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/module.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>
40 * dcache_inode_lock protects:
41 * - i_dentry, d_alias, d_inode
42 * dcache_hash_lock protects:
43 * - the dcache hash table, s_anon lists
44 * dcache_lru_lock protects:
45 * - the dcache lru lists and counters
52 * - d_parent and d_subdirs
53 * - childrens' d_child and d_parent
62 * If there is an ancestor relationship:
63 * dentry->d_parent->...->d_parent->d_lock
65 * dentry->d_parent->d_lock
68 * If no ancestor relationship:
69 * if (dentry1 < dentry2)
73 int sysctl_vfs_cache_pressure __read_mostly
= 100;
74 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
76 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_inode_lock
);
77 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_hash_lock
);
78 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(dcache_lru_lock
);
79 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
81 EXPORT_SYMBOL(rename_lock
);
82 EXPORT_SYMBOL(dcache_inode_lock
);
84 static struct kmem_cache
*dentry_cache __read_mostly
;
86 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
96 #define D_HASHBITS d_hash_shift
97 #define D_HASHMASK d_hash_mask
99 static unsigned int d_hash_mask __read_mostly
;
100 static unsigned int d_hash_shift __read_mostly
;
101 static struct hlist_head
*dentry_hashtable __read_mostly
;
103 /* Statistics gathering. */
104 struct dentry_stat_t dentry_stat
= {
108 static DEFINE_PER_CPU(unsigned int, nr_dentry
);
110 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
111 static int get_nr_dentry(void)
115 for_each_possible_cpu(i
)
116 sum
+= per_cpu(nr_dentry
, i
);
117 return sum
< 0 ? 0 : sum
;
120 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
121 size_t *lenp
, loff_t
*ppos
)
123 dentry_stat
.nr_dentry
= get_nr_dentry();
124 return proc_dointvec(table
, write
, buffer
, lenp
, ppos
);
128 static void __d_free(struct rcu_head
*head
)
130 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
132 WARN_ON(!list_empty(&dentry
->d_alias
));
133 if (dname_external(dentry
))
134 kfree(dentry
->d_name
.name
);
135 kmem_cache_free(dentry_cache
, dentry
);
141 static void d_free(struct dentry
*dentry
)
143 BUG_ON(dentry
->d_count
);
144 this_cpu_dec(nr_dentry
);
145 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
146 dentry
->d_op
->d_release(dentry
);
148 /* if dentry was never inserted into hash, immediate free is OK */
149 if (hlist_unhashed(&dentry
->d_hash
))
150 __d_free(&dentry
->d_u
.d_rcu
);
152 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
156 * Release the dentry's inode, using the filesystem
157 * d_iput() operation if defined.
159 static void dentry_iput(struct dentry
* dentry
)
160 __releases(dentry
->d_lock
)
161 __releases(dcache_inode_lock
)
163 struct inode
*inode
= dentry
->d_inode
;
165 dentry
->d_inode
= NULL
;
166 list_del_init(&dentry
->d_alias
);
167 spin_unlock(&dentry
->d_lock
);
168 spin_unlock(&dcache_inode_lock
);
170 fsnotify_inoderemove(inode
);
171 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
172 dentry
->d_op
->d_iput(dentry
, inode
);
176 spin_unlock(&dentry
->d_lock
);
177 spin_unlock(&dcache_inode_lock
);
182 * dentry_lru_(add|del|move_tail) must be called with d_lock held.
184 static void dentry_lru_add(struct dentry
*dentry
)
186 if (list_empty(&dentry
->d_lru
)) {
187 spin_lock(&dcache_lru_lock
);
188 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
189 dentry
->d_sb
->s_nr_dentry_unused
++;
190 dentry_stat
.nr_unused
++;
191 spin_unlock(&dcache_lru_lock
);
195 static void __dentry_lru_del(struct dentry
*dentry
)
197 list_del_init(&dentry
->d_lru
);
198 dentry
->d_sb
->s_nr_dentry_unused
--;
199 dentry_stat
.nr_unused
--;
202 static void dentry_lru_del(struct dentry
*dentry
)
204 if (!list_empty(&dentry
->d_lru
)) {
205 spin_lock(&dcache_lru_lock
);
206 __dentry_lru_del(dentry
);
207 spin_unlock(&dcache_lru_lock
);
211 static void dentry_lru_move_tail(struct dentry
*dentry
)
213 spin_lock(&dcache_lru_lock
);
214 if (list_empty(&dentry
->d_lru
)) {
215 list_add_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
216 dentry
->d_sb
->s_nr_dentry_unused
++;
217 dentry_stat
.nr_unused
++;
219 list_move_tail(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
221 spin_unlock(&dcache_lru_lock
);
225 * d_kill - kill dentry and return parent
226 * @dentry: dentry to kill
228 * The dentry must already be unhashed and removed from the LRU.
230 * If this is the root of the dentry tree, return NULL.
232 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
235 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
236 __releases(dentry
->d_lock
)
237 __releases(parent
->d_lock
)
238 __releases(dcache_inode_lock
)
240 dentry
->d_parent
= NULL
;
241 list_del(&dentry
->d_u
.d_child
);
243 spin_unlock(&parent
->d_lock
);
246 * dentry_iput drops the locks, at which point nobody (except
247 * transient RCU lookups) can reach this dentry.
254 * d_drop - drop a dentry
255 * @dentry: dentry to drop
257 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
258 * be found through a VFS lookup any more. Note that this is different from
259 * deleting the dentry - d_delete will try to mark the dentry negative if
260 * possible, giving a successful _negative_ lookup, while d_drop will
261 * just make the cache lookup fail.
263 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
264 * reason (NFS timeouts or autofs deletes).
266 * __d_drop requires dentry->d_lock.
268 void __d_drop(struct dentry
*dentry
)
270 if (!(dentry
->d_flags
& DCACHE_UNHASHED
)) {
271 dentry
->d_flags
|= DCACHE_UNHASHED
;
272 spin_lock(&dcache_hash_lock
);
273 hlist_del_rcu(&dentry
->d_hash
);
274 spin_unlock(&dcache_hash_lock
);
277 EXPORT_SYMBOL(__d_drop
);
279 void d_drop(struct dentry
*dentry
)
281 spin_lock(&dentry
->d_lock
);
283 spin_unlock(&dentry
->d_lock
);
285 EXPORT_SYMBOL(d_drop
);
288 * Finish off a dentry we've decided to kill.
289 * dentry->d_lock must be held, returns with it unlocked.
290 * If ref is non-zero, then decrement the refcount too.
291 * Returns dentry requiring refcount drop, or NULL if we're done.
293 static inline struct dentry
*dentry_kill(struct dentry
*dentry
, int ref
)
294 __releases(dentry
->d_lock
)
296 struct dentry
*parent
;
298 if (!spin_trylock(&dcache_inode_lock
)) {
300 spin_unlock(&dentry
->d_lock
);
302 return dentry
; /* try again with same dentry */
307 parent
= dentry
->d_parent
;
308 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
309 spin_unlock(&dcache_inode_lock
);
314 /* if dentry was on the d_lru list delete it from there */
315 dentry_lru_del(dentry
);
316 /* if it was on the hash then remove it */
318 return d_kill(dentry
, parent
);
324 * This is complicated by the fact that we do not want to put
325 * dentries that are no longer on any hash chain on the unused
326 * list: we'd much rather just get rid of them immediately.
328 * However, that implies that we have to traverse the dentry
329 * tree upwards to the parents which might _also_ now be
330 * scheduled for deletion (it may have been only waiting for
331 * its last child to go away).
333 * This tail recursion is done by hand as we don't want to depend
334 * on the compiler to always get this right (gcc generally doesn't).
335 * Real recursion would eat up our stack space.
339 * dput - release a dentry
340 * @dentry: dentry to release
342 * Release a dentry. This will drop the usage count and if appropriate
343 * call the dentry unlink method as well as removing it from the queues and
344 * releasing its resources. If the parent dentries were scheduled for release
345 * they too may now get deleted.
347 void dput(struct dentry
*dentry
)
353 if (dentry
->d_count
== 1)
355 spin_lock(&dentry
->d_lock
);
356 BUG_ON(!dentry
->d_count
);
357 if (dentry
->d_count
> 1) {
359 spin_unlock(&dentry
->d_lock
);
363 if (dentry
->d_op
&& dentry
->d_op
->d_delete
) {
364 if (dentry
->d_op
->d_delete(dentry
))
368 /* Unreachable? Get rid of it */
369 if (d_unhashed(dentry
))
372 /* Otherwise leave it cached and ensure it's on the LRU */
373 dentry
->d_flags
|= DCACHE_REFERENCED
;
374 dentry_lru_add(dentry
);
377 spin_unlock(&dentry
->d_lock
);
381 dentry
= dentry_kill(dentry
, 1);
388 * d_invalidate - invalidate a dentry
389 * @dentry: dentry to invalidate
391 * Try to invalidate the dentry if it turns out to be
392 * possible. If there are other dentries that can be
393 * reached through this one we can't delete it and we
394 * return -EBUSY. On success we return 0.
399 int d_invalidate(struct dentry
* dentry
)
402 * If it's already been dropped, return OK.
404 spin_lock(&dentry
->d_lock
);
405 if (d_unhashed(dentry
)) {
406 spin_unlock(&dentry
->d_lock
);
410 * Check whether to do a partial shrink_dcache
411 * to get rid of unused child entries.
413 if (!list_empty(&dentry
->d_subdirs
)) {
414 spin_unlock(&dentry
->d_lock
);
415 shrink_dcache_parent(dentry
);
416 spin_lock(&dentry
->d_lock
);
420 * Somebody else still using it?
422 * If it's a directory, we can't drop it
423 * for fear of somebody re-populating it
424 * with children (even though dropping it
425 * would make it unreachable from the root,
426 * we might still populate it if it was a
427 * working directory or similar).
429 if (dentry
->d_count
> 1) {
430 if (dentry
->d_inode
&& S_ISDIR(dentry
->d_inode
->i_mode
)) {
431 spin_unlock(&dentry
->d_lock
);
437 spin_unlock(&dentry
->d_lock
);
440 EXPORT_SYMBOL(d_invalidate
);
442 /* This must be called with d_lock held */
443 static inline void __dget_dlock(struct dentry
*dentry
)
448 static inline void __dget(struct dentry
*dentry
)
450 spin_lock(&dentry
->d_lock
);
451 __dget_dlock(dentry
);
452 spin_unlock(&dentry
->d_lock
);
455 struct dentry
*dget_parent(struct dentry
*dentry
)
461 * Don't need rcu_dereference because we re-check it was correct under
465 ret
= dentry
->d_parent
;
470 spin_lock(&ret
->d_lock
);
471 if (unlikely(ret
!= dentry
->d_parent
)) {
472 spin_unlock(&ret
->d_lock
);
477 BUG_ON(!ret
->d_count
);
479 spin_unlock(&ret
->d_lock
);
483 EXPORT_SYMBOL(dget_parent
);
486 * d_find_alias - grab a hashed alias of inode
487 * @inode: inode in question
488 * @want_discon: flag, used by d_splice_alias, to request
489 * that only a DISCONNECTED alias be returned.
491 * If inode has a hashed alias, or is a directory and has any alias,
492 * acquire the reference to alias and return it. Otherwise return NULL.
493 * Notice that if inode is a directory there can be only one alias and
494 * it can be unhashed only if it has no children, or if it is the root
497 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
498 * any other hashed alias over that one unless @want_discon is set,
499 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
501 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
503 struct dentry
*alias
, *discon_alias
;
507 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
508 spin_lock(&alias
->d_lock
);
509 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
510 if (IS_ROOT(alias
) &&
511 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
512 discon_alias
= alias
;
513 } else if (!want_discon
) {
515 spin_unlock(&alias
->d_lock
);
519 spin_unlock(&alias
->d_lock
);
522 alias
= discon_alias
;
523 spin_lock(&alias
->d_lock
);
524 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
525 if (IS_ROOT(alias
) &&
526 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
528 spin_unlock(&alias
->d_lock
);
532 spin_unlock(&alias
->d_lock
);
538 struct dentry
*d_find_alias(struct inode
*inode
)
540 struct dentry
*de
= NULL
;
542 if (!list_empty(&inode
->i_dentry
)) {
543 spin_lock(&dcache_inode_lock
);
544 de
= __d_find_alias(inode
, 0);
545 spin_unlock(&dcache_inode_lock
);
549 EXPORT_SYMBOL(d_find_alias
);
552 * Try to kill dentries associated with this inode.
553 * WARNING: you must own a reference to inode.
555 void d_prune_aliases(struct inode
*inode
)
557 struct dentry
*dentry
;
559 spin_lock(&dcache_inode_lock
);
560 list_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
561 spin_lock(&dentry
->d_lock
);
562 if (!dentry
->d_count
) {
563 __dget_dlock(dentry
);
565 spin_unlock(&dentry
->d_lock
);
566 spin_unlock(&dcache_inode_lock
);
570 spin_unlock(&dentry
->d_lock
);
572 spin_unlock(&dcache_inode_lock
);
574 EXPORT_SYMBOL(d_prune_aliases
);
577 * Try to throw away a dentry - free the inode, dput the parent.
578 * Requires dentry->d_lock is held, and dentry->d_count == 0.
579 * Releases dentry->d_lock.
581 * This may fail if locks cannot be acquired no problem, just try again.
583 static void try_prune_one_dentry(struct dentry
*dentry
)
584 __releases(dentry
->d_lock
)
586 struct dentry
*parent
;
588 parent
= dentry_kill(dentry
, 0);
590 * If dentry_kill returns NULL, we have nothing more to do.
591 * if it returns the same dentry, trylocks failed. In either
592 * case, just loop again.
594 * Otherwise, we need to prune ancestors too. This is necessary
595 * to prevent quadratic behavior of shrink_dcache_parent(), but
596 * is also expected to be beneficial in reducing dentry cache
601 if (parent
== dentry
)
604 /* Prune ancestors. */
607 spin_lock(&dentry
->d_lock
);
608 if (dentry
->d_count
> 1) {
610 spin_unlock(&dentry
->d_lock
);
613 dentry
= dentry_kill(dentry
, 1);
617 static void shrink_dentry_list(struct list_head
*list
)
619 struct dentry
*dentry
;
623 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
624 if (&dentry
->d_lru
== list
)
626 spin_lock(&dentry
->d_lock
);
627 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
628 spin_unlock(&dentry
->d_lock
);
633 * We found an inuse dentry which was not removed from
634 * the LRU because of laziness during lookup. Do not free
635 * it - just keep it off the LRU list.
637 if (dentry
->d_count
) {
638 dentry_lru_del(dentry
);
639 spin_unlock(&dentry
->d_lock
);
645 try_prune_one_dentry(dentry
);
653 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
654 * @sb: superblock to shrink dentry LRU.
655 * @count: number of entries to prune
656 * @flags: flags to control the dentry processing
658 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
660 static void __shrink_dcache_sb(struct super_block
*sb
, int *count
, int flags
)
662 /* called from prune_dcache() and shrink_dcache_parent() */
663 struct dentry
*dentry
;
664 LIST_HEAD(referenced
);
669 spin_lock(&dcache_lru_lock
);
670 while (!list_empty(&sb
->s_dentry_lru
)) {
671 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
672 struct dentry
, d_lru
);
673 BUG_ON(dentry
->d_sb
!= sb
);
675 if (!spin_trylock(&dentry
->d_lock
)) {
676 spin_unlock(&dcache_lru_lock
);
682 * If we are honouring the DCACHE_REFERENCED flag and the
683 * dentry has this flag set, don't free it. Clear the flag
684 * and put it back on the LRU.
686 if (flags
& DCACHE_REFERENCED
&&
687 dentry
->d_flags
& DCACHE_REFERENCED
) {
688 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
689 list_move(&dentry
->d_lru
, &referenced
);
690 spin_unlock(&dentry
->d_lock
);
692 list_move_tail(&dentry
->d_lru
, &tmp
);
693 spin_unlock(&dentry
->d_lock
);
697 cond_resched_lock(&dcache_lru_lock
);
699 if (!list_empty(&referenced
))
700 list_splice(&referenced
, &sb
->s_dentry_lru
);
701 spin_unlock(&dcache_lru_lock
);
703 shrink_dentry_list(&tmp
);
709 * prune_dcache - shrink the dcache
710 * @count: number of entries to try to free
712 * Shrink the dcache. This is done when we need more memory, or simply when we
713 * need to unmount something (at which point we need to unuse all dentries).
715 * This function may fail to free any resources if all the dentries are in use.
717 static void prune_dcache(int count
)
719 struct super_block
*sb
, *p
= NULL
;
721 int unused
= dentry_stat
.nr_unused
;
725 if (unused
== 0 || count
== 0)
730 prune_ratio
= unused
/ count
;
732 list_for_each_entry(sb
, &super_blocks
, s_list
) {
733 if (list_empty(&sb
->s_instances
))
735 if (sb
->s_nr_dentry_unused
== 0)
738 /* Now, we reclaim unused dentrins with fairness.
739 * We reclaim them same percentage from each superblock.
740 * We calculate number of dentries to scan on this sb
741 * as follows, but the implementation is arranged to avoid
743 * number of dentries to scan on this sb =
744 * count * (number of dentries on this sb /
745 * number of dentries in the machine)
747 spin_unlock(&sb_lock
);
748 if (prune_ratio
!= 1)
749 w_count
= (sb
->s_nr_dentry_unused
/ prune_ratio
) + 1;
751 w_count
= sb
->s_nr_dentry_unused
;
754 * We need to be sure this filesystem isn't being unmounted,
755 * otherwise we could race with generic_shutdown_super(), and
756 * end up holding a reference to an inode while the filesystem
757 * is unmounted. So we try to get s_umount, and make sure
760 if (down_read_trylock(&sb
->s_umount
)) {
761 if ((sb
->s_root
!= NULL
) &&
762 (!list_empty(&sb
->s_dentry_lru
))) {
763 __shrink_dcache_sb(sb
, &w_count
,
767 up_read(&sb
->s_umount
);
774 /* more work left to do? */
780 spin_unlock(&sb_lock
);
784 * shrink_dcache_sb - shrink dcache for a superblock
787 * Shrink the dcache for the specified super block. This is used to free
788 * the dcache before unmounting a file system.
790 void shrink_dcache_sb(struct super_block
*sb
)
794 spin_lock(&dcache_lru_lock
);
795 while (!list_empty(&sb
->s_dentry_lru
)) {
796 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
797 spin_unlock(&dcache_lru_lock
);
798 shrink_dentry_list(&tmp
);
799 spin_lock(&dcache_lru_lock
);
801 spin_unlock(&dcache_lru_lock
);
803 EXPORT_SYMBOL(shrink_dcache_sb
);
806 * destroy a single subtree of dentries for unmount
807 * - see the comments on shrink_dcache_for_umount() for a description of the
810 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
812 struct dentry
*parent
;
813 unsigned detached
= 0;
815 BUG_ON(!IS_ROOT(dentry
));
817 /* detach this root from the system */
818 spin_lock(&dentry
->d_lock
);
819 dentry_lru_del(dentry
);
821 spin_unlock(&dentry
->d_lock
);
824 /* descend to the first leaf in the current subtree */
825 while (!list_empty(&dentry
->d_subdirs
)) {
828 /* this is a branch with children - detach all of them
829 * from the system in one go */
830 spin_lock(&dentry
->d_lock
);
831 list_for_each_entry(loop
, &dentry
->d_subdirs
,
833 spin_lock_nested(&loop
->d_lock
,
834 DENTRY_D_LOCK_NESTED
);
835 dentry_lru_del(loop
);
837 spin_unlock(&loop
->d_lock
);
839 spin_unlock(&dentry
->d_lock
);
841 /* move to the first child */
842 dentry
= list_entry(dentry
->d_subdirs
.next
,
843 struct dentry
, d_u
.d_child
);
846 /* consume the dentries from this leaf up through its parents
847 * until we find one with children or run out altogether */
851 if (dentry
->d_count
!= 0) {
853 "BUG: Dentry %p{i=%lx,n=%s}"
855 " [unmount of %s %s]\n",
858 dentry
->d_inode
->i_ino
: 0UL,
861 dentry
->d_sb
->s_type
->name
,
866 if (IS_ROOT(dentry
)) {
868 list_del(&dentry
->d_u
.d_child
);
870 parent
= dentry
->d_parent
;
871 spin_lock(&parent
->d_lock
);
873 list_del(&dentry
->d_u
.d_child
);
874 spin_unlock(&parent
->d_lock
);
879 inode
= dentry
->d_inode
;
881 dentry
->d_inode
= NULL
;
882 list_del_init(&dentry
->d_alias
);
883 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
884 dentry
->d_op
->d_iput(dentry
, inode
);
891 /* finished when we fall off the top of the tree,
892 * otherwise we ascend to the parent and move to the
893 * next sibling if there is one */
897 } while (list_empty(&dentry
->d_subdirs
));
899 dentry
= list_entry(dentry
->d_subdirs
.next
,
900 struct dentry
, d_u
.d_child
);
905 * destroy the dentries attached to a superblock on unmounting
906 * - we don't need to use dentry->d_lock because:
907 * - the superblock is detached from all mountings and open files, so the
908 * dentry trees will not be rearranged by the VFS
909 * - s_umount is write-locked, so the memory pressure shrinker will ignore
910 * any dentries belonging to this superblock that it comes across
911 * - the filesystem itself is no longer permitted to rearrange the dentries
914 void shrink_dcache_for_umount(struct super_block
*sb
)
916 struct dentry
*dentry
;
918 if (down_read_trylock(&sb
->s_umount
))
923 spin_lock(&dentry
->d_lock
);
925 spin_unlock(&dentry
->d_lock
);
926 shrink_dcache_for_umount_subtree(dentry
);
928 while (!hlist_empty(&sb
->s_anon
)) {
929 dentry
= hlist_entry(sb
->s_anon
.first
, struct dentry
, d_hash
);
930 shrink_dcache_for_umount_subtree(dentry
);
935 * Search for at least 1 mount point in the dentry's subdirs.
936 * We descend to the next level whenever the d_subdirs
937 * list is non-empty and continue searching.
941 * have_submounts - check for mounts over a dentry
942 * @parent: dentry to check.
944 * Return true if the parent or its subdirectories contain
947 int have_submounts(struct dentry
*parent
)
949 struct dentry
*this_parent
;
950 struct list_head
*next
;
954 seq
= read_seqbegin(&rename_lock
);
956 this_parent
= parent
;
958 if (d_mountpoint(parent
))
960 spin_lock(&this_parent
->d_lock
);
962 next
= this_parent
->d_subdirs
.next
;
964 while (next
!= &this_parent
->d_subdirs
) {
965 struct list_head
*tmp
= next
;
966 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
969 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
970 /* Have we found a mount point ? */
971 if (d_mountpoint(dentry
)) {
972 spin_unlock(&dentry
->d_lock
);
973 spin_unlock(&this_parent
->d_lock
);
976 if (!list_empty(&dentry
->d_subdirs
)) {
977 spin_unlock(&this_parent
->d_lock
);
978 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
979 this_parent
= dentry
;
980 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
983 spin_unlock(&dentry
->d_lock
);
986 * All done at this level ... ascend and resume the search.
988 if (this_parent
!= parent
) {
990 struct dentry
*child
;
992 tmp
= this_parent
->d_parent
;
994 spin_unlock(&this_parent
->d_lock
);
997 spin_lock(&this_parent
->d_lock
);
998 /* might go back up the wrong parent if we have had a rename
1000 if (this_parent
!= child
->d_parent
||
1001 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1002 spin_unlock(&this_parent
->d_lock
);
1007 next
= child
->d_u
.d_child
.next
;
1010 spin_unlock(&this_parent
->d_lock
);
1011 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1014 write_sequnlock(&rename_lock
);
1015 return 0; /* No mount points found in tree */
1017 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1020 write_sequnlock(&rename_lock
);
1025 write_seqlock(&rename_lock
);
1028 EXPORT_SYMBOL(have_submounts
);
1031 * Search the dentry child list for the specified parent,
1032 * and move any unused dentries to the end of the unused
1033 * list for prune_dcache(). We descend to the next level
1034 * whenever the d_subdirs list is non-empty and continue
1037 * It returns zero iff there are no unused children,
1038 * otherwise it returns the number of children moved to
1039 * the end of the unused list. This may not be the total
1040 * number of unused children, because select_parent can
1041 * drop the lock and return early due to latency
1044 static int select_parent(struct dentry
* parent
)
1046 struct dentry
*this_parent
;
1047 struct list_head
*next
;
1052 seq
= read_seqbegin(&rename_lock
);
1054 this_parent
= parent
;
1055 spin_lock(&this_parent
->d_lock
);
1057 next
= this_parent
->d_subdirs
.next
;
1059 while (next
!= &this_parent
->d_subdirs
) {
1060 struct list_head
*tmp
= next
;
1061 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1064 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1067 * move only zero ref count dentries to the end
1068 * of the unused list for prune_dcache
1070 if (!dentry
->d_count
) {
1071 dentry_lru_move_tail(dentry
);
1074 dentry_lru_del(dentry
);
1078 * We can return to the caller if we have found some (this
1079 * ensures forward progress). We'll be coming back to find
1082 if (found
&& need_resched()) {
1083 spin_unlock(&dentry
->d_lock
);
1088 * Descend a level if the d_subdirs list is non-empty.
1090 if (!list_empty(&dentry
->d_subdirs
)) {
1091 spin_unlock(&this_parent
->d_lock
);
1092 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1093 this_parent
= dentry
;
1094 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1098 spin_unlock(&dentry
->d_lock
);
1101 * All done at this level ... ascend and resume the search.
1103 if (this_parent
!= parent
) {
1105 struct dentry
*child
;
1107 tmp
= this_parent
->d_parent
;
1109 spin_unlock(&this_parent
->d_lock
);
1110 child
= this_parent
;
1112 spin_lock(&this_parent
->d_lock
);
1113 /* might go back up the wrong parent if we have had a rename
1115 if (this_parent
!= child
->d_parent
||
1116 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
1117 spin_unlock(&this_parent
->d_lock
);
1122 next
= child
->d_u
.d_child
.next
;
1126 spin_unlock(&this_parent
->d_lock
);
1127 if (!locked
&& read_seqretry(&rename_lock
, seq
))
1130 write_sequnlock(&rename_lock
);
1137 write_seqlock(&rename_lock
);
1142 * shrink_dcache_parent - prune dcache
1143 * @parent: parent of entries to prune
1145 * Prune the dcache to remove unused children of the parent dentry.
1148 void shrink_dcache_parent(struct dentry
* parent
)
1150 struct super_block
*sb
= parent
->d_sb
;
1153 while ((found
= select_parent(parent
)) != 0)
1154 __shrink_dcache_sb(sb
, &found
, 0);
1156 EXPORT_SYMBOL(shrink_dcache_parent
);
1159 * Scan `nr' dentries and return the number which remain.
1161 * We need to avoid reentering the filesystem if the caller is performing a
1162 * GFP_NOFS allocation attempt. One example deadlock is:
1164 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
1165 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
1166 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
1168 * In this case we return -1 to tell the caller that we baled.
1170 static int shrink_dcache_memory(struct shrinker
*shrink
, int nr
, gfp_t gfp_mask
)
1173 if (!(gfp_mask
& __GFP_FS
))
1178 return (dentry_stat
.nr_unused
/ 100) * sysctl_vfs_cache_pressure
;
1181 static struct shrinker dcache_shrinker
= {
1182 .shrink
= shrink_dcache_memory
,
1183 .seeks
= DEFAULT_SEEKS
,
1187 * d_alloc - allocate a dcache entry
1188 * @parent: parent of entry to allocate
1189 * @name: qstr of the name
1191 * Allocates a dentry. It returns %NULL if there is insufficient memory
1192 * available. On a success the dentry is returned. The name passed in is
1193 * copied and the copy passed in may be reused after this call.
1196 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1198 struct dentry
*dentry
;
1201 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1205 if (name
->len
> DNAME_INLINE_LEN
-1) {
1206 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1208 kmem_cache_free(dentry_cache
, dentry
);
1212 dname
= dentry
->d_iname
;
1214 dentry
->d_name
.name
= dname
;
1216 dentry
->d_name
.len
= name
->len
;
1217 dentry
->d_name
.hash
= name
->hash
;
1218 memcpy(dname
, name
->name
, name
->len
);
1219 dname
[name
->len
] = 0;
1221 dentry
->d_count
= 1;
1222 dentry
->d_flags
= DCACHE_UNHASHED
;
1223 spin_lock_init(&dentry
->d_lock
);
1224 dentry
->d_inode
= NULL
;
1225 dentry
->d_parent
= NULL
;
1226 dentry
->d_sb
= NULL
;
1227 dentry
->d_op
= NULL
;
1228 dentry
->d_fsdata
= NULL
;
1229 dentry
->d_mounted
= 0;
1230 INIT_HLIST_NODE(&dentry
->d_hash
);
1231 INIT_LIST_HEAD(&dentry
->d_lru
);
1232 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1233 INIT_LIST_HEAD(&dentry
->d_alias
);
1234 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1237 spin_lock(&parent
->d_lock
);
1239 * don't need child lock because it is not subject
1240 * to concurrency here
1242 __dget_dlock(parent
);
1243 dentry
->d_parent
= parent
;
1244 dentry
->d_sb
= parent
->d_sb
;
1245 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1246 spin_unlock(&parent
->d_lock
);
1249 this_cpu_inc(nr_dentry
);
1253 EXPORT_SYMBOL(d_alloc
);
1255 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1260 q
.len
= strlen(name
);
1261 q
.hash
= full_name_hash(q
.name
, q
.len
);
1262 return d_alloc(parent
, &q
);
1264 EXPORT_SYMBOL(d_alloc_name
);
1266 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1268 spin_lock(&dentry
->d_lock
);
1270 list_add(&dentry
->d_alias
, &inode
->i_dentry
);
1271 dentry
->d_inode
= inode
;
1272 spin_unlock(&dentry
->d_lock
);
1273 fsnotify_d_instantiate(dentry
, inode
);
1277 * d_instantiate - fill in inode information for a dentry
1278 * @entry: dentry to complete
1279 * @inode: inode to attach to this dentry
1281 * Fill in inode information in the entry.
1283 * This turns negative dentries into productive full members
1286 * NOTE! This assumes that the inode count has been incremented
1287 * (or otherwise set) by the caller to indicate that it is now
1288 * in use by the dcache.
1291 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1293 BUG_ON(!list_empty(&entry
->d_alias
));
1294 spin_lock(&dcache_inode_lock
);
1295 __d_instantiate(entry
, inode
);
1296 spin_unlock(&dcache_inode_lock
);
1297 security_d_instantiate(entry
, inode
);
1299 EXPORT_SYMBOL(d_instantiate
);
1302 * d_instantiate_unique - instantiate a non-aliased dentry
1303 * @entry: dentry to instantiate
1304 * @inode: inode to attach to this dentry
1306 * Fill in inode information in the entry. On success, it returns NULL.
1307 * If an unhashed alias of "entry" already exists, then we return the
1308 * aliased dentry instead and drop one reference to inode.
1310 * Note that in order to avoid conflicts with rename() etc, the caller
1311 * had better be holding the parent directory semaphore.
1313 * This also assumes that the inode count has been incremented
1314 * (or otherwise set) by the caller to indicate that it is now
1315 * in use by the dcache.
1317 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1318 struct inode
*inode
)
1320 struct dentry
*alias
;
1321 int len
= entry
->d_name
.len
;
1322 const char *name
= entry
->d_name
.name
;
1323 unsigned int hash
= entry
->d_name
.hash
;
1326 __d_instantiate(entry
, NULL
);
1330 list_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1331 struct qstr
*qstr
= &alias
->d_name
;
1334 * Don't need alias->d_lock here, because aliases with
1335 * d_parent == entry->d_parent are not subject to name or
1336 * parent changes, because the parent inode i_mutex is held.
1338 if (qstr
->hash
!= hash
)
1340 if (alias
->d_parent
!= entry
->d_parent
)
1342 if (qstr
->len
!= len
)
1344 if (memcmp(qstr
->name
, name
, len
))
1350 __d_instantiate(entry
, inode
);
1354 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1356 struct dentry
*result
;
1358 BUG_ON(!list_empty(&entry
->d_alias
));
1360 spin_lock(&dcache_inode_lock
);
1361 result
= __d_instantiate_unique(entry
, inode
);
1362 spin_unlock(&dcache_inode_lock
);
1365 security_d_instantiate(entry
, inode
);
1369 BUG_ON(!d_unhashed(result
));
1374 EXPORT_SYMBOL(d_instantiate_unique
);
1377 * d_alloc_root - allocate root dentry
1378 * @root_inode: inode to allocate the root for
1380 * Allocate a root ("/") dentry for the inode given. The inode is
1381 * instantiated and returned. %NULL is returned if there is insufficient
1382 * memory or the inode passed is %NULL.
1385 struct dentry
* d_alloc_root(struct inode
* root_inode
)
1387 struct dentry
*res
= NULL
;
1390 static const struct qstr name
= { .name
= "/", .len
= 1 };
1392 res
= d_alloc(NULL
, &name
);
1394 res
->d_sb
= root_inode
->i_sb
;
1395 res
->d_parent
= res
;
1396 d_instantiate(res
, root_inode
);
1401 EXPORT_SYMBOL(d_alloc_root
);
1403 static inline struct hlist_head
*d_hash(struct dentry
*parent
,
1406 hash
+= ((unsigned long) parent
^ GOLDEN_RATIO_PRIME
) / L1_CACHE_BYTES
;
1407 hash
= hash
^ ((hash
^ GOLDEN_RATIO_PRIME
) >> D_HASHBITS
);
1408 return dentry_hashtable
+ (hash
& D_HASHMASK
);
1412 * d_obtain_alias - find or allocate a dentry for a given inode
1413 * @inode: inode to allocate the dentry for
1415 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1416 * similar open by handle operations. The returned dentry may be anonymous,
1417 * or may have a full name (if the inode was already in the cache).
1419 * When called on a directory inode, we must ensure that the inode only ever
1420 * has one dentry. If a dentry is found, that is returned instead of
1421 * allocating a new one.
1423 * On successful return, the reference to the inode has been transferred
1424 * to the dentry. In case of an error the reference on the inode is released.
1425 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1426 * be passed in and will be the error will be propagate to the return value,
1427 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1429 struct dentry
*d_obtain_alias(struct inode
*inode
)
1431 static const struct qstr anonstring
= { .name
= "" };
1436 return ERR_PTR(-ESTALE
);
1438 return ERR_CAST(inode
);
1440 res
= d_find_alias(inode
);
1444 tmp
= d_alloc(NULL
, &anonstring
);
1446 res
= ERR_PTR(-ENOMEM
);
1449 tmp
->d_parent
= tmp
; /* make sure dput doesn't croak */
1452 spin_lock(&dcache_inode_lock
);
1453 res
= __d_find_alias(inode
, 0);
1455 spin_unlock(&dcache_inode_lock
);
1460 /* attach a disconnected dentry */
1461 spin_lock(&tmp
->d_lock
);
1462 tmp
->d_sb
= inode
->i_sb
;
1463 tmp
->d_inode
= inode
;
1464 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1465 tmp
->d_flags
&= ~DCACHE_UNHASHED
;
1466 list_add(&tmp
->d_alias
, &inode
->i_dentry
);
1467 spin_lock(&dcache_hash_lock
);
1468 hlist_add_head(&tmp
->d_hash
, &inode
->i_sb
->s_anon
);
1469 spin_unlock(&dcache_hash_lock
);
1470 spin_unlock(&tmp
->d_lock
);
1471 spin_unlock(&dcache_inode_lock
);
1479 EXPORT_SYMBOL(d_obtain_alias
);
1482 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1483 * @inode: the inode which may have a disconnected dentry
1484 * @dentry: a negative dentry which we want to point to the inode.
1486 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1487 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1488 * and return it, else simply d_add the inode to the dentry and return NULL.
1490 * This is needed in the lookup routine of any filesystem that is exportable
1491 * (via knfsd) so that we can build dcache paths to directories effectively.
1493 * If a dentry was found and moved, then it is returned. Otherwise NULL
1494 * is returned. This matches the expected return value of ->lookup.
1497 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1499 struct dentry
*new = NULL
;
1501 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1502 spin_lock(&dcache_inode_lock
);
1503 new = __d_find_alias(inode
, 1);
1505 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1506 spin_unlock(&dcache_inode_lock
);
1507 security_d_instantiate(new, inode
);
1508 d_move(new, dentry
);
1511 /* already taking dcache_inode_lock, so d_add() by hand */
1512 __d_instantiate(dentry
, inode
);
1513 spin_unlock(&dcache_inode_lock
);
1514 security_d_instantiate(dentry
, inode
);
1518 d_add(dentry
, inode
);
1521 EXPORT_SYMBOL(d_splice_alias
);
1524 * d_add_ci - lookup or allocate new dentry with case-exact name
1525 * @inode: the inode case-insensitive lookup has found
1526 * @dentry: the negative dentry that was passed to the parent's lookup func
1527 * @name: the case-exact name to be associated with the returned dentry
1529 * This is to avoid filling the dcache with case-insensitive names to the
1530 * same inode, only the actual correct case is stored in the dcache for
1531 * case-insensitive filesystems.
1533 * For a case-insensitive lookup match and if the the case-exact dentry
1534 * already exists in in the dcache, use it and return it.
1536 * If no entry exists with the exact case name, allocate new dentry with
1537 * the exact case, and return the spliced entry.
1539 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1543 struct dentry
*found
;
1547 * First check if a dentry matching the name already exists,
1548 * if not go ahead and create it now.
1550 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1552 new = d_alloc(dentry
->d_parent
, name
);
1558 found
= d_splice_alias(inode
, new);
1567 * If a matching dentry exists, and it's not negative use it.
1569 * Decrement the reference count to balance the iget() done
1572 if (found
->d_inode
) {
1573 if (unlikely(found
->d_inode
!= inode
)) {
1574 /* This can't happen because bad inodes are unhashed. */
1575 BUG_ON(!is_bad_inode(inode
));
1576 BUG_ON(!is_bad_inode(found
->d_inode
));
1583 * Negative dentry: instantiate it unless the inode is a directory and
1584 * already has a dentry.
1586 spin_lock(&dcache_inode_lock
);
1587 if (!S_ISDIR(inode
->i_mode
) || list_empty(&inode
->i_dentry
)) {
1588 __d_instantiate(found
, inode
);
1589 spin_unlock(&dcache_inode_lock
);
1590 security_d_instantiate(found
, inode
);
1595 * In case a directory already has a (disconnected) entry grab a
1596 * reference to it, move it in place and use it.
1598 new = list_entry(inode
->i_dentry
.next
, struct dentry
, d_alias
);
1600 spin_unlock(&dcache_inode_lock
);
1601 security_d_instantiate(found
, inode
);
1609 return ERR_PTR(error
);
1611 EXPORT_SYMBOL(d_add_ci
);
1614 * d_lookup - search for a dentry
1615 * @parent: parent dentry
1616 * @name: qstr of name we wish to find
1617 * Returns: dentry, or NULL
1619 * d_lookup searches the children of the parent dentry for the name in
1620 * question. If the dentry is found its reference count is incremented and the
1621 * dentry is returned. The caller must use dput to free the entry when it has
1622 * finished using it. %NULL is returned if the dentry does not exist.
1624 struct dentry
* d_lookup(struct dentry
* parent
, struct qstr
* name
)
1626 struct dentry
* dentry
= NULL
;
1630 seq
= read_seqbegin(&rename_lock
);
1631 dentry
= __d_lookup(parent
, name
);
1634 } while (read_seqretry(&rename_lock
, seq
));
1637 EXPORT_SYMBOL(d_lookup
);
1640 * __d_lookup - search for a dentry (racy)
1641 * @parent: parent dentry
1642 * @name: qstr of name we wish to find
1643 * Returns: dentry, or NULL
1645 * __d_lookup is like d_lookup, however it may (rarely) return a
1646 * false-negative result due to unrelated rename activity.
1648 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1649 * however it must be used carefully, eg. with a following d_lookup in
1650 * the case of failure.
1652 * __d_lookup callers must be commented.
1654 struct dentry
* __d_lookup(struct dentry
* parent
, struct qstr
* name
)
1656 unsigned int len
= name
->len
;
1657 unsigned int hash
= name
->hash
;
1658 const unsigned char *str
= name
->name
;
1659 struct hlist_head
*head
= d_hash(parent
,hash
);
1660 struct dentry
*found
= NULL
;
1661 struct hlist_node
*node
;
1662 struct dentry
*dentry
;
1665 * The hash list is protected using RCU.
1667 * Take d_lock when comparing a candidate dentry, to avoid races
1670 * It is possible that concurrent renames can mess up our list
1671 * walk here and result in missing our dentry, resulting in the
1672 * false-negative result. d_lookup() protects against concurrent
1673 * renames using rename_lock seqlock.
1675 * See Documentation/vfs/dcache-locking.txt for more details.
1679 hlist_for_each_entry_rcu(dentry
, node
, head
, d_hash
) {
1682 if (dentry
->d_name
.hash
!= hash
)
1684 if (dentry
->d_parent
!= parent
)
1687 spin_lock(&dentry
->d_lock
);
1690 * Recheck the dentry after taking the lock - d_move may have
1691 * changed things. Don't bother checking the hash because
1692 * we're about to compare the whole name anyway.
1694 if (dentry
->d_parent
!= parent
)
1697 /* non-existing due to RCU? */
1698 if (d_unhashed(dentry
))
1702 * It is safe to compare names since d_move() cannot
1703 * change the qstr (protected by d_lock).
1705 qstr
= &dentry
->d_name
;
1706 if (parent
->d_op
&& parent
->d_op
->d_compare
) {
1707 if (parent
->d_op
->d_compare(parent
, parent
->d_inode
,
1708 dentry
, dentry
->d_inode
,
1709 qstr
->len
, qstr
->name
, name
))
1712 if (qstr
->len
!= len
)
1714 if (memcmp(qstr
->name
, str
, len
))
1720 spin_unlock(&dentry
->d_lock
);
1723 spin_unlock(&dentry
->d_lock
);
1731 * d_hash_and_lookup - hash the qstr then search for a dentry
1732 * @dir: Directory to search in
1733 * @name: qstr of name we wish to find
1735 * On hash failure or on lookup failure NULL is returned.
1737 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
1739 struct dentry
*dentry
= NULL
;
1742 * Check for a fs-specific hash function. Note that we must
1743 * calculate the standard hash first, as the d_op->d_hash()
1744 * routine may choose to leave the hash value unchanged.
1746 name
->hash
= full_name_hash(name
->name
, name
->len
);
1747 if (dir
->d_op
&& dir
->d_op
->d_hash
) {
1748 if (dir
->d_op
->d_hash(dir
, dir
->d_inode
, name
) < 0)
1751 dentry
= d_lookup(dir
, name
);
1757 * d_validate - verify dentry provided from insecure source (deprecated)
1758 * @dentry: The dentry alleged to be valid child of @dparent
1759 * @dparent: The parent dentry (known to be valid)
1761 * An insecure source has sent us a dentry, here we verify it and dget() it.
1762 * This is used by ncpfs in its readdir implementation.
1763 * Zero is returned in the dentry is invalid.
1765 * This function is slow for big directories, and deprecated, do not use it.
1767 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
1769 struct dentry
*child
;
1771 spin_lock(&dparent
->d_lock
);
1772 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
1773 if (dentry
== child
) {
1774 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1775 __dget_dlock(dentry
);
1776 spin_unlock(&dentry
->d_lock
);
1777 spin_unlock(&dparent
->d_lock
);
1781 spin_unlock(&dparent
->d_lock
);
1785 EXPORT_SYMBOL(d_validate
);
1788 * When a file is deleted, we have two options:
1789 * - turn this dentry into a negative dentry
1790 * - unhash this dentry and free it.
1792 * Usually, we want to just turn this into
1793 * a negative dentry, but if anybody else is
1794 * currently using the dentry or the inode
1795 * we can't do that and we fall back on removing
1796 * it from the hash queues and waiting for
1797 * it to be deleted later when it has no users
1801 * d_delete - delete a dentry
1802 * @dentry: The dentry to delete
1804 * Turn the dentry into a negative dentry if possible, otherwise
1805 * remove it from the hash queues so it can be deleted later
1808 void d_delete(struct dentry
* dentry
)
1812 * Are we the only user?
1815 spin_lock(&dentry
->d_lock
);
1816 isdir
= S_ISDIR(dentry
->d_inode
->i_mode
);
1817 if (dentry
->d_count
== 1) {
1818 if (!spin_trylock(&dcache_inode_lock
)) {
1819 spin_unlock(&dentry
->d_lock
);
1823 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
1824 dentry_iput(dentry
);
1825 fsnotify_nameremove(dentry
, isdir
);
1829 if (!d_unhashed(dentry
))
1832 spin_unlock(&dentry
->d_lock
);
1834 fsnotify_nameremove(dentry
, isdir
);
1836 EXPORT_SYMBOL(d_delete
);
1838 static void __d_rehash(struct dentry
* entry
, struct hlist_head
*list
)
1841 entry
->d_flags
&= ~DCACHE_UNHASHED
;
1842 hlist_add_head_rcu(&entry
->d_hash
, list
);
1845 static void _d_rehash(struct dentry
* entry
)
1847 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
1851 * d_rehash - add an entry back to the hash
1852 * @entry: dentry to add to the hash
1854 * Adds a dentry to the hash according to its name.
1857 void d_rehash(struct dentry
* entry
)
1859 spin_lock(&entry
->d_lock
);
1860 spin_lock(&dcache_hash_lock
);
1862 spin_unlock(&dcache_hash_lock
);
1863 spin_unlock(&entry
->d_lock
);
1865 EXPORT_SYMBOL(d_rehash
);
1868 * dentry_update_name_case - update case insensitive dentry with a new name
1869 * @dentry: dentry to be updated
1872 * Update a case insensitive dentry with new case of name.
1874 * dentry must have been returned by d_lookup with name @name. Old and new
1875 * name lengths must match (ie. no d_compare which allows mismatched name
1878 * Parent inode i_mutex must be held over d_lookup and into this call (to
1879 * keep renames and concurrent inserts, and readdir(2) away).
1881 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
1883 BUG_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
1884 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
1886 spin_lock(&dentry
->d_lock
);
1887 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
1888 spin_unlock(&dentry
->d_lock
);
1890 EXPORT_SYMBOL(dentry_update_name_case
);
1892 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
1894 if (dname_external(target
)) {
1895 if (dname_external(dentry
)) {
1897 * Both external: swap the pointers
1899 swap(target
->d_name
.name
, dentry
->d_name
.name
);
1902 * dentry:internal, target:external. Steal target's
1903 * storage and make target internal.
1905 memcpy(target
->d_iname
, dentry
->d_name
.name
,
1906 dentry
->d_name
.len
+ 1);
1907 dentry
->d_name
.name
= target
->d_name
.name
;
1908 target
->d_name
.name
= target
->d_iname
;
1911 if (dname_external(dentry
)) {
1913 * dentry:external, target:internal. Give dentry's
1914 * storage to target and make dentry internal
1916 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1917 target
->d_name
.len
+ 1);
1918 target
->d_name
.name
= dentry
->d_name
.name
;
1919 dentry
->d_name
.name
= dentry
->d_iname
;
1922 * Both are internal. Just copy target to dentry
1924 memcpy(dentry
->d_iname
, target
->d_name
.name
,
1925 target
->d_name
.len
+ 1);
1926 dentry
->d_name
.len
= target
->d_name
.len
;
1930 swap(dentry
->d_name
.len
, target
->d_name
.len
);
1933 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
1936 * XXXX: do we really need to take target->d_lock?
1938 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
1939 spin_lock(&target
->d_parent
->d_lock
);
1941 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
1942 spin_lock(&dentry
->d_parent
->d_lock
);
1943 spin_lock_nested(&target
->d_parent
->d_lock
,
1944 DENTRY_D_LOCK_NESTED
);
1946 spin_lock(&target
->d_parent
->d_lock
);
1947 spin_lock_nested(&dentry
->d_parent
->d_lock
,
1948 DENTRY_D_LOCK_NESTED
);
1951 if (target
< dentry
) {
1952 spin_lock_nested(&target
->d_lock
, 2);
1953 spin_lock_nested(&dentry
->d_lock
, 3);
1955 spin_lock_nested(&dentry
->d_lock
, 2);
1956 spin_lock_nested(&target
->d_lock
, 3);
1960 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
1961 struct dentry
*target
)
1963 if (target
->d_parent
!= dentry
->d_parent
)
1964 spin_unlock(&dentry
->d_parent
->d_lock
);
1965 if (target
->d_parent
!= target
)
1966 spin_unlock(&target
->d_parent
->d_lock
);
1970 * When switching names, the actual string doesn't strictly have to
1971 * be preserved in the target - because we're dropping the target
1972 * anyway. As such, we can just do a simple memcpy() to copy over
1973 * the new name before we switch.
1975 * Note that we have to be a lot more careful about getting the hash
1976 * switched - we have to switch the hash value properly even if it
1977 * then no longer matches the actual (corrupted) string of the target.
1978 * The hash value has to match the hash queue that the dentry is on..
1981 * d_move - move a dentry
1982 * @dentry: entry to move
1983 * @target: new dentry
1985 * Update the dcache to reflect the move of a file name. Negative
1986 * dcache entries should not be moved in this way.
1988 void d_move(struct dentry
* dentry
, struct dentry
* target
)
1990 if (!dentry
->d_inode
)
1991 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
1993 BUG_ON(d_ancestor(dentry
, target
));
1994 BUG_ON(d_ancestor(target
, dentry
));
1996 write_seqlock(&rename_lock
);
1998 dentry_lock_for_move(dentry
, target
);
2000 /* Move the dentry to the target hash queue, if on different bucket */
2001 spin_lock(&dcache_hash_lock
);
2002 if (!d_unhashed(dentry
))
2003 hlist_del_rcu(&dentry
->d_hash
);
2004 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2005 spin_unlock(&dcache_hash_lock
);
2007 /* Unhash the target: dput() will then get rid of it */
2010 list_del(&dentry
->d_u
.d_child
);
2011 list_del(&target
->d_u
.d_child
);
2013 /* Switch the names.. */
2014 switch_names(dentry
, target
);
2015 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2017 /* ... and switch the parents */
2018 if (IS_ROOT(dentry
)) {
2019 dentry
->d_parent
= target
->d_parent
;
2020 target
->d_parent
= target
;
2021 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2023 swap(dentry
->d_parent
, target
->d_parent
);
2025 /* And add them back to the (new) parent lists */
2026 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2029 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2031 dentry_unlock_parents_for_move(dentry
, target
);
2032 spin_unlock(&target
->d_lock
);
2033 fsnotify_d_move(dentry
);
2034 spin_unlock(&dentry
->d_lock
);
2035 write_sequnlock(&rename_lock
);
2037 EXPORT_SYMBOL(d_move
);
2040 * d_ancestor - search for an ancestor
2041 * @p1: ancestor dentry
2044 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2045 * an ancestor of p2, else NULL.
2047 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2051 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2052 if (p
->d_parent
== p1
)
2059 * This helper attempts to cope with remotely renamed directories
2061 * It assumes that the caller is already holding
2062 * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock
2064 * Note: If ever the locking in lock_rename() changes, then please
2065 * remember to update this too...
2067 static struct dentry
*__d_unalias(struct dentry
*dentry
, struct dentry
*alias
)
2068 __releases(dcache_inode_lock
)
2070 struct mutex
*m1
= NULL
, *m2
= NULL
;
2073 /* If alias and dentry share a parent, then no extra locks required */
2074 if (alias
->d_parent
== dentry
->d_parent
)
2077 /* Check for loops */
2078 ret
= ERR_PTR(-ELOOP
);
2079 if (d_ancestor(alias
, dentry
))
2082 /* See lock_rename() */
2083 ret
= ERR_PTR(-EBUSY
);
2084 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2086 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2087 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2089 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2091 d_move(alias
, dentry
);
2094 spin_unlock(&dcache_inode_lock
);
2103 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2104 * named dentry in place of the dentry to be replaced.
2105 * returns with anon->d_lock held!
2107 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2109 struct dentry
*dparent
, *aparent
;
2111 dentry_lock_for_move(anon
, dentry
);
2113 dparent
= dentry
->d_parent
;
2114 aparent
= anon
->d_parent
;
2116 switch_names(dentry
, anon
);
2117 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2119 dentry
->d_parent
= (aparent
== anon
) ? dentry
: aparent
;
2120 list_del(&dentry
->d_u
.d_child
);
2121 if (!IS_ROOT(dentry
))
2122 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2124 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
2126 anon
->d_parent
= (dparent
== dentry
) ? anon
: dparent
;
2127 list_del(&anon
->d_u
.d_child
);
2129 list_add(&anon
->d_u
.d_child
, &anon
->d_parent
->d_subdirs
);
2131 INIT_LIST_HEAD(&anon
->d_u
.d_child
);
2133 dentry_unlock_parents_for_move(anon
, dentry
);
2134 spin_unlock(&dentry
->d_lock
);
2136 /* anon->d_lock still locked, returns locked */
2137 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2141 * d_materialise_unique - introduce an inode into the tree
2142 * @dentry: candidate dentry
2143 * @inode: inode to bind to the dentry, to which aliases may be attached
2145 * Introduces an dentry into the tree, substituting an extant disconnected
2146 * root directory alias in its place if there is one
2148 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2150 struct dentry
*actual
;
2152 BUG_ON(!d_unhashed(dentry
));
2156 __d_instantiate(dentry
, NULL
);
2161 spin_lock(&dcache_inode_lock
);
2163 if (S_ISDIR(inode
->i_mode
)) {
2164 struct dentry
*alias
;
2166 /* Does an aliased dentry already exist? */
2167 alias
= __d_find_alias(inode
, 0);
2170 /* Is this an anonymous mountpoint that we could splice
2172 if (IS_ROOT(alias
)) {
2173 __d_materialise_dentry(dentry
, alias
);
2177 /* Nope, but we must(!) avoid directory aliasing */
2178 actual
= __d_unalias(dentry
, alias
);
2185 /* Add a unique reference */
2186 actual
= __d_instantiate_unique(dentry
, inode
);
2190 BUG_ON(!d_unhashed(actual
));
2192 spin_lock(&actual
->d_lock
);
2194 spin_lock(&dcache_hash_lock
);
2196 spin_unlock(&dcache_hash_lock
);
2197 spin_unlock(&actual
->d_lock
);
2198 spin_unlock(&dcache_inode_lock
);
2200 if (actual
== dentry
) {
2201 security_d_instantiate(dentry
, inode
);
2208 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2210 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2214 return -ENAMETOOLONG
;
2216 memcpy(*buffer
, str
, namelen
);
2220 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2222 return prepend(buffer
, buflen
, name
->name
, name
->len
);
2226 * Prepend path string to a buffer
2228 * @path: the dentry/vfsmount to report
2229 * @root: root vfsmnt/dentry (may be modified by this function)
2230 * @buffer: pointer to the end of the buffer
2231 * @buflen: pointer to buffer length
2233 * Caller holds the rename_lock.
2235 * If path is not reachable from the supplied root, then the value of
2236 * root is changed (without modifying refcounts).
2238 static int prepend_path(const struct path
*path
, struct path
*root
,
2239 char **buffer
, int *buflen
)
2241 struct dentry
*dentry
= path
->dentry
;
2242 struct vfsmount
*vfsmnt
= path
->mnt
;
2246 br_read_lock(vfsmount_lock
);
2247 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2248 struct dentry
* parent
;
2250 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2252 if (vfsmnt
->mnt_parent
== vfsmnt
) {
2255 dentry
= vfsmnt
->mnt_mountpoint
;
2256 vfsmnt
= vfsmnt
->mnt_parent
;
2259 parent
= dentry
->d_parent
;
2261 spin_lock(&dentry
->d_lock
);
2262 error
= prepend_name(buffer
, buflen
, &dentry
->d_name
);
2263 spin_unlock(&dentry
->d_lock
);
2265 error
= prepend(buffer
, buflen
, "/", 1);
2274 if (!error
&& !slash
)
2275 error
= prepend(buffer
, buflen
, "/", 1);
2277 br_read_unlock(vfsmount_lock
);
2282 * Filesystems needing to implement special "root names"
2283 * should do so with ->d_dname()
2285 if (IS_ROOT(dentry
) &&
2286 (dentry
->d_name
.len
!= 1 || dentry
->d_name
.name
[0] != '/')) {
2287 WARN(1, "Root dentry has weird name <%.*s>\n",
2288 (int) dentry
->d_name
.len
, dentry
->d_name
.name
);
2291 root
->dentry
= dentry
;
2296 * __d_path - return the path of a dentry
2297 * @path: the dentry/vfsmount to report
2298 * @root: root vfsmnt/dentry (may be modified by this function)
2299 * @buf: buffer to return value in
2300 * @buflen: buffer length
2302 * Convert a dentry into an ASCII path name.
2304 * Returns a pointer into the buffer or an error code if the
2305 * path was too long.
2307 * "buflen" should be positive.
2309 * If path is not reachable from the supplied root, then the value of
2310 * root is changed (without modifying refcounts).
2312 char *__d_path(const struct path
*path
, struct path
*root
,
2313 char *buf
, int buflen
)
2315 char *res
= buf
+ buflen
;
2318 prepend(&res
, &buflen
, "\0", 1);
2319 write_seqlock(&rename_lock
);
2320 error
= prepend_path(path
, root
, &res
, &buflen
);
2321 write_sequnlock(&rename_lock
);
2324 return ERR_PTR(error
);
2329 * same as __d_path but appends "(deleted)" for unlinked files.
2331 static int path_with_deleted(const struct path
*path
, struct path
*root
,
2332 char **buf
, int *buflen
)
2334 prepend(buf
, buflen
, "\0", 1);
2335 if (d_unlinked(path
->dentry
)) {
2336 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2341 return prepend_path(path
, root
, buf
, buflen
);
2344 static int prepend_unreachable(char **buffer
, int *buflen
)
2346 return prepend(buffer
, buflen
, "(unreachable)", 13);
2350 * d_path - return the path of a dentry
2351 * @path: path to report
2352 * @buf: buffer to return value in
2353 * @buflen: buffer length
2355 * Convert a dentry into an ASCII path name. If the entry has been deleted
2356 * the string " (deleted)" is appended. Note that this is ambiguous.
2358 * Returns a pointer into the buffer or an error code if the path was
2359 * too long. Note: Callers should use the returned pointer, not the passed
2360 * in buffer, to use the name! The implementation often starts at an offset
2361 * into the buffer, and may leave 0 bytes at the start.
2363 * "buflen" should be positive.
2365 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2367 char *res
= buf
+ buflen
;
2373 * We have various synthetic filesystems that never get mounted. On
2374 * these filesystems dentries are never used for lookup purposes, and
2375 * thus don't need to be hashed. They also don't need a name until a
2376 * user wants to identify the object in /proc/pid/fd/. The little hack
2377 * below allows us to generate a name for these objects on demand:
2379 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2380 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2382 get_fs_root(current
->fs
, &root
);
2383 write_seqlock(&rename_lock
);
2385 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2387 res
= ERR_PTR(error
);
2388 write_sequnlock(&rename_lock
);
2392 EXPORT_SYMBOL(d_path
);
2395 * d_path_with_unreachable - return the path of a dentry
2396 * @path: path to report
2397 * @buf: buffer to return value in
2398 * @buflen: buffer length
2400 * The difference from d_path() is that this prepends "(unreachable)"
2401 * to paths which are unreachable from the current process' root.
2403 char *d_path_with_unreachable(const struct path
*path
, char *buf
, int buflen
)
2405 char *res
= buf
+ buflen
;
2410 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2411 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2413 get_fs_root(current
->fs
, &root
);
2414 write_seqlock(&rename_lock
);
2416 error
= path_with_deleted(path
, &tmp
, &res
, &buflen
);
2417 if (!error
&& !path_equal(&tmp
, &root
))
2418 error
= prepend_unreachable(&res
, &buflen
);
2419 write_sequnlock(&rename_lock
);
2422 res
= ERR_PTR(error
);
2428 * Helper function for dentry_operations.d_dname() members
2430 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2431 const char *fmt
, ...)
2437 va_start(args
, fmt
);
2438 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2441 if (sz
> sizeof(temp
) || sz
> buflen
)
2442 return ERR_PTR(-ENAMETOOLONG
);
2444 buffer
+= buflen
- sz
;
2445 return memcpy(buffer
, temp
, sz
);
2449 * Write full pathname from the root of the filesystem into the buffer.
2451 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2453 char *end
= buf
+ buflen
;
2456 prepend(&end
, &buflen
, "\0", 1);
2463 while (!IS_ROOT(dentry
)) {
2464 struct dentry
*parent
= dentry
->d_parent
;
2468 spin_lock(&dentry
->d_lock
);
2469 error
= prepend_name(&end
, &buflen
, &dentry
->d_name
);
2470 spin_unlock(&dentry
->d_lock
);
2471 if (error
!= 0 || prepend(&end
, &buflen
, "/", 1) != 0)
2479 return ERR_PTR(-ENAMETOOLONG
);
2482 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
2486 write_seqlock(&rename_lock
);
2487 retval
= __dentry_path(dentry
, buf
, buflen
);
2488 write_sequnlock(&rename_lock
);
2492 EXPORT_SYMBOL(dentry_path_raw
);
2494 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2499 write_seqlock(&rename_lock
);
2500 if (d_unlinked(dentry
)) {
2502 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
2506 retval
= __dentry_path(dentry
, buf
, buflen
);
2507 write_sequnlock(&rename_lock
);
2508 if (!IS_ERR(retval
) && p
)
2509 *p
= '/'; /* restore '/' overriden with '\0' */
2512 return ERR_PTR(-ENAMETOOLONG
);
2516 * NOTE! The user-level library version returns a
2517 * character pointer. The kernel system call just
2518 * returns the length of the buffer filled (which
2519 * includes the ending '\0' character), or a negative
2520 * error value. So libc would do something like
2522 * char *getcwd(char * buf, size_t size)
2526 * retval = sys_getcwd(buf, size);
2533 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
2536 struct path pwd
, root
;
2537 char *page
= (char *) __get_free_page(GFP_USER
);
2542 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
2545 write_seqlock(&rename_lock
);
2546 if (!d_unlinked(pwd
.dentry
)) {
2548 struct path tmp
= root
;
2549 char *cwd
= page
+ PAGE_SIZE
;
2550 int buflen
= PAGE_SIZE
;
2552 prepend(&cwd
, &buflen
, "\0", 1);
2553 error
= prepend_path(&pwd
, &tmp
, &cwd
, &buflen
);
2554 write_sequnlock(&rename_lock
);
2559 /* Unreachable from current root */
2560 if (!path_equal(&tmp
, &root
)) {
2561 error
= prepend_unreachable(&cwd
, &buflen
);
2567 len
= PAGE_SIZE
+ page
- cwd
;
2570 if (copy_to_user(buf
, cwd
, len
))
2574 write_sequnlock(&rename_lock
);
2580 free_page((unsigned long) page
);
2585 * Test whether new_dentry is a subdirectory of old_dentry.
2587 * Trivially implemented using the dcache structure
2591 * is_subdir - is new dentry a subdirectory of old_dentry
2592 * @new_dentry: new dentry
2593 * @old_dentry: old dentry
2595 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2596 * Returns 0 otherwise.
2597 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2600 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
2605 if (new_dentry
== old_dentry
)
2609 /* for restarting inner loop in case of seq retry */
2610 seq
= read_seqbegin(&rename_lock
);
2612 * Need rcu_readlock to protect against the d_parent trashing
2616 if (d_ancestor(old_dentry
, new_dentry
))
2621 } while (read_seqretry(&rename_lock
, seq
));
2626 int path_is_under(struct path
*path1
, struct path
*path2
)
2628 struct vfsmount
*mnt
= path1
->mnt
;
2629 struct dentry
*dentry
= path1
->dentry
;
2632 br_read_lock(vfsmount_lock
);
2633 if (mnt
!= path2
->mnt
) {
2635 if (mnt
->mnt_parent
== mnt
) {
2636 br_read_unlock(vfsmount_lock
);
2639 if (mnt
->mnt_parent
== path2
->mnt
)
2641 mnt
= mnt
->mnt_parent
;
2643 dentry
= mnt
->mnt_mountpoint
;
2645 res
= is_subdir(dentry
, path2
->dentry
);
2646 br_read_unlock(vfsmount_lock
);
2649 EXPORT_SYMBOL(path_is_under
);
2651 void d_genocide(struct dentry
*root
)
2653 struct dentry
*this_parent
;
2654 struct list_head
*next
;
2658 seq
= read_seqbegin(&rename_lock
);
2661 spin_lock(&this_parent
->d_lock
);
2663 next
= this_parent
->d_subdirs
.next
;
2665 while (next
!= &this_parent
->d_subdirs
) {
2666 struct list_head
*tmp
= next
;
2667 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
2670 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2671 if (d_unhashed(dentry
) || !dentry
->d_inode
) {
2672 spin_unlock(&dentry
->d_lock
);
2675 if (!list_empty(&dentry
->d_subdirs
)) {
2676 spin_unlock(&this_parent
->d_lock
);
2677 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
2678 this_parent
= dentry
;
2679 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
2682 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
2683 dentry
->d_flags
|= DCACHE_GENOCIDE
;
2686 spin_unlock(&dentry
->d_lock
);
2688 if (this_parent
!= root
) {
2690 struct dentry
*child
;
2692 tmp
= this_parent
->d_parent
;
2693 if (!(this_parent
->d_flags
& DCACHE_GENOCIDE
)) {
2694 this_parent
->d_flags
|= DCACHE_GENOCIDE
;
2695 this_parent
->d_count
--;
2698 spin_unlock(&this_parent
->d_lock
);
2699 child
= this_parent
;
2701 spin_lock(&this_parent
->d_lock
);
2702 /* might go back up the wrong parent if we have had a rename
2704 if (this_parent
!= child
->d_parent
||
2705 (!locked
&& read_seqretry(&rename_lock
, seq
))) {
2706 spin_unlock(&this_parent
->d_lock
);
2711 next
= child
->d_u
.d_child
.next
;
2714 spin_unlock(&this_parent
->d_lock
);
2715 if (!locked
&& read_seqretry(&rename_lock
, seq
))
2718 write_sequnlock(&rename_lock
);
2723 write_seqlock(&rename_lock
);
2728 * find_inode_number - check for dentry with name
2729 * @dir: directory to check
2730 * @name: Name to find.
2732 * Check whether a dentry already exists for the given name,
2733 * and return the inode number if it has an inode. Otherwise
2736 * This routine is used to post-process directory listings for
2737 * filesystems using synthetic inode numbers, and is necessary
2738 * to keep getcwd() working.
2741 ino_t
find_inode_number(struct dentry
*dir
, struct qstr
*name
)
2743 struct dentry
* dentry
;
2746 dentry
= d_hash_and_lookup(dir
, name
);
2748 if (dentry
->d_inode
)
2749 ino
= dentry
->d_inode
->i_ino
;
2754 EXPORT_SYMBOL(find_inode_number
);
2756 static __initdata
unsigned long dhash_entries
;
2757 static int __init
set_dhash_entries(char *str
)
2761 dhash_entries
= simple_strtoul(str
, &str
, 0);
2764 __setup("dhash_entries=", set_dhash_entries
);
2766 static void __init
dcache_init_early(void)
2770 /* If hashes are distributed across NUMA nodes, defer
2771 * hash allocation until vmalloc space is available.
2777 alloc_large_system_hash("Dentry cache",
2778 sizeof(struct hlist_head
),
2786 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2787 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2790 static void __init
dcache_init(void)
2795 * A constructor could be added for stable state like the lists,
2796 * but it is probably not worth it because of the cache nature
2799 dentry_cache
= KMEM_CACHE(dentry
,
2800 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
2802 register_shrinker(&dcache_shrinker
);
2804 /* Hash may have been set up in dcache_init_early */
2809 alloc_large_system_hash("Dentry cache",
2810 sizeof(struct hlist_head
),
2818 for (loop
= 0; loop
< (1 << d_hash_shift
); loop
++)
2819 INIT_HLIST_HEAD(&dentry_hashtable
[loop
]);
2822 /* SLAB cache for __getname() consumers */
2823 struct kmem_cache
*names_cachep __read_mostly
;
2824 EXPORT_SYMBOL(names_cachep
);
2826 EXPORT_SYMBOL(d_genocide
);
2828 void __init
vfs_caches_init_early(void)
2830 dcache_init_early();
2834 void __init
vfs_caches_init(unsigned long mempages
)
2836 unsigned long reserve
;
2838 /* Base hash sizes on available memory, with a reserve equal to
2839 150% of current kernel size */
2841 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
2842 mempages
-= reserve
;
2844 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
2845 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2849 files_init(mempages
);