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 * read_seqbegin_or_lock - begin a sequence number check or locking block
94 * seq : sequence number to be checked
96 * First try it once optimistically without taking the lock. If that fails,
97 * take the lock. The sequence number is also used as a marker for deciding
98 * whether to be a reader (even) or writer (odd).
99 * N.B. seq must be initialized to an even number to begin with.
101 static inline void read_seqbegin_or_lock(seqlock_t
*lock
, int *seq
)
103 if (!(*seq
& 1)) /* Even */
104 *seq
= read_seqbegin(lock
);
109 static inline int need_seqretry(seqlock_t
*lock
, int seq
)
111 return !(seq
& 1) && read_seqretry(lock
, seq
);
114 static inline void done_seqretry(seqlock_t
*lock
, int seq
)
117 write_sequnlock(lock
);
121 * This is the single most critical data structure when it comes
122 * to the dcache: the hashtable for lookups. Somebody should try
123 * to make this good - I've just made it work.
125 * This hash-function tries to avoid losing too many bits of hash
126 * information, yet avoid using a prime hash-size or similar.
128 #define D_HASHBITS d_hash_shift
129 #define D_HASHMASK d_hash_mask
131 static unsigned int d_hash_mask __read_mostly
;
132 static unsigned int d_hash_shift __read_mostly
;
134 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
136 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
139 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
140 hash
= hash
+ (hash
>> D_HASHBITS
);
141 return dentry_hashtable
+ (hash
& D_HASHMASK
);
144 /* Statistics gathering. */
145 struct dentry_stat_t dentry_stat
= {
149 static DEFINE_PER_CPU(long, nr_dentry
);
150 static DEFINE_PER_CPU(long, nr_dentry_unused
);
152 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
155 * Here we resort to our own counters instead of using generic per-cpu counters
156 * for consistency with what the vfs inode code does. We are expected to harvest
157 * better code and performance by having our own specialized counters.
159 * Please note that the loop is done over all possible CPUs, not over all online
160 * CPUs. The reason for this is that we don't want to play games with CPUs going
161 * on and off. If one of them goes off, we will just keep their counters.
163 * glommer: See cffbc8a for details, and if you ever intend to change this,
164 * please update all vfs counters to match.
166 static long get_nr_dentry(void)
170 for_each_possible_cpu(i
)
171 sum
+= per_cpu(nr_dentry
, i
);
172 return sum
< 0 ? 0 : sum
;
175 static long get_nr_dentry_unused(void)
179 for_each_possible_cpu(i
)
180 sum
+= per_cpu(nr_dentry_unused
, i
);
181 return sum
< 0 ? 0 : sum
;
184 int proc_nr_dentry(ctl_table
*table
, int write
, void __user
*buffer
,
185 size_t *lenp
, loff_t
*ppos
)
187 dentry_stat
.nr_dentry
= get_nr_dentry();
188 dentry_stat
.nr_unused
= get_nr_dentry_unused();
189 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
194 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
195 * The strings are both count bytes long, and count is non-zero.
197 #ifdef CONFIG_DCACHE_WORD_ACCESS
199 #include <asm/word-at-a-time.h>
201 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
202 * aligned allocation for this particular component. We don't
203 * strictly need the load_unaligned_zeropad() safety, but it
204 * doesn't hurt either.
206 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
207 * need the careful unaligned handling.
209 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
211 unsigned long a
,b
,mask
;
214 a
= *(unsigned long *)cs
;
215 b
= load_unaligned_zeropad(ct
);
216 if (tcount
< sizeof(unsigned long))
218 if (unlikely(a
!= b
))
220 cs
+= sizeof(unsigned long);
221 ct
+= sizeof(unsigned long);
222 tcount
-= sizeof(unsigned long);
226 mask
= ~(~0ul << tcount
*8);
227 return unlikely(!!((a
^ b
) & mask
));
232 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
246 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
248 const unsigned char *cs
;
250 * Be careful about RCU walk racing with rename:
251 * use ACCESS_ONCE to fetch the name pointer.
253 * NOTE! Even if a rename will mean that the length
254 * was not loaded atomically, we don't care. The
255 * RCU walk will check the sequence count eventually,
256 * and catch it. And we won't overrun the buffer,
257 * because we're reading the name pointer atomically,
258 * and a dentry name is guaranteed to be properly
259 * terminated with a NUL byte.
261 * End result: even if 'len' is wrong, we'll exit
262 * early because the data cannot match (there can
263 * be no NUL in the ct/tcount data)
265 cs
= ACCESS_ONCE(dentry
->d_name
.name
);
266 smp_read_barrier_depends();
267 return dentry_string_cmp(cs
, ct
, tcount
);
270 static void __d_free(struct rcu_head
*head
)
272 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
274 WARN_ON(!hlist_unhashed(&dentry
->d_alias
));
275 if (dname_external(dentry
))
276 kfree(dentry
->d_name
.name
);
277 kmem_cache_free(dentry_cache
, dentry
);
283 static void d_free(struct dentry
*dentry
)
285 BUG_ON((int)dentry
->d_lockref
.count
> 0);
286 this_cpu_dec(nr_dentry
);
287 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
288 dentry
->d_op
->d_release(dentry
);
290 /* if dentry was never visible to RCU, immediate free is OK */
291 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
292 __d_free(&dentry
->d_u
.d_rcu
);
294 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
298 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
299 * @dentry: the target dentry
300 * After this call, in-progress rcu-walk path lookup will fail. This
301 * should be called after unhashing, and after changing d_inode (if
302 * the dentry has not already been unhashed).
304 static inline void dentry_rcuwalk_barrier(struct dentry
*dentry
)
306 assert_spin_locked(&dentry
->d_lock
);
307 /* Go through a barrier */
308 write_seqcount_barrier(&dentry
->d_seq
);
312 * Release the dentry's inode, using the filesystem
313 * d_iput() operation if defined. Dentry has no refcount
316 static void dentry_iput(struct dentry
* dentry
)
317 __releases(dentry
->d_lock
)
318 __releases(dentry
->d_inode
->i_lock
)
320 struct inode
*inode
= dentry
->d_inode
;
322 dentry
->d_inode
= NULL
;
323 hlist_del_init(&dentry
->d_alias
);
324 spin_unlock(&dentry
->d_lock
);
325 spin_unlock(&inode
->i_lock
);
327 fsnotify_inoderemove(inode
);
328 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
329 dentry
->d_op
->d_iput(dentry
, inode
);
333 spin_unlock(&dentry
->d_lock
);
338 * Release the dentry's inode, using the filesystem
339 * d_iput() operation if defined. dentry remains in-use.
341 static void dentry_unlink_inode(struct dentry
* dentry
)
342 __releases(dentry
->d_lock
)
343 __releases(dentry
->d_inode
->i_lock
)
345 struct inode
*inode
= dentry
->d_inode
;
346 dentry
->d_inode
= NULL
;
347 hlist_del_init(&dentry
->d_alias
);
348 dentry_rcuwalk_barrier(dentry
);
349 spin_unlock(&dentry
->d_lock
);
350 spin_unlock(&inode
->i_lock
);
352 fsnotify_inoderemove(inode
);
353 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
354 dentry
->d_op
->d_iput(dentry
, inode
);
360 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
362 static void dentry_lru_add(struct dentry
*dentry
)
364 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
))) {
365 spin_lock(&dcache_lru_lock
);
366 dentry
->d_flags
|= DCACHE_LRU_LIST
;
367 list_add(&dentry
->d_lru
, &dentry
->d_sb
->s_dentry_lru
);
368 dentry
->d_sb
->s_nr_dentry_unused
++;
369 this_cpu_inc(nr_dentry_unused
);
370 spin_unlock(&dcache_lru_lock
);
374 static void __dentry_lru_del(struct dentry
*dentry
)
376 list_del_init(&dentry
->d_lru
);
377 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
378 dentry
->d_sb
->s_nr_dentry_unused
--;
379 this_cpu_dec(nr_dentry_unused
);
383 * Remove a dentry with references from the LRU.
385 static void dentry_lru_del(struct dentry
*dentry
)
387 if (!list_empty(&dentry
->d_lru
)) {
388 spin_lock(&dcache_lru_lock
);
389 __dentry_lru_del(dentry
);
390 spin_unlock(&dcache_lru_lock
);
394 static void dentry_lru_move_list(struct dentry
*dentry
, struct list_head
*list
)
396 spin_lock(&dcache_lru_lock
);
397 if (list_empty(&dentry
->d_lru
)) {
398 dentry
->d_flags
|= DCACHE_LRU_LIST
;
399 list_add_tail(&dentry
->d_lru
, list
);
400 dentry
->d_sb
->s_nr_dentry_unused
++;
401 this_cpu_inc(nr_dentry_unused
);
403 list_move_tail(&dentry
->d_lru
, list
);
405 spin_unlock(&dcache_lru_lock
);
409 * d_kill - kill dentry and return parent
410 * @dentry: dentry to kill
411 * @parent: parent dentry
413 * The dentry must already be unhashed and removed from the LRU.
415 * If this is the root of the dentry tree, return NULL.
417 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
420 static struct dentry
*d_kill(struct dentry
*dentry
, struct dentry
*parent
)
421 __releases(dentry
->d_lock
)
422 __releases(parent
->d_lock
)
423 __releases(dentry
->d_inode
->i_lock
)
425 list_del(&dentry
->d_u
.d_child
);
427 * Inform try_to_ascend() that we are no longer attached to the
430 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
432 spin_unlock(&parent
->d_lock
);
435 * dentry_iput drops the locks, at which point nobody (except
436 * transient RCU lookups) can reach this dentry.
443 * Unhash a dentry without inserting an RCU walk barrier or checking that
444 * dentry->d_lock is locked. The caller must take care of that, if
447 static void __d_shrink(struct dentry
*dentry
)
449 if (!d_unhashed(dentry
)) {
450 struct hlist_bl_head
*b
;
451 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
452 b
= &dentry
->d_sb
->s_anon
;
454 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
457 __hlist_bl_del(&dentry
->d_hash
);
458 dentry
->d_hash
.pprev
= NULL
;
464 * d_drop - drop a dentry
465 * @dentry: dentry to drop
467 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
468 * be found through a VFS lookup any more. Note that this is different from
469 * deleting the dentry - d_delete will try to mark the dentry negative if
470 * possible, giving a successful _negative_ lookup, while d_drop will
471 * just make the cache lookup fail.
473 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
474 * reason (NFS timeouts or autofs deletes).
476 * __d_drop requires dentry->d_lock.
478 void __d_drop(struct dentry
*dentry
)
480 if (!d_unhashed(dentry
)) {
482 dentry_rcuwalk_barrier(dentry
);
485 EXPORT_SYMBOL(__d_drop
);
487 void d_drop(struct dentry
*dentry
)
489 spin_lock(&dentry
->d_lock
);
491 spin_unlock(&dentry
->d_lock
);
493 EXPORT_SYMBOL(d_drop
);
496 * Finish off a dentry we've decided to kill.
497 * dentry->d_lock must be held, returns with it unlocked.
498 * If ref is non-zero, then decrement the refcount too.
499 * Returns dentry requiring refcount drop, or NULL if we're done.
501 static inline struct dentry
*dentry_kill(struct dentry
*dentry
)
502 __releases(dentry
->d_lock
)
505 struct dentry
*parent
;
507 inode
= dentry
->d_inode
;
508 if (inode
&& !spin_trylock(&inode
->i_lock
)) {
510 spin_unlock(&dentry
->d_lock
);
512 return dentry
; /* try again with same dentry */
517 parent
= dentry
->d_parent
;
518 if (parent
&& !spin_trylock(&parent
->d_lock
)) {
520 spin_unlock(&inode
->i_lock
);
525 * The dentry is now unrecoverably dead to the world.
527 lockref_mark_dead(&dentry
->d_lockref
);
530 * inform the fs via d_prune that this dentry is about to be
531 * unhashed and destroyed.
533 if ((dentry
->d_flags
& DCACHE_OP_PRUNE
) && !d_unhashed(dentry
))
534 dentry
->d_op
->d_prune(dentry
);
536 dentry_lru_del(dentry
);
537 /* if it was on the hash then remove it */
539 return d_kill(dentry
, parent
);
545 * This is complicated by the fact that we do not want to put
546 * dentries that are no longer on any hash chain on the unused
547 * list: we'd much rather just get rid of them immediately.
549 * However, that implies that we have to traverse the dentry
550 * tree upwards to the parents which might _also_ now be
551 * scheduled for deletion (it may have been only waiting for
552 * its last child to go away).
554 * This tail recursion is done by hand as we don't want to depend
555 * on the compiler to always get this right (gcc generally doesn't).
556 * Real recursion would eat up our stack space.
560 * dput - release a dentry
561 * @dentry: dentry to release
563 * Release a dentry. This will drop the usage count and if appropriate
564 * call the dentry unlink method as well as removing it from the queues and
565 * releasing its resources. If the parent dentries were scheduled for release
566 * they too may now get deleted.
568 void dput(struct dentry
*dentry
)
570 if (unlikely(!dentry
))
574 if (lockref_put_or_lock(&dentry
->d_lockref
))
577 /* Unreachable? Get rid of it */
578 if (unlikely(d_unhashed(dentry
)))
581 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
582 if (dentry
->d_op
->d_delete(dentry
))
586 dentry
->d_flags
|= DCACHE_REFERENCED
;
587 dentry_lru_add(dentry
);
589 dentry
->d_lockref
.count
--;
590 spin_unlock(&dentry
->d_lock
);
594 dentry
= dentry_kill(dentry
);
601 * d_invalidate - invalidate a dentry
602 * @dentry: dentry to invalidate
604 * Try to invalidate the dentry if it turns out to be
605 * possible. If there are other dentries that can be
606 * reached through this one we can't delete it and we
607 * return -EBUSY. On success we return 0.
612 int d_invalidate(struct dentry
* dentry
)
615 * If it's already been dropped, return OK.
617 spin_lock(&dentry
->d_lock
);
618 if (d_unhashed(dentry
)) {
619 spin_unlock(&dentry
->d_lock
);
623 * Check whether to do a partial shrink_dcache
624 * to get rid of unused child entries.
626 if (!list_empty(&dentry
->d_subdirs
)) {
627 spin_unlock(&dentry
->d_lock
);
628 shrink_dcache_parent(dentry
);
629 spin_lock(&dentry
->d_lock
);
633 * Somebody else still using it?
635 * If it's a directory, we can't drop it
636 * for fear of somebody re-populating it
637 * with children (even though dropping it
638 * would make it unreachable from the root,
639 * we might still populate it if it was a
640 * working directory or similar).
641 * We also need to leave mountpoints alone,
644 if (dentry
->d_lockref
.count
> 1 && dentry
->d_inode
) {
645 if (S_ISDIR(dentry
->d_inode
->i_mode
) || d_mountpoint(dentry
)) {
646 spin_unlock(&dentry
->d_lock
);
652 spin_unlock(&dentry
->d_lock
);
655 EXPORT_SYMBOL(d_invalidate
);
657 /* This must be called with d_lock held */
658 static inline void __dget_dlock(struct dentry
*dentry
)
660 dentry
->d_lockref
.count
++;
663 static inline void __dget(struct dentry
*dentry
)
665 lockref_get(&dentry
->d_lockref
);
668 struct dentry
*dget_parent(struct dentry
*dentry
)
674 * Do optimistic parent lookup without any
678 ret
= ACCESS_ONCE(dentry
->d_parent
);
679 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
681 if (likely(gotref
)) {
682 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
689 * Don't need rcu_dereference because we re-check it was correct under
693 ret
= dentry
->d_parent
;
694 spin_lock(&ret
->d_lock
);
695 if (unlikely(ret
!= dentry
->d_parent
)) {
696 spin_unlock(&ret
->d_lock
);
701 BUG_ON(!ret
->d_lockref
.count
);
702 ret
->d_lockref
.count
++;
703 spin_unlock(&ret
->d_lock
);
706 EXPORT_SYMBOL(dget_parent
);
709 * d_find_alias - grab a hashed alias of inode
710 * @inode: inode in question
711 * @want_discon: flag, used by d_splice_alias, to request
712 * that only a DISCONNECTED alias be returned.
714 * If inode has a hashed alias, or is a directory and has any alias,
715 * acquire the reference to alias and return it. Otherwise return NULL.
716 * Notice that if inode is a directory there can be only one alias and
717 * it can be unhashed only if it has no children, or if it is the root
720 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
721 * any other hashed alias over that one unless @want_discon is set,
722 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
724 static struct dentry
*__d_find_alias(struct inode
*inode
, int want_discon
)
726 struct dentry
*alias
, *discon_alias
;
730 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
731 spin_lock(&alias
->d_lock
);
732 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
733 if (IS_ROOT(alias
) &&
734 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
735 discon_alias
= alias
;
736 } else if (!want_discon
) {
738 spin_unlock(&alias
->d_lock
);
742 spin_unlock(&alias
->d_lock
);
745 alias
= discon_alias
;
746 spin_lock(&alias
->d_lock
);
747 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
748 if (IS_ROOT(alias
) &&
749 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
751 spin_unlock(&alias
->d_lock
);
755 spin_unlock(&alias
->d_lock
);
761 struct dentry
*d_find_alias(struct inode
*inode
)
763 struct dentry
*de
= NULL
;
765 if (!hlist_empty(&inode
->i_dentry
)) {
766 spin_lock(&inode
->i_lock
);
767 de
= __d_find_alias(inode
, 0);
768 spin_unlock(&inode
->i_lock
);
772 EXPORT_SYMBOL(d_find_alias
);
775 * Try to kill dentries associated with this inode.
776 * WARNING: you must own a reference to inode.
778 void d_prune_aliases(struct inode
*inode
)
780 struct dentry
*dentry
;
782 spin_lock(&inode
->i_lock
);
783 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_alias
) {
784 spin_lock(&dentry
->d_lock
);
785 if (!dentry
->d_lockref
.count
) {
787 * inform the fs via d_prune that this dentry
788 * is about to be unhashed and destroyed.
790 if ((dentry
->d_flags
& DCACHE_OP_PRUNE
) &&
792 dentry
->d_op
->d_prune(dentry
);
794 __dget_dlock(dentry
);
796 spin_unlock(&dentry
->d_lock
);
797 spin_unlock(&inode
->i_lock
);
801 spin_unlock(&dentry
->d_lock
);
803 spin_unlock(&inode
->i_lock
);
805 EXPORT_SYMBOL(d_prune_aliases
);
808 * Try to throw away a dentry - free the inode, dput the parent.
809 * Requires dentry->d_lock is held, and dentry->d_count == 0.
810 * Releases dentry->d_lock.
812 * This may fail if locks cannot be acquired no problem, just try again.
814 static void try_prune_one_dentry(struct dentry
*dentry
)
815 __releases(dentry
->d_lock
)
817 struct dentry
*parent
;
819 parent
= dentry_kill(dentry
);
821 * If dentry_kill returns NULL, we have nothing more to do.
822 * if it returns the same dentry, trylocks failed. In either
823 * case, just loop again.
825 * Otherwise, we need to prune ancestors too. This is necessary
826 * to prevent quadratic behavior of shrink_dcache_parent(), but
827 * is also expected to be beneficial in reducing dentry cache
832 if (parent
== dentry
)
835 /* Prune ancestors. */
838 if (lockref_put_or_lock(&dentry
->d_lockref
))
840 dentry
= dentry_kill(dentry
);
844 static void shrink_dentry_list(struct list_head
*list
)
846 struct dentry
*dentry
;
850 dentry
= list_entry_rcu(list
->prev
, struct dentry
, d_lru
);
851 if (&dentry
->d_lru
== list
)
853 spin_lock(&dentry
->d_lock
);
854 if (dentry
!= list_entry(list
->prev
, struct dentry
, d_lru
)) {
855 spin_unlock(&dentry
->d_lock
);
860 * We found an inuse dentry which was not removed from
861 * the LRU because of laziness during lookup. Do not free
862 * it - just keep it off the LRU list.
864 if (dentry
->d_lockref
.count
) {
865 dentry_lru_del(dentry
);
866 spin_unlock(&dentry
->d_lock
);
872 try_prune_one_dentry(dentry
);
880 * prune_dcache_sb - shrink the dcache
882 * @count: number of entries to try to free
884 * Attempt to shrink the superblock dcache LRU by @count entries. This is
885 * done when we need more memory an called from the superblock shrinker
888 * This function may fail to free any resources if all the dentries are in
891 void prune_dcache_sb(struct super_block
*sb
, int count
)
893 struct dentry
*dentry
;
894 LIST_HEAD(referenced
);
898 spin_lock(&dcache_lru_lock
);
899 while (!list_empty(&sb
->s_dentry_lru
)) {
900 dentry
= list_entry(sb
->s_dentry_lru
.prev
,
901 struct dentry
, d_lru
);
902 BUG_ON(dentry
->d_sb
!= sb
);
904 if (!spin_trylock(&dentry
->d_lock
)) {
905 spin_unlock(&dcache_lru_lock
);
910 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
911 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
912 list_move(&dentry
->d_lru
, &referenced
);
913 spin_unlock(&dentry
->d_lock
);
915 list_move_tail(&dentry
->d_lru
, &tmp
);
916 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
917 spin_unlock(&dentry
->d_lock
);
921 cond_resched_lock(&dcache_lru_lock
);
923 if (!list_empty(&referenced
))
924 list_splice(&referenced
, &sb
->s_dentry_lru
);
925 spin_unlock(&dcache_lru_lock
);
927 shrink_dentry_list(&tmp
);
931 * shrink_dcache_sb - shrink dcache for a superblock
934 * Shrink the dcache for the specified super block. This is used to free
935 * the dcache before unmounting a file system.
937 void shrink_dcache_sb(struct super_block
*sb
)
941 spin_lock(&dcache_lru_lock
);
942 while (!list_empty(&sb
->s_dentry_lru
)) {
943 list_splice_init(&sb
->s_dentry_lru
, &tmp
);
944 spin_unlock(&dcache_lru_lock
);
945 shrink_dentry_list(&tmp
);
946 spin_lock(&dcache_lru_lock
);
948 spin_unlock(&dcache_lru_lock
);
950 EXPORT_SYMBOL(shrink_dcache_sb
);
953 * destroy a single subtree of dentries for unmount
954 * - see the comments on shrink_dcache_for_umount() for a description of the
957 static void shrink_dcache_for_umount_subtree(struct dentry
*dentry
)
959 struct dentry
*parent
;
961 BUG_ON(!IS_ROOT(dentry
));
964 /* descend to the first leaf in the current subtree */
965 while (!list_empty(&dentry
->d_subdirs
))
966 dentry
= list_entry(dentry
->d_subdirs
.next
,
967 struct dentry
, d_u
.d_child
);
969 /* consume the dentries from this leaf up through its parents
970 * until we find one with children or run out altogether */
975 * inform the fs that this dentry is about to be
976 * unhashed and destroyed.
978 if ((dentry
->d_flags
& DCACHE_OP_PRUNE
) &&
980 dentry
->d_op
->d_prune(dentry
);
982 dentry_lru_del(dentry
);
985 if (dentry
->d_lockref
.count
!= 0) {
987 "BUG: Dentry %p{i=%lx,n=%s}"
989 " [unmount of %s %s]\n",
992 dentry
->d_inode
->i_ino
: 0UL,
994 dentry
->d_lockref
.count
,
995 dentry
->d_sb
->s_type
->name
,
1000 if (IS_ROOT(dentry
)) {
1002 list_del(&dentry
->d_u
.d_child
);
1004 parent
= dentry
->d_parent
;
1005 parent
->d_lockref
.count
--;
1006 list_del(&dentry
->d_u
.d_child
);
1009 inode
= dentry
->d_inode
;
1011 dentry
->d_inode
= NULL
;
1012 hlist_del_init(&dentry
->d_alias
);
1013 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
1014 dentry
->d_op
->d_iput(dentry
, inode
);
1021 /* finished when we fall off the top of the tree,
1022 * otherwise we ascend to the parent and move to the
1023 * next sibling if there is one */
1027 } while (list_empty(&dentry
->d_subdirs
));
1029 dentry
= list_entry(dentry
->d_subdirs
.next
,
1030 struct dentry
, d_u
.d_child
);
1035 * destroy the dentries attached to a superblock on unmounting
1036 * - we don't need to use dentry->d_lock because:
1037 * - the superblock is detached from all mountings and open files, so the
1038 * dentry trees will not be rearranged by the VFS
1039 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1040 * any dentries belonging to this superblock that it comes across
1041 * - the filesystem itself is no longer permitted to rearrange the dentries
1042 * in this superblock
1044 void shrink_dcache_for_umount(struct super_block
*sb
)
1046 struct dentry
*dentry
;
1048 if (down_read_trylock(&sb
->s_umount
))
1051 dentry
= sb
->s_root
;
1053 dentry
->d_lockref
.count
--;
1054 shrink_dcache_for_umount_subtree(dentry
);
1056 while (!hlist_bl_empty(&sb
->s_anon
)) {
1057 dentry
= hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
);
1058 shrink_dcache_for_umount_subtree(dentry
);
1063 * This tries to ascend one level of parenthood, but
1064 * we can race with renaming, so we need to re-check
1065 * the parenthood after dropping the lock and check
1066 * that the sequence number still matches.
1068 static struct dentry
*try_to_ascend(struct dentry
*old
, unsigned seq
)
1070 struct dentry
*new = old
->d_parent
;
1073 spin_unlock(&old
->d_lock
);
1074 spin_lock(&new->d_lock
);
1077 * might go back up the wrong parent if we have had a rename
1080 if (new != old
->d_parent
||
1081 (old
->d_flags
& DCACHE_DENTRY_KILLED
) ||
1082 need_seqretry(&rename_lock
, seq
)) {
1083 spin_unlock(&new->d_lock
);
1091 * enum d_walk_ret - action to talke during tree walk
1092 * @D_WALK_CONTINUE: contrinue walk
1093 * @D_WALK_QUIT: quit walk
1094 * @D_WALK_NORETRY: quit when retry is needed
1095 * @D_WALK_SKIP: skip this dentry and its children
1105 * d_walk - walk the dentry tree
1106 * @parent: start of walk
1107 * @data: data passed to @enter() and @finish()
1108 * @enter: callback when first entering the dentry
1109 * @finish: callback when successfully finished the walk
1111 * The @enter() and @finish() callbacks are called with d_lock held.
1113 static void d_walk(struct dentry
*parent
, void *data
,
1114 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1115 void (*finish
)(void *))
1117 struct dentry
*this_parent
;
1118 struct list_head
*next
;
1120 enum d_walk_ret ret
;
1124 read_seqbegin_or_lock(&rename_lock
, &seq
);
1125 this_parent
= parent
;
1126 spin_lock(&this_parent
->d_lock
);
1128 ret
= enter(data
, this_parent
);
1130 case D_WALK_CONTINUE
:
1135 case D_WALK_NORETRY
:
1140 next
= this_parent
->d_subdirs
.next
;
1142 while (next
!= &this_parent
->d_subdirs
) {
1143 struct list_head
*tmp
= next
;
1144 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_u
.d_child
);
1147 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1149 ret
= enter(data
, dentry
);
1151 case D_WALK_CONTINUE
:
1154 spin_unlock(&dentry
->d_lock
);
1156 case D_WALK_NORETRY
:
1160 spin_unlock(&dentry
->d_lock
);
1164 if (!list_empty(&dentry
->d_subdirs
)) {
1165 spin_unlock(&this_parent
->d_lock
);
1166 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1167 this_parent
= dentry
;
1168 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1171 spin_unlock(&dentry
->d_lock
);
1174 * All done at this level ... ascend and resume the search.
1176 if (this_parent
!= parent
) {
1177 struct dentry
*child
= this_parent
;
1178 this_parent
= try_to_ascend(this_parent
, seq
);
1181 next
= child
->d_u
.d_child
.next
;
1184 if (need_seqretry(&rename_lock
, seq
)) {
1185 spin_unlock(&this_parent
->d_lock
);
1192 spin_unlock(&this_parent
->d_lock
);
1193 done_seqretry(&rename_lock
, seq
);
1204 * Search for at least 1 mount point in the dentry's subdirs.
1205 * We descend to the next level whenever the d_subdirs
1206 * list is non-empty and continue searching.
1210 * have_submounts - check for mounts over a dentry
1211 * @parent: dentry to check.
1213 * Return true if the parent or its subdirectories contain
1217 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1220 if (d_mountpoint(dentry
)) {
1224 return D_WALK_CONTINUE
;
1227 int have_submounts(struct dentry
*parent
)
1231 d_walk(parent
, &ret
, check_mount
, NULL
);
1235 EXPORT_SYMBOL(have_submounts
);
1238 * Called by mount code to set a mountpoint and check if the mountpoint is
1239 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1240 * subtree can become unreachable).
1242 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1243 * this reason take rename_lock and d_lock on dentry and ancestors.
1245 int d_set_mounted(struct dentry
*dentry
)
1249 write_seqlock(&rename_lock
);
1250 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1251 /* Need exclusion wrt. check_submounts_and_drop() */
1252 spin_lock(&p
->d_lock
);
1253 if (unlikely(d_unhashed(p
))) {
1254 spin_unlock(&p
->d_lock
);
1257 spin_unlock(&p
->d_lock
);
1259 spin_lock(&dentry
->d_lock
);
1260 if (!d_unlinked(dentry
)) {
1261 dentry
->d_flags
|= DCACHE_MOUNTED
;
1264 spin_unlock(&dentry
->d_lock
);
1266 write_sequnlock(&rename_lock
);
1271 * Search the dentry child list of the specified parent,
1272 * and move any unused dentries to the end of the unused
1273 * list for prune_dcache(). We descend to the next level
1274 * whenever the d_subdirs list is non-empty and continue
1277 * It returns zero iff there are no unused children,
1278 * otherwise it returns the number of children moved to
1279 * the end of the unused list. This may not be the total
1280 * number of unused children, because select_parent can
1281 * drop the lock and return early due to latency
1285 struct select_data
{
1286 struct dentry
*start
;
1287 struct list_head dispose
;
1291 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1293 struct select_data
*data
= _data
;
1294 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1296 if (data
->start
== dentry
)
1300 * move only zero ref count dentries to the dispose list.
1302 * Those which are presently on the shrink list, being processed
1303 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1304 * loop in shrink_dcache_parent() might not make any progress
1307 if (dentry
->d_lockref
.count
) {
1308 dentry_lru_del(dentry
);
1309 } else if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
)) {
1310 dentry_lru_move_list(dentry
, &data
->dispose
);
1311 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
1313 ret
= D_WALK_NORETRY
;
1316 * We can return to the caller if we have found some (this
1317 * ensures forward progress). We'll be coming back to find
1320 if (data
->found
&& need_resched())
1327 * shrink_dcache_parent - prune dcache
1328 * @parent: parent of entries to prune
1330 * Prune the dcache to remove unused children of the parent dentry.
1332 void shrink_dcache_parent(struct dentry
*parent
)
1335 struct select_data data
;
1337 INIT_LIST_HEAD(&data
.dispose
);
1338 data
.start
= parent
;
1341 d_walk(parent
, &data
, select_collect
, NULL
);
1345 shrink_dentry_list(&data
.dispose
);
1349 EXPORT_SYMBOL(shrink_dcache_parent
);
1351 static enum d_walk_ret
check_and_collect(void *_data
, struct dentry
*dentry
)
1353 struct select_data
*data
= _data
;
1355 if (d_mountpoint(dentry
)) {
1356 data
->found
= -EBUSY
;
1360 return select_collect(_data
, dentry
);
1363 static void check_and_drop(void *_data
)
1365 struct select_data
*data
= _data
;
1367 if (d_mountpoint(data
->start
))
1368 data
->found
= -EBUSY
;
1370 __d_drop(data
->start
);
1374 * check_submounts_and_drop - prune dcache, check for submounts and drop
1376 * All done as a single atomic operation relative to has_unlinked_ancestor().
1377 * Returns 0 if successfully unhashed @parent. If there were submounts then
1380 * @dentry: dentry to prune and drop
1382 int check_submounts_and_drop(struct dentry
*dentry
)
1386 /* Negative dentries can be dropped without further checks */
1387 if (!dentry
->d_inode
) {
1393 struct select_data data
;
1395 INIT_LIST_HEAD(&data
.dispose
);
1396 data
.start
= dentry
;
1399 d_walk(dentry
, &data
, check_and_collect
, check_and_drop
);
1402 if (!list_empty(&data
.dispose
))
1403 shrink_dentry_list(&data
.dispose
);
1414 EXPORT_SYMBOL(check_submounts_and_drop
);
1417 * __d_alloc - allocate a dcache entry
1418 * @sb: filesystem it will belong to
1419 * @name: qstr of the name
1421 * Allocates a dentry. It returns %NULL if there is insufficient memory
1422 * available. On a success the dentry is returned. The name passed in is
1423 * copied and the copy passed in may be reused after this call.
1426 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1428 struct dentry
*dentry
;
1431 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1436 * We guarantee that the inline name is always NUL-terminated.
1437 * This way the memcpy() done by the name switching in rename
1438 * will still always have a NUL at the end, even if we might
1439 * be overwriting an internal NUL character
1441 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1442 if (name
->len
> DNAME_INLINE_LEN
-1) {
1443 dname
= kmalloc(name
->len
+ 1, GFP_KERNEL
);
1445 kmem_cache_free(dentry_cache
, dentry
);
1449 dname
= dentry
->d_iname
;
1452 dentry
->d_name
.len
= name
->len
;
1453 dentry
->d_name
.hash
= name
->hash
;
1454 memcpy(dname
, name
->name
, name
->len
);
1455 dname
[name
->len
] = 0;
1457 /* Make sure we always see the terminating NUL character */
1459 dentry
->d_name
.name
= dname
;
1461 dentry
->d_lockref
.count
= 1;
1462 dentry
->d_flags
= 0;
1463 spin_lock_init(&dentry
->d_lock
);
1464 seqcount_init(&dentry
->d_seq
);
1465 dentry
->d_inode
= NULL
;
1466 dentry
->d_parent
= dentry
;
1468 dentry
->d_op
= NULL
;
1469 dentry
->d_fsdata
= NULL
;
1470 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1471 INIT_LIST_HEAD(&dentry
->d_lru
);
1472 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1473 INIT_HLIST_NODE(&dentry
->d_alias
);
1474 INIT_LIST_HEAD(&dentry
->d_u
.d_child
);
1475 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1477 this_cpu_inc(nr_dentry
);
1483 * d_alloc - allocate a dcache entry
1484 * @parent: parent of entry to allocate
1485 * @name: qstr of the name
1487 * Allocates a dentry. It returns %NULL if there is insufficient memory
1488 * available. On a success the dentry is returned. The name passed in is
1489 * copied and the copy passed in may be reused after this call.
1491 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1493 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1497 spin_lock(&parent
->d_lock
);
1499 * don't need child lock because it is not subject
1500 * to concurrency here
1502 __dget_dlock(parent
);
1503 dentry
->d_parent
= parent
;
1504 list_add(&dentry
->d_u
.d_child
, &parent
->d_subdirs
);
1505 spin_unlock(&parent
->d_lock
);
1509 EXPORT_SYMBOL(d_alloc
);
1511 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1513 struct dentry
*dentry
= __d_alloc(sb
, name
);
1515 dentry
->d_flags
|= DCACHE_DISCONNECTED
;
1518 EXPORT_SYMBOL(d_alloc_pseudo
);
1520 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1525 q
.len
= strlen(name
);
1526 q
.hash
= full_name_hash(q
.name
, q
.len
);
1527 return d_alloc(parent
, &q
);
1529 EXPORT_SYMBOL(d_alloc_name
);
1531 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1533 WARN_ON_ONCE(dentry
->d_op
);
1534 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1536 DCACHE_OP_REVALIDATE
|
1537 DCACHE_OP_WEAK_REVALIDATE
|
1538 DCACHE_OP_DELETE
));
1543 dentry
->d_flags
|= DCACHE_OP_HASH
;
1545 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1546 if (op
->d_revalidate
)
1547 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1548 if (op
->d_weak_revalidate
)
1549 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1551 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1553 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1556 EXPORT_SYMBOL(d_set_d_op
);
1558 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1560 spin_lock(&dentry
->d_lock
);
1562 if (unlikely(IS_AUTOMOUNT(inode
)))
1563 dentry
->d_flags
|= DCACHE_NEED_AUTOMOUNT
;
1564 hlist_add_head(&dentry
->d_alias
, &inode
->i_dentry
);
1566 dentry
->d_inode
= inode
;
1567 dentry_rcuwalk_barrier(dentry
);
1568 spin_unlock(&dentry
->d_lock
);
1569 fsnotify_d_instantiate(dentry
, inode
);
1573 * d_instantiate - fill in inode information for a dentry
1574 * @entry: dentry to complete
1575 * @inode: inode to attach to this dentry
1577 * Fill in inode information in the entry.
1579 * This turns negative dentries into productive full members
1582 * NOTE! This assumes that the inode count has been incremented
1583 * (or otherwise set) by the caller to indicate that it is now
1584 * in use by the dcache.
1587 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1589 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1591 spin_lock(&inode
->i_lock
);
1592 __d_instantiate(entry
, inode
);
1594 spin_unlock(&inode
->i_lock
);
1595 security_d_instantiate(entry
, inode
);
1597 EXPORT_SYMBOL(d_instantiate
);
1600 * d_instantiate_unique - instantiate a non-aliased dentry
1601 * @entry: dentry to instantiate
1602 * @inode: inode to attach to this dentry
1604 * Fill in inode information in the entry. On success, it returns NULL.
1605 * If an unhashed alias of "entry" already exists, then we return the
1606 * aliased dentry instead and drop one reference to inode.
1608 * Note that in order to avoid conflicts with rename() etc, the caller
1609 * had better be holding the parent directory semaphore.
1611 * This also assumes that the inode count has been incremented
1612 * (or otherwise set) by the caller to indicate that it is now
1613 * in use by the dcache.
1615 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1616 struct inode
*inode
)
1618 struct dentry
*alias
;
1619 int len
= entry
->d_name
.len
;
1620 const char *name
= entry
->d_name
.name
;
1621 unsigned int hash
= entry
->d_name
.hash
;
1624 __d_instantiate(entry
, NULL
);
1628 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_alias
) {
1630 * Don't need alias->d_lock here, because aliases with
1631 * d_parent == entry->d_parent are not subject to name or
1632 * parent changes, because the parent inode i_mutex is held.
1634 if (alias
->d_name
.hash
!= hash
)
1636 if (alias
->d_parent
!= entry
->d_parent
)
1638 if (alias
->d_name
.len
!= len
)
1640 if (dentry_cmp(alias
, name
, len
))
1646 __d_instantiate(entry
, inode
);
1650 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1652 struct dentry
*result
;
1654 BUG_ON(!hlist_unhashed(&entry
->d_alias
));
1657 spin_lock(&inode
->i_lock
);
1658 result
= __d_instantiate_unique(entry
, inode
);
1660 spin_unlock(&inode
->i_lock
);
1663 security_d_instantiate(entry
, inode
);
1667 BUG_ON(!d_unhashed(result
));
1672 EXPORT_SYMBOL(d_instantiate_unique
);
1674 struct dentry
*d_make_root(struct inode
*root_inode
)
1676 struct dentry
*res
= NULL
;
1679 static const struct qstr name
= QSTR_INIT("/", 1);
1681 res
= __d_alloc(root_inode
->i_sb
, &name
);
1683 d_instantiate(res
, root_inode
);
1689 EXPORT_SYMBOL(d_make_root
);
1691 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1693 struct dentry
*alias
;
1695 if (hlist_empty(&inode
->i_dentry
))
1697 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_alias
);
1703 * d_find_any_alias - find any alias for a given inode
1704 * @inode: inode to find an alias for
1706 * If any aliases exist for the given inode, take and return a
1707 * reference for one of them. If no aliases exist, return %NULL.
1709 struct dentry
*d_find_any_alias(struct inode
*inode
)
1713 spin_lock(&inode
->i_lock
);
1714 de
= __d_find_any_alias(inode
);
1715 spin_unlock(&inode
->i_lock
);
1718 EXPORT_SYMBOL(d_find_any_alias
);
1721 * d_obtain_alias - find or allocate a dentry for a given inode
1722 * @inode: inode to allocate the dentry for
1724 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1725 * similar open by handle operations. The returned dentry may be anonymous,
1726 * or may have a full name (if the inode was already in the cache).
1728 * When called on a directory inode, we must ensure that the inode only ever
1729 * has one dentry. If a dentry is found, that is returned instead of
1730 * allocating a new one.
1732 * On successful return, the reference to the inode has been transferred
1733 * to the dentry. In case of an error the reference on the inode is released.
1734 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1735 * be passed in and will be the error will be propagate to the return value,
1736 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1738 struct dentry
*d_obtain_alias(struct inode
*inode
)
1740 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1745 return ERR_PTR(-ESTALE
);
1747 return ERR_CAST(inode
);
1749 res
= d_find_any_alias(inode
);
1753 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1755 res
= ERR_PTR(-ENOMEM
);
1759 spin_lock(&inode
->i_lock
);
1760 res
= __d_find_any_alias(inode
);
1762 spin_unlock(&inode
->i_lock
);
1767 /* attach a disconnected dentry */
1768 spin_lock(&tmp
->d_lock
);
1769 tmp
->d_inode
= inode
;
1770 tmp
->d_flags
|= DCACHE_DISCONNECTED
;
1771 hlist_add_head(&tmp
->d_alias
, &inode
->i_dentry
);
1772 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1773 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1774 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1775 spin_unlock(&tmp
->d_lock
);
1776 spin_unlock(&inode
->i_lock
);
1777 security_d_instantiate(tmp
, inode
);
1782 if (res
&& !IS_ERR(res
))
1783 security_d_instantiate(res
, inode
);
1787 EXPORT_SYMBOL(d_obtain_alias
);
1790 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1791 * @inode: the inode which may have a disconnected dentry
1792 * @dentry: a negative dentry which we want to point to the inode.
1794 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1795 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1796 * and return it, else simply d_add the inode to the dentry and return NULL.
1798 * This is needed in the lookup routine of any filesystem that is exportable
1799 * (via knfsd) so that we can build dcache paths to directories effectively.
1801 * If a dentry was found and moved, then it is returned. Otherwise NULL
1802 * is returned. This matches the expected return value of ->lookup.
1804 * Cluster filesystems may call this function with a negative, hashed dentry.
1805 * In that case, we know that the inode will be a regular file, and also this
1806 * will only occur during atomic_open. So we need to check for the dentry
1807 * being already hashed only in the final case.
1809 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
1811 struct dentry
*new = NULL
;
1814 return ERR_CAST(inode
);
1816 if (inode
&& S_ISDIR(inode
->i_mode
)) {
1817 spin_lock(&inode
->i_lock
);
1818 new = __d_find_alias(inode
, 1);
1820 BUG_ON(!(new->d_flags
& DCACHE_DISCONNECTED
));
1821 spin_unlock(&inode
->i_lock
);
1822 security_d_instantiate(new, inode
);
1823 d_move(new, dentry
);
1826 /* already taking inode->i_lock, so d_add() by hand */
1827 __d_instantiate(dentry
, inode
);
1828 spin_unlock(&inode
->i_lock
);
1829 security_d_instantiate(dentry
, inode
);
1833 d_instantiate(dentry
, inode
);
1834 if (d_unhashed(dentry
))
1839 EXPORT_SYMBOL(d_splice_alias
);
1842 * d_add_ci - lookup or allocate new dentry with case-exact name
1843 * @inode: the inode case-insensitive lookup has found
1844 * @dentry: the negative dentry that was passed to the parent's lookup func
1845 * @name: the case-exact name to be associated with the returned dentry
1847 * This is to avoid filling the dcache with case-insensitive names to the
1848 * same inode, only the actual correct case is stored in the dcache for
1849 * case-insensitive filesystems.
1851 * For a case-insensitive lookup match and if the the case-exact dentry
1852 * already exists in in the dcache, use it and return it.
1854 * If no entry exists with the exact case name, allocate new dentry with
1855 * the exact case, and return the spliced entry.
1857 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
1860 struct dentry
*found
;
1864 * First check if a dentry matching the name already exists,
1865 * if not go ahead and create it now.
1867 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
1868 if (unlikely(IS_ERR(found
)))
1871 new = d_alloc(dentry
->d_parent
, name
);
1873 found
= ERR_PTR(-ENOMEM
);
1877 found
= d_splice_alias(inode
, new);
1886 * If a matching dentry exists, and it's not negative use it.
1888 * Decrement the reference count to balance the iget() done
1891 if (found
->d_inode
) {
1892 if (unlikely(found
->d_inode
!= inode
)) {
1893 /* This can't happen because bad inodes are unhashed. */
1894 BUG_ON(!is_bad_inode(inode
));
1895 BUG_ON(!is_bad_inode(found
->d_inode
));
1902 * Negative dentry: instantiate it unless the inode is a directory and
1903 * already has a dentry.
1905 new = d_splice_alias(inode
, found
);
1916 EXPORT_SYMBOL(d_add_ci
);
1919 * Do the slow-case of the dentry name compare.
1921 * Unlike the dentry_cmp() function, we need to atomically
1922 * load the name and length information, so that the
1923 * filesystem can rely on them, and can use the 'name' and
1924 * 'len' information without worrying about walking off the
1925 * end of memory etc.
1927 * Thus the read_seqcount_retry() and the "duplicate" info
1928 * in arguments (the low-level filesystem should not look
1929 * at the dentry inode or name contents directly, since
1930 * rename can change them while we're in RCU mode).
1932 enum slow_d_compare
{
1938 static noinline
enum slow_d_compare
slow_dentry_cmp(
1939 const struct dentry
*parent
,
1940 struct dentry
*dentry
,
1942 const struct qstr
*name
)
1944 int tlen
= dentry
->d_name
.len
;
1945 const char *tname
= dentry
->d_name
.name
;
1947 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
1949 return D_COMP_SEQRETRY
;
1951 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
1952 return D_COMP_NOMATCH
;
1957 * __d_lookup_rcu - search for a dentry (racy, store-free)
1958 * @parent: parent dentry
1959 * @name: qstr of name we wish to find
1960 * @seqp: returns d_seq value at the point where the dentry was found
1961 * Returns: dentry, or NULL
1963 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1964 * resolution (store-free path walking) design described in
1965 * Documentation/filesystems/path-lookup.txt.
1967 * This is not to be used outside core vfs.
1969 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1970 * held, and rcu_read_lock held. The returned dentry must not be stored into
1971 * without taking d_lock and checking d_seq sequence count against @seq
1974 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1977 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1978 * the returned dentry, so long as its parent's seqlock is checked after the
1979 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1980 * is formed, giving integrity down the path walk.
1982 * NOTE! The caller *has* to check the resulting dentry against the sequence
1983 * number we've returned before using any of the resulting dentry state!
1985 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
1986 const struct qstr
*name
,
1989 u64 hashlen
= name
->hash_len
;
1990 const unsigned char *str
= name
->name
;
1991 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
1992 struct hlist_bl_node
*node
;
1993 struct dentry
*dentry
;
1996 * Note: There is significant duplication with __d_lookup_rcu which is
1997 * required to prevent single threaded performance regressions
1998 * especially on architectures where smp_rmb (in seqcounts) are costly.
1999 * Keep the two functions in sync.
2003 * The hash list is protected using RCU.
2005 * Carefully use d_seq when comparing a candidate dentry, to avoid
2006 * races with d_move().
2008 * It is possible that concurrent renames can mess up our list
2009 * walk here and result in missing our dentry, resulting in the
2010 * false-negative result. d_lookup() protects against concurrent
2011 * renames using rename_lock seqlock.
2013 * See Documentation/filesystems/path-lookup.txt for more details.
2015 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2020 * The dentry sequence count protects us from concurrent
2021 * renames, and thus protects parent and name fields.
2023 * The caller must perform a seqcount check in order
2024 * to do anything useful with the returned dentry.
2026 * NOTE! We do a "raw" seqcount_begin here. That means that
2027 * we don't wait for the sequence count to stabilize if it
2028 * is in the middle of a sequence change. If we do the slow
2029 * dentry compare, we will do seqretries until it is stable,
2030 * and if we end up with a successful lookup, we actually
2031 * want to exit RCU lookup anyway.
2033 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2034 if (dentry
->d_parent
!= parent
)
2036 if (d_unhashed(dentry
))
2039 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2040 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2043 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2046 case D_COMP_NOMATCH
:
2053 if (dentry
->d_name
.hash_len
!= hashlen
)
2056 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2063 * d_lookup - search for a dentry
2064 * @parent: parent dentry
2065 * @name: qstr of name we wish to find
2066 * Returns: dentry, or NULL
2068 * d_lookup searches the children of the parent dentry for the name in
2069 * question. If the dentry is found its reference count is incremented and the
2070 * dentry is returned. The caller must use dput to free the entry when it has
2071 * finished using it. %NULL is returned if the dentry does not exist.
2073 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2075 struct dentry
*dentry
;
2079 seq
= read_seqbegin(&rename_lock
);
2080 dentry
= __d_lookup(parent
, name
);
2083 } while (read_seqretry(&rename_lock
, seq
));
2086 EXPORT_SYMBOL(d_lookup
);
2089 * __d_lookup - search for a dentry (racy)
2090 * @parent: parent dentry
2091 * @name: qstr of name we wish to find
2092 * Returns: dentry, or NULL
2094 * __d_lookup is like d_lookup, however it may (rarely) return a
2095 * false-negative result due to unrelated rename activity.
2097 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2098 * however it must be used carefully, eg. with a following d_lookup in
2099 * the case of failure.
2101 * __d_lookup callers must be commented.
2103 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2105 unsigned int len
= name
->len
;
2106 unsigned int hash
= name
->hash
;
2107 const unsigned char *str
= name
->name
;
2108 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2109 struct hlist_bl_node
*node
;
2110 struct dentry
*found
= NULL
;
2111 struct dentry
*dentry
;
2114 * Note: There is significant duplication with __d_lookup_rcu which is
2115 * required to prevent single threaded performance regressions
2116 * especially on architectures where smp_rmb (in seqcounts) are costly.
2117 * Keep the two functions in sync.
2121 * The hash list is protected using RCU.
2123 * Take d_lock when comparing a candidate dentry, to avoid races
2126 * It is possible that concurrent renames can mess up our list
2127 * walk here and result in missing our dentry, resulting in the
2128 * false-negative result. d_lookup() protects against concurrent
2129 * renames using rename_lock seqlock.
2131 * See Documentation/filesystems/path-lookup.txt for more details.
2135 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2137 if (dentry
->d_name
.hash
!= hash
)
2140 spin_lock(&dentry
->d_lock
);
2141 if (dentry
->d_parent
!= parent
)
2143 if (d_unhashed(dentry
))
2147 * It is safe to compare names since d_move() cannot
2148 * change the qstr (protected by d_lock).
2150 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2151 int tlen
= dentry
->d_name
.len
;
2152 const char *tname
= dentry
->d_name
.name
;
2153 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2156 if (dentry
->d_name
.len
!= len
)
2158 if (dentry_cmp(dentry
, str
, len
))
2162 dentry
->d_lockref
.count
++;
2164 spin_unlock(&dentry
->d_lock
);
2167 spin_unlock(&dentry
->d_lock
);
2175 * d_hash_and_lookup - hash the qstr then search for a dentry
2176 * @dir: Directory to search in
2177 * @name: qstr of name we wish to find
2179 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2181 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2184 * Check for a fs-specific hash function. Note that we must
2185 * calculate the standard hash first, as the d_op->d_hash()
2186 * routine may choose to leave the hash value unchanged.
2188 name
->hash
= full_name_hash(name
->name
, name
->len
);
2189 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2190 int err
= dir
->d_op
->d_hash(dir
, name
);
2191 if (unlikely(err
< 0))
2192 return ERR_PTR(err
);
2194 return d_lookup(dir
, name
);
2196 EXPORT_SYMBOL(d_hash_and_lookup
);
2199 * d_validate - verify dentry provided from insecure source (deprecated)
2200 * @dentry: The dentry alleged to be valid child of @dparent
2201 * @dparent: The parent dentry (known to be valid)
2203 * An insecure source has sent us a dentry, here we verify it and dget() it.
2204 * This is used by ncpfs in its readdir implementation.
2205 * Zero is returned in the dentry is invalid.
2207 * This function is slow for big directories, and deprecated, do not use it.
2209 int d_validate(struct dentry
*dentry
, struct dentry
*dparent
)
2211 struct dentry
*child
;
2213 spin_lock(&dparent
->d_lock
);
2214 list_for_each_entry(child
, &dparent
->d_subdirs
, d_u
.d_child
) {
2215 if (dentry
== child
) {
2216 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
2217 __dget_dlock(dentry
);
2218 spin_unlock(&dentry
->d_lock
);
2219 spin_unlock(&dparent
->d_lock
);
2223 spin_unlock(&dparent
->d_lock
);
2227 EXPORT_SYMBOL(d_validate
);
2230 * When a file is deleted, we have two options:
2231 * - turn this dentry into a negative dentry
2232 * - unhash this dentry and free it.
2234 * Usually, we want to just turn this into
2235 * a negative dentry, but if anybody else is
2236 * currently using the dentry or the inode
2237 * we can't do that and we fall back on removing
2238 * it from the hash queues and waiting for
2239 * it to be deleted later when it has no users
2243 * d_delete - delete a dentry
2244 * @dentry: The dentry to delete
2246 * Turn the dentry into a negative dentry if possible, otherwise
2247 * remove it from the hash queues so it can be deleted later
2250 void d_delete(struct dentry
* dentry
)
2252 struct inode
*inode
;
2255 * Are we the only user?
2258 spin_lock(&dentry
->d_lock
);
2259 inode
= dentry
->d_inode
;
2260 isdir
= S_ISDIR(inode
->i_mode
);
2261 if (dentry
->d_lockref
.count
== 1) {
2262 if (!spin_trylock(&inode
->i_lock
)) {
2263 spin_unlock(&dentry
->d_lock
);
2267 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2268 dentry_unlink_inode(dentry
);
2269 fsnotify_nameremove(dentry
, isdir
);
2273 if (!d_unhashed(dentry
))
2276 spin_unlock(&dentry
->d_lock
);
2278 fsnotify_nameremove(dentry
, isdir
);
2280 EXPORT_SYMBOL(d_delete
);
2282 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2284 BUG_ON(!d_unhashed(entry
));
2286 entry
->d_flags
|= DCACHE_RCUACCESS
;
2287 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2291 static void _d_rehash(struct dentry
* entry
)
2293 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2297 * d_rehash - add an entry back to the hash
2298 * @entry: dentry to add to the hash
2300 * Adds a dentry to the hash according to its name.
2303 void d_rehash(struct dentry
* entry
)
2305 spin_lock(&entry
->d_lock
);
2307 spin_unlock(&entry
->d_lock
);
2309 EXPORT_SYMBOL(d_rehash
);
2312 * dentry_update_name_case - update case insensitive dentry with a new name
2313 * @dentry: dentry to be updated
2316 * Update a case insensitive dentry with new case of name.
2318 * dentry must have been returned by d_lookup with name @name. Old and new
2319 * name lengths must match (ie. no d_compare which allows mismatched name
2322 * Parent inode i_mutex must be held over d_lookup and into this call (to
2323 * keep renames and concurrent inserts, and readdir(2) away).
2325 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2327 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2328 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2330 spin_lock(&dentry
->d_lock
);
2331 write_seqcount_begin(&dentry
->d_seq
);
2332 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2333 write_seqcount_end(&dentry
->d_seq
);
2334 spin_unlock(&dentry
->d_lock
);
2336 EXPORT_SYMBOL(dentry_update_name_case
);
2338 static void switch_names(struct dentry
*dentry
, struct dentry
*target
)
2340 if (dname_external(target
)) {
2341 if (dname_external(dentry
)) {
2343 * Both external: swap the pointers
2345 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2348 * dentry:internal, target:external. Steal target's
2349 * storage and make target internal.
2351 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2352 dentry
->d_name
.len
+ 1);
2353 dentry
->d_name
.name
= target
->d_name
.name
;
2354 target
->d_name
.name
= target
->d_iname
;
2357 if (dname_external(dentry
)) {
2359 * dentry:external, target:internal. Give dentry's
2360 * storage to target and make dentry internal
2362 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2363 target
->d_name
.len
+ 1);
2364 target
->d_name
.name
= dentry
->d_name
.name
;
2365 dentry
->d_name
.name
= dentry
->d_iname
;
2368 * Both are internal. Just copy target to dentry
2370 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2371 target
->d_name
.len
+ 1);
2372 dentry
->d_name
.len
= target
->d_name
.len
;
2376 swap(dentry
->d_name
.len
, target
->d_name
.len
);
2379 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2382 * XXXX: do we really need to take target->d_lock?
2384 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2385 spin_lock(&target
->d_parent
->d_lock
);
2387 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2388 spin_lock(&dentry
->d_parent
->d_lock
);
2389 spin_lock_nested(&target
->d_parent
->d_lock
,
2390 DENTRY_D_LOCK_NESTED
);
2392 spin_lock(&target
->d_parent
->d_lock
);
2393 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2394 DENTRY_D_LOCK_NESTED
);
2397 if (target
< dentry
) {
2398 spin_lock_nested(&target
->d_lock
, 2);
2399 spin_lock_nested(&dentry
->d_lock
, 3);
2401 spin_lock_nested(&dentry
->d_lock
, 2);
2402 spin_lock_nested(&target
->d_lock
, 3);
2406 static void dentry_unlock_parents_for_move(struct dentry
*dentry
,
2407 struct dentry
*target
)
2409 if (target
->d_parent
!= dentry
->d_parent
)
2410 spin_unlock(&dentry
->d_parent
->d_lock
);
2411 if (target
->d_parent
!= target
)
2412 spin_unlock(&target
->d_parent
->d_lock
);
2416 * When switching names, the actual string doesn't strictly have to
2417 * be preserved in the target - because we're dropping the target
2418 * anyway. As such, we can just do a simple memcpy() to copy over
2419 * the new name before we switch.
2421 * Note that we have to be a lot more careful about getting the hash
2422 * switched - we have to switch the hash value properly even if it
2423 * then no longer matches the actual (corrupted) string of the target.
2424 * The hash value has to match the hash queue that the dentry is on..
2427 * __d_move - move a dentry
2428 * @dentry: entry to move
2429 * @target: new dentry
2431 * Update the dcache to reflect the move of a file name. Negative
2432 * dcache entries should not be moved in this way. Caller must hold
2433 * rename_lock, the i_mutex of the source and target directories,
2434 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2436 static void __d_move(struct dentry
* dentry
, struct dentry
* target
)
2438 if (!dentry
->d_inode
)
2439 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2441 BUG_ON(d_ancestor(dentry
, target
));
2442 BUG_ON(d_ancestor(target
, dentry
));
2444 dentry_lock_for_move(dentry
, target
);
2446 write_seqcount_begin(&dentry
->d_seq
);
2447 write_seqcount_begin(&target
->d_seq
);
2449 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2452 * Move the dentry to the target hash queue. Don't bother checking
2453 * for the same hash queue because of how unlikely it is.
2456 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2458 /* Unhash the target: dput() will then get rid of it */
2461 list_del(&dentry
->d_u
.d_child
);
2462 list_del(&target
->d_u
.d_child
);
2464 /* Switch the names.. */
2465 switch_names(dentry
, target
);
2466 swap(dentry
->d_name
.hash
, target
->d_name
.hash
);
2468 /* ... and switch the parents */
2469 if (IS_ROOT(dentry
)) {
2470 dentry
->d_parent
= target
->d_parent
;
2471 target
->d_parent
= target
;
2472 INIT_LIST_HEAD(&target
->d_u
.d_child
);
2474 swap(dentry
->d_parent
, target
->d_parent
);
2476 /* And add them back to the (new) parent lists */
2477 list_add(&target
->d_u
.d_child
, &target
->d_parent
->d_subdirs
);
2480 list_add(&dentry
->d_u
.d_child
, &dentry
->d_parent
->d_subdirs
);
2482 write_seqcount_end(&target
->d_seq
);
2483 write_seqcount_end(&dentry
->d_seq
);
2485 dentry_unlock_parents_for_move(dentry
, target
);
2486 spin_unlock(&target
->d_lock
);
2487 fsnotify_d_move(dentry
);
2488 spin_unlock(&dentry
->d_lock
);
2492 * d_move - move a dentry
2493 * @dentry: entry to move
2494 * @target: new dentry
2496 * Update the dcache to reflect the move of a file name. Negative
2497 * dcache entries should not be moved in this way. See the locking
2498 * requirements for __d_move.
2500 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2502 write_seqlock(&rename_lock
);
2503 __d_move(dentry
, target
);
2504 write_sequnlock(&rename_lock
);
2506 EXPORT_SYMBOL(d_move
);
2509 * d_ancestor - search for an ancestor
2510 * @p1: ancestor dentry
2513 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2514 * an ancestor of p2, else NULL.
2516 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2520 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2521 if (p
->d_parent
== p1
)
2528 * This helper attempts to cope with remotely renamed directories
2530 * It assumes that the caller is already holding
2531 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2533 * Note: If ever the locking in lock_rename() changes, then please
2534 * remember to update this too...
2536 static struct dentry
*__d_unalias(struct inode
*inode
,
2537 struct dentry
*dentry
, struct dentry
*alias
)
2539 struct mutex
*m1
= NULL
, *m2
= NULL
;
2540 struct dentry
*ret
= ERR_PTR(-EBUSY
);
2542 /* If alias and dentry share a parent, then no extra locks required */
2543 if (alias
->d_parent
== dentry
->d_parent
)
2546 /* See lock_rename() */
2547 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2549 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2550 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2552 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2554 if (likely(!d_mountpoint(alias
))) {
2555 __d_move(alias
, dentry
);
2559 spin_unlock(&inode
->i_lock
);
2568 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2569 * named dentry in place of the dentry to be replaced.
2570 * returns with anon->d_lock held!
2572 static void __d_materialise_dentry(struct dentry
*dentry
, struct dentry
*anon
)
2574 struct dentry
*dparent
;
2576 dentry_lock_for_move(anon
, dentry
);
2578 write_seqcount_begin(&dentry
->d_seq
);
2579 write_seqcount_begin(&anon
->d_seq
);
2581 dparent
= dentry
->d_parent
;
2583 switch_names(dentry
, anon
);
2584 swap(dentry
->d_name
.hash
, anon
->d_name
.hash
);
2586 dentry
->d_parent
= dentry
;
2587 list_del_init(&dentry
->d_u
.d_child
);
2588 anon
->d_parent
= dparent
;
2589 list_move(&anon
->d_u
.d_child
, &dparent
->d_subdirs
);
2591 write_seqcount_end(&dentry
->d_seq
);
2592 write_seqcount_end(&anon
->d_seq
);
2594 dentry_unlock_parents_for_move(anon
, dentry
);
2595 spin_unlock(&dentry
->d_lock
);
2597 /* anon->d_lock still locked, returns locked */
2598 anon
->d_flags
&= ~DCACHE_DISCONNECTED
;
2602 * d_materialise_unique - introduce an inode into the tree
2603 * @dentry: candidate dentry
2604 * @inode: inode to bind to the dentry, to which aliases may be attached
2606 * Introduces an dentry into the tree, substituting an extant disconnected
2607 * root directory alias in its place if there is one. Caller must hold the
2608 * i_mutex of the parent directory.
2610 struct dentry
*d_materialise_unique(struct dentry
*dentry
, struct inode
*inode
)
2612 struct dentry
*actual
;
2614 BUG_ON(!d_unhashed(dentry
));
2618 __d_instantiate(dentry
, NULL
);
2623 spin_lock(&inode
->i_lock
);
2625 if (S_ISDIR(inode
->i_mode
)) {
2626 struct dentry
*alias
;
2628 /* Does an aliased dentry already exist? */
2629 alias
= __d_find_alias(inode
, 0);
2632 write_seqlock(&rename_lock
);
2634 if (d_ancestor(alias
, dentry
)) {
2635 /* Check for loops */
2636 actual
= ERR_PTR(-ELOOP
);
2637 spin_unlock(&inode
->i_lock
);
2638 } else if (IS_ROOT(alias
)) {
2639 /* Is this an anonymous mountpoint that we
2640 * could splice into our tree? */
2641 __d_materialise_dentry(dentry
, alias
);
2642 write_sequnlock(&rename_lock
);
2646 /* Nope, but we must(!) avoid directory
2647 * aliasing. This drops inode->i_lock */
2648 actual
= __d_unalias(inode
, dentry
, alias
);
2650 write_sequnlock(&rename_lock
);
2651 if (IS_ERR(actual
)) {
2652 if (PTR_ERR(actual
) == -ELOOP
)
2653 pr_warn_ratelimited(
2654 "VFS: Lookup of '%s' in %s %s"
2655 " would have caused loop\n",
2656 dentry
->d_name
.name
,
2657 inode
->i_sb
->s_type
->name
,
2665 /* Add a unique reference */
2666 actual
= __d_instantiate_unique(dentry
, inode
);
2670 BUG_ON(!d_unhashed(actual
));
2672 spin_lock(&actual
->d_lock
);
2675 spin_unlock(&actual
->d_lock
);
2676 spin_unlock(&inode
->i_lock
);
2678 if (actual
== dentry
) {
2679 security_d_instantiate(dentry
, inode
);
2686 EXPORT_SYMBOL_GPL(d_materialise_unique
);
2688 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2692 return -ENAMETOOLONG
;
2694 memcpy(*buffer
, str
, namelen
);
2699 * prepend_name - prepend a pathname in front of current buffer pointer
2700 * buffer: buffer pointer
2701 * buflen: allocated length of the buffer
2702 * name: name string and length qstr structure
2704 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2705 * make sure that either the old or the new name pointer and length are
2706 * fetched. However, there may be mismatch between length and pointer.
2707 * The length cannot be trusted, we need to copy it byte-by-byte until
2708 * the length is reached or a null byte is found. It also prepends "/" at
2709 * the beginning of the name. The sequence number check at the caller will
2710 * retry it again when a d_move() does happen. So any garbage in the buffer
2711 * due to mismatched pointer and length will be discarded.
2713 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2715 const char *dname
= ACCESS_ONCE(name
->name
);
2716 u32 dlen
= ACCESS_ONCE(name
->len
);
2719 if (*buflen
< dlen
+ 1)
2720 return -ENAMETOOLONG
;
2721 *buflen
-= dlen
+ 1;
2722 p
= *buffer
-= dlen
+ 1;
2734 * prepend_path - Prepend path string to a buffer
2735 * @path: the dentry/vfsmount to report
2736 * @root: root vfsmnt/dentry
2737 * @buffer: pointer to the end of the buffer
2738 * @buflen: pointer to buffer length
2740 * The function tries to write out the pathname without taking any lock other
2741 * than the RCU read lock to make sure that dentries won't go away. It only
2742 * checks the sequence number of the global rename_lock as any change in the
2743 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2744 * number. If the sequence number had been change, it will restart the whole
2745 * pathname back-tracing sequence again. It performs a total of 3 trials of
2746 * lockless back-tracing sequences before falling back to take the
2749 static int prepend_path(const struct path
*path
,
2750 const struct path
*root
,
2751 char **buffer
, int *buflen
)
2753 struct dentry
*dentry
= path
->dentry
;
2754 struct vfsmount
*vfsmnt
= path
->mnt
;
2755 struct mount
*mnt
= real_mount(vfsmnt
);
2765 read_seqbegin_or_lock(&rename_lock
, &seq
);
2766 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2767 struct dentry
* parent
;
2769 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2771 if (mnt_has_parent(mnt
)) {
2772 dentry
= mnt
->mnt_mountpoint
;
2773 mnt
= mnt
->mnt_parent
;
2778 * Filesystems needing to implement special "root names"
2779 * should do so with ->d_dname()
2781 if (IS_ROOT(dentry
) &&
2782 (dentry
->d_name
.len
!= 1 ||
2783 dentry
->d_name
.name
[0] != '/')) {
2784 WARN(1, "Root dentry has weird name <%.*s>\n",
2785 (int) dentry
->d_name
.len
,
2786 dentry
->d_name
.name
);
2789 error
= is_mounted(vfsmnt
) ? 1 : 2;
2792 parent
= dentry
->d_parent
;
2794 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
2802 if (need_seqretry(&rename_lock
, seq
)) {
2806 done_seqretry(&rename_lock
, seq
);
2808 if (error
>= 0 && bptr
== *buffer
) {
2810 error
= -ENAMETOOLONG
;
2820 * __d_path - return the path of a dentry
2821 * @path: the dentry/vfsmount to report
2822 * @root: root vfsmnt/dentry
2823 * @buf: buffer to return value in
2824 * @buflen: buffer length
2826 * Convert a dentry into an ASCII path name.
2828 * Returns a pointer into the buffer or an error code if the
2829 * path was too long.
2831 * "buflen" should be positive.
2833 * If the path is not reachable from the supplied root, return %NULL.
2835 char *__d_path(const struct path
*path
,
2836 const struct path
*root
,
2837 char *buf
, int buflen
)
2839 char *res
= buf
+ buflen
;
2842 prepend(&res
, &buflen
, "\0", 1);
2843 br_read_lock(&vfsmount_lock
);
2844 error
= prepend_path(path
, root
, &res
, &buflen
);
2845 br_read_unlock(&vfsmount_lock
);
2848 return ERR_PTR(error
);
2854 char *d_absolute_path(const struct path
*path
,
2855 char *buf
, int buflen
)
2857 struct path root
= {};
2858 char *res
= buf
+ buflen
;
2861 prepend(&res
, &buflen
, "\0", 1);
2862 br_read_lock(&vfsmount_lock
);
2863 error
= prepend_path(path
, &root
, &res
, &buflen
);
2864 br_read_unlock(&vfsmount_lock
);
2869 return ERR_PTR(error
);
2874 * same as __d_path but appends "(deleted)" for unlinked files.
2876 static int path_with_deleted(const struct path
*path
,
2877 const struct path
*root
,
2878 char **buf
, int *buflen
)
2880 prepend(buf
, buflen
, "\0", 1);
2881 if (d_unlinked(path
->dentry
)) {
2882 int error
= prepend(buf
, buflen
, " (deleted)", 10);
2887 return prepend_path(path
, root
, buf
, buflen
);
2890 static int prepend_unreachable(char **buffer
, int *buflen
)
2892 return prepend(buffer
, buflen
, "(unreachable)", 13);
2896 * d_path - return the path of a dentry
2897 * @path: path to report
2898 * @buf: buffer to return value in
2899 * @buflen: buffer length
2901 * Convert a dentry into an ASCII path name. If the entry has been deleted
2902 * the string " (deleted)" is appended. Note that this is ambiguous.
2904 * Returns a pointer into the buffer or an error code if the path was
2905 * too long. Note: Callers should use the returned pointer, not the passed
2906 * in buffer, to use the name! The implementation often starts at an offset
2907 * into the buffer, and may leave 0 bytes at the start.
2909 * "buflen" should be positive.
2911 char *d_path(const struct path
*path
, char *buf
, int buflen
)
2913 char *res
= buf
+ buflen
;
2918 * We have various synthetic filesystems that never get mounted. On
2919 * these filesystems dentries are never used for lookup purposes, and
2920 * thus don't need to be hashed. They also don't need a name until a
2921 * user wants to identify the object in /proc/pid/fd/. The little hack
2922 * below allows us to generate a name for these objects on demand:
2924 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
)
2925 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
2927 get_fs_root(current
->fs
, &root
);
2928 br_read_lock(&vfsmount_lock
);
2929 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
2930 br_read_unlock(&vfsmount_lock
);
2932 res
= ERR_PTR(error
);
2936 EXPORT_SYMBOL(d_path
);
2939 * Helper function for dentry_operations.d_dname() members
2941 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
2942 const char *fmt
, ...)
2948 va_start(args
, fmt
);
2949 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
2952 if (sz
> sizeof(temp
) || sz
> buflen
)
2953 return ERR_PTR(-ENAMETOOLONG
);
2955 buffer
+= buflen
- sz
;
2956 return memcpy(buffer
, temp
, sz
);
2959 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
2961 char *end
= buffer
+ buflen
;
2962 /* these dentries are never renamed, so d_lock is not needed */
2963 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
2964 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
2965 prepend(&end
, &buflen
, "/", 1))
2966 end
= ERR_PTR(-ENAMETOOLONG
);
2971 * Write full pathname from the root of the filesystem into the buffer.
2973 static char *__dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
2983 prepend(&end
, &len
, "\0", 1);
2989 read_seqbegin_or_lock(&rename_lock
, &seq
);
2990 while (!IS_ROOT(dentry
)) {
2991 struct dentry
*parent
= dentry
->d_parent
;
2995 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3004 if (need_seqretry(&rename_lock
, seq
)) {
3008 done_seqretry(&rename_lock
, seq
);
3013 return ERR_PTR(-ENAMETOOLONG
);
3016 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3018 return __dentry_path(dentry
, buf
, buflen
);
3020 EXPORT_SYMBOL(dentry_path_raw
);
3022 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3027 if (d_unlinked(dentry
)) {
3029 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3033 retval
= __dentry_path(dentry
, buf
, buflen
);
3034 if (!IS_ERR(retval
) && p
)
3035 *p
= '/'; /* restore '/' overriden with '\0' */
3038 return ERR_PTR(-ENAMETOOLONG
);
3042 * NOTE! The user-level library version returns a
3043 * character pointer. The kernel system call just
3044 * returns the length of the buffer filled (which
3045 * includes the ending '\0' character), or a negative
3046 * error value. So libc would do something like
3048 * char *getcwd(char * buf, size_t size)
3052 * retval = sys_getcwd(buf, size);
3059 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3062 struct path pwd
, root
;
3063 char *page
= (char *) __get_free_page(GFP_USER
);
3068 get_fs_root_and_pwd(current
->fs
, &root
, &pwd
);
3071 br_read_lock(&vfsmount_lock
);
3072 if (!d_unlinked(pwd
.dentry
)) {
3074 char *cwd
= page
+ PAGE_SIZE
;
3075 int buflen
= PAGE_SIZE
;
3077 prepend(&cwd
, &buflen
, "\0", 1);
3078 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3079 br_read_unlock(&vfsmount_lock
);
3084 /* Unreachable from current root */
3086 error
= prepend_unreachable(&cwd
, &buflen
);
3092 len
= PAGE_SIZE
+ page
- cwd
;
3095 if (copy_to_user(buf
, cwd
, len
))
3099 br_read_unlock(&vfsmount_lock
);
3105 free_page((unsigned long) page
);
3110 * Test whether new_dentry is a subdirectory of old_dentry.
3112 * Trivially implemented using the dcache structure
3116 * is_subdir - is new dentry a subdirectory of old_dentry
3117 * @new_dentry: new dentry
3118 * @old_dentry: old dentry
3120 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3121 * Returns 0 otherwise.
3122 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3125 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3130 if (new_dentry
== old_dentry
)
3134 /* for restarting inner loop in case of seq retry */
3135 seq
= read_seqbegin(&rename_lock
);
3137 * Need rcu_readlock to protect against the d_parent trashing
3141 if (d_ancestor(old_dentry
, new_dentry
))
3146 } while (read_seqretry(&rename_lock
, seq
));
3151 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3153 struct dentry
*root
= data
;
3154 if (dentry
!= root
) {
3155 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3158 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3159 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3160 dentry
->d_lockref
.count
--;
3163 return D_WALK_CONTINUE
;
3166 void d_genocide(struct dentry
*parent
)
3168 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3171 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3173 inode_dec_link_count(inode
);
3174 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3175 !hlist_unhashed(&dentry
->d_alias
) ||
3176 !d_unlinked(dentry
));
3177 spin_lock(&dentry
->d_parent
->d_lock
);
3178 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3179 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3180 (unsigned long long)inode
->i_ino
);
3181 spin_unlock(&dentry
->d_lock
);
3182 spin_unlock(&dentry
->d_parent
->d_lock
);
3183 d_instantiate(dentry
, inode
);
3185 EXPORT_SYMBOL(d_tmpfile
);
3187 static __initdata
unsigned long dhash_entries
;
3188 static int __init
set_dhash_entries(char *str
)
3192 dhash_entries
= simple_strtoul(str
, &str
, 0);
3195 __setup("dhash_entries=", set_dhash_entries
);
3197 static void __init
dcache_init_early(void)
3201 /* If hashes are distributed across NUMA nodes, defer
3202 * hash allocation until vmalloc space is available.
3208 alloc_large_system_hash("Dentry cache",
3209 sizeof(struct hlist_bl_head
),
3218 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3219 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3222 static void __init
dcache_init(void)
3227 * A constructor could be added for stable state like the lists,
3228 * but it is probably not worth it because of the cache nature
3231 dentry_cache
= KMEM_CACHE(dentry
,
3232 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3234 /* Hash may have been set up in dcache_init_early */
3239 alloc_large_system_hash("Dentry cache",
3240 sizeof(struct hlist_bl_head
),
3249 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3250 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3253 /* SLAB cache for __getname() consumers */
3254 struct kmem_cache
*names_cachep __read_mostly
;
3255 EXPORT_SYMBOL(names_cachep
);
3257 EXPORT_SYMBOL(d_genocide
);
3259 void __init
vfs_caches_init_early(void)
3261 dcache_init_early();
3265 void __init
vfs_caches_init(unsigned long mempages
)
3267 unsigned long reserve
;
3269 /* Base hash sizes on available memory, with a reserve equal to
3270 150% of current kernel size */
3272 reserve
= min((mempages
- nr_free_pages()) * 3/2, mempages
- 1);
3273 mempages
-= reserve
;
3275 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3276 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3280 files_init(mempages
);