4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
41 #include <linux/kasan.h>
48 * dcache->d_inode->i_lock protects:
49 * - i_dentry, d_u.d_alias, d_inode of aliases
50 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
54 * dentry->d_sb->s_dentry_lru_lock protects:
55 * - the dcache lru lists and counters
62 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
64 * - d_u.d_alias, d_inode
67 * dentry->d_inode->i_lock
69 * dentry->d_sb->s_dentry_lru_lock
70 * dcache_hash_bucket lock
73 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
76 * dentry->d_parent->d_lock
79 * If no ancestor relationship:
80 * if (dentry1 < dentry2)
84 int sysctl_vfs_cache_pressure __read_mostly
= 100;
85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
87 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
89 EXPORT_SYMBOL(rename_lock
);
91 static struct kmem_cache
*dentry_cache __read_mostly
;
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
102 static unsigned int d_hash_mask __read_mostly
;
103 static unsigned int d_hash_shift __read_mostly
;
105 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
107 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
110 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
111 return dentry_hashtable
+ hash_32(hash
, d_hash_shift
);
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat
= {
119 static DEFINE_PER_CPU(long, nr_dentry
);
120 static DEFINE_PER_CPU(long, nr_dentry_unused
);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
125 * Here we resort to our own counters instead of using generic per-cpu counters
126 * for consistency with what the vfs inode code does. We are expected to harvest
127 * better code and performance by having our own specialized counters.
129 * Please note that the loop is done over all possible CPUs, not over all online
130 * CPUs. The reason for this is that we don't want to play games with CPUs going
131 * on and off. If one of them goes off, we will just keep their counters.
133 * glommer: See cffbc8a for details, and if you ever intend to change this,
134 * please update all vfs counters to match.
136 static long get_nr_dentry(void)
140 for_each_possible_cpu(i
)
141 sum
+= per_cpu(nr_dentry
, i
);
142 return sum
< 0 ? 0 : sum
;
145 static long get_nr_dentry_unused(void)
149 for_each_possible_cpu(i
)
150 sum
+= per_cpu(nr_dentry_unused
, i
);
151 return sum
< 0 ? 0 : sum
;
154 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
155 size_t *lenp
, loff_t
*ppos
)
157 dentry_stat
.nr_dentry
= get_nr_dentry();
158 dentry_stat
.nr_unused
= get_nr_dentry_unused();
159 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
164 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165 * The strings are both count bytes long, and count is non-zero.
167 #ifdef CONFIG_DCACHE_WORD_ACCESS
169 #include <asm/word-at-a-time.h>
171 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172 * aligned allocation for this particular component. We don't
173 * strictly need the load_unaligned_zeropad() safety, but it
174 * doesn't hurt either.
176 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177 * need the careful unaligned handling.
179 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
181 unsigned long a
,b
,mask
;
184 a
= *(unsigned long *)cs
;
185 b
= load_unaligned_zeropad(ct
);
186 if (tcount
< sizeof(unsigned long))
188 if (unlikely(a
!= b
))
190 cs
+= sizeof(unsigned long);
191 ct
+= sizeof(unsigned long);
192 tcount
-= sizeof(unsigned long);
196 mask
= bytemask_from_count(tcount
);
197 return unlikely(!!((a
^ b
) & mask
));
202 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
216 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
218 const unsigned char *cs
;
220 * Be careful about RCU walk racing with rename:
221 * use ACCESS_ONCE to fetch the name pointer.
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
235 cs
= ACCESS_ONCE(dentry
->d_name
.name
);
236 smp_read_barrier_depends();
237 return dentry_string_cmp(cs
, ct
, tcount
);
240 struct external_name
{
243 struct rcu_head head
;
245 unsigned char name
[];
248 static inline struct external_name
*external_name(struct dentry
*dentry
)
250 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
253 static void __d_free(struct rcu_head
*head
)
255 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
257 kmem_cache_free(dentry_cache
, dentry
);
260 static void __d_free_external(struct rcu_head
*head
)
262 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
263 kfree(external_name(dentry
));
264 kmem_cache_free(dentry_cache
, dentry
);
267 static inline int dname_external(const struct dentry
*dentry
)
269 return dentry
->d_name
.name
!= dentry
->d_iname
;
272 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
274 spin_lock(&dentry
->d_lock
);
275 if (unlikely(dname_external(dentry
))) {
276 struct external_name
*p
= external_name(dentry
);
277 atomic_inc(&p
->u
.count
);
278 spin_unlock(&dentry
->d_lock
);
279 name
->name
= p
->name
;
281 memcpy(name
->inline_name
, dentry
->d_iname
, DNAME_INLINE_LEN
);
282 spin_unlock(&dentry
->d_lock
);
283 name
->name
= name
->inline_name
;
286 EXPORT_SYMBOL(take_dentry_name_snapshot
);
288 void release_dentry_name_snapshot(struct name_snapshot
*name
)
290 if (unlikely(name
->name
!= name
->inline_name
)) {
291 struct external_name
*p
;
292 p
= container_of(name
->name
, struct external_name
, name
[0]);
293 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
294 kfree_rcu(p
, u
.head
);
297 EXPORT_SYMBOL(release_dentry_name_snapshot
);
299 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
305 dentry
->d_inode
= inode
;
306 flags
= READ_ONCE(dentry
->d_flags
);
307 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
309 WRITE_ONCE(dentry
->d_flags
, flags
);
312 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
314 unsigned flags
= READ_ONCE(dentry
->d_flags
);
316 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
317 WRITE_ONCE(dentry
->d_flags
, flags
);
318 dentry
->d_inode
= NULL
;
321 static void dentry_free(struct dentry
*dentry
)
323 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
324 if (unlikely(dname_external(dentry
))) {
325 struct external_name
*p
= external_name(dentry
);
326 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
327 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
331 /* if dentry was never visible to RCU, immediate free is OK */
332 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
333 __d_free(&dentry
->d_u
.d_rcu
);
335 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
339 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
340 * @dentry: the target dentry
341 * After this call, in-progress rcu-walk path lookup will fail. This
342 * should be called after unhashing, and after changing d_inode (if
343 * the dentry has not already been unhashed).
345 static inline void dentry_rcuwalk_invalidate(struct dentry
*dentry
)
347 lockdep_assert_held(&dentry
->d_lock
);
348 /* Go through am invalidation barrier */
349 write_seqcount_invalidate(&dentry
->d_seq
);
353 * Release the dentry's inode, using the filesystem
354 * d_iput() operation if defined. Dentry has no refcount
357 static void dentry_iput(struct dentry
* dentry
)
358 __releases(dentry
->d_lock
)
359 __releases(dentry
->d_inode
->i_lock
)
361 struct inode
*inode
= dentry
->d_inode
;
363 __d_clear_type_and_inode(dentry
);
364 hlist_del_init(&dentry
->d_u
.d_alias
);
365 spin_unlock(&dentry
->d_lock
);
366 spin_unlock(&inode
->i_lock
);
368 fsnotify_inoderemove(inode
);
369 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
370 dentry
->d_op
->d_iput(dentry
, inode
);
374 spin_unlock(&dentry
->d_lock
);
379 * Release the dentry's inode, using the filesystem
380 * d_iput() operation if defined. dentry remains in-use.
382 static void dentry_unlink_inode(struct dentry
* dentry
)
383 __releases(dentry
->d_lock
)
384 __releases(dentry
->d_inode
->i_lock
)
386 struct inode
*inode
= dentry
->d_inode
;
388 raw_write_seqcount_begin(&dentry
->d_seq
);
389 __d_clear_type_and_inode(dentry
);
390 hlist_del_init(&dentry
->d_u
.d_alias
);
391 raw_write_seqcount_end(&dentry
->d_seq
);
392 spin_unlock(&dentry
->d_lock
);
393 spin_unlock(&inode
->i_lock
);
395 fsnotify_inoderemove(inode
);
396 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
397 dentry
->d_op
->d_iput(dentry
, inode
);
403 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
404 * is in use - which includes both the "real" per-superblock
405 * LRU list _and_ the DCACHE_SHRINK_LIST use.
407 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
408 * on the shrink list (ie not on the superblock LRU list).
410 * The per-cpu "nr_dentry_unused" counters are updated with
411 * the DCACHE_LRU_LIST bit.
413 * These helper functions make sure we always follow the
414 * rules. d_lock must be held by the caller.
416 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
417 static void d_lru_add(struct dentry
*dentry
)
419 D_FLAG_VERIFY(dentry
, 0);
420 dentry
->d_flags
|= DCACHE_LRU_LIST
;
421 this_cpu_inc(nr_dentry_unused
);
422 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
425 static void d_lru_del(struct dentry
*dentry
)
427 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
428 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
429 this_cpu_dec(nr_dentry_unused
);
430 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
433 static void d_shrink_del(struct dentry
*dentry
)
435 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
436 list_del_init(&dentry
->d_lru
);
437 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
438 this_cpu_dec(nr_dentry_unused
);
441 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
443 D_FLAG_VERIFY(dentry
, 0);
444 list_add(&dentry
->d_lru
, list
);
445 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
446 this_cpu_inc(nr_dentry_unused
);
450 * These can only be called under the global LRU lock, ie during the
451 * callback for freeing the LRU list. "isolate" removes it from the
452 * LRU lists entirely, while shrink_move moves it to the indicated
455 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
457 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
458 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
459 this_cpu_dec(nr_dentry_unused
);
460 list_lru_isolate(lru
, &dentry
->d_lru
);
463 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
464 struct list_head
*list
)
466 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
467 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
468 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
472 * dentry_lru_(add|del)_list) must be called with d_lock held.
474 static void dentry_lru_add(struct dentry
*dentry
)
476 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
481 * d_drop - drop a dentry
482 * @dentry: dentry to drop
484 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
485 * be found through a VFS lookup any more. Note that this is different from
486 * deleting the dentry - d_delete will try to mark the dentry negative if
487 * possible, giving a successful _negative_ lookup, while d_drop will
488 * just make the cache lookup fail.
490 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
491 * reason (NFS timeouts or autofs deletes).
493 * __d_drop requires dentry->d_lock.
495 void __d_drop(struct dentry
*dentry
)
497 if (!d_unhashed(dentry
)) {
498 struct hlist_bl_head
*b
;
500 * Hashed dentries are normally on the dentry hashtable,
501 * with the exception of those newly allocated by
502 * d_obtain_alias, which are always IS_ROOT:
504 if (unlikely(IS_ROOT(dentry
)))
505 b
= &dentry
->d_sb
->s_anon
;
507 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
510 __hlist_bl_del(&dentry
->d_hash
);
511 dentry
->d_hash
.pprev
= NULL
;
513 dentry_rcuwalk_invalidate(dentry
);
516 EXPORT_SYMBOL(__d_drop
);
518 void d_drop(struct dentry
*dentry
)
520 spin_lock(&dentry
->d_lock
);
522 spin_unlock(&dentry
->d_lock
);
524 EXPORT_SYMBOL(d_drop
);
526 static void __dentry_kill(struct dentry
*dentry
)
528 struct dentry
*parent
= NULL
;
529 bool can_free
= true;
530 if (!IS_ROOT(dentry
))
531 parent
= dentry
->d_parent
;
534 * The dentry is now unrecoverably dead to the world.
536 lockref_mark_dead(&dentry
->d_lockref
);
539 * inform the fs via d_prune that this dentry is about to be
540 * unhashed and destroyed.
542 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
543 dentry
->d_op
->d_prune(dentry
);
545 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
546 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
549 /* if it was on the hash then remove it */
551 __list_del_entry(&dentry
->d_child
);
553 * Inform d_walk() that we are no longer attached to the
556 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
558 spin_unlock(&parent
->d_lock
);
561 * dentry_iput drops the locks, at which point nobody (except
562 * transient RCU lookups) can reach this dentry.
564 BUG_ON(dentry
->d_lockref
.count
> 0);
565 this_cpu_dec(nr_dentry
);
566 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
567 dentry
->d_op
->d_release(dentry
);
569 spin_lock(&dentry
->d_lock
);
570 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
571 dentry
->d_flags
|= DCACHE_MAY_FREE
;
574 spin_unlock(&dentry
->d_lock
);
575 if (likely(can_free
))
580 * Finish off a dentry we've decided to kill.
581 * dentry->d_lock must be held, returns with it unlocked.
582 * If ref is non-zero, then decrement the refcount too.
583 * Returns dentry requiring refcount drop, or NULL if we're done.
585 static struct dentry
*dentry_kill(struct dentry
*dentry
)
586 __releases(dentry
->d_lock
)
588 struct inode
*inode
= dentry
->d_inode
;
589 struct dentry
*parent
= NULL
;
591 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
594 if (!IS_ROOT(dentry
)) {
595 parent
= dentry
->d_parent
;
596 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
598 spin_unlock(&inode
->i_lock
);
603 __dentry_kill(dentry
);
607 spin_unlock(&dentry
->d_lock
);
608 return dentry
; /* try again with same dentry */
611 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
613 struct dentry
*parent
= dentry
->d_parent
;
616 if (unlikely(dentry
->d_lockref
.count
< 0))
618 if (likely(spin_trylock(&parent
->d_lock
)))
621 spin_unlock(&dentry
->d_lock
);
623 parent
= ACCESS_ONCE(dentry
->d_parent
);
624 spin_lock(&parent
->d_lock
);
626 * We can't blindly lock dentry until we are sure
627 * that we won't violate the locking order.
628 * Any changes of dentry->d_parent must have
629 * been done with parent->d_lock held, so
630 * spin_lock() above is enough of a barrier
631 * for checking if it's still our child.
633 if (unlikely(parent
!= dentry
->d_parent
)) {
634 spin_unlock(&parent
->d_lock
);
638 if (parent
!= dentry
)
639 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
646 * Try to do a lockless dput(), and return whether that was successful.
648 * If unsuccessful, we return false, having already taken the dentry lock.
650 * The caller needs to hold the RCU read lock, so that the dentry is
651 * guaranteed to stay around even if the refcount goes down to zero!
653 static inline bool fast_dput(struct dentry
*dentry
)
656 unsigned int d_flags
;
659 * If we have a d_op->d_delete() operation, we sould not
660 * let the dentry count go to zero, so use "put_or_lock".
662 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
663 return lockref_put_or_lock(&dentry
->d_lockref
);
666 * .. otherwise, we can try to just decrement the
667 * lockref optimistically.
669 ret
= lockref_put_return(&dentry
->d_lockref
);
672 * If the lockref_put_return() failed due to the lock being held
673 * by somebody else, the fast path has failed. We will need to
674 * get the lock, and then check the count again.
676 if (unlikely(ret
< 0)) {
677 spin_lock(&dentry
->d_lock
);
678 if (dentry
->d_lockref
.count
> 1) {
679 dentry
->d_lockref
.count
--;
680 spin_unlock(&dentry
->d_lock
);
687 * If we weren't the last ref, we're done.
693 * Careful, careful. The reference count went down
694 * to zero, but we don't hold the dentry lock, so
695 * somebody else could get it again, and do another
696 * dput(), and we need to not race with that.
698 * However, there is a very special and common case
699 * where we don't care, because there is nothing to
700 * do: the dentry is still hashed, it does not have
701 * a 'delete' op, and it's referenced and already on
704 * NOTE! Since we aren't locked, these values are
705 * not "stable". However, it is sufficient that at
706 * some point after we dropped the reference the
707 * dentry was hashed and the flags had the proper
708 * value. Other dentry users may have re-gotten
709 * a reference to the dentry and change that, but
710 * our work is done - we can leave the dentry
711 * around with a zero refcount.
714 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
715 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
717 /* Nothing to do? Dropping the reference was all we needed? */
718 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
722 * Not the fast normal case? Get the lock. We've already decremented
723 * the refcount, but we'll need to re-check the situation after
726 spin_lock(&dentry
->d_lock
);
729 * Did somebody else grab a reference to it in the meantime, and
730 * we're no longer the last user after all? Alternatively, somebody
731 * else could have killed it and marked it dead. Either way, we
732 * don't need to do anything else.
734 if (dentry
->d_lockref
.count
) {
735 spin_unlock(&dentry
->d_lock
);
740 * Re-get the reference we optimistically dropped. We hold the
741 * lock, and we just tested that it was zero, so we can just
744 dentry
->d_lockref
.count
= 1;
752 * This is complicated by the fact that we do not want to put
753 * dentries that are no longer on any hash chain on the unused
754 * list: we'd much rather just get rid of them immediately.
756 * However, that implies that we have to traverse the dentry
757 * tree upwards to the parents which might _also_ now be
758 * scheduled for deletion (it may have been only waiting for
759 * its last child to go away).
761 * This tail recursion is done by hand as we don't want to depend
762 * on the compiler to always get this right (gcc generally doesn't).
763 * Real recursion would eat up our stack space.
767 * dput - release a dentry
768 * @dentry: dentry to release
770 * Release a dentry. This will drop the usage count and if appropriate
771 * call the dentry unlink method as well as removing it from the queues and
772 * releasing its resources. If the parent dentries were scheduled for release
773 * they too may now get deleted.
775 void dput(struct dentry
*dentry
)
777 if (unlikely(!dentry
))
784 if (likely(fast_dput(dentry
))) {
789 /* Slow case: now with the dentry lock held */
792 /* Unreachable? Get rid of it */
793 if (unlikely(d_unhashed(dentry
)))
796 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
799 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
800 if (dentry
->d_op
->d_delete(dentry
))
804 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
805 dentry
->d_flags
|= DCACHE_REFERENCED
;
806 dentry_lru_add(dentry
);
808 dentry
->d_lockref
.count
--;
809 spin_unlock(&dentry
->d_lock
);
813 dentry
= dentry_kill(dentry
);
822 /* This must be called with d_lock held */
823 static inline void __dget_dlock(struct dentry
*dentry
)
825 dentry
->d_lockref
.count
++;
828 static inline void __dget(struct dentry
*dentry
)
830 lockref_get(&dentry
->d_lockref
);
833 struct dentry
*dget_parent(struct dentry
*dentry
)
839 * Do optimistic parent lookup without any
843 ret
= ACCESS_ONCE(dentry
->d_parent
);
844 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
846 if (likely(gotref
)) {
847 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
854 * Don't need rcu_dereference because we re-check it was correct under
858 ret
= dentry
->d_parent
;
859 spin_lock(&ret
->d_lock
);
860 if (unlikely(ret
!= dentry
->d_parent
)) {
861 spin_unlock(&ret
->d_lock
);
866 BUG_ON(!ret
->d_lockref
.count
);
867 ret
->d_lockref
.count
++;
868 spin_unlock(&ret
->d_lock
);
871 EXPORT_SYMBOL(dget_parent
);
874 * d_find_alias - grab a hashed alias of inode
875 * @inode: inode in question
877 * If inode has a hashed alias, or is a directory and has any alias,
878 * acquire the reference to alias and return it. Otherwise return NULL.
879 * Notice that if inode is a directory there can be only one alias and
880 * it can be unhashed only if it has no children, or if it is the root
881 * of a filesystem, or if the directory was renamed and d_revalidate
882 * was the first vfs operation to notice.
884 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
885 * any other hashed alias over that one.
887 static struct dentry
*__d_find_alias(struct inode
*inode
)
889 struct dentry
*alias
, *discon_alias
;
893 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
894 spin_lock(&alias
->d_lock
);
895 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
896 if (IS_ROOT(alias
) &&
897 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
898 discon_alias
= alias
;
901 spin_unlock(&alias
->d_lock
);
905 spin_unlock(&alias
->d_lock
);
908 alias
= discon_alias
;
909 spin_lock(&alias
->d_lock
);
910 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
912 spin_unlock(&alias
->d_lock
);
915 spin_unlock(&alias
->d_lock
);
921 struct dentry
*d_find_alias(struct inode
*inode
)
923 struct dentry
*de
= NULL
;
925 if (!hlist_empty(&inode
->i_dentry
)) {
926 spin_lock(&inode
->i_lock
);
927 de
= __d_find_alias(inode
);
928 spin_unlock(&inode
->i_lock
);
932 EXPORT_SYMBOL(d_find_alias
);
935 * Try to kill dentries associated with this inode.
936 * WARNING: you must own a reference to inode.
938 void d_prune_aliases(struct inode
*inode
)
940 struct dentry
*dentry
;
942 spin_lock(&inode
->i_lock
);
943 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
944 spin_lock(&dentry
->d_lock
);
945 if (!dentry
->d_lockref
.count
) {
946 struct dentry
*parent
= lock_parent(dentry
);
947 if (likely(!dentry
->d_lockref
.count
)) {
948 __dentry_kill(dentry
);
953 spin_unlock(&parent
->d_lock
);
955 spin_unlock(&dentry
->d_lock
);
957 spin_unlock(&inode
->i_lock
);
959 EXPORT_SYMBOL(d_prune_aliases
);
961 static void shrink_dentry_list(struct list_head
*list
)
963 struct dentry
*dentry
, *parent
;
965 while (!list_empty(list
)) {
967 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
968 spin_lock(&dentry
->d_lock
);
969 parent
= lock_parent(dentry
);
972 * The dispose list is isolated and dentries are not accounted
973 * to the LRU here, so we can simply remove it from the list
974 * here regardless of whether it is referenced or not.
976 d_shrink_del(dentry
);
979 * We found an inuse dentry which was not removed from
980 * the LRU because of laziness during lookup. Do not free it.
982 if (dentry
->d_lockref
.count
> 0) {
983 spin_unlock(&dentry
->d_lock
);
985 spin_unlock(&parent
->d_lock
);
990 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
991 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
992 spin_unlock(&dentry
->d_lock
);
994 spin_unlock(&parent
->d_lock
);
1000 inode
= dentry
->d_inode
;
1001 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1002 d_shrink_add(dentry
, list
);
1003 spin_unlock(&dentry
->d_lock
);
1005 spin_unlock(&parent
->d_lock
);
1009 __dentry_kill(dentry
);
1012 * We need to prune ancestors too. This is necessary to prevent
1013 * quadratic behavior of shrink_dcache_parent(), but is also
1014 * expected to be beneficial in reducing dentry cache
1018 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1019 parent
= lock_parent(dentry
);
1020 if (dentry
->d_lockref
.count
!= 1) {
1021 dentry
->d_lockref
.count
--;
1022 spin_unlock(&dentry
->d_lock
);
1024 spin_unlock(&parent
->d_lock
);
1027 inode
= dentry
->d_inode
; /* can't be NULL */
1028 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1029 spin_unlock(&dentry
->d_lock
);
1031 spin_unlock(&parent
->d_lock
);
1035 __dentry_kill(dentry
);
1041 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1042 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1044 struct list_head
*freeable
= arg
;
1045 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1049 * we are inverting the lru lock/dentry->d_lock here,
1050 * so use a trylock. If we fail to get the lock, just skip
1053 if (!spin_trylock(&dentry
->d_lock
))
1057 * Referenced dentries are still in use. If they have active
1058 * counts, just remove them from the LRU. Otherwise give them
1059 * another pass through the LRU.
1061 if (dentry
->d_lockref
.count
) {
1062 d_lru_isolate(lru
, dentry
);
1063 spin_unlock(&dentry
->d_lock
);
1067 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1068 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1069 spin_unlock(&dentry
->d_lock
);
1072 * The list move itself will be made by the common LRU code. At
1073 * this point, we've dropped the dentry->d_lock but keep the
1074 * lru lock. This is safe to do, since every list movement is
1075 * protected by the lru lock even if both locks are held.
1077 * This is guaranteed by the fact that all LRU management
1078 * functions are intermediated by the LRU API calls like
1079 * list_lru_add and list_lru_del. List movement in this file
1080 * only ever occur through this functions or through callbacks
1081 * like this one, that are called from the LRU API.
1083 * The only exceptions to this are functions like
1084 * shrink_dentry_list, and code that first checks for the
1085 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1086 * operating only with stack provided lists after they are
1087 * properly isolated from the main list. It is thus, always a
1093 d_lru_shrink_move(lru
, dentry
, freeable
);
1094 spin_unlock(&dentry
->d_lock
);
1100 * prune_dcache_sb - shrink the dcache
1102 * @sc: shrink control, passed to list_lru_shrink_walk()
1104 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1105 * is done when we need more memory and called from the superblock shrinker
1108 * This function may fail to free any resources if all the dentries are in
1111 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1116 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1117 dentry_lru_isolate
, &dispose
);
1118 shrink_dentry_list(&dispose
);
1122 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1123 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1125 struct list_head
*freeable
= arg
;
1126 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1129 * we are inverting the lru lock/dentry->d_lock here,
1130 * so use a trylock. If we fail to get the lock, just skip
1133 if (!spin_trylock(&dentry
->d_lock
))
1136 d_lru_shrink_move(lru
, dentry
, freeable
);
1137 spin_unlock(&dentry
->d_lock
);
1144 * shrink_dcache_sb - shrink dcache for a superblock
1147 * Shrink the dcache for the specified super block. This is used to free
1148 * the dcache before unmounting a file system.
1150 void shrink_dcache_sb(struct super_block
*sb
)
1157 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1158 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1160 this_cpu_sub(nr_dentry_unused
, freed
);
1161 shrink_dentry_list(&dispose
);
1162 } while (freed
> 0);
1164 EXPORT_SYMBOL(shrink_dcache_sb
);
1167 * enum d_walk_ret - action to talke during tree walk
1168 * @D_WALK_CONTINUE: contrinue walk
1169 * @D_WALK_QUIT: quit walk
1170 * @D_WALK_NORETRY: quit when retry is needed
1171 * @D_WALK_SKIP: skip this dentry and its children
1181 * d_walk - walk the dentry tree
1182 * @parent: start of walk
1183 * @data: data passed to @enter() and @finish()
1184 * @enter: callback when first entering the dentry
1185 * @finish: callback when successfully finished the walk
1187 * The @enter() and @finish() callbacks are called with d_lock held.
1189 void d_walk(struct dentry
*parent
, void *data
,
1190 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1191 void (*finish
)(void *))
1193 struct dentry
*this_parent
;
1194 struct list_head
*next
;
1196 enum d_walk_ret ret
;
1200 read_seqbegin_or_lock(&rename_lock
, &seq
);
1201 this_parent
= parent
;
1202 spin_lock(&this_parent
->d_lock
);
1204 ret
= enter(data
, this_parent
);
1206 case D_WALK_CONTINUE
:
1211 case D_WALK_NORETRY
:
1216 next
= this_parent
->d_subdirs
.next
;
1218 while (next
!= &this_parent
->d_subdirs
) {
1219 struct list_head
*tmp
= next
;
1220 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1223 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1225 ret
= enter(data
, dentry
);
1227 case D_WALK_CONTINUE
:
1230 spin_unlock(&dentry
->d_lock
);
1232 case D_WALK_NORETRY
:
1236 spin_unlock(&dentry
->d_lock
);
1240 if (!list_empty(&dentry
->d_subdirs
)) {
1241 spin_unlock(&this_parent
->d_lock
);
1242 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1243 this_parent
= dentry
;
1244 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1247 spin_unlock(&dentry
->d_lock
);
1250 * All done at this level ... ascend and resume the search.
1254 if (this_parent
!= parent
) {
1255 struct dentry
*child
= this_parent
;
1256 this_parent
= child
->d_parent
;
1258 spin_unlock(&child
->d_lock
);
1259 spin_lock(&this_parent
->d_lock
);
1261 /* might go back up the wrong parent if we have had a rename. */
1262 if (need_seqretry(&rename_lock
, seq
))
1264 /* go into the first sibling still alive */
1266 next
= child
->d_child
.next
;
1267 if (next
== &this_parent
->d_subdirs
)
1269 child
= list_entry(next
, struct dentry
, d_child
);
1270 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1274 if (need_seqretry(&rename_lock
, seq
))
1281 spin_unlock(&this_parent
->d_lock
);
1282 done_seqretry(&rename_lock
, seq
);
1286 spin_unlock(&this_parent
->d_lock
);
1294 EXPORT_SYMBOL(d_walk
);
1297 * Search for at least 1 mount point in the dentry's subdirs.
1298 * We descend to the next level whenever the d_subdirs
1299 * list is non-empty and continue searching.
1302 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1305 if (d_mountpoint(dentry
)) {
1309 return D_WALK_CONTINUE
;
1313 * have_submounts - check for mounts over a dentry
1314 * @parent: dentry to check.
1316 * Return true if the parent or its subdirectories contain
1319 int have_submounts(struct dentry
*parent
)
1323 d_walk(parent
, &ret
, check_mount
, NULL
);
1327 EXPORT_SYMBOL(have_submounts
);
1330 * Called by mount code to set a mountpoint and check if the mountpoint is
1331 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1332 * subtree can become unreachable).
1334 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1335 * this reason take rename_lock and d_lock on dentry and ancestors.
1337 int d_set_mounted(struct dentry
*dentry
)
1341 write_seqlock(&rename_lock
);
1342 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1343 /* Need exclusion wrt. d_invalidate() */
1344 spin_lock(&p
->d_lock
);
1345 if (unlikely(d_unhashed(p
))) {
1346 spin_unlock(&p
->d_lock
);
1349 spin_unlock(&p
->d_lock
);
1351 spin_lock(&dentry
->d_lock
);
1352 if (!d_unlinked(dentry
)) {
1354 if (!d_mountpoint(dentry
)) {
1355 dentry
->d_flags
|= DCACHE_MOUNTED
;
1359 spin_unlock(&dentry
->d_lock
);
1361 write_sequnlock(&rename_lock
);
1366 * Search the dentry child list of the specified parent,
1367 * and move any unused dentries to the end of the unused
1368 * list for prune_dcache(). We descend to the next level
1369 * whenever the d_subdirs list is non-empty and continue
1372 * It returns zero iff there are no unused children,
1373 * otherwise it returns the number of children moved to
1374 * the end of the unused list. This may not be the total
1375 * number of unused children, because select_parent can
1376 * drop the lock and return early due to latency
1380 struct select_data
{
1381 struct dentry
*start
;
1382 struct list_head dispose
;
1386 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1388 struct select_data
*data
= _data
;
1389 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1391 if (data
->start
== dentry
)
1394 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1397 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1399 if (!dentry
->d_lockref
.count
) {
1400 d_shrink_add(dentry
, &data
->dispose
);
1405 * We can return to the caller if we have found some (this
1406 * ensures forward progress). We'll be coming back to find
1409 if (!list_empty(&data
->dispose
))
1410 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1416 * shrink_dcache_parent - prune dcache
1417 * @parent: parent of entries to prune
1419 * Prune the dcache to remove unused children of the parent dentry.
1421 void shrink_dcache_parent(struct dentry
*parent
)
1424 struct select_data data
;
1426 INIT_LIST_HEAD(&data
.dispose
);
1427 data
.start
= parent
;
1430 d_walk(parent
, &data
, select_collect
, NULL
);
1434 shrink_dentry_list(&data
.dispose
);
1438 EXPORT_SYMBOL(shrink_dcache_parent
);
1440 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1442 /* it has busy descendents; complain about those instead */
1443 if (!list_empty(&dentry
->d_subdirs
))
1444 return D_WALK_CONTINUE
;
1446 /* root with refcount 1 is fine */
1447 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1448 return D_WALK_CONTINUE
;
1450 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1451 " still in use (%d) [unmount of %s %s]\n",
1454 dentry
->d_inode
->i_ino
: 0UL,
1456 dentry
->d_lockref
.count
,
1457 dentry
->d_sb
->s_type
->name
,
1458 dentry
->d_sb
->s_id
);
1460 return D_WALK_CONTINUE
;
1463 static void do_one_tree(struct dentry
*dentry
)
1465 shrink_dcache_parent(dentry
);
1466 d_walk(dentry
, dentry
, umount_check
, NULL
);
1472 * destroy the dentries attached to a superblock on unmounting
1474 void shrink_dcache_for_umount(struct super_block
*sb
)
1476 struct dentry
*dentry
;
1478 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1480 dentry
= sb
->s_root
;
1482 do_one_tree(dentry
);
1484 while (!hlist_bl_empty(&sb
->s_anon
)) {
1485 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1486 do_one_tree(dentry
);
1490 struct detach_data
{
1491 struct select_data select
;
1492 struct dentry
*mountpoint
;
1494 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1496 struct detach_data
*data
= _data
;
1498 if (d_mountpoint(dentry
)) {
1499 __dget_dlock(dentry
);
1500 data
->mountpoint
= dentry
;
1504 return select_collect(&data
->select
, dentry
);
1507 static void check_and_drop(void *_data
)
1509 struct detach_data
*data
= _data
;
1511 if (!data
->mountpoint
&& !data
->select
.found
)
1512 __d_drop(data
->select
.start
);
1516 * d_invalidate - detach submounts, prune dcache, and drop
1517 * @dentry: dentry to invalidate (aka detach, prune and drop)
1521 * The final d_drop is done as an atomic operation relative to
1522 * rename_lock ensuring there are no races with d_set_mounted. This
1523 * ensures there are no unhashed dentries on the path to a mountpoint.
1525 void d_invalidate(struct dentry
*dentry
)
1528 * If it's already been dropped, return OK.
1530 spin_lock(&dentry
->d_lock
);
1531 if (d_unhashed(dentry
)) {
1532 spin_unlock(&dentry
->d_lock
);
1535 spin_unlock(&dentry
->d_lock
);
1537 /* Negative dentries can be dropped without further checks */
1538 if (!dentry
->d_inode
) {
1544 struct detach_data data
;
1546 data
.mountpoint
= NULL
;
1547 INIT_LIST_HEAD(&data
.select
.dispose
);
1548 data
.select
.start
= dentry
;
1549 data
.select
.found
= 0;
1551 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1553 if (data
.select
.found
)
1554 shrink_dentry_list(&data
.select
.dispose
);
1556 if (data
.mountpoint
) {
1557 detach_mounts(data
.mountpoint
);
1558 dput(data
.mountpoint
);
1561 if (!data
.mountpoint
&& !data
.select
.found
)
1567 EXPORT_SYMBOL(d_invalidate
);
1570 * __d_alloc - allocate a dcache entry
1571 * @sb: filesystem it will belong to
1572 * @name: qstr of the name
1574 * Allocates a dentry. It returns %NULL if there is insufficient memory
1575 * available. On a success the dentry is returned. The name passed in is
1576 * copied and the copy passed in may be reused after this call.
1579 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1581 struct dentry
*dentry
;
1584 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1589 * We guarantee that the inline name is always NUL-terminated.
1590 * This way the memcpy() done by the name switching in rename
1591 * will still always have a NUL at the end, even if we might
1592 * be overwriting an internal NUL character
1594 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1595 if (name
->len
> DNAME_INLINE_LEN
-1) {
1596 size_t size
= offsetof(struct external_name
, name
[1]);
1597 struct external_name
*p
= kmalloc(size
+ name
->len
, GFP_KERNEL
);
1599 kmem_cache_free(dentry_cache
, dentry
);
1602 atomic_set(&p
->u
.count
, 1);
1604 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1605 kasan_unpoison_shadow(dname
,
1606 round_up(name
->len
+ 1, sizeof(unsigned long)));
1608 dname
= dentry
->d_iname
;
1611 dentry
->d_name
.len
= name
->len
;
1612 dentry
->d_name
.hash
= name
->hash
;
1613 memcpy(dname
, name
->name
, name
->len
);
1614 dname
[name
->len
] = 0;
1616 /* Make sure we always see the terminating NUL character */
1618 dentry
->d_name
.name
= dname
;
1620 dentry
->d_lockref
.count
= 1;
1621 dentry
->d_flags
= 0;
1622 spin_lock_init(&dentry
->d_lock
);
1623 seqcount_init(&dentry
->d_seq
);
1624 dentry
->d_inode
= NULL
;
1625 dentry
->d_parent
= dentry
;
1627 dentry
->d_op
= NULL
;
1628 dentry
->d_fsdata
= NULL
;
1629 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1630 INIT_LIST_HEAD(&dentry
->d_lru
);
1631 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1632 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1633 INIT_LIST_HEAD(&dentry
->d_child
);
1634 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1636 this_cpu_inc(nr_dentry
);
1642 * d_alloc - allocate a dcache entry
1643 * @parent: parent of entry to allocate
1644 * @name: qstr of the name
1646 * Allocates a dentry. It returns %NULL if there is insufficient memory
1647 * available. On a success the dentry is returned. The name passed in is
1648 * copied and the copy passed in may be reused after this call.
1650 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1652 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1655 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1656 spin_lock(&parent
->d_lock
);
1658 * don't need child lock because it is not subject
1659 * to concurrency here
1661 __dget_dlock(parent
);
1662 dentry
->d_parent
= parent
;
1663 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1664 spin_unlock(&parent
->d_lock
);
1668 EXPORT_SYMBOL(d_alloc
);
1671 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1672 * @sb: the superblock
1673 * @name: qstr of the name
1675 * For a filesystem that just pins its dentries in memory and never
1676 * performs lookups at all, return an unhashed IS_ROOT dentry.
1678 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1680 return __d_alloc(sb
, name
);
1682 EXPORT_SYMBOL(d_alloc_pseudo
);
1684 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1689 q
.len
= strlen(name
);
1690 q
.hash
= full_name_hash(q
.name
, q
.len
);
1691 return d_alloc(parent
, &q
);
1693 EXPORT_SYMBOL(d_alloc_name
);
1695 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1697 WARN_ON_ONCE(dentry
->d_op
);
1698 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1700 DCACHE_OP_REVALIDATE
|
1701 DCACHE_OP_WEAK_REVALIDATE
|
1703 DCACHE_OP_SELECT_INODE
|
1709 dentry
->d_flags
|= DCACHE_OP_HASH
;
1711 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1712 if (op
->d_revalidate
)
1713 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1714 if (op
->d_weak_revalidate
)
1715 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1717 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1719 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1720 if (op
->d_select_inode
)
1721 dentry
->d_flags
|= DCACHE_OP_SELECT_INODE
;
1723 dentry
->d_flags
|= DCACHE_OP_REAL
;
1726 EXPORT_SYMBOL(d_set_d_op
);
1730 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1731 * @dentry - The dentry to mark
1733 * Mark a dentry as falling through to the lower layer (as set with
1734 * d_pin_lower()). This flag may be recorded on the medium.
1736 void d_set_fallthru(struct dentry
*dentry
)
1738 spin_lock(&dentry
->d_lock
);
1739 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1740 spin_unlock(&dentry
->d_lock
);
1742 EXPORT_SYMBOL(d_set_fallthru
);
1744 static unsigned d_flags_for_inode(struct inode
*inode
)
1746 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1749 return DCACHE_MISS_TYPE
;
1751 if (S_ISDIR(inode
->i_mode
)) {
1752 add_flags
= DCACHE_DIRECTORY_TYPE
;
1753 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1754 if (unlikely(!inode
->i_op
->lookup
))
1755 add_flags
= DCACHE_AUTODIR_TYPE
;
1757 inode
->i_opflags
|= IOP_LOOKUP
;
1759 goto type_determined
;
1762 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1763 if (unlikely(inode
->i_op
->follow_link
)) {
1764 add_flags
= DCACHE_SYMLINK_TYPE
;
1765 goto type_determined
;
1767 inode
->i_opflags
|= IOP_NOFOLLOW
;
1770 if (unlikely(!S_ISREG(inode
->i_mode
)))
1771 add_flags
= DCACHE_SPECIAL_TYPE
;
1774 if (unlikely(IS_AUTOMOUNT(inode
)))
1775 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1779 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1781 unsigned add_flags
= d_flags_for_inode(inode
);
1783 spin_lock(&dentry
->d_lock
);
1785 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1786 raw_write_seqcount_begin(&dentry
->d_seq
);
1787 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1788 raw_write_seqcount_end(&dentry
->d_seq
);
1789 spin_unlock(&dentry
->d_lock
);
1790 fsnotify_d_instantiate(dentry
, inode
);
1794 * d_instantiate - fill in inode information for a dentry
1795 * @entry: dentry to complete
1796 * @inode: inode to attach to this dentry
1798 * Fill in inode information in the entry.
1800 * This turns negative dentries into productive full members
1803 * NOTE! This assumes that the inode count has been incremented
1804 * (or otherwise set) by the caller to indicate that it is now
1805 * in use by the dcache.
1808 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1810 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1812 spin_lock(&inode
->i_lock
);
1813 __d_instantiate(entry
, inode
);
1815 spin_unlock(&inode
->i_lock
);
1816 security_d_instantiate(entry
, inode
);
1818 EXPORT_SYMBOL(d_instantiate
);
1821 * d_instantiate_unique - instantiate a non-aliased dentry
1822 * @entry: dentry to instantiate
1823 * @inode: inode to attach to this dentry
1825 * Fill in inode information in the entry. On success, it returns NULL.
1826 * If an unhashed alias of "entry" already exists, then we return the
1827 * aliased dentry instead and drop one reference to inode.
1829 * Note that in order to avoid conflicts with rename() etc, the caller
1830 * had better be holding the parent directory semaphore.
1832 * This also assumes that the inode count has been incremented
1833 * (or otherwise set) by the caller to indicate that it is now
1834 * in use by the dcache.
1836 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1837 struct inode
*inode
)
1839 struct dentry
*alias
;
1840 int len
= entry
->d_name
.len
;
1841 const char *name
= entry
->d_name
.name
;
1842 unsigned int hash
= entry
->d_name
.hash
;
1845 __d_instantiate(entry
, NULL
);
1849 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1851 * Don't need alias->d_lock here, because aliases with
1852 * d_parent == entry->d_parent are not subject to name or
1853 * parent changes, because the parent inode i_mutex is held.
1855 if (alias
->d_name
.hash
!= hash
)
1857 if (alias
->d_parent
!= entry
->d_parent
)
1859 if (alias
->d_name
.len
!= len
)
1861 if (dentry_cmp(alias
, name
, len
))
1867 __d_instantiate(entry
, inode
);
1871 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1873 struct dentry
*result
;
1875 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1878 spin_lock(&inode
->i_lock
);
1879 result
= __d_instantiate_unique(entry
, inode
);
1881 spin_unlock(&inode
->i_lock
);
1884 security_d_instantiate(entry
, inode
);
1888 BUG_ON(!d_unhashed(result
));
1893 EXPORT_SYMBOL(d_instantiate_unique
);
1896 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1897 * @entry: dentry to complete
1898 * @inode: inode to attach to this dentry
1900 * Fill in inode information in the entry. If a directory alias is found, then
1901 * return an error (and drop inode). Together with d_materialise_unique() this
1902 * guarantees that a directory inode may never have more than one alias.
1904 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1906 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1908 spin_lock(&inode
->i_lock
);
1909 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1910 spin_unlock(&inode
->i_lock
);
1914 __d_instantiate(entry
, inode
);
1915 spin_unlock(&inode
->i_lock
);
1916 security_d_instantiate(entry
, inode
);
1920 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1922 struct dentry
*d_make_root(struct inode
*root_inode
)
1924 struct dentry
*res
= NULL
;
1927 static const struct qstr name
= QSTR_INIT("/", 1);
1929 res
= __d_alloc(root_inode
->i_sb
, &name
);
1931 d_instantiate(res
, root_inode
);
1937 EXPORT_SYMBOL(d_make_root
);
1939 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1941 struct dentry
*alias
;
1943 if (hlist_empty(&inode
->i_dentry
))
1945 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1951 * d_find_any_alias - find any alias for a given inode
1952 * @inode: inode to find an alias for
1954 * If any aliases exist for the given inode, take and return a
1955 * reference for one of them. If no aliases exist, return %NULL.
1957 struct dentry
*d_find_any_alias(struct inode
*inode
)
1961 spin_lock(&inode
->i_lock
);
1962 de
= __d_find_any_alias(inode
);
1963 spin_unlock(&inode
->i_lock
);
1966 EXPORT_SYMBOL(d_find_any_alias
);
1968 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1970 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1976 return ERR_PTR(-ESTALE
);
1978 return ERR_CAST(inode
);
1980 res
= d_find_any_alias(inode
);
1984 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1986 res
= ERR_PTR(-ENOMEM
);
1990 spin_lock(&inode
->i_lock
);
1991 res
= __d_find_any_alias(inode
);
1993 spin_unlock(&inode
->i_lock
);
1998 /* attach a disconnected dentry */
1999 add_flags
= d_flags_for_inode(inode
);
2002 add_flags
|= DCACHE_DISCONNECTED
;
2004 spin_lock(&tmp
->d_lock
);
2005 __d_set_inode_and_type(tmp
, inode
, add_flags
);
2006 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
2007 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
2008 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
2009 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
2010 spin_unlock(&tmp
->d_lock
);
2011 spin_unlock(&inode
->i_lock
);
2012 security_d_instantiate(tmp
, inode
);
2017 if (res
&& !IS_ERR(res
))
2018 security_d_instantiate(res
, inode
);
2024 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2025 * @inode: inode to allocate the dentry for
2027 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2028 * similar open by handle operations. The returned dentry may be anonymous,
2029 * or may have a full name (if the inode was already in the cache).
2031 * When called on a directory inode, we must ensure that the inode only ever
2032 * has one dentry. If a dentry is found, that is returned instead of
2033 * allocating a new one.
2035 * On successful return, the reference to the inode has been transferred
2036 * to the dentry. In case of an error the reference on the inode is released.
2037 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2038 * be passed in and the error will be propagated to the return value,
2039 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2041 struct dentry
*d_obtain_alias(struct inode
*inode
)
2043 return __d_obtain_alias(inode
, 1);
2045 EXPORT_SYMBOL(d_obtain_alias
);
2048 * d_obtain_root - find or allocate a dentry for a given inode
2049 * @inode: inode to allocate the dentry for
2051 * Obtain an IS_ROOT dentry for the root of a filesystem.
2053 * We must ensure that directory inodes only ever have one dentry. If a
2054 * dentry is found, that is returned instead of allocating a new one.
2056 * On successful return, the reference to the inode has been transferred
2057 * to the dentry. In case of an error the reference on the inode is
2058 * released. A %NULL or IS_ERR inode may be passed in and will be the
2059 * error will be propagate to the return value, with a %NULL @inode
2060 * replaced by ERR_PTR(-ESTALE).
2062 struct dentry
*d_obtain_root(struct inode
*inode
)
2064 return __d_obtain_alias(inode
, 0);
2066 EXPORT_SYMBOL(d_obtain_root
);
2069 * d_add_ci - lookup or allocate new dentry with case-exact name
2070 * @inode: the inode case-insensitive lookup has found
2071 * @dentry: the negative dentry that was passed to the parent's lookup func
2072 * @name: the case-exact name to be associated with the returned dentry
2074 * This is to avoid filling the dcache with case-insensitive names to the
2075 * same inode, only the actual correct case is stored in the dcache for
2076 * case-insensitive filesystems.
2078 * For a case-insensitive lookup match and if the the case-exact dentry
2079 * already exists in in the dcache, use it and return it.
2081 * If no entry exists with the exact case name, allocate new dentry with
2082 * the exact case, and return the spliced entry.
2084 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2087 struct dentry
*found
;
2091 * First check if a dentry matching the name already exists,
2092 * if not go ahead and create it now.
2094 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2096 new = d_alloc(dentry
->d_parent
, name
);
2098 found
= ERR_PTR(-ENOMEM
);
2100 found
= d_splice_alias(inode
, new);
2111 EXPORT_SYMBOL(d_add_ci
);
2114 * Do the slow-case of the dentry name compare.
2116 * Unlike the dentry_cmp() function, we need to atomically
2117 * load the name and length information, so that the
2118 * filesystem can rely on them, and can use the 'name' and
2119 * 'len' information without worrying about walking off the
2120 * end of memory etc.
2122 * Thus the read_seqcount_retry() and the "duplicate" info
2123 * in arguments (the low-level filesystem should not look
2124 * at the dentry inode or name contents directly, since
2125 * rename can change them while we're in RCU mode).
2127 enum slow_d_compare
{
2133 static noinline
enum slow_d_compare
slow_dentry_cmp(
2134 const struct dentry
*parent
,
2135 struct dentry
*dentry
,
2137 const struct qstr
*name
)
2139 int tlen
= dentry
->d_name
.len
;
2140 const char *tname
= dentry
->d_name
.name
;
2142 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2144 return D_COMP_SEQRETRY
;
2146 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2147 return D_COMP_NOMATCH
;
2152 * __d_lookup_rcu - search for a dentry (racy, store-free)
2153 * @parent: parent dentry
2154 * @name: qstr of name we wish to find
2155 * @seqp: returns d_seq value at the point where the dentry was found
2156 * Returns: dentry, or NULL
2158 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2159 * resolution (store-free path walking) design described in
2160 * Documentation/filesystems/path-lookup.txt.
2162 * This is not to be used outside core vfs.
2164 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2165 * held, and rcu_read_lock held. The returned dentry must not be stored into
2166 * without taking d_lock and checking d_seq sequence count against @seq
2169 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2172 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2173 * the returned dentry, so long as its parent's seqlock is checked after the
2174 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2175 * is formed, giving integrity down the path walk.
2177 * NOTE! The caller *has* to check the resulting dentry against the sequence
2178 * number we've returned before using any of the resulting dentry state!
2180 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2181 const struct qstr
*name
,
2184 u64 hashlen
= name
->hash_len
;
2185 const unsigned char *str
= name
->name
;
2186 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2187 struct hlist_bl_node
*node
;
2188 struct dentry
*dentry
;
2191 * Note: There is significant duplication with __d_lookup_rcu which is
2192 * required to prevent single threaded performance regressions
2193 * especially on architectures where smp_rmb (in seqcounts) are costly.
2194 * Keep the two functions in sync.
2198 * The hash list is protected using RCU.
2200 * Carefully use d_seq when comparing a candidate dentry, to avoid
2201 * races with d_move().
2203 * It is possible that concurrent renames can mess up our list
2204 * walk here and result in missing our dentry, resulting in the
2205 * false-negative result. d_lookup() protects against concurrent
2206 * renames using rename_lock seqlock.
2208 * See Documentation/filesystems/path-lookup.txt for more details.
2210 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2215 * The dentry sequence count protects us from concurrent
2216 * renames, and thus protects parent and name fields.
2218 * The caller must perform a seqcount check in order
2219 * to do anything useful with the returned dentry.
2221 * NOTE! We do a "raw" seqcount_begin here. That means that
2222 * we don't wait for the sequence count to stabilize if it
2223 * is in the middle of a sequence change. If we do the slow
2224 * dentry compare, we will do seqretries until it is stable,
2225 * and if we end up with a successful lookup, we actually
2226 * want to exit RCU lookup anyway.
2228 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2229 if (dentry
->d_parent
!= parent
)
2231 if (d_unhashed(dentry
))
2234 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2235 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2238 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2241 case D_COMP_NOMATCH
:
2248 if (dentry
->d_name
.hash_len
!= hashlen
)
2251 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2258 * d_lookup - search for a dentry
2259 * @parent: parent dentry
2260 * @name: qstr of name we wish to find
2261 * Returns: dentry, or NULL
2263 * d_lookup searches the children of the parent dentry for the name in
2264 * question. If the dentry is found its reference count is incremented and the
2265 * dentry is returned. The caller must use dput to free the entry when it has
2266 * finished using it. %NULL is returned if the dentry does not exist.
2268 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2270 struct dentry
*dentry
;
2274 seq
= read_seqbegin(&rename_lock
);
2275 dentry
= __d_lookup(parent
, name
);
2278 } while (read_seqretry(&rename_lock
, seq
));
2281 EXPORT_SYMBOL(d_lookup
);
2284 * __d_lookup - search for a dentry (racy)
2285 * @parent: parent dentry
2286 * @name: qstr of name we wish to find
2287 * Returns: dentry, or NULL
2289 * __d_lookup is like d_lookup, however it may (rarely) return a
2290 * false-negative result due to unrelated rename activity.
2292 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2293 * however it must be used carefully, eg. with a following d_lookup in
2294 * the case of failure.
2296 * __d_lookup callers must be commented.
2298 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2300 unsigned int len
= name
->len
;
2301 unsigned int hash
= name
->hash
;
2302 const unsigned char *str
= name
->name
;
2303 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2304 struct hlist_bl_node
*node
;
2305 struct dentry
*found
= NULL
;
2306 struct dentry
*dentry
;
2309 * Note: There is significant duplication with __d_lookup_rcu which is
2310 * required to prevent single threaded performance regressions
2311 * especially on architectures where smp_rmb (in seqcounts) are costly.
2312 * Keep the two functions in sync.
2316 * The hash list is protected using RCU.
2318 * Take d_lock when comparing a candidate dentry, to avoid races
2321 * It is possible that concurrent renames can mess up our list
2322 * walk here and result in missing our dentry, resulting in the
2323 * false-negative result. d_lookup() protects against concurrent
2324 * renames using rename_lock seqlock.
2326 * See Documentation/filesystems/path-lookup.txt for more details.
2330 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2332 if (dentry
->d_name
.hash
!= hash
)
2335 spin_lock(&dentry
->d_lock
);
2336 if (dentry
->d_parent
!= parent
)
2338 if (d_unhashed(dentry
))
2342 * It is safe to compare names since d_move() cannot
2343 * change the qstr (protected by d_lock).
2345 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2346 int tlen
= dentry
->d_name
.len
;
2347 const char *tname
= dentry
->d_name
.name
;
2348 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2351 if (dentry
->d_name
.len
!= len
)
2353 if (dentry_cmp(dentry
, str
, len
))
2357 dentry
->d_lockref
.count
++;
2359 spin_unlock(&dentry
->d_lock
);
2362 spin_unlock(&dentry
->d_lock
);
2370 * d_hash_and_lookup - hash the qstr then search for a dentry
2371 * @dir: Directory to search in
2372 * @name: qstr of name we wish to find
2374 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2376 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2379 * Check for a fs-specific hash function. Note that we must
2380 * calculate the standard hash first, as the d_op->d_hash()
2381 * routine may choose to leave the hash value unchanged.
2383 name
->hash
= full_name_hash(name
->name
, name
->len
);
2384 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2385 int err
= dir
->d_op
->d_hash(dir
, name
);
2386 if (unlikely(err
< 0))
2387 return ERR_PTR(err
);
2389 return d_lookup(dir
, name
);
2391 EXPORT_SYMBOL(d_hash_and_lookup
);
2394 * When a file is deleted, we have two options:
2395 * - turn this dentry into a negative dentry
2396 * - unhash this dentry and free it.
2398 * Usually, we want to just turn this into
2399 * a negative dentry, but if anybody else is
2400 * currently using the dentry or the inode
2401 * we can't do that and we fall back on removing
2402 * it from the hash queues and waiting for
2403 * it to be deleted later when it has no users
2407 * d_delete - delete a dentry
2408 * @dentry: The dentry to delete
2410 * Turn the dentry into a negative dentry if possible, otherwise
2411 * remove it from the hash queues so it can be deleted later
2414 void d_delete(struct dentry
* dentry
)
2416 struct inode
*inode
;
2419 * Are we the only user?
2422 spin_lock(&dentry
->d_lock
);
2423 inode
= dentry
->d_inode
;
2424 isdir
= S_ISDIR(inode
->i_mode
);
2425 if (dentry
->d_lockref
.count
== 1) {
2426 if (!spin_trylock(&inode
->i_lock
)) {
2427 spin_unlock(&dentry
->d_lock
);
2431 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2432 dentry_unlink_inode(dentry
);
2433 fsnotify_nameremove(dentry
, isdir
);
2437 if (!d_unhashed(dentry
))
2440 spin_unlock(&dentry
->d_lock
);
2442 fsnotify_nameremove(dentry
, isdir
);
2444 EXPORT_SYMBOL(d_delete
);
2446 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2448 BUG_ON(!d_unhashed(entry
));
2450 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2454 static void _d_rehash(struct dentry
* entry
)
2456 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2460 * d_rehash - add an entry back to the hash
2461 * @entry: dentry to add to the hash
2463 * Adds a dentry to the hash according to its name.
2466 void d_rehash(struct dentry
* entry
)
2468 spin_lock(&entry
->d_lock
);
2470 spin_unlock(&entry
->d_lock
);
2472 EXPORT_SYMBOL(d_rehash
);
2475 * dentry_update_name_case - update case insensitive dentry with a new name
2476 * @dentry: dentry to be updated
2479 * Update a case insensitive dentry with new case of name.
2481 * dentry must have been returned by d_lookup with name @name. Old and new
2482 * name lengths must match (ie. no d_compare which allows mismatched name
2485 * Parent inode i_mutex must be held over d_lookup and into this call (to
2486 * keep renames and concurrent inserts, and readdir(2) away).
2488 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2490 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2491 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2493 spin_lock(&dentry
->d_lock
);
2494 write_seqcount_begin(&dentry
->d_seq
);
2495 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2496 write_seqcount_end(&dentry
->d_seq
);
2497 spin_unlock(&dentry
->d_lock
);
2499 EXPORT_SYMBOL(dentry_update_name_case
);
2501 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2503 if (unlikely(dname_external(target
))) {
2504 if (unlikely(dname_external(dentry
))) {
2506 * Both external: swap the pointers
2508 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2511 * dentry:internal, target:external. Steal target's
2512 * storage and make target internal.
2514 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2515 dentry
->d_name
.len
+ 1);
2516 dentry
->d_name
.name
= target
->d_name
.name
;
2517 target
->d_name
.name
= target
->d_iname
;
2520 if (unlikely(dname_external(dentry
))) {
2522 * dentry:external, target:internal. Give dentry's
2523 * storage to target and make dentry internal
2525 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2526 target
->d_name
.len
+ 1);
2527 target
->d_name
.name
= dentry
->d_name
.name
;
2528 dentry
->d_name
.name
= dentry
->d_iname
;
2531 * Both are internal.
2534 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2535 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2536 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2537 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2538 swap(((long *) &dentry
->d_iname
)[i
],
2539 ((long *) &target
->d_iname
)[i
]);
2543 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2546 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2548 struct external_name
*old_name
= NULL
;
2549 if (unlikely(dname_external(dentry
)))
2550 old_name
= external_name(dentry
);
2551 if (unlikely(dname_external(target
))) {
2552 atomic_inc(&external_name(target
)->u
.count
);
2553 dentry
->d_name
= target
->d_name
;
2555 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2556 target
->d_name
.len
+ 1);
2557 dentry
->d_name
.name
= dentry
->d_iname
;
2558 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2560 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2561 kfree_rcu(old_name
, u
.head
);
2564 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2567 * XXXX: do we really need to take target->d_lock?
2569 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2570 spin_lock(&target
->d_parent
->d_lock
);
2572 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2573 spin_lock(&dentry
->d_parent
->d_lock
);
2574 spin_lock_nested(&target
->d_parent
->d_lock
,
2575 DENTRY_D_LOCK_NESTED
);
2577 spin_lock(&target
->d_parent
->d_lock
);
2578 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2579 DENTRY_D_LOCK_NESTED
);
2582 if (target
< dentry
) {
2583 spin_lock_nested(&target
->d_lock
, 2);
2584 spin_lock_nested(&dentry
->d_lock
, 3);
2586 spin_lock_nested(&dentry
->d_lock
, 2);
2587 spin_lock_nested(&target
->d_lock
, 3);
2591 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2593 if (target
->d_parent
!= dentry
->d_parent
)
2594 spin_unlock(&dentry
->d_parent
->d_lock
);
2595 if (target
->d_parent
!= target
)
2596 spin_unlock(&target
->d_parent
->d_lock
);
2597 spin_unlock(&target
->d_lock
);
2598 spin_unlock(&dentry
->d_lock
);
2602 * When switching names, the actual string doesn't strictly have to
2603 * be preserved in the target - because we're dropping the target
2604 * anyway. As such, we can just do a simple memcpy() to copy over
2605 * the new name before we switch, unless we are going to rehash
2606 * it. Note that if we *do* unhash the target, we are not allowed
2607 * to rehash it without giving it a new name/hash key - whether
2608 * we swap or overwrite the names here, resulting name won't match
2609 * the reality in filesystem; it's only there for d_path() purposes.
2610 * Note that all of this is happening under rename_lock, so the
2611 * any hash lookup seeing it in the middle of manipulations will
2612 * be discarded anyway. So we do not care what happens to the hash
2616 * __d_move - move a dentry
2617 * @dentry: entry to move
2618 * @target: new dentry
2619 * @exchange: exchange the two dentries
2621 * Update the dcache to reflect the move of a file name. Negative
2622 * dcache entries should not be moved in this way. Caller must hold
2623 * rename_lock, the i_mutex of the source and target directories,
2624 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2626 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2629 if (!dentry
->d_inode
)
2630 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2632 BUG_ON(d_ancestor(dentry
, target
));
2633 BUG_ON(d_ancestor(target
, dentry
));
2635 dentry_lock_for_move(dentry
, target
);
2637 write_seqcount_begin(&dentry
->d_seq
);
2638 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2640 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2643 * Move the dentry to the target hash queue. Don't bother checking
2644 * for the same hash queue because of how unlikely it is.
2647 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2650 * Unhash the target (d_delete() is not usable here). If exchanging
2651 * the two dentries, then rehash onto the other's hash queue.
2656 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2659 /* Switch the names.. */
2661 swap_names(dentry
, target
);
2663 copy_name(dentry
, target
);
2665 /* ... and switch them in the tree */
2666 if (IS_ROOT(dentry
)) {
2667 /* splicing a tree */
2668 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2669 dentry
->d_parent
= target
->d_parent
;
2670 target
->d_parent
= target
;
2671 list_del_init(&target
->d_child
);
2672 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2674 /* swapping two dentries */
2675 swap(dentry
->d_parent
, target
->d_parent
);
2676 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2677 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2679 fsnotify_d_move(target
);
2680 fsnotify_d_move(dentry
);
2683 write_seqcount_end(&target
->d_seq
);
2684 write_seqcount_end(&dentry
->d_seq
);
2686 dentry_unlock_for_move(dentry
, target
);
2690 * d_move - move a dentry
2691 * @dentry: entry to move
2692 * @target: new dentry
2694 * Update the dcache to reflect the move of a file name. Negative
2695 * dcache entries should not be moved in this way. See the locking
2696 * requirements for __d_move.
2698 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2700 write_seqlock(&rename_lock
);
2701 __d_move(dentry
, target
, false);
2702 write_sequnlock(&rename_lock
);
2704 EXPORT_SYMBOL(d_move
);
2707 * d_exchange - exchange two dentries
2708 * @dentry1: first dentry
2709 * @dentry2: second dentry
2711 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2713 write_seqlock(&rename_lock
);
2715 WARN_ON(!dentry1
->d_inode
);
2716 WARN_ON(!dentry2
->d_inode
);
2717 WARN_ON(IS_ROOT(dentry1
));
2718 WARN_ON(IS_ROOT(dentry2
));
2720 __d_move(dentry1
, dentry2
, true);
2722 write_sequnlock(&rename_lock
);
2726 * d_ancestor - search for an ancestor
2727 * @p1: ancestor dentry
2730 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2731 * an ancestor of p2, else NULL.
2733 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2737 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2738 if (p
->d_parent
== p1
)
2745 * This helper attempts to cope with remotely renamed directories
2747 * It assumes that the caller is already holding
2748 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2750 * Note: If ever the locking in lock_rename() changes, then please
2751 * remember to update this too...
2753 static int __d_unalias(struct inode
*inode
,
2754 struct dentry
*dentry
, struct dentry
*alias
)
2756 struct mutex
*m1
= NULL
, *m2
= NULL
;
2759 /* If alias and dentry share a parent, then no extra locks required */
2760 if (alias
->d_parent
== dentry
->d_parent
)
2763 /* See lock_rename() */
2764 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2766 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2767 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2769 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2771 __d_move(alias
, dentry
, false);
2782 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2783 * @inode: the inode which may have a disconnected dentry
2784 * @dentry: a negative dentry which we want to point to the inode.
2786 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2787 * place of the given dentry and return it, else simply d_add the inode
2788 * to the dentry and return NULL.
2790 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2791 * we should error out: directories can't have multiple aliases.
2793 * This is needed in the lookup routine of any filesystem that is exportable
2794 * (via knfsd) so that we can build dcache paths to directories effectively.
2796 * If a dentry was found and moved, then it is returned. Otherwise NULL
2797 * is returned. This matches the expected return value of ->lookup.
2799 * Cluster filesystems may call this function with a negative, hashed dentry.
2800 * In that case, we know that the inode will be a regular file, and also this
2801 * will only occur during atomic_open. So we need to check for the dentry
2802 * being already hashed only in the final case.
2804 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2807 return ERR_CAST(inode
);
2809 BUG_ON(!d_unhashed(dentry
));
2812 __d_instantiate(dentry
, NULL
);
2815 spin_lock(&inode
->i_lock
);
2816 if (S_ISDIR(inode
->i_mode
)) {
2817 struct dentry
*new = __d_find_any_alias(inode
);
2818 if (unlikely(new)) {
2819 /* The reference to new ensures it remains an alias */
2820 spin_unlock(&inode
->i_lock
);
2821 write_seqlock(&rename_lock
);
2822 if (unlikely(d_ancestor(new, dentry
))) {
2823 write_sequnlock(&rename_lock
);
2825 new = ERR_PTR(-ELOOP
);
2826 pr_warn_ratelimited(
2827 "VFS: Lookup of '%s' in %s %s"
2828 " would have caused loop\n",
2829 dentry
->d_name
.name
,
2830 inode
->i_sb
->s_type
->name
,
2832 } else if (!IS_ROOT(new)) {
2833 int err
= __d_unalias(inode
, dentry
, new);
2834 write_sequnlock(&rename_lock
);
2840 __d_move(new, dentry
, false);
2841 write_sequnlock(&rename_lock
);
2842 security_d_instantiate(new, inode
);
2848 /* already taking inode->i_lock, so d_add() by hand */
2849 __d_instantiate(dentry
, inode
);
2850 spin_unlock(&inode
->i_lock
);
2852 security_d_instantiate(dentry
, inode
);
2856 EXPORT_SYMBOL(d_splice_alias
);
2858 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2862 return -ENAMETOOLONG
;
2864 memcpy(*buffer
, str
, namelen
);
2869 * prepend_name - prepend a pathname in front of current buffer pointer
2870 * @buffer: buffer pointer
2871 * @buflen: allocated length of the buffer
2872 * @name: name string and length qstr structure
2874 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2875 * make sure that either the old or the new name pointer and length are
2876 * fetched. However, there may be mismatch between length and pointer.
2877 * The length cannot be trusted, we need to copy it byte-by-byte until
2878 * the length is reached or a null byte is found. It also prepends "/" at
2879 * the beginning of the name. The sequence number check at the caller will
2880 * retry it again when a d_move() does happen. So any garbage in the buffer
2881 * due to mismatched pointer and length will be discarded.
2883 * Data dependency barrier is needed to make sure that we see that terminating
2884 * NUL. Alpha strikes again, film at 11...
2886 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2888 const char *dname
= ACCESS_ONCE(name
->name
);
2889 u32 dlen
= ACCESS_ONCE(name
->len
);
2892 smp_read_barrier_depends();
2894 *buflen
-= dlen
+ 1;
2896 return -ENAMETOOLONG
;
2897 p
= *buffer
-= dlen
+ 1;
2909 * prepend_path - Prepend path string to a buffer
2910 * @path: the dentry/vfsmount to report
2911 * @root: root vfsmnt/dentry
2912 * @buffer: pointer to the end of the buffer
2913 * @buflen: pointer to buffer length
2915 * The function will first try to write out the pathname without taking any
2916 * lock other than the RCU read lock to make sure that dentries won't go away.
2917 * It only checks the sequence number of the global rename_lock as any change
2918 * in the dentry's d_seq will be preceded by changes in the rename_lock
2919 * sequence number. If the sequence number had been changed, it will restart
2920 * the whole pathname back-tracing sequence again by taking the rename_lock.
2921 * In this case, there is no need to take the RCU read lock as the recursive
2922 * parent pointer references will keep the dentry chain alive as long as no
2923 * rename operation is performed.
2925 static int prepend_path(const struct path
*path
,
2926 const struct path
*root
,
2927 char **buffer
, int *buflen
)
2929 struct dentry
*dentry
;
2930 struct vfsmount
*vfsmnt
;
2933 unsigned seq
, m_seq
= 0;
2939 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
2946 dentry
= path
->dentry
;
2948 mnt
= real_mount(vfsmnt
);
2949 read_seqbegin_or_lock(&rename_lock
, &seq
);
2950 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2951 struct dentry
* parent
;
2953 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2954 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
2956 if (dentry
!= vfsmnt
->mnt_root
) {
2963 if (mnt
!= parent
) {
2964 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
2970 error
= is_mounted(vfsmnt
) ? 1 : 2;
2973 parent
= dentry
->d_parent
;
2975 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
2983 if (need_seqretry(&rename_lock
, seq
)) {
2987 done_seqretry(&rename_lock
, seq
);
2991 if (need_seqretry(&mount_lock
, m_seq
)) {
2995 done_seqretry(&mount_lock
, m_seq
);
2997 if (error
>= 0 && bptr
== *buffer
) {
2999 error
= -ENAMETOOLONG
;
3009 * __d_path - return the path of a dentry
3010 * @path: the dentry/vfsmount to report
3011 * @root: root vfsmnt/dentry
3012 * @buf: buffer to return value in
3013 * @buflen: buffer length
3015 * Convert a dentry into an ASCII path name.
3017 * Returns a pointer into the buffer or an error code if the
3018 * path was too long.
3020 * "buflen" should be positive.
3022 * If the path is not reachable from the supplied root, return %NULL.
3024 char *__d_path(const struct path
*path
,
3025 const struct path
*root
,
3026 char *buf
, int buflen
)
3028 char *res
= buf
+ buflen
;
3031 prepend(&res
, &buflen
, "\0", 1);
3032 error
= prepend_path(path
, root
, &res
, &buflen
);
3035 return ERR_PTR(error
);
3041 char *d_absolute_path(const struct path
*path
,
3042 char *buf
, int buflen
)
3044 struct path root
= {};
3045 char *res
= buf
+ buflen
;
3048 prepend(&res
, &buflen
, "\0", 1);
3049 error
= prepend_path(path
, &root
, &res
, &buflen
);
3054 return ERR_PTR(error
);
3059 * same as __d_path but appends "(deleted)" for unlinked files.
3061 static int path_with_deleted(const struct path
*path
,
3062 const struct path
*root
,
3063 char **buf
, int *buflen
)
3065 prepend(buf
, buflen
, "\0", 1);
3066 if (d_unlinked(path
->dentry
)) {
3067 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3072 return prepend_path(path
, root
, buf
, buflen
);
3075 static int prepend_unreachable(char **buffer
, int *buflen
)
3077 return prepend(buffer
, buflen
, "(unreachable)", 13);
3080 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3085 seq
= read_seqcount_begin(&fs
->seq
);
3087 } while (read_seqcount_retry(&fs
->seq
, seq
));
3091 * d_path - return the path of a dentry
3092 * @path: path to report
3093 * @buf: buffer to return value in
3094 * @buflen: buffer length
3096 * Convert a dentry into an ASCII path name. If the entry has been deleted
3097 * the string " (deleted)" is appended. Note that this is ambiguous.
3099 * Returns a pointer into the buffer or an error code if the path was
3100 * too long. Note: Callers should use the returned pointer, not the passed
3101 * in buffer, to use the name! The implementation often starts at an offset
3102 * into the buffer, and may leave 0 bytes at the start.
3104 * "buflen" should be positive.
3106 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3108 char *res
= buf
+ buflen
;
3113 * We have various synthetic filesystems that never get mounted. On
3114 * these filesystems dentries are never used for lookup purposes, and
3115 * thus don't need to be hashed. They also don't need a name until a
3116 * user wants to identify the object in /proc/pid/fd/. The little hack
3117 * below allows us to generate a name for these objects on demand:
3119 * Some pseudo inodes are mountable. When they are mounted
3120 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3121 * and instead have d_path return the mounted path.
3123 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3124 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3125 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3128 get_fs_root_rcu(current
->fs
, &root
);
3129 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3133 res
= ERR_PTR(error
);
3136 EXPORT_SYMBOL(d_path
);
3139 * Helper function for dentry_operations.d_dname() members
3141 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3142 const char *fmt
, ...)
3148 va_start(args
, fmt
);
3149 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3152 if (sz
> sizeof(temp
) || sz
> buflen
)
3153 return ERR_PTR(-ENAMETOOLONG
);
3155 buffer
+= buflen
- sz
;
3156 return memcpy(buffer
, temp
, sz
);
3159 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3161 char *end
= buffer
+ buflen
;
3162 /* these dentries are never renamed, so d_lock is not needed */
3163 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3164 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3165 prepend(&end
, &buflen
, "/", 1))
3166 end
= ERR_PTR(-ENAMETOOLONG
);
3169 EXPORT_SYMBOL(simple_dname
);
3172 * Write full pathname from the root of the filesystem into the buffer.
3174 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3176 struct dentry
*dentry
;
3189 prepend(&end
, &len
, "\0", 1);
3193 read_seqbegin_or_lock(&rename_lock
, &seq
);
3194 while (!IS_ROOT(dentry
)) {
3195 struct dentry
*parent
= dentry
->d_parent
;
3198 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3207 if (need_seqretry(&rename_lock
, seq
)) {
3211 done_seqretry(&rename_lock
, seq
);
3216 return ERR_PTR(-ENAMETOOLONG
);
3219 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3221 return __dentry_path(dentry
, buf
, buflen
);
3223 EXPORT_SYMBOL(dentry_path_raw
);
3225 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3230 if (d_unlinked(dentry
)) {
3232 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3236 retval
= __dentry_path(dentry
, buf
, buflen
);
3237 if (!IS_ERR(retval
) && p
)
3238 *p
= '/'; /* restore '/' overriden with '\0' */
3241 return ERR_PTR(-ENAMETOOLONG
);
3244 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3250 seq
= read_seqcount_begin(&fs
->seq
);
3253 } while (read_seqcount_retry(&fs
->seq
, seq
));
3257 * NOTE! The user-level library version returns a
3258 * character pointer. The kernel system call just
3259 * returns the length of the buffer filled (which
3260 * includes the ending '\0' character), or a negative
3261 * error value. So libc would do something like
3263 * char *getcwd(char * buf, size_t size)
3267 * retval = sys_getcwd(buf, size);
3274 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3277 struct path pwd
, root
;
3278 char *page
= __getname();
3284 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3287 if (!d_unlinked(pwd
.dentry
)) {
3289 char *cwd
= page
+ PATH_MAX
;
3290 int buflen
= PATH_MAX
;
3292 prepend(&cwd
, &buflen
, "\0", 1);
3293 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3299 /* Unreachable from current root */
3301 error
= prepend_unreachable(&cwd
, &buflen
);
3307 len
= PATH_MAX
+ page
- cwd
;
3310 if (copy_to_user(buf
, cwd
, len
))
3323 * Test whether new_dentry is a subdirectory of old_dentry.
3325 * Trivially implemented using the dcache structure
3329 * is_subdir - is new dentry a subdirectory of old_dentry
3330 * @new_dentry: new dentry
3331 * @old_dentry: old dentry
3333 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3334 * Returns 0 otherwise.
3335 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3338 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3343 if (new_dentry
== old_dentry
)
3347 /* for restarting inner loop in case of seq retry */
3348 seq
= read_seqbegin(&rename_lock
);
3350 * Need rcu_readlock to protect against the d_parent trashing
3354 if (d_ancestor(old_dentry
, new_dentry
))
3359 } while (read_seqretry(&rename_lock
, seq
));
3364 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3366 struct dentry
*root
= data
;
3367 if (dentry
!= root
) {
3368 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3371 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3372 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3373 dentry
->d_lockref
.count
--;
3376 return D_WALK_CONTINUE
;
3379 void d_genocide(struct dentry
*parent
)
3381 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3384 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3386 inode_dec_link_count(inode
);
3387 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3388 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3389 !d_unlinked(dentry
));
3390 spin_lock(&dentry
->d_parent
->d_lock
);
3391 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3392 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3393 (unsigned long long)inode
->i_ino
);
3394 spin_unlock(&dentry
->d_lock
);
3395 spin_unlock(&dentry
->d_parent
->d_lock
);
3396 d_instantiate(dentry
, inode
);
3398 EXPORT_SYMBOL(d_tmpfile
);
3400 static __initdata
unsigned long dhash_entries
;
3401 static int __init
set_dhash_entries(char *str
)
3405 dhash_entries
= simple_strtoul(str
, &str
, 0);
3408 __setup("dhash_entries=", set_dhash_entries
);
3410 static void __init
dcache_init_early(void)
3414 /* If hashes are distributed across NUMA nodes, defer
3415 * hash allocation until vmalloc space is available.
3421 alloc_large_system_hash("Dentry cache",
3422 sizeof(struct hlist_bl_head
),
3431 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3432 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3435 static void __init
dcache_init(void)
3440 * A constructor could be added for stable state like the lists,
3441 * but it is probably not worth it because of the cache nature
3444 dentry_cache
= KMEM_CACHE(dentry
,
3445 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3447 /* Hash may have been set up in dcache_init_early */
3452 alloc_large_system_hash("Dentry cache",
3453 sizeof(struct hlist_bl_head
),
3462 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3463 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3466 /* SLAB cache for __getname() consumers */
3467 struct kmem_cache
*names_cachep __read_mostly
;
3468 EXPORT_SYMBOL(names_cachep
);
3470 EXPORT_SYMBOL(d_genocide
);
3472 void __init
vfs_caches_init_early(void)
3474 dcache_init_early();
3478 void __init
vfs_caches_init(void)
3480 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3481 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3486 files_maxfiles_init();