
);
894 shrink_dentry_list(&tmp
);
895 spin_lock(&dcache_lru_lock
);
897 spin_unlock(&dcache_lru_lock
);
899 EXPORT_SYMBOL(shrink_dcache_sb
);
902 * destroy a single subtree of dentries for unmount
903 * - see the comments on shrink_dcache_for_umount() for a description of the
906 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
908 struct dentry
*parent
;
910 BUG_ON(!IS_ROOT(dentry
));
913 /* descend to the first leaf in the current subtree */
914 while (!list_empty(&dentry
->d_subdirs
))
915 dentry
= list_entry(dentry
->d_subdirs
.next
,
916 struct dentry
, d_u
.d_child
);
918 /* consume the dentries from this leaf up through its parents
919 * until we find one with children or run out altogether */
924 * remove the dentry from the lru, and inform
925 * the fs that this dentry is about to be
926 * unhashed and destroyed.
928 dentry_lru_prune(dentry
);
931 if (dentry
->d_count
!= 0) {
933 "BUG: Dentry %p{i=%lx,n=%s}"
935 " [unmount of %s %s]\n",
938 dentry
->d_inode
->i_ino
: 0UL,
941 dentry
->d_sb
->s_type
->name
,
946 if (IS_ROOT(dentry
)) {
948 list_del(&dentry
->d_u
.d_child
);
950 parent
= dentry
->d_parent
;
952 list_del(&dentry
->d_u
.d_child
);
955 inode
= dentry
->d_inode
;
957 dentry
->d_inode
= NULL
;
958 hlist_del_init(&dentry
->d_alias
);
959 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
960 dentry
->d_op
->d_iput(dentry
, inode
);
967 /* finished when we fall off the top of the tree,
968 * otherwise we ascend to the parent and move to the
969 * next sibling if there is one */
973 } while (list_empty(&dentry
->d_subdirs
));
975 dentry
= list_entry(dentry
->d_subdirs
.next
,
976 struct dentry
, d_u
.d_child
);
981 * destroy the dentries attached to a superblock on unmounting
982 * - we don't need to use dentry->d_lock because:
983 * - the superblock is detached from all mountings and open files, so the
984 * dentry trees will not be rearranged by the VFS
985 * - s_umount is write-locked, so the memory pressure shrinker will ignore
986 * any dentries belonging to this superblock that it comes across
987 * - the filesystem itself is no longer permitted to rearrange the dentries
990 void shrink_dcache_for_umount(struct super_block
*sb
)
992 struct dentry
*dentry
;
994 if (down_read_trylock(&sb
->s_umount
))
1000 shrink_dcache_for_umount_subtree(dentry
);
1002 while (!hlist_bl_empty(&sb
->s_anon
)) {
1003 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
1004 shrink_dcache_for_umount_subtree(dentry
);
1009 * This tries to ascend one level of parenthood, but
1010 * we can race with renaming, so we need to re-check
1011 * the parenthood after dropping the lock and check
1012 * that the sequence number still matches.
1014 static struct dentry
*try_to_ascend(struct dentry
*old
, int locked
, unsigned seq
)
1016 struct dentry
*new = old
->d_parent
;
1019 spin_unlock(&old
->d_lock
);
1020 spin_lock(&new->d_lock
);
1023 * might go back up the wrong parent if we have had a rename
1026 if (new != old
->d_parent
||
1027 (old
->d_flags
& DCACHE_DENTRY_KILLED
) ||
1028 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1029 spin_unlock(&new->d_lock
);
1038 * Search for at least 1 mount point in the dentry's subdirs.
1039 * We descend to the next level whenever the d_subdirs
1040 * list is non-empty and continue searching.
1044 * have_submounts - check for mounts over a dentry
1045 * @parent: dentry to check.
1047 * Return true if the parent or its subdirectories contain
1050 int have_submounts(struct dentry
*parent
)
1052 struct dentry
*this_parent
;
1053 struct list_head
*next
;
1057 seq
= read_seqbegin(&rename_lock
);
1059 this_parent
= parent
;
1061 if (d_mountpoint(parent
))
1063 spin_lock(&this_parent
->d_lock
);
1065 next
= this_parent
->d_subdirs
.next
;
1067 while (next
!= &this_parent
->d_subdirs
) {
1068 struct list_head
*tmp
= next
;
1069 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1072 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1073 /* Have we found a mount point ? */
1074 if (d_mountpoint(dentry
)) {
1075 spin_unlock(&dentry
->d_lock
);
1076 spin_unlock(&this_parent
->d_lock
);
1079 if (!list_empty(&dentry
->d_subdirs
)) {
1080 spin_unlock(&this_parent
->d_lock
);
1081 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1082 this_parent
= dentry
;
1083 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1086 spin_unlock(&dentry
->d_lock
);
1089 * All done at this level ... ascend and resume the search.
1091 if (this_parent
!= parent
) {
1092 struct dentry
*child
= this_parent
;
1093 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1096 next
= child
->d_u
.d_child
.next
;
1099 spin_unlock(&this_parent
->d_lock
);
1100 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1103 write_sequnlock(&rename_lock
);
1104 return 0; /* No mount points found in tree */
1106 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1109 write_sequnlock(&rename_lock
);
1116 write_seqlock(&rename_lock
);
1119 EXPORT_SYMBOL(have_submounts
);
1122 * Search the dentry child list of the specified parent,
1123 * and move any unused dentries to the end of the unused
1124 * list for prune_dcache(). We descend to the next level
1125 * whenever the d_subdirs list is non-empty and continue
1128 * It returns zero iff there are no unused children,
1129 * otherwise it returns the number of children moved to
1130 * the end of the unused list. This may not be the total
1131 * number of unused children, because select_parent can
1132 * drop the lock and return early due to latency
1135 static int select_parent(struct dentry
*parent
, struct list_head
*dispose
)
1137 struct dentry
*this_parent
;
1138 struct list_head
*next
;
1143 seq
= read_seqbegin(&rename_lock
);
1145 this_parent
= parent
;
1146 spin_lock(&this_parent
->d_lock
);
1148 next
= this_parent
->d_subdirs
.next
;
1150 while (next
!= &this_parent
->d_subdirs
) {
1151 struct list_head
*tmp
= next
;
1152 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1155 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1158 * move only zero ref count dentries to the dispose list.
1160 * Those which are presently on the shrink list, being processed
1161 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1162 * loop in shrink_dcache_parent() might not make any progress
1165 if (dentry
->d_count
) {
1166 dentry_lru_del(dentry
);
1167 } else if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
)) {
1168 dentry_lru_move_list(dentry
, dispose
);
1169 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
1173 * We can return to the caller if we have found some (this
1174 * ensures forward progress). We'll be coming back to find
1177 if (found
&& need_resched()) {
1178 spin_unlock(&dentry
->d_lock
);
1183 * Descend a level if the d_subdirs list is non-empty.
1185 if (!list_empty(&dentry
->d_subdirs
)) {
1186 spin_unlock(&this_parent
->d_lock
);
1187 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1188 this_parent
= dentry
;
1189 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1193 spin_unlock(&dentry
->d_lock
);
1196 * All done at this level ... ascend and resume the search.
1198 if (this_parent
!= parent
) {
1199 struct dentry
*child
= this_parent
;
1200 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
1203 next
= child
->d_u
.d_child
.next
;
1207 spin_unlock(&this_parent
->d_lock
);
1208 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1211 write_sequnlock(&rename_lock
);
1220 write_seqlock(&rename_lock
);
1225 * shrink_dcache_parent - prune dcache
1226 * @parent: parent of entries to prune
1228 * Prune the dcache to remove unused children of the parent dentry.
1230 void shrink_dcache_parent(struct dentry
* parent
)
1235 while ((found
= select_parent(parent
, &dispose
)) != 0)
1236 shrink_dentry_list(&dispose
);
1238 EXPORT_SYMBOL(shrink_dcache_parent
);
1241 * __d_alloc - allocate a dcache entry
1242 * @sb: filesystem it will belong to
1243 * @name: qstr of the name
1245 * Allocates a dentry. It returns %NULL if there is insufficient memory
1246 * available. On a success the dentry is returned. The name passed in is
1247 * copied and the copy passed in may be reused after this call.
1250 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1252 struct dentry
*dentry
;
1255 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1260 * We guarantee that the inline name is always NUL-terminated.
1261 * This way the memcpy() done by the name switching in rename
1262 * will still always have a NUL at the end, even if we might
1263 * be overwriting an internal NUL character
1265 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1266 if (name
->len
> DNAME_INLINE_LEN
-1) {
1267 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1269 kmem_cache_free(dentry_cache
, dentry
);
1273 dname
= dentry
->d_iname
;
1276 dentry
->d_name
.len
= name
->len
;
1277 dentry
->d_name
.hash
= name
->hash
;
1278 memcpy(dname
, name
->name
, name
->len
);
1279 dname
[name
->len
] = 0;
1281 /* Make sure we always see the terminating NUL character */
1283 dentry
->d_name
.name
= dname
;
1285 dentry
->d_count
= 1;
1286 dentry
->d_flags
= 0;
1287 spin_lock_init(&dentry
->d_lock
);
1288 seqcount_init(&dentry
->d_seq
);
1289 dentry
->d_inode
= NULL
;
1290 dentry
->d_parent
= dentry
;
1292 dentry
->d_op
= NULL
;
1293 dentry
->d_fsdata
= NULL
;
1294 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1295 INIT_LIST_HEAD(&dentry
->d_lru
);
1296 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1297 INIT_HLIST_NODE(&dentry
->d_alias
);
1298 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1299 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1301 this_cpu_inc(nr_dentry
);
1307 * d_alloc - allocate a dcache entry
1308 * @parent: parent of entry to allocate
1309 * @name: qstr of the name
1311 * Allocates a dentry. It returns %NULL if there is insufficient memory
1312 * available. On a success the dentry is returned. The name passed in is
1313 * copied and the copy passed in may be reused after this call.
1315 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1317 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1321 spin_lock(&parent
->d_lock
);
1323 * don't need child lock because it is not subject
1324 * to concurrency here
1326 __dget_dlock(parent
);
1327 dentry
->d_parent
= parent
;
1328 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1329 spin_unlock(&parent
->d_lock
);
1333 EXPORT_SYMBOL(d_alloc
);
1335 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1337 struct dentry
*dentry
= __d_alloc(sb
, name
);
1339 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1342 EXPORT_SYMBOL(d_alloc_pseudo
);
1344 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1349 q
.len
= strlen(name
);
1350 q
.hash
= full_name_hash(q
.name
, q
.len
);
1351 return d_alloc(parent
, &q
);
1353 EXPORT_SYMBOL(d_alloc_name
);
1355 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1357 WARN_ON_ONCE(dentry
->d_op
);
1358 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1360 DCACHE_OP_REVALIDATE
|
1361 DCACHE_OP_WEAK_REVALIDATE
|
1362 DCACHE_OP_DELETE
));
1367 dentry
->d_flags
|= DCACHE_OP_HASH
;
1369 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1370 if (op
->d_revalidate
)
1371 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1372 if (op
->d_weak_revalidate
)
1373 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1375 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1377 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1380 EXPORT_SYMBOL(d_set_d_op
);
1382 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1384 spin_lock(&dentry
->d_lock
);
1386 if (unlikely(IS_AUTOMOUNT(inode
)))
1387 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1388 hlist_add_head(&dentry
->d_alias
, &inode
->i_dentry
);
1390 dentry
->d_inode
= inode
;
1391 dentry_rcuwalk_barrier(dentry
);
1392 spin_unlock(&dentry
->d_lock
);
1393 fsnotify_d_instantiate(dentry
, inode
);
1397 * d_instantiate - fill in inode information for a dentry
1398 * @entry: dentry to complete
1399 * @inode: inode to attach to this dentry
1401 * Fill in inode information in the entry.
1403 * This turns negative dentries into productive full members
1406 * NOTE! This assumes that the inode count has been incremented
1407 * (or otherwise set) by the caller to indicate that it is now
1408 * in use by the dcache.
1411 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1413 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1415 spin_lock(&inode
->i_lock
);
1416 __d_instantiate(entry
, inode
);
1418 spin_unlock(&inode
->i_lock
);
1419 security_d_instantiate(entry
, inode
);
1421 EXPORT_SYMBOL(d_instantiate
);
1424 * d_instantiate_unique - instantiate a non-aliased dentry
1425 * @entry: dentry to instantiate
1426 * @inode: inode to attach to this dentry
1428 * Fill in inode information in the entry. On success, it returns NULL.
1429 * If an unhashed alias of "entry" already exists, then we return the
1430 * aliased dentry instead and drop one reference to inode.
1432 * Note that in order to avoid conflicts with rename() etc, the caller
1433 * had better be holding the parent directory semaphore.
1435 * This also assumes that the inode count has been incremented
1436 * (or otherwise set) by the caller to indicate that it is now
1437 * in use by the dcache.
1439 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1440 struct inode
*inode
)
1442 struct dentry
*alias
;
1443 int len
= entry
->d_name
.len
;
1444 const char *name
= entry
->d_name
.name
;
1445 unsigned int hash
= entry
->d_name
.hash
;
1446 struct hlist_node
*p
;
1449 __d_instantiate(entry
, NULL
);
1453 hlist_for_each_entry(alias
, p
, &inode
->i_dentry
, d_alias
) {
1455 * Don't need alias->d_lock here, because aliases with
1456 * d_parent == entry->d_parent are not subject to name or
1457 * parent changes, because the parent inode i_mutex is held.
1459 if (alias
->d_name
.hash
!= hash
)
1461 if (alias
->d_parent
!= entry
->d_parent
)
1463 if (alias
->d_name
.len
!= len
)
1465 if (dentry_cmp(alias
, name
, len
))
1471 __d_instantiate(entry
, inode
);
1475 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1477 struct dentry
*result
;
1479 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1482 spin_lock(&inode
->i_lock
);
1483 result
= __d_instantiate_unique(entry
, inode
);
1485 spin_unlock(&inode
->i_lock
);
1488 security_d_instantiate(entry
, inode
);
1492 BUG_ON(!d_unhashed(result
));
1497 EXPORT_SYMBOL(d_instantiate_unique
);
1499 struct dentry
*d_make_root(struct inode
*root_inode
)
1501 struct dentry
*res
= NULL
;
1504 static const struct qstr name
= QSTR_INIT("/", 1);
1506 res
= __d_alloc(root_inode
->i_sb
, &name
);
1508 d_instantiate(res
, root_inode
);
1514 EXPORT_SYMBOL(d_make_root
);
1516 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1518 struct dentry
*alias
;
1520 if (hlist_empty(&inode
->i_dentry
))
1522 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_alias
);
1528 * d_find_any_alias - find any alias for a given inode
1529 * @inode: inode to find an alias for
1531 * If any aliases exist for the given inode, take and return a
1532 * reference for one of them. If no aliases exist, return %NULL.
1534 struct dentry
*d_find_any_alias(struct inode
*inode
)
1538 spin_lock(&inode
->i_lock
);
1539 de
= __d_find_any_alias(inode
);
1540 spin_unlock(&inode
->i_lock
);
1543 EXPORT_SYMBOL(d_find_any_alias
);
1546 * d_obtain_alias - find or allocate a dentry for a given inode
1547 * @inode: inode to allocate the dentry for
1549 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1550 * similar open by handle operations. The returned dentry may be anonymous,
1551 * or may have a full name (if the inode was already in the cache).
1553 * When called on a directory inode, we must ensure that the inode only ever
1554 * has one dentry. If a dentry is found, that is returned instead of
1555 * allocating a new one.
1557 * On successful return, the reference to the inode has been transferred
1558 * to the dentry. In case of an error the reference on the inode is released.
1559 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1560 * be passed in and will be the error will be propagate to the return value,
1561 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1563 struct dentry
*d_obtain_alias(struct inode
*inode
)
1565 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1570 return ERR_PTR(-ESTALE
);
1572 return ERR_CAST(inode
);
1574 res
= d_find_any_alias(inode
);
1578 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1580 res
= ERR_PTR(-ENOMEM
);
1584 spin_lock(&inode
->i_lock
);
1585 res
= __d_find_any_alias(inode
);
1587 spin_unlock(&inode
->i_lock
);
1592 /* attach a disconnected dentry */
1593 spin_lock(&tmp
->d_lock
);
1594 tmp
->d_inode
= inode
;
1595 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1596 hlist_add_head(&tmp
->d_alias
, &inode
->i_dentry
);
1597 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1598 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1599 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1600 spin_unlock(&tmp
->d_lock
);
1601 spin_unlock(&inode
->i_lock
);
1602 security_d_instantiate(tmp
, inode
);
1607 if (res
&& !IS_ERR(res
))
1608 security_d_instantiate(res
, inode
);
1612 EXPORT_SYMBOL(d_obtain_alias
);
1615 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1616 * @inode: the inode which may have a disconnected dentry
1617 * @dentry: a negative dentry which we want to point to the inode.
1619 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1620 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1621 * and return it, else simply d_add the inode to the dentry and return NULL.
1623 * This is needed in the lookup routine of any filesystem that is exportable
1624 * (via knfsd) so that we can build dcache paths to directories effectively.
1626 * If a dentry was found and moved, then it is returned. Otherwise NULL
1627 * is returned. This matches the expected return value of ->lookup.
1630 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1632 struct dentry
*new = NULL
;
1635 return ERR_CAST(inode
);
1637 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1638 spin_lock(&inode
->i_lock
);
1639 new = __d_find_alias(inode
, 1);
1641 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1642 spin_unlock(&inode
->i_lock
);
1643 security_d_instantiate(new, inode
);
1644 d_move(new, dentry
);
1647 /* already taking inode->i_lock, so d_add() by hand */
1648 __d_instantiate(dentry
, inode
);
1649 spin_unlock(&inode
->i_lock
);
1650 security_d_instantiate(dentry
, inode
);
1654 d_add(dentry
, inode
);
1657 EXPORT_SYMBOL(d_splice_alias
);
1660 * d_add_ci - lookup or allocate new dentry with case-exact name
1661 * @inode: the inode case-insensitive lookup has found
1662 * @dentry: the negative dentry that was passed to the parent's lookup func
1663 * @name: the case-exact name to be associated with the returned dentry
1665 * This is to avoid filling the dcache with case-insensitive names to the
1666 * same inode, only the actual correct case is stored in the dcache for
1667 * case-insensitive filesystems.
1669 * For a case-insensitive lookup match and if the the case-exact dentry
1670 * already exists in in the dcache, use it and return it.
1672 * If no entry exists with the exact case name, allocate new dentry with
1673 * the exact case, and return the spliced entry.
1675 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1678 struct dentry
*found
;
1682 * First check if a dentry matching the name already exists,
1683 * if not go ahead and create it now.
1685 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1686 if (unlikely(IS_ERR(found
)))
1689 new = d_alloc(dentry
->d_parent
, name
);
1691 found
= ERR_PTR(-ENOMEM
);
1695 found
= d_splice_alias(inode
, new);
1704 * If a matching dentry exists, and it's not negative use it.
1706 * Decrement the reference count to balance the iget() done
1709 if (found
->d_inode
) {
1710 if (unlikely(found
->d_inode
!= inode
)) {
1711 /* This can't happen because bad inodes are unhashed. */
1712 BUG_ON(!is_bad_inode(inode
));
1713 BUG_ON(!is_bad_inode(found
->d_inode
));
1720 * Negative dentry: instantiate it unless the inode is a directory and
1721 * already has a dentry.
1723 new = d_splice_alias(inode
, found
);
1734 EXPORT_SYMBOL(d_add_ci
);
1737 * Do the slow-case of the dentry name compare.
1739 * Unlike the dentry_cmp() function, we need to atomically
1740 * load the name, length and inode information, so that the
1741 * filesystem can rely on them, and can use the 'name' and
1742 * 'len' information without worrying about walking off the
1743 * end of memory etc.
1745 * Thus the read_seqcount_retry() and the "duplicate" info
1746 * in arguments (the low-level filesystem should not look
1747 * at the dentry inode or name contents directly, since
1748 * rename can change them while we're in RCU mode).
1750 enum slow_d_compare
{
1756 static noinline
enum slow_d_compare
slow_dentry_cmp(
1757 const struct dentry
*parent
,
1758 struct inode
*inode
,
1759 struct dentry
*dentry
,
1761 const struct qstr
*name
)
1763 int tlen
= dentry
->d_name
.len
;
1764 const char *tname
= dentry
->d_name
.name
;
1765 struct inode
*i
= dentry
->d_inode
;
1767 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
1769 return D_COMP_SEQRETRY
;
1771 if (parent
->d_op
->d_compare(parent
, inode
,
1774 return D_COMP_NOMATCH
;
1779 * __d_lookup_rcu - search for a dentry (racy, store-free)
1780 * @parent: parent dentry
1781 * @name: qstr of name we wish to find
1782 * @seqp: returns d_seq value at the point where the dentry was found
1783 * @inode: returns dentry->d_inode when the inode was found valid.
1784 * Returns: dentry, or NULL
1786 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1787 * resolution (store-free path walking) design described in
1788 * Documentation/filesystems/path-lookup.txt.
1790 * This is not to be used outside core vfs.
1792 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1793 * held, and rcu_read_lock held. The returned dentry must not be stored into
1794 * without taking d_lock and checking d_seq sequence count against @seq
1797 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1800 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1801 * the returned dentry, so long as its parent's seqlock is checked after the
1802 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1803 * is formed, giving integrity down the path walk.
1805 * NOTE! The caller *has* to check the resulting dentry against the sequence
1806 * number we've returned before using any of the resulting dentry state!
1808 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
1809 const struct qstr
*name
,
1810 unsigned *seqp
, struct inode
*inode
)
1812 u64 hashlen
= name
->hash_len
;
1813 const unsigned char *str
= name
->name
;
1814 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
1815 struct hlist_bl_node
*node
;
1816 struct dentry
*dentry
;
1819 * Note: There is significant duplication with __d_lookup_rcu which is
1820 * required to prevent single threaded performance regressions
1821 * especially on architectures where smp_rmb (in seqcounts) are costly.
1822 * Keep the two functions in sync.
1826 * The hash list is protected using RCU.
1828 * Carefully use d_seq when comparing a candidate dentry, to avoid
1829 * races with d_move().
1831 * It is possible that concurrent renames can mess up our list
1832 * walk here and result in missing our dentry, resulting in the
1833 * false-negative result. d_lookup() protects against concurrent
1834 * renames using rename_lock seqlock.
1836 * See Documentation/filesystems/path-lookup.txt for more details.
1838 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1843 * The dentry sequence count protects us from concurrent
1844 * renames, and thus protects inode, parent and name fields.
1846 * The caller must perform a seqcount check in order
1847 * to do anything useful with the returned dentry,
1848 * including using the 'd_inode' pointer.
1850 * NOTE! We do a "raw" seqcount_begin here. That means that
1851 * we don't wait for the sequence count to stabilize if it
1852 * is in the middle of a sequence change. If we do the slow
1853 * dentry compare, we will do seqretries until it is stable,
1854 * and if we end up with a successful lookup, we actually
1855 * want to exit RCU lookup anyway.
1857 seq
= raw_seqcount_begin(&dentry
->d_seq
);
1858 if (dentry
->d_parent
!= parent
)
1860 if (d_unhashed(dentry
))
1864 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
1865 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
1867 switch (slow_dentry_cmp(parent
, inode
, dentry
, seq
, name
)) {
1870 case D_COMP_NOMATCH
:
1877 if (dentry
->d_name
.hash_len
!= hashlen
)
1879 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
1886 * d_lookup - search for a dentry
1887 * @parent: parent dentry
1888 * @name: qstr of name we wish to find
1889 * Returns: dentry, or NULL
1891 * d_lookup searches the children of the parent dentry for the name in
1892 * question. If the dentry is found its reference count is incremented and the
1893 * dentry is returned. The caller must use dput to free the entry when it has
1894 * finished using it. %NULL is returned if the dentry does not exist.
1896 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
1898 struct dentry
*dentry
;
1902 seq
= read_seqbegin(&rename_lock
);
1903 dentry
= __d_lookup(parent
, name
);
1906 } while (read_seqretry(&rename_lock
, seq
));
1909 EXPORT_SYMBOL(d_lookup
);
1912 * __d_lookup - search for a dentry (racy)
1913 * @parent: parent dentry
1914 * @name: qstr of name we wish to find
1915 * Returns: dentry, or NULL
1917 * __d_lookup is like d_lookup, however it may (rarely) return a
1918 * false-negative result due to unrelated rename activity.
1920 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1921 * however it must be used carefully, eg. with a following d_lookup in
1922 * the case of failure.
1924 * __d_lookup callers must be commented.
1926 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
1928 unsigned int len
= name
->len
;
1929 unsigned int hash
= name
->hash
;
1930 const unsigned char *str
= name
->name
;
1931 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
1932 struct hlist_bl_node
*node
;
1933 struct dentry
*found
= NULL
;
1934 struct dentry
*dentry
;
1937 * Note: There is significant duplication with __d_lookup_rcu which is
1938 * required to prevent single threaded performance regressions
1939 * especially on architectures where smp_rmb (in seqcounts) are costly.
1940 * Keep the two functions in sync.
1944 * The hash list is protected using RCU.
1946 * Take d_lock when comparing a candidate dentry, to avoid races
1949 * It is possible that concurrent renames can mess up our list
1950 * walk here and result in missing our dentry, resulting in the
1951 * false-negative result. d_lookup() protects against concurrent
1952 * renames using rename_lock seqlock.
1954 * See Documentation/filesystems/path-lookup.txt for more details.
1958 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
1960 if (dentry
->d_name
.hash
!= hash
)
1963 spin_lock(&dentry
->d_lock
);
1964 if (dentry
->d_parent
!= parent
)
1966 if (d_unhashed(dentry
))
1970 * It is safe to compare names since d_move() cannot
1971 * change the qstr (protected by d_lock).
1973 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
1974 int tlen
= dentry
->d_name
.len
;
1975 const char *tname
= dentry
->d_name
.name
;
1976 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1977 dentry
, dentry
->d_inode
,
1981 if (dentry
->d_name
.len
!= len
)
1983 if (dentry_cmp(dentry
, str
, len
))
1989 spin_unlock(&dentry
->d_lock
);
1992 spin_unlock(&dentry
->d_lock
);
2000 * d_hash_and_lookup - hash the qstr then search for a dentry
2001 * @dir: Directory to search in
2002 * @name: qstr of name we wish to find
2004 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2006 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2009 * Check for a fs-specific hash function. Note that we must
2010 * calculate the standard hash first, as the d_op->d_hash()
2011 * routine may choose to leave the hash value unchanged.
2013 name
->hash
= full_name_hash(name
->name
, name
->len
);
2014 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2015 int err
= dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
);
2016 if (unlikely(err
< 0))
2017 return ERR_PTR(err
);
2019 return d_lookup(dir
, name
);
2021 EXPORT_SYMBOL(d_hash_and_lookup
);
2024 * d_validate - verify dentry provided from insecure source (deprecated)
2025 * @dentry: The dentry alleged to be valid child of @dparent
2026 * @dparent: The parent dentry (known to be valid)
2028 * An insecure source has sent us a dentry, here we verify it and dget() it.
2029 * This is used by ncpfs in its readdir implementation.
2030 * Zero is returned in the dentry is invalid.
2032 * This function is slow for big directories, and deprecated, do not use it.
2034 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
2036 struct dentry
*child
;
2038 spin_lock(&dparent
->d_lock
);
2039 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
2040 if (dentry
== child
) {
2041 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2042 __dget_dlock(dentry
);
2043 spin_unlock(&dentry
->d_lock
);
2044 spin_unlock(&dparent
->d_lock
);
2048 spin_unlock(&dparent
->d_lock
);
2052 EXPORT_SYMBOL(d_validate
);
2055 * When a file is deleted, we have two options:
2056 * - turn this dentry into a negative dentry
2057 * - unhash this dentry and free it.
2059 * Usually, we want to just turn this into
2060 * a negative dentry, but if anybody else is
2061 * currently using the dentry or the inode
2062 * we can't do that and we fall back on removing
2063 * it from the hash queues and waiting for
2064 * it to be deleted later when it has no users
2068 * d_delete - delete a dentry
2069 * @dentry: The dentry to delete
2071 * Turn the dentry into a negative dentry if possible, otherwise
2072 * remove it from the hash queues so it can be deleted later
2075 void d_delete(struct dentry
* dentry
)
2077 struct inode
*inode
;
2080 * Are we the only user?
2083 spin_lock(&dentry
->d_lock
);
2084 inode
= dentry
->d_inode
;
2085 isdir
= S_ISDIR(inode
->i_mode
);
2086 if (dentry
->d_count
== 1) {
2087 if (!spin_trylock(&inode
->i_lock
)) {
2088 spin_unlock(&dentry
->d_lock
);
2092 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2093 dentry_unlink_inode(dentry
);
2094 fsnotify_nameremove(dentry
, isdir
);
2098 if (!d_unhashed(dentry
))
2101 spin_unlock(&dentry
->d_lock
);
2103 fsnotify_nameremove(dentry
, isdir
);
2105 EXPORT_SYMBOL(d_delete
);
2107 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2109 BUG_ON(!d_unhashed(entry
));
2111 entry
->d_flags
|= DCACHE_RCUACCESS
;
2112 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2116 static void _d_rehash(struct dentry
* entry
)
2118 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2122 * d_rehash - add an entry back to the hash
2123 * @entry: dentry to add to the hash
2125 * Adds a dentry to the hash according to its name.
2128 void d_rehash(struct dentry
* entry
)
2130 spin_lock(&entry
->d_lock
);
2132 spin_unlock(&entry
->d_lock
);
2134 EXPORT_SYMBOL(d_rehash
);
2137 * dentry_update_name_case - update case insensitive dentry with a new name
2138 * @dentry: dentry to be updated
2141 * Update a case insensitive dentry with new case of name.
2143 * dentry must have been returned by d_lookup with name @name. Old and new
2144 * name lengths must match (ie. no d_compare which allows mismatched name
2147 * Parent inode i_mutex must be held over d_lookup and into this call (to
2148 * keep renames and concurrent inserts, and readdir(2) away).
2150 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2152 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2153 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2155 spin_lock(&dentry
->d_lock
);
2156 write_seqcount_begin(&dentry
->d_seq
);
2157 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2158 write_seqcount_end(&dentry
->d_seq
);
2159 spin_unlock(&dentry
->d_lock
);
2161 EXPORT_SYMBOL(dentry_update_name_case
);
2163 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2165 if (dname_external(target
)) {
2166 if (dname_external(dentry
)) {
2168 * Both external: swap the pointers
2170 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2173 * dentry:internal, target:external. Steal target's
2174 * storage and make target internal.
2176 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2177 dentry
->d_name
.len
+ 1);
2178 dentry
->d_name
.name
= target
->d_name
.name
;
2179 target
->d_name
.name
= target
->d_iname
;
2182 if (dname_external(dentry
)) {
2184 * dentry:external, target:internal. Give dentry's
2185 * storage to target and make dentry internal
2187 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2188 target
->d_name
.len
+ 1);
2189 target
->d_name
.name
= dentry
->d_name
.name
;
2190 dentry
->d_name
.name
= dentry
->d_iname
;
2193 * Both are internal. Just copy target to dentry
2195 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2196 target
->d_name
.len
+ 1);
2197 dentry
->d_name
.len
= target
->d_name
.len
;
2201 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2204 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2207 * XXXX: do we really need to take target->d_lock?
2209 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2210 spin_lock(&target
->d_parent
->d_lock
);
2212 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2213 spin_lock(&dentry
->d_parent
->d_lock
);
2214 spin_lock_nested(&target
->d_parent
->d_lock
,
2215 DENTRY_D_LOCK_NESTED
);
2217 spin_lock(&target
->d_parent
->d_lock
);
2218 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2219 DENTRY_D_LOCK_NESTED
);
2222 if (target
< dentry
) {
2223 spin_lock_nested(&target
->d_lock
, 2);
2224 spin_lock_nested(&dentry
->d_lock
, 3);
2226 spin_lock_nested(&dentry
->d_lock
, 2);
2227 spin_lock_nested(&target
->d_lock
, 3);
2231 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2232 struct dentry
*target
)
2234 if (target
->d_parent
!= dentry
->d_parent
)
2235 spin_unlock(&dentry
->d_parent
->d_lock
);
2236 if (target
->d_parent
!= target
)
2237 spin_unlock(&target
->d_parent
->d_lock
);
2241 * When switching names, the actual string doesn't strictly have to
2242 * be preserved in the target - because we're dropping the target
2243 * anyway. As such, we can just do a simple memcpy() to copy over
2244 * the new name before we switch.
2246 * Note that we have to be a lot more careful about getting the hash
2247 * switched - we have to switch the hash value properly even if it
2248 * then no longer matches the actual (corrupted) string of the target.
2249 * The hash value has to match the hash queue that the dentry is on..
2252 * __d_move - move a dentry
2253 * @dentry: entry to move
2254 * @target: new dentry
2256 * Update the dcache to reflect the move of a file name. Negative
2257 * dcache entries should not be moved in this way. Caller must hold
2258 * rename_lock, the i_mutex of the source and target directories,
2259 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2261 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2263 if (!dentry
->d_inode
)
2264 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2266 BUG_ON(d_ancestor(dentry
, target
));
2267 BUG_ON(d_ancestor(target
, dentry
));
2269 dentry_lock_for_move(dentry
, target
);
2271 write_seqcount_begin(&dentry
->d_seq
);
2272 write_seqcount_begin(&target
->d_seq
);
2274 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2277 * Move the dentry to the target hash queue. Don't bother checking
2278 * for the same hash queue because of how unlikely it is.
2281 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2283 /* Unhash the target: dput() will then get rid of it */
2286 list_del(&dentry
->d_u
.d_child
);
2287 list_del(&target
->d_u
.d_child
);
2289 /* Switch the names.. */
2290 switch_names(dentry
, target
);
2291 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2293 /* ... and switch the parents */
2294 if (IS_ROOT(dentry
)) {
2295 dentry
->d_parent
= target
->d_parent
;
2296 target
->d_parent
= target
;
2297 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2299 swap(dentry
->d_parent
, target
->d_parent
);
2301 /* And add them back to the (new) parent lists */
2302 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2305 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2307 write_seqcount_end(&target
->d_seq
);
2308 write_seqcount_end(&dentry
->d_seq
);
2310 dentry_unlock_parents_for_move(dentry
, target
);
2311 spin_unlock(&target
->d_lock
);
2312 fsnotify_d_move(dentry
);
2313 spin_unlock(&dentry
->d_lock
);
2317 * d_move - move a dentry
2318 * @dentry: entry to move
2319 * @target: new dentry
2321 * Update the dcache to reflect the move of a file name. Negative
2322 * dcache entries should not be moved in this way. See the locking
2323 * requirements for __d_move.
2325 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2327 write_seqlock(&rename_lock
);
2328 __d_move(dentry
, target
);
2329 write_sequnlock(&rename_lock
);
2331 EXPORT_SYMBOL(d_move
);
2334 * d_ancestor - search for an ancestor
2335 * @p1: ancestor dentry
2338 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2339 * an ancestor of p2, else NULL.
2341 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2345 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2346 if (p
->d_parent
== p1
)
2353 * This helper attempts to cope with remotely renamed directories
2355 * It assumes that the caller is already holding
2356 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2358 * Note: If ever the locking in lock_rename() changes, then please
2359 * remember to update this too...
2361 static struct dentry
*__d_unalias(struct inode
*inode
,
2362 struct dentry
*dentry
, struct dentry
*alias
)
2364 struct mutex
*m1
= NULL
, *m2
= NULL
;
2365 struct dentry
*ret
= ERR_PTR(-EBUSY
);
2367 /* If alias and dentry share a parent, then no extra locks required */
2368 if (alias
->d_parent
== dentry
->d_parent
)
2371 /* See lock_rename() */
2372 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2374 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2375 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2377 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2379 if (likely(!d_mountpoint(alias
))) {
2380 __d_move(alias
, dentry
);
2384 spin_unlock(&inode
->i_lock
);
2393 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2394 * named dentry in place of the dentry to be replaced.
2395 * returns with anon->d_lock held!
2397 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2399 struct dentry
*dparent
;
2401 dentry_lock_for_move(anon
, dentry
);
2403 write_seqcount_begin(&dentry
->d_seq
);
2404 write_seqcount_begin(&anon
->d_seq
);
2406 dparent
= dentry
->d_parent
;
2408 switch_names(dentry
, anon
);
2409 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2411 dentry
->d_parent
= dentry
;
2412 list_del_init(&dentry
->d_u
.d_child
);
2413 anon
->d_parent
= dparent
;
2414 list_del(&anon
->d_u
.d_child
);
2415 list_add(&anon
->d_u
.d_child
, &dparent
->d_subdirs
);
2417 write_seqcount_end(&dentry
->d_seq
);
2418 write_seqcount_end(&anon
->d_seq
);
2420 dentry_unlock_parents_for_move(anon
, dentry
);
2421 spin_unlock(&dentry
->d_lock
);
2423 /* anon->d_lock still locked, returns locked */
2424 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2428 * d_materialise_unique - introduce an inode into the tree
2429 * @dentry: candidate dentry
2430 * @inode: inode to bind to the dentry, to which aliases may be attached
2432 * Introduces an dentry into the tree, substituting an extant disconnected
2433 * root directory alias in its place if there is one. Caller must hold the
2434 * i_mutex of the parent directory.
2436 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2438 struct dentry
*actual
;
2440 BUG_ON(!d_unhashed(dentry
));
2444 __d_instantiate(dentry
, NULL
);
2449 spin_lock(&inode
->i_lock
);
2451 if (S_ISDIR(inode
->i_mode
)) {
2452 struct dentry
*alias
;
2454 /* Does an aliased dentry already exist? */
2455 alias
= __d_find_alias(inode
, 0);
2458 write_seqlock(&rename_lock
);
2460 if (d_ancestor(alias
, dentry
)) {
2461 /* Check for loops */
2462 actual
= ERR_PTR(-ELOOP
);
2463 spin_unlock(&inode
->i_lock
);
2464 } else if (IS_ROOT(alias
)) {
2465 /* Is this an anonymous mountpoint that we
2466 * could splice into our tree? */
2467 __d_materialise_dentry(dentry
, alias
);
2468 write_sequnlock(&rename_lock
);
2472 /* Nope, but we must(!) avoid directory
2473 * aliasing. This drops inode->i_lock */
2474 actual
= __d_unalias(inode
, dentry
, alias
);
2476 write_sequnlock(&rename_lock
);
2477 if (IS_ERR(actual
)) {
2478 if (PTR_ERR(actual
) == -ELOOP
)
2479 pr_warn_ratelimited(
2480 "VFS: Lookup of '%s' in %s %s"
2481 " would have caused loop\n",
2482 dentry
->d_name
.name
,
2483 inode
->i_sb
->s_type
->name
,
2491 /* Add a unique reference */
2492 actual
= __d_instantiate_unique(dentry
, inode
);
2496 BUG_ON(!d_unhashed(actual
));
2498 spin_lock(&actual
->d_lock
);
2501 spin_unlock(&actual
->d_lock
);
2502 spin_unlock(&inode
->i_lock
);
2504 if (actual
== dentry
) {
2505 security_d_instantiate(dentry
, inode
);
2512 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2514 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2518 return -ENAMETOOLONG
;
2520 memcpy(*buffer
, str
, namelen
);
2524 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2526 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2530 * prepend_path - Prepend path string to a buffer
2531 * @path: the dentry/vfsmount to report
2532 * @root: root vfsmnt/dentry
2533 * @buffer: pointer to the end of the buffer
2534 * @buflen: pointer to buffer length
2536 * Caller holds the rename_lock.
2538 static int prepend_path(const struct path
*path
,
2539 const struct path
*root
,
2540 char **buffer
, int *buflen
)
2542 struct dentry
*dentry
= path
->dentry
;
2543 struct vfsmount
*vfsmnt
= path
->mnt
;
2544 struct mount
*mnt
= real_mount(vfsmnt
);
2548 br_read_lock(&vfsmount_lock
);
2549 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2550 struct dentry
* parent
;
2552 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2554 if (!mnt_has_parent(mnt
))
2556 dentry
= mnt
->mnt_mountpoint
;
2557 mnt
= mnt
->mnt_parent
;
2561 parent
= dentry
->d_parent
;
2563 spin_lock(&dentry
->d_lock
);
2564 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2565 spin_unlock(&dentry
->d_lock
);
2567 error
= prepend(buffer
, buflen
, "/", 1);
2575 if (!error
&& !slash
)
2576 error
= prepend(buffer
, buflen
, "/", 1);
2579 br_read_unlock(&vfsmount_lock
);
2584 * Filesystems needing to implement special "root names"
2585 * should do so with ->d_dname()
2587 if (IS_ROOT(dentry
) &&
2588 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2589 WARN(1, "Root dentry has weird name <%.*s>\n",
2590 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2593 error
= prepend(buffer
, buflen
, "/", 1);
2595 error
= is_mounted(vfsmnt
) ? 1 : 2;
2600 * __d_path - return the path of a dentry
2601 * @path: the dentry/vfsmount to report
2602 * @root: root vfsmnt/dentry
2603 * @buf: buffer to return value in
2604 * @buflen: buffer length
2606 * Convert a dentry into an ASCII path name.
2608 * Returns a pointer into the buffer or an error code if the
2609 * path was too long.
2611 * "buflen" should be positive.
2613 * If the path is not reachable from the supplied root, return %NULL.
2615 char *__d_path(const struct path
*path
,
2616 const struct path
*root
,
2617 char *buf
, int buflen
)
2619 char *res
= buf
+ buflen
;
2622 prepend(&res
, &buflen
, "\0", 1);
2623 write_seqlock(&rename_lock
);
2624 error
= prepend_path(path
, root
, &res
, &buflen
);
2625 write_sequnlock(&rename_lock
);
2628 return ERR_PTR(error
);
2634 char *d_absolute_path(const struct path
*path
,
2635 char *buf
, int buflen
)
2637 struct path root
= {};
2638 char *res
= buf
+ buflen
;
2641 prepend(&res
, &buflen
, "\0", 1);
2642 write_seqlock(&rename_lock
);
2643 error
= prepend_path(path
, &root
, &res
, &buflen
);
2644 write_sequnlock(&rename_lock
);
2649 return ERR_PTR(error
);
2654 * same as __d_path but appends "(deleted)" for unlinked files.
2656 static int path_with_deleted(const struct path
*path
,
2657 const struct path
*root
,
2658 char **buf
, int *buflen
)
2660 prepend(buf
, buflen
, "\0", 1);
2661 if (d_unlinked(path
->dentry
)) {
2662 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2667 return prepend_path(path
, root
, buf
, buflen
);
2670 static int prepend_unreachable(char **buffer
, int *buflen
)
2672 return prepend(buffer
, buflen
, "(unreachable)", 13);
2676 * d_path - return the path of a dentry
2677 * @path: path to report
2678 * @buf: buffer to return value in
2679 * @buflen: buffer length
2681 * Convert a dentry into an ASCII path name. If the entry has been deleted
2682 * the string " (deleted)" is appended. Note that this is ambiguous.
2684 * Returns a pointer into the buffer or an error code if the path was
2685 * too long. Note: Callers should use the returned pointer, not the passed
2686 * in buffer, to use the name! The implementation often starts at an offset
2687 * into the buffer, and may leave 0 bytes at the start.
2689 * "buflen" should be positive.
2691 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2693 char *res
= buf
+ buflen
;
2698 * We have various synthetic filesystems that never get mounted. On
2699 * these filesystems dentries are never used for lookup purposes, and
2700 * thus don't need to be hashed. They also don't need a name until a
2701 * user wants to identify the object in /proc/pid/fd/. The little hack
2702 * below allows us to generate a name for these objects on demand:
2704 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2705 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2707 get_fs_root(current
->fs
, &root
);
2708 write_seqlock(&rename_lock
);
2709 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2711 res
= ERR_PTR(error
);
2712 write_sequnlock(&rename_lock
);
2716 EXPORT_SYMBOL(d_path
);
2719 * Helper function for dentry_operations.d_dname() members
2721 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2722 const char *fmt
, ...)
2728 va_start(args
, fmt
);
2729 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2732 if (sz
> sizeof(temp
) || sz
> buflen
)
2733 return ERR_PTR(-ENAMETOOLONG
);
2735 buffer
+= buflen
- sz
;
2736 return memcpy(buffer
, temp
, sz
);
2740 * Write full pathname from the root of the filesystem into the buffer.
2742 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2744 char *end
= buf
+ buflen
;
2747 prepend(&end
, &buflen
, "\0", 1);
2754 while (!IS_ROOT(dentry
)) {
2755 struct dentry
*parent
= dentry
->d_parent
;
2759 spin_lock(&dentry
->d_lock
);
2760 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2761 spin_unlock(&dentry
->d_lock
);
2762 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2770 return ERR_PTR(-ENAMETOOLONG
);
2773 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2777 write_seqlock(&rename_lock
);
2778 retval
= __dentry_path(dentry
, buf
, buflen
);
2779 write_sequnlock(&rename_lock
);
2783 EXPORT_SYMBOL(dentry_path_raw
);
2785 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2790 write_seqlock(&rename_lock
);
2791 if (d_unlinked(dentry
)) {
2793 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2797 retval
= __dentry_path(dentry
, buf
, buflen
);
2798 write_sequnlock(&rename_lock
);
2799 if (!IS_ERR(retval
) && p
)
2800 *p
= '/'; /* restore '/' overriden with '\0' */
2803 return ERR_PTR(-ENAMETOOLONG
);
2807 * NOTE! The user-level library version returns a
2808 * character pointer. The kernel system call just
2809 * returns the length of the buffer filled (which
2810 * includes the ending '\0' character), or a negative
2811 * error value. So libc would do something like
2813 * char *getcwd(char * buf, size_t size)
2817 * retval = sys_getcwd(buf, size);
2824 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2827 struct path pwd
, root
;
2828 char *page
= (char *) __get_free_page(GFP_USER
);
2833 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2836 write_seqlock(&rename_lock
);
2837 if (!d_unlinked(pwd
.dentry
)) {
2839 char *cwd
= page
+ PAGE_SIZE
;
2840 int buflen
= PAGE_SIZE
;
2842 prepend(&cwd
, &buflen
, "\0", 1);
2843 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
2844 write_sequnlock(&rename_lock
);
2849 /* Unreachable from current root */
2851 error
= prepend_unreachable(&cwd
, &buflen
);
2857 len
= PAGE_SIZE
+ page
- cwd
;
2860 if (copy_to_user(buf
, cwd
, len
))
2864 write_sequnlock(&rename_lock
);
2870 free_page((unsigned long) page
);
2875 * Test whether new_dentry is a subdirectory of old_dentry.
2877 * Trivially implemented using the dcache structure
2881 * is_subdir - is new dentry a subdirectory of old_dentry
2882 * @new_dentry: new dentry
2883 * @old_dentry: old dentry
2885 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2886 * Returns 0 otherwise.
2887 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2890 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2895 if (new_dentry
== old_dentry
)
2899 /* for restarting inner loop in case of seq retry */
2900 seq
= read_seqbegin(&rename_lock
);
2902 * Need rcu_readlock to protect against the d_parent trashing
2906 if (d_ancestor(old_dentry
, new_dentry
))
2911 } while (read_seqretry(&rename_lock
, seq
));
2916 void d_genocide(struct dentry
*root
)
2918 struct dentry
*this_parent
;
2919 struct list_head
*next
;
2923 seq
= read_seqbegin(&rename_lock
);
2926 spin_lock(&this_parent
->d_lock
);
2928 next
= this_parent
->d_subdirs
.next
;
2930 while (next
!= &this_parent
->d_subdirs
) {
2931 struct list_head
*tmp
= next
;
2932 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2935 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2936 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2937 spin_unlock(&dentry
->d_lock
);
2940 if (!list_empty(&dentry
->d_subdirs
)) {
2941 spin_unlock(&this_parent
->d_lock
);
2942 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2943 this_parent
= dentry
;
2944 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2947 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2948 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2951 spin_unlock(&dentry
->d_lock
);
2953 if (this_parent
!= root
) {
2954 struct dentry
*child
= this_parent
;
2955 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2956 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2957 this_parent
->d_count
--;
2959 this_parent
= try_to_ascend(this_parent
, locked
, seq
);
2962 next
= child
->d_u
.d_child
.next
;
2965 spin_unlock(&this_parent
->d_lock
);
2966 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2969 write_sequnlock(&rename_lock
);
2976 write_seqlock(&rename_lock
);
2981 * find_inode_number - check for dentry with name
2982 * @dir: directory to check
2983 * @name: Name to find.
2985 * Check whether a dentry already exists for the given name,
2986 * and return the inode number if it has an inode. Otherwise
2989 * This routine is used to post-process directory listings for
2990 * filesystems using synthetic inode numbers, and is necessary
2991 * to keep getcwd() working.
2994 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2996 struct dentry
* dentry
;
2999 dentry
= d_hash_and_lookup(dir
, name
);
3000 if (!IS_ERR_OR_NULL(dentry
)) {
3001 if (dentry
->d_inode
)
3002 ino
= dentry
->d_inode
->i_ino
;
3007 EXPORT_SYMBOL(find_inode_number
);
3009 static __initdata
unsigned long dhash_entries
;
3010 static int __init
set_dhash_entries(char *str
)
3014 dhash_entries
= simple_strtoul(str
, &str
, 0);
3017 __setup("dhash_entries=", set_dhash_entries
);
3019 static void __init
dcache_init_early(void)
3023 /* If hashes are distributed across NUMA nodes, defer
3024 * hash allocation until vmalloc space is available.
3030 alloc_large_system_hash("Dentry cache",
3031 sizeof(struct hlist_bl_head
),
3040 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3041 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3044 static void __init
dcache_init(void)
3049 * A constructor could be added for stable state like the lists,
3050 * but it is probably not worth it because of the cache nature
3053 dentry_cache
= KMEM_CACHE(dentry
,
3054 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3056 /* Hash may have been set up in dcache_init_early */
3061 alloc_large_system_hash("Dentry cache",
3062 sizeof(struct hlist_bl_head
),
3071 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3072 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3075 /* SLAB cache for __getname() consumers */
3076 struct kmem_cache
*names_cachep __read_mostly
;
3077 EXPORT_SYMBOL(names_cachep
);
3079 EXPORT_SYMBOL(d_genocide
);
3081 void __init
vfs_caches_init_early(void)
3083 dcache_init_early();
3087 void __init
vfs_caches_init(unsigned long mempages
)
3089 unsigned long reserve
;
3091 /* Base hash sizes on available memory, with a reserve equal to
3092 150% of current kernel size */
3094 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3095 mempages
-= reserve
;
3097 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3098 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3102 files_init(mempages
);